The transition toward high-density medical diagnostic hardware has created a critical dependency on cooling system integrity. Traditional hardware troubleshooting often relies on internal thermal sensors and RPM telemetry, which are inherently reactive—alerting operators only after mechanical failure has initiated thermal runaway. This research proposes an unsupervised, non-invasive framework for proactive hardware troubleshooting using Acoustic Side-Channel Analysis (ASCA). By converting raw acoustic emissions from cooling fans and power supply units into Mel-Spectrograms, we utilize a Deep Convolutional Neural Network (CNN) to identify "pre-failure" acoustic signatures. Validated on the MIMII (Malfunctioning Industrial Machine Investigation and Inspection) dataset, the proposed model identifies mechanical fatigue, such as bearing dry-out and impeller imbalance, with a 96.4% accuracy rate. This approach provides a significant lead-time of 48–72 hours over traditional reactive alerts, ensuring uninterrupted operational uptime for critical medical imaging pipelines.
Mustaquim Ahmad (Thu,) studied this question.