Data Mining Solutions to Diagnostics
1999 AAAI Spring Symposium on AI in Equipment Maintenance Service & Support (Stanford University, CA, 3/99) |Home|Background|Solutions|SW Architecture|Real Examples|
Web-based computer product diagnosis
Web-based customer support (problem resolution) tool - AI on the Web
Diagnose an equipment for ST14 steel plate production
Diagnose a distillation tower to increase gasoline yield
Diagnose a platinum reforming system for ethylbenzene production
Industrial diagnosis using hyperspace data mining (a Microsoft Powerpoint file, 0.9MB) - a presentation at AAAI'99 Symposium: AI in Equipment Diagnosis and Maintenance Services.
Background on Diagnosis
A fault is an abnormal state of a machine or a system, including dysfunction or misfunction of a part, an assembly, or the whole system. The occurrence of a fault is associated with a number of factors, which in turn is related to a number of symptoms. Fault diagnostics is the study of the relationship of fault, factors and symptoms, and it is used to predict and control the performance of a system, be it a telecommunication system, a semiconductor manufacturing equipment, or even a human body.
Fault Diagnosis Methods
Many methods are available for fault diagnosis. They include
Time series analysis - use history data to predict future events
Fuzzy logic methods - fuzzy diagnosis matrix, clustering, evaluation, segmentation and learning
Neural networks - system adaptation by self learning
Multisource & multisensor data fusion - combination of numeric, logic, linguistic information
Case-based reasoning - heuristic method using experience and history data
Probability reasoning - Bayesian networks, Fisher's discrimination functions
Hybrid method - combination of above methods
Data mining techniques - seperate data and build a mathematical model in the the sub-space of factors. The model is used to isolate the fault, identify the factors that contribute to the fault, and provide advisory for fixing the fault.
Steps in Fault Diagnosis
Collect history data (in a database) diagnosis expertise from experts
Mechanism analysis - study fault physics
Collect primary cases - gather samples that prove or disapprove
Establish the relationship between factors and symptoms
Build the diagnosis matrix
Determine the inference rules from known experience and cases
Fix the fault
Update fault database
System self-learning - update inference rules and the diagnosis matrix
Diagnose an equipment for ST14 steel plate production
Diagnose a distillation tower to increase gasoline yield
Diagnose a platinum reforming system for ethylbenzene production
Last updated March 30, 1999, Copyright © 1997 - 2000, ZAPTRON Systems, Inc.