WO2021000556A1 - 一种工业设备剩余有效寿命预测方法、系统及电子设备 - Google Patents

一种工业设备剩余有效寿命预测方法、系统及电子设备 Download PDF

Info

Publication number
WO2021000556A1
WO2021000556A1 PCT/CN2019/130600 CN2019130600W WO2021000556A1 WO 2021000556 A1 WO2021000556 A1 WO 2021000556A1 CN 2019130600 W CN2019130600 W CN 2019130600W WO 2021000556 A1 WO2021000556 A1 WO 2021000556A1
Authority
WO
WIPO (PCT)
Prior art keywords
vibration signal
signal data
data
scaled
function
Prior art date
Application number
PCT/CN2019/130600
Other languages
English (en)
French (fr)
Inventor
阳文斯
么庆丰
叶可江
须成忠
Original Assignee
中国科学院深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Publication of WO2021000556A1 publication Critical patent/WO2021000556A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/04Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/20Administration of product repair or maintenance
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/04Ageing analysis or optimisation against ageing
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02PCLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30Computing systems specially adapted for manufacturing

Definitions

  • This application belongs to the technical field of equipment failure prediction, and particularly relates to a method, system and electronic equipment for predicting the remaining effective life of industrial equipment.
  • the existing failure prediction methods mainly include model-based methods and data-driven methods:
  • Model-based prediction methods mathematical models are established to predict the remaining useful life of industrial equipment based on empirical knowledge and collected data. The evaluation and prediction results of this type of method are generally intuitive and accurate, but the equipment degradation information needs to be known in advance And failure mechanism.
  • Data-driven method It aims to convert equipment detection and operating data into information related to equipment degradation, system operating status and its degradation mechanism model.
  • the basic steps of this type of method are feature extraction, feature selection, degradation state evaluation and remaining effective life prediction.
  • Artificial intelligence and statistical methods are used to extract effective data features from collected sensor signals, and then learn the degradation mode of the device and predict the device The effective remaining life (RUL).
  • RUL effective remaining life
  • the existing problems of the existing fault prediction methods are:
  • Model-based prediction methods rely on analytical models (algebraic or differential equations) to represent the operating status of industrial equipment and its aging process. Although this kind of method can provide more accurate results, the real equipment system is usually non-linear, and the aging mechanism of equipment is usually random and difficult to obtain in the form of an analytical model. In actual working conditions, it is difficult to establish a The model can adapt to complex environmental noise and degradation mechanisms.
  • support vector machine (SVM) modeling is useful to predict the remaining effective service life of industrial equipment, but for large-scale training data and multi-class problems, SVM consumes a lot of running memory and calculations Time, resulting in lower prediction accuracy and slower speed.
  • CNN convolutional neural network
  • CNN convolutional neural network
  • Convolutional neural network (CNN) ) Can not solve the time series prediction well, resulting in low prediction accuracy.
  • the model is more difficult to converge and the prediction speed is slow.
  • This application provides a method, system and electronic device for predicting the remaining effective life of industrial equipment, which aims to solve one of the above technical problems in the prior art at least to a certain extent.
  • a method for predicting the remaining useful life of industrial equipment including the following steps:
  • Step a Normalize the original vibration signal data of the device
  • Step b Use the empirical mode decomposition method to perform feature expansion on the normalized vibration signal data, and then extract the data feature of the vibration signal data;
  • Step c construct a time series convolutional network according to the extracted data features
  • Step d Use the time series convolutional network to output the remaining effective life prediction result of the device.
  • the technical solution adopted in the embodiment of the present application further includes: in the step a, the normalization formula is:
  • X is the original vibration signal data
  • X scaled is the normalized vibration signal data
  • X std is the normalized vibration signal data
  • X min is the mean square error, minimum and maximum of the original vibration signal data respectively.
  • the standardized vibration signal data X scaled [x(1), x(2),...,x(N)].
  • the technical solution adopted in the embodiment of the present application further includes: in the step b, the feature expansion of the normalized vibration signal data using the empirical mode decomposition method specifically includes:
  • Step b1 Find all the maximum points of the vibration signal data X scaled , and fit all the maximum points with the cubic spline interpolation function to form the upper envelope of the vibration signal data X scaled ; find the vibration signal data X All the minimum value points of scaled are fitted by cubic spline interpolation function to form the lower envelope of vibration signal data X scaled ;
  • Step b3 Determine whether the signal sequence x 1 is an eigenmode function, if x 1 is not an eigenmode function, re-execute step b1 to re-decompose the vibration signal data; if x 1 is an eigenmode function, perform step b4;
  • N represents the signal length in the delay time slice
  • c(i) represents the signal amplitude of the i-th data point in a certain eigenmode function.
  • the technical solution adopted by the embodiment of the application further includes: in the step b3, the judgment method for judging whether the signal sequence x 1 is an eigenmode function is: the eigenmode function must satisfy two necessary conditions: one is a function In the entire time range, the number of local extreme points and zero-crossing points must be equal, or at most one difference; second, at any point in time, the average value of the upper envelope of the local maximum and the lower envelope of the local minimum must be Is 0 or close to 0.
  • the technical solution adopted by the embodiment of the application further includes: in the step d, the prediction result of the remaining useful life of the output device using the time-series convolutional network is specifically: the time-series convolutional network uses a one-dimensional causal convolution sum
  • the dilated convolution is used as a standard convolutional layer, and every two such convolutional layers and the identity map are encapsulated into a residual module containing the RELU function.
  • the residual module stacks a deep sequential convolutional network, and at the end Use full convolution instead of a fully connected layer to make the output and input dimensions consistent to achieve end-to-end prediction.
  • Data processing module used to normalize the original vibration signal data of the device
  • Feature extraction module used to use empirical mode decomposition to perform feature expansion on the normalized vibration signal data, and then extract the data features of the vibration signal data;
  • Model building module used to construct a time series convolutional network according to the extracted data features
  • Result output module used to output the prediction result of the remaining effective life of the device using the time series convolutional network.
  • the technical solution adopted in the embodiment of the present application further includes: the normalization formula is:
  • X is the original vibration signal data
  • X scaled is the normalized vibration signal data
  • X std is the normalized vibration signal data
  • X min is the mean square error, minimum and maximum of the original vibration signal data respectively.
  • the standardized vibration signal data X scaled [x(1), x(2),...,x(N)].
  • N represents the signal length in the delay time slice
  • c(i) represents the signal amplitude of the i-th data point in a certain eigenmode function.
  • the technical solution adopted in the embodiment of the present application also includes: the method for determining whether the signal sequence x 1 is an eigenmode function is: the eigenmode function must meet two necessary conditions: one is that the function is locally extreme in the entire time range. The number of value points and zero-crossing points must be equal, or at most one difference; the second is that at any point in time, the average value of the upper envelope of the local maximum and the lower envelope of the local minimum must be 0 or close to 0.
  • the technical solution adopted in the embodiment of this application also includes: the time-series convolutional network uses one-dimensional causal convolution and dilated convolution as standard convolutional layers, and encapsulates every two such convolutional layers and identity mappings into A residual module containing the RELU function.
  • the residual module stacks a deep time series convolutional network, and at the end uses full convolution instead of the fully connected layer, so that the output and input dimensions are consistent, and end-to-end prediction is realized.
  • an electronic device including:
  • At least one processor At least one processor
  • a memory communicatively connected with the at least one processor; wherein,
  • the memory stores instructions executable by the one processor, and the instructions are executed by the at least one processor, so that the at least one processor can perform the following operations of the above-mentioned method for predicting the remaining useful life of industrial equipment :
  • Step a Normalize the original vibration signal data of the device
  • Step b Use the empirical mode decomposition method to perform feature expansion on the normalized vibration signal data, and then extract the data feature of the vibration signal data;
  • Step c construct a time series convolutional network according to the extracted data features
  • Step d Use the time series convolutional network to output the remaining effective life prediction result of the device.
  • the beneficial effects produced by the embodiments of the present application are: the method, system and electronic equipment for predicting the remaining useful life of industrial equipment in the embodiments of the present application extract and enrich the data features of the original signal through empirical model decomposition and then use time series convolution Neural network training and prediction obtain the remaining effective service life prediction model.
  • the application of time series convolutional network considers and utilizes the time series characteristics of the original signal of industrial equipment, so that the trained model learns data characteristics more accurately and representatively, has better generalization ability, and greatly improves the industrial equipment
  • the prediction speed and accuracy of the remaining life are achievable in the actual manufacturing process.
  • FIG. 1 is a flowchart of a method for predicting the remaining effective life of industrial equipment according to an embodiment of the present application
  • Figure 2 (a)- Figure 2 (c) are diagrams of structural elements of time series convolutional networks
  • FIG. 3 is a schematic structural diagram of a system for predicting the remaining useful life of industrial equipment according to an embodiment of the present application
  • Figure 4 is a schematic diagram of verification results
  • FIG. 5 is a schematic diagram of the hardware device structure of the method for predicting the remaining effective life of industrial equipment provided by an embodiment of the present application.
  • FIG. 1 is a flowchart of a method for predicting the remaining useful life of industrial equipment according to an embodiment of the present application.
  • the method for predicting the remaining effective life of industrial equipment in the embodiment of the application includes the following steps:
  • Step 100 Collect raw vibration signal data of the equipment
  • Step 200 normalize the original vibration signal data to obtain a sample data set, and divide the sample data set into a training set and a test set;
  • step 200 because different data have different specification units, it is necessary to perform MinMax normalization processing on the original vibration signal data.
  • the normalization formula is as follows:
  • X is the original vibration signal data
  • X scaled is the normalized vibration signal data
  • X std is the normalized vibration signal data
  • X min is the mean square error and minimum value of the original vibration signal data, respectively
  • X max is the mean square error and minimum value of the original vibration signal data, respectively
  • Step 300 Use Empirical Mode Decomposition (EMD) to perform feature expansion on the training set data, and then extract the data features of the training set data;
  • EMD Empirical Mode Decomposition
  • IMF Intransic Mode Function
  • Step 301 Find all the maximum points of the vibration signal data X scaled , and fit all the maximum points with the cubic spline interpolation function to form the upper envelope of the vibration signal data X scaled ; similarly, find the vibration signal All the minimum value points of the data X scaled are fitted by cubic spline interpolation function to form the lower envelope of the vibration signal data X scaled ;
  • Step 303 Determine whether the signal sequence x 1 is an eigenmode function, if x 1 is not an eigenmode function, re-execute step 301 to re-decompose the vibration signal data; if x 1 is an eigenmode function, go to step 304;
  • the eigenmode function must meet two necessary conditions: one is that the number of local extreme points and zero-crossing points of the function must be equal, or at most one difference, in the entire time range; the other is that the local maximum at any time point
  • the average value of the envelope (upper envelope) and the envelope of the local minimum (lower envelope) must be zero or close to zero.
  • EMD decomposition feature For several eigenmode function signal sequences IMF or residual term RES obtained by decomposition, the energy of these components of each time slice is obtained through calculation, and the corresponding energy is called EMD decomposition feature.
  • the energy expression is as follows:
  • N represents the signal length in the delay time slice
  • c(i) represents the signal amplitude of the i-th data point in a certain IMF.
  • Step 400 Train a temporal convolutional network (TCN, Temporal Convolutional Network, temporal convolutional network) according to the extracted data features, and use a test set to test the temporal convolutional network;
  • TCN Temporal Convolutional Network
  • temporal convolutional network TCN, Temporal Convolutional Network, temporal convolutional network
  • Step 500 Use the time series convolutional network to output the prediction result of the remaining effective life of the device
  • Time sequence prediction requires that the prediction y t at time t can only be judged by the input x 1 to x t-1 before time t , and has nothing to do with x t+1 ,..., x T.
  • FIG. 2 is a structural element diagram of a time series convolutional network.
  • the TCN convolution operation performs an expansion convolution operation on the basis of one-dimensional convolution. The deeper the number of layers, the greater the expansion.
  • Filter F (f 1 ,f 2 ,...,f K )
  • sequence X ⁇ x 0 ,...,x T ⁇
  • the causal convolution at x t is:
  • the temporal convolutional network uses one-dimensional causal convolution and expansion convolution as the standard convolutional layer, and each two such convolutional layers are mapped to the identity (identical[identity]mapping ) Is packaged as a residual module (including RELU function). Introduce the jump connection of residual convolution and in order to add the same number of feature maps, that is, the number of channels, through 1*1 convolution to ensure the same shape of the two tensors.
  • the residual module stacks up a deep time series convolutional network, and at the end uses full convolution instead of the fully connected layer, so that the output and input dimensions are consistent, and end-to-end prediction is realized.
  • FIG. 3 is a schematic structural diagram of a system for predicting the remaining useful life of industrial equipment according to an embodiment of the present application.
  • the system for predicting the remaining useful life of industrial equipment in the embodiment of the application includes a data acquisition module, a data processing module, a feature extraction module, a model construction module, and a result output module.
  • Data collection module used to collect the original vibration signal data of the equipment
  • Data processing module used to normalize the original vibration signal data to obtain a sample data set, and divide the sample data set into a training set and a test set; among them, because different data have different specification units, the original vibration
  • the signal data is normalized by MinMax, and the normalization formula is as follows:
  • X is the original vibration signal data
  • X scaled is the normalized vibration signal data
  • X std is the normalized vibration signal data
  • X min is the mean square error and minimum value of the original vibration signal data, respectively
  • X max is the mean square error and minimum value of the original vibration signal data, respectively
  • EMD Empirical Mode Decomposition
  • the eigenmode function must meet two necessary conditions: one is that the number of local extreme points and zero-crossing points of the function must be equal or at most one difference in the entire time range; the other is that at any time point, the package of the local maximum
  • the average value of the envelope (upper envelope) and the envelope of the local minimum (lower envelope) must be zero or close to zero.
  • EMD decomposition feature For several eigenmode function signal sequences IMF or residual term RES obtained by decomposition, the energy of these components of each time slice is obtained through calculation, and the corresponding energy is called EMD decomposition feature.
  • the energy expression is as follows:
  • N represents the signal length in the delay time slice
  • c(i) represents the signal amplitude of the i-th data point in a certain IMF.
  • Model building module used to train the temporal convolutional network (TCN, Temporal Convolutional Network) according to the extracted data features, and use the test set to test the temporal convolutional network;
  • TCN Temporal Convolutional Network
  • FIG. 2 is a structural element diagram of a time series convolutional network.
  • the TCN convolution operation performs an expansion convolution operation on the basis of one-dimensional convolution. The deeper the number of layers, the greater the expansion.
  • Filter F (f 1 ,f 2 ,...,f K )
  • sequence X ⁇ x 0 ,...,x T ⁇
  • the causal convolution at x t is:
  • the temporal convolutional network uses one-dimensional causal convolution and expansion convolution as the standard convolutional layer, and each two such convolutional layers are mapped to the identity (identical[identity]mapping ) Is packaged as a residual module (including RELU function). Introduce the jump connection of residual convolution and in order to add the same number of feature maps, that is, the number of channels, through 1*1 convolution to ensure the same shape of the two tensors.
  • a deep sequential convolutional network is stacked by the residual module, and at the end, full convolution is used to replace the fully connected layer, so that the output and input dimensions are consistent, and end-to-end prediction is realized.
  • FIG. 5 is a schematic diagram of the hardware device structure of the method for predicting the remaining effective life of industrial equipment provided by an embodiment of the present application.
  • the device includes one or more processors and memory. Taking a processor as an example, the device may also include: an input system and an output system.
  • the processor, the memory, the input system, and the output system may be connected by a bus or other methods.
  • the connection by a bus is taken as an example.
  • the memory can be used to store non-transitory software programs, non-transitory computer executable programs, and modules.
  • the processor executes various functional applications and data processing of the electronic device by running non-transitory software programs, instructions, and modules stored in the memory, that is, realizing the processing methods of the foregoing method embodiments.
  • the memory may include a program storage area and a data storage area, where the program storage area can store an operating system and an application program required by at least one function; the data storage area can store data and the like.
  • the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid state storage devices.
  • the storage may optionally include storage remotely arranged with respect to the processor, and these remote storages may be connected to the processing system through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • the input system can receive input digital or character information, and generate signal input.
  • the output system may include display devices such as a display screen.
  • the one or more modules are stored in the memory, and when executed by the one or more processors, the following operations of any of the foregoing method embodiments are performed:
  • Step a Normalize the original vibration signal data of the device
  • Step b Use the empirical mode decomposition method to perform feature expansion on the normalized vibration signal data, and then extract the data feature of the vibration signal data;
  • Step c construct a time series convolutional network according to the extracted data features
  • Step d Use the time series convolutional network to output the remaining effective life prediction result of the device.
  • the embodiments of the present application provide a non-transitory (non-volatile) computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can perform the following operations:
  • Step a Normalize the original vibration signal data of the device
  • Step b Use the empirical mode decomposition method to perform feature expansion on the normalized vibration signal data, and then extract the data feature of the vibration signal data;
  • Step c construct a time series convolutional network according to the extracted data features
  • Step d Use the time series convolutional network to output the remaining effective life prediction result of the device.
  • the embodiment of the present application provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions are executed by a computer To make the computer do the following:
  • Step a Normalize the original vibration signal data of the device
  • Step b Use the empirical mode decomposition method to perform feature expansion on the normalized vibration signal data, and then extract the data feature of the vibration signal data;
  • Step c construct a time series convolutional network according to the extracted data features
  • Step d Use the time series convolutional network to output the remaining effective life prediction result of the device.
  • the method, system, and electronic device for predicting the remaining useful life of industrial equipment in the embodiments of the application extract and enrich the data features of the original signal through empirical model decomposition, and then use the time series convolutional neural network to train and predict the remaining useful life prediction model.
  • the application of time series convolutional network considers and utilizes the time series characteristics of the original signal of industrial equipment, so that the trained model learns data characteristics more accurately and representatively, has better generalization ability, and greatly improves the industrial equipment The prediction speed and accuracy of the remaining life are achievable in the actual manufacturing process.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • General Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Business, Economics & Management (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Hardware Design (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Geometry (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

一种工业设备剩余有效寿命预测方法、系统及电子设备。该方法包括:采集设备的原始振动信号数据(100);对原始振动信号数据进行归一化处理,得到样本数据集,并将样本数据集分为训练集和测试集(200);使用经验模态分解方式对训练集数据进行特征扩充后,提取训练集数据的数据特征(300);根据提取的数据特征训练时序卷积网络,并采用测试集对时序卷积网络进行测试(400);使用时序卷积网络输出设备的剩余有效寿命预测结果(500)。该方法通过经验模型分解提取并丰富原始信号的数据特征再用时序卷积神经网络训练预测得到剩余有效使用寿命预测模型,可以大大提高工业设备剩余寿命的预测速度和预测精度,在实际制造的过程中具有可实现性。

Description

一种工业设备剩余有效寿命预测方法、系统及电子设备 技术领域
本申请领属于设备故障预测技术域,特别涉及一种工业设备剩余有效寿命预测方法、系统及电子设备。
背景技术
工业生产过程中,设备的老化过程是不可避免的。为了保持竞争力,工业生产企业必须让其生产设备长期保持良好的工况,需要在减少设备维护费用的前提下提高设备的可用性、稳定性和安全性,而设备故障预测则成为其关键环节。准确的设备故障预测能够提前为设备维护人员提供设备安全预警,维护人员依据预警提前确定设备维护时间、减少由于设备故障产生的废品率、缩短维护周期,进而极大的减少企业的损失,具有极大的社会和经济效益。为此,工业生产企业需要采取适当的设备维护策略来满足这一需求。近年来,众多故障预测方法、工具以及应用涌现出来。
现有的故障预测方法主要包括基于模型的方法和数据驱动的方法:
1、基于模型的预测方法:主要根据经验知识和收集的数据建立数学模型来预测工业设备的剩余有效寿命,这类方法的评估和预测结果一般比较直观、准确,但是需要提前知道设备的退化信息和故障机理。
2、数据驱动的方法:旨在将设备的检测和运行数据转换成与设备退化有关的信息、系统运行状态及其退化机制模型。这类方法基本步骤为特征提取,特征选择,退化状态评估和剩余有效寿命预测,运用人工智能方法以及统计方 法等技术从采集传感器信号中提取有效的数据特征,然后学习设备的退化模式并预测设备的有效剩余寿命(RUL)。数据驱动方法可以应用于一些获取并处理检测数据易于构建物理和分析模型的场景。
综上所述,现有故障预测方法存在的问题在于:
1、基于模型的预测方法依赖于分析模型(代数或微分方程)来代表工业设备运转状态及其老化过程。这类方法虽能能够提供比较准确的结果,但真实的设备系统通常是非线性的,设备的老化机制通常是随机的且很难以分析模型的形式得到,在实际工况中,很难建立一种模型能够适应复杂的环境噪声和退化机理。
2、现有数据驱动的方法中,有用支持向量机(SVM)建模预测工业设备的剩余有效使用寿命,但是对于大规模的训练数据以及多分类的问题中,SVM耗费大量的运行内存和运算时间,导致预测的精度较低而且速度较慢。另外也有用卷积神经网络(CNN)建模分析用以对剩余有效寿命的预测,但是从工业设备收集的原始振动信号数据从本质上来看是一种时序运行状态数据,卷积神经网络(CNN)无法很好的解决时间序列预测,导致预测精度不高,另一方面,若直接将原始信号应用于卷积神经网络(CNN),模型是比较难以收敛,且预测速度较慢。
发明内容
本申请提供了一种工业设备剩余有效寿命预测方法、系统及电子设备,旨在至少在一定程度上解决现有技术中的上述技术问题之一。
为了解决上述问题,本申请提供了如下技术方案:
一种工业设备剩余有效寿命预测方法,包括以下步骤:
步骤a:对设备的原始振动信号数据进行归一化处理;
步骤b:使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
步骤c:根据提取的数据特征构建时序卷积网络;
步骤d:使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
本申请实施例采取的技术方案还包括:在所述步骤a中,所述归一化公式为:
X std=X-X min/X max-X min
X scaled=X std*(X max-X min)+X min
上述公式中,X为原始振动信号数据,X scaled为归一化处理后的振动信号数据,X std、X min、X max分别是原始振动信号数据的均方差、最小值和最大值,假设归一化后的振动信号数据X scaled=[x(1),x(2),...,x(N)]。
本申请实施例采取的技术方案还包括:在所述步骤b中,所述使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充具体包括:
步骤b1:找出振动信号数据X scaled的所有极大值点,并用三次样条插值函数拟合所有的极大值点,形成振动信号数据X scaled的上包络线;找出振动信号数据X scaled的所有极小值点,并通过三次样条插值函数拟合所有的极小值点,形成振动信号数据X scaled的下包络线;
步骤b2:根据上包络线和下包络线计算包络线均值x’,记作x’=[a1,a2,…,a(N)],将振动信号数据X scaled减去该包络线均值x’,得到一个新的信号序列x 1
步骤b3:判断信号序列x 1是否是本征模函数,如果x 1不是本征模函数, 重新执行步骤b1对振动信号数据进行重新分解;如果x 1是本征模函数,执行步骤b4;
步骤b4:将所述信号序列x 1表示为c=[c(1),c(2),…,c(N)],将归一化后振动信号数据X scaled和x 1相减,得到一个新的信号序列,并重新执行步骤b1至步骤b3,对新的信号继续往下分解,直到经过多次分解之后的x i是单调的,则经验模态分解结束,剩下的x i称之为余项;对于分解得到的若干个本征模函数信号序列或者余项,经过计算得到每个时间片的分量的能量,所述能量表达式为:
Figure PCTCN2019130600-appb-000001
上述公式中,N表示耽搁时间片内的信号长度;c(i)表示某本征模函数内第i个数据点的信号幅值。
本申请实施例采取的技术方案还包括:在所述步骤b3中,所述判断信号序列x 1是否是本征模函数的判断方式为:本征模函数必须满足两个必要条件:一是函数在整个时间范围内局部极值点和过零点的数目必须相等,或最多差一个;二是在任意时刻点,局部最大值的上包络线和局部最小值的下包络线的平均值必须为0或者接近0。
本申请实施例采取的技术方案还包括:在所述步骤d中,所述使用时序卷积网络输出设备的剩余有效寿命预测结果具体为:所述时序卷积网络是用一维因果卷积和扩张卷积作为标准卷积层,并将每两个这样的卷积层与恒等映射封装为一个包含RELU函数的残差模块,由残差模块堆叠起深度的时序卷积网络,并在最后使用全卷积代替全连接层,使输出与输入维度一致,实现端到端的预测。
本申请实施例采取的另一技术方案为:一种工业设备剩余有效寿命预测系 统,包括:
数据处理模块:用于对设备的原始振动信号数据进行归一化处理;
特征提取模块:用于使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
模型构建模块:用于根据提取的数据特征构建时序卷积网络;
结果输出模块:用于使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
本申请实施例采取的技术方案还包括:所述归一化公式为:
X std=X-X min/X max-X min
X scaled=X std*(X max-X min)+X min
上述公式中,X为原始振动信号数据,X scaled为归一化处理后的振动信号数据,X std、X min、X max分别是原始振动信号数据的均方差、最小值和最大值,假设归一化后的振动信号数据X scaled=[x(1),x(2),...,x(N)]。
本申请实施例采取的技术方案还包括:所述特征提取模块使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充具体:找出振动信号数据X scaled的所有极大值点,并用三次样条插值函数拟合所有的极大值点,形成振动信号数据X scaled的上包络线;找出振动信号数据X scaled的所有极小值点,并通过三次样条插值函数拟合所有的极小值点,形成振动信号数据X scaled的下包络线;根据上包络线和下包络线计算包络线均值x’,记作x’=[a1,a2,…,a(N)],将振动信号数据X scaled减去该包络线均值x’,得到一个新的信号序列x 1;判断信号序列x 1是否是本征模函数,如果x 1不是本征模函数,重新对振动信号数据进行重新分解;如果x 1是本征模函数,将所述信号序列x 1表示为c=[c(1),c(2),…,c(N)],将归一化后振动信号数据X scaled和x 1相减,得到一个新的 信号序列,并对新的信号继续往下分解,直到经过多次分解之后的x i是单调的,则经验模态分解结束,剩下的x i称之为余项;对于分解得到的若干个本征模函数信号序列或者余项,经过计算得到每个时间片的分量的能量,所述能量表达式为:
Figure PCTCN2019130600-appb-000002
上述公式中,N表示耽搁时间片内的信号长度;c(i)表示某本征模函数内第i个数据点的信号幅值。
本申请实施例采取的技术方案还包括:所述判断信号序列x 1是否是本征模函数的判断方式为:本征模函数必须满足两个必要条件:一是函数在整个时间范围内局部极值点和过零点的数目必须相等,或最多差一个;二是在任意时刻点,局部最大值的上包络线和局部最小值的下包络线的平均值必须为0或者接近0。
本申请实施例采取的技术方案还包括:所述时序卷积网络是用一维因果卷积和扩张卷积作为标准卷积层,并将每两个这样的卷积层与恒等映射封装为一个包含RELU函数的残差模块,由残差模块堆叠起深度的时序卷积网络,并在最后使用全卷积代替全连接层,使输出与输入维度一致,实现端到端的预测。
本申请实施例采取的又一技术方案为:一种电子设备,包括:
至少一个处理器;以及
与所述至少一个处理器通信连接的存储器;其中,
所述存储器存储有可被所述一个处理器执行的指令,所述指令被所述至少一个处理器执行,以使所述至少一个处理器能够执行上述的工业设备剩余有效寿命预测方法的以下操作:
步骤a:对设备的原始振动信号数据进行归一化处理;
步骤b:使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
步骤c:根据提取的数据特征构建时序卷积网络;
步骤d:使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
相对于现有技术,本申请实施例产生的有益效果在于:本申请实施例的工业设备剩余有效寿命预测方法、系统及电子设备通过经验模型分解提取并丰富原始信号的数据特征再用时序卷积神经网络训练预测得到剩余有效使用寿命预测模型。时序卷积网络的应用将工业设备原始信号的时序特点进行考虑并加以利用,使训练好的模型对数据特征的学习更加精确和具有代表性,具有更好的泛化能力,并大大提高工业设备剩余寿命的预测速度和预测精度,在实际制造的过程中具有可实现性。
附图说明
图1是本申请实施例的工业设备剩余有效寿命预测方法的流程图;
图2(a)-图2(c)是为时序卷积网络的结构元素图;
图3是本申请实施例的工业设备剩余有效寿命预测系统的结构示意图;
图4为验证结果示意图;
图5是本申请实施例提供的工业设备剩余有效寿命预测方法的硬件设备结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本申请进行进一步详细说明。应当理解,此处所描述的具体实施例仅 用以解释本申请,并不用于限定本申请。
请参阅图1,是本申请实施例的工业设备剩余有效寿命预测方法的流程图。本申请实施例的工业设备剩余有效寿命预测方法包括以下步骤:
步骤100:采集设备的原始振动信号数据;
步骤200:对原始振动信号数据进行归一化处理,得到样本数据集,并将样本数据集分为训练集和测试集;
步骤200中,由于不同数据的规格单位不同,因此需要对原始振动信号数据进行MinMax归一化处理,归一化公式如下:
X std=X-X min/X max-X min             (1)
X scaled=X std*(X max-X min)+X min     (2)
公式(1)、(2)中,X为原始振动信号数据,X scaled为归一化处理后的振动信号数据,X std、X min、X max分别是原始振动信号数据的均方差、最小值和最大值,假设归一化后的振动信号数据X scaled=[x(1),x(2),...,x(N)]。
步骤300:使用经验模态分解(Empirical Mode Decomposition,EMD)方式对训练集数据进行特征扩充后,提取训练集数据的数据特征;
步骤300中,对于振动信号数据为X=[x(1),x(2),…x(N)],经过每次EMD分解得到的函数称之为本征模函数(Intrinsic Mode Function,简称IMF),EMD分解出来的各IMF分量包含了振动信号数据的不同时间尺度的局部特征信号。特征扩充具体过程为:
步骤301:找出振动信号数据X scaled的所有极大值点,并用三次样条插值函数拟合所有的极大值点,形成振动信号数据X scaled的上包络线;同样,找出振动信号数据X scaled的所有极小值点,并通过三次样条插值函数拟合所有的极小值点,形成振动信号数据X scaled的下包络线;
步骤302:根据上包络线和下包络线计算包络线均值x’,记作x’=[a1,a2,…,a(N)],将振动信号数据X scaled减去该包络线均值x’,得到一个新的信号序列x 1
步骤303:判定信号序列x 1是否是本征模函数,如果x 1不是本征模函数,重新执行步骤301对振动信号数据进行重新分解;如果x 1是本征模函数,执行步骤304;
步骤303中,本征模函数必须满足两个必要条件:一是函数在整个时间范围内局部极值点和过零点的数目必须相等,或最多差一个;二是在任意时刻点,局部最大值的包络(上包络)和局部最小值的包络(下包络)平均值必须为0或者接近0。
步骤304:将该信号序列表示为c=[c(1),c(2),…,c(N)],将归一化后振动信号数据X scaled和x 1相减,得到一个新的信号序列,并重新执行步骤301至步骤303,对新的信号继续往下分解,直到经过多次分解之后的x i是单调的,则EMD分解结束,剩下的x i称之为余项(Residue,简称RES)。
对于分解得到的若干个本征模函数信号序列IMF或者余项RES,经过计算得到每个时间片的这些分量的能量,并将其对应的能量称之为EMD分解特征。能量表达式如下:
Figure PCTCN2019130600-appb-000003
公式(3)中,N表示耽搁时间片内的信号长度;c(i)表示某IMF内第i个数据点的信号幅值。
步骤400:根据提取的数据特征训练时序卷积网络(TCN,Temporal Convolutional Network,时序卷积网络),并采用测试集对时序卷积网络进行测试;
步骤500:使用时序卷积网络输出设备的剩余有效寿命预测结果;
步骤500中,时序卷积网络的输入为X={x 0,...,x T}的时序数据,网络的输出也是一样大小的Y={y 0,...,y T}的预测;时序预测要求对时刻t的预测y t只能通过t时刻之前的输入x 1到x t-1来判别,而与x t+1,…,x T无关。
具体请一并参阅图2,为时序卷积网络的结构元素图。在图2(a)中,TCN卷积操作在一维卷积的基础上进行了扩张卷积操作,层数越深,扩张的幅度越大。首先,通过第一层因果卷积层(相当于扩张因子d=1的扩张卷积卷积层)对数据集进行卷积,得到卷积特征。滤波器F=(f 1,f 2,…,f K),序列X={x 0,...,x T},在x t处的因果卷积为:
Figure PCTCN2019130600-appb-000004
如图2(b)所示,得到卷积特征后通过权重归一化,RELU非线性函数以及Dropout以实现正则化后得到第一层的卷积特征。然后通过第二次的扩张卷积对第一层的卷积特征进行卷积得到更深层次的卷积特征。扩张卷积运作在x s元素上,滤波器F=(f 1,f 2,…,f K),序列X={x 0,...,x T},在x s处扩张因子为d的扩张卷积为:
Figure PCTCN2019130600-appb-000005
得到更深层次的卷积特征后再次通过权重归一化,RELU非线性函数以及Dropout以实现正则化后得到第二层的卷积特征。如图2(c)所示,时序卷积网络是用一维因果卷积和扩张卷积作为标准卷积层,并将每两个这样的卷积层与恒等映射(identical[identity]mapping)封装为一个残差模块(包含RELU函数)。引入残差卷积的跳跃连接并且为了两层加和时特征图数量,即通道数数量相同,通过1*1的卷积进行元素合并来保证两个张量的形状相同。由残差模块堆叠起深度的时序卷积网络,并在最后使用全卷积代替全连接层,使输出与输入维度 一致,实现端到端的预测。
将模型的预测结果
Figure PCTCN2019130600-appb-000006
和真实的轴承剩余有效寿命ActRUL i通过两个度量指标进行计算,第i个测试数据的错误率由等式(3)计算得到:
Figure PCTCN2019130600-appb-000007
预测不足和过度预测将以不同的形式进行处理:好的预测性能是模型能够较早的预测出RUL(即:%Er i>0或
Figure PCTCN2019130600-appb-000008
),而较差的预测性能则是模型产生了高于实际RUL的预测值(即:%Er i<0或
Figure PCTCN2019130600-appb-000009
)。因此RUL的精度分值及最终所有的测试集预测RUL由等式(4)、(5)计算得到:
Figure PCTCN2019130600-appb-000010
Figure PCTCN2019130600-appb-000011
请参阅图3,是本申请实施例的工业设备剩余有效寿命预测系统的结构示意图。本申请实施例的工业设备剩余有效寿命预测系统包括数据采集模块、数据处理模块、特征提取模块、模型构建模块和结果输出模块。
数据采集模块:用于采集设备的原始振动信号数据;
数据处理模块:用于对原始振动信号数据进行归一化处理,得到样本数据集,并将样本数据集分为训练集和测试集;其中,由于不同数据的规格单位不同,因此需要对原始振动信号数据进行MinMax归一化处理,归一化公式如下:
X std=X-X min/X max-X miu             (1)
X scaled=X std*(X max-X min)+X min      (2)
公式(1)、(2)中,X为原始振动信号数据,X scaled为归一化处理后的振动信号数据,X std、X min、X max分别是原始振动信号数据的均方差、最小值和最 大值,假设归一化后的振动信号数据X scaled=[x(1),x(2),...,x(N)]。
特征提取模块:用于使用经验模态分解(Empirical Mode Decomposition,EMD)方式对训练集数据进行特征扩充后,提取训练集数据的数据特征;其中,对于振动信号数据为X=[x(1),x(2),…x(N)],经过每次EMD分解得到的函数称之为本征模函数(Intrinsic Mode Function,简称IMF),EMD分解出来的各IMF分量包含了振动信号数据的不同时间尺度的局部特征信号。特征扩充具体过程为:
1:找出振动信号数据X scaled的所有极大值点,并用三次样条插值函数拟合所有的极大值点,形成振动信号数据X scaled的上包络线;同样,找出振动信号数据X scaled的所有极小值点,并通过三次样条插值函数拟合所有的极小值点,形成振动信号数据X scaled的下包络线;
2:根据上包络线和下包络线计算包络线均值x’,记作x’=[a1,a2,…,a(N)],将振动信号数据X scaled减去该包络线均值x’,得到一个新的信号序列x 1
3:判定信号序列x 1是否是本征模函数,如果x 1不是本征模函数,重新对振动信号数据进行重新分解;如果x 1是本征模函数,将该信号序列表示为c=[c(1),c(2),…,c(N)],将归一化后振动信号数据X scaled和x 1相减,得到一个新的信号序列,并重新执行步骤301至步骤303,对新的信号继续往下分解,直到经过多次分解之后的x i是单调的,则EMD分解结束,剩下的x i称之为余项(Residue,简称RES)。其中,本征模函数必须满足两个必要条件:一是函数在整个时间范围内局部极值点和过零点的数目必须相等,或最多差一个;二是在任意时刻点,局部最大值的包络(上包络)和局部最小值的包络(下包络)平均值必须为0或者接近0。
对于分解得到的若干个本征模函数信号序列IMF或者余项RES,经过计算得到每个时间片的这些分量的能量,并将其对应的能量称之为EMD分解特征。能量表达式如下:
Figure PCTCN2019130600-appb-000012
公式(3)中,N表示耽搁时间片内的信号长度;c(i)表示某IMF内第i个数据点的信号幅值。
模型构建模块:用于根据提取的数据特征训练时序卷积网络(TCN,Temporal Convolutional Network,时序卷积网络),并采用测试集对时序卷积网络进行测试;
结果输出模块:用于使用时序卷积网络输出设备的剩余有效寿命预测结果;其中,时序卷积网络的输入为X={x 0,...,x T}的时序数据,网络的输出也是一样大小的Y={y 0,...,y T}的预测;时序预测要求对时刻t的预测y t只能通过t时刻之前的输入x 1到x t-1来判别,而与x t+1,…,x T无关。
具体请一并参阅图2,为时序卷积网络的结构元素图。在图2(a)中,TCN卷积操作在一维卷积的基础上进行了扩张卷积操作,层数越深,扩张的幅度越大。首先,通过第一层因果卷积层(相当于扩张因子d=1的扩张卷积卷积层)对数据集进行卷积,得到卷积特征。滤波器F=(f 1,f 2,…,f K),序列X={x 0,...,x T},在x t处的因果卷积为:
Figure PCTCN2019130600-appb-000013
如图2(b)所示,得到卷积特征后通过权重归一化,RELU非线性函数以及Dropout以实现正则化后得到第一层的卷积特征。然后通过第二次的扩张卷积对第一层的卷积特征进行卷积得到更深层次的卷积特征。扩张卷积运作在x s元素上,滤波器F=(f 1,f 2,…,f K),序列X={x 0,...,x T},在x s处扩张因子为 d的扩张卷积为:
Figure PCTCN2019130600-appb-000014
得到更深层次的卷积特征后再次通过权重归一化,RELU非线性函数以及Dropout以实现正则化后得到第二层的卷积特征。如图2(c)所示,时序卷积网络是用一维因果卷积和扩张卷积作为标准卷积层,并将每两个这样的卷积层与恒等映射(identical[identity]mapping)封装为一个残差模块(包含RELU函数)。引入残差卷积的跳跃连接并且为了两层加和时特征图数量,即通道数数量相同,通过1*1的卷积进行元素合并来保证两个张量的形状相同。由残差模块堆叠起深度的时序卷积网络,并在最后使用全卷积代替全连接层,使输出与输入维度一致,实现端到端的预测。
将模型的预测结果
Figure PCTCN2019130600-appb-000015
和真实的轴承剩余有效寿命ActRUL i通过两个度量指标进行计算,第i个测试数据的错误率由等式(3)计算得到:
Figure PCTCN2019130600-appb-000016
预测不足和过度预测将以不同的形式进行处理:好的预测性能是模型能够较早的预测出RUL(即:%Er i>0或
Figure PCTCN2019130600-appb-000017
),而较差的预测性能则是模型产生了高于实际RUL的预测值(即:%Er i<0或
Figure PCTCN2019130600-appb-000018
)。因此RUL的精度分值及最终所有的测试集预测RUL由等式(4)、(5)计算得到:
Figure PCTCN2019130600-appb-000019
Figure PCTCN2019130600-appb-000020
本申请已在PRONOSTIA实验装置上进行验证。验证结果如图4所示,本申请与CNN和LSTM实验效果相比分别有15%到20%的效果提升,与直接使用时序卷积网络对原始信号处理的结果也有一定的提升。
图5是本申请实施例提供的工业设备剩余有效寿命预测方法的硬件设备结构示意图。如图5所示,该设备包括一个或多个处理器以及存储器。以一个处理器为例,该设备还可以包括:输入系统和输出系统。
处理器、存储器、输入系统和输出系统可以通过总线或者其他方式连接,图5中以通过总线连接为例。
存储器作为一种非暂态计算机可读存储介质,可用于存储非暂态软件程序、非暂态计算机可执行程序以及模块。处理器通过运行存储在存储器中的非暂态软件程序、指令以及模块,从而执行电子设备的各种功能应用以及数据处理,即实现上述方法实施例的处理方法。
存储器可以包括存储程序区和存储数据区,其中,存储程序区可存储操作系统、至少一个功能所需要的应用程序;存储数据区可存储数据等。此外,存储器可以包括高速随机存取存储器,还可以包括非暂态存储器,例如至少一个磁盘存储器件、闪存器件、或其他非暂态固态存储器件。在一些实施例中,存储器可选包括相对于处理器远程设置的存储器,这些远程存储器可以通过网络连接至处理系统。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入系统可接收输入的数字或字符信息,以及产生信号输入。输出系统可包括显示屏等显示设备。
所述一个或者多个模块存储在所述存储器中,当被所述一个或者多个处理器执行时,执行上述任一方法实施例的以下操作:
步骤a:对设备的原始振动信号数据进行归一化处理;
步骤b:使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
步骤c:根据提取的数据特征构建时序卷积网络;
步骤d:使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
上述产品可执行本申请实施例所提供的方法,具备执行方法相应的功能模块和有益效果。未在本实施例中详尽描述的技术细节,可参见本申请实施例提供的方法。
本申请实施例提供了一种非暂态(非易失性)计算机存储介质,所述计算机存储介质存储有计算机可执行指令,该计算机可执行指令可执行以下操作:
步骤a:对设备的原始振动信号数据进行归一化处理;
步骤b:使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
步骤c:根据提取的数据特征构建时序卷积网络;
步骤d:使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
本申请实施例提供了一种计算机程序产品,所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算机程序,所述计算机程序包括程序指令,当所述程序指令被计算机执行时,使所述计算机执行以下操作:
步骤a:对设备的原始振动信号数据进行归一化处理;
步骤b:使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
步骤c:根据提取的数据特征构建时序卷积网络;
步骤d:使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
本申请实施例的工业设备剩余有效寿命预测方法、系统及电子设备通过经验模型分解提取并丰富原始信号的数据特征再用时序卷积神经网络训练预测得到剩余有效使用寿命预测模型。时序卷积网络的应用将工业设备原始信号的时序特点进行考虑并加以利用,使训练好的模型对数据特征的学习更加精确和具有代表性,具有更好的泛化能力,并大大提高工业设备剩余寿命的预测速度和预测精度,在实际制造的过程中具有可实现性。
对所公开的实施例的上述说明,使本领域专业技术人员能够实现或使用本申请。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的,本申请中所定义的一般原理可以在不脱离本申请的精神或范围的情况下,在其它实施例中实现。因此,本申请将不会被限制于本申请所示的这些实施例,而是要符合与本申请所公开的原理和新颖特点相一致的最宽的范围。

Claims (11)

  1. 一种工业设备剩余有效寿命预测方法,其特征在于,包括以下步骤:
    步骤a:对设备的原始振动信号数据进行归一化处理;
    步骤b:使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
    步骤c:根据提取的数据特征构建时序卷积网络;
    步骤d:使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
  2. 根据权利要求1所述的工业设备剩余有效寿命预测方法,其特征在于,在所述步骤a中,所述归一化公式为:
    X std=X-X min/X max-X min
    X scaled=X std*(X max-X min)+X min
    上述公式中,X为原始振动信号数据,X scaled为归一化处理后的振动信号数据,X std、X min、X max分别是原始振动信号数据的均方差、最小值和最大值,假设归一化后的振动信号数据X scaled=[x(1),x(2),...,x(N)]。
  3. 根据权利要求2所述的工业设备剩余有效寿命预测方法,其特征在于,在所述步骤b中,所述使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充具体包括:
    步骤b1:找出振动信号数据X scaled的所有极大值点,并用三次样条插值函数拟合所有的极大值点,形成振动信号数据X scaled的上包络线;找出振动信号数据X scaled的所有极小值点,并通过三次样条插值函数拟合所有的极小值点,形成振动信号数据X scaled的下包络线;
    步骤b2:根据上包络线和下包络线计算包络线均值x’,记作x’=[a1,a2,…, a(N)],将振动信号数据X scaled减去该包络线均值x’,得到一个新的信号序列x 1
    步骤b3:判断信号序列x 1是否是本征模函数,如果x 1不是本征模函数,重新执行步骤b1对振动信号数据进行重新分解;如果x 1是本征模函数,执行步骤b4;
    步骤b4:将所述信号序列x 1表示为c=[c(1),c(2),…,c(N)],将归一化后振动信号数据X scaled和x 1相减,得到一个新的信号序列,并重新执行步骤b1至步骤b3,对新的信号继续往下分解,直到经过多次分解之后的x i是单调的,则经验模态分解结束,剩下的x i称之为余项;对于分解得到的若干个本征模函数信号序列或者余项,经过计算得到每个时间片的分量的能量,所述能量表达式为:
    Figure PCTCN2019130600-appb-100001
    上述公式中,N表示耽搁时间片内的信号长度;c(i)表示某本征模函数内第i个数据点的信号幅值。
  4. 根据权利要求3所述的工业设备剩余有效寿命预测方法,其特征在于,在所述步骤b3中,所述判断信号序列x 1是否是本征模函数的判断方式为:本征模函数必须满足两个必要条件:一是函数在整个时间范围内局部极值点和过零点的数目必须相等,或最多差一个;二是在任意时刻点,局部最大值的上包络线和局部最小值的下包络线的平均值必须为0或者接近0。
  5. 根据权利要求1至4任一项所述的工业设备剩余有效寿命预测方法,其特征在于,在所述步骤d中,所述使用时序卷积网络输出设备的剩余有效寿命预测结果具体为:所述时序卷积网络是用一维因果卷积和扩张卷积作为标准卷积层,并将每两个这样的卷积层与恒等映射封装为一个包含RELU函数的残差模块,由残差模块堆叠起深度的时序卷积网络,并在最后使用全卷积代替全连接层,使输出与输入维度一致,实现端到端的预测。
  6. 一种工业设备剩余有效寿命预测系统,其特征在于,包括:
    数据处理模块:用于对设备的原始振动信号数据进行归一化处理;
    特征提取模块:用于使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
    模型构建模块:用于根据提取的数据特征构建时序卷积网络;
    结果输出模块:用于使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
  7. 根据权利要求6所述的工业设备剩余有效寿命预测系统,其特征在于,所述归一化公式为:
    X std=X-X min/X max-X min
    X scaled=X std*(X max-X min)+X min
    上述公式中,X为原始振动信号数据,X scaled为归一化处理后的振动信号数据,X std、X min、X max分别是原始振动信号数据的均方差、最小值和最大值,假设归一化后的振动信号数据X scaled=[x(1),x(2),...,x(N)]。
  8. 根据权利要求7所述的工业设备剩余有效寿命预测系统,其特征在于,所述特征提取模块使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充具体:找出振动信号数据X scaled的所有极大值点,并用三次样条插值函数拟合所有的极大值点,形成振动信号数据X scaled的上包络线;找出振动信号数据X scaled的所有极小值点,并通过三次样条插值函数拟合所有的极小值点,形成振动信号数据X scaled的下包络线;根据上包络线和下包络线计算包络线均值x’,记作x’=[a1,a2,…,a(N)],将振动信号数据X scaled减去该包络线均值x’,得到一个新的信号序列x 1;判断信号序列x 1是否是本征模函数,如果x 1不是本征模函数,重新对振动信号数据进行重新分解;如果x 1是本征模函数,将所述信号 序列x 1表示为c=[c(1),c(2),…,c(N)],将归一化后振动信号数据X scaled和x 1相减,得到一个新的信号序列,并对新的信号继续往下分解,直到经过多次分解之后的x i是单调的,则经验模态分解结束,剩下的x i称之为余项;对于分解得到的若干个本征模函数信号序列或者余项,经过计算得到每个时间片的分量的能量,所述能量表达式为:
    Figure PCTCN2019130600-appb-100002
    上述公式中,N表示耽搁时间片内的信号长度;c(i)表示某本征模函数内第i个数据点的信号幅值。
  9. 根据权利要求8所述的工业设备剩余有效寿命预测系统,其特征在于,所述判断信号序列x 1是否是本征模函数的判断方式为:本征模函数必须满足两个必要条件:一是函数在整个时间范围内局部极值点和过零点的数目必须相等,或最多差一个;二是在任意时刻点,局部最大值的上包络线和局部最小值的下包络线的平均值必须为0或者接近0。
  10. 根据权利要求6至9任一项所述的工业设备剩余有效寿命预测方法,其特征在于,所述时序卷积网络是用一维因果卷积和扩张卷积作为标准卷积层,并将每两个这样的卷积层与恒等映射封装为一个包含RELU函数的残差模块,由残差模块堆叠起深度的时序卷积网络,并在最后使用全卷积代替全连接层,使输出与输入维度一致,实现端到端的预测。
  11. 一种电子设备,包括:
    至少一个处理器;以及
    与所述至少一个处理器通信连接的存储器;其中,
    所述存储器存储有可被所述一个处理器执行的指令,所述指令被所述至少一个处理器执行,以使所述至少一个处理器能够执行上述1至5任一项所述的工业 设备剩余有效寿命预测方法的以下操作:
    步骤a:对设备的原始振动信号数据进行归一化处理;
    步骤b:使用经验模态分解方式对归一化处理后的振动信号数据进行特征扩充后,提取所述振动信号数据的数据特征;
    步骤c:根据提取的数据特征构建时序卷积网络;
    步骤d:使用所述时序卷积网络输出设备的剩余有效寿命预测结果。
PCT/CN2019/130600 2019-07-02 2019-12-31 一种工业设备剩余有效寿命预测方法、系统及电子设备 WO2021000556A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910588139.5A CN110377984B (zh) 2019-07-02 2019-07-02 一种工业设备剩余有效寿命预测方法、系统及电子设备
CN201910588139.5 2019-07-02

Publications (1)

Publication Number Publication Date
WO2021000556A1 true WO2021000556A1 (zh) 2021-01-07

Family

ID=68251547

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/130600 WO2021000556A1 (zh) 2019-07-02 2019-12-31 一种工业设备剩余有效寿命预测方法、系统及电子设备

Country Status (2)

Country Link
CN (1) CN110377984B (zh)
WO (1) WO2021000556A1 (zh)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113255780A (zh) * 2021-05-28 2021-08-13 润联软件系统(深圳)有限公司 一种减速箱故障预测方法、装置、计算机设备及存储介质
CN113487084A (zh) * 2021-07-06 2021-10-08 新智数字科技有限公司 一种设备使用寿命的预测方法、装置、计算机设备及计算机可读存储介质
CN113657012A (zh) * 2021-07-21 2021-11-16 西安理工大学 一种基于tcn和粒子滤波的关键设备剩余寿命预测方法
CN113779882A (zh) * 2021-09-10 2021-12-10 中国石油大学(北京) 设备剩余使用寿命的预测方法、装置、设备及存储介质
CN114580101A (zh) * 2022-02-23 2022-06-03 华中科技大学 一种旋转机械剩余使用寿命预测方法及系统
CN114942066A (zh) * 2022-05-26 2022-08-26 清华大学 电主轴误差间接预测方法及装置
CN116992393A (zh) * 2023-09-27 2023-11-03 联通(江苏)产业互联网有限公司 一种基于工业物联网的安全生产监测方法
CN117114516A (zh) * 2023-10-25 2023-11-24 湖南省水务规划设计院有限公司 一种长距离小断面引水隧洞安全评估方法
CN117331339A (zh) * 2023-12-01 2024-01-02 南京华视智能科技股份有限公司 基于时序神经网络模型的涂布机模头电机控制方法及装置
CN118035733A (zh) * 2024-04-12 2024-05-14 聊城市中瑞轴承配件股份有限公司 一种轴承保持架的寿命检测方法及装置

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110377984B (zh) * 2019-07-02 2023-08-18 中国科学院深圳先进技术研究院 一种工业设备剩余有效寿命预测方法、系统及电子设备
CN110942194A (zh) * 2019-11-27 2020-03-31 徐州上若科技有限公司 一种基于tcn的风电预测误差区间评估方法
TWI753338B (zh) * 2019-12-23 2022-01-21 財團法人工業技術研究院 資料整合方法及資料整合系統
CN111695452B (zh) * 2020-05-28 2023-08-18 国网河北省电力有限公司电力科学研究院 基于rbf神经网络的并联电抗器内部老化程度评估方法
CN111783362B (zh) * 2020-07-09 2022-07-26 哈尔滨工程大学 一种电动闸阀剩余使用寿命确定方法及系统
CN112785051A (zh) * 2021-01-14 2021-05-11 武汉纺织大学 一种基于emd和tcn相结合的云资源预测方法
CN115221769B (zh) * 2021-04-15 2023-10-10 广州中国科学院先进技术研究所 一种故障预测方法、系统、电子设备和存储介质
CN115476199A (zh) * 2021-05-31 2022-12-16 沈阳中科数控技术股份有限公司 一种基于时间卷积网络的数控机床刀具磨损预测方法
CN113486578B (zh) * 2021-06-28 2023-10-20 北京科技大学 一种工业过程中设备剩余寿命的预测方法
CN113570138B (zh) * 2021-07-28 2022-04-26 朗坤智慧科技股份有限公司 一种时间卷积网络的设备剩余使用寿命预测方法及装置
CN113837433A (zh) * 2021-08-12 2021-12-24 裴平 有源数字阵列雷达装备射频系统剩余使用寿命预测方法
CN113743541B (zh) * 2021-11-04 2022-02-08 华中科技大学 一种基于退化模式预测船用动力系统轴承剩余寿命的方法
CN115423221B (zh) * 2022-11-03 2023-02-07 北京建筑大学 一种设施运行趋势预测方法
CN115600512B (zh) * 2022-12-01 2023-04-14 深圳先进技术研究院 一种基于分布式学习的刀具寿命预测方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2008132066A1 (en) * 2007-04-27 2008-11-06 Siemens Aktiengesellschaft A method for computer-assisted learning of one or more neural networks
CN101846753A (zh) * 2010-04-29 2010-09-29 南京信息工程大学 基于经验模态分解和支持向量机的气候时序预测方法
CN106126896A (zh) * 2016-06-20 2016-11-16 中国地质大学(武汉) 基于经验模态分解和深度学习的混合模型风速预测方法及系统
CN109657851A (zh) * 2018-12-12 2019-04-19 哈尔滨工业大学(深圳) 事件驱动型的股价预测方法、装置、设备及存储介质
CN110377984A (zh) * 2019-07-02 2019-10-25 中国科学院深圳先进技术研究院 一种工业设备剩余有效寿命预测方法、系统及电子设备

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9335290B2 (en) * 2011-12-23 2016-05-10 Abbott Point Of Care, Inc. Integrated test device for optical and electrochemical assays
CN109460618B (zh) * 2018-11-13 2023-02-10 华中科技大学 一种滚动轴承剩余寿命在线预测方法及系统
CN109726524B (zh) * 2019-03-01 2022-11-01 哈尔滨理工大学 一种基于cnn和lstm的滚动轴承剩余使用寿命预测方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2008132066A1 (en) * 2007-04-27 2008-11-06 Siemens Aktiengesellschaft A method for computer-assisted learning of one or more neural networks
CN101846753A (zh) * 2010-04-29 2010-09-29 南京信息工程大学 基于经验模态分解和支持向量机的气候时序预测方法
CN106126896A (zh) * 2016-06-20 2016-11-16 中国地质大学(武汉) 基于经验模态分解和深度学习的混合模型风速预测方法及系统
CN109657851A (zh) * 2018-12-12 2019-04-19 哈尔滨工业大学(深圳) 事件驱动型的股价预测方法、装置、设备及存储介质
CN110377984A (zh) * 2019-07-02 2019-10-25 中国科学院深圳先进技术研究院 一种工业设备剩余有效寿命预测方法、系统及电子设备

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
SHUMIN DENG; NINGYU ZHANG; WEN ZHANG; JIAOYAN CHEN; JEFF Z PAN; HUAJUN CHEN: "Knowledge-Driven Stock Trend Prediction and Explanation via Temporal Convolutional Network", WWW '19: COMPANION PROCEEDINGS OF THE 2019 WORLD WIDE WEB CONFERENCE, 13 May 2019 (2019-05-13), pages 678 - 685, XP058436916, DOI: 10.1145/3308560.3317701 *
WENTAO MAO, JIANLIANG HE, JIAMEI TANG , YUAN LI: "Predicting remaining useful life of rolling bearings based on deep feature representation and long short-term memory neural network", ADVANCES IN MECHANICAL ENGINEERING, vol. 10, no. 12, 11 December 2018 (2018-12-11), pages 1 - 18, XP055770875, ISSN: 1687-8140, DOI: 10.1177/1687814018817184 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113255780B (zh) * 2021-05-28 2024-05-03 润联智能科技股份有限公司 一种减速箱故障预测方法、装置、计算机设备及存储介质
CN113255780A (zh) * 2021-05-28 2021-08-13 润联软件系统(深圳)有限公司 一种减速箱故障预测方法、装置、计算机设备及存储介质
CN113487084A (zh) * 2021-07-06 2021-10-08 新智数字科技有限公司 一种设备使用寿命的预测方法、装置、计算机设备及计算机可读存储介质
CN113657012A (zh) * 2021-07-21 2021-11-16 西安理工大学 一种基于tcn和粒子滤波的关键设备剩余寿命预测方法
CN113779882A (zh) * 2021-09-10 2021-12-10 中国石油大学(北京) 设备剩余使用寿命的预测方法、装置、设备及存储介质
CN113779882B (zh) * 2021-09-10 2024-05-17 中国石油大学(北京) 设备剩余使用寿命的预测方法、装置、设备及存储介质
CN114580101A (zh) * 2022-02-23 2022-06-03 华中科技大学 一种旋转机械剩余使用寿命预测方法及系统
CN114942066B (zh) * 2022-05-26 2023-08-04 清华大学 电主轴误差间接预测方法及装置
CN114942066A (zh) * 2022-05-26 2022-08-26 清华大学 电主轴误差间接预测方法及装置
CN116992393A (zh) * 2023-09-27 2023-11-03 联通(江苏)产业互联网有限公司 一种基于工业物联网的安全生产监测方法
CN116992393B (zh) * 2023-09-27 2023-12-22 联通(江苏)产业互联网有限公司 一种基于工业物联网的安全生产监测方法
CN117114516A (zh) * 2023-10-25 2023-11-24 湖南省水务规划设计院有限公司 一种长距离小断面引水隧洞安全评估方法
CN117114516B (zh) * 2023-10-25 2024-05-28 湖南省水务规划设计院有限公司 一种长距离小断面引水隧洞安全评估方法
CN117331339A (zh) * 2023-12-01 2024-01-02 南京华视智能科技股份有限公司 基于时序神经网络模型的涂布机模头电机控制方法及装置
CN117331339B (zh) * 2023-12-01 2024-02-06 南京华视智能科技股份有限公司 基于时序神经网络模型的涂布机模头电机控制方法及装置
CN118035733A (zh) * 2024-04-12 2024-05-14 聊城市中瑞轴承配件股份有限公司 一种轴承保持架的寿命检测方法及装置

Also Published As

Publication number Publication date
CN110377984A (zh) 2019-10-25
CN110377984B (zh) 2023-08-18

Similar Documents

Publication Publication Date Title
WO2021000556A1 (zh) 一种工业设备剩余有效寿命预测方法、系统及电子设备
CN107480789B (zh) 一种深度学习模型的高效转换方法及装置
WO2016101628A1 (zh) 一种数据建模中的数据处理方法及装置
CN107632590A (zh) 一种基于优先级的底事件排序方法
CN115204491A (zh) 基于数字孪生与lstm的生产线工况预测方法及系统
CN113626612A (zh) 一种基于知识图谱推理的预测方法和系统
CN115150471B (zh) 数据处理方法、装置、设备、存储介质及程序产品
Ye et al. Big data processing framework for manufacturing
CN117596191A (zh) 一种电力物联网异常检测方法、装置及存储介质
CN112926269A (zh) 电厂边缘节点数据分组与清洗的方法及系统
CN102184136A (zh) 一种验证aadl模型运行状态与需求一致性的方法
CN116975081A (zh) 一种日志诊断集更新方法、装置、设备及存储介质
CN115186738B (zh) 模型训练方法、装置和存储介质
CN115965160A (zh) 一种数据中心能耗预测方法、装置、存储介质及电子设备
CN115935822A (zh) 计算机参数长期退化趋势预测方法、系统、设备及介质
CN114385403A (zh) 基于双层知识图谱架构的分布式协同故障诊断方法
CN117435901B (zh) 一种工业互联网数据获取方法、系统、终端及储存介质
CN117560275B (zh) 基于图神经网络模型的微服务系统根因定位方法及装置
CN114237915B (zh) 分布式环境下基于机器学习的数据修复方法、装置及设备
CN116723083B (zh) 一种云服务器在线故障诊断方法及装置
CN113139490B (zh) 一种图像特征匹配方法、装置、计算机设备及存储介质
CN117435870B (zh) 一种负荷数据实时填充方法、系统、设备和介质
CN116307159A (zh) 一种负荷预测方法、装置、电子设备和存储介质
Cui et al. Data Link Fault Location Model Based on Machine Learning
Li et al. FAITH: Frequency-domain Attention In Two Horizons for Time Series Forecasting

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19936294

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19936294

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19936294

Country of ref document: EP

Kind code of ref document: A1