Optimization of symptom observation matrix in vibration condition monitoring

In diagnostics of complex machines, for their condition assessment we often use many dasiawould bepsila symptoms at the beginning, especially at the diagnostic startup of a new machine. The discrete observation of this dasiawould bepsila symptom vector creates so called symptom observation matrix (SOM). Using next the singular value decomposition (SVD) for the given SOM, one can extract the generalized fault symptoms, describing the fault evolution in a given case, and also diagnostic contribution of measured symptoms. Using the symptom reliability concepts further, and the grey system forecast methodology, it is possible to asses the generalized symptoms limit value. In this way one can establish the needed dimensionality of symptom observation matrix, and moreover assess the residual system life. However, doing this we have to establish new criteria for the dimensionality of SOM, based not on the number of symptoms in use, but the quality of diagnostic decision. This concept was verified in the paper using the data taken from real cases of vibration condition monitoring practice.