RECOVERY OF IMPACT SIGNATURES IN MACHINE STRUCTURES

Reciprocating machines are difficult to diagnose using traditional frequency domain techniques because they generate predominantly transient vibrations which have a broad frequency content. We have found that transient vibrations, like those generated by valve impacts in a reciprocating compressor, are best analysed in the time-domain. Unfortunately, by the time the vibrations reach the surface of the machine where we can measure them non-invasively, reverberation and dispersion have disfigured them so that they look nothing like their originating forces. In order to convert them into a useful form, where they can be characterized in terms of timing and strength on a cycle-by-cycle basis, the vibration signal can be compressed by filtering it with the inverse of the structural transfer function. This is a straightforward process if the exact transfer function is known; however, in practice we expect to have simply a 'typical' transfer function from a nominally identical machine. Additional transfer function variability results from changes in machine operating conditions such as temperature and load. The inverse-filtering process can be made robust to transfer function variability through a combination of cepstral-smoothing and minimum-phase processing. In addition, if a cepstral comb window is incorporated into the signal processing scheme, multiple impact signatures can be removed. (C) 1995 Academic Press Limited