Data analytics and machine learning modeling of energy consumption-surface integrity relationship in wire electrical discharge machining

Wire Electrical Discharge Machining (Wire-EDM) is a non-traditional machining process that enables the high-precision machining of hard, conductive materials, making it widely applicable in the aerospace, tooling, and medical industries. Wire-EDM faces a trade-off between surface quality and efficiency, making it essential to understand how process parameters influence surface integrity. Although the bulk of the research focuses on single-discharge parameters, the complex relationship between machine-level energy consumption and surface quality remains largely unknown. The advancement in real-time process monitoring techniques enables machine learning (ML) to be an innovative approach to establishing the energy consumption-surface quality relationship.To address these gaps, this dissertation investigates the multi-scale relationship between Wire-EDM energy signatures and surface integrity. First, multiple experiments were conducted using different workpiece materials and cutting conditions to characterize machine-level power consumption characteristics in different cutting modes and examine their relationships with surface roughness and topographical features. A machine learning model was then developed to predict post-machining surface roughness based on overall power consumption, yielding promising predictive performance. Furthermore, this work has created a physics-informed neural network (PINN) to predict microscale thermal behavior during a single EDM discharge. The physics-informed machine learning (PIML) model predicts the transient temperature field and melt-pool size by embedding the governing heat conduction equations and boundary conditions into the learning network. This provides a mechanistic understanding of how discharge energy translates into localized thermal effects that influence surface formation. Together, the energy data analytics, the data-driven ML model, and the physics-informed ML models offer an in-depth understanding of the energy–surface integrity relationship in Wire-EDM.