基于特征融合的机械核心部件剩余寿命预测  

RUL Prediction of Mechanical Core Components Based on Feature Fusion

作　　者：刘宗宇 廖雪超[1,2] LIU Zong-yu;LIAO Xue-chao(School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,China;Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System,Wuhan 430065,China)

机构地区：[1]武汉科技大学计算机科学与技术学院,湖北武汉430065 [2]智能信息处理与实时工业系统湖北省重点实验室,湖北武汉430065

出　　处：《计算机技术与发展》2025年第3期165-171,共7页Computer Technology and Development

基　　金：国家自然科学基金项目(62273264)。

摘　　要：工程机械设备的预测性维护能确保设备高效运行和维护计划的合理安排,其关键在于准确预测设备系统或核心部件的剩余寿命。针对工程机械设备预测性维护中的特征提取和预测精度难题,该文提出了基于特征融合的Informer机械设备核心部件剩余寿命预测框架。首先,采用按寿命比例的训练样本构造方法优化数据利用率,采用巴特沃斯高通滤波器和小波降噪对原始数据进行滤波和降噪,并对原始数据进行特征扩展来提取关键特征,使用XGBoost算法进行特征选择。然后,按设备类型将特征选择后的数据分类,设计了基于Informer的机械核心部件剩余寿命预测模型进行分类训练。使用公开数据集对模型进行验证,与其他预测模型的预测结果进行比较,验证了基于特征融合的Informer预测模型能够实现最准确的预测。Predictive maintenance of engineering machinery equipment can ensure the efficient operation of the equipment and the rational arrangement of maintenance plans,with the key lying in accurately predicting the RUL of equipment systems or core components.To address the challenges of feature extraction and prediction accuracy in predictive maintenance of engineering machinery equipment,we propose an Informer-based framework for the prediction of the RUL of core components of mechanical equipment based on feature fusion.Firstly,an optimized data utilization method is adopted to construct training samples according to the proportion of life.Bartlett high-pass filter and wavelet denoising are used to filter and denoise the raw data,and feature extension is performed to extract key features from the raw data.The XGBoost algorithm is used for feature selection.Then,the selected data is classified by equipment type,and an Informer-based predictive model for the RUL of mechanical core components is designed for classification training.The model is validated using public datasets,demonstrating that the Informer predictive model based on feature fusion achieves the highest prediction accuracy compared to other models.

关 键 词：预测性维护 剩余寿命预测 特征提取 特征融合 深度学习 自注意力机制 

分 类 号：TP390[自动化与计算机技术—计算机应用技术]