Development of a system for part quality prediction and process parameters adjustment in wire electrical discharge machining

In the past decades, wire electric discharge machining (W-EDM) has undergone a number of techno-logical developments aimed at improving its process performance. Still the current practice of W-EDM is not defect-free and results in a number of surface related defects on the machined parts. The main defects related to surface quality of parts produced by the W-EDM process include the occurrence of surface lines, surface roughness and the occurrence of recast layer on the machined surface. The experimental evidence suggests that surface defects in W-EDM occur mainly due to short-circuiting or abnormal sparks. If the W-EDM process is controlled to minimize the duration of short circuit or abnormal sparks, the occurrence of surface defects can be minimized. Furthermore, online monitoring of the spark quality can help minimize the occurrence of surface lines by first predicting the possible occurrence of surface lines in W-EDM as a function of certain critical process variables, namely, discharge gap and spark frequency, and then proactively implementing the corrective action in the form of adjustment of vital process parameters, like pulse-off time, before the sequence of abnormal sparks could develop into a visible surface line. Two systems are proposed to achieve the objectives mentioned above: an offline multiple-input multiple-output (MIMO) fuzzy-nets prediction system and a multi-objective optimization system for process parameters adjustment and an online prediction and adjustment system based on the short circuit duration. For the offline prediction system, the two part quality characteristics considered in this thesis are surface roughness and recast layer thickness. A fuzzy logic based approach called fuzzy-nets approach for the prediction of these part quality characteristics on the basis of the initial conditions is proposed. A hybrid multi-objective optimization method combining fuzzy-nets prediction technique and genetic algorithm is applied to find best setting of process parameters based on desired surface roughness and minimium recast layer thickness. The online prediction and adjustment system comprises: (i) a procedure for the online prediction of the occurrence of surface lines in W-EDM based on the short circuit duration, and (ii) a procedure for online adjustment pulse-off time (electrical process parameter) to prevent the occurrence of surface lines if and when their occurrence is predicted. Both systems required intensive experiments based on carefully designed experiment plans to collect and assess the training data for the different fuzzy rule bases. A case study from GF Machining Solutions Ltd. illustrating the different issues related to the proposed approaches is described. The experimental results validate the proposed approach for both online and offline systems.