基于WPT-MCNN的电梯门机故障诊断方法  

A Fault Diagnosis Method for Elevator DoorSystem Based on WPT-MCNN

作　　者：童鑫 万安平 程晓民 TONG Xin;WAN Anping;CHENG Xiaomin(School of Electromechanical Engineering,Anhui University of Science&Technology,Huainan,Anhui 232000;Department of Mechanical Engineering,City College of Zhejiang University,Hangzhou,Zhejiang 310015,China)

机构地区：[1]安徽理工大学机电工程学院,安徽淮南232000 [2]浙大城市学院机械工程学系,浙江杭州310015

出　　处：《九江学院学报(自然科学版)》2025年第1期30-36,共7页Journal of Jiujiang University:Natural Science Edition

基　　金：国家自然科学基金资助项目(编号52372420);宁波市科创甬江2035关键技术突破计划项目(编号2024Z177);浙江省科技计划项目(编号2024C01039);宁波市重大专项(编号2022Z071,2023Z032)的研究成果之一。

摘　　要：电梯门机系统作为电梯运行的关键组件,其健康状态直接关系到电梯的安全性与运行效率。针对传统专家系统在电梯门机故障诊断中面临的诊断精度不高、效率低下等挑战,深入研究了人工智能深度学习技术在该领域的创新应用。鉴于电梯门机故障数据集的稀缺性,自主设计并搭建了实验平台,全面收集了六种典型的电梯门机故障案例,涵盖了多种工况下的振动信号。为充分挖掘这些振动信号中的故障特征,提出了一种融合小波包变换(WPT)与多尺度卷积神经网络(MCNN)的先进诊断方法。该方法首先利用WPT技术,根据故障频率特性将三通道振动信号进行二级分解,重构为三个具有不同频率特性的子带(subband),并通过归一化处理确保数据的一致性。之后,这三个子带被作为多尺度输入送入MCNN模型,该模型能够有效捕捉并融合振动信号的全局与局部特征,从而实现对故障类型的精准识别。实验结果表明,所提出的方法在电梯门机系统振动故障诊断中展现出了卓越的性能,经过十次独立实验的验证,平均故障诊断准确率高达99.05%。这一成果不仅验证了深度学习技术在电梯门机故障诊断中的巨大潜力,也为电梯行业的智能化维护与管理提供了有力的技术支持。The elevator door operator system,a crucial component for elevator operation,directly impacts the safety and efficiency of the elevator.Addressing the challenges faced by traditional expert systems,such as low diagnostic accuracy and inefficiency in elevator door operator fault diagnosis,an innovative applications of artificial intelligence and deep learning technologies was explored in this field.Due to the scarcity of fault data sets for elevator door operators,a self-designed experimental platform was established to comprehensively collect six typical fault cases,covering vibration signals under various operating conditions.To thoroughly extract fault features from these vibration signals,an advanced diagnostic method combining Wavelet Packet Transform(WPT)and Multi-Scale Convolutional Neural Networks(MCNN)was proposed.This method first employed WPT to decompose the three-channel vibration signals into three subbands with distinct frequency characteristics,followed by normalization to ensure data consistency.These subbands were then fed into the MCNN model as multi-scale inputs,enabling the model to effectively capture and integrate both global and local features of the vibration signals for precise fault type identification.Experimental results demonstrated that the proposed method exhibits exceptional performance in diagnosing vibration faults in elevator door operator systems,achieving an average fault diagnosis accuracy of 99.05%after ten independent tests.This result not only validated the immense potential of deep learning technology in elevator door operator fault diagnosis but also provided robust technical support for intelligent maintenance and management in the elevator industry.

关 键 词：小波包变换 多尺度卷积网络 振动故障诊断 电梯门机系统 

分 类 号：TH17[机械工程—机械制造及自动化] TU857[建筑科学]