1

7.1 Knowledge-based supervision: selected issues

Intelligent process supervision: basic issues

fusion, generalisation, etc., situation assessment, modelling the analysed system at the knowledge level, monitoring its qualitative behaviour, semantic evaluation of system performance, fault detection, diagnosis, decision support, etc.

Assessing the need for intelligent process supervision

The need for intelligent process supervision is a consequence of arriving at the limits of reasonable use and interpretation of large amounts of numerical data obtained during process monitoring on one hand, and the increasing requirements for knowledge generation, analysis and utilisation in achieving quality and safety of computer controlled systems.

The following factors are decisive for the need of knowledge-based methods in supervision:

new types of data can be acquired, specified, used; numerical data consists only part of the data accessible,
symbolic and qualitative knowledge enables analysis at higher levels of abstraction,
complex relations cannot be expressed with simple numerical conditions,
complex activities require complex tools (e.g. situation recognition, diagnosis),
man-machine communication at symbolic level is more efficient and comfortable for human operators.

What is knowledge-based supervision

Finally, let us approach the answer what is knowledge-based supervision. The answer is mostly descriptive, i.e. it is attempted by listing the necessary requirements and possible tasks to be accomplished. In particular, we shall refer to the idea of supervision in automatic control, however, contrary to some of the approaches we do not speak about supervisory control as, e.g. in [Bernard and Williams, 1995], [Lane et al, 1995], [Stock 1989], or [Martinez, 1997], but about general task of supervision, assuming higher-level control to be one of the possible tasks only.

Knowledge-based supervision of dynamic process (usually a complex one) is a continuous process constituting an upper-level activity with the ultimate goal to assure that ''everything goes well''. It consists of collecting the data from the object, performing data-to-knowledge transformation, processing the knowledge with use of knowledge engineering (AI) methods, and possibly, applying the resulting knowledge to control and decision support for the process in order to assure better performance, safety, reliability, traceability, and to obtain better knowledge about the process itself. This can be achieved through direct reaction of the system (reactive systems), decision aid to process operator, reports on, tracing and analysis of process history, and, finally, aid in improvement and redesign of the system. Hence, some minimal requirements of knowledge-based supervision should include:

in-line architecture, or on-line real-time architecture if supervisory control is also one of the assumed tasks,
data acquisition and abstraction in order to generate qualitative or symbolic knowledge; this implies the use of symbolic knowledge representation at the resulting level,
knowledge processing with use of knowledge engineering methods; this implies that symbolic inference methods will be applied and knowledge-based models will be used to model the system properties,
finally, the results, if they take the form of control action or decisions, may be applied to control of the systems or dealing with certain problems (e.g. diagnosis).

The basic concept of knowledge-based supervision consists in continuous evaluation of the current state and the trajectory of states with specific KB methods in order to work out the required output. The list of possible tasks for knowledge-based supervision is abundant. Some of the goals may include, but certainly are not limited to, to following issues:

knowledge level tracing of system behaviour over time, which provides high-level semantic information to human operators and can be used in qualitative state assessment and multi-purpose on-line and off-line analysis,
analysis of data correctness, including indirect checking with other data in case of data redundancy, direct checking of satisfaction of domain constraints and possibly, model-based checking with use of model for data generation,
generation, storage, and report generation of knowledge-based traces for analysis of system performance,
intelligent process monitoring, by on-line analysis of the current state of the system and its evolution, state evaluation, classification, and processing,
higher level control and decision support, both on-line and of line,
on-line state assessment, checking of imposed constraints, alarm generation and reactive decision making,
model tracing and adaptation,
on-line and of line diagnosis (fault detection and isolation),
reliability and safety verification and improvement.

Finally, the need and role of knowledge-based model in supervision can be summarised as follows: the model plus current observations are used to situation assignment, inferring knowledge and finally producing decisions. At all the stages different part of the domain knowledge forming the model can be used.