Predictive Maintenance for Filling Machines with Online Evolving Spiking Neural Networks

Accurate anomaly detection in streaming data is of paramount interest in the field of predictive maintenance. This paper presents an approach based on Online evolving Spiking Neural Network for Unsupervised Anomaly Detection (OeSNN-UAD) for predictive maintenance, with reference to real functioning of filling machines. The novelty in this work lies in the application of OeSNN-UAD to a predictive maintenance problem. It also integrates the calculation of the Remaining Useful Life (RUL) in the model. Different anomalies in filler motors functioning were selected to compare the accuracy of the proposed approach with other machine learning algorithms. The corresponding data were collected from sensors installed on machines in operation. The experimental results show that the OeSNN-UAD model can achieve better or comparable accuracy with less computational effort.