64 G. Colford et al. Fig. 6.4 Remote sensing schematic 6.4 Data Collection and Signal Processing Key frequencies that would show separation of features between machine operating conditions are dependent on the type of fault. In the case of the combination bearing fault as provided in the fault simulator kit, new peaks appear at the inner and outer ball pass frequencies. The geometry of the provided combination fault bearing was used to calculate possible fault frequencies (Eq. 6.1). In this investigation, the ball pass inner frequency (BPIF) of 279.7 Hz showed the most significant difference between the healthy and faulty operating conditions. A frequency range of 278–282 Hz was selected to calculate features. Ball Pass Inner Frequecy = n 2 ∗ 1− BD PD cos(θ) ∗f (6.1) Where n is the number of balls, BD and PD are the ball and pitch diameter, respectively, theta is the contact angle, and f is the mechanical fundamental frequency – in our case the motor speed of 3400 RPM (or 56.6 Hz). Once the frequency range of interest for separation was identified in the initial data sets, the recorded time data was downsampled by a factor of 20. A Fourier transform was performed on the down-sampled data, preserving a frequency resolution of 0.0056 Hz but reducing the Nyquist frequency down to 1250 Hz. Autopower spectra were calculated for each 3-minute set within a test. The ten sets of autopower spectra were averaged together to produce one column of data for each sensor in a single test. This was repeated for all healthy, faulty, and motor off tests at local, 7 m, and 16 m from the MFS. Even though tests were recorded in a healthy/faulty/motor off order, some variation was noticeable between each 30-minute test. Averaging of the autopower spectra was used to attenuate background noise and transients that may have appeared within a set of data, but nowhere else in the test. 6.5 Data Analysis 6.5.1 Feature Selection An investigation into features that exhibited high variation between the healthy and faulty operating state of the induction motor was important for (1) maximizing the observability of fault features and (2) reducing the feature space required for classification. In this work, 70–80 data points were collected at three different locations for each operating state: healthy, faulty, and motor off. Here, a single high-performing feature is selected to demonstrate the diminishment of the observability
RkJQdWJsaXNoZXIy MTMzNzEzMQ==