On the application of recurrent neural network techniques for detecting instability trends in an industrial process

This paper analyses the use of the recurrent neural network approach to diagnose degraded cutting regimes in wire electrical discharge machining (WEDM) Process. The main objective of this work is to detect in advance the degradation of the cutting process since this can lead to the breakage of the cutting tool (the wire), reducing the process productivity and the required accuracy. Besides, the quantification of the grade of influence of different types of degraded behaviours is meant in this work. In order to achieve all these challenges, a configuration of three Elman neural networks has been selected due to the memorization capability and the dynamic character of the Elman architecture. Each network is dedicated to specific process functions. The results of this work show a satisfactory performance of the presented approach.