采用多尺度Mamba架构的装备剩余寿命预测方法

打开文本图片集

关键词：剩余寿命预测;Mamba模型；门控循环单元；多尺度卷积层;互补特征动态融合中图分类号：TH17文献标志码：ADOI：10.7652/xjtuxb202605015 文章编号：0253-987X（2026）05-0153-11

Equipment Remaining Useful Life Prediction Method Using Multi-Scale Mamba Architecture

TIAN Congbaol，WANG Zhaoqiang1'²，HU Changhual ，FAN Hongdong1 ， ZHENG Ziyangl ，KONG Xiangyu1 （1.LaboratoryofInteligentControl，RocketForceUniversityofEngineeing，XianO25，China;.StateKeyLaboratoryof Fluid Power Components and Mechatronic Systems， Zhejiang University，Hangzhou 31oo58，China）

Abstract： To address the issues of fixed single convolution kernels and insufficient local temporal feature extraction in existing Mamba-based equipment remaining useful life （RUL） prediction methods， a novel RUL prediction approach based on a multi-scale Mamba architecture is proposed. A parallel network combining multi-scale Mamba and gated recurrent units （GRUs） was constructed. The multi-scale Mamba captures long-term degradation trends through the synergistic interaction of multi-scale convolutional layers and structured state-space models，while the GRUs extract shortterm dynamic features via gating mechanisms， achieving complementary long-and short-term feature representation. A module for dynamic fusion of complementary features was designed to perform weighted fusion of these two types of features through dynamic gating weights，effectively suppressing redundant information and enhancing key features. Fully connected layers were then employed to map degradation features and predict RUL. Validation on the NASA’s aircraft engine dataset demonstrates that the proposed method achieves superior root mean square error compared to mainstream models，with the scoring function reduced by 22%-73% compared to the best baseline model. Additionally，it improves computational efficiency by 57.5% over the latest Mamba-based methods.

Keywords： remaining useful life prediction； Mamba model； gated recurrent unit；multiscale convolutional layer;dynamic fusion of complementary features

随着装备服役时间的增加，由于疲劳磨损、腐蚀、蠕变等原因，装备的性能不可避免地发生退化，装备一旦发生故障，常会造成巨大的人员和财产损失，尤其是在核电、航空航天、武器装备、高速列车等对可靠性、安全性要求较高的领域。（剩余19718字）