Synthesizing Rolling Bearing Fault Samples in New Conditions: A framework based on a modified CGAN

TL;DR

The paper tackles the challenge of scarce fault data for bearing condition monitoring across varying operating conditions. It introduces Normal to Fault GAN (N2FGAN), a Pix2Pix-inspired conditional GAN that converts abundant normal vibration signals into fault data under target conditions without using random noise in the generator. Trained on one condition with both normal and fault examples, the model generates fault data for other speeds, and the synthesized data closely matches real fault data as shown by t-SNE visualizations and high classifier performance (often >97%). Compared against CGAN, WGAN-GP, and classical augmentation, N2FGAN demonstrates higher realism and improves fault-classification accuracy, enabling more robust data-driven diagnosis in unseen conditions and addressing the small-sample problem in industrial settings.

Abstract

Bearings are one of the vital components of rotating machines that are prone to unexpected faults. Therefore, bearing fault diagnosis and condition monitoring is essential for reducing operational costs and downtime in numerous industries. In various production conditions, bearings can be operated under a range of loads and speeds, which causes different vibration patterns associated with each fault type. Normal data is ample as systems usually work in desired conditions. On the other hand, fault data is rare, and in many conditions, there is no data recorded for the fault classes. Accessing fault data is crucial for developing data-driven fault diagnosis tools that can improve both the performance and safety of operations. To this end, a novel algorithm based on Conditional Generative Adversarial Networks (CGANs) is introduced. Trained on the normal and fault data on any actual fault conditions, this algorithm generates fault data from normal data of target conditions. The proposed method is validated on a real-world bearing dataset, and fault data are generated for different conditions. Several state-of-the-art classifiers and visualization models are implemented to evaluate the quality of the synthesized data. The results demonstrate the efficacy of the proposed algorithm.

Figures (10)

• Figure 1: A flow diagram of using N2FGAN for fault data generation.
• Figure 2: A basic structure of LSTM Network.
• Figure 3: A schematic representation of the major CNN layers.
• Figure 4: A comparison between GAN, CGAN, and our proposed methods(N2FGAN).
• Figure 5: The propose data generation framework (N2FGAN).