Removing Resonances in Automotive Crash Test Instrumentation

Abstract

Unwanted resonances can make analysis of crash instrumentation extremely difficult. These resonances are a natural part of the acceleration measurements and in many cases are allowed to be present to maintain the needed high-frequency responses. Crystal accelerometers are, for instance, essentially undamped, and have resonance humps 40 dB above unity in order to maintain a flat response to approximately one-half the resonance or natural frequency of the accelerometer. This resonance also allows the phase angle response to be close to zero well out towards the natural frequency. Additional resonance problems exist in the mounting brackets, or as frame resonances which carry little or no information. The process of removing them, however, can produce extreme errors in both amplitude and phase. The SAE J211a Recommended Practice recommends four channel classes for impact tests: c1asses 60, 180, 600, and 1000. The frequency response is flat to +1/2, -1 dB to these frequencies, and to +1, -4 dB to the break frequencies of 100, 300, 1000, and 1650 Hertz. The filter rolloff is nominally 12 dB/octave (second order) from these break points. Second order filters are normally used for impact tests since accelerometers have second order response. This filtering will provide satisfactory results, if no resonance is present at less than several times the class frequency. Often the next lower class is used to remove a resonance, a step that may cause large errors. A better solution is to follow the typical class filter with a sharper cutoff filter that will remove the resonance without affecting the amplitude and phase of the initial impulse. A method that determines when data is lost by excessive filtering is presented and demonstrated on two crashes. One crash has a resonance about 15 times higher than the class; one is less than 3 times higher.