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

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Predictive maintenance is crucial for preventing equipment failures and minimizing downtime in modern industries.
• Anomaly detection using machine learning offers a novel approach to anticipate equipment malfunctions by identifying patterns in historical data.

Purpose of the Study:

• To develop an advanced deep learning model for anomaly detection in the predictive maintenance of cyber-physical systems.
• To enhance detection precision and operational efficacy through optimized deep learning techniques.

Main Methods:

• The study proposes the TFsTDO-SNN system, incorporating trend factor smoothing with Tasmanian devil optimization (TFsTDO) and siamese neural networks (SNN).
• Methodology includes Box-Cox transformation for preprocessing, TFsTDO for feature selection, oversampling for anomaly injection, and TFsTDO-trained SNN for anomaly detection.
• The model was evaluated on the CNC Mill Tool Wear dataset.

Main Results:

• The TFsTDO-SNN model achieved high performance metrics: 96.5% precision, 97.2% recall, and 96.9% F1-score.
• The proposed model outperformed traditional methods such as GA, AdaBoost, LSTM-autoencoder, and CNN-LSTM.
• Experimental findings confirm the model's superior detection precision and operational efficacy.

Conclusions:

• The TFsTDO-SNN model presents a robust solution for anomaly identification in predictive maintenance applications.
• The study highlights the potential for future enhancements using explainable AI and advanced optimization methods.
• This approach offers significant improvements in predicting equipment failures and optimizing maintenance strategies.