An Efficient Approach for Synthetic Data Generation and Fault Diagnosis for Rotating Machinery

Ritabrata Chakraborty, T. Mian, P. Kundu
July 2025
DOI PDF Code

Abstract

This study explores the use of an Auxiliary Classifier Wasserstein GAN with Gradient Penalty (ACWGAN-GP) for synthetic data generation and fault diagnosis in rotating machinery. The proposed framework addresses data scarcity and class imbalance by generating diverse time-series signals using a 1D-CNN generator and a Temporal Convolutional Network (TCN)-based discriminator, both enhanced with positional embeddings. The discriminator also functions as a fault classifier. Statistical similarity metrics, PCC, Cosine Similarity, KL Divergence, and MMD, are used to validate sample quality. Experiments on the CWRU-bearing dataset demonstrate improved classification robustness and effectiveness.

Type
Conference paper
Publication
The Institution of Engineering and Technology (IET) Conference Proceedings

This work addresses the challenge of limited fault data in rotating machinery by developing efficient synthetic data generation techniques using auxiliary classifier Wasserstein GANs and conditional latent denoising diffusion models.

Fault Diagnosis Synthetic Data Rotating Machinery ACWGAN-GP Temporal Convolutional Networks deep-learning Predictive Maintenance
Ritabrata Chakraborty
Ritabrata Chakraborty
CV Research Intern

Research Engineer specializing in robotics, computer vision, and autonomous systems. Currently developing automated data annotation solutions with foundation models for autonomous vehicles at Uber.