A technique for the vibration signal analysis in vehicle diagnostics

作者：Puchalski Andrzej^*

来源：Mechanical Systems and Signal Processing, 2015, 56-57: 173-180.

DOI：10.1016/j.ymssp.2014.11.007

• Vibrodiagnostics
• Empirical models
• Statistical feature vector
• Singular value decomposition
• Features classification

The method of utilising signals of vibration acceleration in the on-line and off-line diagnostics of mechanical defects of internal combustion engines is presented in the paper. The monitored vibration signals of the spark ignition (SI) engine in various maintenance states of the valve system were investigated. <br xmlns:set="http://exslt.org/sets">The suggested technique is based on mathematical methods of the lower triangular-orthogonal (LQ) factorisation and the singular value decomposition (SVD) of observation subspaces computed on a vibration time series after their angular resampling without any transformations in the frequency domain. The applied algorithm of data processing filters excessive information and allows the selection of diagnostic features (essential from the maintenance point of view) and generates the empirical model and matrix residuals assessed in the no-fault state as being 'zero'. Then, statistical feature vectors, for which the averaged successive singular values of the residuals of the observation subspaces of the vibration signals were assumed as components, were analysed. As a result of this procedure the vectors of lower dimensions reduced to components, allowing the classification of observations within all defined classes, were obtained. On the basis of these vectors a scalar measure - sensitive to the kind of defect - was proposed and verified.

• 出版日期2015-5