Original Date: 08/08/1994
Revision Date: 01/18/2007
Best Practice : Data Analysis
To accommodate the large volume of data collected during vehicle testing, CSTA has developed software for anomaly analysis in near real time. The Vibration Expert System Analyzer (VESA) is a series of C programs that performs the quick look function and recommends continuing or halting the test to correct a problem with the data acquisition system.
During vehicle testing on the test tracks, CSTA instruments the vehicle to measure vibration. The transducer signals run through a signal conditioner and are converted from analog to digital signals which are then transmitted to a receiver. The data is transmitted from the test vehicle (see graphic) to eliminate the need for a second test collection vehicle. Thirty to sixty channels of data are typically collected during each test, and each channel is sampled at a nominal rate of 2000 samples per second. These are typical rates although additional capability exists. Each project contains 600 million to one billion data points with dozens of projects per year. The data is verified onsite using "frmer" and "amdst." Frame errors (data greater than 98% of full scale), wild points (sudden change in adjacent data points), and DC shifts (shifts of greater than 25 counts for 1/6 second) are detected using "frmer," and acceleration amplitude distribution is verified with "amdst." The data is further analyzed postsite for time domain verification and frequency domain verification.
VESA examines the output file which includes the total number of frame errors on all channels, determining if the file is editable or not (contains more than ten errors), identifying the largest track interval that contains no frame errors, and identifying any channel that is noneditable. A histogram error checking verification is performed to inspect for items such as missing data, noisy data, onesided data, clipped signal data, discontinuities in the data, and insufficient resolution. A recommendation to proceed with testing is made by VESA based on this analysis.
The postsite time domain verification uses "statn" to check for stationarity or time invariance; "kurst" to compute the minimum, maximum, mean, standard deviation kurtosis, and the coefficient of variation of the peak and rms values of each data channel; and "skewk" to compute the weighted moments or weighted excursions about the mean. The postsite frequency domain verification is performed using "psdfl" to compute the linear average, standard deviation, and peak spectra over the length of the run, "sigmu" which calculates spectral scatter during the averaging process, and "psgmu" which plots the coefficient of variation as a function of frequency for each data channel.
By using these tools that were developed at CSTA, problems with invalid data are eliminated; problems during the testing are detected and can be corrected before proceeding with the testing; the analysis of the data can be performed in short periods of time; and the data collected can also be used to develop the testing specifications for similar vehicles in one and a half days instead of the normal six months.
For more information see the
Point of Contact for this survey.
