2026
Book
978-3-032-02716-0 (978-3-032-02717-7)
Industrial companies pay great attention to the quality of their products. This requires the implementation of state-of-the-art monitoring and optimization tools to organize the production process in such a way to obtain products of the required quality. Special attention is given to fault detection and diagnosis since faults threaten the stability and productivity of the processes. Therefore, the objective of this research was to develop a soft sensor to monitor the quality of manufactured products in real time and to effectively diagnose faults in the presence of strong interrelationships between process parameters. This chapter proposes a computational procedure for fault detection and classification based on the soft sensor predictions. During operation, its predictions could indicate significant changes in the technological process. A Hampel identifier is used to detect abnormal behavior of the soft sensor prediction. Causal analysis of process variables for the identified deviations is performed based on Shapley values. The presented approach is validated for propylene production. From the results obtained, it can be concluded that the cause of the abnormal behavior of the soft sensor prediction is mostly influenced by the two technological parameters. In some cases, there is a need for process corrective actions.