ML19011A444
| ML19011A444 | |
| Person / Time | |
|---|---|
| Issue date: | 01/16/2019 |
| From: | Office of Nuclear Regulatory Research |
| To: | |
| Nathan Siu 415-0744 | |
| Shared Package | |
| ML19011A416 | List:
|
| References | |
| Download: ML19011A444 (31) | |
Text
The Frontier: Grand Challenges and Advanced Methods Lecture 9-3 1
Key Topics
- PRA Grand Challenges
- General Methods
- AI/Big Data
- Dynamic PRA 2
Overview
Resources N. Siu, K. Coyne, and F. Gonzalez, Knowledge Management and Knowledge Engineering at a Risk-Informed Regulatory Agency:
Challenges and Suggestions, U.S. Nuclear Regulatory Commission, March 2017. (ADAMS ML17089A538)
N. Siu, "Risk assessment for dynamic systems: an overview,"
Reliability Engineering and System Safety, 43, 43-73, 1994.
T. Aldemir (ed.), Advanced Concepts in Nuclear Energy Risk Assessment and Management, World Scientific Publishing Co (2018). (Available from:
https://www.worldscientific.com/worldscibooks/10.1142/10587) 3 Overview
Other References N. Siu and K. Coyne, Knowledge Engineering at a Risk-Informed Regulatory Agency: Challenges and Suggestions, in Knowledge in Risk Assessment and Management, T. Aven and E. Zio, eds., Wiley, 2018.
NIST annual conference on text analytics (https://tac.nist.gov/)
N. Siu, "Dynamic accident sequence analysis in PRA: A Comment on
'Human Reliability Analysis - Where Shouldst Thou Turn?'," Technical Note, Reliability Engineering and System Safety, 29, No. 3, 359-364, 1990.
D. Helton, Scoping Study on Advanced Modeling Techniques for Level 2/3 PRA, U.S. Nuclear Regulatory Commission, May 2009. (ADAMS ML091320447)
J. LaChance, et al., Discrete Dynamic Probabilistic Risk Assessment Model Development and Application, SAND2012-9346, Sandia National Laboratories, October 2012. (ADAMS ML12305A351) 4 Overview
The PRA Frontier
- PRA is an integrative, systems engineering enterprise
- Can stimulate advances in specific disciplines, but typically
- Dont own phenomena
- Advances involve new collection, adaptation, integration, assembly
- Frontier
- Long-standing completeness issues (Grand Challenges)
- New tools and approaches 5
Overview
Grand Challenges
- Errors of commission
- TMI-2: throttling safety injection
- Chernobyl 4: disabling automatic trips
- Fukushima Dai-ichi 1: isolating passive system
- Organizational factors
- Major role in accidents
- Arguably major role in most if not all incidents (e.g., Browns Ferry, Blayais, Davis-Besse)
- Security-related events; design faults
- Controversies over fundamental framework
- Information availability 6
Grand Challenges
Technological Advances
- Artificial Intelligence (AI)/Big Data
- Simulation Methods and Dynamic PRA 7
General Methods
AI/Big Data
- Historical attempts
- Expert systems, e.g., fault diagnosis
- Automated model construction
- Recent interest
- Prognostics
- Text mining 8
Big Data: Prognostics and Reliability Concept: use field data and physics of failure models to anticipate failures and develop prevention strategies Longstanding interest in nuclear community, gaining ground with technological advances and applications to conventional power plants A non-nuclear example: Formula 1 race cars
- 1600 lbs, turbocharged hybrids
- Engineering models
- Real-time support during race
- Empirically calibrated through testing
- ~100 supporting staff at home office
- Miniscule performance D => major effects 9
A. Gilbertson, 2016 Big Data: Prognostics and Reliability
Real-Time Monitoring: Qualification at Circuit of the Americas (COTA) 10 Footage courtesy of A. Gilbertson COTA Turn 15 Entry speed ~210 km/h Speed at apex ~84 km/h Braking distance ~30m Braking power ~800 kW Load ~3g Big Data: Prognostics and Reliability Formula 1 Video
11 Data Streams 120+ sensors (car and driver) 1000-2000 wireless channels Low latency - o(ms) 2 GB/lap, 3 TB/race Thermal cameras during qualification Big Data: Prognostics and Reliability
Data Use
- Engineering models
- Empirically calibrated through testing
- Predict performance and wear over time
- Used to develop/modify (in real time) race strategies
- Use in a Formula 1 PRA? Models dont cover all factors
- Road debris
- Other drivers
- Would an analogous approach for NPPs catch risk-significant conditions?
12 Big Data: Prognostics and Reliability
Sample Old But Interesting Events 13 Year Plant Feature 1975 Browns Ferry Cable fire affects multiple units 1985 Hatch HVAC water falls into MCR panel; SRV cycles, sticks open 1993 Cooper External flood, one evacuation route blocked 1997 Fort Calhoun Steam line rupture, intermittent electrical grounds 1999 Blayais High wind and external flood affect multiple units, site access 2001 Maanshan High energy arc fault, station blackout Big Data: Prognostics and Reliability
More Interesting Events - Multi-Unit Precursors 14 Big Data: Prognostics and Reliability Date Plant Description 6/22/82 Quad Cities LOOP, Maintenance 8/11/83 Salem LOOP, Clogged screens 7/26/84 Susquehanna SBO, Bkr mis-aligned 5/17/85 Turkey Point LOOP, Brush fires 7/23/87 Calvert Cliffs LOOP, Offsite tree 3/20/90 Vogtle SBO, Truck hit support 8/24/92 Turkey Point LOOP, Hurricane 12/31/92 Sequoyah LOOP, Switchyard fault 10/12/93 Beaver Valley LOOP, Offsite fault 6/28/96 LaSalle Trip, Foreign material in SW Tunnel 6/29/96 Prairie Island LOOP, High winds Date Plant Description 8/14/03 6 Sites LOOP, NE Blackout 6/14/04 Palo Verde LOOP, Offsite fault 9/25/04 St. Lucie LOOP, Hurricane 5/20/06 Catawba LOOP, Switchyard fault 3/26/09 Sequoyah LOOP, Bus fault 4/16/11 Surry LOOP, Tornado 4/27/11 Browns Ferry LOOP, Winds/tornadoes 8/23/11 North Anna LOOP, Earthquake 3/31/13 ANO LOOP/Trip, Load drop 4/17/13 LaSalle LOOP, Lightning 5/25/14 Millstone LOOP, Offsite fault
Observation
- Prognostics could be useful for some PRA scenarios but probably not for many others
- Need to think about specific needs of PRA when formulating an N+1 project 15 Big Data: Prognostics and Reliability
A Different Use of AI/Big Data Concept: use smart tools to identify and enable efficient use of documents for PRA/RIDM Responses to queries Active alerts Example: Le Blayais (1999)*
International Nuclear Event Scale (INES) Level 2 (out of 7): Incident with significant failures in safety provisions Reported in several venues Missed by general PRA community Actual vs. potential losses Busy (IPE/IPEEE; RG 1.174, revised ROP, other events)
Perceived importance of flooding Now generally considered a precursor to Fukushima 16 E. De Fraguier, Lessons learned from 1999 Blayais flood: overview of EDF flood risk management plan, U.S. NRC Regulatory Information Conference, March 11, 2010.
AI/Big Data: Text Mining
- See Lecture 7-2
Fundamental knowledge engineering challenge - understanding the message(s) 17 Challenge Type Example Phrase Challenge for KE Tool Ambiguity (multiple meanings for the same word or phrase) pumps of Train A were lost lost = failed (versus missing, bewildered, etc)
Context dependence (meaning depends on other factors, including document type, purpose, structure, and surrounding text) essential service water pumps essential is part of system name (vs. descriptor, as in references to non-essential service water pumps)
Implicitness (meaning is not stated directly, and must be inferred from other facts in document)
Of the facilities which were flooded the following should be noted: rooms containing the essential service water pumps Widely separated text needs to be combined to infer that the ESW pump rooms were flooded Non-uniqueness (multiple ways of making a statement with the same meaning)
The essential service water system of each unit comprises four pumps on two independent trains (A and B) alternatives include different words, e.g., has four pumps in two separate trains or different constructions, e.g., There are four pumps in the essential service water system, arranged in independent trains A and B.
Other Challenges
- Enormous (and rapidly growing) volume
- Information from embedded structures
- Database limitations
- Legacy documents only in hardcopy
- Faults in documents
- Poor OCR, metadata
- Enormous interest
- Hype
- Rapid advances 18 AI/Big Data: Text Mining
An old vision Intelligent aides
- Library resource (e.g., Star Trek, 1966-1969)
- Active agent (e.g., 2001: A Space Odyssey, 1968)
Key features include:
- Natural language information processing
- Data collection, validation, understanding, reporting
- Performed in interactive mode (even dialog!)
- Prioritization for attention, action 19 Im worried about the mission, Dave.
Cmon HAL, open the pod bay door AI/Big Data: Text Mining
Risk Information: Special Characteristics
- Systems viewpoint
- Multiple technical disciplines
- Problem scale and complexity (multitude of scenarios, interacting pieces)
- Diverse and implicit sources of information (licensing basis, operating experience, past analyses, )
- Rare events
- Sparse data
- Large uncertainties
- Broad user base
- Technical backgrounds
- Approaches to cope with uncertainties
- Responsibilities 20 What can go wrong?
AI/Big Data: Text Mining
Status Rapidly growing interest in NRC, industry Small-scale exploratory studies (NRC, EPRI)
Human-in-the-loop approaches likely most efficient and effective (at least in near term)
Developments will depend on intended mode of application 21 Searcher Explorer Aide Oracle Tool Toy Servant Partner AI/Big Data: Text Mining
Dynamic PRA - Not If But When?
- Concept
- General: Explicitly model driving forces on NPP elements and element responses
- Typical: Simulation-oriented analysis
- Origin
- Fast reactor uncertainty analyses
- Control systems
- Human reliability analysis 22 Dynamic PRA
Attractive Features
- As with simulation approaches in general
- Potential for improved realism
- Consistency with current directions in engineering
- Natural, holistic framework for integrating multiple disciplines
- Chronological orientation - understandable stories
- Better use of rich information from events
- For PRA/RIDM, potential to address difficult problems (including Grand Challenges)
- Errors of commission
- Passive system reliability 23 Dynamic PRA
1F1, 3/11/2011 24 Time Relative Time Hazard Systems Indications Operators/Workers ERC/ER team EP 14:46 0:00 Earthquake Scram 14:47 0:01 MSIVs close, turbine trips, EDGs start and load Rx level drops 14:52 0:06 ICs start automatically RV pressure decreases; RV level in normal range 15:03 0:17 ICs removed from service Cooldown rate exceeding tech spec limits Manually remove IC from service 15:06 0:20 Disaster HQ established in TEPCO Tokyo 15:10 0:24 Determine only 1 train IC needed; cycle A train 15:27 0:41 First tsunami arrives 15:35 0:49 Second tsunami arrives 15:37 0:51 Loss of AC 15:37 0:51 1537-1550: Gradual loss of instrumentation, indications (including IC valve status, RV level),
alarms, MCR main lighting Determine HPCI unavailable 15:42 0:56 TEPCO enters emergency plan (loss of AC power); ERC established 16:35 1:49 D/DFP indicator lamp indicates "halted" 16:36 1:50 Cannot determine RV level or injection status; work to restore level indication; do not put IC in service Review accident management procedures, start developing procedure to open containment vent valves without power Declared emergency (inability to determine level or injection)
Dynamic PRA
25 Time Relative Time Hazard Systems Indications Operators/Workers ERC/ER team EP 16:45 1:59 Determine RV level Emergency cancelled 16:55 2:09 Tsunami alert Workers on way to check D/DFP had to turn back 17:07 2:21 Lose ability to determine RV level or injection status Reentered emergency plan 17:12 2:26 Site superintendent directs investigation of using fire protection to inject water 17:15 2:29 Estimated core uncovery in 1 hr 17:19 2:33 Tsunami alert cleared 17:30 2:44 Diesel-driven fire pump started and left to idle Pressure above 100 psi Manually open valves (in dark) from fire protection system to core spray system; take turns holding D/DFP switch to keep in standby 18:00 3:14 Govt orders seawater injection 18:18 3:32 DC power partially returned MO-3A and MO-2A indicate closed 18:18 3:32 MO-3A and MO-2A opened Open IC valves MO-3A and 2A. Steam from condenser observed 18:25 3:39 MO-3A closed Remove IC from service (concerned about failing lines). Entered R/B and T/B to manually open MOV for FP lineup.
Hard time finding valve, had wrong key, hard to operate hand wheel.
Long time.
Dynamic PRA 1F1, 3/11/2011 (cont.)
26 Time Relative Time Hazard Systems Indications Operators/Workers ERC/ER team EP 18:46 4:00 Core degradation and major release (4-5 hr after trip) 19:00 4:14 Close valves for broken outdoor FP pipes.
Broke lock to allow passage between Units 2 and 3.
Ask Tokyo for more fire engines Following PM question about possible recriticality, TEPCO makes strong request to site superintendent to suspend seawater injection 19:03 4:17 Govt. declares nuclear emergency 20:07 5:21 No pressure indication in MCR; Reactor pressure
= 6.89 MPa (1000 psi) local indication 20:49 6:03 Small portable generator installed MCR has temporary lighting 20:50 6:04 Local authorities order evacuation within 2 km 21:19 6:33 Level indication restored; level = 0.20 m (8) above TAF 21:23 6:37 Prime minister orders evacuation within 3 km; sheltering out to 10 km 21:30 6:44 MO-3A opened Place IC in service; steam observed 21:51 7:05 Access to RB restricted due to dose rates -
indirect indication of core uncover 22:00 7:14 Level = 0.55 m (21.7) above TAF 23:50 9:04 Drywell pressure = 0.50 MPa (87 psi) above design Restoration team from ERC enables reading 23:59 9:13 Offsite power supply trucks arrive by midnight Dynamic PRA 1F1, 3/11/2011 (cont.)
Errors of Commission - Importance of Context Can always add a basic event Operator Stops RCIC.
Possible but sufficiently probable? Why or why not?
27 Dynamic PRA
Predominant Approach: Discrete Dynamic Event Trees 28 Dynamic PRA Adapted from: N. Siu, "Risk assessment for dynamic systems: an overview,"
Reliability Engineering and System Safety, 43, 43-73, 1994 Historical J. LaChance, et al., Discrete Dynamic Probabilistic Risk Assessment Model Development and Application, SAND2012-9346, Sandia National Laboratories, October 2012.
More Recently
International R&D Organization Tool(s)
Electricité de France (EDF)
RAVEN, PyCATSHOO Gesellschaft für Anlagen und Reaktorsicherheit (GRS)
MC-DET Idaho National Laboratory (INL)
RAVEN Ohio State University (OSU)
ADAPT Sandia National Laboratories (SNL)
ADAPT University of California at Los Angeles (UCLA)
ADS/IDAC University of Maryland (UMD)
ADS/IDAC, ADAPT Universidad Politécnica de Madrid (UPM)
SCAIS 29 Dynamic PRA
Implementation Challenges
- Technical (many being addressed)
- Phenomenological sub-models
- Data
- V&V
- Computational resources
- Aids to support searches
- Aids to support sensemaking
- Economic
- Demonstrating added value
- Demonstrating acceptable resource requirements 30 Dynamic PRA
Implementation Challenges (cont.)
- Socio-organizational
- Perception that dynamic PRA is necessarily complex
- Developer community mindset
- Increased detail > increased realism
- Importance of insights (vs. bottom line results)
- Openness to concerns raised by skeptics
- User community mindset
- Potential value of different approaches
- Targeting of development activities
- R&D => product development
- Increased emphasis on actual problem solving (beyond demos)
- Role in PRA toolbox
- What expertise is needed, how to develop and maintain 31 Dynamic PRA