Trend Factor Smoothing and Tasmanian Devil Optimization based Siamese Neural Network for anomaly detection in predictive maintenance

Abstract

Abstract In today’s business world, predictive maintenance is essential since it helps organizations prevent equipment breakdowns and minimize downtime. A novel technique that employs machine learning to anticipate equipment failures is anomaly detection-based predictive maintenance. This method helps maintenance teams to foresee and prevent problems by looking for patterns and anomalies in historical data. This lowers the possibility of unplanned downtime and boosts overall productivity. Using optimized deep learning, this study is intended to create an advanced deep learning model for anomaly detection in the predictive maintenance of cyber-physical systems, incorporating trend factor smoothing with Tasmanian devil optimization (TFsTDO) and siamese neural networks (SNN) to enhance detection precision and operational efficacy. The proposed TFsTDO-SNN system encompasses preprocessing through Box–Cox transformation, feature selection utilizing an innovative TFsTDO algorithm, anomaly injection via oversampling, and anomaly detection employing an SNN trained by TFsTDO. The model was assessed utilizing the CNC Mill Tool Wear dataset, with performance criteria comprising precision, recall, and F1-score. Experimental findings indicate that TFsTDO-SNN attains a precision of 96.5%, a recall of 97.2%, and an F1-score of 96.9%, surpassing traditional approaches including GA, AdaBoost, LSTM-autoencoder, and CNN-LSTM. The TFsTDO-SNN model provides a strong solution for anomaly identification in predictive maintenance, with prospects for future improvements using explainable AI and sophisticated optimization methods.