AI-Assisted Explainable Multimodal Predictive Maintenance of Industrial Bearings Using Vibration, Acoustic Analysis, Vision-Based Fault Detection, And Intelligent Maintenance Recommendation

Abstract

The emergence of industry 4.0 and smart manufacturing is resulting in the need for intelligent predictive maintenance systems that will help in the prevention of breakdowns and decrease maintenance costs. Even though vibration-based condition monitoring is commonly applied for diagnosis of bearing failure in industrial systems, it is faced by many challenges such as being incapable of diagnosing early stage and external faults. In this work, an AI-based multimodal predictive maintenance system that utilizes vibration analysis, acoustic analysis, thermal imaging and RGB vision-based inspection for precise, timely and non-invasive detection of faults is proposed. The performance of the hybrid CNN–LSTM model applied in the experiment was compared to SVM, Random Forest (RF), XGBoost, and traditional CNN. Contrary to conventional prediction maintenance methods which are focused on detection of faults, the proposed system utilizes an AI-assisted decision support system that will help in identifying the cause of the fault, the fault severity, and recommend appropriate maintenance strategies. The developed architecture had the accuracy of 98.2% and surpassed each of the individual sensors and traditional machine learning approaches. Vision-based AI algorithm identified the presence of thermal hot spots, lubricant leaks, bearing housing cracks, misalignment signals, and surface corrosion without taking apart the equipment while acoustic sensing increased the identification of early faults that had poor vibration signals. In general, the developed architecture is an intelligent approach to predictive maintenance using multiple sensors coupled with AI-based decision support system.