8 / 2025-04-05 19:31:20

Fault Sample Generation Method for Industrial Processes with a Limited Number of Samples

process industry, generative adversarial networks, fault diagnosis, deep learning

Abstract—Process industry serves as a pivotal sector in China's national economy and social development. To address the challenges of limited fault samples and difficulties in sample collection in this field, this paper proposes a fault sample generation method for process industries based on Generative Adversarial Networks (GANs). First, the random forest algorithm is employed to analyze fault feature importance. Subsequently, the original adversarial network is optimized by integrating deep convolution and maximum mean discrepancy (MMD). Finally, case validation is conducted using the Tennessee-Eastman simulation dataset and a collected compressor unit dataset. Experimental results demonstrate that the generated samples significantly improve diagnostic accuracy and outperform the original GANs in effectiveness.