ML20356A235

Day 3 - Program and Presentations of Workshop: Digital Twin Applications for Advanced Nuclear Technologies, December 1 - 4, 2020 - Part 4 of 5

ML20356A235
Person / Time
Issue date:	12/01/2020
From:	NRC/OCM, Ontario Power Generation
To:
Shared Package
ML20356A241	List: ML20356A235
References
Download: ML20356A235 (278)

Text

Qualification of the Pickering A Test Facility Workshop on Digital Twin Applications for Advanced Nuclear Technologies Richard Henry (OPG), John Sladek (CNSC) l 1-4 December 2020

Overview 1 l What is the Pickering A Test Facility (PATF) and how is it used?

2 l What are the regulatory requirements and process for qualification of Software Tools?

3 l How was the PATF qualified?

p2

Context

• Pickering A 4x540 MW CANDU Pressurized Heavy Water Reactors

• Placed in service 1971-73. Currently two units in service. Two units in safe store.

• Digital Control Computers (DCC) control major plant processes (Reactor power, Boiler pressure, Online fuel handling)

• Obsolescence issues with legacy test facilities used for software Verification and Validation testing.

• A software-based test facility (digital twin) was developed.

• Qualified September 2011

• Followed existing qualification processes

• No new processes required p3

DCC System Context p4

Pickering A Test Facility Overview

• High-fidelity DCC Emulation

• Simulated Operator Human Machine

• Plant Simulation Interface

• Function and timing

• Process models

• Annunciation computer, keyboard

• Instructions, peripherals,

• Simulated relay logic and displays process input/output, interrupts

• Simulated stand-alone devices

• Control room panels (e.g., PLCs, digital meters) p5

• Contact alarm scanner

Test Execution for V&V Activities

• Software engineering tool used within the development lifecycle

1. Restore 100% full power storepoint

• Test execution controlled by scripting language (Python) restore('/home/patf/storept/FPSS_ibm1800.stp')

1. Set Digital Input 0x41, bit 10 on DCC1 off

• Able to control DCC, simulation, etc.

1. Set Digital Input 0x41, bit 10 on DCC2 off

• Can simulate operator actions (keyboard, panel switches) dcc1.di[0x41].b10=0 dcc2.di[0x41].b10=0

• Can override simulation with substitute values # Print simulated reactor neutron power print "Starting Power: ", getcdb ('RNTPOW')

• Observe/record process parameters or internal DCC states # On ACP1, pull up the RRS display and

1. decrease reactor power by 10% at rate 3

• Save and restore storepoints pace.acp1.keypress("<RRS>")

pace.acp1.keypress("RD3<ENTER><EXECUTE>")

• Available for all major plant modes of operation (0%-100%FP) # Run for 30 seconds run_for (seconds=30.0)

• Process model state

1. Print simulated reactor neutron power

• Internal DCC state (memory, disk, registers) print "Ending Power:", getcdb ('RNTPOW')

1. Print DCC1 Analog Input @Address /1D01

• Allows for development of fully-automated repeatable tests # in ADC counts and eng. units print dcc1.ai[0x1D01], dcc1.ai[0x1D01].eng

• Useful for training new staff in DCC fundamentals # Display core location 0x1800 print dcc1.core [0x01800]

• Scenario based analysis of control algorithm changes # Print dcc registers A, Q and I print dcc1.a, dcc1.q, dcc1.i p6

Regulatory Requirements for Software QA

• The Pickering Nuclear Generating Station License and Condition Handbook (Effective 17 April 2020):

• The licensee shall implement and maintain a management system

• Licensing basis publication: CSA N286, NPP QA Program Requirements.

• Compliance Verification Criteria include:

• OPG N-CHAR-AS-002, Nuclear Management System,

• OPG N-PROG-MP-0006, Software

• The licensee shall implement and maintain a design program.

• Guidance: N290.14, Qualification of Digital Hardware and Software for Use in Instrumentation and Control Applications for Nuclear Power Plants

• The licensee shall implement and maintain a safety analysis program.

• Guidance: N286.7, Quality assurance of analytical, scientific and design computer programs for nuclear power plants p7

Software Qualification

• N-PROG-MP-0006, Software, defines processes for all types of software including,

• Software Engineering Tools follows OPG N-STI-69000-10002, Qualification of Software Engineering Tools

• This is key to the qualification of PATF for the use case for V&V of software

• Testing requirements based on categorization of Target software (RTPC or SESA)

• Real Time Process Computing (RTPC). QA Requirements based upon software classification:

Reference:

http://www.world-nuclear.org/uploadedFiles/org/WNA/Publications/Working_Group_Reports/safety-classification-for-iandc-systems-in-npps.pdf p8

• Scientific, Engineering and Safety Analysis (SESA) software follow CSA N286.7 QA requirements

N-STI-69000-10002

• Qualification requirements for software engineering tools:

Software engineering tools are those used to support any aspect of the software engineering lifecycle, including: requirements gathering and specification; design and code production; review and static verification of requirements, design and code; test case generation, execution, and results analysis; configuration management and change control; and training.

• Method

1. Identify target software classification and categorization

2. Select tools

3. Determine impact severity of tool failure

4. Determine mitigating circumstances

5. Select a qualification approach

6. Perform qualification activities

7. Configuration management of software engineering tools

8. Software engineering tool qualification report p9

Determine Severity of Tool Failure

• Guidance: Consider the failure modes for each use of the software engineering tool, and identify the relevant failure effects on the target software. Classify each failure effect based on its potential impact on the safety, functional, reliability, performance or security requirements of the target software.

• Classification Scheme:

Failure Type Description Direct the tool is incorporated in the target application. Tool is to be considered pre-developed software and qualified to the same degree of rigor as the target software.

Indirect-Causal Tool failure can introduce errors in the target software which if undetected could result in the target software failing to meet the above requirements.

Indirect-Preventive A tool failure effect can result in the non-detection of errors in the target software which could result in the failure of the target software to meet the above requirements.

Minimal A tool failure could have an impact on the target software but no mechanism has been identified that could result in the target software failing to meet the above requirements.

No Impact A tool failure can have no impact on the target software in meeting the above requirements.

p10

Identify and Classify Mitigations

• Identify any mitigations that eliminate or reduce the impact of the failure effect.

• Classify mitigations as one of the following:

Class Description None Single Single reliable mitigating activity or procedure which defends against impact.

Must be independent of the failure effect (efficacy of mitigation not diminished or nullified by the failure effect).

Examples: Testing, review, checksum comparison Multiple Multiple reliable mitigating activities or procedures which defend against impact.

Must be independent of the failure effect.

Must be independent of each other (having no other common failure mechanism)

Example: Review of outputs by two independent individuals using different methods.

p11

Determine Qualification Approach

• Qualification grade is determined based upon:

• Target software classification and categorization

• Impact severity of tool failure

• Mitigations

• Result is: NSR2, NSR3, O, and A

• Qualification method is based on grade

• NSR2 / NSR3: Follow RTPC Category II/III qualification method (e.g., CSA N290.14)

• A: Follow SESA qualification method (e.g., CSA N286.7)

• O: Select qualification method from. For example:

• Acceptance testing

• Widespread industry usage (for same purpose)

• Operating history from third party p12

Results for PATF

• Target Software Classification and Categorization: System is used for testing of DCC software which is Categorized.

Category 2.

• Impact: failure of the software test tool could result in non-detection of errors in the target software.

Indirect-Preventive

• Mitigation:

• Several sets of tests are performed by independent individuals (e.g., unit testing, subsystem testing, integration testing and validation testing)

• However, some of these tests could all potentially make use of the PATF, so there could be a common failure in the mitigating activities. Single

• Qualification approach is O (as per lookup table)

• Qualification method selected to be Acceptance p13 Testing

Acceptance Testing of PATF

• PATF development team had extensive experience with qualification of DCC hardware and with the software QA processes

• DCCs previously replaced with hardware emulators (1995-2001)

• Qualification tests focused on:

• Quality of emulation:

• Included execution of all OEM diagnostics

• Test cases for DCC features, based on OEM documentation

• Test cases to test all features of scripting language

• Testing of integration of Emulated DCC with plant simulation

• Testing of transfer of control mechanism

• Qualification tests were implemented using scripting language

• Simplified qualification testing of a new release of PATF in 2014 p14

Copyright 2020 Nuclear Virtual Engineering Capability Dr. Albrecht Kyrieleis

Agenda Context for NVEC NVEC elements Case studies Future developments 2 ©Jacobs 2020

The Nuclear Innovation Programme ~£30M ~£150M Research Theme Apr 18 Apr 19 Apr 20 Apr 21 Accident Tolerant Fuels Coated Particle Fuels Advanced Fuels Pu containing fast reactor fuels Reactor physics Nuclear Data Thermal hydraulic model development Thermal hydraulic facility development Reactor Design Reactor safety and security Virtual engineering Modelling and simulation Spent fuel recycle Development of proliferation resistant spent fuel recycle technology Materials testing and development Advanced component manufacturing Materials and Manufacturing Large scale manufacturing / assembly Prefabrication module development Codes and standards Strategic assessments Nuclear facilities and Fast reactor knowledge capture strategic toolkit Regulatory engagement Access to irradiation facilities Feasibility Study Advanced Modular Reactors Design Development 3

©Jacobs 2020

Challenge Innovative nuclear power plants needed to meet the UK Government commitment of net zero carbon emissions by 2050 By 2 0 3 0 deliver Cost savings of 30% on new build, 20% on decommissioning

£2bn domestic and international contract wins Silo practices Information sharing Innovation Cost management 4 ©Jacobs 2020

The NVEC Partnership Jacobs Lead Wider partners include: Digital Catapult, University of Bangor, University of Bristol, EvoMetric Collaboration includes: UKAEA, Fraser-Nash, Sellafield Ltd, Menai Science Park 5 ©Jacobs 2020

NVEC Elements Develop collaborative digital environment to support the nuclear life cycle Use existing technology where possible Open and highly flexible Develop operating model, standards and guidance Demonstrate benefits of digital environment in various case studies Involve stakeholders Early adoption Develop community which can assume responsibility for issuing guidance, maintaining standards, discussing and resolving common technological issues ensuring a common approach across the sector 6 ©Jacobs 2020

NVEC Phase 2 Environment Design Build Operation Decommissioning Waste storage NVEC Digital Environment Common Data Common Modelling Standards for virtual Environment Environment engineering Run distributed Enable single Securely share Integrate Enable analyses on source of truth information analysis tools digital twins common data 7 ©Jacobs 2 0 2 0

Benefits Increased

• oReduced return on investment through efficient operation & maintenance costs o oLower Singlerisk leading source to reduction of design in financing data; collaborative cost environment o Increased return on investment through efficient operation & maintenance o Lower risk leading to reduction in financing costs

• Shortened development times o Efficient Design & licensing ; Integrated multi-physics approach o More reliable prediction of development times, allowing better synchronisation

• Enhanced credibility, operability, reliability & safety o Real time understanding of plant, better planning and predictive maintenance o Enhanced training & skills development o Reduced risk and perception of risk

• Cross-discipline transfer of expertise; joined-up industry

• Enables innovation and new technology adoption; diverse users 8 ©Jacobs 2020

Architecture Status Initial implementation of all components complete Continuous improvement Deployed on different systems Application to various cases on-going 9 ©Jacobs 2020

Networked Architecture 10 ©Jacobs 2020

Change Control using the Data System 0

Baseline 1 2 head Study A 1

Project 2

3 BoM Materials Uncertainty /

Sensitivity Pipe Pum p Concept A 1 2 Baseline Study B 1 2 3 4 5 Changed Diam eter Flow Rate Optimisation Concept B 1 2 Length CAD Concept Study C Rejected 1 CAD 2

3 4

DOE 11 ©Jacobs 2020

Graphical User Interface 12 ©Jacobs 2020

NVEC Multi-Scale Simulation Benefits Features Single tool can analyse many Break down model into different designs with few changes hierarchy of components SYSTEM Value and efficiency Rapid turn-around from concept to Equation Orientated outcome from an analysis modelling approach or dedicated code SUB-SYSTEM Detailed component analysis via Balance of Plant dedicated code where required Code coupling via plug and play modular design COMPONENT Analyse faults faster as plant simulator and control system can Scalable to allow Thermal Hydraulics have common features deployment in a range of applications Sub-component Core Analysis 13 ©Jacobs 2020

Case Study: System Level Modelling Component libraries can be re-used Complex system model of Molten Salt AMR Simulation of operational sequences developed and implemented Reactor analysis optionally using WIMS code 14 ©Jacobs 2020

Multi -Scale Simulation GUI

[Screen shot Sys Lev Sim GUI]

15 ©Jacobs 2020

Case Study: AGR Graphite Workbench Comprehensive simulation of reactor graphite properties over time X

Workbench 2.0 MoFEM Workbench 1.0 NVEC software environment Supervisor NVEC Code HPC Aster MoFEM Code Supervisor Aster Status: EDF Project Collaborator Basic implementation of Workbench 2 .0 functional repository Benefits of integration with NVEC Enables sharing of computational infrastructure / increased fidelity / inclusion of future modules Improved collaboration between sub-contractors Standardisation of analyses: less QA effort and training 16 ©Jacobs 2020

Integration with IIoT Processing of sensor data enabling improved data use Integration of static and time series data Status:

Design in progress Plant at NAMRC setup to export sensor data Model of plant sub-system Enabling NVEC software Connector developed (Frazer-Nash)

IIOT system Digital Twin environement Collaboration with Digital Catapult / SMEs on related methods/ technologies Simulation based on Import of sensor data real-time sensor data Manage design data of plant 17 ©Jacobs 2 0 2 0

Further Applications Decommissioning Integration of simulations with point cloud data from innovative decommissioning project (IIND)

Reactor Physics Completed design of coupling of codes for key workflow (WIMS-ENIGMA) in NVEC THOR(Thermal Hydraulic Open-Access Research Facility), University of Bangor Collaboration started aiming at involving NVEC from design stage onwards Data Model for THOR developed FAITH Application of NVEC approach and tools on-going 18 ©Jacobs 2020

SME Discovery Workshop (held in Sept 20)

Exploring opportunities for collaboration NVEC enabling innovation through SMEs 19 ©Jacobs 2020

Future Developments SMR Use NVEC for key requirements: e.g. engineering data management, design, change control Initial NVEC evaluation version in development for Rolls-Royce SMR Fusion (STEP, CHIMERA)

Develop requirements / information model System level digital representations enabling optimisation strategies e.g. AMR, process heat, Hydrogen-Nuclear combinations Implementation of operational Digital Twins for existing facilities Further development of application for FAITH and THOR/ THUNDER 20 ©Jacobs 2020

Future Developments Advanced Materials & Manufacturing: more effective structural integrity management Increased effectiveness of safety case support Social factors study of benefits/ obstacles of digitalisation in nuclear Develop working practices to support an information management strategy between disciplines Develop standards/ guidance aiming at a NVEC Community Link to the Construction Sector Deal 21 ©Jacobs 2020

Summary NVEC to help deliver UK Government net zero carbon emissions target by 2050 Key challenges: Silo practices, information sharing, innovation, cost management Development of collaborative digital environment along with standards/guidance NVEC community: responsible for issuing guidance, maintaining standards, ensuring a common approach across the sector Various case studies on-going demonstrating benefits of NVEC Broad range of future opportunities 22 ©Jacobs 2020

Acknowledgments C. Phelps, A. Aslam (Jacobs)

D. Bowman, K. Vikhorev (University of Liverpool - VEC)

M. Bankhead (NNL)

C. Jackson (Rolls-Royce)

J. Draup, P. Martinuzzi (EDF-Energy)

S. Marr (NAMRC) 23 ©Jacobs 2020

Thank you 1

Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications NRC&INL&ORNL 2020 - December 3rd Susana López (slopez@tecnatom.es)

Pablo Rey (prey@tecnatom.es) www.tecnatom.es All right reserved 2017

2 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications CONTENTS Who we are The concern Success Story DT P40 Other success stories Conclusions www.tecnatom.es All right reserved 2017

3 Who we are www.tecnatom.es All right reserved 2017

4 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Who we are MARKETS ENERGY INDUSTRIAL Training REACTORS NEW COMBINED CYCLE AEROSPACE TECHNOLOGICAL CAPABILITIES IN OPERATION REACTORS THERMAL PETROCHEMICAL RENEWABLE RAILWAY Simulation Control Rooms Operation Support Emergency Support Engineering Inspection Tests NDT Systems NDT Products www.tecnatom.es All right reserved 2017

5 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Who we are www.tecnatom.es All right reserved 2017

6 The concern www.tecnatom.es All right reserved 2017

7 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications The concern BWR/6 manufactured by General Electric Electrical power 1,092.02 MW Located in Spain Key facts

• 1st coupling: 14 October 1984

• Commercial operation: 11 March 1985 BOP DCS

• Honeywell TDC 3000 since 1988

• Installed in Full-scope simulator in 2002

• Migrated to Experion in 2005

• Design modifications: control room digitalization www.tecnatom.es All right reserved 2017

8 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications The concern : BWR NPP heat sinks Normal Operation: Main Sink

• 2 natural draught cooling towers (Main condenser)

• Forced draught towers (Auxiliary systems)

Alternative: Ultimate Heat Sink (UHS)

• Pond: 30 days autonomy

• ESW: 3 cooling water pumping and distribution sub-systems

• LOCA or LOOP DIV I & II www.tecnatom.es All right reserved 2017

9 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications The concern: Essential Services Water pipes effective section Carbon Steel Pipes Iron Oxide accumulation Chemically pretreated water www.tecnatom.es All right reserved 2017

10 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications The concern: Essential Services Water pipes effective section Carbon Hinder Tech Specs Steel Pipes Efective compliance section reduction Iron Oxide accumulation Chemically ESW Flow & P periodic pretreated Corrective Valves Actions surveillance water position Cleaning &

maintenance www.tecnatom.es All right reserved 2017

11 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications The concern : the solution Digital Twin to ensure compliance with Technical Specifications www.tecnatom.es All right reserved 2017

12 Success story DT P40 www.tecnatom.es All right reserved 2017

13 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT P40 ESW integration into DCS Tools for the automated fulfillment of surveillance reports.

Digital Twin www.tecnatom.es All right reserved 2017

14 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT P40 : ESW integration into DCS ESW Data Acquisition Flow, P Modbus Valves PLC TCP DCS UHS Level, servers T

HSI displays:

• Division I, II, III

• Heat exchangers pressure drop (K factor) www.tecnatom.es All right reserved 2017

15 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT P40 : Tools for automatization of surveillance reports DCS Station DCS displays HMI ESW DATA MS Excel ACQUISITION OPC client Datasheets in Execution Each division surveillance books:

• 24M: Minimum flow surveillance (all components)

• 03M: Pumps functional capacity (current and historical working point)

• Pressure drop factor (K= dP/Q2)

• Heat exchangers & filters

• Graph over time www.tecnatom.es All right reserved 2017

16 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT P40 : Digital Twin Simulator Excel Files Bernuilli net model ESW DATA 24M: last global ACQUISITION surveillance test Baseline 3M: Partial current Updates working points Partial pressure drops www.tecnatom.es All right reserved 2017

17 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT P40 : Digital Twin www.tecnatom.es All right reserved 2017

18 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT P40 : Digital Twin RESULTS www.tecnatom.es All right reserved 2017

19 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT P40 : Digital Twin Simulator Excel Files Bernuilli net model ESW DATA 24M: last global ACQUISITION surveillance test Baseline 3M: Partial current Updates working points Partial pressure What IF?

drops Prediction

• Quantitative result of eventual system use or complete test Best system configuration

• Simulate valves and pumps operation www.tecnatom.es All right reserved 2017

20 Other success stories www.tecnatom.es All right reserved 2017

21 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT TecOS SOLCEP ASME PTC PM 1993(2010).

Performance Monitoring Guidelines for Power Plants Efficiency Calculation Energy TecOS Deviation Accounting SOLCEP Diagnosis Operation Optimization Assisted Diagnostic (Tecnatom Engineers) www.tecnatom.es All right reserved 2017

22 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Project DT TecOS SOLCEP www.tecnatom.es All right reserved 2017

23 Conclusions www.tecnatom.es All right reserved 2017

24 Benefits of Digitalizing and Employing Simulation to Increase Plant System Performance and Ensure Compliance with Technical Specifications Conclusions Benefits:

• Maximizing operation increasing safety margins

• No need for maintenance shutdown

• Eases the evaluation of blockages and soiling

• On-line information (digitalization)

• Optimizes preventive maintenance tasks (cleaning)

• Optimizes operator workload:

• Automates calculations and surveillance and test requirements reports Key Factors:

• Integrated approach for design modifications

• Simulation, I&C, HFE, Operation www.tecnatom.es All right reserved 2017

www.tecnatom.es

©2017 Tecnatom, S.A.

Todos los derechos reservados. El contenido de esta obra está protegido por la Ley y no podrá ser reproducida, ni en todo ni en parte, ni transmitida, ni registrada por ningún sistema de recuperación de información, en ninguna forma ni por ningún medio, sin el consentimiento previo y por escrito de Tecnatom, S.A. y de sus autores.

Tecnatom y el logotipo de Tecnatom son marcas registradas de Tecnatom, S.A.

THE EU FRAMEWORK PROGRAMME FOR RESEARCH AND INNOVATION EURATOM RESEARCH AND TRAINING PROGRAMME FISSION RESEARCH Panagiotis MANOLATOS DG RTD Clean Planet panagiotis.manolatos@ec.europa.eu DIGITAL TWIN 4-5 December 2020

Horizon 2020 Challenges - Overall budgets Total budget H2020:

EUR 74,83 billion Budget of the Energy Challenge:

EUR 5,69 billion

H2020 EURATOM Council Regulation Euratom indirect actions specific objectives:

(a) supporting safety of nuclear systems; (b) contributing to the development of safe, longer term solutions for the management of ultimate nuclear waste, including final geological disposal as well as partitioning and transmutation; (c) supporting the development and sustainability of nuclear expertise and excellence in the Union; (d) supporting radiation protection and development of medical applications of radiation, including, inter alia, the secure and safe supply and use of radioisotopes;

Council Regulation Euratom indirect actions specific objectives:

(e) moving towards demonstration of feasibility of fusion as a power source by exploiting existing and future fusion facilities; (f) laying the foundations for future fusion power plants by developing materials, technologies and conceptual design; (g) promoting innovation and industrial competitiveness; (h) ensuring availability and use of research infrastructures of pan-European relevance.

Current Euratom Nuclear Fission and Radiation Protection budget share 6

Source: EC

• Types of Actions - Research/Innovation Research and Innovation Actions They are actions with Research and Development activities as the core of the project intending to establish new scientific and technical knowledge and/or explore the feasibility of a new or improved technology, product, process, service or solution

- may include basic and applied research, technology development and integration, testing and validation on a small-scale prototype in a laboratory or simulated environment

- may contain closely connected but limited demonstration or pilot activities aiming to show technical feasibility in a near to operational environment

• 100% funding rate "Pure" Innovation Actions

"'Innovation action' means an action primarily consisting of activities directly aiming at producing plans and arrangements or designs for new, altered or improved products, processes or services. For this purpose they may include prototyping, testing, demonstrating, piloting, large-scale product validation and market replication"

• 70% funding rate (100% for non-profit legal entities)

• Types of Actions - Coordination and Support Coordination and Support Action Actions consisting primarily of accompanying measures such as standardisation, dissemination, awareness-raising and communication, networking, coordination or support services, policy dialogues and mutual learning exercises and studies, including design studies for new infrastructure and may also include complementary activities of strategic planning, networking and coordination between programmes in different countries.

Nuclear Fission & Radiation Protection Research (NRFP)

Call 2019-2020 Calendar WP Adoption: 14 December 2018 Call Open: 15 May 2019 Submission deadline: 25 September 2019 Evaluation: November 2019 62 proposals received EC requested : EUR 265 million EC budget : EUR 134 million Signature of GAs: May 2020 9

Research and Innovation Actions (RIA)

Topic Budgets (EUR million)

Nuclear safety - NFRP-01: Ageing phenomena of 16 components and structures and operational issues Nuclear safety - NFRP-02: Safety assessments for LTO 12 upgrades of Generation II and III reactors Nuclear safety - NFRP-03: Safety margins determination 8 for design basis-exceeding external hazards Nuclear safety - NFRP-05: Support for safety research of 8 Small Modular Reactors Nuclear safety - NFRP-06: Safety Research and 7.6 Innovation for advanced nuclear systems Nuclear safety - NFRP-07: Safety Research and 6 Innovation for Partitioning and/or Transmutation 10

Coordination and Support Actions (CSA)

Topic Budgets (EUR million)

Nuclear safety - NFRP-08: Towards joint European effort in 1.1 area of nuclear materials Education and Training - NFRP-11: Advancing nuclear 5 education Research Infrastructure - NFRP-16: Roadmap for use of 1.1 Euratom access rights to JHR experimental capacity Research Infrastructure - NFRP 17: Optimised use of 1.1 European research reactors Innovation Action (IA)

Nuclear safety - NFRP-04: Innovation for Generation II and 12 III reactors 11

Nuclear safety Topics NRFP - 1, 2 Max EC Duration Total cost Topic Acronym Title contribution (Months) (M)

(M)

Towards improved assessment of safety NFRP-01 ACES performance for long-term operation of 48 4 5,5 nuclear civil engineering structures European database for multiscale ENTENTE 48 4 5 modelling of radiation damage Increasing safety in npps by covering INCEFA-SCALE gaps in environmental fatigue 60 4 6,8 assessment - focusing on gaps between laboratory data and component SCALE STRUMAT-LTO Structural materials research for safe 48 4 4,8 long term operation of LWR npps NFRP-02 AMHYCO Towards an enhanced accident 48 4 4 management of the hydrogen/co combustion risk APAL Advanced PTS analysis for LTO 48 4 4,6 Codes and methods CAMIVVER 4 improvements for VVER comprehensive 36 4 safety assessment 12

Abstracts, coordinator, and further info is published as soon as the Grant Agreements are signed and can be found at :

https://cordis.europa.eu/projects/en 13

International Cooperation Multilateral International Energy Agency (IEA)

Nuclear Energy Agency (OECD-NEA)

International Atomic Energy Agency (IAEA)

Bilateral Association Agreements with Switzerland and Ukraine Cooperation with Japan, Canada, US, China, Korea, Brazil, Argentina 14

• International Cooperation Participation Funding Open for all legal entities Third country identified in established in third countries the Work Programme and for international or organisations.

participation deemed by Restrictions only possible if the Commission essential introduced in the work in the action programme. or For reciprocity reasons when provided under a bilateral scientific and For security reasons technological agreement

EU priorities: 2021-2027 MFF proposal 16 Source: EC

Commission proposal for Horizon Europe THE NEXT EU RESEARCH & INNOVATION PROGRAMME (2021 - 2027)

Horizon Europe budget proposal (2021-2027) 100 B including 3.5 from InvestEU 18 Source: EC

Thank you!

1. HorizonEU http://ec.europa.eu/horizon-europe

© European Union, 2019. l Images source: © darkovujic, #82863476; © Konovalov Pavel, #109031193; 2018. Fotolia.com

European R&D&I towards Digital Twins A. Al Mazouzi (General Secretariat)

SNETP in a nutshell SNETP was set up in 2007 under the auspices of the European Commission with the goal to support technological 105 full members development for enhancing safe and competitive nuclear as of oct20 fission in a climate-neutral and sustainable energy mix.

In line with the objectives of the SET-Plan, SNETP aims to contribute to: 11%

Research organisations Lowering European greenhouse gas emissions Industry 15% 39%

Assuring security of energy supply for Europe SMEs Stabilizing electricity prices in Europe 17% Academia The association gathers various types of stakeholders: Other 18%

industry, research centres, safety organisations, universities, non-governmental organisations, SMEs, etc.

2

Objectives Providing solutions to Industry Promoting Scientific Excellence Foster industrial-driven research addressing the needs Agree on, implement and promote common R&I of SNETP industrial members in particular regarding priorities within the SNETP community safety, supply chain, licensing and cost-representing the three pillars competitiveness Boosting Innovation Cooperating closely with Regulators Facilitate industrial-driven and intersectoral innovation (digital, robotics, materials, etc.) in Reinforce cooperation between SNETP and the nuclear for current and new applications (non- different regulatory and standardization bodies.

power, hydrogen, etc.) Supporting R&D infrastructures Support projects and initiatives aiming at Representing nuclear fission R&D in European Affairs maintaining/refurbishing/building the needed Promote SNETP expertise and research priorities infrastructure to perform R&D&I in the nuclear field.

towards European institutions Strenghtening International Relations Sharing Experience with European Associations Promote SNETP expertise and research priorities Fostering and coordinating interactions with towards international nuclear institutions (IAEA, European associations in the field of nuclear, and any OECD/NEA, GIF, etc.) other sector with potential mutual interests with nuclear.

Engaging with Civil Society Engage with civil society and non-nuclear stakeholders to rationalize the debate on the European energy mix and enhance the acceptability of nuclear.

3

SNETP-Strategic Research and Innovation agenda Establishes long-term research priorities for its members Provides a clear research plan for industry, policy makers and research centers Provides state of the art analysis on nuclear research

& innovation topics in line with European foreseen electricity mix in 2050 and the Green deal Prioritizes the topics of added value to the end users Create a synergy between various industrial sectors:

cross-sectorial innovation (digital, material, space, ocean, robotics, etc.)

Establish win-win relationship with national/European and international stakeholders Initiate and disseminate innovation within the nuclear sector 4

Wo is SNETP?

5

Current state of Digital Reactor

Different uses in Reactor Simulations Higher representativity Simulators Best estimate Best efforts High Fidelity

• Operators training

• Quantification of

• Driver assistance systems simulation biases

• Operations studies

• Reference for safety studies

• Reference calculations

• Design Studies

• Studies in extreme situations

• Reactor Design (accidents)

• Accidents and safety

• Substitute for experiments studies where no data are available The European Nuclear sector has a long standing experience in developing a lot of physics codes including state of the art thanks to the EURATOM support and 7

international collaboration

Emphasis on codes - Neutronics neutronics code

• Both lattice and core calculations

• Transport solvers on unstructured meshes

• Parallelization on thousands of nodes

• Depletion chain with more than a thousand isotopes

• Allows advanced calculation such as direct calculation (on going work)

H2020 projects:

- ARIEL (2019-2023)

- SANDA (2020-2024)

Emphasis on codes - Thermalhydraulics CFD code

• Single and multiphase flows

• RANS and LES turbulent models

• Unstructured meshes and parallelization on tens of thousands of nodes

• Multiphysics: Fire, Severe Accidents, turbomachinery, ground water flows,

• Single and multiphase flows

• Based on the porous media assumption

• Used for Cores, Steam Generators, Heat exchangers

• used for Safety analysis, Core refueling operations ad R&D studies H2020 projects (exemples):

- McSafe (2017-2020); McSafer (2020-2024)

- Cortex (2017_2021)

- PIACE (2020-2024)

- CAMVVER (2020-2024)

Advanced Modeling Applications Typical use case : Steam-line break accident (SLB)

This transient has been studied for decades by with different simulation tools.

Very complex situation with strong physics coupling and 3D effects :

good candidate for advanced simulation codes (CFD, neutronics transport with unstructured meshes)

Allows benchmarking between legacy and new generation of codes (test for code interchangeability)

Possibility benchmarking with other international software (VERA from CASL,)

Good candidate for advanced visualization techniques to help understand the physics.

H2020 projects (examples):

- INCEFA+ & INCEFA-SCALE (2015-2025)

- MEACTOS (2017-2021)

- MUSA (2020-2024)

- APAL (2020-2024)

Improvements in Reactor Simulations Representativity/quality Simulators Best estimate Best efforts High Fidelity

• Operators training

• Quantification of

• Driver assistance systems simulation biases

• Operations studies

• Design Studies

• Reference for safety studies

• Reference calculations

• Reactor Design

• Studies in extreme situations

• Accidents and safety (accidents) studies

• Substitute for experiments where no data are available Interoperability / Interchangeability Exploitation R&D / Expertise 11

Codes and Data integration DATA SOFTWARE INTEROPERABILITY & MODULARITY Data acquisition from physical tests and exploitation Minimization of Validation of the experiments tests Digital Reactor CAD and Geometries calculation Advanced Modeling of the whole core Operating point Optimization of Physics coupling including Physical tests the number of Uncertainties and database calculations Quantification validation Validation of Models reduction modelling Real time modeling of the power plant Improving physics Visualization of complex representativity + physics phenomena Users Automation of the validation Needs Nugenia 2019

Some Scientific and technical challenges Goals / Challenges Need of collaboration (European and international)

Building a multi-physics (interoperability) and multi-scale (interchangeability) platform where all relevant physics codes should be able to plug in seamlessly.

Being able to come together with a common standard (API, data model exchange) for both new and legacy codes.

Building bridges to allow advanced codes to be used in simulators as well.

Using reduction models techniques for at least real-time simulation .

Taking into account, from the ground up, the possibility to quantify uncertainties.

Developing the right methodology for propagating uncertainties when doing multi-physics.

Being able to understand the physics involved as complexity increases dramatically.

Using advanced, ergonomic, visualization techniques (metaphors, AR, VR) as a helping tool.

V&V of the whole platform when using strongly coupled physics.

SNETP added value SNETP is the only European wide association dedicated to collaborative nuclear research.

All major European R&D organisations involved in nuclear are members of the association.

Various events are organised and online tools are deployed to facilitate collaboration of the community on new projects proposals. Since its creation in 2007, SNETP has supported discussions on approximately 300 project ideas.

The specific European Technology & Innovation Platform (ETIP) status provides an important visibility to SNETP and its members, with privileged access to relevant high-level managers within EU institutions, international organisations, and member states.

SNETP and its members contribute to the shaping of European energy policies, by exchanging with peers on research priority topics, by producing reference documents (e.g.

SRIA) on the state of R&D&I in Europe, by publishing position papers, etc.

14

Contact us www.snetp.eu secretariat@snetp.eu www.linkedin.com/company/snetp

@SNE_TP 15

October 18, 2020 Chris Spirito Nuclear Cybersecurity Specialist Digital Twins and Cyber Capability Development

2 War Operations Plan Response WOPR (circa 1983)

Simulation Software / AI:

- Joshua Simulation Models:

- Basic Strategy (Tic-Tac-Toe)

- Complex Strategy (Chess)

- Basic Warfare (Air-to-Ground Actions)

- Tactical Warfare (Theaterwide Biotoxic and Chem)

- Digital Twin (Global Thermonuclear War)

3 Cyber Capability Development (Digital Twins // Systems)

Reactor Simulators:

- IAEA Asherah, GSE GPWR, Digital Twin Targets:

- Systems (Pressurizer, Condenser, )

- Components (PLCs, FPGAs, )

- Comm Mediums (Analog, Digital, )

- Functional Targets (Diodes, Proto. Converters)

4 Cyber Capability Development (Digital Twins / Humans)

Personality Characteristics:

- Curiosity & Relentlessness

- Novelty & Creativity, Motivation and Ethics:

- Mercenary & Ideology Strategies

- Weakness Exploitation

- Denial & Deception Enumerate Interaction Pathways

5 Cyber Capability Development (Digital Twins / APTs)

Interactive Test Ranges:

- Integrated AI

- Infrastructure Modeling Attack Library:

- Validation of Capabilities

- Validation of Processes

- Theoretical Testbed

Cyber Security for Digital Twins Cynthia DeBisschop and Alan (AI) Konkal Senior Cyber Security Analysts NRC Contractors, Cyber Security Branch (CSB)

Division of Physical and Cyber Security Policy (DPCP)

Office of Nuclear Security and Incident Response (NSIR) 1

Overview

• Background

• Motivation

• Considerations for Entire Life Cycle of Digital Assets ICON LEGEND

• Cyber Security Vulnerabilities Real World and Protective Strategies Virtual Model Technology Gap

• Summary 2

Collaborative Review INL DICE Glossary1 of Terms: What Does Cyber Analyst Hear?

• Digital Twin. The computational simulation of a physical process or system that has a live link to the physical system, enabling enhanced verification of the simulation, control of the physical system, and analysis of trends via artificial intelligence and machine learning.

• Artificial Intelligence (AI). The simulation of human intelligence in computers or computer-controlled robots, allowing them to perform tasks commonly associated with intelligent beings.

• Machine Learning. The application of AI that provides systems the ability to automatically learn and improve from experience without being explicitly programmed.

• Operational AI. The application of artificial intelligence (AI) in energy systems to automate expensive and manual human activities and improve the efficiency of asset operations.

• Next Gen AI. The simulation of human intelligence in computers or computer-controlled robots, allowing them to perform tasks commonly associated with intelligent beings.

1 Source: Idaho National Laboratory (INL) Digital Innovation Center of Excellence (DICE) at https://dice.inl.gov/glossary-of-terms 3

Technology Gap Must Close Safely, If at All INL DICE Operational Artificial Intelligence2 Description The use of artificial intelligence (AI) in energy applications has a game-changing potential in automating expensive and manual human activities in various types of industries. In the energy industry, power plants (especially nuclear) rely on staff performing several types of manual activities on a regular basis.

Future energy plants, including advanced nuclear reactors, are designed to reduce the dependence on people for the operations, maintenance, and support activities of a plant. A light water nuclear power plant is typically full of analog gauges and manual actuators. By comparing a nuclear power plant control room to a modern airplane cockpit where the plane can fly itself and the pilots role can be reduced to simply monitoring the airplane, it is obvious that a significant technology gap exists that needs to be closed. Human intelligence needs to be replaced by machine intelligence in various forms of AI if this vision is to be realized.

• Comparison of nuclear power plant (NPP) control room to modern self-flying-airplane cockpit

• Mention of need to close significant technology gap if this Vision is to be realized for NPPs

• Cyber Security Analyst: Technology gap must close safely, if at all 2 Source: Idaho National Laboratory (INL) Digital Innovation Center of Excellence (DICE) at https://dice.inl.gov/operational-artificial-intelligence 4

Motivation Pop Quiz3 from October 2019 Forbes Article How to Protect Your Digital Twin Q. Which of the following is more valuable: a Boeing 777 or the digital twin of a Boeing 777?

A. The first option, the physical plane, is an expensive item - buying a new one will cost you around

$344 million. Yet, the digital twin of a 777 is far more valuable. Its the digital simulation of the plane that constantly collects situational awareness data and is used to understand and improve the ongoing performance of various parts and systems. If you control the digital twin, you control every 777 on (and above) the planet.

BONUS. Fill in the Blank.

Cyber Security Analyst: To effectively ________, protect think carefully throughout the evolution. Keep the vision in mind. Evaluate protections with every step along the road!

3 Source: Kawalec, Andrzej, How to Protect Your Digital Twin, Forbes Technology Councill Post, October 21, 2019, at https://www.forbes.com/sites/forbestechcouncil/2019/10/21/how-to-protect-your-digital-twin 5

Need to Understand/Consider (Now, Throughout Evolution, and Before Procurement or Use)

• Technology Itself

• Security Gaps

• Threat You Are Designing Against Throughout Life Cycle of

• Changes to Environment of Digital Assets Digital Assets

• Attack Surfaces of Digital Assets

• Cyber Risk

• Consequences of Cyber Attacks

• Defense-in-Depth Protective Strategies 6

Cyber Security Vulnerabilities

• Data Exfiltration. Plant sensors and data streams need to be connected to virtual model to realize concept. Digital twin is intended to be near-perfect blueprint of its real twin. Potential exists for monitoring and exfiltration of information about types of systems and sensors used by plant.

• Man in the Middle Attack. Early component failures may result due to alternated maintenance cycles based on faulty data after a compromise, if data is used for predictive maintenance. Scenarios that involved predictive component failure were used in the now famous Stuxnet attack. Untimely failure of a key component could be used as an element of a kinetic attack.

• Supply Chain Attack. Digital twins can be used to model new components, testing how they will perform under real-world conditions. Data obtained from components of digital twin models can be used in manufacturing. Compromised components data could lead to manufacturing faulty components.

7

Protective Strategies Fully Implement a Sound Cyber Security Framework

• NEI 08-09, RG 5.71, NIST 800-53, and NIST 800-82

• Implement security patches and remediate vulnerabilities quickly

• Harden digital twin platforms

- Utilize hash code-based allowlisting

- Remove all unnecessary files and services

- Implement Anti-Virus and Host Intrusion Detection Systems

• Identify security and remediate gaps between the twin and the physical hardware 8

Protective Strategies Secure Software Development Environments

• Develop in secure isolated environment

• Verify all Third Party and Open Source Code

• Test for language conformance, known vulnerabilities and flaws

• Conduct peer code reviews

• Use a secure repository control

• Utilize security testing techniques, fuzzing and penetration testing 9

Protective Strategies Implement Software Hardening

• Harden software to make the binary resistant to hacking

• Use coded cyclic redundancy checks (CRC) or embedded hash codes checks, binary runtime encryption

• Utilize inline coding and merge functions to minimize modular code, and altered code flow to make reverse engineering difficult

• Implement glass box techniques (Binary code can only run on designated hardware) 10

Protective Strategies Supply Chain and Intellectual Property

• Supply Chain Protection

- Audit vendors for compliance with cyber security best practices

- Test and verify third-party code releases prior to introducing them into your environment

- Purchase hardware and software from trusted sources

• Protection of All Intellectual Property

- Secure all digital twin artifacts, documents, schematics, etc.

- Protect all information flow to and from the digital twin platform

- Minimize access to source code and critical design elements 11

Summary

• This presentation offers considerations from a regulatory perspective while digital twin technology is in development.

• Before procurement or use of technology and throughout its evolution, there is a need to understand the attack surfaces and environments associated with digital assets.

• Nuclear power plant operators maintain the following throughout the life cycle of digital assets: a security defensive architecture to address the attack surfaces and environments, and multiple layers of cyber security protections to establish sufficient defense-in depth.

Defense-in-depth protective strategies are maintained to ensure the capability to detect, respond, and recover from cyber attacks.

• Such an objective depends on understanding and careful consideration of technology before procurement or use.

12

Asherah NPP Simulator Cybersecurity in a Digital Closed-Loop Environment The 2020 Workshop on Digital Twin Applications in the Nuclear Industry Rodney Busquim e Silva December 3, 2020

Nuclear Power Plants Rely on Digital-based Systems

• NPPs are among the most complex energy systems ever built.

• NPP functions and processes rely on a myriad of digital IT and I&C systems.

• NPPs capital cost and the radioactive nature of nuclear fuel demand computational tools for licensing, operation and accident analysis.

• NPPs are among the most emblematic examples of critical infrastructure cyber-targets.

2

The Challenge How can we improve cybersecurity capabilities, conduct IT and I&C research, increase awareness, and perform training and hands-on exercises in an integrated nuclear power plant environment?

3

In a NPP environment, how do we:

• assess the facility impact of a system being compromised?

• evaluate the effectiveness of segregating facility functions?

• assess computer security (CS) vulnerabilities in the systems that perform functions?

• evaluate the use of de-coupling mechanisms?

• test the effectiveness of firewall rules?

• and assess many other CS related issues?

4

IAEA CRP: Asherah NPP Simulator (ANS)

A hypothetical/neutral PWR named Asherah was defined based upon several NPP existing designs.

• The results were combined to produce a technological neutral facility.

• USP developed ANS model to be the heart of a cyber security assessment test bed.

• The simulator was designed specifically for the simulation of cyber-attacks.

5

IAEA CRP: Asherah NPP Simulator (ANS)

• ANS reproduces the Asherah NPP behavior using dynamic models.

• It is based on:

The TMI core.

Typical industry systems and equipment.

Standard control logic.

• It has been implemented using the Matlab/Simulink environment.

• It has the capability to interface with IT/OT equipment for cyber security assessment.

Local HMI & Simulation Control 6

IAEA CRP: Asherah NPP Simulator (ANS)

WATCH THE VIDEOS Local HMI & Simulation Control 7

ANS Interfaces

• ANS has been connected to physical and virtual controllers -

and other equipment.

• USP developed I/O interfaces for Modbus and OPC-UA & DA communication.

• USP has also developed a light OPC UA Server & client, i4BrSrv, for ANS communication.

8

ANS Deployment Modes Abstract &

Portable ANS (plant processes & controllers)

Standalone running in one VM Model-Based ANS plant processes & controllers running in many VMs ANS plant processes & controllers CLDT-Based running in a Closed-Loop Digital Twin (CLDT) test bed ANS plant processes & controllers HIL-Based running in a Hardware-In-the-Loop (HIL) test bed Concrete &

Complex 9

IAEA ICONS 2020 DEMO: HIL and Model-based Run Easy to run by any user Model & HIL based setups 4 Virtual machines per setup Easy to analyze the network 10

Closed-Loop Digital Twin (CLDT)

• A DT is a simulated/emulated device/system that replicates in detail their physical counterparts on the logic and network layer.

Virtual Product Virtual

• A DT may be leveraged for CS purposes in two ways: Production Simulation mode Real Product Replication mode Real Production CS approach

• CS can be introduced during the product design and production phases.

• CS can be seamlessly integrated in the entire digital-based systems lifecycle.

11

ANS PLC CLDT: Successful Attack Example

• Attack scenario where a PLC is compromised from outside the I&C controllers network.

• A PLC DT integrated with the ANS CLDT test bed allows for assessment of the network indicator of compromise and of the facility impact.

WATCH THE VIDEOS 12

ANS PLC CLDT: Unsuccessful Attack Example The ANS PLC CLDT Simulation mode test bed allowed for:

• Monitoring of NPP facility functions.

• Assessment of the effectiveness of a computer security strategy.

• Monitoring I/O tags at the PLC (process).

• Checking integrity and availability of PLC I/O tags and HMI tags (network). WATCH THE VIDEOS

• Introduction of CS from the design phase.

13

Final Remarks

• DTs create new possibilities for monitoring, simulating, estimating and assessing states of real systems.

• ANS was developed for an IAEA CRP (17 teams of 13 MS) for computer security research and it has been supporting graduate and postdoctoral studies.

• ANS has been integrated in test beds, applied in CS exercises and demonstration in Austria, Brazil, Canada, China, Germany, ROK and USA.

• DTs can be leveraged for computer security purposes when integrated with nuclear simulators like the ANS.

14

Thank you!

Advanced Modeling and Simulation and its Future Role in Nuclear Systems Digital Twin Technology Dave Kropaczek Oak Ridge National Laboratory Technical Session: Multiphysics Modeling Digital Twin Applications for Advanced Nuclear Technologies December 1-4, 2020 ORNL is managed by UT-Battelle, LLC for the US Department of Energy

Digital Twin - Role of Modeling and Simulation The digital twin is the virtual representation of a physical object or system across its life-cycle. It uses real-time data and other sources to enable learning, reasoning, and dynamically recalibrating for improved decision making.

• In Nuclear Systems, the digital twin may be characterized by:

- Virtual simulator for the plant systems and subsystems, including the reactor core and fuel

- Use of a wide-range of sensors (in-core/ex-core detectors, thermocouples, pressure, flow, etc.)

- Mapping of sensor data onto the virtual model through update of the simulator model parameters

- Recalibration of the virtual simulator based on real-time data

- Use of the virtual simulator to monitor operational limits (e.g. core, fuel)

- Use of the virtual simulator to make future projections regarding reactor behavior under what if scenarios

- Use of the virtual simulator as part of the reactor control system (human or autonomous)

By these definitions, the digital twin for nuclear systems has existed for decades in the form of on-line core monitoring systems. What has changed are the advances in modeling, sensors, calibration techniques, and predictive analytics to enable a step change in decision-making capabilities for reactor operation.

• https://www.ibm.com/blogs/internet-of-things/iot-cheat-sheet-digital-twin/

with credit to Josh Kaizer, NRC 2

Digital Twin - Virtual Simulator

• High fidelity predictive simulation for quantities of interest

• Safety parameters (temperatures, power deposition)

• Operational parameters (power response, energy output)

• Component behavior (lifetime analysis)

• Physics-based modeling for key phenomena

• First-principles based (elimination of correlations)

• Multi-physics response for coupled physics

• Includes neutronics, thermal-hydraulics, chemistry, and materials modeling

• High geometrical resolution

• Sufficient resolution to make use of real-time sensor data

• Modeling across length scales - atomistic to engineering scale

• Comprehensive, usable and extensible software system

• Verified software - code and solution verification

• Validated software - single and integral effects tests

• Quantified uncertainties for model parameters and input data 3

Multi-Physics Coupled Simulation Reactor Thermal Physics Hydraulics (n + ) Transport Mass, Momentum, and Energy Transport Nuclear XS Feedback Whole-core flow field On-the-fly XS Processing resolution Isotopic Transmutation Multi-phase flow Fuel Coolant Performance Chemistry Thermal Transport Crud source term Fuel Evolution Surface deposition (relocation, swelling) and growth Clad Evolution (creep) Equilibrium Mechanical Contact thermodynamics Crud distribution in a PWR Chemical Kinetics Physics Phenomena of Interest are Common to All Reactor Types 4

VERA - A Fully Integrated Capability for Reactor Analysis Virtual Environment for Reactor Applications VERA Neutronics

• High Resolution: MPACT Shift SCALE/

Mesh / Solution DAKOTA AMPX

- Fully coupled and pin-resolved neutronic, T/H, and Transfer ORIGEN MOOSE crud growth physics DTK Trilinos

- Detailed rod-wise fuel performance analyses libMesh Thermal-Hydraulics Star-CCM+

CTF PETSc System Codes

• Integrated Applications:

- Modeling in-core and ex-core detector prediction Solvers / UQ Chemistry of axial offset (AO) due to CRUD deposition MAMBA

- Identification of PCI failure risk during load follow VeraIn/Out VERAView Fuel Performance operation with accident tolerance fuel and Common I/O & Visualization BISON cladding

- Accumulation of radiation damage in the reactor vessel due to neutron fluence

- Prediction of cladding integrity during reactivity-initiated transient using coupled neutronics and T/H with offline fuels analysis

• Performance & Usability:

- User-friendly I/O (e.g. automated mesh generation and data transfers)

- Integrated visualization tools 5

VERA Key Physics Codes ORIGEN MPACT Shift Isotopic Reactor Monte Carlo depletion physics neutron &

and gamma decay transport CTF BISON MAMBA Fuel rod Thermal-hydraulics CRUD chemistry performance 6

MPACT

• Advanced 3D Neutronics

- Method-of-Characteristics

- 51 energy group nuclear data library

- Whole pin-wise resolution, including intra-pellet power and isotopic distributions

• Steady-state and transient capability

• Integrated explicit isotopic depletion and decay with ORIGEN

• 3D accuracy comparable to continuous-energy Monte Carlo methods, including Shift and MCNP

• Core shuffling and control rod movement

• In-core detector responses

• Validated against critical experiments and over 150 fuel cycles 7

Shift

• Accurate and efficient neutron and gamma transport

- Continuous-energy Monte Carlo neutron &

gamma transport to any region outside of the reactor core

- State-of-the-art hybrid methods focus particles toward the regions of interest

• General geometry capability for ex-vessel region

• MPACT provides accurate 3D fission source &

isotopics

• Enables best-estimate vessel fluence analysis and coupon irradiation

• Ex-core detector response calculations and weighting factor generation

• Coupling with materials models allows for calculation of concrete degradation and core structure embrittlement 8

BISON

• VERA can be used to perform detailed fuel rod performance analysis with BISON

- Finite element-based engineering scale fuel performance code

- Solves the fully-coupled thermo-mechanics and species diffusion equations in 1D symmetric, 1.5D, 2D axisymmetric or generalized plane strain, or 3D

• Lower length scale and mechanistic models for key physics phenomenon (e.g. fission gas release, thermal conductivity) applicable to existing and future ATF fuel forms and clad

• Fuel rod geometry and power histories used to automatically create BISON inputs for any or all fuel rods in a reactor core

• BISON results are collected into VERAOut format for whole-core fuel rod performance analysis or screening Ref. K. Gamble, G. Pastore, M. Cooper, D. Andersson, ATF material model development and validation for priority fuel concepts, CASL-U-2019-1870-000, July 2019.

9

CTF Sub-Channel

• Whole-Core Two-Phase Subchannel Thermal-Hydraulics Discretization

- Three-field representation of two-phase flow for the entire

- Continuous vapor (mass, momentum and energy) core

- Continuous liquid (mass, momentum and energy)

- Entrained liquid drops (mass and momentum)

- Non-condensable gas mixture (mass)

- Native, transient fuel temperature model

• Cross flow between channels

• Coupling with Systems Codes (TRACE, RELAP) via inlet and exit boundary conditions

• Spacer grid pressure loses and blockages and intra-grid form losses

• Use of higher resolution computational fluid dynamics (CFD) simulation to improve the subchannel modeling 10

Addressing Modeling Gaps

• Use of high resolution, high fidelity methods to improve lower resolution model for key System Response Quantities (SRQs)

- STAR-CCM+ informs CTF

- SRQs include azimuthal heat flux and TKE Ref. Salko, R., S. Slattery, T. Lange, M. Delchini, W. Gurecky, E. Tatli, and B. Collins, Development of Preliminary VERA-CS Crud-Induced Localized Corrosion Modeling Capability, CASL-U- 2018-1617-000, June 2018.

• Use of integral experiments and system level data to calibrate fundamental model parameters where Single Effects Tests (SETs) data does not exist

- Bayesian calibration allows for establishment of uncertainty bounds on calibrated parameters Ref. B. Kuwaileh and P. Turinsky, Data Assimilation and Uncertainty Quantification Using VERA-CS for a Core Wide LWR Problem with Depletion, CASL-U-2016-1054-000, April 2016.

11

Watts Bar Unit 2 Power Ascension First US reactor startup in over two decades modeled in near real time as a blind prediction Watts Bar Unit 2 Power Ascension 100%

- 4,130 hourly state-points 80%

- 13.5 days of runtime on 2,784 cores Core Power Level (%)

60%

- 892,837 core-hours 40%

- 16,605 fully-coupled neutronics/TH iterations 20%

Accurate comparison to measurement, including a 0%

new Vanadium-wire, in-core flux map system (+/- 2.4%)

Measured Power Distribution (M/P)

Pin-by-pin spatial detail of non-measurable (Full Power, 5.6 GWD/T Exposure) quantities of interest (e.g. Xe-135)

Watts Bar 2 predicted, transient Xenon-135 distribution at 28% power level

1. Ref. A. Godfrey, B. Collins, C. Gentry, S. Stimpson, J. Ritchie, Watts Bar Unit 2 Startup Results with VERA, CASL-U-2017-1306-000, March 2017.

12

VERA Simulation of Signal Response

• First-of-a-kind capability demonstrated for VERA-Shift applied to coupled in-core/ex-core calculations

• Addresses a concern over secondary source signal strength as seen by the source range detectors (SRD) during refueling

• In this application, virtual detector signals were generated for the WBN1 C8 Thermal Flux with 9 Assemblies Loaded for Southern refueling shuffle sequence with direct SRD comparison against measured count rates

• Excellent agreement between measured and predicted signal Ref. Godfrey, E. Davidson, G. Wolfram, B. Collins, C. Gentry, G. Ilas, S. Palmtag, T. Pandya, K. Royston, Watts Bar Unit 1 Source Range Detector Response Validation During Refueling, CASL-U-2018-1561-000, December 2018.

13

DOE-NE Advanced Modeling Simulation Light Water Reactors - Near-term focus

• Provide support for advanced LWR nuclear technologies and target areas for which current LWR modeling and simulation capabilities cannot be used Ejected

• Areas include: Control Rod

• Accident tolerant fuels

• High burnup, high enrichment fuel

• Materials fabrication and performance, including advanced manufacturing

• Two-phase fluid flow, including flow regime transitions

• Reactor operational performance

• Reactor safety performance AP1000 RIA fuel rod enthalpy and energy deposition evolution (ATF fuel form)

Ref. S. Ray, V. Kucukboyaci, Y. Sung, P. Kersting, R. Brewster, Industry Use of CASL Tools, 14 CASL-U-2019-1739-000, September 2018.

DOE-NE Advanced Modeling Simulation Advanced Non-Light Water Reactors - Near-term focus

• Target areas identified by industry, GAIN Technical Working Groups, and the US NRC to support their activities including molten salt, HTGR, and fast reactor technologies

• Support industry and the NRC for the rapid development and demonstration of microreactors in the 3-5 years time frame

• Areas include:

• Fuels

• Materials fabrication and performance, including advanced manufacturing

• Chemistry Turbulent Heat Flux - Nek5000/BISON

• Reactor systems 15

Summary

• The virtual reactor simulator is one aspect of the Digital Twin for Nuclear Systems

• High fidelity, high resolution virtual simulator technology has rapidly evolved to the level of high predictability for reactor quantities of interest based on coupled, multi-physics modeling

- First principles combined with multi-scale approach can capture the relevant physics phenomena

• Uncertainties in input parameters and closure relations may nevertheless be an issue for a particular reactor configuration (fuel form, coolant).

- Model gaps can be addressed through use of formal calibration methods

- Such methods benefit from availability of measured data required for calibration

• Integration of high fidelity, high resolution simulation with advanced sensors result in unprecedent detailed of reactor behavior 16

Modeling and Simulation to Support Digital Twins Dr. Jeffrey W. Lane Chief Engineer and Principal Consultant Zachry Nuclear Engineering lanejw@zachrynuclear.com 919-903-6763 Digital Twin Applications for Advanced Nuclear Technologies Online Workshop December 1-4, 2020

© 2020 Zachry Brands, Inc. All rights reserved.

1

INTRODUCTION Digital Twin is a virtual replica of a physical asset

• Can be a plant, system or specific component

• Enhanced understanding of physical asset integrating data + simulation

• Can use Machine Learning (ML) & Artificial Intelligence (AI) to identify causal relationships and produce reduced-order models (ROM)

• Predict performance and expected response of the asset

• Identify vulnerabilities Types of Digital Twins

• Design - identify issues before construction and optimize system

• Construction - support scheduling and evaluate as-built deviations

• Operations - monitor performance degradation and maintenance

• Others

© 2020 Zachry Brands, Inc. All rights reserved.

2

EXAMPLE - DESIGN DIGITAL TWIN Design by simulation

• Physics based modeling provides access to a wealth of data, including unmeasurable quantities

• Can be more cost effective than testing Attributes:

• Identify system faults before the system is built

• System optimization Example - Ventilation System

• Toxic gas and room habitability assessments

• Location and sizing of HVAC and filtration

• Complex geometry and recirculation patterns

• Simulation identified local pockets of higher concentration for original design

© 2020 Zachry Brands, Inc. All rights reserved.

3

EXAMPLE - OPERATIONS DIGITAL TWIN Attributes:

• Connected to the physical asset by continuously monitoring and collecting information

• Continuously learning and dynamically updating

• Simulation used to fill in knowledge gaps (e.g., non-existent data for fault scenarios)

Example - Vacuum transfer system

• Time critical transfer of fluid

• Performance degradation of seals and vacuum system

• Elongate time between maintenance and minimize downtime

• Pre-emptively schedule maintenance before failure

© 2020 Zachry Brands, Inc. All rights reserved.

4

REQUIREMENTS

• Modeling & Simulation (M&S) plays an important role in digital twins to:

- Fill in knowledge gaps or lack of available data

- Provide access to unmeasurable quantities

• However, the M&S results must be obtainable and meaningful

- Design by simulation requires VVUQ of M&S tool

- Will need to assimilate M&S results with information obtained directly from the asset (I&C signals) and resolve discrepancies

• The following must be considered with respect to applying M&S for digital twins

- Establish Applicability

- Data Assessment From - J. Rowley, The Wisdom Hierarchy:

Representations of the DIKW hierarchy, Journal of

- Software Requirements information Science, pp. 163-180, 2006

- Computational Requirements

© 2020 Zachry Brands, Inc. All rights reserved.

5

APPLICABILITY OF M&S TOOLS

• Just like any other application of M&S, one must establish the credibility and applicability of the evaluation model (code + inputs) for the intended application

- Provides confidence that the simulation includes the necessary physics and produces accurate results throughout the application domain

- Establish the uncertainty and trustworthiness of the results

• Established methods for assessment

- US NRC Code Scaling, Applicability and Uncertainty (CSAU) using Phenomena Identification and Ranking Table (PIRT)

- Evaluation Model Development and Assessment Process (EMDAP) from Reg. Guide 1.203

• Provide good frameworks for evaluating the adequacy and sufficiency of a result; however, in practice many applications of these methods tend to rely heavily on engineering judgement.

© 2020 Zachry Brands, Inc. All rights reserved.

6

APPLICABILITY OF M&S TOOLS

• Need quantitative approaches to assess

- Accuracy of results, including uncertainty and effects of scale

- Domain coverage - where are the holes?

- Adequacy - what level of agreement is sufficient?

• Quantitative approaches

- Reduce reliance on engineering judgement

- Rank and prioritize areas for improvement both in the simulation and experimental needs

• Example:

- Predictive Capability Maturity Quantification (PCMQ)

© 2020 Zachry Brands, Inc. All rights reserved.

7

DATA ASSESSMENT Model of EBR-II Simulation results from range of transients

• Must assess the quality and trustworthiness of M&S data prior to training ML algorithms.

- Establish confidence or identify unexpected results

• Automated tool to parse results

- Search against multiple criteria and types

- Limits (>,<), logical (AND/OR), inflection points, Unexpected behavior in etc. simulation results?

Not Desirable

- Reverse flow Inflection point #1

• Scanning tool and criteria developed to identify - No unusual inflections (expected) PT1 unexpected or anomalous behavior in 700 simulation results 600 Inflection point 500 #3 (unexpected)

- Reinforces that samples and training data 400 cannot be treated as a black-box 300 Inflection point 200 #4 100 (unexpected)

Unacceptable Inflection 0point #2 Acceptable

- Reverse flow (expected) 0 50 100 150 200 250

- Only forward flow

- Unusual 8 inflections

© 2020 Zachry Brands, Inc. All rights reserved.

- Pump stopped - No unusual inflections

DIGITAL TWINS AS SOFTWARE

• Digital twin functionality is similar to M&S

- Stores prototypical conditions, physics, and closures to make inferences and predictions for real systems

- But, generally able to provide results much faster than M&S

• Digital twins include different types of software

- Computational Engine or simulator

- Training of ML/AI algorithms

- Digital twin itself

• ML/AI are new paradigms relative to existing nuclear SQA standards (e.g., 10 CFR 50 Appendix B, ASME NQA-1)

- Not static and continuously learning, but must be able to verify results

- Must provide transparency & traceability to gain confidence in these technologies

• Depending on functionality or role of the digital twin, may also need to consider software reliability, hazard analysis and cybersecurity.

© 2020 Zachry Brands, Inc. All rights reserved.

9

COMPUTATIONAL REQUIREMENTS

• Several different considerations for computational performance

- Effectiveness of process for generating training data

• Adaptive sampling and coverage assessment

• Assisted using other available knowledge bases

- Digital Twin Training Process

• Balance accuracy with potential for overfitting

• Hyperparameters represent an additional sensitivity/uncertainty

- Execution time for Digital Twin

• Depends on the time scale for the event, but initial response must be real-time

• Potential for recommendations to change during processing time

• Need a general purpose, validated and robust simulation engine to generate training data

- Can involve O(104-106) or more simulations, so even a small fraction of simulations that fail to run to completion can be problematic.

- Requires a 3-D, coarse-grid CFD code that can model all facets of the plant (reactor vessel, piping systems and containment) using a variable mesh and is applicable to both LWR and non-LWRs

- GOTHIC is an industry trusted multi-physics, multi-scale M&S tool that supports digital twin development

© 2020 Zachry Brands, Inc. All rights reserved.

10

CONCLUSIONS Digital twin solutions support decision Cost Savings making and provide a variety of benefits.

Increased Safety Modeling & simulation plays an Equipment Reliability & Loss important role in digital twins Avoidance

• Must establish the credibility of M&S results Operations Flexibility as it directly impacts the credibility of digital twins Reduced Reactive Maintenance

• Therefore, this is a critical element to the adoption, and regulatory approval, of ML based technologies for nuclear applications Higher Efficiency Optimized Design/Construction

© 2020 Zachry Brands, Inc. All rights reserved.

11

WORKSHOP ON DIGITAL TWIN APPLICATIONS FOR ADVANCED NUCLEAR TECHNOLOGIES December 1-4, 2020 MULTI-PHYSICS MODELING FOR ADVANCED REACTOR SAFETY erhtjhtyhy RUI HU Nuclear Science and Engineering Division Argonne National Laboratory

Safety Characteristics of Advanced Reactors Pursuing high levels of inherent (walk-away) safety Inherent reactivity feedback Multi-physics calculation for Passive decay heat removal unprotected transients?

Ultimate heat sink (ambient air) Accurate modeling of in-vessel heat transport (from the core to vessel Advanced fuel wall)

- TRISO, metallic, liquid Detailed simulation vs. lumped parameter approach SMR and Micro-Reactor Integrated modeling of reactor

- Small nuclear fuel inventories cavity cooling system or vessel

- Large surface to volume ratio cooling system 2

Needs for Multi-scale Multi-physics Capability (1)

Analysis of the transient behavior of a nuclear reactor requires coupled simulation of reactor kinetics and thermal-hydraulics of the reactor core In advanced nuclear reactors, e.g. Sodium-cooled fast reactor, the reactivity feedback due to core radial and axial thermal expansion are important The coupled simulation of thermal-hydraulics and thermal-mechanics is important for the multi-physics simulations of advance reactors for accurate prediction of thermal reactivity feedbacks 3

Needs for Multi-scale Multi-physics Capability (2)

Decay heat removal

- Most advanced reactor designs rely on passive safety system, such as RCCS

- Decay heat must be conducted from core to surface: fuels/structures are strongly thermally-coupled, and requires multi-dimensional modeling and simulation capabilities PB HTGR MHTGR Heat-pipe Reactor RCCS 4

Sources: High Temperature Gas-cooled Reactor Technology Training Curriculum; ANL/NSE-19/25; UCBTH-14-009

MULTI-PHYSICS SIMULATION OF HEAT PIPE MICRO-REACTOR

COUPLED CODE SIMULATIONS Joint Argonne-INL-NRC efforts using BlueCRAB Coupled codes in reference heat pipe microreactor model

- Reactor Kinetics (MAMMOTH/Rattlesnake)

- Thermomechanics (MOOSE Tensor Mechanics)

- 3D Heat Transfer (SAM)

- Heat Pipe Heat Exchanger (SAM)

- Reactor Cavity Cooling System (SAM)

MOOSE: multi-physics framework MAMMOTH: INL neutronics code SAM: ANL system code 6

SAM MODELS Reactor core Heat pipe heat exchanger RCCS 7

MAMMOTH REACTOR PHYSICS MODEL Beginning-of-life (BOL) conditions Multi-group diffusion solver with MAMMOTH/Rattlesnake Correction with the super-homogenization (SPH) equivalence scheme Cross-section preparation with SERPENT Monte Carlo code Reactivity feedback effects

- Doppler effect: fuel temperature

- Radial expansion: radial core mesh deformation

- Axial expansion: fuel axial mesh deformation 8

STEADY STATE The model works very well for the steady Parameters Value state operation analysis Eigenvalue 0.99990492 Total power 5.0 MW Average fuel temperature keeps very well the Power to heat pipes 4.8942 MW Power to RCCS 0.05291 MW symmetry of the reactor core Average fuel temperature 914.7 K Average hex can temperature 912.8 K Heat pipe near the center removes roughly Average bottom/top reflector temperature 866.9 K 1.5 times heat compared with heat pipe near Average side reflector temperature 765.6 K the periphery of the core (average 26 kW) Average plate temperature 803.6 K Average vessel wall temperature 674.5 K Reactor core temperature Average fuel temperature Heat pipe heat removal rate 9

LOSS OF HEAT SINK Heat pipe heat removal rate drops quickly to a lower level

- Flow rate drops to 0.1% of steady-state value

- Slow decrease due to the thermal inertial of the heat pipes Reactor power drops quickly due to the strong negative reactivity Average solid temperature feedback Decay power was not considered yet in the reactor physics model 10 Fuel average temperature

MULTI-PHYSICS MODELING FOR DIGITAL TWIN DEVELOPMENT

SCALABLE DIGITAL TWIN IN SAFARI PROJECT Physics-based to ensure robustness over the entire range of operations and data-enabled to enhance predictive capabilities.

SAFARI: Secure Automation for Advanced Reactor Innovation, ARPA-E GEMINA Award (Courtesy of A. Manera, UM) 12

MULTI-PHYSICS MODELING FOR DIGITAL TWIN DEVELOPMENT Multi-physics simulations including plant control and protection systems To build the ML-augmented, physics-based reduced order models of the FHR To demonstrate the accuracy of the digital twin and the commercial benefit

+ Control systems

+ Component Models https: //kairospower.com 13

MULTI-SCALE MULTI-PHYSICS MODELING CAPABILITY NEEDED FOR ADVANCED REACTOR SAFETY AND SCALABLE DIGITAL TWIN DEVELOPMENT

Hybrid Physics-Informed Neural Networks, Cumulative Damage Models, and Digital Twins Felipe A. C. Viana, PhD Probabilistic Mechanics Laboratory PML Principal Investigator pml-ucf.github.io Assistant Professor Mechanical and Aerospace Engineering E-mail: viana@ucf.edu University of Central Florida

Prognosis and digital twins Maintenance costs (a) Onshore wind turbine example

• Production lost

• Component

• Equipment rent, labor, etc.

Prognosis and digital twin challenges:

• Physics not fully understood Gearbox-LSS Repair/replacement cost

• Data is highly unstructured $300K Main bearing

• Operation/controls vastly available (?) $150-$300K

• Poor inspection and failure data Gen. bearing Gearbox-HSS

1. Digital twins must bridge the gap between $10-$20K $25-90K model predictions (understanding) and Frequency of occurrence observations (reality)

Sethuraman, L., Guo, Y., & Sheng, S. (2015). Main bearing dynamics in three-

2. Hybrid models can be really helpful point suspension drivetrains for wind turbines. American Wind Energy Association Conference & Exhibition, May 18-21, Orlando, FL.

2

Physics-informed neural networks are not new 2018 Advances in Neural Information Processing Systems (NeurIPS 2018 - Best paper)

How can we leverage this concept to build digital twins?

3

Cumulative damage models and uncertainty quantification (a) Fatigue crack growth at fuselage panel Fatigue crack growth

=

where:

• : number of cycles

• and : material properties (coupon tests)

• = (b) Finite element modeling

• : engineering analysis (e.g., FEM)

What if or are not accurate?

We propose using hybrid models for uncertainty quantification 4

Physics-informed neural networks are perfect for prognosis digital twin Use case: (a) Typical training (b) Typical prediction

• Very few output observations

• Inputs observed throughout

• Sequences are VERY long

• Cell models transition never observed years years Cell Cell Cell Cell Cell Cell If output is observed throughout, data-driven recurrent neural networks (LSTM, GRU, etc.) might be useful, otherwise years years Very hard (impossible) without physics Blue: observed data Gray: desired output (never fully observed)

R. G. Nascimento and F. A. C. Viana, Cumulative damage modeling with recurrent neural Orange: Recurrent neural network prediction networks, AIAA Journal, Online First, 13 pages, 2020, DOI: 10.2514/1.J059250.

5

Cumulative damage model with recurrent neural networks (a) Long short-term memory (LSTM) cell (b) Euler integrator cell (cumulative damage)

, 1

R. G. Nascimento, K. Fricke, and F. A. C. Viana, A tutorial on solving ordinary differential

• RNNs are perfect fit for damage accumulation, equations using Python and hybrid physics-informed neural network, Engineering

• ( , 1 ) can be customized. Applications of Artificial Intelligence, Vol. 96, 2020, 103996, DOI:

10.1016/j.engappai.2020.103996.

6

Wind turbine main bearing fatigue (a) SCADA data Model-form uncertainty:

• Bearing fatigue: relatively well-understood

• Grease degradation: difficult to model with physics Damage inspection:

• Bearing: not always measurable

• Grease:

• Laboratory: accurate but expensive

• Visual: large uncertainty but affordable (b) Visual grease inspection ranking (high variability)

Unbalanced data: Example of ranking

• Supervisory control and data acquisition (SCADA) 1 2 3 system (per 10 mins)

• Inspection depends on operator inspection policy 4 5 Y. A. Yucesan and F. A. C. Viana, A physics-informed neural network for wind turbine main bearing fatigue, International Journal of Prognostics and Health Management, Vol. 11 (1), 2020.

Y. A. Yucesan and F. A. C. Viana, Hybrid physics-informed neural networks for main bearing fatigue prognosis with visual grease inspection, Computers in Industry, Accepted.

7

Hybrid physics-informed neural network (a) Hybrid model (b) Turbine-level service optimization 10 turbines used for training Regreasing optimization @ 120 turbines

• Wind speed and bearing temperature

• Visual grease inspection ranking

• Wind speed and bearing temperature every month Data-driven layers Physics layers We optimized service intervals on a turbine-by-turbine basis 8

Lithium-ion battery aging modeling (a) Example of random loading conditions Key technology for electric vehicles (b) Aging can cause models to diverge from observations Challenges:

• Prognosis models depend on a number of empirically adjusted factors

• Hard to account for aging R. G. Nascimento, M. Corbetta, C. S. Kulkarni, and F. A. C. Viana, Hybrid Physics-Informed Neural Networks for Lithium-Ion Battery Modeling and Prognosis, Applied Energy, submitted.

9

Hybrid physics-informed neural network (a) Hybrid model (b) Aging model 8 batteries used for training

• Current and voltage time histories

• Internal voltage adjusted with constant discharge

• Battery aging is a probabilistic model adjusted using hundreds of hours worth of data (c) Probabilistic forecast data 10

Model-form uncertainty in corrosion fatigue (b) Fleet prediction at the end of 5th year.

Challenge

• Assumed: pure mechanical fatigue

• After 5 years: corrosion-fatigue Data

• Load history of 5 years: 150 aircraft

• Crack length: 15 aircraft at end of 5th year.

Damage accumulation grossly underestimated!!!

(a) Hybrid physics-informed neural network cell (c) Probability of failure forecast A. Dourado and F. A. C. Viana, Physics-informed neural networks for missing physics estimation in cumulative damage models: a case study in corrosion fatigue, ASME Journal of Computing and Information Science in Engineering, Vol. 20 (6), 10 pages, 2020.

11

Probabilistic Mechanics Laboratory Credit really goes to my PhD students Publications:

pml-ucf.github.io/publications Andre Von Zuben Arinan Dourado Kajetan Fricke Physics-informed neural networks package github.com/PML-UCF/pinn Ordinary differential equation solver:

https://github.com/PML-UCF/pinn_ode_tutorial Renato Nascimento Yigit Yucesan Wind turbine main bearing fatigue github.com/PML-UCF/pinn_wind_bearing Sponsors and Collaborators Corrosion-fatigue prognosis github.com/PML-UCF/pinn_corrosion_fatigue 12

A Quantitative Framework to Assess Tradeoffs in Alternative Models and Algorithms for Prognostics and Health Management Saikath Bhattacharya and Lance Fiondella 1

Introduction

• Prognostics and health management

- Modernizing system reliability engineering with sensing, models, and algorithms to accurately estimate remaining useful life

- Promotes nonfunctional RAM+C (reliability, availability, maintainability, and cost) requirements 2

Motivation

• Previous studies

- Emphasize development of

• Degradation models

• Algorithms to estimate model parameters

- Typically

• Restricted to single maintenance cycle and focused on enhancing prediction

• Do not assess long term performance of competing methods 3

Limitations of Academic Modeling Studies Number of cycles used to fit models (500)

Often hand-picked to make a proposed model appear favorable 4

Motivation (2)

• Fewer studies

- Assess impact of PHM decisions on cost and other derived reliability measures

- Restricted to simulation and analytical techniques (not data-driven) 5

Proposed Approach

• Objective framework to assess

- Performance decisions made by alternative combinations of models and algorithms

- Adapts analytical methods from maintenance theory to data-driven approach

• Average cost per unit time

• Utilization

• Safety

• Availability 6

Capacity (Battery) Degradation Models Some parametric models

• Polynomial model

- = 1 2 + 2 + 3 (1)

• Exponential model

- = 1 2 + 3 4 (2)

• Hybrid model

- = 1 2 + 3 2 + 4 (3) 7

Filtering for Battery Degradation Models

• Unscented Kalman filter

- Recursively updates degradation model parameters ()

based on capacity in past and present cycles ( ) to estimate RUL

• Particle Filtering

- Based on Bayesian Monte Carlo simulation with importance sampling to update parameters 8

Preventive Maintenance

• Based on present model parameter estimates

• Recommends maintenance

- If remaining useful life (RUL) prediction less than prognostic distance

• Continues operation otherwise 9

Reliability, Availability, and Maintainability Measures

• Given unit lifetime and maintenance interval ,

inter-renewal time = min(, ) such that

= 1 ()

0

- = 1 () - Unit reliability (complement of CDF) 10

Age Replacement Maintenance Model

• Average cost per unit time

+ 1

=

0 1

- () - Probability of failure before maintenance

- - Cost of emergency repair

- - Cost of preventive maintenance 11

Age Replacement Maintenance Model (2)

• Average cost per cycle

=1 + 1

=

=1{ + [1 ( )]}

- - Prognostic distance

- - Number of units

- - Indicator function of unit

- - Cycle at which preventive maintenance performed on unit with prognostic distance 12

ILLUSTRATIONS 13

Data and Methodology

• Utilized Li-ion battery data set ( = 4)

- Performed least squares estimation on battery exhibiting most cycles prior to failure and used as initial estimates for UKF and PF (also considered battery with fewest cycles) 14

Data and Methodology (2)

• Ratio of emergency and preventive repair costs

= 1,000

• Mean times to repair

= 3 and = 8 15

Point Example: Equation (2) under UKF with = 150 Measure CX2-34 CX2-36 CX2-38 True EUL (cycles) 505 560 524 Maintenance 505 527 511 Predicted EUL (cycles) 679 677 661 Unused life (cycles) 0 33 13 Cost () 10,000 10 10 Safety () 0 1 1 Time to repair (cycles) 8 3 3 10,020 1,543 2 150 = = 6.494, 150 = = 97.11%, 150 = , 150 = 99.1%

1,543 1,589 3 16

Average Cost per Cycle (UKF)

Prognostic distance 175,678 minimizes cost 17

Utilization (UKF)

Utilization decreases monotonically as larger prognostic distance initiates earlier maintenance 18

Safety (UKF)

Safety and average cost per cycle exhibit inverse trends 19

Availability (UKF)

Low utilization corresponds to low availability 20

Example tradeoff (UKF)

Safety and utilization competing constraints 21

Summary and Conclusions

• Proposed framework to assess

- Quantitative performance of PHM decisions made by alternative combinations of models and algorithms

• Developed RAM+C measures for PHM

- Average cost per unit time, utilization, safety, and availability 22

Summary and Conclusions (2)

• Applied to combinations of three degradation models and two filtering methods with Li-ion data set

• Proposed approach

- Offers method to select prognostic distance to balance stakeholder needs

- Can be applied to other domains, degradation models (physics of failure), and algorithms (deep learning) 23

Future work

• Open source framework

- Crowdsource contribution of

• Models

• Algorithm

• Datasets/Challenges

- Raise academic standards for comparison

- Promote collaboration between academic, industry, and government stakeholders 24

• Formulation of additional quantitative measures

• Performance of particle filtering on quantitative measures

• Comparison of quantitative measures in window minimizing cost BACK UP SLIDES 25

Additional Measures

• Utilization

=1 + 1

=

=1

- - End of useful life of unit

• Can take values in interval 0,1

• Poses competing objective with cost and safety 26

Additional Measures (2)

• Safety

=1

=

- Fraction of units that undergo preventive maintenance

• Minimizing cost corresponds to maximizing safety 27

Additional Measures (3)

• Availability

=

+

• Mean time to failure (MTTF) 1

= + (1 )

=1 28

Additional Measures (4)

• Mean time to repair (MTTR) 1

= +

- - Number of units subject to ,

- - Mean time to repair given that unit underwent (, )

29

Average Cost per Cycle (PF)

Prognostic distance 202,288 minimizes cost 30

Utilization (PF)

Prognostic distances that produce low utilization correspond to low cost 31

Safety (PF)

Safety and average cost per cycle exhibit inverse trends 32

Availability (PF)

Low utilization corresponds to low availability 33

Measures in Prognostic Window with Cx2-37 (most cycles to failure)

Unscented Particle Filter Kalman Filter Left Mid Right Left Mid Right Eq (2) 175 412 660 142 215 288

() 0.020 0.039 0.810 0.025 0.031 0.071

() 0.939 0.482 0.023 0.734 0.596 0.262

() 0.994 0.988 0.804 0.992 0.990 0.978 Eq (3) 193 435 678 202 267 332

() 0.020 0.039 0.810 0.025 0.030 0.037

() 0.939 0.482 0.023 0.751 0.628 0.500

() 0.994 0.988 0.804 0.992 0.991 0.988 Conservative strategy selects at midpoint 34

Measures in Prognostic Window with Cx2-34 (most cycles to failure)

Unscented Particle Filter Kalman Filter Left Mid Right Left Mid Right Eq (1) 101 343 586 330 361 392

() 0.019 0.035 0.267 0.035 0.039 0.044

() 0.937 0.503 0.067 0.512 0.456 0.4013

() 0.994 0.989 0.925 0.989 0.988 0.986 Eq (3) 136 374 613 160 261 362

() 0.020 0.039 0.576 0.022 0.028 0.044

() 0.887 0.459 0.031 0.815 0.634 0.401

() 0.994 0.988 0.852 0.993 0.991 0.986 Equation (3) with UKF stable in both scenarios 35

Digital Twins in a Nearly Autonomous Management and Control System for Advanced Reactors Nam Dinh, Linyu Lin Department of Nuclear Engineering North Carolina State University 12/08/2020 1

Nearly Autonomous Management and Control (NAMAC)

• A comprehensive control system to assist plant operations

• Knowledge integration

• Scenario-based model of plant (systems, success paths)

• plant operating procedures, tech. specs., etc.

• Real-time measurements

• Digital twin technology

• Power of AI/ML

• NAMAC

• Diagnoses the plant state

• Searches for all available mitigation strategies

• Projects the effects of actions and uncertainties into the future behavior

• Determines the best strategy considering plant safety, performance, and cost.

2

Guiding Principles and Development Philosophy

• High-level requirements

• Technology neutral

• Accurate representation (twin) of the plant

• Dynamic and real-time: diagnosis, prognosis, and evaluation during operations

• Adaptive (or continuously learning)

• Explainable: outputs are traceable and justifiable

• Design principles for an intelligent autonomous control system

• Three-Level Architecture

• Knowledge Base

• Digital Twin

• Digital Twin Development and Assessment Process

• Trustworthiness Assessment 3

Three-Layer Architecture

• Individual Digital Twins (DT) are assembled into a DT-Hub to support decisions in operation, maintenance, safety management, etc. in the Operational Layer

• Each Digital Twin (DT) is a knowledge acquisition system to support specific functions

• Digital Twin for Diagnosis (DT-D)

• Digital Twin for Strategy Inventory (DT-SI)

• Digital Twin for Prognosis (DT-P)

• Digital Twin for Strategy Assessment (DT-SA)

• Developmental Layer extracts useful information from the knowledge base and creates Digital Twins (DT)

• Knowledge base stores data from simulations, operations, documents, procedures, etc.

4

Knowledge Base

• Knowledge base is the foundation of DTs and NAMAC M&S Set

• Integrate knowledge from a variety of sources I&C Points

• Plant monitoring systems Scenario based modeling & simulation (M&S)

Operating limits and control procedures Probabilistic assessment of the risk DT

• Emergency Operating Procedures (EOP) Tech.

EOP Specs

• Knowledge base will transit from simulation-based data to assimilating sensor data as a new plant comes on-line and operating history becomes available PRA

• M&S will always be a key contributor to the knowledge base, particularly for accidents and other low frequency events where actual plant data may not be available.

• Not just raw data signals, but these sources are vital knowledge bases

• Leverage existing information

• Minimize propensity to treat ML and DTs as black-box 5

Database Generation in Knowledge Base

• NAMAC Database generation:

• Training databases are generated by NAMAC use cases Mathematical sampling scenarios to populate and issue spaces representation information in the application domain RAVEN framework RAVEN

• The Digital Twin are constructed EBR-II model input according to the databases for supporting diagnosis, prognosis, NAMAC etc.

Digital Twin Knowledge base GCF GOTHIC framework GOTHIC nominal- Restart GOTHIC CSV state calculation File output 6

Digital Twin Definitions for DTs [1]

• Digital Twin technology - construct a digital replica (twin) for the real MODEL INTERFACE reactors and transients

• Data-driven model

• API

• Mechanistic model

• I/O

• Reasoning-based model

• User Interface

• DTs must provide insights equivalent to Modeling and Simulation (M&S), but need to learn and provide those insights much faster than Digital Twin Digital Twin the development and uses of M&S Prototype (DTP) Instances (DTI)

• But DTs are tightly coupled with operation

• Assimilating and adapting to real-time information from the operating environment Digital Twin

• Interacting with user for specific objectives Environment (DTE)

FUNCTION

• Use cases

• Objectives

• Output types 7

[1] F. Kahlen, et al., Transdisciplinary perspectives on complex systems - new findings and approaches, Springer, 2017

Digital Twin Training and Algorithms

• Artificial Neural Network (ANNs) is currently the major technology in constructing Digital Twins and NAMAC system.

• As complexity of NAMAC case studies increases, advanced algorithms are required to support DTs

• Modular framework allows for multi-tiered implementation

• Do not need a single, monolithic solution to cover all conditions

• Two classes of advanced algorithms are being investigated:

1. Knowledge/reasoning-based methods

• Provide explainability and transparency

2. Model free methods

• Deep learning capability that is needed for diversity and complexity

• Need both types for NAMAC 8

Digital Twin Training and Algorithms

• Advanced Algorithms

• Answer Set Programming (ASP) is a form of declarative programming oriented towards difficult search problems

• Discrepancy Checker (DC)

• Ensemble modeling employs a voting technique to aggregate/select predictions from a set of base models

• Digital twin for diagnosis (DT-D)

• Reinforcement Learning (RL) interacts with the environment and is time aware

• Wholistic NAMAC for furnishing recommendations

• Adaptive sampling techniques for data generation

• Efficient process to support Strategy Inventory (DT-SI)

• Meta-Learning to accelerate and optimize development 9

Modular Framework Function Modeling Recover full reactor states by assimilating plant sensor Neural nets (feedforward & recurrent);

Diagnosis data with the knowledge base Logic programming (Answer Set Programming)

Uniform sampling Strategy Inventory Find all available control/mitigation strategies Reinforcement Learning Predict the transients of state variables over a time Prognosis Neural nets (feedforward & recurrent) range Rank possible mitigations strategies and make Safety margin/limiting surface; Strategy Assessment recommendations considering preference structure Expected utility; Detect unexpected transient during operations Distance metrics; Discrepancy Checker considering DT trustworthiness for current conditions Logic programming (Answer Set Programming) 10

• NAMAC Operational Workflow NAMAC @ recommendation NAMAC @ discrepancy Plant Simulator Sensory data time checkingpoint 1 , 2 ,

Observable state Unobservable safety Multi-Physics and DT for Diagnosis significant factors System Simulation Unobservable safety

• Thermal Hydraulics significant factors

• Neutronics

• Structural Complete reactor information Distance

+ discrepancy DT for Strategy +

Initial - DT for Prognosis -

Condition Inventory All available Predicted reactor Constraints on control actions state for all options control actions of control actions Safety Constraints on Constraints on 1 > limit Challenges state variables Constraints reactor states Yes Instrument/Control Recommended DT for Strategy Predicted maximum

• I/C Layout Operator Actions Assessment temperature SCRAM

• Operator Model Command 11

The Development and Assessment Process (DAP)

• Instead of claiming to have a perfect autonomous system for a specific reactor during a specific scenario, our objective is to have a smart Development and Assessment Process (DAP) that produces NAMAC systems for generic types of reactors based on requirements from all stakeholders.

1924 - Ford assembly line 1965 - Ford assembly line 2019 - Tesla smart factory Evolution of Development and Assessment Process (DAP) for Automobile

[1] [2] Picture by Ford, The evolution of assembly lines: A brief history, https://robohub.org/the-evolution-of-assembly-lines-a-brief-history/, 2014

[3] Picture from Popular Mechanics, https://ottomotors.com/blog/what-is-the-smart-factory-manufacturing, 2019 12

Digital Twin Development and Assessment Process

• DT-DAP for a scalable and robust application of digital twins and NAMAC concept to generic types of use cases and advanced reactors.

• The DAP is conducted iteratively to deliver a reliable NAMAC with a set of credible DTs Element 1: Refined requirements Element 2: More complex and realistic knowledge base Element 3: Different machine-learning algorithms Element 4: ML uncertainty quantification, software reliability analysis Element 5: Digital twin trustworthiness assessment Adopted from U.S. NRC RG 1.203 Transient and Accident Analysis Methods 13

Digital Twin Trustworthiness

• In fundamental, the NAMAC make recommendations by extracting A gap between knowledge from knowledge base and assimilating them with real-the development & assessment of a digital twin time sensor signals - Digital Twin

• Considering the complexity and heterogeneity of knowledge base, we and investigate data-driven models and machine-learning algorithms for the use & regulation of a digital twin

• However, for complex systems and difficult tasks, the uncertainty of the DTs in NAMAC, if being overlooked, could introduce additional risks and degrade the trustworthiness of NAMAC recommendations, especially when the DT itself is complicated and black-box

• As a result, we need a trustworthiness assessment framework for DTs in NAMAC (ongoing)

(1) monitor uncertainty that could complicate the determination of mitigation strategies (2) make uses of information from the DT development and assessment process (3) do this in real time 14

Summary

• Implementation of digital twins for extracting and assimilating the knowledge base with real-time information

• Proof-of-concept of NAMAC for one class of transients

• Pump malfunction ranging from flow anomaly to complete loss of flow accident

• NAMAC provides recommendations during the event consistent to human operator norm

• The design of a digital twin development and assessment process (DT-DAP) for implementing, improving, and collecting evidence of a generic types of digital twins in autonomous systems

• DT-DAP at scoping stage that is driven by user experiences and sensitivity analysis

• Informed by EMDAP, but necessarily seeks to provide quantitative basis to support NAMAC decision making

• Next steps - the trustworthiness and robustness of DTs based on both intrinsic (i.e., uncertainty quantification, reliability) and contextual properties (i.e.,

confidence, safety-related vs. non-safety related).

15

Acknowledgement

• This work is performed with the support of ARPA-E MEITNER program under the project entitled: Development of a Nearly Autonomous Management and Control System for Advanced Reactors

• The NAMAC project team

• Nam Dinh (PI), Abhinav Gupta, Maria Avramova, Min Chi (NCSU)

• Linyu Lin, Pascal Rouxelin, Paridhi Athe, Joomyung Lee, Anil Gurgen, Edward Chen, Longcong Wang, Harleen Sandhu, Yeojin Kim (NCSU)

• Son Tran, Botros Hanna (NMSU)

• Carol Smidts, Xiaoxu Diao, Yunfei Zhao, Boyuan Li (OSU)

• Robert Youngblood, Cristian Rabiti (INL)

• David Pointer, Sacit Cetiner , Birdy Phathanapirom (ORNL)

• Jeff Lane, John Link (Zachry)

• Olu Omotowa, Eric Williams (TerraPower)

• Richard Vilim (ANL), Andrea Alfonsi (INL), Askin Yigitoglu (ORNL) [Resource Team]

• GOTHIC license is provided by Zachry Nuclear Engineering, Inc. GOTHIC incorporates technology developed for the electric power industry under the sponsorship of EPRI, the Electric Power Research Institute.

16

Structural Condition Monitoring with a Digital Twin:

Explorations on a Nuclear Containment Vessel Model Presenter: Dr. Timothy Kernicky, EPIC Research Assistant Professor of Civil Engineering, University of North Carolina Charlotte Contributors: Dr. Matthew Whelan

Digital Twin

• A digital twin is more than a digital model that faithfully represents physical assets and processes

• The primary distinguishing feature of a digital twin is its connection to the real-world asset with the ability to inform the state of the physical asset Dynamic/operational data Environmental exposure Acceleration Physical asset Settlement Digital twin Strain Etc.

Data cleansing Data analysis Uncertainty quantification Behavior modeling Visualization Simulation Feedback Lifecycle assessment Process optimization Decision making Forecasting

First Step: Structural Identification

• The first step towards successful digital twin deployment is the development of a trusted model, which faithfully replicates the performance of the physical asset

• This simple study leverages vibration-based structural identification to calibrate a set of uncertain material parameters of the digital twin using synthetic measurement data

Vibration-Based Structural Identification Physical Asset

• Finite element (FE) model of physical asset created

• Initial model suffers from

• Parameter uncertainties

• Geometries

• Material properties

• Idealization errors Digital Twin

• Discretization errors

• FE model may be leveraged to develop appropriate sensor array for physical structure

Vibration-Based Structural Identification Physical Asset Vibration Testing

• Vibration data is acquired from sensor array on the physical structure

• Structure excitation can be ambient (operational) or forced Digital Twin

Vibration-Based Structural Identification Physical Asset Vibration Testing Modal Properties Estimates Digital Twin Analytical Modal Analysis

• Modal properties of the physical asset are identified

• Modal properties of the digital twin are extracted from the system matrices, [K] & [M]

Vibration-Based Structural Identification Physical Asset Vibration Testing Modal Properties Estimates Digital Twin Analytical Modal Analysis

• Correlation between Model Correlation the physical and FE modal properties is determined by a modal measure of fit, which accepts natural frequencies and mode shapes as input.

Vibration-Based Structural Identification Physical Asset Vibration Testing Modal Properties Estimates

• FE model parameters are adjusted if the correlation between the physical and FE modal parameters is Digital Twin poor Analytical Modal Analysis Model Correlation Acceptable?

No Update uncertain parameters in [K] & [M]

via optimization scheme

Vibration-Based Structural Identification Physical Asset Vibration Testing Modal Properties Estimates Performance and Lifecycle Assessment Digital Twin Analytical Modal Analysis Yes Model Correlation Acceptable?

No Update uncertain parameters in [K] & [M]

via optimization scheme

Digital Twin Study of a Containment Vessel This study will demonstrate a potential capability to track changes/deterioration in a concrete containment vessel The structure explored is based on the 1:4 scale model of the Ohi-3 containment vessel in Japan, which was funded by NUPEC and the NRC and tested by SNL [NUREG/CR-6810, SAND2003-0840P]

Two FE models will be utilized in this study.

One represents the physical asset from which in-service, synthetic measurements are obtained The second will serve as the digital twin to be updated Measurements of dynamic properties (modal parameters) will be used by the digital twin to inform [NUREG/CR-6810, SAND2003-0840P]

changes in the structural condition, while synthetic response measurements obtained from the physical structure will be used to correct the digital twin

Concrete Containment Model A simplified finite element model was created using ABAQUS Four locations of interest where penetrations exist in the vessel were chosen as uncertain parameters to identify main steam penetration (M/S) feed water penetration (F/W) E/H A/L equipment hatch (E/H) air lock (A/L)

The modulus of elasticity of each section of elements was used as the uncertain parameter to be updated Four deterioration scenarios were examined to demonstrate the ability of the methodology to identify material degradation M/S F/W

Development of Synthetic Dataset Synthetic measurement data was extracted from the finite element model in the form of natural frequencies and mode shapes from 42 biaxial sensors Noise was added to the synthetic measurements by adding 0.5% Gaussian noise to generate 10 sets of data Synthetic natural frequencies and mode shapes for Case 1

Bayesian Model Updating

• Probabilistic updating was utilized as the model updating method, which accounts for measurement and modeling uncertainties

• Each uncertain parameter was assigned lower and upper bounds to which an adaptive Markov Chain Monte Carlo sampling method was used to generate 5000 posterior probability distributions

Bayesian Model Updating

• The discrete distributions of the points samples clearly indicate a successful identification of deterioration in the modulus of elasticity of the M/S elements, with negligible changes identified in the other parameters

Bayesian Model Updating

• Posterior probability density functions may be analyzed from which confidence bounds may be placed on the parameter identification

Trusted ModelNow What?

Once a faithful digital representation of a physical structure has been realized:

Performance of the structure may be monitored by using historical and present-day data streams Critical limit states may be evaluated in a digital environment Lifecycle analyses may be performed to inform maintenance outside of routinely scheduled programs t0 t1 t2 t3

Trusted ModelNow What?

Once a faithful digital representation of a physical structure has been realized:

Performance of the structure may be monitored by using historical and present-day data streams Critical limit states may be evaluated in a digital environment Lifecycle analyses may be performed to inform maintenance outside of routinely scheduled programs Threshold

Challenges

• Physical Asset

• Digital Twin

• Development of appropriate

• Development of data pipeline to performance metrics connect physical sensors to digital twin

• Deployment of suitable sensor net to capture relevant physical

• Creation of routines to process phenomena and interpret operational data

• Development of end-user application of methodology

• Instruction of end-user knowledge-base

Advantages of Methodology

• May provide near real-time assessment

• Not inhibited by outages as other periodic inspections

• Can incorporate data from periodic inspections

• Capable of identifying hidden/local deterioration

• Identifies potential areas of preventative maintenance

Digital Twins for Prognostic Health Management (PHM) in Nuclear Energy: Opportunities and Challenges Pradeep Ramuhalli Distinguished Scientist Virtual Workshop on Digital Twin Applications for Advanced Nuclear Technologies December 3, 2020 ORNL is managed by UT-Battelle, LLC for the US Department of Energy

Outline

• Background - drivers for prognostics health management in nuclear power

• Diagnostics, prognostics and decision making - An integrated solution using intelligent digital twins

• Examples

• Research Needs and Summary 2

The Big Picture Capital Costs Median Operations and Maintenance Costs in Nuclear Energy 25 20 15

$/MWh 10 Operating Costs 5

0 Year O&M Cost/MWh Operations Cost/MWh Maintenance Cost/MWh Energy Options Network Report (2019) What Will Advanced Nuclear Power Plants Cost? A Data from: Broken: Costs to Operate, Maintain Electricity Generation Have Soared Over Two Decades Standardized Cost Analysis of Advanced Nuclear Technologies in Commercial Development (uptake.com/energy)

Operating Plants Advanced Reactors Need: Information-driven Asset Management Technologies and best practices to lower operating and maintenance costs while maintaining safety and reliability 3

Diagnostics and Prognostics Enable Information-Driven Asset Management O&M Decision Making and Execution Safety Risk Economic Description (/yr) Risk (40 yrs)

EOL 6.21E-07 441 ERM - 6.60E-07 328 safety goal ERM - 5.26E-07 315 Real-time Model safety &

Complex System Calibration economics Predictive Risk Physics Constrained and Cost Dynamic & Surrogate Models Estimates A2 n =

A12 0 Condition Measurements Process Measurements Uncertainty Quantification

% Level Correlation Coefficients, Bad4 0.8 0.6 Equipment 0.4 0.2 0 7 14 0

-0.2 Time (Months)

Causal Diagnostics: Degradation

-0.4

-0.6

-0.8 Analysis Feature Engineering & Degradation Prognostics Signature Extraction Detection & ID & Confidence 4

Intelligent Digital Twins Enable PHM Risk Models and O&M Decision Making Metrics; Predictive Cost and Execution

& Risk Estimates Safety Risk Economic Description (/yr) Risk (40 yrs)

EOL 6.21E-07 441 ERM - 6.60E-07 328 safety goal ERM - 5.26E-07 315 Real-time Model Digital Twins safety &

Complex System Calibration economics Predictive Risk Physics Constrained and Cost Sensor Design and Dynamic & Surrogate Models Estimates A2 n =

Survivability for A12 0 Condition Condition Measurements Process Measurements Uncertainty Monitoring Quantification Inverse

% Level Data/Physics Online Problems and Correlation Coefficients, Bad4 Driven 0.8 0.6 Equipment 0.4 0.2 0 7 14 Condition Data Fusion Prognostics 0

-0.2 Time (Months)

Causal Diagnostics: Degradation

-0.4

-0.6 Monitoring

-0.8 Analysis Feature Engineering & Degradation Prognostics Signature Extraction Detection & ID & Confidence 5

Together with Advances in

• Sensors and instrumentation

>2Mrad JFET-based Sensor

• Modeling and simulation methods Interface Electronics (DOE NEET) and high performance computing Temperature, Pressure and Level Sensors SAW Chemical Sensors

• Data analytics, especially domain-aware data analytics

• Communication technologies Novel Ex-Vessel, In-Vessel, and In-Core Sensors and Electronics

• Advanced manufacturing 316L sheathed sensor in AM 316L build High Temperature Compatible and Embedded Sensors for Nuclear Process and Component Health Monitoring Self-powered Through-Wall Communication 6 3D Printing Passive Wireless Sensors

Digital Twin Neutron Flux

• A software design pattern that Temperature represents a physical object with the objective of understanding the assets state, responding to Steaming Rate &

changes, improving business Crud Thickness operations and adding value (Gartner)

• Potential for different levels of fidelity and for different uses, and spanning the range from fully data-driven to physics-based

- What is good enough for the problem?

7 ALMR PRISM Power Block

Intelligent Digital Twins for Diagnostics and Prognostics

• Hybrid (Data-driven, with domain information) can serve as digital twins for diagnostics and prognostics

• Reliability assessment and prediction

- Sensors

- Active components (pumps, valves, etc.)

- Passive components (piping, vessel, etc.)

- Sub-system (power conversion unit, etc.)

Physics Informed Machine Learning Reduced Order Model

• Risk-informed operational decision making for autonomous operations

• Risk-informed maintenance decision making for cost reduction 8

Robust Virtual Sensor Models Can Improve Sensor Drift Detection and Compensation Performance Example of Sensor Calibration Drift Detection and Compensation ITEM ID SENSOR TYPE MANUFACTURER 1 FT-4-1 DIFFERENTIAL PRESSURE ROSEMOUNT I&C026-10 2 FT-3-1 DIFFERENTIAL PRESSURE (SMART) ROSEMOUNT 3 FT-3-2 DIFFERENTIAL PRESSURE BARTON 10 16 4 FT-1-1 DIFFERENTIAL PRESSURE FOXBORO 7

5 FT-1-2 DIFFERENTIAL PRESSURE FOXBORO 17 9 6 FT-1-4 DIFFERENTIAL PRESSURE (SMART) BARTON 8 7 TE-1-2 RTD (SMART) ROSEMOUNT 11 4 5 6 12 8 TC-2-1 THERMOCOUPLE TYPE-J (SMART) ROSEMOUNT 13 9 FT-2-1 DIFFERENTIAL PRESSURE SCHLUMBERGER 1 2 15 3 14 10 CTRL-TEMP RTD (SMART) ROSEMOUNT 11 TC-HX-OUT THERMOCOUPLE TYPE-J OMEGA 12 FT-2-3 DIFFERENTIAL PRESSURE HONEYWELL Heat Exchanger Motor Pump 13 TC-HX-IN THERMOCOUPLE TYPE-J OMEGA 14 CTRL-PSR GAUGE PRESSURE FOXBORO 15 PT-2 GAUGE PRESSURE ROSEMOUNT 16 TC-LOOP-FAR THERMOCOUPLE TYPE-E OMEGA 17 TC-PUMP-OUT THERMOCOUPLE TYPE-K OMEGA 9 Tipireddy, Ramuhalli et al, ANS NPIC-HMIT 2017

Data-driven, Physics-Inspired Models for Diagnostics and Predictive Maintenance Pneumatic Valve Input Pressure History Top Chamber Bottom Chamber Valve Position s[k+1] = s[k] + t Prognostic Result: RUL s[k]

Data shown from Daigle and Goebel, IJPHM, 2008 Roy, Ramuhalli et al, ANS NPIC-HMIT 2015 10 Dib, Roy, et al, ANS NPIC-HMIT 2017

Data-driven, Physics-Inspired Models for Diagnostics and Predictive Maintenance Pneumatic Valve Input Pressure History Top Chamber Bottom Chamber Valve Position s[k+1] = s[k] + t Prognostic Result: RUL s[k]

Failure Threshold Region Monitored Quantity Increase in Loads Time Diagnostic/Predictive Model Selection Data shown from Daigle and Goebel, IJPHM, 2008 Roy, Ramuhalli et al, ANS NPIC-HMIT 2015 11 Dib, Roy, et al, ANS NPIC-HMIT 2017

Complex Multi-scale Physics of Failure Models Challenge PHM for Materials Failure; Data-driven Models Show Promise Visual Examination Ultrasonic Examination Irradiation (A533B Steel) Low Cycle Fatigue (A36 Steel)

(Matlack et al, 2014) (Walker et al, 2011)

(Jacobs 2015) High Cycle Fatigue (304 SS)

(Ramuhalli et al, 2014)

Nonlinear Ultrasonics MBN Peak 3500 Perpendicular 3000 Parallel 2500 2000 mV 1500 1000 MBN (Tensile Strain, 410 Steel) 500 (Ramuhalli et al, 2015) 0

-2% 8% 18% 28% 38%

12 Magnetic Barkhausen Noise Strain

Bayesian Methods Allow Integration of Failure Physics Information, Condition Data, and enable Uncertainty Quantification

• Underlying models can be at desired level of fidelity

• Prediction updates with new measurements

• Model updates over time also possible to reflect reality Evolution of Posterior Probability Density with Time 13

Example: Predicted Time-to-Failure (TTF) for Fatigue Crack Initiation

• Diagnostics and prognostics using data-driven models of

- Damage growth Predicted Damage Predicted TTF using 1

- Measurement Index (DI) using 1 Measurement Measurement Measurement

• Necessary data may Instant be difficult to acquire

• Physics-inspired models (damage growth and Measurement Instants measurement) have been used in other Predicted TTF instances with good Predicted DI using using 3 accuracy 3 Measurements Measurements 14 Ramuhalli et al, ANS NPIC-HMIT 2012

Digital Twin Model Updates are Essential for Many Applications

• Limited examples for certain fault conditions, Failure Threshold Region and limited data Increase in Loads Creep Strain

• Operational conditions may vary over time

• System or component condition may vary, requiring different models Primary Secondary Tertiary

• New failure modes with longer term Stage I Stage II Stage III operation Time Degradation Growth Characteristics -

• Continuous learning, with model selection, Function of Time and Load will be necessary Models Model Likelihood 15 RJMCMC for Model Selection Roy, Ramuhalli et al, ANS NPIC-HMIT 2015

Integrating Prognostic Results with Risk-Informed Operational Decision Making Sensing Prognostic Health and Risk Control Logic Decision Making Actuation Assessment 16 Cetiner et al, ANS NPIC-HMIT 2017

Integrating PHM and Risk Monitors with Plant Control Logic

• Risk: Measure of probability of some undesirable consequence Core damage frequency, large early release frequency, health consequences to the public Simplified Diagram of Multimodule Reactor Simplified Reactor PRA Event Tree

• Elements of PHM integration with risk monitors Equipment condition assessment (ECA) and prognostics for predictive health assessment Predictive risk assessment (safety and economic)

Uncertainty quantification 17 Cetiner, Muhlheim et al, Nuclear News 2015

Risk-informed Decisions: Economics and Safety

• Methodology for using cost metrics for Reduction in component replacement scheduling Expected Economic Case # Description CDF Risk Over 40

• Hypothetical cost and failure rates used in (/yr) yrs (Relative analysis to Case A)

A Expected end-

• Assessment computes safety related risk of-life 6.21E-07 -

metric (CDF) and normalized cost over 40 replacement B

years for three cases ERM - safety goal based 6.60E-07 25.6%

- Case A: Run to end-of-service-life; replace maintenance during scheduled outage. C ERM - safety and economics

- Case B: Use diagnostics/prognostics; replace based 5.26E-07 28.6%

equipment just prior to plant exceeding safety maintenance limit.

- Case C: Use diagnostics/prognostics; replace equipment if risk of unplanned outage at a future time. Schedule based on optimizing cost metric.

Ramuhalli, Veeramany et al, ICONE25, 2016 18

Summary

• Digital twin solutions for intelligent asset management and autonomous operations

- Enabled by technology advances in sensing, data analysis, modeling and simulation, and machine learning

• Technical challenges still exist and are targets for ongoing research

- Research leveraging advances in machine learning

• Resulting technologies enable sustainable nuclear power by improving the reliability and economics of nuclear plants 19

Looking Forward: Some Challenges

• Data

- Data access and data quality

- Optimal sensor type and placement

- Testbeds for data generation and verification and validation (V&V)

• Technology Development

- Robust digital twin development

- Model selection and model updates

- Robust diagnostics and prognostics in the presence of concurrent mechanisms, influencing factors, interacting subsystems, and measurement drift

- Methods for semi-autonomous decision making

• Deployment

- V&V approaches

- Uncertainty quantification

- Cybersecurity 20

Acknowledgments

• A number of collaborators have contributed to the work presented here, and include staff from National Laboratories (ORNL, PNNL, ANL, Bettis, INL),

Universities (UT-Knoxville, PSU, WSU, ISU, CSU-LB, WUSTL, Ajou University),

and Industry (AMS Corp.)

• A portion of the research presented here was supported by the USDOE Office of Nuclear Energy through the Advanced Reactor Technologies (ART), Nuclear Energy Enabling Technologies (NEET), and the National Scientific User Facility (ATR-NSUF) programs. A portion of the research was supported by the NNSA Office of Defense Nuclear Nonproliferation (NA22). Parts of this work were supported by Ajou University (S. Korea).

• Oak Ridge National Laboratory is operated by UT-Battelle for the US Department of Energy.

21 21

Questions?

22