Speakers
Description
Electrical Discharge Machining (EDM) is a widely applied manufacturing process where performance is strongly influenced by multiple interacting technological parameters. This study presents a data-driven approach for analysing and reducing the dimensionality of EDM process parameters using feature selection methods. An industrial dataset containing six input parameters and time-related outputs across three machining stages is investigated. Two analysis tasks are defined: classification of machining regimes into high-speed and low-speed categories, and refinement within the high-speed regime. Adaptive Hybrid Feature Selection (AHFS) and Model Input-Output Configuration based Search (MICS-EFS) methods are applied to identify influential parameters and optimal feature subsets. The results show that high classification accuracy of up to 96% can be achieved for regime separation, while the high-speed regime analysis remains more challenging, with 75–85% accuracy. An optimal number of features is identified for most cases, although full parameter sets are required in certain scenarios. In addition, the reduced parameter space enables simpler process control with fewer variables to be optimally tuned and significantly decreases the number of required experiments in design of experiments (DoE) for process optimisation. The findings reveal both globally dominant and stage-dependent parameters, supporting efficient process optimisation and highlighting the benefits of a two-stage analytical framework.