Fault Detection and Diagnosis of the Tennessee Eastman

  • Slides: 18
Download presentation
Fault Detection and Diagnosis of the Tennessee Eastman Process using Multivariate Control Charts Jeremy

Fault Detection and Diagnosis of the Tennessee Eastman Process using Multivariate Control Charts Jeremy Ash, JMP Division, SAS Institute Copyright © SAS Inst itute Inc. All rig hts reserved.

Preliminaries I am assuming some knowledge of statistical process control • If you haven’t

Preliminaries I am assuming some knowledge of statistical process control • If you haven’t already seen these talks on MDMCC: • • Large-Scale Process Monitoring Using JMP - Jianfeng Ding and Laura Lancaster • Monitoring, Diagnosing and Improving Large-Scale Complex Processes - Byron Wingerd • Focus areas of this talk • Application of MDMCC to a simulation of a real world industrial chemical process • On-line monitoring of data in a database Copyright © SAS Inst itute Inc. All rig hts reserved.

Why Not Use Univariate Control Charts? Quality control data sets are often high dimensional

Why Not Use Univariate Control Charts? Quality control data sets are often high dimensional • Multivariate outliers can go undetected: • Univariate in-control Multivariate out-of-control Copyright © SAS Inst itute Inc. All rig hts reserved.

PCA and PLS Projection Methods a p • PCA X Process n a a

PCA and PLS Projection Methods a p • PCA X Process n a a << p n P` PLS p X Process n a T W` Copyright © SAS Inst itute Inc. All rig hts reserved. a T m a U n Max(Cov (tk, uk)) a Y Quality Q`

T 2 and SPE Charts Copyright © SAS Inst itute Inc. All rig hts

T 2 and SPE Charts Copyright © SAS Inst itute Inc. All rig hts reserved.

Historical Versus Current Data Build models Apply models Historical process New process data Copyright

Historical Versus Current Data Build models Apply models Historical process New process data Copyright © SAS Inst itute Inc. All rig hts reserved.

Tennessee Eastman Process • Simulation of an industrial chemical process • Developed by Downs

Tennessee Eastman Process • Simulation of an industrial chemical process • Developed by Downs and Vogel 1 at Eastman Chemical • The simulated process: • Used extensively in the benchmarking of multivariate process control methods Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Process Diagram Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical

Process Diagram Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Manipulated Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical

Manipulated Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Manipulated Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical

Manipulated Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Manipulated Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical

Manipulated Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Manipulated Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical

Manipulated Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Measurement Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical

Measurement Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Composition Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical

Composition Variables Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Demo Overview Process Oracle, My. SQL, etc. Streaming Session JMP Data Table Monitoring Session

Demo Overview Process Oracle, My. SQL, etc. Streaming Session JMP Data Table Monitoring Session Copyright © SAS Inst itute Inc. All rig hts reserved. MDMCC Report

Process Control Goals • Downs and Vogel didn’t provide a single performance metric •

Process Control Goals • Downs and Vogel didn’t provide a single performance metric • “We felt that the tradeoffs among the possible control strategies and techniques involve much more than a mathematical expression. ” • Goals: • Maintain the process variables at desired values • Minimize variability of product quality during disturbances • Recover quickly and smoothly from disturbances Downs, J. J. , & Vogel, E. F. (1993). Computers & chemical engineering, 17(3), 245 -255. Copyright © SAS Inst itute Inc. All rig hts reserved.

Key MDMCC Features Multivariate Control Charts • On-line fault detection and off-line fault diagnosis.

Key MDMCC Features Multivariate Control Charts • On-line fault detection and off-line fault diagnosis. • Methods to “drill down” to the root cause of a fault. • Switching between multivariate and univariate control chats in an interactive and user-friendly way. Chiang, L. H. , Russell, E. L. , & Braatz, R. D. (2000). Copyright © SAS Inst itute Inc. All rig hts reserved. Univariate Control Charts

Email me with any further questions! • Jeremy. Ash@jmp. com Copyright © SAS Inst

Email me with any further questions! • Jeremy. Ash@jmp. com Copyright © SAS Inst itute Inc. All rig hts reserved.

TENNESSEE HISTORY TENNESSEE Tennessee State Seal TENNESSEE TennesseeTENNESSEE HISTORY TENNESSEE Tennessee State Seal TENNESSEE Tennessee
Fault Management Detection and Diagnosis Outline Fault managementFault Management Detection and Diagnosis Outline Fault management
Fault Detection and Diagnosis Outline l Fault managementFault Detection and Diagnosis Outline l Fault management
A Shakeup at Eastman Kodak MANAGEMENT CASE EastmanA Shakeup at Eastman Kodak MANAGEMENT CASE Eastman
Mrs Eastman Hi My name is Mary EastmanMrs Eastman Hi My name is Mary Eastman
Fault Diagnosis SATBased Diagnosis The Diagnosis Problem GeneralFault Diagnosis SATBased Diagnosis The Diagnosis Problem General
Active Fault An active fault is a faultActive Fault An active fault is a fault
Normal Fault Normal Fault A normal fault actsNormal Fault Normal Fault A normal fault acts
Fault Diagnosis Using Fault Dictionaries and Probability AdamFault Diagnosis Using Fault Dictionaries and Probability Adam
Fault Modeling Simulation and Diagnosis Fault Modeling FaultsFault Modeling Simulation and Diagnosis Fault Modeling Faults
Objective Fault Detection capturing the fault and estimatingObjective Fault Detection capturing the fault and estimating
Transmission line fault detection INTRODUCTION When fault occursTransmission line fault detection INTRODUCTION When fault occurs
m SWAT LowCost Hardware Fault Detection and Diagnosism SWAT LowCost Hardware Fault Detection and Diagnosis
Chiller Fault Detection and Diagnosis FDD Paul RiemerChiller Fault Detection and Diagnosis FDD Paul Riemer
MSWAT Hardware Fault Detection and Diagnosis for MulticoreMSWAT Hardware Fault Detection and Diagnosis for Multicore
Module 4 Screening and Diagnosis Diagnosis Definitive diagnosisModule 4 Screening and Diagnosis Diagnosis Definitive diagnosis
DIAGNOSIS EPIDEMIOLOGI Pengertian diagnosis epidemiologi adalah diagnosis yangDIAGNOSIS EPIDEMIOLOGI Pengertian diagnosis epidemiologi adalah diagnosis yang