Semiphysical Models of a Hydraulic Servo Axis for Fault Detection

This paper is concerned with the model-based fault detection and diagnosis of hydraulic servo axes. For the application of model-based methods, a very detailed model of the plant must be derived. Such a model can be derived by means of physical equations (i.e. white-box models) or by data-driven methods (so-termed black box models). The paper at hand tries to combine the best of both worlds by applying the LOLIMOT neural net and deriving semi-physical models. Since only physical relationships but no physical laws must be known, the model structure can be set up easily. On the contrary, the data driven modeling can be expedited since a model structure can be assumed. In this paper, models will be derived for the pressure supply of the hydraulic servo axis as well as the valve and cylinder unit. Based on these models, parity equations will be compiled, which in turn will be used for fault detection and diagnosis. Measurements at a testbed verify these results.