Model goodness of fit evaluation based on a fuzzy inference system in virtual commissioning

Abstract

Modern industrial plants are becoming increasingly complex, resulting in the need for rapid testing and validation of industrial
automation systems. To meet the requirements mentioned above, new simulation techniques, like virtual commissioning (VC), can be employed, as they allow for identifying process bottlenecks at the very beginning of the commissioning process. Moreover, it has also been used for maintenance operator training. The essential stage of VC is verification of the model of a commissioned plant quality – model goodness of fit. A plethora of measures are used for model goodness of fit evaluation, but each is characterized by a different range of values and interpretations. Thus, the best idea is to use the hybrid approach for model goodness of fit evaluation, combining the information from different measures. In order to create a flexible system for decision-making, if a model quality is good and sufficient to be used in VC, the Virtual-Commissioning-Model Fuzzy Coefficient (VCMF) is introduced based on the Takagi-Sugeno-Kang fuzzy-inference system. It considers knowledge of virtual commissioning
of industrial automation systems and information carried by different methods of goodness of fit evaluation (NRMSE, ME, MAE, and MIA).
VCMF was based on data from the belt conveyor, which was thoroughly analyzed. Current, velocity, and torque time series underwent the data pre-processing and analysis methods, which resulted in obtaining a model. VCMF allows for differentiating models into those that can be used in VC and those that cannot. The threshold value was defined by Gaussian Mixture Modeling and Bayesian Information Criterion.