A PREDICTIVE OPERATING CONTROL SYSTEM BASED ON DATA DRIVEN BAYESIAN NETWORKS

This paper reports a first step towards the implementation of a digital twin of an upper tier Seveso plant, which can predict the behavior of the system (failures, risks, malfunctions, errors) in order to operate effectively in safety. The system, based on machine learning algorithms and Bayesian reasoning, learns continuously from the data provided by the physical system. From the operational experience of the coastal storage facility, it is clear how most of the accidental events are due to a wrong arrangement of the valves, to abnormal transfer pressures, to pump failures and pipe deterioration. This paper is focused on building an operational management system, based on the operational instruction, suitable to predict operational errors and accordingly avoiding them and thus protecting asset integrity and improve aging management.