A Data Fusion Analysis and Random Forest Learning for Enhanced Control and Failure Diagnosis in Rotating Machinery

In a heavily-subjected-to-failure field of rotating machinery, the need for accurate and reliable detection methods is paramount. This paper aims to advance fault detection capabilities through a novel approach that integrates current–temperature–vibration data fusion analysis and also by employing a Random Forest (RF) artificial intelligence methodology. An experimental investigation was conducted to acquire temperature variations, vibration signals, and electrical current readings from a rotating machine subjected to various types of faults. These faults include conditions of normal operation, bearing faults (inner and outer races), shaft misalignment, and rotor unbalance, all monitored under a constant rotating speed of 680 RPM and zero load. Four ceramic shear ICP-based accelerometers, two thermocouples, and three current transformers, were collectively used to collect the data while adhering to the International Organization for Standardization (ISO) standards. To refine the data for the RF algorithm, the standard deviation of the three datasets was calculated at specific intervals, revealing a significant enhancement in diagnostic accuracy when combinedly used. The resulted accuracies varied as follows: 45.0% accuracy using current data alone, 19.4% with temperature data, 62.0% with vibration data, and a remarkable 96.0% when using data fusion. Thus, data fusion is promising in thermal, electrical, and mechanical condition monitoring. Integrating diagnostic approaches in control systems can significantly improve the reliability of rotating machinery.