Anomaly detection in energy consumption for Condition-Based maintenance of Compressed Air Generation systems: an approach based on artificial neural networks

The topic of early detection of faults has great relevance for the implementation of more rational and efficient management in industrial plants. Since most energy systems and machines show a degradation in energy performance when a failure is starting to develop, monitoring this parameter can result in early behavioral anomalies identification and consequently in the implementation of more timely countermeasures. As the presence of energy consumption monitoring systems is becoming increasingly more widespread, the use of energy consumption monitoring as a way to detect failures is quite interesting because it can enable the introduction of early detection methods for a broad spectrum of failures without the need of specific instrumentation. Furthermore, the promotion of energy performance is also useful as a quality indicator of the process itself. Thus, this paper describes a methodology to monitor the energy consumption performance of a compressed air generation system in an industrial plant to detect anomalies using artificial neural networks. The use of ANNs allows an accurate characterization of the system in a healthy state. Then, by comparing the model prediction and the actual energy consumption, residuals are calculated and plotted in a control chart. The setting of limits and frequency of control enables the management to decide the sensitivity of the control system itself.