Rotodynamics Multi-Fault Diagnosis through Time Domain Parameter Analysis with MLP: A Comprehensive Study

Rotary machines are crucial to various industrial and renewable energy applications such as wind turbines, and their efficient operation is vital for maintaining productivity of energy and minimizing downtime. Bearings, as integral components of rotary machines, are particularly prone to faults, which can lead to inefficiencies or even catastrophic failures. Detecting and diagnosing faults in bearings in real-time is therefore essential for ensuring the reliability and longevity of rotating systems. This investigation explores the application of machine learning technique for fault detection of misalignment and unbalance cases in rotary machines, with a specific focus on vibration analysis. The dataset utilized, COMFAULDA, comprises vibration signals recorded under various operating conditions, including normal operation and different fault scenarios. Employing a multi-layer perceptron (MLP) architecture, the analysis of vibration signals from rotary machines centers on multi-fault detection. The results demonstrate the effectiveness of this approach across various operating conditions, highlighting specific channels such as the tachometer (channel 2) and accelerometer in horizontal direction (channel 8) signals. Techniques such as root mean square, peak-to-peak amplitude, shape factor, variance, envelope mean, standard deviation and Wilson amplitude prove particularly effective in detecting anomalies. The study underscores the potential of machine learning and signal analysis for predictive maintenance in industrial applications, emphasizing a focus on low computational cost achieving 99% for analyzing 7 different faults. Ultimately, the aim is to improve bearing health monitoring and reduce downtime in industrial systems.