High Voltage Circuit Breaker Fault Detection Using Vibration and Acoustic Signals Analysis and Machine Learning

Effective maintenance is essential for power plants to ensure consistent electricity generation and overall system stability. Circuit breakers play a key role in the power system and require plant shutdowns for fault inspection, even though most operate under normal conditions. This leads to unnecessary downtime and energy loss. To address this issue, this study proposes an alternative condition-monitoring approach using vibration and acoustic signals recorded during circuit breaker operations. The aim is to assess the feasibility of determining circuit breaker health status through signal analysis combined with machine learning techniques without unnecessary downtime. The collected signals are processed using various techniques, including signal decomposition, time–frequency transformation, and normalization. Dimension reduction methods are also applied to simplify the data while preserving essential features. These preprocessed signals are then used to train a deep neural network model. To further optimize the model's performance, an Adaptive Tabu Search algorithm is integrated into the training process, enabling more accurate learning of fault patterns. The resulting model estimates the probability of circuit breaker degradation and effectively distinguishes between normal and fault conditions without interrupting plant operations. This approach offers a promising solution for reducing maintenance costs and minimizing power generation losses.