Multilayer Simulation and Expert Systems in Multi-level Process Control

Abstract A hierarchical simulation model consisting of four calculating levels, ranging from known parameters and formulae to detailed simulation models, is developed for ferroalloy processes. In order to gradually increase the amount of detail in selected parts of the process a decomposition procedure is used. Uncertainties are taken into account in the formula based nonlinear models of electrical flow. In other applications simplified fuzzy models are developed on the basis of simulation experiments by fuzzy linear regression. The models are too extensive to be modified for fuzzy calculations. Accuracy of the results in the lower level models depends totally on the input data values. Fuzziness is included in the input data, and the simulation models are not changed. Discretized values, each with its own certainty factor, are generated on the basis of a triangular distribution. In the fuzzy linear regression, the independent parameters are the design parameters of the optimization. The effects of the vague input data are transferred to the fuzzy parameters. Dependent parameters are used in decision making. The simplified simulation models are used in conjunction with a rule-based expert system. These models, corresponding to the procedural knowledge, are embedded in the rules of the knowledge base. The models are occasionally updated, which significantly reduces the vagueness of the results. In addition to simulation results, experimental results can also be utilized. The rules may also contain qualitative models. The usage of the systems on the optimization level can be classified into four types of application: feasibility studies, process design, process studies and process development. In other parts of the multilevel process control, communication with the data oriented systems, e.g. automation systems in monitoring and control and decision support systems in management and coordination appl ications, becomes important. User interface is very important in a multilayer system with a variety of properties. The computer modelling of the simulation system is carried out by using standard FORTRAN 77 and various hardware has been used in development and applications . The simulation experiments for fuzzy linear regression require a mainframe computer. The present expert system is implemented on microcomputers.