CN113191245B - Migration intelligent diagnosis method for multi-source rolling bearing health state fusion - Google Patents

Migration intelligent diagnosis method for multi-source rolling bearing health state fusion Download PDF

Info

Publication number
CN113191245B
CN113191245B CN202110449135.6A CN202110449135A CN113191245B CN 113191245 B CN113191245 B CN 113191245B CN 202110449135 A CN202110449135 A CN 202110449135A CN 113191245 B CN113191245 B CN 113191245B
Authority
CN
China
Prior art keywords
rolling bearing
vibration signal
signal sample
sample set
bearing vibration
Prior art date
Legal status (The legal status 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 status listed.)
Active
Application number
CN202110449135.6A
Other languages
Chinese (zh)
Other versions
CN113191245A (en
Inventor
雷亚国
赵军
杨彬
李乃鹏
王文彬
何平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
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 Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN202110449135.6A priority Critical patent/CN113191245B/en
Publication of CN113191245A publication Critical patent/CN113191245A/en
Application granted granted Critical
Publication of CN113191245B publication Critical patent/CN113191245B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • Y—GENERAL 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
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

一种多源滚动轴承健康状态融合的迁移智能诊断方法，首先，同时训练多个由单一源滚动轴承‑目标滚动轴承振动信号样本集对构建的局部分布适配子模型，获得目标滚动轴承基于每个源滚动轴承的健康状态诊断结果，其中对于每一个局部分布适配子模型，包括用于提取深度迁移特征的域共享深度残差网络和用于提取领域混淆特征的参数共享的领域混淆网络；最后，通过对多滚动轴承融合迁移诊断模型的训练，融合基于不同源滚动轴承对目标滚动轴承的诊断结果获取目标滚动轴承振动信号样本集健康状态的最终诊断结果；本发明克服源滚动轴承的诊断知识无法涵盖目标滚动轴承的故障类型和目标滚动轴承振动信号样本不平衡的影响，显著提高了迁移诊断模型的诊断精度。

Figure 202110449135

A migration intelligent diagnosis method for multi-source rolling bearing health state fusion. First, multiple local distribution adaptation submodels constructed from a single source rolling bearing-target rolling bearing vibration signal sample set are trained simultaneously to obtain the target rolling bearing based on each source rolling bearing. Health status diagnosis results, where for each local distribution fit sub-model, including a domain-shared deep residual network for extracting deep transfer features and a parameter-shared domain confusion network for extracting domain confusion features; finally, by pairing multiple The training of the rolling bearing fusion migration diagnosis model, the final diagnosis result of the health state of the target rolling bearing vibration signal sample set is obtained by fusing the diagnostic results of the target rolling bearing based on different source rolling bearings; the present invention overcomes that the diagnostic knowledge of the source rolling bearing cannot cover the fault type and target The influence of the unbalanced sample of the vibration signal of the rolling bearing significantly improves the diagnostic accuracy of the migration diagnostic model.

Figure 202110449135

Description

一种多源滚动轴承健康状态融合的迁移智能诊断方法A migration intelligent diagnosis method based on multi-source rolling bearing health status fusion

技术领域Technical Field

本发明属于滚动轴承故障诊断技术领域，具体涉及一种多源滚动轴承健康状态融合的迁移智能诊断方法。The present invention belongs to the technical field of rolling bearing fault diagnosis, and in particular relates to a migration intelligent diagnosis method for multi-source rolling bearing health status fusion.

背景技术Background Art

滚动轴承是各类旋转机械设备中使用频率最高的关键部件之一，在机械工业生产中起着重要的作用，它作为机械的“关节”，承担着维持旋转功能正常的重要工作，因此，滚动轴承的健康状况直接关系着机械设备的性能。机械设备工作环境复杂多变，导致滚动轴承的健康状况随着运行时间的延长，容易发生故障，为了提高机械设备的安全性和可靠性，研究滚动轴承的故障诊断技术十分必要。Rolling bearings are one of the most frequently used key components in various types of rotating mechanical equipment. They play an important role in the production of mechanical industry. As the "joint" of the machine, they undertake the important work of maintaining the normal rotation function. Therefore, the health of rolling bearings is directly related to the performance of mechanical equipment. The working environment of mechanical equipment is complex and changeable, which leads to the health of rolling bearings being prone to failure as the running time increases. In order to improve the safety and reliability of mechanical equipment, it is necessary to study the fault diagnosis technology of rolling bearings.

近年来，滚动轴承的智能故障诊断无需过分依赖专家诊断知识就能提供直观的诊断结果而备受关注，特别是随着机器学习的快速发展，深度学习被引入到滚动轴承智能故障诊断中，并取得了一些出色的成果。但传统的滚动轴承智能诊断方法的基本假设是要求训练数据与测试数据服从相同的概率分布，在很多实际问题中，这个假设常不能被满足，使得这些方法的性能显著下降。迁移学习方法通过实现不同领域之间的知识迁移可以缓解这个问题，因此基于迁移学习的滚动轴承故障诊断方法近年来受到广泛关注。In recent years, intelligent fault diagnosis of rolling bearings has attracted much attention because it can provide intuitive diagnostic results without relying too much on expert diagnostic knowledge. In particular, with the rapid development of machine learning, deep learning has been introduced into intelligent fault diagnosis of rolling bearings and has achieved some outstanding results. However, the basic assumption of traditional intelligent diagnosis methods for rolling bearings is that the training data and test data must follow the same probability distribution. In many practical problems, this assumption is often not met, which significantly reduces the performance of these methods. Transfer learning methods can alleviate this problem by realizing knowledge transfer between different fields. Therefore, rolling bearing fault diagnosis methods based on transfer learning have received widespread attention in recent years.

然而，现有的滚动轴承迁移诊断技术存在显著的局限性：1)源滚动轴承的健康状态集应与目标滚动轴承的健康状态集重叠；2)目标滚动轴承样本的数量在健康状态之间保持平衡。但是，在实际应用中，此类假设很难成立：首先，目标滚动轴承不可避免地会遭受源滚动轴承从未经历过的故障类型的困扰，因此，来自单个源滚动轴承的诊断知识不足以识别从所有健康状态中提取的目标滚动轴承样本；其次，目标滚动轴承在寿命周期内长期处于正常状态，因此，收集的目标滚动轴承数据集由大量健康样本和少量故障样本组成，上述两点因素降低了传统迁移诊断技术对滚动滚动轴承故障的诊断精度。However, existing rolling bearing migration diagnosis techniques have significant limitations: 1) the health state set of the source rolling bearing should overlap with the health state set of the target rolling bearing; 2) the number of target rolling bearing samples should be balanced between the health states. However, in practical applications, such assumptions are difficult to hold true: first, the target rolling bearing will inevitably suffer from fault types that the source rolling bearing has never experienced, so the diagnostic knowledge from a single source rolling bearing is not sufficient to identify the target rolling bearing samples extracted from all health states; second, the target rolling bearing is in a normal state for a long time during its life cycle, so the collected target rolling bearing dataset consists of a large number of healthy samples and a small number of faulty samples. The above two factors reduce the diagnostic accuracy of traditional migration diagnosis techniques for rolling bearing faults.

受来自源滚动轴承的诊断知识可能无法涵盖目标滚动轴承的所有故障类型和目标滚动轴承样本不平衡的影响，现有迁移智能诊断技术的性能显著下降。Due to the fact that the diagnostic knowledge from the source rolling bearing may not cover all fault types of the target rolling bearing and the imbalance of the target rolling bearing samples, the performance of existing migration intelligent diagnosis technology is significantly reduced.

发明内容Summary of the invention

为了克服上述现有技术的缺点，本发明的目的在于提出一种多源滚动轴承健康状态融合的迁移智能诊断方法，提高源滚动轴承的诊断知识可能无法涵盖目标滚动轴承的所有故障类型和目标滚动轴承样本不平衡条件下滚动轴承迁移诊断的精度，推动智能诊断技术的实际应用。In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to propose a migration intelligent diagnosis method that integrates the health status of multiple source rolling bearings, improve the accuracy of rolling bearing migration diagnosis under the condition that the diagnostic knowledge of the source rolling bearing may not cover all fault types of the target rolling bearing and the target rolling bearing sample is unbalanced, and promote the practical application of intelligent diagnosis technology.

为了达到上述目的，本发明采取的技术方案为：In order to achieve the above object, the technical solution adopted by the present invention is:

一种多源滚动轴承健康状态融合的迁移智能诊断方法，包括以下步骤：A migration intelligent diagnosis method for multi-source rolling bearing health status fusion includes the following steps:

1)获取多个源滚动轴承振动信号样本集和目标滚动轴承振动信号样本集；1) Acquire multiple source rolling bearing vibration signal sample sets and target rolling bearing vibration signal sample sets;

2)由单一源滚动轴承-目标滚动轴承振动信号样本集对构建局部分布适配子模型；2) constructing a local distribution adapter model from a single source rolling bearing-target rolling bearing vibration signal sample set;

3)同时训练多个局部分布适配子模型；3) Training multiple local distribution adapter models simultaneously;

4)重复执行步骤3至网络参数{θsk|sk∈S}收敛；4) Repeat step 3 until the network parameters {θ sk |s k ∈S} converge;

5)融合基于不同源滚动轴承对目标滚动轴承的诊断结果获取目标滚动轴承振动信号样本集健康状态的最终诊断结果：5) The final diagnosis result of the health status of the target rolling bearing vibration signal sample set is obtained by fusing the diagnosis results of the target rolling bearing based on different source rolling bearings:

Figure GDA0004018962480000031
Figure GDA0004018962480000031

其中，

Figure GDA00040189624800000312
表示目标滚动轴承振动信号样本集中第i个样本健康状态诊断结果，
Figure GDA0004018962480000032
表示第sk个源滚动轴承振动信号样本集的状态标签，
Figure GDA0004018962480000033
in,
Figure GDA00040189624800000312
represents the health status diagnosis result of the i-th sample in the target rolling bearing vibration signal sample set,
Figure GDA0004018962480000032
represents the state label of the s kth source rolling bearing vibration signal sample set,
Figure GDA0004018962480000033

所述的步骤1)具体为：获取N个源滚动轴承振动信号样本集

Figure GDA0004018962480000034
其中，
Figure GDA0004018962480000035
表示第sk个源滚动轴承振动信号样本集，包含的样本量为
Figure GDA0004018962480000036
表示第sk个源滚动轴承振动信号样本集的状态标签，
Figure GDA0004018962480000037
其中
Figure GDA0004018962480000038
表示第sk个源滚动轴承振动信号样本集的状态类别总数；获取目标滚动轴承振动信号样本集
Figure GDA0004018962480000039
其中
Figure GDA00040189624800000310
表示目标滚动轴承振动信号样本集，包含的样本量为nt。The step 1) is specifically as follows: obtaining N source rolling bearing vibration signal sample sets
Figure GDA0004018962480000034
in,
Figure GDA0004018962480000035
represents the s kth source rolling bearing vibration signal sample set, which contains the number of samples:
Figure GDA0004018962480000036
represents the state label of the s kth source rolling bearing vibration signal sample set,
Figure GDA0004018962480000037
in
Figure GDA0004018962480000038
Represents the total number of state categories of the s kth source rolling bearing vibration signal sample set; obtains the target rolling bearing vibration signal sample set
Figure GDA0004018962480000039
in
Figure GDA00040189624800000310
represents the target rolling bearing vibration signal sample set, which contains n t samples.

所述的步骤2)具体为：将N个源滚动轴承振动信号样本集与单一目标滚动轴承振动信号样本集配对，共组成N对

Figure GDA00040189624800000311
构建N个局部分布适配子模型与这N个源滚动轴承-目标滚动轴承振动信号样本集对一一对应。The step 2) is specifically as follows: pairing N source rolling bearing vibration signal sample sets with a single target rolling bearing vibration signal sample set to form a total of N pairs.
Figure GDA00040189624800000311
N local distribution adapter sub-models are constructed to correspond one-to-one to the N source rolling bearing-target rolling bearing vibration signal sample set pairs.

所述的步骤3)具体为：执行如下步骤同时训练这N个局部分布适配子模型，以第sk个源滚动轴承-目标滚动轴承振动信号样本集对

Figure GDA0004018962480000041
说明每个局部分布适配子模型的训练过程，包括如下步骤：The step 3) is specifically as follows: perform the following steps to simultaneously train the N local distribution adapter sub-models, using the s kth source rolling bearing-target rolling bearing vibration signal sample set pair
Figure GDA0004018962480000041
The training process of each local distribution adapter model is described, including the following steps:

3.1)计算第sk个源滚动轴承-目标滚动轴承振动信号样本集对的深度迁移故障特征：3.1) Calculate the deep migration fault feature of the s kth source rolling bearing-target rolling bearing vibration signal sample set pair:

同时从第sk个源滚动轴承与目标滚动轴承振动信号样本集中提取深度迁移故障特征

Figure GDA0004018962480000042
其中，
Figure GDA0004018962480000043
为第sk个源滚动轴承振动信号样本集第i个样本的深度迁移故障特征，
Figure GDA0004018962480000044
为目标滚动轴承振动信号样本集第i个样本的深度迁移故障特征，上/下标F2代表域共享深度残差网络的F2层；At the same time, deep migration fault features are extracted from the vibration signal sample sets of the s kth source rolling bearing and the target rolling bearing.
Figure GDA0004018962480000042
in,
Figure GDA0004018962480000043
is the deep migration fault feature of the ith sample of the s k th source rolling bearing vibration signal sample set,
Figure GDA0004018962480000044
is the deep migration fault feature of the i-th sample in the target rolling bearing vibration signal sample set, and the superscript/subscript F 2 represents the F 2 layer of the domain-shared deep residual network;

3.2)预测健康状态的概率分布：3.2) Predict the probability distribution of health status:

将提取的特征同时输入到F3层，其中，目标滚动轴承对应的F3层神经元个数为

Figure GDA0004018962480000045
表示第sk个滚动轴承振动信号样本集的状态类别总数，则第
Figure GDA0004018962480000046
个神经元代表目标滚动轴承未知健康状态；利用Softmax激活函数预测域共享深度残差网络F3层特征属于源滚动轴承振动信号样本集健康状态的概率分布
Figure GDA0004018962480000047
属于目标滚动轴承未知健康状态的概率分布
Figure GDA0004018962480000048
其中，
Figure GDA0004018962480000049
为第sk个滚动轴承振动信号样本集中第i个样本属于源滚动轴承振动信号样本集健康状态的概率分布，
Figure GDA00040189624800000410
为目标滚动轴承振动信号样本集中第i个样本属于源滚动轴承信号集健康状态的概率分布，
Figure GDA00040189624800000411
为目标滚动轴承振动信号样本集中第i个样本属于未知健康状态概率分布，上/下标F3代表域共享深度残差网络的输出层F3层；根据下式，输出目标样本属于未知健康状态的诊断结果：The extracted features are simultaneously input into the F3 layer, where the number of neurons in the F3 layer corresponding to the target rolling bearing is
Figure GDA0004018962480000045
represents the total number of state categories of the s kth rolling bearing vibration signal sample set, then
Figure GDA0004018962480000046
The neurons represent the unknown health status of the target rolling bearing; the Softmax activation function is used to predict the probability distribution of the health status of the three- layer feature of the domain-sharing deep residual network F belonging to the source rolling bearing vibration signal sample set.
Figure GDA0004018962480000047
Probability distribution of unknown health states of target rolling bearing
Figure GDA0004018962480000048
in,
Figure GDA0004018962480000049
is the probability distribution that the i-th sample in the s k -th rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing vibration signal sample set,
Figure GDA00040189624800000410
is the probability distribution that the i-th sample in the target rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing signal set,
Figure GDA00040189624800000411
is the probability distribution of the i-th sample in the target rolling bearing vibration signal sample set belonging to the unknown health state. The superscript/subscript F 3 represents the output layer F 3 of the domain-shared deep residual network. According to the following formula, the diagnostic result that the target sample belongs to the unknown health state is output:

Figure GDA0004018962480000051
Figure GDA0004018962480000051

其中，m表示网络每次训练输入的样本量；

Figure GDA0004018962480000052
表示目标滚动轴承振动信号样本集中第i个样本属于未知健康状态的诊断结果，1表示属于未知健康状态，0表示不属于未知健康状态；Among them, m represents the number of samples input into the network for each training;
Figure GDA0004018962480000052
Indicates the diagnosis result that the i-th sample in the target rolling bearing vibration signal sample set belongs to an unknown health state, 1 indicates that it belongs to an unknown health state, and 0 indicates that it does not belong to an unknown health state;

3.3)计算域共享深度残差网络预测损失：3.3) Compute domain shared depth residual network prediction loss:

3.3.1)分别计算域共享深度残差网络预测源滚动轴承振动信号样本集和目标滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的交叉熵和信息熵损失：3.3.1) The cross entropy and information entropy loss of the health state of the source rolling bearing vibration signal sample set and the target rolling bearing vibration signal sample set are respectively calculated using the domain shared deep residual network:

Figure GDA0004018962480000053
Figure GDA0004018962480000053

Figure GDA0004018962480000054
Figure GDA0004018962480000054

其中，

Figure GDA0004018962480000055
表示预测第sk个源滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的交叉熵损失，
Figure GDA0004018962480000056
表示预测目标滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的信息熵损失，j表示源滚动轴承振动信号样本集中第j种健康状态，I(·)表示指示函数，
Figure GDA0004018962480000057
表示第sk个源滚动轴承振动信号样本集中第i个样本属于源滚动轴承振动信号样本集健康状态的诊断结果；in,
Figure GDA0004018962480000055
represents the cross entropy loss for predicting that the s kth source rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing vibration signal sample set,
Figure GDA0004018962480000056
represents the information entropy loss of predicting the health status of the target rolling bearing vibration signal sample set to the source rolling bearing vibration signal sample set, j represents the jth health status in the source rolling bearing vibration signal sample set, I(·) represents the indicator function,
Figure GDA0004018962480000057
Indicates that the i-th sample in the s k -th source rolling bearing vibration signal sample set belongs to the diagnosis result of the health state of the source rolling bearing vibration signal sample set;

3.3.2)计算域共享深度残差网络预测目标滚动轴承振动信号样本集属于未知健康状态的交叉熵损失：3.3.2) The cross entropy loss of the computational domain shared deep residual network predicts the target rolling bearing vibration signal sample set belonging to the unknown health state:

Figure GDA0004018962480000061
Figure GDA0004018962480000061

其中，

Figure GDA0004018962480000062
表示预测目标滚动轴承振动信号样本集属于未知健康状态的交叉熵损失；in,
Figure GDA0004018962480000062
The cross entropy loss represents the prediction that the target rolling bearing vibration signal sample set belongs to an unknown health state;

3.3.3)最小化上式交叉熵损失，更新域共享深度残差网络参数

Figure GDA0004018962480000063
即：3.3.3) Minimize the cross entropy loss above and update the domain shared deep residual network parameters
Figure GDA0004018962480000063
Right now:

Figure GDA0004018962480000064
Figure GDA0004018962480000064

3.4)构建参数共享的领域混淆网络并多次迭代更新网络参数：3.4) Construct a parameter-sharing domain confusion network and iterate and update the network parameters multiple times:

构建参数共享的领域混淆网络，领域混淆网络的待训练参数为θadv，该网络的输入为深度迁移故障特征

Figure GDA0004018962480000065
输出为领域混淆特征
Figure GDA0004018962480000066
其中，
Figure GDA0004018962480000067
为第sk个源滚动轴承振动信号样本集第i个样本的领域混淆特征，
Figure GDA0004018962480000068
为目标滚动轴承振动信号样本集第i个样本的领域混淆特征，上/下标adv代表领域混淆网络；最大化如下目标函数更新领域混淆网络的参数θadv，即：Construct a parameter-sharing domain confusion network. The parameter to be trained of the domain confusion network is θ adv . The input of the network is the deep migration fault feature.
Figure GDA0004018962480000065
The output is the domain confusion feature
Figure GDA0004018962480000066
in,
Figure GDA0004018962480000067
is the domain confusion feature of the ith sample in the s k th source rolling bearing vibration signal sample set,
Figure GDA0004018962480000068
is the domain confusion feature of the i-th sample in the target rolling bearing vibration signal sample set, and the superscript/subscript adv represents the domain confusion network; the parameter θ adv of the domain confusion network is updated by maximizing the following objective function, namely:

Figure GDA0004018962480000069
Figure GDA0004018962480000069

迭代更新nadv次领域混淆网络参数θadv，每次迭代更新后，将领域混淆网络的待训练参数θadv截断在范围{-ξ,ξ}内；Iteratively update n adv domain confusion network parameters θ adv . After each iterative update, the parameters θ adv to be trained of the domain confusion network are truncated within the range {-ξ,ξ}.

3.5)计算源滚动轴承振动信号样本集和目标滚动轴承振动信号样本集的深度迁移故障特征的局部分布差异：3.5) Calculate the local distribution difference of the deep migration fault characteristics of the source rolling bearing vibration signal sample set and the target rolling bearing vibration signal sample set:

3.5.1)计算目标滚动轴承振动信号样本集平衡导向的加权系数：3.5.1) Calculate the weighted coefficient of the target rolling bearing vibration signal sample set for balance guidance:

Figure GDA00040189624800000610
Figure GDA00040189624800000610

其中，

Figure GDA00040189624800000611
表示目标滚动轴承振动信号样本集第i个样本的加权系数，
Figure GDA0004018962480000071
为目标滚动轴承振动信号样本集第i个样本的领域混淆特征,
Figure GDA0004018962480000072
为目标滚动轴承振动信号样本集中第i个样本属于未知健康状态概率分布，上/下标F3代表域共享深度残差网络的输出层(F3层),σs(·)表示激活函数；in,
Figure GDA00040189624800000611
represents the weighting coefficient of the i-th sample of the target rolling bearing vibration signal sample set,
Figure GDA0004018962480000071
is the domain confusion feature of the i-th sample in the target rolling bearing vibration signal sample set,
Figure GDA0004018962480000072
is the probability distribution of the i-th sample in the target rolling bearing vibration signal sample set belonging to the unknown health state, the superscript/subscript F 3 represents the output layer (F 3 layer) of the domain-shared deep residual network, and σ s (·) represents the activation function;

3.5.2)对目标滚动轴承振动信号样本集深度迁移故障特征加权，计算局部分布的最大均值差异：3.5.2) Weight the deep migration fault features of the target rolling bearing vibration signal sample set and calculate the maximum mean difference of the local distribution:

Figure GDA0004018962480000073
Figure GDA0004018962480000073

其中，

Figure GDA0004018962480000074
表示多项式核最大均值差异函数；in,
Figure GDA0004018962480000074
represents the polynomial kernel maximum mean difference function;

3.6)多滚动轴承融合迁移诊断模型训练：3.6) Training of multi-roller bearing fusion migration diagnosis model:

3.6.1)对基于不同源滚动轴承振动信号样本集健康状态知识学习到的目标滚动轴承深度迁移故障特征进行特征分布适配，计算其最大均值差异为：3.6.1) Perform feature distribution adaptation on the target rolling bearing deep migration fault feature learned based on the health status knowledge of the rolling bearing vibration signal sample set from different sources, and calculate the maximum mean difference as:

Figure GDA0004018962480000075
Figure GDA0004018962480000075

其中，S＝{s1,s2,...,sN}表示N个源滚动轴承振动信号样本集，

Figure GDA0004018962480000076
分别表示为基于第si、sj个源滚动轴承振动信号样本集知识提取的目标滚动轴承振动信号样本集的深度迁移故障特征，上/下标F2代表域共享深度残差网络的F2层；Where, S = {s 1 ,s 2 ,...,s N } represents N source rolling bearing vibration signal sample sets,
Figure GDA0004018962480000076
They are respectively represented as the deep migration fault features of the target rolling bearing vibration signal sample set extracted based on the knowledge of the s i th and s j th source rolling bearing vibration signal sample sets, and the superscript/subscript F 2 represents the F 2 layer of the domain-shared deep residual network;

3.6.2)最小化如下目标函数同时更新N个域共享深度残差网络的参数集

Figure GDA0004018962480000077
即：3.6.2) Minimize the following objective function and update the parameter set of N domains shared deep residual network at the same time
Figure GDA0004018962480000077
Right now:

Figure GDA0004018962480000078
Figure GDA0004018962480000078

其中，λ,β,γ是权衡参数。Among them, λ, β, γ are trade-off parameters.

本发明的有益效果为：本发明提出了一种多源滚动轴承健康状态融合的迁移智能诊断方法，一方面，通过融合多个源滚动轴承的健康状态知识对目标滚动轴承的健康状态进行诊断，解决了源滚动轴承的诊断知识可能无法涵盖目标滚动轴承的所有故障类型的问题；另一方面，通过计算目标滚动轴承振动信号样本集平衡导向的加权系数并对特征分布适配中的目标滚动轴承振动信号样本集深度迁移故障特征加权，改善目标滚动轴承样本不平衡的问题，进而提高了现有迁移智能诊断技术的诊断精度。The beneficial effects of the present invention are as follows: the present invention proposes a migration intelligent diagnosis method for multi-source rolling bearing health status fusion. On the one hand, the health status of the target rolling bearing is diagnosed by fusing the health status knowledge of multiple source rolling bearings, thereby solving the problem that the diagnostic knowledge of the source rolling bearings may not cover all fault types of the target rolling bearings; on the other hand, by calculating the balance-oriented weighting coefficient of the target rolling bearing vibration signal sample set and weighting the deep migration fault features of the target rolling bearing vibration signal sample set in the feature distribution adaptation, the problem of target rolling bearing sample imbalance is improved, thereby improving the diagnostic accuracy of the existing migration intelligent diagnosis technology.

附图说明BRIEF DESCRIPTION OF THE DRAWINGS

图1为本发明的整体流程图。FIG1 is an overall flow chart of the present invention.

图2为基于第sk个源滚动轴承-目标滚动轴承振动信号样本集对构建的局部分布适配子模型框架图。FIG2 is a framework diagram of a local distribution adaptation sub-model constructed based on the s kth source rolling bearing-target rolling bearing vibration signal sample set pair.

图3为多滚动轴承融合迁移诊断模型框架图。Figure 3 is a framework diagram of the multi-roller bearing fusion migration diagnosis model.

具体实施方式DETAILED DESCRIPTION

下面结合附图和实施例对本发明进一步详细描述。The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

如图1所示，一种多源滚动轴承健康状态融合的迁移智能诊断方法，包括以下步骤：As shown in FIG1 , a migration intelligent diagnosis method for multi-source rolling bearing health status fusion includes the following steps:

1)获取多个源滚动轴承振动信号样本集和目标滚动轴承振动信号样本集：1) Obtain multiple source rolling bearing vibration signal sample sets and target rolling bearing vibration signal sample sets:

获取N个源滚动轴承振动信号样本集

Figure GDA0004018962480000081
其中，
Figure GDA0004018962480000082
表示第sk个源滚动轴承样本集，包含的样本量为
Figure GDA0004018962480000083
表示第sk个源滚动轴承样本集的状态标签，
Figure GDA0004018962480000091
其中
Figure GDA0004018962480000092
表示第sk个源滚动轴承样本集的状态类别总数；获取目标滚动轴承的振动信号样本集
Figure GDA0004018962480000093
其中
Figure GDA0004018962480000094
表示目标滚动轴承样本集，包含的样本量为nt；Obtain N source rolling bearing vibration signal sample sets
Figure GDA0004018962480000081
in,
Figure GDA0004018962480000082
represents the s kth source rolling bearing sample set, which contains the number of samples:
Figure GDA0004018962480000083
represents the state label of the s kth source rolling bearing sample set,
Figure GDA0004018962480000091
in
Figure GDA0004018962480000092
Represents the total number of state categories of the s kth source rolling bearing sample set; obtains the vibration signal sample set of the target rolling bearing
Figure GDA0004018962480000093
in
Figure GDA0004018962480000094
represents the target rolling bearing sample set, which contains n t samples;

2)由单一源滚动轴承-目标滚动轴承振动信号样本集对构建局部分布适配子模型，如图2所示；2) A local distribution adapter model is constructed from a single source rolling bearing-target rolling bearing vibration signal sample set pair, as shown in FIG2 ;

将N个源滚动轴承振动信号样本集与单一目标滚动轴承振动信号样本集配对，共组成N对

Figure GDA0004018962480000095
构建N个如图2所示的局部分布适配子模型与这N个源滚动轴承-目标滚动轴承振动信号样本集对一一对应；Pair N source rolling bearing vibration signal sample sets with a single target rolling bearing vibration signal sample set to form a total of N pairs
Figure GDA0004018962480000095
Construct N local distribution adapter sub-models as shown in FIG2 to correspond one-to-one to the N source rolling bearing-target rolling bearing vibration signal sample set pairs;

3)同时训练多个局部分布适配子模型：3) Training multiple local distribution adapter models simultaneously:

执行如下步骤同时训练这N个局部分布适配子模型，以第sk个源滚动轴承-目标滚动轴承振动信号样本集对

Figure GDA0004018962480000096
详细说明每个局部分布适配子模型的训练过程，包括如下步骤：Perform the following steps to train the N local distribution adapter sub-models simultaneously, taking the s kth source rolling bearing-target rolling bearing vibration signal sample set pair
Figure GDA0004018962480000096
The training process of each local distribution adapter model is described in detail, including the following steps:

3.1)计算第sk个源滚动轴承-目标滚动轴承振动信号样本集对的深度迁移故障特征：3.1) Calculate the deep migration fault feature of the s kth source rolling bearing-target rolling bearing vibration signal sample set pair:

同时从第sk个源滚动轴承与目标滚动轴承振动信号样本集中提取深度迁移故障特征

Figure GDA0004018962480000097
其中，
Figure GDA0004018962480000098
为第sk个源滚动轴承振动信号样本集第i个样本的深度迁移故障特征，
Figure GDA0004018962480000099
为目标滚动轴承振动信号样本集第i个样本的深度迁移故障特征，上/下标F2代表域共享深度残差网络的F2层；At the same time, deep migration fault features are extracted from the vibration signal sample sets of the s kth source rolling bearing and the target rolling bearing.
Figure GDA0004018962480000097
in,
Figure GDA0004018962480000098
is the deep migration fault feature of the ith sample of the s k th source rolling bearing vibration signal sample set,
Figure GDA0004018962480000099
is the deep migration fault feature of the i-th sample in the target rolling bearing vibration signal sample set, and the superscript/subscript F 2 represents the F 2 layer of the domain-shared deep residual network;

3.2)预测健康状态的概率分布：3.2) Predict the probability distribution of health status:

将提取的特征同时输入到F3层，其中，目标滚动轴承对应的F3层神经元个数为

Figure GDA0004018962480000101
表示第sk个滚动轴承振动信号样本集的状态类别总数，则第
Figure GDA0004018962480000102
个神经元代表目标滚动轴承未知健康状态；利用Softmax激活函数预测域共享深度残差网络F3层特征属于源滚动轴承振动信号样本集健康状态的概率分布
Figure GDA0004018962480000103
属于目标滚动轴承未知健康状态的概率分布
Figure GDA0004018962480000104
其中，
Figure GDA0004018962480000105
为第sk个滚动轴承振动信号样本集中第i个样本属于源滚动轴承振动信号样本集健康状态的概率分布，
Figure GDA0004018962480000106
为目标滚动轴承振动信号样本集中第i个样本属于源滚动轴承信号集健康状态的概率分布，
Figure GDA0004018962480000107
为目标滚动轴承振动信号样本集中第i个样本属于未知健康状态概率分布，上/下标F3代表域共享深度残差网络的输出层F3层；根据下式，输出目标样本属于未知健康状态的诊断结果：The extracted features are simultaneously input into the F3 layer, where the number of neurons in the F3 layer corresponding to the target rolling bearing is
Figure GDA0004018962480000101
represents the total number of state categories of the s kth rolling bearing vibration signal sample set, then
Figure GDA0004018962480000102
The neurons represent the unknown health status of the target rolling bearing; the Softmax activation function is used to predict the probability distribution of the health status of the three- layer feature of the domain-sharing deep residual network F belonging to the source rolling bearing vibration signal sample set.
Figure GDA0004018962480000103
Probability distribution of unknown health states of target rolling bearing
Figure GDA0004018962480000104
in,
Figure GDA0004018962480000105
is the probability distribution that the i-th sample in the s k -th rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing vibration signal sample set,
Figure GDA0004018962480000106
is the probability distribution that the i-th sample in the target rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing signal set,
Figure GDA0004018962480000107
is the probability distribution of the i-th sample in the target rolling bearing vibration signal sample set belonging to the unknown health state. The superscript/subscript F 3 represents the output layer F 3 of the domain-shared deep residual network. According to the following formula, the diagnostic result that the target sample belongs to the unknown health state is output:

Figure GDA0004018962480000108
Figure GDA0004018962480000108

其中，m表示网络每次训练输入的样本量；

Figure GDA0004018962480000109
表示目标滚动轴承振动信号样本集中第i个样本属于未知健康状态的诊断结果，1表示属于未知健康状态，0表示不属于未知健康状态；Among them, m represents the number of samples input into the network for each training;
Figure GDA0004018962480000109
Indicates the diagnosis result that the i-th sample in the target rolling bearing vibration signal sample set belongs to an unknown health state, 1 indicates that it belongs to an unknown health state, and 0 indicates that it does not belong to an unknown health state;

3.3)计算域共享深度残差网络预测损失：3.3) Compute domain shared depth residual network prediction loss:

3.3.1)分别计算域共享深度残差网络预测源滚动轴承振动信号样本集和目标滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的交叉熵和信息熵损失：3.3.1) The cross entropy and information entropy loss of the health state of the source rolling bearing vibration signal sample set and the target rolling bearing vibration signal sample set are respectively calculated using the domain shared deep residual network:

Figure GDA0004018962480000111
Figure GDA0004018962480000111

Figure GDA0004018962480000112
Figure GDA0004018962480000112

其中，

Figure GDA0004018962480000113
表示预测第sk个源滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的交叉熵损失，
Figure GDA0004018962480000114
表示预测目标滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的信息熵损失，j表示源滚动轴承振动信号样本集中第j种健康状态，I(·)表示指示函数，
Figure GDA0004018962480000115
表示第sk个源滚动轴承振动信号样本集中第i个样本属于源滚动轴承振动信号样本集健康状态的诊断结果；in,
Figure GDA0004018962480000113
represents the cross entropy loss for predicting that the s kth source rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing vibration signal sample set,
Figure GDA0004018962480000114
represents the information entropy loss of predicting the health status of the target rolling bearing vibration signal sample set to the source rolling bearing vibration signal sample set, j represents the jth health status in the source rolling bearing vibration signal sample set, I(·) represents the indicator function,
Figure GDA0004018962480000115
Indicates that the i-th sample in the s k -th source rolling bearing vibration signal sample set belongs to the diagnosis result of the health state of the source rolling bearing vibration signal sample set;

3.3.2)计算域共享深度残差网络预测目标滚动轴承振动信号样本集属于未知健康状态的交叉熵损失：3.3.2) The cross entropy loss of the computational domain shared deep residual network predicts the target rolling bearing vibration signal sample set belonging to the unknown health state:

Figure GDA0004018962480000116
Figure GDA0004018962480000116

其中，

Figure GDA0004018962480000117
表示预测目标滚动轴承振动信号样本集属于未知健康状态的交叉熵损失；in,
Figure GDA0004018962480000117
The cross entropy loss represents the prediction that the target rolling bearing vibration signal sample set belongs to an unknown health state;

3.3.3)最小化上式交叉熵损失，更新域共享深度残差网络参数

Figure GDA0004018962480000118
即：3.3.3) Minimize the cross entropy loss above and update the domain shared deep residual network parameters
Figure GDA0004018962480000118
Right now:

Figure GDA0004018962480000119
Figure GDA0004018962480000119

3.4)构建参数共享的领域混淆网络并多次迭代更新网络参数：3.4) Construct a parameter-sharing domain confusion network and iterate and update the network parameters multiple times:

构建参数共享的领域混淆网络，领域混淆网络的待训练参数为θadv，该网络的输入为深度迁移故障特征

Figure GDA00040189624800001110
输出为领域混淆特征
Figure GDA00040189624800001111
其中，
Figure GDA00040189624800001112
为第sk个源滚动轴承振动信号样本集第i个样本的领域混淆特征，
Figure GDA0004018962480000121
为目标滚动轴承振动信号样本集第i个样本的领域混淆特征，上/下标adv代表领域混淆网络；最大化如下目标函数更新领域混淆网络的参数θadv，即：Construct a parameter-sharing domain confusion network. The parameter to be trained of the domain confusion network is θ adv . The input of the network is the deep migration fault feature.
Figure GDA00040189624800001110
The output is the domain confusion feature
Figure GDA00040189624800001111
in,
Figure GDA00040189624800001112
is the domain confusion feature of the ith sample in the s k th source rolling bearing vibration signal sample set,
Figure GDA0004018962480000121
is the domain confusion feature of the i-th sample in the target rolling bearing vibration signal sample set, and the superscript/subscript adv represents the domain confusion network; the parameter θ adv of the domain confusion network is updated by maximizing the following objective function, namely:

Figure GDA0004018962480000122
Figure GDA0004018962480000122

迭代更新nadv次领域混淆网络参数θadv，每次迭代更新后，将领域混淆网络的待训练参数θadv截断在范围{-ξ,ξ}内；Iteratively update n adv domain confusion network parameters θ adv . After each iterative update, the parameters θ adv to be trained of the domain confusion network are truncated within the range {-ξ,ξ}.

3.5)计算源滚动轴承振动信号样本集和目标滚动轴承振动信号样本集的深度迁移故障特征的局部分布差异：3.5) Calculate the local distribution difference of the deep migration fault characteristics of the source rolling bearing vibration signal sample set and the target rolling bearing vibration signal sample set:

3.5.1)计算目标滚动轴承振动信号样本集平衡导向的加权系数：3.5.1) Calculate the weighted coefficient of the target rolling bearing vibration signal sample set for balance guidance:

Figure GDA0004018962480000123
Figure GDA0004018962480000123

其中，

Figure GDA0004018962480000124
表示目标滚动轴承振动信号样本集第i个样本的加权系数，
Figure GDA0004018962480000125
为目标滚动轴承振动信号样本集第i个样本的领域混淆特征,
Figure GDA0004018962480000126
为目标滚动轴承振动信号样本集中第i个样本属于未知健康状态概率分布，上/下标F3代表域共享深度残差网络的输出层(F3层),σs(·)表示激活函数；in,
Figure GDA0004018962480000124
represents the weighting coefficient of the i-th sample of the target rolling bearing vibration signal sample set,
Figure GDA0004018962480000125
is the domain confusion feature of the i-th sample in the target rolling bearing vibration signal sample set,
Figure GDA0004018962480000126
is the probability distribution of the i-th sample in the target rolling bearing vibration signal sample set belonging to the unknown health state, the superscript/subscript F 3 represents the output layer (F 3 layer) of the domain-shared deep residual network, and σ s (·) represents the activation function;

3.5.2)对目标滚动轴承振动信号样本集深度迁移故障特征加权，计算局部分布的最大均值差异：3.5.2) Weight the deep migration fault features of the target rolling bearing vibration signal sample set and calculate the maximum mean difference of the local distribution:

Figure GDA0004018962480000127
Figure GDA0004018962480000127

其中，

Figure GDA0004018962480000128
表示多项式核最大均值差异函数；in,
Figure GDA0004018962480000128
represents the polynomial kernel maximum mean difference function;

3.6)如图3所示，多滚动轴承融合迁移诊断模型训练：3.6) As shown in Figure 3, multi-roller bearing fusion migration diagnosis model training:

3.6.1)对基于不同源滚动轴承振动信号样本集健康状态知识学习到的目标滚动轴承深度迁移故障特征进行特征分布适配，计算其最大均值差异为：3.6.1) Perform feature distribution adaptation on the target rolling bearing deep migration fault feature learned based on the health status knowledge of the rolling bearing vibration signal sample set from different sources, and calculate the maximum mean difference as:

Figure GDA0004018962480000131
Figure GDA0004018962480000131

其中，S＝{s1,s2,...,sN}表示N个源滚动轴承振动信号样本集，

Figure GDA0004018962480000132
分别表示为基于第si、sj个源滚动轴承振动信号样本集知识提取的目标滚动轴承振动信号样本集的深度迁移故障特征，上/下标F2代表域共享深度残差网络的F2层；Where, S = {s 1 ,s 2 ,...,s N } represents N source rolling bearing vibration signal sample sets,
Figure GDA0004018962480000132
They are respectively represented as the deep migration fault features of the target rolling bearing vibration signal sample set extracted based on the knowledge of the s i th and s j th source rolling bearing vibration signal sample sets, and the superscript/subscript F 2 represents the F 2 layer of the domain-shared deep residual network;

3.6.2)最小化如下目标函数同时更新N个域共享深度残差网络的参数集

Figure GDA0004018962480000133
即：3.6.2) Minimize the following objective function and update the parameter set of N domains shared deep residual network at the same time
Figure GDA0004018962480000133
Right now:

Figure GDA0004018962480000134
Figure GDA0004018962480000134

其中，λ,β,γ是权衡参数。Among them, λ, β, γ are trade-off parameters.

4)重复执行步骤3)至网络参数

Figure GDA0004018962480000135
收敛；4) Repeat step 3) until network parameters are reached
Figure GDA0004018962480000135
convergence;

5)融合基于不同源滚动轴承对目标滚动轴承的诊断结果获取目标滚动轴承振动信号样本集健康状态的最终诊断结果：5) The final diagnosis result of the health status of the target rolling bearing vibration signal sample set is obtained by fusing the diagnosis results of the target rolling bearing based on different source rolling bearings:

Figure GDA0004018962480000136
Figure GDA0004018962480000136

其中，

Figure GDA0004018962480000137
表示目标滚动轴承振动信号样本集中第i个样本健康状态诊断结果，
Figure GDA0004018962480000138
表示第sk个源滚动轴承振动信号样本集的状态标签，
Figure GDA0004018962480000139
in,
Figure GDA0004018962480000137
represents the health status diagnosis result of the i-th sample in the target rolling bearing vibration signal sample set,
Figure GDA0004018962480000138
represents the state label of the s kth source rolling bearing vibration signal sample set,
Figure GDA0004018962480000139

实施例：应用三组滚动轴承振动信号样本集，验证本发明的可行性。Embodiment: Three groups of rolling bearing vibration signal sample sets are used to verify the feasibility of the present invention.

数据集D来自滚动轴承加速寿命试验台，加速计用于监测被测轴承(LDK UER 204)的退化情况，在试验期间，驱动试验轴承的交流电机的速度设定为2400r/min，液压缸在轴承外圈上提供11kN的径向负载，采样频率设定为25.6kHz，实验人员在试验台发生故障后，停机并检查了实验轴承，确认了轴承外圈裂纹，因此，收集的数据具有从正常(N)阶段到外圈故障(OF)阶段的降级信息，对健康状态下轴承整个寿命早期阶段的数据和故障状态下轴承最终阶段的数据进行了标记，数据集D中有724个样本，它们在健康和故障状态上是平衡的，每个样本包含1200个采样点。Dataset D is from a rolling bearing accelerated life test bench. An accelerometer is used to monitor the degradation of the bearing under test (LDK UER 204). During the test, the speed of the AC motor driving the test bearing was set to 2400r/min, and the hydraulic cylinder provided a radial load of 11kN on the outer ring of the bearing. The sampling frequency was set to 25.6kHz. After a failure occurred on the test bench, the experimenter stopped and inspected the test bearing and confirmed the crack in the outer ring of the bearing. Therefore, the collected data has degradation information from the normal (N) stage to the outer ring failure (OF) stage. The data of the early stage of the entire life of the bearing in the healthy state and the data of the final stage of the bearing in the faulty state are marked. There are 724 samples in dataset D, which are balanced in the healthy and faulty states, and each sample contains 1200 sampling points.

数据集E由机械故障预防技术协会提供，该测试台配备了RBC NICE轴承，这些轴承是健康的或包含内圈故障(IF)的，工程师在实验室中模拟了滚动轴承的内圈磨损，输入轴转速设置为1500r/min，正常样品在270lbs的负载下采集，采样频率为97.656kHz，在200lbs、250lbs和300lbs的载荷下采集IF样品，采样频率为48.828kHz。数据集E包含732个样本，样本数平衡。Dataset E is provided by the Association for Mechanical Failure Prevention Technology. The test bench is equipped with RBC NICE bearings, which are either healthy or contain inner race faults (IF). Engineers simulated the inner race wear of rolling bearings in the laboratory. The input shaft speed is set to 1500r/min. Normal samples are collected under a load of 270lbs, with a sampling frequency of 97.656kHz. IF samples are collected under loads of 200lbs, 250lbs, and 300lbs, with a sampling frequency of 48.828kHz. Dataset E contains 732 samples, and the number of samples is balanced.

数据集F来自机车轴承自动试验台(197726)，该试验台可直接对装有轴承的机车轮对进行试验，被测轴承由液压马达驱动，由液压缸加载，有三种轮对：普通轮对和轴承内圈或外圈有表面抛光的轮对，试验转速450r/min，径向载荷680kg，振动传感器以76.8kHz的采样频率采集监测数据，如表1所示，数据集F由三种健康状态的样本组成，其中健康样本832个，故障样本1664×m％，注意m％∈(0,1)使得数据集F处于不同的不平衡水平，详细信息如表1所示：Dataset F comes from the locomotive bearing automatic test bench (197726). The test bench can directly test locomotive wheels equipped with bearings. The tested bearings are driven by hydraulic motors and loaded by hydraulic cylinders. There are three types of wheelsets: ordinary wheelsets and wheelsets with polished inner or outer rings. The test speed is 450r/min and the radial load is 680kg. The vibration sensor collects monitoring data at a sampling frequency of 76.8kHz, as shown in Table 1. Dataset F consists of samples in three healthy states, including 832 healthy samples and 1664×m% faulty samples. Note that m%∈(0,1) makes the dataset F at different imbalance levels. The detailed information is shown in Table 1:

表1.三组轴承数据集详细信息Table 1. Detailed information of the three bearing datasets

Figure GDA0004018962480000151
Figure GDA0004018962480000151

注意：随机选择m％∈(0,1)错误样本以使数据集F不平衡。Note: m%∈(0,1) error samples are randomly selected to make the dataset F unbalanced.

创建多源迁移学习任务(D，E)→F来验证提出的方法。它用于模拟诊断知识从实验室转移到现实的情况，因为在工程场景中始终没有足够的标记样本，源轴承数据集D或数据集E无法提供目标轴承数据集F所需的足够的诊断知识，此外，数据集F中的样本在健康状态之间不平衡，取m＝20,即数据集F样本不平衡度为20％。选取F-scroe、平均精度均值(meanaverage precision，mAP)、与AUC分类指标量化本发明在迁移诊断任务上的效果，重复实验10次，计算诊断结果的统计值，如表2所示，本发明的三种指标分别为0.936、0.983、0.968，指标接近于1，说明本发明方法的诊断准确性高，验证了本发明在解决工程实际中源域健康状态类型无法覆盖目标域健康状态类型且目标域样本不平衡情况下的迁移诊断问题中的可行性。A multi-source transfer learning task (D, E) → F is created to verify the proposed method. It is used to simulate the situation where diagnostic knowledge is transferred from the laboratory to reality. Because there are always not enough labeled samples in the engineering scenario, the source bearing dataset D or dataset E cannot provide sufficient diagnostic knowledge required by the target bearing dataset F. In addition, the samples in dataset F are unbalanced between health states. Take m = 20, that is, the sample imbalance of dataset F is 20%. The F-scroe, mean average precision (mAP), and AUC classification indicators are selected to quantify the effect of the present invention on the migration diagnosis task. The experiment is repeated 10 times, and the statistical values of the diagnosis results are calculated. As shown in Table 2, the three indicators of the present invention are 0.936, 0.983, and 0.968, respectively. The indicators are close to 1, indicating that the diagnostic accuracy of the method of the present invention is high, which verifies the feasibility of the present invention in solving the migration diagnosis problem in the case where the source domain health state type cannot cover the target domain health state type and the target domain samples are unbalanced in engineering practice.

表2不同方法的诊断效果对比Table 2 Comparison of diagnostic effects of different methods

Figure GDA0004018962480000161
Figure GDA0004018962480000161

另选取P-ResNet与普通ResNet方法对比本发明方法的效果，P-ResNet的平均精度均值为0.516，接近随机诊断模型，F-score和AUC均明显低于本发明方法，普通ResNet方法平均精度均值、F-score和AUC在三种方法中最低。P-ResNet and ordinary ResNet methods were selected to compare the effects of the method of the present invention. The average precision mean of P-ResNet was 0.516, close to the random diagnosis model, and the F-score and AUC were significantly lower than those of the method of the present invention. The average precision mean, F-score and AUC of the ordinary ResNet method were the lowest among the three methods.

通过对比本发明方法与普通迁移诊断方法(P-ResNet)和普通深度智能诊断方法(ResNet)，表明本发明方法有效地克服源域健康状态类型无法覆盖目标域健康状态类型且目标域样本不平衡的影响，提高了迁移诊断模型的性能。By comparing the method of the present invention with the common migration diagnosis method (P-ResNet) and the common deep intelligent diagnosis method (ResNet), it is shown that the method of the present invention effectively overcomes the influence of the source domain health status type being unable to cover the target domain health status type and the target domain sample imbalance, and improves the performance of the migration diagnosis model.

Claims (1)

1.一种多源滚动轴承健康状态融合的迁移智能诊断方法，其特征在于，包括以下步骤：1. A migration intelligent diagnosis method for multi-source rolling bearing health status fusion, characterized by comprising the following steps: 1)获取多个源滚动轴承振动信号样本集和目标轴承振动信号样本集；1) Acquire multiple source rolling bearing vibration signal sample sets and target bearing vibration signal sample sets; 2)由单一源滚动轴承-目标滚动轴承振动信号样本集对构建局部分布适配子模型；2) constructing a local distribution adapter model from a single source rolling bearing-target rolling bearing vibration signal sample set; 3)同时训练多个局部分布适配子模型；3) Training multiple local distribution adapter models simultaneously; 4)重复执行步骤3至网络参数
Figure FDA0004018962470000011
收敛；4) Repeat step 3 to network parameters
Figure FDA0004018962470000011
convergence; 5)融合基于不同源滚动轴承对目标滚动轴承的诊断结果获取目标滚动轴承振动信号样本集健康状态的最终诊断结果：5) The final diagnosis result of the health status of the target rolling bearing vibration signal sample set is obtained by fusing the diagnosis results of the target rolling bearing based on different source rolling bearings:
Figure FDA0004018962470000012
Figure FDA0004018962470000012
 其中，
Figure FDA00040189624700000111
表示目标滚动轴承振动信号样本集中第i个样本健康状态诊断结果，
Figure FDA0004018962470000013
表示第sk个源滚动轴承振动信号样本集的状态标签，
Figure FDA0004018962470000014
in,
Figure FDA00040189624700000111
represents the health status diagnosis result of the i-th sample in the target rolling bearing vibration signal sample set,
Figure FDA0004018962470000013
represents the state label of the s kth source rolling bearing vibration signal sample set,
Figure FDA0004018962470000014
 所述的步骤1)具体为：获取N个源滚动轴承振动信号样本集
Figure FDA0004018962470000015
其中，
Figure FDA0004018962470000016
表示第sk个源滚动轴承振动信号样本集，包含的样本量为
Figure FDA0004018962470000017
Figure FDA0004018962470000018
表示第sk个源滚动轴承振动信号样本集的状态标签，
Figure FDA0004018962470000019
其中
Figure FDA00040189624700000110
表示第sk个源滚动轴承振动信号样本集的状态类别总数；获取目标滚动轴承振动信号样本集
Figure FDA0004018962470000021
其中
Figure FDA0004018962470000022
表示目标滚动轴承振动信号样本集，包含的样本量为nt；The step 1) is specifically as follows: obtaining N source rolling bearing vibration signal sample sets
Figure FDA0004018962470000015
in,
Figure FDA0004018962470000016
represents the s kth source rolling bearing vibration signal sample set, which contains the number of samples:
Figure FDA0004018962470000017
Figure FDA0004018962470000018
represents the state label of the s kth source rolling bearing vibration signal sample set,
Figure FDA0004018962470000019
in
Figure FDA00040189624700000110
Represents the total number of state categories of the s kth source rolling bearing vibration signal sample set; obtains the target rolling bearing vibration signal sample set
Figure FDA0004018962470000021
in
Figure FDA0004018962470000022
represents the target rolling bearing vibration signal sample set, which contains n t samples; 所述的步骤2)具体为：将N个源滚动轴承振动信号样本集与单一目标滚动轴承振动信号样本集配对，共组成N对
Figure FDA0004018962470000023
构建N个局部分布适配子模型与这N个源滚动轴承-目标滚动轴承振动信号样本集对一一对应；The step 2) is specifically as follows: pairing N source rolling bearing vibration signal sample sets with a single target rolling bearing vibration signal sample set to form a total of N pairs.
Figure FDA0004018962470000023
Constructing N local distribution adapter sub-models corresponding to the N source rolling bearing-target rolling bearing vibration signal sample set pairs one by one; 所述的步骤3)具体为：执行如下步骤同时训练这N个局部分布适配子模型，以第sk个源滚动轴承-目标滚动轴承振动信号样本集对
Figure FDA0004018962470000024
说明每个局部分布适配子模型的训练过程，包括如下步骤：The step 3) is specifically as follows: perform the following steps to simultaneously train the N local distribution adapter sub-models, using the s kth source rolling bearing-target rolling bearing vibration signal sample set pair
Figure FDA0004018962470000024
The training process of each local distribution adapter model is described, including the following steps: 3.1)计算第sk个源滚动轴承-目标滚动轴承振动信号样本集对的深度迁移故障特征：3.1) Calculate the deep migration fault feature of the s kth source rolling bearing-target rolling bearing vibration signal sample set pair: 同时从第sk个源滚动轴承与目标滚动轴承振动信号样本集中提取深度迁移故障特征
Figure FDA0004018962470000025
其中，
Figure FDA0004018962470000026
为第sk个源滚动轴承振动信号样本集第i个样本的深度迁移故障特征，
Figure FDA0004018962470000027
为目标滚动轴承振动信号样本集第i个样本的深度迁移故障特征，上/下标F2代表域共享深度残差网络的F2层；At the same time, deep migration fault features are extracted from the vibration signal sample sets of the s kth source rolling bearing and the target rolling bearing.
Figure FDA0004018962470000025
in,
Figure FDA0004018962470000026
is the deep migration fault feature of the ith sample of the s k th source rolling bearing vibration signal sample set,
Figure FDA0004018962470000027
is the deep migration fault feature of the i-th sample in the target rolling bearing vibration signal sample set, and the superscript/subscript F 2 represents the F 2 layer of the domain-shared deep residual network; 3.2)预测健康状态的概率分布：3.2) Predict the probability distribution of health status: 将提取的特征同时输入到F3层，其中，目标滚动轴承对应的F3层神经元个数为
Figure FDA0004018962470000028
Figure FDA0004018962470000029
表示第sk个滚动轴承振动信号样本集的状态类别总数，则第
Figure FDA00040189624700000210
个神经元代表目标滚动轴承未知健康状态；利用Softmax激活函数预测域共享深度残差网络F3层特征属于源滚动轴承振动信号样本集健康状态的概率分布
Figure FDA00040189624700000211
属于目标滚动轴承未知健康状态的概率分布
Figure FDA0004018962470000031
其中，
Figure FDA0004018962470000032
为第sk个滚动轴承振动信号样本集中第i个样本属于源滚动轴承振动信号样本集健康状态的概率分布，
Figure FDA0004018962470000033
为目标滚动轴承振动信号样本集中第i个样本属于源滚动轴承信号集健康状态的概率分布，
Figure FDA0004018962470000034
为目标滚动轴承振动信号样本集中第i个样本属于未知健康状态概率分布，上/下标F3代表域共享深度残差网络的输出层F3层；根据下式，输出目标样本属于未知健康状态的诊断结果：The extracted features are simultaneously input into the F3 layer, where the number of neurons in the F3 layer corresponding to the target rolling bearing is
Figure FDA0004018962470000028
Figure FDA0004018962470000029
represents the total number of state categories of the s kth rolling bearing vibration signal sample set, then
Figure FDA00040189624700000210
The neurons represent the unknown health status of the target rolling bearing; the Softmax activation function is used to predict the probability distribution of the health status of the three- layer feature of the domain-sharing deep residual network F belonging to the source rolling bearing vibration signal sample set.
Figure FDA00040189624700000211
Probability distribution of unknown health states of target rolling bearing
Figure FDA0004018962470000031
in,
Figure FDA0004018962470000032
is the probability distribution that the i-th sample in the s k -th rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing vibration signal sample set,
Figure FDA0004018962470000033
is the probability distribution that the i-th sample in the target rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing signal set,
Figure FDA0004018962470000034
is the probability distribution of the i-th sample in the target rolling bearing vibration signal sample set belonging to the unknown health state. The superscript/subscript F 3 represents the output layer F 3 of the domain-shared deep residual network. According to the following formula, the diagnostic result that the target sample belongs to the unknown health state is output:
Figure FDA0004018962470000035
Figure FDA0004018962470000035
 其中，m表示网络每次训练输入的样本量；
Figure FDA0004018962470000036
表示目标滚动轴承振动信号样本集中第i个样本属于未知健康状态的诊断结果，1表示属于未知健康状态，0表示不属于未知健康状态；Among them, m represents the number of samples input into the network for each training;
Figure FDA0004018962470000036
Indicates the diagnosis result that the i-th sample in the target rolling bearing vibration signal sample set belongs to an unknown health state, 1 indicates that it belongs to an unknown health state, and 0 indicates that it does not belong to an unknown health state; 3.3)计算域共享深度残差网络预测损失：3.3) Compute domain shared depth residual network prediction loss: 3.3.1)分别计算域共享深度残差网络预测源滚动轴承振动信号样本集和目标滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的交叉熵和信息熵损失：3.3.1) The cross entropy and information entropy loss of the health state of the source rolling bearing vibration signal sample set and the target rolling bearing vibration signal sample set are respectively calculated using the domain shared deep residual network:
Figure FDA0004018962470000037
Figure FDA0004018962470000037
Figure FDA0004018962470000038
Figure FDA0004018962470000038
 其中，
Figure FDA0004018962470000039
表示预测第sk个源滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的交叉熵损失，
Figure FDA00040189624700000310
表示预测目标滚动轴承振动信号样本集属于源滚动轴承振动信号样本集健康状态的信息熵损失，j表示源滚动轴承振动信号样本集中第j种健康状态，I(·)表示指示函数，
Figure FDA0004018962470000041
表示第sk个源滚动轴承振动信号样本集中第i个样本属于源滚动轴承振动信号样本集健康状态的诊断结果；in,
Figure FDA0004018962470000039
represents the cross entropy loss for predicting that the s kth source rolling bearing vibration signal sample set belongs to the healthy state of the source rolling bearing vibration signal sample set,
Figure FDA00040189624700000310
represents the information entropy loss of predicting the health status of the target rolling bearing vibration signal sample set to the source rolling bearing vibration signal sample set, j represents the jth health status in the source rolling bearing vibration signal sample set, I(·) represents the indicator function,
Figure FDA0004018962470000041
Indicates that the i-th sample in the s k -th source rolling bearing vibration signal sample set belongs to the diagnosis result of the health state of the source rolling bearing vibration signal sample set; 3.3.2)计算域共享深度残差网络预测目标滚动轴承振动信号样本集属于未知健康状态的交叉熵损失：3.3.2) The cross entropy loss of the computational domain shared deep residual network predicts the target rolling bearing vibration signal sample set belonging to the unknown health state:
Figure FDA0004018962470000042
Figure FDA0004018962470000042
 其中，
Figure FDA0004018962470000043
表示预测目标滚动轴承振动信号样本集属于未知健康状态的交叉熵损失；in,
Figure FDA0004018962470000043
The cross entropy loss represents the prediction that the target rolling bearing vibration signal sample set belongs to an unknown health state; 3.3.3)最小化上式交叉熵损失，更新域共享深度残差网络参数
Figure FDA0004018962470000044
即：3.3.3) Minimize the cross entropy loss above and update the domain shared deep residual network parameters
Figure FDA0004018962470000044
Right now:
Figure FDA0004018962470000045
Figure FDA0004018962470000045
 3.4)构建参数共享的领域混淆网络并多次迭代更新网络参数：3.4) Construct a parameter-sharing domain confusion network and iterate and update the network parameters multiple times: 构建参数共享的领域混淆网络，领域混淆网络的待训练参数为θadv，该网络的输入为深度迁移故障特征
Figure FDA0004018962470000046
输出为领域混淆特征
Figure FDA0004018962470000047
其中，
Figure FDA0004018962470000048
为第sk个源滚动轴承振动信号样本集第i个样本的领域混淆特征，
Figure FDA0004018962470000049
为目标滚动轴承振动信号样本集第i个样本的领域混淆特征，上/下标adv代表领域混淆网络；最大化如下目标函数更新领域混淆网络的参数θadv，即：Construct a parameter-sharing domain confusion network. The parameter to be trained of the domain confusion network is θ adv . The input of the network is the deep migration fault feature.
Figure FDA0004018962470000046
The output is the domain confusion feature
Figure FDA0004018962470000047
in,
Figure FDA0004018962470000048
is the domain confusion feature of the ith sample in the s k th source rolling bearing vibration signal sample set,
Figure FDA0004018962470000049
is the domain confusion feature of the i-th sample in the target rolling bearing vibration signal sample set, and the superscript/subscript adv represents the domain confusion network; the parameter θ adv of the domain confusion network is updated by maximizing the following objective function, namely:
Figure FDA00040189624700000410
Figure FDA00040189624700000410
 迭代更新nadv次领域混淆网络参数θadv，每次迭代更新后，将领域混淆网络的待训练参数θadv截断在范围{-ξ,ξ}内；Iteratively update n adv domain confusion network parameters θ adv . After each iterative update, the parameters θ adv to be trained of the domain confusion network are truncated within the range {-ξ,ξ}. 3.5)计算源滚动轴承振动信号样本集和目标滚动轴承振动信号样本集的深度迁移故障特征的局部分布差异：3.5) Calculate the local distribution difference of the deep migration fault characteristics of the source rolling bearing vibration signal sample set and the target rolling bearing vibration signal sample set: 3.5.1)计算目标滚动轴承振动信号样本集平衡导向的加权系数：3.5.1) Calculate the weighted coefficient of the target rolling bearing vibration signal sample set for balance guidance:
Figure FDA0004018962470000051
Figure FDA0004018962470000051
 其中，
Figure FDA0004018962470000052
表示目标滚动轴承振动信号样本集第i个样本的加权系数，
Figure FDA0004018962470000053
为目标滚动轴承振动信号样本集第i个样本的领域混淆特征,
Figure FDA0004018962470000054
为目标滚动轴承振动信号样本集中第i个样本属于未知健康状态概率分布，上/下标F3代表域共享深度残差网络的输出层F3层,σs(·)表示激活函数；in,
Figure FDA0004018962470000052
represents the weighting coefficient of the i-th sample of the target rolling bearing vibration signal sample set,
Figure FDA0004018962470000053
is the domain confusion feature of the i-th sample in the target rolling bearing vibration signal sample set,
Figure FDA0004018962470000054
is the probability distribution of the i-th sample in the target rolling bearing vibration signal sample set belonging to the unknown health state, the superscript/subscript F 3 represents the output layer F 3 layer of the domain-shared deep residual network, and σ s (·) represents the activation function; 3.5.2)对目标滚动轴承振动信号样本集深度迁移故障特征加权，计算局部分布的最大均值差异：3.5.2) Weight the deep migration fault features of the target rolling bearing vibration signal sample set and calculate the maximum mean difference of the local distribution:
Figure FDA0004018962470000055
Figure FDA0004018962470000055
 其中，
Figure FDA0004018962470000056
表示多项式核最大均值差异函数；in,
Figure FDA0004018962470000056
represents the polynomial kernel maximum mean difference function; 3.6)多滚动轴承融合迁移诊断模型训练：3.6) Training of multi-roller bearing fusion migration diagnosis model: 3.6.1)对基于不同源滚动轴承振动信号样本集健康状态知识学习到的目标滚动轴承深度迁移故障特征进行特征分布适配，计算其最大均值差异为：3.6.1) Perform feature distribution adaptation on the target rolling bearing deep migration fault feature learned based on the health status knowledge of the rolling bearing vibration signal sample set from different sources, and calculate the maximum mean difference as:
Figure FDA0004018962470000057
Figure FDA0004018962470000057
 其中，S＝{s1,s2,...,sN}表示N个源滚动轴承振动信号样本集，
Figure FDA0004018962470000058
分别表示为基于第si、sj个源滚动轴承振动信号样本集知识提取的目标滚动轴承振动信号样本集的深度迁移故障特征，上/下标F2代表域共享深度残差网络的F2层；Where, S = {s 1 ,s 2 ,...,s N } represents N source rolling bearing vibration signal sample sets,
Figure FDA0004018962470000058
They are respectively represented as the deep migration fault features of the target rolling bearing vibration signal sample set extracted based on the knowledge of the s i th and s j th source rolling bearing vibration signal sample sets, and the superscript/subscript F 2 represents the F 2 layer of the domain-shared deep residual network; 3.6.2)最小化如下目标函数同时更新N个域共享深度残差网络的参数集
Figure FDA0004018962470000061
即：3.6.2) Minimize the following objective function and update the parameter set of N domains shared deep residual network at the same time
Figure FDA0004018962470000061
Right now:
Figure FDA0004018962470000062
Figure FDA0004018962470000062
 其中，λ,β,γ是权衡参数。Among them, λ, β, γ are trade-off parameters.
CN202110449135.6A 2021-04-25 2021-04-25 Migration intelligent diagnosis method for multi-source rolling bearing health state fusion Active CN113191245B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110449135.6A CN113191245B (en) 2021-04-25 2021-04-25 Migration intelligent diagnosis method for multi-source rolling bearing health state fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110449135.6A CN113191245B (en) 2021-04-25 2021-04-25 Migration intelligent diagnosis method for multi-source rolling bearing health state fusion

Publications (2)

Publication Number Publication Date
CN113191245A CN113191245A (en) 2021-07-30
CN113191245B true CN113191245B (en) 2023-04-11

Family

ID=76978831

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110449135.6A Active CN113191245B (en) 2021-04-25 2021-04-25 Migration intelligent diagnosis method for multi-source rolling bearing health state fusion

Country Status (1)

Country Link
CN (1) CN113191245B (en)

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105758644A (en) * 2016-05-16 2016-07-13 上海电力学院 Rolling bearing fault diagnosis method based on variation mode decomposition and permutation entropy
CN106596107B (en) * 2016-12-27 2019-03-26 华南理工大学 A kind of flexibility precise thin-wall bearing failure diagnosis life test machine
CN110285969B (en) * 2019-07-10 2020-05-26 西安交通大学 A fault migration diagnosis method for rolling bearings based on feature distribution adaptation of polynomial kernel implantation
CN111337256B (en) * 2020-03-27 2020-12-29 西安交通大学 Diagnosis method for local migration of rolling bearing fault depth based on domain asymmetry factor
CN112308147B (en) * 2020-11-02 2024-02-09 西安电子科技大学 Rotary machinery fault diagnosis method based on multi-source domain anchor adapter integrated migration
AU2020103681A4 (en) * 2020-11-26 2021-02-04 Anhui University Of Technology Rolling Bearing Fault Diagnosis Method Based on Fourier Decomposition and Multi-scale Arrangement Entropy Partial Mean Value

Also Published As

Publication number Publication date
CN113191245A (en) 2021-07-30

Similar Documents

Publication Publication Date Title
CN110866314B (en) Rotating Machinery Remaining Lifetime Prediction Method Based on Multilayer Bidirectionally Gated Recurrent Unit Networks
CN109522600B (en) Complex equipment residual service life prediction method based on combined deep neural network
CN111337256B (en) Diagnosis method for local migration of rolling bearing fault depth based on domain asymmetry factor
WO2021042935A1 (en) Bearing service life prediction method based on hidden markov model and transfer learning
CN101915234B (en) Method for diagnosing compressor-associated failure based on Bayesian network
CN110376522B (en) A data fusion deep learning network based motor fault diagnosis method
CN109376620A (en) A migration diagnosis method for wind turbine gearbox faults
CN110046409B (en) ResNet-based steam turbine component health state evaluation method
CN103147972A (en) Reciprocating-type compressor fault diagnosis method based on multi-sensor information fusion
CN106769032B (en) Method for predicting service life of slewing bearing
CN114564987B (en) Rotary machine fault diagnosis method and system based on graph data
CN114742122A (en) Equipment fault diagnosis method and device, electronic equipment and storage medium
CN115809595A (en) Digital twin model construction method reflecting rolling bearing defect expansion
CN116108346A (en) A Lifelong Learning Method for Incremental Bearing Fault Diagnosis Based on Generative Feature Replay
CN118934705A (en) A method for predicting the fault evolution trend of main helium blower combined with fault mechanism analysis
CN117171907A (en) Rolling bearing residual life prediction method and system
CN114997046B (en) A Dynamics Simulation-Guided Domain-Adversarial Bearing Fault Diagnosis Method
CN117332367A (en) Small sample rotary machine intelligent diagnosis method based on mechanism data fusion
CN113191245B (en) Migration intelligent diagnosis method for multi-source rolling bearing health state fusion
CN107122609A (en) A kind of electromechanical product quality evaluation method based on mass property gene theory
CN117574078A (en) A hierarchical diagnosis method for mechanical seal failure modes based on vibration analysis
Parziale et al. Anomaly characterization for the condition monitoring of rotating shafts exploiting data fusion and explainable convolutional neural networks
CN113469066B (en) Unbalanced sample multitask self-optimization rolling bearing fault migration diagnosis method
Elkihel et al. Artificial Intelligence Based on the Neurons Networks at the Service Predictive Bearing
CN114838932A (en) A slow time-varying weak fault diagnosis method for RV reducer

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant