Multi symptom condition monitoring of a critical mechanical systems as a first approach to design condition inference agent (CIA)

The paper presents the introductory results in application to multi fault condition monitoring of mechanical systems in operation, in particular internal combustion engines. This generalization to multi dimensionality and multi fault condition monitoring is possible by utilizing transformed symptom observation matrix, and by successive application of singular value decomposition (SVD) and based on it principal component analysis (PCA). On this basis one can make full ex-traction of fault related information taken from symptom observation matrix, which can be created by traditional monitoring technology. Moreover, by SVD/PCA we can create some independent fault measures and indices, and of overall system condition. In another words, full utilization of SVD/PCA enable us to pass from multi dimensional - non orthogonal symptom space, to or-thogonal generalized fault space, of much reduced dimension. This seems to be important, as it can increase the scope and the reliability of condition monitoring of critical system in operation. It enables also to maximize the amount of condition related information, and to redesign the tradi-tional condition monitoring system. At the end of the paper some introductory consideration are presented leading to a design of Condition Inference Agent (CIA), which will enable to infer in real time on condition of critical objects in operation.