Monitoring of induction motor load by neural network techniques

This paper deals with the electric tracing of the load variation of an indu ction machine supplied by the mains. A load trouble, like a torque dip, aff ects the machine supply current and consequently it should be possible to u se the current pattern to detect features of the torque pattern, using the machine itself as a torque sensor. But current signature depends on many ph enomena and misunderstandings are possible. At first the effect of different load anomalies on current spectrum, in com parison with other machine troubles like rotor asymmetries, are investigate d. Reference is made to low frequency torque disturbances, which cause a qu asistationary machine behavior Simplified relationships, validated by simul ation results and by experimental results, are developed to address the cur rent spectrum features. In order to detect on-lines anomalies, a current signature extraction is pe rformed by the time-frequency spectrum approach. This method allows the det ection of random fault as well, Finally it is shown that a Neural Network approach can help the torque patt ern recognition, improving the interpretation of machine anomalies effects.