Monitoring System for Diesel Generator Operation Condition via Sound Signal

The Antarctic Scientific Research Support Platform provides energy guarantee for automatic astronomical observation and other scientific exploration equipment in Antarctica. The energy of the platform mainly comes from diesel generators. Therefore, it is very important and urgent to monitor the operating condition of diesel generators. This paper presents a generator operation condition monitoring system based on Variational Mode Decomposition - Fast Fourier Transform (VMD-FFT) algorithm and One-Class Support Vector Machine-Back Propagation neural network (OCSVM-BP) cascaded classifier. First, aiming at the problem of feature extraction of audio signal, VMD-FFT algorithm is proposed to perform noise reduction and feature extraction on the original audio signal. Then, aiming at the problem of insufficient types and number of negative samples, a cascaded classifier based on OCSVM-BP is designed. Experiments show that the features obtained by VMD-FFT can well represent the feature of the original audio data. The OCSVM-BP cascaded classifier based on polynomial kernel can effectively identify faults and different operating conditions.