A Robust Data-Driven Predictive Maintenance Framework for Industrial Machinery using Explainable Machine Learning Techniques

The best performance and longevity of industrial machinery are dependent on predictive maintenance. The proposed framework is a reliable and effective predictive maintenance tool that makes use of the widely used NASA-bearing dataset. It uses advanced machine learning algorithms in a multi-step procedure that includes data preparation, anomaly identification, and fault categorization. Then we tried to learn the underlying patterns in the data using an autoencoder, a potent unsupervised learning technique, and then set a threshold to detect anomalies. Anomalies are defined as deviations from the reconstructed signal, which the autoencoder is trained to recreate from the input data. Next, we employed k-shape clustering, a state-of-the-art time-series clustering method, to group similar patterns together in the latent representation obtained from the autoencoder. This enables us to obtain more accurate anomaly labels and improve fault classification accuracy. To further enhance the accuracy of the framework, we extracted several features from the data, including the poisson process, latent representation, frequency, and time domain features. These features are then used as inputs to a decision tree classifier, a popular machine-learning algorithm that can handle both categorical and numerical data. Additionally, we experimented with the gradient boost classifier, a boosting algorithm that improves the performance of the decision tree classifier. The experimental results demonstrate that our proposed framework achieves an accuracy of 95 percent using the gradient boost classifier, outperforming traditional approaches. The combination of the autoencoder-based anomaly detection and the decision tree classifier with the extracted features provides a powerful predictive maintenance framework that can improve the reliability and performance of machinery, reduce downtime and maintenance costs, and ultimately enhance overall productivity in various industrial settings.