Test Ideas: Find vibrations in defective rotating parts

Data-acquisition systems need signal conditioning to prevent unwanted signals from masking important data.

By Jim Axelson, Bauer Controls -- Test & Measurement World, 7/1/2010 12:00:00 AM

Have a test idea you'd like to share? Publish it here and receive $150. Send us your ideas. Also: Read other Test Ideas Read Design Ideas published by EDN

Identifying gear, shaft, or bearing quality problems in rotating systems such as automobile transmissions typically requires laser vibrometers, microphones, or accelerometers. During a test, these sensors produce analog signals that can represent defects as the mechanical system rotates. Each rotating part transfers energy, or vibrates, at a frequency associated with its rotating speed and its physical characteristics. The number of teeth, number of vanes on a pump, bearing dimensions, and gear-set design all contribute to the vibration characteristics of the rotating component.
Defective components produce a set of energies, or vibrations, different from those generated by good parts. Some of these defects appear at certain speed and torque conditions only. Thus, tests are performed while the speed or torque is ramped very quickly, typically from several hundred rpms to several thousand rpms within a few seconds, throughout the assembly's operating range.

Figure 1. A low-pass filter removes noise from an accelerometer's signal before a data-acquisition card digitizes the signal.

To eliminate measurement errors, you must collect and analyze data in a way that eliminates errors by aliasing, which produces distortion and unwanted frequency components. To remove aliasing, the signal needs filtering to reduce its bandwidth to a frequency range lower than the Nyquist frequency, which is defined as one-half the data-acquisition sampling rate.

Figure 2. This plot shows the intensity of the energy relative to rotational speed and order. Courtesy of Bauer Controls.

You can overcome this problem by using a position-based data-acquisition system with a low-pass anti-aliasing filter. Figure 1 shows the measurement system. The accelerometer signal consists of frequency components that contain low frequencies and other irrelevant content (two sine waves in the upper left box). The system uses an Alligator Technologies anti-aliasing filter, a National Instruments data-acquisition card, and a National Instruments FPGA (field-programmable gate array) card. The filter dynamically adjusts its cutoff frequency as the transmission's rotational speed changes under software control.

The speed signal—a TTL-level pulse train—passes through the FPGA that multiplies the signal's frequency by 100 so it can become the input to the programmable low-pass filter. The filter limits the accelerometer signal's bandwidth, keeping it below the Nyquist frequency.  The filtered signal goes to a National Instruments' analog input card (box in the upper right). As the speed of the system changes, the pass bandwidth changes dynamically, eliminating data distortion.

After filtering and digitizing the signal, the system converts the data from the raw time-based domain into the frequency domain through FFTs (fast Fourier transforms). The FFT shows the energy generated by the gears, bearings, and shafts making up the physical system. Because the speed is known, the software can divide all the frequencies of the FFTs by the speed, which converts frequency to "order." Order is the number of times a vibration is produced per revolution of the tracking shaft. The equation is:

Order = Frequency / Speed

where Speed is revolutions/s.

Figure 3. Peaks in the white areas indicate failed parts.

The spectral plot of order will be the same regardless of rotation speed. For example, a rotating 60-tooth gear will always produce a 60th order vibration. In Figure 1, each trigger pulse initiates a signal sample.

Figure 2 shows the rotational speed (vertical scale) and the order (horizontal scale). At each speed interval, say every 50 rpm, the system calculates an FFT. It then produces an intensity plot consisting of overlaid FFT "slices" much in the way oscilloscopes use color to differentiate the frequency in which time-domain waveforms occur. You can also view the FFTs individually in a waterfall plot. The plot also produces a "slice" for each rotational speed. Seen from a speed perspective, a speed slice shows the energy of each order at that speed. Seen from the order perspective, an order slice shows the speed at which the energy for that order is the greatest. Each physical component (gear, shaft, or bearing) correlates to a unique order slice.

By analyzing the data, the system can identify components that produce abnormal energy levels, which signify defects. A test engineer evaluates the data to find the condition where the energy or vibration is most pronounced.

The production test is optimized by varying the speed and torque around the condition that produces the most vibrations. Figure 3 shows the frequency peaks that correspond to problem parts.

Because the amplitude of the slices may vary only slightly from one UUT (unit under test) to another, we use templates to differentiate between good and defective assemblies. The template shape and limits are statistically created from a sampling of known-good assemblies.

Figure 4. This part failed because the accelerometer signal at just over 700 rpm is too high.

For speed slices, we developed tests of the energy content at orders corresponding to the physical components within the assembly. We established combinations of statistically based limits and fixed limits to measure quality. Violation of a limit easily distinguishes the good assembly from the bad. In Figure 3, all of the part's energy is within the upper limit, thus all the test bands remain green.

The test setup also uses this method to test order slices. Figure 4 shows the order for a gear mesh from a rejected assembly. You can easily recognize at which speed the noise level exceeded the limit compared to the good assemblies. The differences between assemblies might be slight, but the templates accurately separate normal from abnormal assemblies.