An Enhanced Health Index Construction Method Based on Noise Reduction

Le résumé fourni par la source

The prediction of Remaining Useful Life (RUL) is one of the core research directions in the PHM field, and the construction of Health Indicators (HI) is the foundation for accurate RUL prediction. Although some modern construction methods employ deep learning methods to extract prior knowledge or features, the noise in original data exhibit a significant problem. Those research that take noise into account have not considered the impact of multiple types of noise, overlooking the complexity of the noise problem. This paper proposes an enhanced noise reduction HI construction method to address this issue. The method is an integration of Stacked Denoising Autoencoders (SDAE) and Singular Value Decomposition (SVD), without the need for any manual prior knowledge. Two approaches are used to reduce the effect of noise. Firstly, two types of noise are used to corrupt the input data, training the SDAE to reconstruct noise-free outputs to achieve the goal of enhanced noise resistance. Furthermore, feature selection is used to eliminate the features significantly affected by the noise. Finally, the SVD integrated features and extracts one-dimensional feature value, that best represents the original data trend and is least affected by noise to construct the HI. Comparative experiments were conducted on a bearing dataset, and the results indicate that our method can effectively extract HI, and compared to other common methods, the extracted HI is of superior quality.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.