BEARING-FDD: An early detection and diagnosis tool for bearing faults in rotating machinery

This paper presents the design and implementation of a web tool offering an innovative method for detecting, diagnosing and classifying bearing faults in rotating machinery under limited data conditions, providing explainability and interpretability of the results obtained. The tool uses a machine learning model to detect and diagnose bearing faults. A monotonic smoothed stacked autoencoder builds a health indicator without requiring feature extraction, making the tool useful without the need for specialized staff. The tool generates explainability and interpretability reports with a correlation analysis between the health indicator and well-known engineering features and easily interpretable details on the diagnosed faults. The tool includes the option to use preloaded state-of-the-art datasets, while also allowing users to upload their own datasets to analyze vibration data from real industrial equipment. • Bearing-FDD operates effectively with limited data, as only healthy data is required for training. • Bearing-FDD eliminates the need for manual feature extraction, as the MS2AE model comprises the raw vibration data into a single health indicator used for fault classification and diagnosis. • Bearing-FDD demonstrates robustness under varying operating conditions and motor loads. • Bearing-FDD provides explainability of the health indicator obtained using well-known engineering features. • Bearing-FDD offers interpretability, making the system accessible to non-specialized staff, as diagnosed faults are clearly described. • Bearing-FDD is an open-source tool, without requiring any installation or license to work properly, as other frameworks such as Matlab.