Addressing Unbalanced Data Through Generative AI in High-Pressure Die Casting Fault Diagnosis

Abstract

Porosity defects in high-pressure die casting (HPDC) are a significant quality challenge whose diagnosis is exacerbated by data imbalance. Raw data from HPDC processes can be high-dimensional, noisy, and contain nonlinear relationships. Augmenting rare faults can be challenging, potentially leading to the generation of unrealistic or irrelevant fault signatures. This paper introduces an explainable AI framework for robust porosity diagnosis, validated using industrial data. Our method uses an autoencoder and a weighted generative adversarial network (WGAN-GP) for high-fidelity latent space augmentation of rare fault data, with a novel Fréchet Inception Distance (FID) approach to ensure synth..