Using Optimized Time-Frequency Representations for Acoustic Quality Control of Motors

We investigate the application of time-frequency representations to acoustic quality control of electric motors. The signals exhibit a cyclostationary characteristic. Its periodicity is due to the process of noise generation. For two types of faults we compare different statistics based on first and second order properties of the data with a recently introduced method based on time-frequency representations. This procedure takes advantage of the freedom to choose a smoothing kernel by optimizing it in a data dependent manner. Since the spectrum and the local variance are special cases of the time-frequency representations the optimization procedure might also indicate that these simpler quantities are sufficient to perform the classification task. This was the case in one of the faults considered.