The evolution of generalized fault symptoms and fault intensities as indicators of observation redundancy and coming system breakdown

Application of SVD to fault extraction from the machine symptom observation matrix (SOM) seems to be validated enough, especially by data taken from many real diagnostic cases. However, decomposition has two sets of components, singular vectors, and singular vales. The first component we obtain directly as the lifetime discrete function and it has direct diagnostic meaning in condition monitoring. The second component has not so direct interpretation but with some software update one can see how singular value evolves along the system lifetime. Strangely, it is a good indicator of observation redundancy, and it is the measure of generalized fault intensity. More importantly, this measure is not sensitive to the changing condition of machine work, like working load, and we do not need to filter our observation or generalized symptoms in any way. This seems to be the most important conclusion of this paper, but needs more validation.

► Machine with multidimensional fault space assessed by creating symptoms observation matrix.
► Singular value decomposition used to extract diagnostic information from it, by special software.
► New descriptors, generalized fault symptom and singular value, immune for the machine load change.