Machine learning based classification and analysis of wire-EDM discharge pulses

Wire electrical discharge machining (wire-EDM) process is having immense potential over conventional machining methods due to its non-contact nature of material removal. However, frequent and unanticipated machining failures like wire breakages negatively affect the productivity, sustainability and efficiency of the process. In this context, there is a wide scope to improve the process efficiency through online condition monitoring. A prominent aspect of EDM condition monitoring is discharge pulse discrimination. The threshold based methods which are currently being used has low accuracy and is reliant on operator’s experience. In this study, a machine learning (ML) based pulse classification based on the extracted discharge characteristics is proposed. The features are extracted from the raw voltage and current senor signals collected from the machining zone during the wire EDM operation. Among the various ML models, Artificial Neural Network (ANN) classifier is found to have the maximum prediction accuracy of 98 %. Also, the effects of different discharge pulses on the productivity, surface finish and machining failures are investigated. The short circuit and arc discharges are found to cause wire breakage failure if they predominate the pulse cycle by more than 80 %. Also, short and arc sparks increase the surface roughness significantly, by up to 70 % .