7.7 An integrated environment for CASSD

Object-variables, embedding data and methods, implementing numeric to qualitative interfaces, based on abstraction tools, for dealing with ALCMEN qualitative relations and CEES expert rules, have been implemented by means of Simulink block-based graphical representation. As a result, MATLAB/Simulink, the numeric CACSD framework, has been improved to support expert supervisory strategies design becoming a CASSD framework.

Architecture and tools.

The previous described tools, abstractors, ALCMEN and SimCEES, have been implemented in Simulink under an object oriented approach to facilitate the task of designing and testing supervisory structures, based on expert knowledge [Melendez et al., 1996b][Melendez et al., 1997a]. Object-variables have been defined to encapsulate information from process variables at different levels of abstraction. Then, abstractors also called abstraction tools, are encapsulated in object-variables to obtain a significant information from those variables related to the process. This information is encapsulated using both numerical and qualitative representations in the object-variables. This data can be directly supplied to a modular ES, SimCEES, by building facts that join all the information encapsulated in object-variables and subjective appreciation about certainty of this information with methods needed for the inference engine of ES. Parameters are supplied using a dialogue window.

 

Fig. .29 Mechanisms for obtaining significant information involved in expert reasoning.

 
Because not all expert knowledge concerns numerical information, ALCMEN is added to deal with qualitative information, allowing the designer to implement simple relations in order to obtain a rough evolution of not measurable process variables. In the implementation of ALCMEN blocks, object-variables structure have been used, despite of only qualitative fields are used and abstraction tools are only used in the interface blocks (filtering), to preserve the structure and encapsulation of information used with numerical data.

Fig. 5.29 shows how those tools are involved in the procedure of obtaining significant information to be supplied to the ES. The heterogeneous kind of information merged to reason about complex systems, needs the object-oriented approach to be used as integration mechanism. In this case, it has been used in the representation of information flow (object-variables and facts) and also to build a modular ES able to deal with this encapsulated information.

Fig. .30 Set of Simulink ToolBoxes of the proposed CASSD framework.
Abstraction tools, ALCMEN and CEES are available.

The previous schema is implemented as sets of blocks grouped in ToolBoxes for each tool (CEES, ALCMEN and Abstraction Tools) forming the complete CASSD framework.. The use of this CASSD framework for designing supervisory structures allows partial validations of the structure when working with complex systems. Simulink capabilities allow to group set of blocks into a new one simplifying the appearance and allowing knowledge encapsulation. At the same time connectivity between tools is quickly solved by tracing lines between inputs and outputs of blocks.

Knowledge representation.

Processing and administration of expert knowledge requires of knowledge formalising and structuring. Rule-based programming is not the only technique used with this goal, within artificially intelligent programs, but nowadays, it is the more adequate technique for us to directly codify expertise. In spite of this, some other classical formalism are more or less present in the proposed framework taking in account the particular point of view of process supervision domain. Procedures typically used to represent objects and their relationship are used to represent process variables :

• Production rules, for describing knowledge in the form of "IF...THEN..." rules are present in CEES. Knowledge base associated to any ES block is described as sets of IF-THE rules, as showing Fig. 5.31.

Fig. .31 SimCEES rules

• Semantic networks : Graphic representation of knowledge based on objects and their relationship can be compared with the block-based graphical distribution of knowledge provided by the graphical user interface of Simulink. Interconnection of blocks represent dependencies between process variables and the flow of information. This is the case of ALCMEN relations or dependencies between several ES blocks.
• Frames, are thought to be data structures for representing objects. In this work the object-oriented approach has been used for representing process variables and information related to process behaviour. All of these information is embedded into object-variables.


The modular conception of all those tools presented to help supervisory systems design, supported by Simulink blocks, simplifies modular design of reasoning systems. Simplicity and modularity, which combines quantitative and qualitative knowledge, are important advantages when building on-line reasoning systems that are embedded in conventional applications. According to the degree of abstraction of knowledge about process variables the tools presented in the previous sections are especially conceived to assist in three hierarchical levels (Fig. 5.32):
 

• Level of Signal : This is the interface level between the supervisory structure and the process. It is the acquisition level where measures from process and process engineers knowledge join together to supply other levels with significant information extracted from measures. The tools used to obtain qualitative information from measured signals are the abstraction tools or abstractors and their output are a qualitative representation of the trends of process variables (tendencies, oscillation degrees, alarms, degree of transient states, ...) needed not only in supervision, but also in fault detection and diagnosis tasks. The use of object-variables for this purpose is similar to a representation based on frames.
• Level of Qualitative Relations : At this level is where engineers could represent dependencies or relations between qualitative variables, obtained or not from the first level. These relations could be used to deduce roughly inaccessible dynamics according to a qualitative model, physics or just observed dependencies between variables that are difficult to join in a numeric equation. ALCMEN is used to perform these tasks (and knowledge is represented by semantic networks).
• Level of Rules : This level reflects dependencies between facts that could be represented by rules describing expert KBs. Thus, ES are used at this level. This is the final destination of inferior levels.

Fig. .32 Levels of abstraction

Each level is able to deal with significant information coming from equal and inferior levels (See "Fig. 5.32") . The mechanisms used in the representation of knowledge at each level are associated with the respective tools, described in previous subsections. These mechanisms, are always implemented using object-oriented programming and encapsulated as blocks in a graphical representation supported by the Simulink graphical user interface. Knowledge modularisation and encapsulation is represented in Fig. 5.33.

Row numerical data coming from process (measures or estimations) are encapsulated into object-variables together with abstraction tools used to obtain qualitative descriptions of process variables based on significant information. Qualitative description of those signals are stored in temporal windows to be used in reasoning task by other blocks. Finally they, can be supplied to ES as facts encapsulating additional information.