A complementary continual semi-supervised learning scheme using contrastive variational autoencoder for remaining useful life estimation

Tiancheng Wang

doi:10.1016/j.engappai.2026.114870

A complementary continual semi-supervised learning scheme using contrastive variational autoencoder for remaining useful life estimation

2026-06-09

0

PRE

AI

DOI：10.1016/j.engappai.2026.114870

Original

Original request for help

Save

Abstract

En 中文

As a solution in the field of remaining useful life, deep learning has improved its accuracy compared to traditional machine learning, but it also has some common problems in application, such as the labor-intensive process of data labeling and the catastrophic forgetting of continual learning. In order to solve the above problems, this paper proposes a complementary continual semi-supervised remaining useful life estimation method. First, this paper utilizes unsupervised pre-training and supervised training fine-tuning to fully utilize unlabeled data, and uses a simplified contrastive variational autoencoder to improve the accuracy of the method. Second, a complementary continual learning strategy is proposed for semi-supervised structure, where a regularization-based approach is used for pre-training and a memory-based approach is employed for supervised training. Finally, Kolmogorov–Arnold Networks which have continual learning capability are also applied to further improve the continual learning performance. In addition, comparison experiments are designed to verify the validity of the model through Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) and High Intensity Radiated Field (HIRF) battery dataset.

A complementary continual semi-supervised learning scheme using contrastive variational autoencoder for remaining useful life estimation

Abstract

Journal

Organization