Improving Predictive Maintenance Performance in Limited Data Environment with CGAN

Predictive Maintenance is critical to the efficiency and reliability of industrial systems. This work aims to improve the performance of failure prediction models under limited training data conditions. The proposed approach involves the integration of feature engineering techniques and Generative Adversarial Networks (GAN) architectures. The methodology involves pre-processing the existing limited dataset and then extending it with GAN-based synthetic data generation. The ultimate goal of the work is to improve the accuracy and reliability of equipment Remaining Useful Life (RUL) estimates. This approach resulted in a 5 percent increase in industrial maintenance and failure prediction performance, where data limitations are a common problem.