Utilizing Autoregressive Networks for Full Lifecycle Data Generation of Rolling Bearings for RUL Prediction (2401.01119v1)

Abstract: The prediction of rolling bearing lifespan is of significant importance in industrial production. However, the scarcity of high-quality, full lifecycle data has been a major constraint in achieving precise predictions. To address this challenge, this paper introduces the CVGAN model, a novel framework capable of generating one-dimensional vibration signals in both horizontal and vertical directions, conditioned on historical vibration data and remaining useful life. In addition, we propose an autoregressive generation method that can iteratively utilize previously generated vibration information to guide the generation of current signals. The effectiveness of the CVGAN model is validated through experiments conducted on the PHM 2012 dataset. Our findings demonstrate that the CVGAN model, in terms of both MMD and FID metrics, outperforms many advanced methods in both autoregressive and non-autoregressive generation modes. Notably, training using the full lifecycle data generated by the CVGAN model significantly improves the performance of the predictive model. This result highlights the effectiveness of the data generated by CVGans in enhancing the predictive power of these models.

• The paper introduces a CVGAN model that combines CVAE, GAN, and autoregressive methods to generate full lifecycle bearing data for enhanced RUL predictions.
• The model utilizes historical vibration data to produce realistic one-dimensional signals in both horizontal and vertical directions, achieving lower MMD and FID metrics on the PHM 2012 dataset.
• The approach significantly improves data authenticity and predictive maintenance, paving the way for future improvements in complex signal generation.

Introduction

Rolling bearings are critical in machinery and their lifespan prediction is an essential aspect of maintenance. Despite the advancements in predictive models for estimating the Remaining Useful Life (RUL) of rolling bearings, the quality of predictions often suffers due to limited data availability. This paper introduces the CVGAN (Conditional Variational Generative Adversarial Network) model which effectively generates full lifecycle vibration data of rolling bearings, improving the quality of lifespan predictions.

Methodology

CVGAN leverages historical vibration data and RUL to produce one-dimensional vibration signals, encompassing both horizontal and vertical dimensions. The model incorporates Conditional Variational Autoencoder (CVAE) with the Generative Adversarial Network (GAN), taking advantage of their respective strengths. Additionally, an autoregressive method is utilized, which iteratively guides current data generation with previous outputs, enhancing temporal coherence in the data.

Validation

Experiments using the PHM 2012 dataset show that CVGAN-generated data significantly outperforms other respected methods in achieving lower MMD (Maximum Mean Discrepancy) and FID (Fréchet Inception Distance) metrics. This proves the generated data's authenticity and suitability for predictive modeling purposes. Utilizing autoregressive generation, the model consistently produces data that closely mirrors real measurements during bearing operation.

Conclusion and Future Work

The paper establishes that using CVGAN to generate complete data cycles for rolling bearings significantly boosts the efficacy of predictive models. In future endeavors, the initial data generator will be improved, aiming to minimize discrepancies generated during the initial steps of data synthesis. Further investigations are also directed towards more complex data structures, such as two-dimensional vibration signals, and refining network architectures to maximize the efficiency of autoregressive generation methods.

By effectively addressing the challenge of limited data in rolling bearing lifespan prediction, the CVGAN model stands to enhance predictive maintenance practices significantly in industrial settings.