Condition monitoring and fault diagnosis of operating machinery have been of the main interests for many years. Existing systems employ data driven models which has limited prediction capability. On the contrary, a physics based model can predict phenomena that the system has not seen before. It also provides a means for wot cause analysis. However, major obstacles include the complexity involved and real-time feasibility. In this paper, we use the bond graph technique to construct an analytical, physics-based model. We use a reciprocating compressor as an example since it is a good representation of multi-domain processes with a variety of possible failure modes. The bond graph model is cross examined with a finite element model and ultimately validated. To show how one can benefit from this approach, cracked piston rod is simulated to generate vibration signals that can potentially be picked up from accelerometers mounted on the exterior surface of the compressor.
展开▼
