Modeling and Sensing Issues for Machine Diagnostics

As motivated by the need to develop automated machine diagnostics for manufacturing systems, modeling and sensing issues for model based diagnostic systems are explored. Developing models of machine systems are argued to be important for accurately isolating and interpreting machine fault signals obtained from remote sensors. The idea of modeling faults as an unknown inputs to a model of the normally operating machine is introduced. A simultaneous state and input observer can then be used to determine the number of sensors and possible sensor locations required to observe the fault inputs. This technique is demonstrated by detecting a bearing fault in a computer simulated machine tool spindle system. Good results are obtained although the results are shown to be sensitive to measurement noise. Methods for dealing with the noise are discussed. Modeling faults as inputs to a model and monitoring them with a state and input observer appears to be a promising machine diagnostics technique.