Data processing of experimental and computational results in machinery diagnostics

Abstract

Dynamic properties of machines influence on generated noise and vibrations, moreover they influence on service life, precision and quality of products; hence condition monitoring is a key factor in machine maintenance. They are two main methods in machinery diagnostics. The first relays mostly on experimental results, whereas the second method relays on computational results. The first approach is called symptom diagnostics, whereas the second is called holistic diagnosis. The experimentally obtained failure modes (symptoms) are functions of measured parameters. They can be non-dimensional without a physical interpretation, but they should be sensitive to condition of machines, which is a key factor. Whereas in holistic diagnostics two kinds of models are used: linear and non-linear. Linear models are wide spread, because they are relatively simple to use. Non-linearity is neglected in these models, which introduces an error, but makes computation easier. In contrast to this, non-linear models are more accurate, but their application is far more difficult. It should be mentioned that, those non-linear models have an important advantage; they can explain phenomena like multi-stability, fluctuations of measured amplitudes and jumps of amplitudes. These phenomena do not take place in linear models. A few examples are presented, to illustrate linear and non-linear models in machinery diagnostics. Moreover practical approach to signal processing is presented.