WO2021042749A1 - 一种基于有监督lle算法的轴承故障诊断方法及装置 - Google Patents
一种基于有监督lle算法的轴承故障诊断方法及装置 Download PDFInfo
- Publication number
- WO2021042749A1 WO2021042749A1 PCT/CN2020/087799 CN2020087799W WO2021042749A1 WO 2021042749 A1 WO2021042749 A1 WO 2021042749A1 CN 2020087799 W CN2020087799 W CN 2020087799W WO 2021042749 A1 WO2021042749 A1 WO 2021042749A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- training data
- fault
- dimensionality reduction
- preferred
- data
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/04—Bearings
- G01M13/045—Acoustic or vibration analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
Definitions
- the invention relates to the technical field of fault diagnosis, in particular to a bearing fault diagnosis method and device based on a supervised LLE algorithm.
- bearing fault diagnosis technology has initially formed a relatively complete subject system.
- vibration diagnosis technology has become the mainstream technology for bearing fault diagnosis.
- the rapid progress of computer technology and signal information processing technology has greatly promoted the development of bearing fault diagnosis and monitoring technology in the direction of science and practicality.
- the purpose of the present invention is to provide a bearing fault diagnosis method and device based on the supervised LLE algorithm, aiming to improve the online prediction rate of bearing fault diagnosis.
- a bearing fault diagnosis method based on supervised LLE algorithm including:
- training data is historical data characterizing bearing vibration signals, and extracting characteristic values of the training data and fault types corresponding to the characteristic values;
- the ratio of the inter-class dispersion to the intra-class dispersion of all fault types is the largest
- the characteristic values include vibration displacement, vibration velocity, vibration acceleration, and high-frequency acceleration
- the failure types include wear failure, fatigue failure, and corrosion failure.
- the determining the preferred dimensionality reduction training data of the training data includes:
- the dimensionality reduction training data corresponding to the preferred number of neighbors and the preferred fault dimension is used as the preferred dimensionality reduction training data.
- the use of the LLE algorithm to reduce the dimensionality of the training data to obtain the dimensionality reduction training data, and to determine the preferred number of neighbors and the preferred fault dimension of the dimensionality reduction training data includes:
- Step 310 Set the value range of the number of neighbors p and the value range of the fault dimension q;
- Step 320 Select a p value and a q value as a parameter group, and form all the parameter groups into a parameter set, and the parameter set includes all combinations of the p value and the q value;
- Step 330 Select a parameter group in sequence as the number of neighbors p and the fault dimension q of the training data
- Step 350 Calculate the evaluation index F by using the reduced dimensionality data set and the fault set, specifically:
- the intra-class dispersion matrix S i of all categories is calculated by the following formula:
- the inter-class dispersion matrix is calculated by the following formula:
- the evaluation index F is calculated by the following formula:
- Step 360 Judge whether all parameter groups in the parameter set have calculated evaluation indexes, if not, jump to step 330, if yes, perform the following steps;
- Step 370 Compare the size of the evaluation index in each parameter group, select the parameter group with the largest evaluation index as the preferred parameter group, the p value of the parameter group as the preferred number of neighbors, and the q value of the parameter group as the preferred fault dimension.
- the dimensionality reduction of the test data received in real time to obtain the dimensionality reduction test data includes:
- the preferred number of neighbors is used as the number of neighbors of the test data
- the preferred fault dimension is used as the fault dimension of the test data
- the LLE algorithm is used to reduce the dimensionality of the test data to obtain the reduced dimensionality test data.
- a bearing fault diagnosis device based on a supervised LLE algorithm.
- the device includes a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor executes the computer
- the program runs in the modules of the following devices:
- An extraction module configured to obtain training data, the training data being historical data characterizing bearing vibration signals, and extracting the characteristic value of the training data and the fault type corresponding to the characteristic value;
- the determining module is used to determine the preferred dimensionality reduction training data of the training data.
- the ratio of the inter-class dispersion to the intra-class dispersion of all fault types is the largest;
- the dimensionality reduction module is used to reduce the dimensionality of the test data received in real time to obtain the dimensionality reduction test data;
- the diagnosis module is configured to calculate the probability value of the dimensionality reduction data under each fault type according to the mean value and the covariance matrix, and use the fault type with the largest probability value as the fault type for bearing fault diagnosis.
- the characteristic values include vibration displacement, vibration velocity, vibration acceleration, and high-frequency acceleration
- the failure types include wear failure, fatigue failure, and corrosion failure.
- determining module is specifically used for:
- the dimensionality reduction training data corresponding to the preferred number of neighbors and the preferred fault dimension is used as the preferred dimensionality reduction training data.
- the present invention discloses a bearing fault diagnosis method and device based on a supervised LLE algorithm. Firstly, training data is acquired, and the training data is historical data representing bearing vibration signals, and the characteristics of the training data are extracted Value and the fault type corresponding to the eigenvalue, and then determine the preferred dimensionality reduction training data of the training data, and then calculate the mean value and covariance matrix corresponding to each fault type in the preferred dimensionality reduction training data. The dimensionality reduction of the test data is performed to obtain the dimensionality reduction test data, the probability value of the dimensionality reduction data under each fault type is calculated according to the mean value and the covariance matrix, and the fault type with the largest probability value is used as the fault type for bearing fault diagnosis. The invention improves the online prediction rate of bearing fault diagnosis.
- FIG. 1 is a schematic flowchart of a bearing fault diagnosis method based on a supervised LLE algorithm according to an embodiment of the present invention
- FIG. 2 is a schematic flowchart of step S200 in an embodiment of the present invention.
- FIG. 3 is a schematic flowchart of step S210 in an embodiment of the present invention.
- Fig. 4 is a schematic structural diagram of a bearing fault diagnosis device based on a supervised LLE algorithm according to an embodiment of the present invention.
- a bearing fault diagnosis method based on a supervised LLE algorithm includes the following steps:
- Step S100 Obtain training data, where the training data is historical data characterizing bearing vibration signals, and extract the characteristic value of the training data and the fault type corresponding to the characteristic value;
- Step S200 Determine preferred dimensionality reduction training data of the training data.
- the ratio of the inter-class dispersion to the intra-class dispersion of all fault types is the largest;
- Step S300 Calculate the mean value and the covariance matrix corresponding to each fault type in the preferred dimensionality reduction training data
- Step S400 Perform dimensionality reduction on the test data received in real time to obtain dimensionality reduction test data
- Step S500 Calculate the probability value of the dimensionality reduction data under each fault type according to the mean value and the covariance matrix, and use the fault type with the largest probability value as the fault type for bearing fault diagnosis.
- a supervised dimension reduction method is used to implement bearing fault diagnosis.
- the feature values and fault types in the high-dimensional data are extracted, so that the training data has a very good degree of discrimination in the low-dimensional space.
- the characteristic value includes vibration displacement (peak-to-peak value), vibration velocity (true effective value), vibration acceleration (peak value), high-frequency acceleration, and the failure types include wear failure, fatigue failure, and corrosion failure.
- the step S200 includes the following steps:
- Step S210 Use the LLE algorithm to perform dimensionality reduction on the training data to obtain dimensionality reduction training data, and determine the preferred number of neighbors and the preferred fault dimension of the dimensionality reduction training data;
- Step S220 Use the dimensionality reduction training data corresponding to the preferred number of neighbors and the preferred fault dimension as the preferred dimensionality reduction training data.
- the step S210 includes:
- Step S211 Set the value range of the number of neighbors p and the value range of the fault dimension q.
- the value range of the neighbor number p and the value range of the fault dimension q can be set according to historical records or artificially according to the diagnosis requirements of bearing faults.
- the number of nearest neighbors p is the first important parameter in the LLE algorithm.
- the premise of the LLE algorithm is that each training data point is locally linear, that is, each training data point can be expressed by a linear combination of its neighbors, and the training data is maintained during the process of high-dimensional to low-dimensional mapping The relationship between neighbors. If the value of p is too large, the local linear range is too large, which cannot well reflect the local characteristics of the LLE algorithm. When the value of p is too small, it is difficult for the LLE algorithm to guarantee the topological structure of the training data in the low-dimensional space.
- the fault dimension q is the second important parameter in the LLE algorithm. If the value of the fault dimension q is too large, the training data after dimensionality reduction will contain too much redundancy. On the contrary, if the value of the fault dimension q is too large Small, so that the training data separated from each other in the high-dimensional space overlap in the low-dimensional space.
- Step S212 Select a p value and a q value as a parameter group, and form all the parameter groups into a parameter set, and the parameter set includes all combinations of the p value and the q value.
- Step S213 Select a parameter group in sequence as the number of neighbors p and the fault dimension q of the training data.
- Step S214 Use the LLE algorithm to reduce the dimensionality of the training sample data obtained in step S213 to obtain a reduced-dimensional data set and a fault set.
- the data set Y ⁇ y 1 , y 2 ,..., y N ⁇ , Y is an N ⁇ m matrix, N is the number of samples, and m is the fault dimension; the fault set s is the total number of fault categories;
- the training sample data after dimensionality reduction does not need to train the classifier. Instead, the fault type of the test data is directly determined through the sample distribution of the data of different fault types.
- Step S215 Calculate the evaluation index F using the reduced-dimensional data set and the fault set.
- the intra-class dispersion matrix S i of all categories is calculated by the following formula:
- the inter-class dispersion matrix is calculated by the following formula:
- the evaluation index F is calculated by the following formula:
- Step S216 Judge whether all parameter groups in the parameter set have calculated evaluation indexes, if not, jump to step S213, if yes, execute the following steps.
- Step S217 Compare the size of the evaluation index in each parameter group, select the parameter group with the largest evaluation index as the preferred parameter group, the p value of the parameter group as the preferred number of neighbors, and the q value of the parameter group as the preferred fault dimension.
- the optimal parameter group is selected to maximize the inter-class spacing of data of different fault types and minimize the intra-class spacing of data of different fault types.
- the LLE algorithm is a typical unsupervised learning method.
- this embodiment by traversing the number of neighbors p and the fault dimension q, the inter-class spacing of data of different fault categories is maximized, and the intra-class spacing is minimized.
- the feature selection after dimensionality reduction is guided, and the supervised LLE algorithm is realized.
- this embodiment needs to store fewer types of parameters, has a faster prediction rate, and is suitable for online prediction.
- the step S400 includes:
- the preferred number of neighbors is used as the number of neighbors of the test data
- the preferred fault dimension is used as the fault dimension of the test data
- the LLE algorithm is used to reduce the dimensionality of the test data to obtain the reduced dimensionality test data.
- this embodiment also provides a bearing fault diagnosis device based on a supervised LLE algorithm.
- the device includes a memory, a processor, and a computer program stored in the memory and running on the processor. , The processor executes the computer program and runs in the module of the following device:
- the extraction module 100 is configured to obtain training data, where the training data is historical data characterizing bearing vibration signals, and extract the characteristic value of the training data and the fault type corresponding to the characteristic value;
- the determining module 200 is configured to determine the preferred dimensionality reduction training data of the training data.
- the ratio of the inter-class dispersion to the intra-class dispersion of all fault types is the largest;
- the calculation module 300 is used for the mean value and covariance matrix corresponding to each fault type in the preferred dimensionality reduction training data
- the dimensionality reduction module 400 is used to reduce the dimensionality of the test data received in real time to obtain the dimensionality reduction test data;
- the diagnosis module 500 is configured to calculate the probability value of the dimensionality reduction data under each fault type according to the mean value and the covariance matrix, and use the fault type with the largest probability value as the fault type for bearing fault diagnosis.
- the characteristic values include vibration displacement, vibration velocity, vibration acceleration, and high-frequency acceleration
- the failure types include wear failure, fatigue failure, and corrosion failure.
- the determining module 200 is specifically configured to:
- the dimensionality reduction training data corresponding to the preferred number of neighbors and the preferred fault dimension is used as the preferred dimensionality reduction training data.
- the bearing fault diagnosis device based on the supervised LLE algorithm can be run in computing devices such as desktop computers, mobile phones, notebooks, tablets, and cloud servers.
- the operable system may include, but is not limited to, a processor and a memory.
- a bearing fault diagnosis device based on the supervised LLE algorithm may include a scale More or fewer components, or a combination of some components, or different components.
- the bearing fault diagnosis device based on the supervised LLE algorithm may also include input and output devices, network access devices, buses, and the like.
- the so-called processor can be a central processing unit (Central-Processing-Unit, CPU), other general-purpose processors, digital signal processors (Digital-Signal-Processor, DSP), application-specific integrated circuits (Application-Specific-Integrated -Circuit, ASIC), ready-made programmable gate array (Field-Programmable-Gate-Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
- the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.
- the processor is the control center of the operating system of the bearing fault diagnosis device based on the supervised LLE algorithm. Interfaces and lines connect the entire operating system of a bearing fault diagnosis device based on supervised LLE algorithm.
- the memory may be used to store the computer program and/or module, and the processor implements the one by running or executing the computer program and/or module stored in the memory and calling data stored in the memory.
- the memory may mainly include a storage program area and a storage data area.
- the storage program area may store an operating system, an application program required by at least one function (such as a sound playback function, an image playback function, etc.), etc.; the storage data area may store Data created based on the use of mobile phones (such as audio data, phone book, etc.), etc.
- the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disks, memory, plug-in hard disks, Smart-Media-Card (SMC), and Secure-Digital (Secure-Digital, SD) card, flash memory card (Flash-Card), at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.
- non-volatile memory such as hard disks, memory, plug-in hard disks, Smart-Media-Card (SMC), and Secure-Digital (Secure-Digital, SD) card, flash memory card (Flash-Card), at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computing Systems (AREA)
- Acoustics & Sound (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
一种基于有监督LLE算法的轴承故障诊断方法及装置,其中的方法包括:首先,获取训练数据,训练数据为表征轴承振动信号的历史数据,提取训练数据的特征值和特征值对应的故障类型(S100);接着,确定训练数据的优选降维训练数据,进而计算优选降维训练数据中各个故障类型对应的均值和协方差矩阵(S300);通过对实时接收的测试数据进行降维,得到降维测试数据(S400);根据均值和协方差矩阵计算降维数据在各个故障类型下的概率值,将概率值最大的故障类型作为轴承故障诊断的故障类型(S500)。由此,提高了轴承故障诊断的在线预测速率。
Description
本发明涉及故障诊断技术领域,具体涉及一种基于有监督LLE算法的轴承故障诊断方法及装置。
作为新兴的综合性的边缘学科,轴承故障诊断技术已初步形成了比较完整的学科体系。就其技术手段而言,振动诊断技术已经成为轴承故障诊断的主流技术。而计算机技术与信号信息处理技术的飞速进步,极大地推动了轴承故障诊断和监测技术向着科学化和实用化的方向发展。
然而,在目前的轴承故障诊断领域,往往存在大规模数据并发的情况,对故障诊断的实时性要求带来极大的挑战,迫切需要提高轴承故障诊断的在线预测速率。
发明内容
本发明的目的在于提供一种基于有监督LLE算法的轴承故障诊断方法及装置,旨在提高轴承故障诊断的在线预测速率。
为了实现上述目的,本发明提供以下技术方案:
一种基于有监督LLE算法的轴承故障诊断方法,包括:
获取训练数据,所述训练数据为表征轴承振动信号的历史数据,提取所述训练数据的特征值和所述特征值对应的故障类型;
确定所述训练数据的优选降维训练数据,所述优选降维训练数据中,所有故障类型的类间离散度与类内离散度的比值最大;
计算所述优选降维训练数据中各个故障类型对应的均值和协方差矩阵;
对实时接收的测试数据进行降维,得到降维测试数据;
根据所述均值和协方差矩阵计算所述降维数据在各个故障类型下的概率 值,将概率值最大的故障类型作为轴承故障诊断的故障类型。
进一步,所述特征值包括振动位移、振动速度、振动加速度、高频加速度,所述故障类型包括磨损失效、疲劳失效、腐蚀失效。
进一步,所述确定所述训练数据的优选降维训练数据,包括:
利用LLE算法对所述训练数据进行降维,得到降维训练数据,并确定所述降维训练数据的优选近邻数和优选故障维度;
将所述优选近邻数和优选故障维度对应的降维训练数据作为优选降维训练数据。
进一步,所述利用LLE算法对所述训练数据进行降维,得到降维训练数据,并确定所述降维训练数据的优选近邻数和优选故障维度,包括:
步骤310、设置近邻数p的取值范围和故障维度q的取值范围;
步骤320、选择一个p值和一个q值作为一个参数组,将所有参数组形成参数集合,所述参数集合包括p值和q值的所有组合形式;
步骤330、依次选择一个参数组,作为所述训练数据的近邻数p和故障维度q;
步骤340、利用LLE算法对步骤330得到的训练样本数据进行降维,得到降维后的数据集Y和故障集φ,其中,数据集Y={y
1,y
2,...,y
N},Y为一个N×m的矩阵,N为样本个数,m为故障维度;故障集
s为故障类别总数;
步骤350、利用所述降维后的数据集和故障集计算评价指标F,具体为:
通过以下公式计算每个故障类别的均值向量c
i:
通过以下公式计算所有类别的类内离散度矩阵S
i:
对所有类内离散度矩阵的求和,得到混合类内离散度矩阵Sw:
Sw=S
1+S
2+...+Ss;
通过以下公式计算类间离散度矩阵:
通过以下公式计算评价指标F:
F=S
b/S
w;
步骤360、判断所述参数集合中的所有参数组是否均计算出评价指标,若否,跳转到步骤330,若是,执行以下步骤;
步骤370、比较各个参数组中评价指标的大小,选择评价指标最大的参数组作为优选参数组,将该参数组的p值作为优选近邻数,将该参数组的q值作为优选故障维度。
进一步,所述对实时接收的测试数据进行降维,得到降维测试数据,包括:
将所述优选近邻数作为所述测试数据的近邻数,将所述优选故障维度作为所述测试数据的故障维度,利用LLE算法对所述测试数据进行降维,得到降维测试数据。
一种基于有监督LLE算法的轴承故障诊断装置,所述装置包括:存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序运行在以下装置的模块中:
提取模块,用于获取训练数据,所述训练数据为表征轴承振动信号的历史数据,提取所述训练数据的特征值和所述特征值对应的故障类型;
确定模块,用于确定所述训练数据的优选降维训练数据,所述优选降维训练数据中,所有故障类型的类间离散度与类内离散度的比值最大;
计算模块,用于所述优选降维训练数据中各个故障类型对应的均值和协方差矩阵;
降维模块,用于对实时接收的测试数据进行降维,得到降维测试数据;
诊断模块,用于根据所述均值和协方差矩阵计算所述降维数据在各个故障类型下的概率值,将概率值最大的故障类型作为轴承故障诊断的故障类型。
进一步,所述特征值包括振动位移、振动速度、振动加速度、高频加速度,所述故障类型包括磨损失效、疲劳失效、腐蚀失效。
进一步,所述确定模块具体用于:
利用LLE算法对所述训练数据进行降维,得到降维训练数据,并确定所述降维训练数据的优选近邻数和优选故障维度;
将所述优选近邻数和优选故障维度对应的降维训练数据作为优选降维训练数据。
本发明的有益效果是:本发明公开一种基于有监督LLE算法的轴承故障诊断方法及装置,首先获取训练数据,所述训练数据为表征轴承振动信号的历史数据,提取所述训练数据的特征值和所述特征值对应的故障类型,接着确定所述训练数据的优选降维训练数据,进而计算所述优选降维训练数据中各个故障类型对应的均值和协方差矩阵,通过对实时接收的测试数据进行降维,得到降维测试数据,根据所述均值和协方差矩阵计算所述降维数据在各个故障类型下的概率值,将概率值最大的故障类型作为轴承故障诊断的故障类型。本发明提高了轴承故障诊断的在线预测速率。
为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实施例中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1是本发明实施例一种基于有监督LLE算法的轴承故障诊断方法的流程示意图;
图2是本发明实施例步骤S200的流程示意图;
图3是本发明实施例步骤S210的流程示意图;
图4是本发明实施例一种基于有监督LLE算法的轴承故障诊断装置的结构示意图。
下面将结合附图对本发明的技术方案进行清楚、完整的描述,显然,所描述的实施例是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所以其他实施例,都属于本发明的保护范围。
参考图1,本发明实施例提供的一种基于有监督LLE算法的轴承故障诊断方法,包括以下步骤:
步骤S100、获取训练数据,所述训练数据为表征轴承振动信号的历史数据,提取所述训练数据的特征值和所述特征值对应的故障类型;
步骤S200、确定所述训练数据的优选降维训练数据,所述优选降维训练数据中,所有故障类型的类间离散度与类内离散度的比值最大;
步骤S300、计算所述优选降维训练数据中各个故障类型对应的均值和协方差矩阵;
步骤S400、对实时接收的测试数据进行降维,得到降维测试数据;
步骤S500、根据所述均值和协方差矩阵计算所述降维数据在各个故障类型下的概率值,将概率值最大的故障类型作为轴承故障诊断的故障类型。
本实施例利用有监督的降维方法实现轴承故障诊断。通过对训练数据进行训练,将高维数据中的特征值和故障类型提取出来,使得训练数据在低维空间中有非常好的区分度,本实施例需要存储的参数类型更少,预测速率更快,适合在线预测。
在一个实施例中,所述特征值包括振动位移(峰峰值)、振动速度(真有效值)、振动加速度(峰值)、高频加速度,所述故障类型包括磨损失效、疲劳失效、腐蚀失效。
参考图2,作为本实施例的进一步改进,所述步骤S200包括以下步骤:
步骤S210、利用LLE算法对所述训练数据进行降维,得到降维训练数据,并确定所述降维训练数据的优选近邻数和优选故障维度;
步骤S220、将所述优选近邻数和优选故障维度对应的降维训练数据作为优选降维训练数据。
参考图3,作为本实施例的进一步改进,所述步骤S210包括:
步骤S211、设置近邻数p的取值范围和故障维度q的取值范围。
本实施例中,需要找出降维训练数据中每个训练数据的p个近邻点。把每个训练数据点中欧式距离最近的p个训练数据点找出,p即所谓近邻数。
近邻数p的取值范围和故障维度q的取值范围可以根据历史记录,或者根据轴承故障的诊断需求人为设置,p和q的取值范围越大,则训练时间越长、诊断更全面,p和q的取值范围越小,则训练时间越短。
近邻数p是LLE算法中的第一个重要参数。LLE算法的前提假设是每一个训练数据点都是局部线性的,即每一个训练数据点都可以用它的近邻点线性组合来表达,在高维向低维映射的过程中,保持了训练数据之间的近邻关系。p的取值过大使得局部线性的范围过大,无法很好的体现LLE算法的局部特征。而当p的取值过小,LLE算法就很难保证训练数据在低维空间中的拓扑结构。
故障维度q是LLE算法中的第二个重要的参数,故障维度q的取值过大,将会使降维后的训练数据中含有过多的冗余,反之如果故障维度q的取值过小,使得在高维空间中彼此分开的训练数据在低维空间中交叠。
步骤S212、选择一个p值和一个q值作为一个参数组,将所有参数组形成参数集合,所述参数集合包括p值和q值的所有组合形式。
步骤S213、依次选择一个参数组,作为所述训练数据的近邻数p和故障维度q。
步骤S214、利用LLE算法对步骤S213得到的训练样本数据进行降维,得到降维后的数据集和故障集。
降维后的训练样本数据不用再训练分类器,而是通过不同故障类型数据的 样本分布来直接确定测试数据的故障类型。
步骤S215、利用所述降维后的数据集和故障集计算评价指标F。
具体为:
通过以下公式计算每个故障类别的均值向量c
i:
通过以下公式计算所有类别的类内离散度矩阵S
i:
对所有类内离散度矩阵的求和,得到混合类内离散度矩阵Sw:
Sw=S
1+S
2+...+Ss;
通过以下公式计算类间离散度矩阵:
通过以下公式计算评价指标F:
F=S
b/S
w;
步骤S216、判断所述参数集合中的所有参数组是否均计算出评价指标,若否,跳转到步骤S213,若是,执行以下步骤。
步骤S217、比较各个参数组中评价指标的大小,选择评价指标最大的参数组作为优选参数组,将该参数组的p值作为优选近邻数,将该参数组的q值作为优选故障维度。
选择优选参数组是为了最大化不同故障类型数据的类间间距、最小化不同故障类型数据间的类内间距。
LLE算法是一种典型的无监督学习方法,本实施例中,通过遍历近邻数p和故障维度q,使得最大化不同故障类别数据的类间间距,最小化类内间距。这样通过已知的特征值和故障类型,引导降维后的特征选取,实现了有监督的LLE算法。与传统的轴承故障诊断方法相比,本实施例需要存储的参数类型更 少,预测速率更快,适合在线预测。
作为本实施例的进一步改进,所述步骤S400包括:
将所述优选近邻数作为所述测试数据的近邻数,将所述优选故障维度作为所述测试数据的故障维度,利用LLE算法对所述测试数据进行降维,得到降维测试数据。
参考图4,本实施例还提供一种基于有监督LLE算法的轴承故障诊断装置,所述装置包括:存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序运行在以下装置的模块中:
提取模块100,用于获取训练数据,所述训练数据为表征轴承振动信号的历史数据,提取所述训练数据的特征值和所述特征值对应的故障类型;
确定模块200,用于确定所述训练数据的优选降维训练数据,所述优选降维训练数据中,所有故障类型的类间离散度与类内离散度的比值最大;
计算模块300,用于所述优选降维训练数据中各个故障类型对应的均值和协方差矩阵;
降维模块400,用于对实时接收的测试数据进行降维,得到降维测试数据;
诊断模块500,用于根据所述均值和协方差矩阵计算所述降维数据在各个故障类型下的概率值,将概率值最大的故障类型作为轴承故障诊断的故障类型。
作为本实施例的进一步改进,所述特征值包括振动位移、振动速度、振动加速度、高频加速度,所述故障类型包括磨损失效、疲劳失效、腐蚀失效。
作为本实施例的进一步改进,所述确定模块200具体用于:
利用LLE算法对所述训练数据进行降维,得到降维训练数据,并确定所述降维训练数据的优选近邻数和优选故障维度;
将所述优选近邻数和优选故障维度对应的降维训练数据作为优选降维训练数据。
所述一种基于有监督LLE算法的轴承故障诊断装置可以运行于桌上型计算机、手机、笔记本、平板电脑及云端服务器等计算设备中。所述一种基于有监 督LLE算法的轴承故障诊断装置,可运行的系统可包括,但不仅限于,处理器、存储器。本领域技术人员可以理解,所述例子仅仅是一种基于有监督LLE算法的轴承故障诊断装置的示例,并不构成对一种基于有监督LLE算法的轴承故障诊断装置的限定,可以包括比例子更多或更少的部件,或者组合某些部件,或者不同的部件,例如所述一种基于有监督LLE算法的轴承故障诊断装置还可以包括输入输出设备、网络接入设备、总线等。
所称处理器可以是中央处理单元(Central-Processing-Unit,CPU),还可以是其他通用处理器、数字信号处理器(Digital-Signal-Processor,DSP)、专用集成电路(Application-Specific-Integrated-Circuit,ASIC)、现成可编程门阵列(Field-Programmable-Gate-Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等,所述处理器是所述一种基于有监督LLE算法的轴承故障诊断装置运行系统的控制中心,利用各种接口和线路连接整个一种基于有监督LLE算法的轴承故障诊断装置可运行系统的各个部分。
所述存储器可用于存储所述计算机程序和/或模块,所述处理器通过运行或执行存储在所述存储器内的计算机程序和/或模块,以及调用存储在存储器内的数据,实现所述一种基于有监督LLE算法的轴承故障诊断装置的各种功能。所述存储器可主要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、至少一个功能所需的应用程序(比如声音播放功能、图像播放功能等)等;存储数据区可存储根据手机的使用所创建的数据(比如音频数据、电话本等)等。此外,存储器可以包括高速随机存取存储器,还可以包括非易失性存储器,例如硬盘、内存、插接式硬盘,智能存储卡(Smart-Media-Card,SMC),安全数字(Secure-Digital,SD)卡,闪存卡(Flash-Card)、至少一个磁盘存储器件、闪存器件、或其他易失性固态存储器件。
尽管本公开的描述已经相当详尽且特别对几个所述实施例进行了描述,但其并非旨在局限于任何这些细节或实施例或任何特殊实施例,而是应当将其视 作是通过参考所附权利要求,考虑到现有技术为这些权利要求提供广义的可能性解释,从而有效地涵盖本公开的预定范围。此外,上文以发明人可预见的实施例对本公开进行描述,其目的是为了提供有用的描述,而那些目前尚未预见的对本公开的非实质性改动仍可代表本公开的等效改动。
Claims (8)
- 一种基于有监督LLE算法的轴承故障诊断方法,其特征在于,包括:获取训练数据,所述训练数据为表征轴承振动信号的历史数据,提取所述训练数据的特征值和所述特征值对应的故障类型;确定所述训练数据的优选降维训练数据,所述优选降维训练数据中,所有故障类型的类间离散度与类内离散度的比值最大;计算所述优选降维训练数据中各个故障类型对应的均值和协方差矩阵;对实时接收的测试数据进行降维,得到降维测试数据;根据所述均值和协方差矩阵计算所述降维数据在各个故障类型下的概率值,将概率值最大的故障类型作为轴承故障诊断的故障类型。
- 根据权利要求1所述的一种基于有监督LLE算法的轴承故障诊断方法,其特征在于,所述特征值包括振动位移、振动速度、振动加速度、高频加速度,所述故障类型包括磨损失效、疲劳失效、腐蚀失效。
- 根据权利要求2所述的一种基于有监督LLE算法的轴承故障诊断方法,其特征在于,所述确定所述训练数据的优选降维训练数据,包括:利用LLE算法对所述训练数据进行降维,得到降维训练数据,并确定所述降维训练数据的优选近邻数和优选故障维度;将所述优选近邻数和优选故障维度对应的降维训练数据作为优选降维训练数据。
- 根据权利要求3所述的一种基于有监督LLE算法的轴承故障诊断方法,其特征在于,所述利用LLE算法对所述训练数据进行降维,得到降维训练数据,并确定所述降维训练数据的优选近邻数和优选故障维度,包括:步骤310、设置近邻数p的取值范围和故障维度q的取值范围;步骤320、选择一个p值和一个q值作为一个参数组,将所有参数组形成参数集合,所述参数集合包括p值和q值的所有组合形式;步骤330、依次选择一个参数组,作为所述训练数据的近邻数p和故障维度q;步骤340、利用LLE算法对步骤330得到的训练样本数据进行降维,得到 降维后的数据集Y和故障集φ,其中,数据集Y={y 1,y 2,...,y N},Y为一个N×m的矩阵,N为样本个数,m为故障维度;故障集 s为故障类别总数;步骤350、利用所述降维后的数据集和故障集计算评价指标F,具体为:通过以下公式计算每个故障类别的均值向量c i:通过以下公式计算所有类别的类内离散度矩阵S i:对所有类内离散度矩阵的求和,得到混合类内离散度矩阵Sw:Sw=S 1+S 2+...+Ss;通过以下公式计算类间离散度矩阵:通过以下公式计算评价指标F:F=S b/S w;步骤360、判断所述参数集合中的所有参数组是否均计算出评价指标,若否,跳转到步骤330,若是,执行以下步骤;步骤370、比较各个参数组中评价指标的大小,选择评价指标最大的参数组作为优选参数组,将该参数组的p值作为优选近邻数,将该参数组的q值作为优选故障维度。
- 根据权利要求4所述的一种基于有监督LLE算法的轴承故障诊断方法,其特征在于,所述对实时接收的测试数据进行降维,得到降维测试数据,包括:实时接收测试数据,将所述优选近邻数作为所述测试数据的近邻数,将所述优选故障维度作为所述测试数据的故障维度,利用LLE算法对所述测试数据进行降维,得到降维测试数据。
- 一种基于有监督LLE算法的轴承故障诊断装置,其特征在于,所述装置包括: 存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序运行在以下装置的模块中:提取模块,用于获取训练数据,所述训练数据为表征轴承振动信号的历史数据,提取所述训练数据的特征值和所述特征值对应的故障类型;确定模块,用于确定所述训练数据的优选降维训练数据,所述优选降维训练数据中,所有故障类型的类间离散度与类内离散度的比值最大;计算模块,用于所述优选降维训练数据中各个故障类型对应的均值和协方差矩阵;降维模块,用于对实时接收的测试数据进行降维,得到降维测试数据;诊断模块,用于根据所述均值和协方差矩阵计算所述降维数据在各个故障类型下的概率值,将概率值最大的故障类型作为轴承故障诊断的故障类型。
- 根据权利要求6所述的一种基于有监督LLE算法的轴承故障诊断装置,其特征在于,所述特征值包括振动位移、振动速度、振动加速度、高频加速度,所述故障类型包括磨损失效、疲劳失效、腐蚀失效。
- 根据权利要求7所述的一种基于有监督LLE算法的轴承故障诊断装置,其特征在于,所述确定模块具体用于:利用LLE算法对所述训练数据进行降维,得到降维训练数据,并确定所述降维训练数据的优选近邻数和优选故障维度;将所述优选近邻数和优选故障维度对应的降维训练数据作为优选降维训练数据。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910824743.3A CN110672324B (zh) | 2019-09-02 | 2019-09-02 | 一种基于有监督lle算法的轴承故障诊断方法及装置 |
CN201910824743.3 | 2019-09-02 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021042749A1 true WO2021042749A1 (zh) | 2021-03-11 |
Family
ID=69075909
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/087799 WO2021042749A1 (zh) | 2019-09-02 | 2020-04-29 | 一种基于有监督lle算法的轴承故障诊断方法及装置 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN110672324B (zh) |
WO (1) | WO2021042749A1 (zh) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110672324B (zh) * | 2019-09-02 | 2021-03-26 | 佛山科学技术学院 | 一种基于有监督lle算法的轴承故障诊断方法及装置 |
CN111896254A (zh) * | 2020-08-10 | 2020-11-06 | 山东大学 | 一种变速变载大型滚动轴承故障预测系统及方法 |
CN114239867A (zh) * | 2021-11-08 | 2022-03-25 | 深圳供电局有限公司 | 故障类型确定方法、装置、计算机设备和可读存储介质 |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001356808A (ja) * | 2000-06-16 | 2001-12-26 | Ntn Corp | 機械部品の監視・診断・販売システム |
CN102930285A (zh) * | 2012-09-18 | 2013-02-13 | 四川大学 | 基于有监督增量式局部线性嵌入(sille)维数化简的早期故障辨识方法 |
CN105043766A (zh) * | 2015-06-24 | 2015-11-11 | 北京航空航天大学 | 一种基于Hessian局部线性嵌入的轴承变工况故障诊断方法 |
CN105973594A (zh) * | 2016-04-25 | 2016-09-28 | 西北工业大学 | 一种基于连续深度置信网络的滚动轴承故障预测方法 |
CN107024352A (zh) * | 2017-05-03 | 2017-08-08 | 哈尔滨理工大学 | 一种基于滑动熵‑ica算法的滚动轴承故障特征提取方法 |
CN107036817A (zh) * | 2017-04-05 | 2017-08-11 | 哈尔滨理工大学 | 基于磷虾群算法的svr滚动轴承性能衰退预测方法 |
CN110163261A (zh) * | 2019-04-28 | 2019-08-23 | 平安科技(深圳)有限公司 | 不平衡数据分类模型训练方法、装置、设备及存储介质 |
CN110672324A (zh) * | 2019-09-02 | 2020-01-10 | 佛山科学技术学院 | 一种基于有监督lle算法的轴承故障诊断方法及装置 |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH04254013A (ja) * | 1990-12-27 | 1992-09-09 | Yaskawa Electric Corp | 軸受破損防止方法 |
CN105760839A (zh) * | 2016-02-22 | 2016-07-13 | 重庆大学 | 基于多特征流形学习与支持向量机的轴承故障诊断方法 |
CN109582003B (zh) * | 2018-12-03 | 2021-04-20 | 东北林业大学 | 基于伪标签半监督核局部费舍尔判别分析轴承故障诊断方法 |
CN109827777B (zh) * | 2019-04-01 | 2020-12-18 | 哈尔滨理工大学 | 基于偏最小二乘法极限学习机的滚动轴承故障预测方法 |
CN110031226A (zh) * | 2019-04-12 | 2019-07-19 | 佛山科学技术学院 | 一种轴承故障的诊断方法及装置 |
-
2019
- 2019-09-02 CN CN201910824743.3A patent/CN110672324B/zh active Active
-
2020
- 2020-04-29 WO PCT/CN2020/087799 patent/WO2021042749A1/zh active Application Filing
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001356808A (ja) * | 2000-06-16 | 2001-12-26 | Ntn Corp | 機械部品の監視・診断・販売システム |
CN102930285A (zh) * | 2012-09-18 | 2013-02-13 | 四川大学 | 基于有监督增量式局部线性嵌入(sille)维数化简的早期故障辨识方法 |
CN105043766A (zh) * | 2015-06-24 | 2015-11-11 | 北京航空航天大学 | 一种基于Hessian局部线性嵌入的轴承变工况故障诊断方法 |
CN105973594A (zh) * | 2016-04-25 | 2016-09-28 | 西北工业大学 | 一种基于连续深度置信网络的滚动轴承故障预测方法 |
CN107036817A (zh) * | 2017-04-05 | 2017-08-11 | 哈尔滨理工大学 | 基于磷虾群算法的svr滚动轴承性能衰退预测方法 |
CN107024352A (zh) * | 2017-05-03 | 2017-08-08 | 哈尔滨理工大学 | 一种基于滑动熵‑ica算法的滚动轴承故障特征提取方法 |
CN110163261A (zh) * | 2019-04-28 | 2019-08-23 | 平安科技(深圳)有限公司 | 不平衡数据分类模型训练方法、装置、设备及存储介质 |
CN110672324A (zh) * | 2019-09-02 | 2020-01-10 | 佛山科学技术学院 | 一种基于有监督lle算法的轴承故障诊断方法及装置 |
Also Published As
Publication number | Publication date |
---|---|
CN110672324A (zh) | 2020-01-10 |
CN110672324B (zh) | 2021-03-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021042749A1 (zh) | 一种基于有监督lle算法的轴承故障诊断方法及装置 | |
WO2022126971A1 (zh) | 基于密度的文本聚类方法、装置、设备及存储介质 | |
CN110111113B (zh) | 一种异常交易节点的检测方法及装置 | |
CN111507470A (zh) | 一种异常账户的识别方法及装置 | |
WO2020250730A1 (ja) | 不正検知装置、不正検知方法および不正検知プログラム | |
CN110672323A (zh) | 一种基于神经网络的轴承健康状态评估方法及装置 | |
CN111798047A (zh) | 风控预测方法、装置、电子设备及存储介质 | |
CN111444807A (zh) | 目标检测方法、装置、电子设备和计算机可读介质 | |
Yang et al. | A real-time image forensics scheme based on multi-domain learning | |
WO2021231826A1 (en) | System and method for time series pattern recognition | |
Yang et al. | Fast and robust key frame extraction method for gesture video based on high-level feature representation | |
Liang et al. | LDAMSS: Fast and efficient undersampling method for imbalanced learning | |
CN109902731B (zh) | 一种基于支持向量机的性能故障的检测方法及装置 | |
US11663295B2 (en) | Methods and systems for generating synthetic microstructure images of material with desired features | |
WO2021244105A1 (zh) | 一种特征向量维度压缩方法、装置、设备、介质 | |
Yan et al. | Improving multi-scale detection layers in the deep learning network for wheat spike detection based on interpretive analysis | |
CN111127407B (zh) | 一种基于傅里叶变换的风格迁移伪造图像检测装置及方法 | |
Gharehchopogh et al. | Automatic data clustering using farmland fertility metaheuristic algorithm | |
CN114722941A (zh) | 信贷违约识别方法、装置、设备和介质 | |
CN111428510B (zh) | 一种基于口碑的p2p平台风险分析方法 | |
CN114611592A (zh) | 一种半监督特征选择方法、系统、介质、设备及终端 | |
CN112651513A (zh) | 一种基于零样本学习的信息抽取方法及系统 | |
Xu et al. | X2-Softmax: Margin adaptive loss function for face recognition | |
CN110827919A (zh) | 一种应用于基因表达谱数据的降维方法 | |
CN111310842A (zh) | 一种密度自适应的快速聚类方法 |
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: 20861280 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: 20861280 Country of ref document: EP Kind code of ref document: A1 |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 26.10.2022) |