IMMUNITY INSPIRED DIAGNOSTICS AND PROGNOSTICS IN SYSTEMS OF INTERACTING DYNAMIC SUBSYSTEMS

Transcription

1 NSF I-UCRC on Intelligent Maintenance Systems; Planning Workshop at the University of Texas; May 15, 2012, Austin, TX IMMUNITY INSPIRED DIAGNOSTICS AND PROGNOSTICS IN SYSTEMS OF INTERACTING DYNAMIC SUBSYSTEMS PROF. DRAGAN DJURDJANOVIC

2 2 PRESENTATION OUTLINE» Motivation» Methodology» Preliminary Work» Research Plan NSF I-UCRC on Intelligent Maintenance Sys. Planning Workshop at the University of Texas

3 3 MOTIVATION: SEVERAL MONTHS OF PECVD TOOL OPERATION Particle failures tool down for several weeks Time Coulomb crystals and particle failures Again tool down for several weeks Each dot is a FOUP of wafers

4 4 PRESENTATION OUTLINE» Motivation» Methodology» Preliminary Work» Research Plan

5 5 FAULT ISOLATION VIA DISTRIBUTED ANOMALY DETECTION AD1 Compressor Charge Air Cooler Throttle Valve AD2 Intake Manifold Engine *ADA - Anomaly Detector Neng AD3 Xth WF UE L XE GR WHFM Engine Control Unit (ECU) TCAC α Pm λ Neng EGR Valve EGR Cooler Exhaust Manifold VNT- Turbine ADA4 XVNT Block diagram of the mass airflow in a diesel engine provided by the Bosch Research Center, Pittsburgh

6 6 AD11 Compressor FAULT ISOLATION VIA DISTRIBUTED ANOMALY DETECTION AD12 Charge Air Cooler AD13 Throttle Valve AD2 Intake Manifold Engine Neng *ADA - Anomaly Detector Xth WF UE L XE GR WHFM Engine Control Unit (ECU) TCAC α Pm λ Neng EGR Valve EGR Cooler Exhaust Manifold AD3 VNT- Turbine ADA4 XVNT Negative selection in natural immune systems (Dasgupta, 2006) Block diagram of the mass airflow in a diesel engine provided by the Bosch Research Center, Pittsburgh

7 7 PRESENTATION OUTLINE» Motivation» Methodology» Preliminary Work» Research Plan

8 Anomaly Detection using Piecewise Dynamic Models Local model 6 Local model 7 Local model 1 Local model 5 Plant Local model 2 Local model 4 Piecewise Dyn. Model - Operating Region Current behavior Compute CV Normal behavior J. Liu, D. Djurdjanovic, K. Marko and J. Ni, Growing Structure Multiple Model System for Anomaly Detection and Fault Diagnosis, Transactions of ASME, Journal of Dynamic Systems, Measurements and Control, Vol. 131, No. 5, pp , 2009 Anomalous Behavior

9 Precedent Free Fault Isolation in the Exhaust Gas Recirculation (EGR) System of a Diesel Engine Injected Fuel Engine Speed Look-up Table φecu ADover all Look-up Table PI Controller Exhaust Pressure Intake Man. Pressure Exhaust Temp EGR Valve Ambient Temperature EGR Cooler TESIS DYNAware, 2006 M. Cholette and D. Djurdjanovic, Precedent-Free Fault Isolation in a Diesel Engine EGR System, to appear in ASME J. of Dynamic Systems Measurements & Control

10 Precedent Free Fault Isolation in the Exhaust Gas Recirculation (EGR) System of a Diesel Engine Injected Fuel Engine Speed Look-up Table φecu ADmass flow Look-up Table PI Controller Exhaust Pressure Intake Man. Pressure Exhaust Temp EGR Valve Ambient Temperature ADcooler EGR Cooler TESIS DYNAware, 2006 M. Cholette and D. Djurdjanovic, Precedent-Free Fault Isolation in a Diesel Engine EGR System, to appear in ASME J. of Dynamic Systems Measurements & Control

11 Precedent Free Fault Isolation in the Exhaust Gas Recirculation (EGR) System of a Diesel Engine Injected Fuel Engine Speed AD3 Look-up Table φecu Exhaust Pressure Intake Man. Pressure Exhaust Temp Look-up Table AD2 AD4 AD5 PI Controller EGR Valve Ambient Temperature ADcooler EGR Cooler TESIS DYNAware, 2006 M. Cholette and D. Djurdjanovic, Precedent-Free Fault Isolation in a Diesel Engine EGR System, to appear in ASME J. of Dynamic Systems Measurements & Control

12 Anomalies in the EGR Valve (Clogging of the Valve) Injected Fuel Engine Speed Look-up Table φecu Look-up Table Exhaust Pressure Intake Man. Pressure Exhaust Temp PI Controller EGR Valve Ambient Temperature EGR Cooler TESIS DYNAware, 2006 M. Cholette and D. Djurdjanovic, Precedent-Free Fault Isolation in a Diesel Engine EGR System, to appear in ASME J. of Dynamic Systems Measurements & Control

13 Detection and Isolation of an Anomaly in the EGR Valve ADmass flow ADcooler AD2 (Look up table) AD3 (Look up table) AD4 AD4 (PI Controller) AD5 (Valve) 13

14 14 PRESENTATION OUTLINE» Motivation» Methodology» Preliminary Work» Research Plan

15 15 TASK 1: DISTRIBUTED ANOMALY DETECTION WHEN CAUSAL CONNECTIONS ARE NOT CLEAR Physics based reasoning and expert knowledge Bayesian causation discovery approaches

16 16 TASK 2: POLICY FOR DISTRIBUTING ANOMALY DETECTORS IN A COMPLEX SYSTEM d0 Detection rate for grouped anomaly detector d1 Detection rate for individual FRU anomaly detectors CC Computational cost per unit time Ci Cost of failing to localize the fault per unit time 16

17 17 TASK 3: SIGNATURE PREDICTION FOR MAINTENANCE DECISION-MAKING Root cause of the degraded state Similarity based prediction methods for predicting signature distributions Warning about a degraded state Existing methods are either dealing with stationary time-series or are too slow to provide real-time information Combine analytics with simulation based tools to obtain accurate predictions fast

18 18 POTENTIAL APPLICATION AREAS Automotive engines Aircraft jet engines Advanced manufacturing equipment Electrical generator systems Modular data centers

19 19 SUMMARY OF PROPOSED WORK Task 1: Identification of causal Method for identifying relations relations between subsystems in a between subsystems in a highly complex system complex system of interacting Task 2: Optimization of policies for dynamic subsystems distribution of anomaly detectors Demonstration of fault that achieve tradeoff between localization in a complex speed and accuracy of fault system in which causal localization and computation load relations between subsystems are not a priori known Task 3: Implementation of (semicond. dataset) degradation monitoring techniques to a real system Mentors: Prof. Dragan Djurdjanovic & industrial partners Researchers: One graduate student per application area in year 1 and 2 graduate students in year 2 (possible application areas include manufacturing equipment, engines, electrical generators) YEAR 1 Ability to deal with unprecedented faults (faults never seen in the past) Data mining to understand relations between subsystems of a highly complex (multi-physics) system Potentials to dramatically shorten equipment downtimes and their impacts T A S K S

20 20 BUDGET SUMMARY Development of the method for causation discovery and one application area, such as plasma based semiconductor manufacturing tool $40K (student tuition and stipend for a year) Budget for a second application example Additional $20K (student tuition and support for 6 months) Total: $40-60K Year 1 Budget Year 2 Budget Development of the method for optimization of policies for distribution of anomaly detectors, with one application area, such as plasma based semiconductor manufacturing tool or automotive engine $40K (student tuition and stipend for a year) Development of the method for predicting signatures of the monitored system, with one application area (semiconductor manufacturing dataset) 40K (student tuition and stipend for a year) Additional $20K for another application dataset (student support and tuition for 6 months) Total: $80-100K (could be split into 2 years)

21 THANK YOU