Signal transforms for feature extraction from vibration signal for air compressor monitoring

Machine monitoring and fault diagnostics have become a major and a prominent area for research now days. For doing the same, there is great scope for using better signal processing tools in extracting key features from machine's acoustic and vibration data. This paper provides a brief survey and comparison of various transforms that can be performed on vibration data for extracting features. The transforms used are Fast Fourier Transform, Discrete Cosine Transform, Autocorrelation function, Convolution with Sinusoidal, Short Time Fourier Transform, Cohen's Class Distributions, S-Transform and various Wavelet Transforms. A case study of fault diagnosis was performed on an air compressor in three different states namely Healthy, Leakage Outlet Valve fault and Leakage Inlet Valve fault. The features from these transforms have been compared with respect to their precision in recognizing the three states. Results showed that instead of using a huge feature set, finding out the right transform for recognizing a certain fault could be a very good course of action.