Selective decimation and analysis of oversampled data
View Patent ↗Useful and meaningful machine characteristic information may be derived through analysis of oversampled digital data collected using dynamic signal analyzers, such as vibration analyzers. Such data have generally been discarded in prior art systems. In addition to peak values and decimated values, other oversampled values are used that are associated with characteristics of the machine being monitored and the sensors and circuits that gather the data. This provides more useful information than has previously been derived from oversampled data within a sampling interval.
1. A method for avoiding distortion while processing oversampled dynamic machine vibration data in a non-linear selective decimation process, wherein the distortion contains frequency components higher than a Nyquist frequency (F s /2) that alias back into a frequency range from zero to F s /2 and thus contaminate spectral purity, the method comprising:
(a) measuring a vibration signal using one or more vibration sensors attached to a machine;
(b) sampling the vibration signal using an analog-to-digital converter at a sampling rate of F s to generate oversampled dynamic machine vibration data;
(c) up-sampling the oversampled dynamic machine vibration data at an integer up-sampling rate N using a digital signal processor that inserts N−1 zeros between adjacent data samples in the oversampled dynamic machine vibration data, thereby generating up-sampled data;
(d) low-pass filtering the up-sampled data using a low-pass filter having a cutoff frequency of
Fs
2
,
thereby generating low-pass-filtered up-sampled data having no spectral images above
Fs
2
;
(e) for a fractional up-sampling rate (N/L), where N>L and L>1, down-sampling the low-pass-filtered up-sampled data using the digital signal processor by retaining every Lth sample and discarding L−1 samples between each Lth sample, thereby generating low-pass-filtered down-sampled data free of spectral images at an up-sampling frequency of F s ×(N/L);
(f) performing the nonlinear selective decimation process on the low-pass-filtered up-sampled data using a data decimator module of the digital signal processor, thereby generating data having distortion components abased above
Fs
2
;
(g) filtering the data generated in step (f) using a low-pass filter having a cutoff frequency of
Fs
2
,
thereby generating data in which aliased components above
Fs
2
are removed; and
(h) down-sampling the data generated in step (g) using the digital signal processor by retaining every Nth sample and discarding N−1 samples between each Nth sample, thereby generating post-nonlinear-processing machine vibration data in which aliasing effects are mitigated and spectral purity is preserved.
2. The method of claim 1 wherein the non-linear selective decimation process of step (f) comprises a PeakVue process.
3. An apparatus for avoiding distortion while processing oversampled dynamic machine vibration data in a non-linear selective decimation process, wherein the distortion contains frequency components higher than a Nyquist frequency (F s /2) that alias back into a frequency range from zero to F s /2 and thus contaminate spectral purity, the apparatus comprising:
one or more vibration sensors attached to a machine for measuring a vibration signal;
an analog-to-digital converter for sampling the vibration signal at a sampling rate of F s to generate oversampled dynamic machine vibration data;
a digital signal processor for up-sampling the oversampled dynamic machine vibration data at an integer up-sampling rate N by inserting N−1 zeros between adjacent data samples in the oversampled dynamic machine vibration data, thereby generating up-sampled data;
a first low-pass filter having a cutoff frequency of
Fs
2
for low-pass filtering the up-sampled data, thereby generating low-pass-filtered up-sampled data having no spectral images above
Fs
2
;
the digital signal processor further for down-sampling the low-pass-filtered up-sampled data by retaining every Lth sample and discarding L−1 samples between each Lth sample, thereby generating low-pass-filtered down-sampled data free of spectral images at an up-sampling frequency of F s ×(N/L);
a data decimator module of the digital signal processor for performing the nonlinear selective decimation process on the low-pass filtered up-sampled data, thereby generating decimated data having distortion components aliased above
Fs
2
;
a second low-pass filter having a cutoff frequency of
Fs
2
for filtering the decimated data, thereby generating low-pass filtered decimated data in which aliased components above
Fs
2
are removed; and
the digital signal processor further for down-sampling the low-pass filtered decimated data by retaining every Nth sample and discarding N-1 samples between each Nth sample, thereby generating post-nonlinear-processing machine vibration data in which aliasing effects are mitigated and spectral purity is preserved.
4. The apparatus of claim 3 wherein the non-linear selective decimation process comprises a PeakVue process.