Predictive Maintenance in Fleet Management: Analyzing Audio Data for Hazard Detection

Abstract

    Audio-based predictive maintenance is a fleet maintenance paradigm based on sound signatures for identifying mechanical faults before failure. The paper explains why sound analysis is better than preventive and reaction-based maintenance strategies since it makes early hazard identification in motorcar systems possible. Experiments have established sound-based monitoring for identifying incipient faults in mechanical systems at a level of 87% compared to vibration analysis at 62%. It consists of installing sensors in key locations, advanced signal processing, and machine learning for generating sound fingerprints and deviation identification. Cost savings are in maintenance at 25-30%, breakdown at 70-75%, and time loss at 35-45%. Areas for future research are advanced signal processing techniques for signal-to-noise ratio improvement at 400%, real-time IoT-based monitoring, and multiple modes involving acoustics and vibration analysis, temperature sensing, and operability telemetry for better diagnostic capability at 27-38% compared to monode solutions.