Remaining useful life prediction considering correlated multi-parameter nonlinear degradation and small sample conditions

This study establishes a RUL prediction method based on an improved Wasserstein GAN, a nonlinear Wiener process, and a Copula function (IWNC) to address correlated multi-parameter nonlinear degradation with small samples. Initially, the proposed IWNC method develops a correlation-aware multi-sequence degradation data augmentation model using an improved Wasserstein Generative Adversarial Network (WGAN) that combines an LSTM-based generator and a 1D CNN-based discriminator. Time series consistency and multi-parameter correlation terms are incorporated into the generator’s loss function to enhance the quality of the augmented degradation data. A nonlinear Wiener process model, integrated with a Copula-based correlation model, is then developed to construct a joint RUL prediction model. Experimental results demonstrated that the IWNC method effectively addresses the challenges of small sample sizes and correlated multi-parameter nonlinear degradation. The augmented data generated by the IWNC method significantly contributes to improving RUL prediction accuracy. Due to the IWNC method’s ease of implementation and broad applicability, it holds considerable potential for widespread adoption across various domains, including digital twins.