A Diagnostic and Control Expert System 
Based On a Plant Model 

ABSTRACT 

A conventional plant control system based on heuristics cannot deal with 
"unforeseen abnormal situations" in a plant due to the lack of heuristics. This 
system compiles plant control rules from a plant model, and deals with such 
situations. 

KEY FEATURES 

(1) Hybrid reasoning architecture with both heuristics-based approach and 
    model-based approach. 
(2) Diagnosis by the Qualitative Causal Model and the symptoms clustering 
    technique. 
(3) Compilation of plant control rules by the Device Model and the Operation 
    Principle Model. 
(4) Verification of compiled rules by the Dynamics Model. 





Figure 1: System configuration 
- 19 -

Outline of MMI outputs 

Case: the Deep Inference Subsystem is active 

(1) Plant configuration display window 
    This window shows plant devices modeled in the system. This window 
    highlights the target devices while the modules run. 
(2) Active module indication window 
    This window indicates the currently activated module. 
(3) DIS/SIS indicator panel 
    This panel indicates which subsystem is currently activated. 

Case: the Shallow Inference Subsystem is active 

(4) Trend monitoring window 
    This window shows both the predicted plant behavior and the observed 
    one by the Plant Monitor. 


- 20 -

Demonstration Schedule 

Only the Deep Inference Subsystem is demonstrated. The experimental results of the whole 
system, which consists of the Deep Inference Subsystem, the Shallow Inference Subsystem and the 
plant simulator, is demonstrated using the video tape. 

Step1) Occurrence of an unforeseen abnormal plant situation 
 
The case of multiple faults is demonstrated below. The video tape reproduces the scene of the 
abnormal plant situation caused by the malfunction of the plant simulator. In this demonstration, 
this occurrence is a manual input into the system. 

performance drop of a boiler-feed-water-pump & stuck close of a circulation-water-control-valve 
 

Step2) Detection of the symptoms 

The Plant Monitor detects the unallowable deviations between the predicted plant behavior and 
the observed one. After that, the Deep Inference Subsystem is activated. 





Step3) Activation of the Diagnosor 

The Diagnosor calculates the possible disorders for the detected symptoms by using the Qualitative 
Causal Model and the symptoms clustering technique. After that, the Diagnosor outputs the plant 
operations to stop the malfunctioning devices. 


- 21 -
Step4) Activation of the Operation Generator 

The Operation Generator figures out the necessary plant operations to deal with the unforeseen 
abnormal situation. This module utilizes the Device Model and the Operation Principle Model. 





Step5) Activation of the Precondition Generator 

This module generates the preconditions for each plant operation, and compiles plant control rules 
as follows. 



Step6) Activation of the Simulation-Verifier 

This module verifies the compiled rules by predicting the plant behavior to be observed when the 
plant is operated according to the compiled rules. 
 
Step7) Execution of the plant operation 

The video tape reproduces the scene of a plant whose operations are executed by the Shallow 
Inference Subsystem according to the compiled rules. To clarify the system's capability, the video 
tape also reproduces the scene of a plant being shutdown when none of the generated operations are 
executed. 
- 22 -