Predictive maintenance with digital model

Abstract

This paper addresses the development of a Predictive Maintenance (PdM) detection algorithm applied to a digital model intended for PdM purposes of a heat exchanger station. The detection algorithm has been designed utilising data from a PLC measurement programme in conjunction with a digital model developed using MATLAB Simulink. Machine Learning (ML) techniques, specifically Support Vector Machines (SVM), were employed like two class classificator to identify anomalies. The SVM algorithm classified the measurement points into fault and normal operating states based on modelled temperature values, the Root Mean Square Error (RMSE) of temperatures within the primary circuit. The normal operating states is defined by digital model introduced in [4]. Anomaly state is simulated by serial clogging valve V3 in primary circuit.