Acoustic Sensing for Anomaly Detection: Spectral-Temporal Anomaly Detection for Remote, High-Altitude Hydro-Power Plants

In the context of industrial factories and energy producers, unplanned outages are costly and difficult to service. This thesis explores the application of acoustic sensing for anomaly detection in predictive maintenance (PdM) systems, with a focus on remote high-altitude hydropower plants. The thesis addresses the challenges of monitoring legacy industrial machinery in harsh environments by proposing a holistic hybrid framework that integrates spectral-temporal analysis with Machine Learning (ML). The framework consists of three complementary stages. (1) A rigorous exploratory data analysis (EDA) module that denoises, normalizes, and segments audio recordings before extracting timeand frequency- domain insights (waveforms, FFT/STFT, Mel-spectrograms, MFCCs, wavelet coefficients, and high-level spectral statistics).