Machine performance degradation monitoring using fuzzy CMAC

Abstract

Conventional approaches to failure detection use NN, fuzzy or expert systems to detect failures (the machine is already down). We believe that if we can detect the machine performance degradation (early signs of failures), then we can prevent the occurrence of failures. Our idea is use a new type of NN, called fuzzy CMAC. We put a smooth hyperbolic tangent (tanh) function at the output of the fuzzy CMAC network with 1 denoting normal and -1 denoting the failure. The training of the network is performed by feeding known patterns of normal and failure conditions to it. When the network is applied to detect faults, if the output lies anywhere in between -1 and 1, it means the machine is in degraded state. If the output is close to 1, it means the system is close to normal but it is also on the verge of degrading. One major advantage of this method is its simplicity in implementation. A simple robot trajectory tracking example is given to illustrate the idea.