Model Adjustment and Multi-Model Based Fault Diagnosis for Hydraulic Servo Axis

This paper presents a model adjustment and a multi-model based fault diagnosis approach. Both methods are based on the same idea. A physical model of the process is altered such that it mimics the behavior of the process in the presence of certain faults. A number of these modified models, each governing a different fault condition, are evaluated and the model with the smallest output error is determined. As this model is assumed to best govern the current process dynamics, it can be used to diagnose the actual state of the process. Two variations of this idea are presented, tailored specifically to online and offline operation respectively. For online applications, multiple models with fixed parameters are evaluated in parallel, whereas for offline application, an optimization approach is employed. Here, one model with several fault size parameters is regarded and the optimal fault size parameters are determined by means of an interval halving technique. Both techniques have been evaluated at a testbed and have shown very good fault detection and diagnosis capabilities.