【作者】 郑琪；

【导师】 赵春晖；

【作者基本信息】 浙江大学 ， 控制科学与工程， 2022， 硕士

【摘要】 随着城市的发展,载人直梯成为了生活中常见的一种设备。然而,电梯在方便乘客上下楼的同时,也带来了一定的安全隐患。由于电梯系统复杂,日常使用频率很高,很容易出现一些故障,例如急停故障、轿厢振动过大故障、超速和自动复位故障,造成财产损失、人员伤亡。此外,位于电梯机房中的曳引电机轴承是很容易损坏的零部件之一,一旦发生故障,将会影响曳引电机运转,直接影响到电梯的正常运行。因此,如何保障电梯的安全运行是亟需解决的社会问题。目前,维保人员定期对电梯进行检修是一种常规手段,然而人工经验的不足经常会导致无法及时有效地检测出电梯故障;曳引电机轴承在长期运行过程中会发生性能退化,其剩余寿命难以被技术人员直接评估和量化,导致不能及时更换轴承,容易引发电梯故障。因此,本文针对载人直梯这一集机械、控制、电气为一体的复杂特种设备,通过数据特征挖掘的方法对电梯运行数据进行分析,结合机器学习、神经网络等方法,对电梯运行过程中发生的几类典型故障进行有效诊断;进一步地,对电梯曳引电机轴承的剩余寿命进行预测,提前预警,避免故障的发生,从故障诊断与故障预警两个角度保障电梯的安全运行。本文的研究内容如下:(1)针对电梯这一复杂特种设备,阐述了电梯系统的结构组成、几种电梯典型故障的含义,以及二者之间的关联关系。针对所采集到的电梯运行数据,尤其是易受噪声影响的轿厢加速度信号,按照电梯型号、故障标签等进行归类整理,并剔除其中的脏数据,提升数据质量,为后续算法研究提供有力的数据支撑。(2)针对四种电梯典型故障,提出了一种基于轿厢加速度特征挖掘的诊断方法。首先,利用时域分析法挖掘其中的时域特征,通过设置合理的时域特征阈值,得到急停故障诊断模型,对运行过程中发生的急停故障进行诊断,确认样本发生急停故障之后发出告警信息;若样本未发生急停故障,则在时域分析法的基础上,通过小波包分解提取时频域特征,结合神经网络判断样本是否发生轿厢振动过大故障;同时,对加速度信号样本进行积分,得到速度信号,利用高斯混合模型对超速和自动复位故障进行诊断,并给出相应的告警信息。该方法针对电梯急停故障、轿厢振动过大故障、超速和自动复位故障这四种典型故障有针对性地提出了相应的诊断模型,并综合起来形成完整的在线故障诊断流程,实现对四种典型故障的有效诊断。(3)针对电梯曳引电机轴承剩余寿命难以被维保人员直接评估和量化的问题,提出了一种基于深度残差收缩网络的轴承分段剩余寿命预测方法。首先,以轴承工作机理知识为基础,对轴承的磨损阶段进行划分,有针对性地考虑处于“稳定磨损阶段”的轴承数据;然后,对剩余寿命进行片段划分,将分段预测精度作为模型的优化目标;最后,利用深度残差收缩网络对轴承振动信号中的特征进行自动提取,构建完整的剩余寿命预测模型。该方法从多个角度有效地提升了针对电梯轴承剩余寿命的预测精度,保障了电梯的安全运行。更多还原

【Abstract】 Nowadays,the manned elevator has become common equipment in our daily life.However,while the elevator brings convenience to the citizens,it also brings certain safety hazards.As the elevator system is complex and used frequently,some faults are prone to occur,such as emergency stop faults,excessive car vibration faults,overspeed and automatic reset faults,which can cause property damage,personal injury and death.In addition,the traction motor bearing in the elevator machine room is one of the components that are easily damaged.It will affect the operation of the traction motor and the elevator once a failure occurs.Therefore,ensuring the safe operation of elevators is a social problem needing to be solved.It is a common method for maintenance personnel to regularly check and repair elevators.However,lack of manual experience often results in the failure to detect elevator failures timely and effectively;traction motor bearings will experience performance degradation during long-term operation,and their remaining life is difficult to be directly evaluated and quantified by the technicians,resulting in the inability to replace the bearings in time,and it is easy to cause elevator failures.Therefore,this article focuses on the manned elevator which is complex special equipment that integrates machinery,control,and electricity.It analyzes the elevator operation data through the method of data feature mining,and combines machine learning to diagnosis several types of typical faults which occur during the elevator operation process.Moreover,It predicts the remaining life of the elevator traction motor bearings to avoid the occurrence of faults.It ensures the safe operation of the elevator from the perspective of fault diagnosis and fault warning.The research content of this article is as follows:(1)Aiming at the complex special equipment of elevator,the structure of the elevator system,the meaning of several typical elevator failures,and the relationship between them are explained.According to the collected elevator operation data,especially the car acceleration signals that are susceptible to noise,they are sorted according to the elevator model,fault label,etc.And the dirty data are eliminated to improve the data quality and provide powerful support for subsequent algorithm research.(2)Aiming at four typical elevator faults,a diagnosis method based on car acceleration feature mining is proposed.Firstly,the time domain analysis method is used to mine the time domain features,and by setting a reasonable time domain feature threshold,the emergency stop fault diagnosis model is obtained.If a sample is diagnosised as emergency stop fault,the model will send alarm information.On the contrary,if the sample does not have an emergency stop failure,based on the time domain analysis method,the time-frequency domain characteristics are extracted through wavelet packet decomposition,and the neural network is used to determine whether the sample has excessive car vibration failure.At the same time,the acceleration signal is integrated to obtain the velocity signal,then the Gaussian Mixture Model is used to diagnose the overspeed and automatic reset faults,and the corresponding warning information is given.This method proposes corresponding diagnosis models for the four typical failures of elevator emergency stop failure,excessive car vibration failure,overspeed and automatic reset failure,and combines them to form a complete online fault diagnosis process to achieve effective diagnosis of the four typical failures.(3)Considering that the remaining life of the elevator traction motor bearing is difficult to be directly evaluated and quantified by the maintenance personnel,a method for predicting the remaining life of the bearing segment based on the deep residual shrinkage network is proposed.First of all,based on the knowledge of bearing working mechanism,the wear stage of the bearing is divided,and the bearing data in the "stable wear stage" is considered in a targeted manner;then,the remaining life is segmented,and the segmented prediction accuracy is used as the model.Optimize the goal;finally,use the deep residual shrinkage network to automatically extract the features in the bearing vibration signal to build a complete remaining life prediction model.This method effectively improves the prediction accuracy for the remaining life of the elevator bearing from multiple angles,and ensures the safe operation of the elevator.更多还原