A Real-Time Expert System for Process Control - ACS Symposium

5 A Real-Time Expert System for Process Control

Downloaded by UNIV OF MASSACHUSETTS AMHERST on October 7, 2015 | http://pubs.acs.org Publication Date: April 30, 1986 | doi: 10.1021/bk-1986-0306.ch005

Lowell B. Hawkinson, Carl G. Knickerbocker, and Robert L. Moore LISP Machine Inc., Los Angeles, C A 90045

Expert systems technology can provide improvements i n analysis of process information, i n t e l l i g e n t alarming, process diagnosis, c o n t r o l and optimization of processes. However, t o r e a l i z e these benefits, a real-time expert system c a p a b i l i t y i s required. A program design i s described which supports forward and backward chaining inference i n a real-time environment, with dynamic measurement data. The knowledge base for the program i s implemented i n structured natural language form for application to a broad range of process expert systems. Plant t e s t r e s u l t s are described.

In the real-time a p p l i c a t i o n of expert systems, a number of design considerations, beyond those usually considered i n expert systems, become important. Execution e f f i c i e n c y i s a prime consideration. In conventional expert systems, the facts and knowledge upon which the inference i s based are s t a t i c . In the i n d u s t r i a l a p p l i c a t i o n , the facts or process measurements are dynamic. In an i n d u s t r i a l a p p l i c a t i o n there may be several thousand measurements and alarms which may s i g n i f i c a n t l y change i n value or status i n a few minutes. The problem posed by an operator advisor, to give expert diagnosis of plant condition and to recommend emergency actions or economic optimization adjustments, i l l u s t r a t e s these real-time requirements. Some of the plant conditions which can occur include : 1. C r i t i c a l measurement f a i l u r e . In t h i s case, the information presented to the operator i s incorrect. An expert system would use a process knowledge base to detect inconsistencies and t o a l e r t the operator. 2. Process upset. In t h i s case, the expert system would i d e n t i f y underlying process problems, distinguishing causes from e f f e c t s , and would advise the operator accordingly. Heuristic 0097-6156/ 86/0306-O069$06.00/0 © 1986 American Chemical Society

In Artificial Intelligence Applications in Chemistry; Pierce, T., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 1986.

70

ARTIFICIAL INTELLIGENCE APPLICATIONS IN CHEMISTRY

rules of optimization would be applied by the expert system to give control advice.


In these examples, the expert system i s simply applying the expertise used i n i t s development. The p o t e n t i a l advantage of the operator advisor i s that t h i s expertise i s available q u i c k l y , on any s h i f t , for providing organized advice to the operator. To meet these requirements, several design considerations must be addressed: 1. Data access. An e f f i c i e n t real-time data interface must be established with the d i s t r i b u t e d measurement system. 2. Inference paradigms. The basic inference mechanisms of forward-chaining and backward-chaining must be integrated into a real-time execution environment. 3. Computational e f f i c i e n c y . The e f f i c i e n c y of inference i s enhanced by program and knowledge-base structure and by machine speed. Also, h e u r i s t i c procedures, as used by experts, can augment the deductive procedures of conventional inference. The program developed by LMI i n response to these design requirements i s c a l l e d Process I n t e l l i g e n t Control (PICON). The i n d i v i d u a l design considerations are addressed i n the following discussion. Process I n t e l l i g e n t Control The expert system package i s designed to operate on a LISP machine interfaced with a conventional d i s t r i b u t e d c o n t r o l system. The design assumes that up to 20,000 measurement points and alarms may be accessed. The Lambda machine from LMI was u t i l i z e d . The r e a l time data interface i s v i a an i n t e g r a l Multibus connected to a computer gateway i n the d i s t r i b u t e d system. Data t r a n s f e r s , i n f l o a t i n g point engineering u n i t s or i n status states, are requested by the expert system. Thus the d i s t r i b u t e d system does not transmit a l l measurements and alarms on a fixed scan basis, but rather the process data are accessed as required for inference. In a sense, the expert system i s acting l i k e an expert operator, who focuses attention or scans the process operation s e l e c t i v e l y , using expertise to determine s p e c i f i c areas of attention. The basic inference paradigms supported by the expert system are forward-chaining and backward-chaining. Within the context of an alarm advisor, there are requirements for both of these paradigms. An expert process operator, during normal plant operation, w i l l scan key process information. This i s for purposes of monitoring c o n t r o l performance and detecting problems which may not cause e x p l i c i t alarms. The programming paradigm which r e f l e c t s t h i s approach i s a scanned forward-chaining inference. The h e u r i s t i c rules which determine possibly-significant-events are scanned, and rule condition matching t r i g g e r s an a l e r t to the expert system monitor program. Conventional alarms also may trigger an a l e r t , i f they are h e u r i s t i c a l l y ranked as p o s s i b l y significant-events.



5.

HAWKINSON E T A L .

A Real- Time Expert System for Process Control

71

An expert process operator, once a l e r t e d , w i l l focus a t t e n t i o n on the problem. This may involve invoking procedure rules f o r safety or other reasons, and i t may involve assembling information and primary analyses to allow inference about the problem. Logic rules and procedures are used when required for the diagnostic inference. The expert system mimics the expert process operator in t h i s regard: Logic rules and procedures are invoked s p e c i f i c a l l y when they are required for diagnosis of a process problem, or as requested for a s p e c i f i c step i n inference. In working through process c o n t r o l examples, we found that many c a l c u l a t i o n s , data checks, rate checks and other computationally intensive tasks are done at the f i r s t l e v e l of inference. Considerations of computational e f f i c i e n c y led to a design u t i l i z i n g two p a r a l l e l processors with a shared memory (Figure 1). One of the processors i s a 68010 programmed i n C code. This processor performs computationally intensive, low l e v e l tasks which are directed by the expert system i n the LISP processor. The processing of data applies a l e v e l of i n t e l l i g e n c e . Instead of mere measurement values, the expert may base inference on trends or patterns of measurements. Thus the system must be able to access p r i m i t i v e functions of data, such as averages and trends of values, and q u a l i t y information, such as the presence of noise or discontinuous values. Such functions are conveniently calculated i n the p a r a l l e l 68010 processor, coded i n C language for execution e f f i c i e n c y . An expert, given time to do so, may u t i l i z e c a l c u l a t i o n s to develop inference r e s u l t s . For example, a material balance c a l c u l a t i o n around a process unit may indicate a measurement inconsistency. To mimic t h i s expertise, general mathematical operations on combinations of measurements or functions of measurements are implemented i n the p a r a l l e l processor also. Higher l e v e l s of inference depend on the truth conditions of the f i r s t l e v e l antecedent conditions, and thus higher l e v e l s of inference involve pattern matching and chained-inference l o g i c . Higher l e v e l inference i s done i n the LISP processor, using various expert system paradigms, while the f i r s t l e v e l antecedents, which are computationally intensive, are evaluated i n the p a r a l l e l 68010 processor. The expert system package i s designed so that an algorithm of reasonably a r b i t r a r y structure can be dynamically loaded into the 68010 from the LISP processor. This allows, for example, the expert system to implement process-monitoring f u n c t i o n a l i t y i n a dynamic fashion, the equivalent o f : "look c l o s e l y at the energy balance around the s p e c i f i c process unit for the next few minutes." The expert system design includes the a b i l i t y to change the time period of measurement and algorithm processing i n i n d i v i d u a l cases. Thus, i n e f f e c t , the system can "focus attention" to a s p e c i f i c area of the process p l a n t , and put a l l associated measurements and rules for that area on frequent scan. This can be done under c o n t r o l of the LISP program. Thus, for example:



F i g u r e 1.

D e s i g n f o r the LMI system f o r p r o c e s s c o n t r o l u s i n g two p a r a l l e l p r o c e s s o r s w i t h a s h a r e d memory.


5.

HAWKINSON E T A L .

A Real-Time Expert System for Process Control

73


A back-chaining diagnostic expert system could reach a point where an inference test i s required. The LISP program would t e l l the 68010 processor to "focus" on the measurements and low-level inferences required around a process u n i t . The inference could then be tested. Another use of t h i s "focus" f a c i l i t y i s to scan the plant i n a background mode, focusing attention on parts of the plant to evaluate unit process performance and detect subtle problems, u t i l i z i n g both the programmed knowledge of the the expert process operator and the expert process engineer. I t i s not p r a c t i c a l to examine an e n t i r e plant continuously with t h i s i n t e n s i t y , but the i n d i v i d u a l parts of the plant could be scanned i n a background mode. This i s equivalent t o the way a process engineer would analyze plant performance during normal plant operation. I t should be noted that the a b i l i t y to focus not only emulates the way a human expert works, but also i t avoids the problem associated with overloading the d i s t r i b u t e d process system with requests for information. While the expert system knows about a l l 20,000 measurement and alarm points i n the process environment, only those of i n t e r e s t to the expert system need be accessed. The LISP environment contains the h i g h e r - l e v e l f u n c t i o n a l i t y of the expert system. A truth-maintenance design structure i s used. The design assumption i s that lower-level i n t e l l i g e n t processing, done i n the 68010, w i l l s i g n a l p o t e n t i a l l y s i g n i f i c a n t process events. Thus, only a table of truth condition t r i g g e r s needs to be checked by the LISP programs. Some general examples of inference using the system: - detecting process problems, p a r t i c u l a r l y on complex combinations of conditions which require expertise for proper interpretation. - focus inference, i n which rules of a l l p r i o r i t i e s are activated for a unit process. In the t y p i c a l use, a possibly-significant-event (detected by a high p r i o r i t y procedure rule) would t r i g g e r a focus on the process u n i t , thus i n i t i a t i n g the gathering of information required for inference around the process unit. - diagnosis, a backward chaining inference procedure, which would be triggered by a possibly-significant-event or by operator request. Diagnosis uses the focus mechanism. An explanation i s then given of the diagnostic conclusion. Summary and Future

Extensions

V i r t u a l l y a l l tasks which require the routine a p p l i c a t i o n of human expertise, i n an organized way, are candidates for expert systems. The computer implementation of expertise has such advantages as speed, around-the-clock a v a i l a b i l i t y , and ease of expansion of the knowledge base. As such, expert systems represent the next generation of higher l e v e l software, performing tasks presently done by human operators.


ARTIFICIAL INTELLIGENCE APPLICATIONS IN CHEMISTRY


74

Expert systems have been investigated for 20 years. The implementation of expert systems i s now being undertaken on a widespread basis, due to the a v a i l a b i l i t y of hardware and software tools which a l l e v i a t e the "knowledge-engineer bottleneck", allowing cost e f f e c t i v e implementation. In a s i m i l a r way, real-time applications of expert systems require tools to allow s t r a i g h t forward implementation. We have presented a software/hardware structure which supports knowledge-base capture and real-time inference for process a p p l i c a t i o n s . In general, the LMI package (Figure 2) provides a knowledge-base structure, f a c i l i t i e s for acquiring the knowledge base i n an organized manner, and real-time c o l l e c t i o n of data with some p a r a l l e l processing of inference, and higher-level inference t o o l s . The i n d i v i d u a l applications require s p e c i f i c knowledge engineering, which i s f a c i l i t a t e d using the t o o l s we have described. The system i s currently i n s t a l l e d at Texaco and Exxon f a c i l i t i e s and i s i n p i l o t plant or laboratory t e s t i n g at seven additional sites.

CAPTURE

y

y

RULES

DIAGRAM

/

/ I/O

RTIME

/

MEMORY

Figure 2.

General structure

o f t h e LMI package.

R E C E I V E D December 17, 1985


A Real-Time Expert System for Process Control - ACS Symposium

Recommend Documents