Electric excavator bearing fault diagnosis method and device

By constructing a new adversarial domain adaptive meta-learning network model, using the bearing vibration signals obtained by the accelerometer for standardized preprocessing and feature extraction, the problems of low diagnostic accuracy and poor model generalization in the bearing fault diagnosis of electric excavator are solved. High accuracy and fast cross-domain fault diagnosis are achieved.

CN118913695APending Publication Date: 2024-11-08GUANGXI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410949369.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing deep learning methods have problems with low diagnostic accuracy and poor generalization of the model in the fault diagnosis of electric excavator bearings, especially in the conditions of scarce data and complex working conditions, which are difficult to meet actual needs.

Method used

Accelerometers are used to obtain bearing vibration signals, and the data is divided into source domain and target domain through standardized preprocessing. A new adversarial domain adaptive meta-learning network model is built, and a support set and query set are used for training, building a feature extraction network and metric classification. and domain identifiers to realize cross-domain fault diagnosis.

Benefits of technology

It significantly improves the cross-domain fault diagnosis performance of the model, has strong ability to extract feature information, fast diagnosis speed and high accuracy, and can realize intelligent electric excavator bearing fault diagnosis, solving the low diagnostic accuracy and model caused by lag in the existing technology The problem of poor generalization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118913695A_ABST
    Figure CN118913695A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of engineering machinery bearing diagnosis, in particular to an electric excavator bearing fault diagnosis method and device, and the method comprises the steps: extracting the features of a source domain data set and a target domain data set, obtaining a query set and a support set, and constructing a novel adversarial domain adaptive meta-learning network; training and optimizing the novel adversarial domain adaptive meta-learning network by using the support set, and then inputting the query set into the trained novel adversarial domain adaptive meta-learning network to obtain a diagnosis model; and carrying out classified diagnosis on the electric excavator bearing fault by using the diagnosis model. According to the method, the cross-domain fault diagnosis performance of the model can be effectively improved, the feature information extraction capacity is high, the structure is simple, the diagnosis speed is high, the accuracy rate is high, intelligent electric excavator bearing fault diagnosis can be achieved, and therefore the problems that due to lag of an existing detection technology, the diagnosis accuracy rate is low, and model generalization is poor are solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of engineering machinery bearing diagnosis, and in particular to a method and device for diagnosing bearing faults of an electric excavator. Background Art

[0002] Electric excavator motor bearings have a critical load and support function in the operation of the excavator, and they function in a variety of environments and operating conditions. Bearings on electric excavators must be able to withstand huge loads and vibrations, and exhibit excellent strength and fatigue resistance. In electric construction machinery, reliable operation under complex conditions is a must for bearings, such as frequent starts and stops, curved slopes, etc. Considering the huge weight of the cargo involved, the bearings need to work effectively under high load and severe weather conditions. Because electric excavator bearings need to work in harsh environments, they are prone to failure. If a failure occurs, it will affect the normal operation of the excavator. Therefore, it is crucial to perform bearing fault diagnosis during real-time monitoring of electric excavator operation to ensure the normal and safe operation of the excavator.

[0003] Deep learning fault diagnosis methods have been applied to the field of excavator bearing fault diagnosis and have achieved remarkable results. However, deep learning methods require a large amount of labeled data during training and are prone to overfitting problems. Since electric excavator motor bearing fault samples are scarce and their operating conditions are complex and changeable, there are obvious differences in regional distribution, resulting in low diagnostic accuracy and poor generalization performance of the model. Therefore, the current fault diagnosis methods based on deep learning can no longer meet the needs of electric excavator bearing fault diagnosis. Summary of the invention

[0004] The purpose of the present invention is to provide a method and device for diagnosing bearing faults of an electric excavator, aiming to solve the problems of low diagnostic accuracy and poor model generalization caused by the lag of existing detection technology.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for diagnosing bearing faults of an electric excavator, comprising the following steps:

[0006] Using accelerometer to obtain vibration signal of electric excavator motor bearing;

[0007] Performing standardized preprocessing on the bearing vibration signal, and dividing the data into a source domain data set and a target domain data set;

[0008] Extracting features of the source domain dataset and the target domain dataset to obtain a query set and a support set;

[0009] Construct a new adversarial domain adaptation meta-learning network model;

[0010] Using the support set to train and optimize the novel adversarial domain adaptation meta-learning network, and then inputting the query set into the trained novel adversarial domain adaptation meta-learning network to obtain a diagnostic model;

[0011] The diagnostic model is used to classify and diagnose the faults of excavator bearings.

[0012] Among them, in “performing standardized preprocessing on the bearing vibration signal, dividing the data into a source domain data set and a target domain data set, and including a support set and a query set”, it includes:

[0013] Using a normalization function to normalize the bearing vibration signal data to obtain normalized data;

[0014] Number the corresponding bearing fault types;

[0015] The normalized data is divided into a source domain dataset and a target domain dataset using an N-way K-shot meta-learning training strategy.

[0016] Among them, in "Constructing a new adversarial domain adaptive meta-learning network", including

[0017] Build feature extraction networks, metric classifiers, and domain identifiers;

[0018] A novel adversarial domain adaptation meta-learning network is constructed based on the feature extraction network, the metric classifier and the domain identifier.

[0019] Among them, in "Building feature extraction network, metric classifier and domain identifier",

[0020] The feature extraction network consists of 6 convolution blocks and 1 lightweight and efficient channel attention module, wherein each convolution block consists of a one-dimensional convolution layer, a batch normalization layer, an activation function layer and a pooling layer;

[0021] The domain discriminator consists of a batch normalization layer, a ReLU activation function layer and a linear layer.

[0022] In a second aspect, a bearing fault diagnosis device for an electric excavator is provided, which is used in the bearing fault diagnosis method for an electric excavator according to the first aspect.

[0023] It includes a bearing vibration signal acquisition module, a bearing signal preprocessing module, a bearing signal feature extraction module and a bearing fault diagnosis module, wherein the bearing vibration signal acquisition module, the bearing signal preprocessing module, the bearing signal feature extraction module and the bearing fault diagnosis module are electrically connected;

[0024] The bearing vibration signal acquisition module is used to acquire the bearing vibration signal;

[0025] The bearing signal preprocessing module is used to perform normalization preprocessing on the bearing vibration signal, and divide the processed data into source domain data and target domain data, including a support set and a query set;

[0026] The bearing signal feature extraction module is used to input the vibration data of the electric excavator motor bearing into the feature extraction network to extract features, and output the extracted features to the high-dimensional quantity space, and calculate the distance between the predicted sample and the initial prototype through the distance function;

[0027] The bearing fault diagnosis module is used to perform fault diagnosis on the motor bearing data of the electric excavator to be tested by using a novel adversarial domain adaptive meta-learning network.

[0028] The present invention provides a method for diagnosing bearing faults of electric excavators, comprising the following steps: using an accelerometer to obtain a vibration signal of an electric excavator motor bearing; performing standardized preprocessing on the bearing vibration signal, and dividing the data into a source domain data set and a target domain data set; extracting features of the source domain data set and the target domain data set to obtain a query set and a support set; constructing a new adversarial domain adaptive meta-learning network; using the support set to train and optimize the new adversarial domain adaptive meta-learning network, and then inputting the query set into the trained new adversarial domain adaptive meta-learning network to obtain a diagnostic model; and using the diagnostic model to classify and diagnose excavator bearing faults. The method of the present invention can effectively improve the cross-domain fault diagnosis performance of the model, has a strong ability to extract feature information, has a simple structure, a fast diagnostic speed and a high accuracy, and can realize intelligent electric excavator bearing fault diagnosis, thereby solving the problem of low diagnostic accuracy and poor model generalization caused by the lag of existing detection technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 The present invention provides a flow chart of a method for diagnosing bearing faults of an electric excavator.

[0031] Figure 2 This is a schematic diagram of the LECA module structure.

[0032] Figure 3 This is a schematic diagram of the NADMN model structure.

[0033] Figure 4The present invention provides a structural block diagram of a bearing fault diagnosis device for an electric excavator.

[0034] 1-bearing vibration signal acquisition module, 2-bearing signal preprocessing module, 3-bearing signal feature extraction module, 4-bearing fault diagnosis module. DETAILED DESCRIPTION

[0035] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0036] See also Figures 1 to 3 In a first aspect, the present invention provides a method for diagnosing bearing faults of an electric excavator, comprising the following steps:

[0037] S1 uses an accelerometer to obtain the vibration signal of the motor bearing of the electric excavator;

[0038] Specifically, the vibration signals of eight bearing states with different fault sizes and speeds were collected through the electric excavator motor bearing test bench.

[0039] S2 performs standardized preprocessing on the bearing vibration signal to divide the data into a source domain data set and a target domain data set;

[0040] S21 uses a normalization function to normalize the bearing vibration signal data to obtain normalized data;

[0041] Specifically, a normalization function is used to normalize the collected bearing vibration signal data so that its value is distributed between 0 and 1. This can significantly reduce the amount of model calculation and improve the model's computing performance. The normalization function is shown in the expression:

[0042]

[0043] Among them, x represents the input bearing signal data, and σ(x) represents the softmax function.

[0044] S22 numbers the corresponding bearing fault types;

[0045] Specifically, the corresponding bearing fault types are numbered, such as the eight bearing types {normal state, inner ring fault, outer ring fault, rolling element fault, inner ring and outer ring fault, inner ring and rolling element fault, outer ring and rolling element fault, inner ring, outer ring and rolling element fault} are numbered {0, 1, 2, 3, 4, 5, 6, 7} respectively.

[0046] S23 uses an N-way K-shot meta-learning training strategy to divide the normalized data into a source domain dataset and a target domain dataset.

[0047] Specifically, in the "N-way K-shot" training strategy, N represents the number of categories involved in each diagnostic task, and K represents the number of samples in each category. After normalization, the data is divided into source domain data and target domain data.

[0048] S3 extracts features of the source domain dataset and the target domain dataset to obtain a query set and a support set;

[0049] Specifically, a feature extraction network is built, which consists of 6 convolution blocks and 1 lightweight and efficient channel attention (LECA) module, where each convolution block consists of 4 layers, namely: one-dimensional convolution layer, batch normalization layer, activation function layer and pooling layer. The vibration data of the electric excavator motor bearing is marked. The vibration data contains 8 types, each type contains 50 samples, and 20 samples are taken from each type as the support set, and the remaining 30 samples are taken as the query set.

[0050] S4 builds a new adversarial domain adaptation meta-learning network model;

[0051] The novel adversarial domain-adaptation meta-learning network model (NADMN)

[0052] The feature extraction network consists of 6 convolution blocks and 1 lightweight and efficient channel attention module, wherein each convolution block consists of a one-dimensional convolution layer, a batch normalization layer, an activation function layer and a pooling layer;

[0053] The domain discriminator consists of a batch normalization layer, a ReLU activation function layer and a linear layer.

[0054] S41 builds feature extraction network, metric classifier and domain identifier;

[0055] Specifically, the one-dimensional convolution layer, batch normalization layer, and Relu activation function expressions are as follows:

[0056]

[0057] Among them, ω represents the weight matrix of the convolution kernel, x i represents the i-th sample of the input, b i represents the offset, and * represents the convolution operation.

[0058]

[0059] Among them, x i represents the i-th sample of the input, the initial values ​​of the learnable parameters γ and β are set to 1 and 0 respectively, the hyperparameter ε=1×10-5, μ(x) and σ(x) represent the mean and variance respectively.

[0060] Activation function f relu It can be defined as:

[0061] f relu (x)=max(0,x)

[0062] The LECA module consists of five layers, namely, the maximum pooling layer, the one-dimensional convolution layer, the batch normalization layer, the Relu activation function layer, and the Sigmoid activation function layer. The LECA module structure is as follows: Figure 2 shown.

[0063] The metric classifier classifies the target samples by calculating the similarity between the target domain support samples and the query samples in the feature space, as follows:

[0064] The category prototype of each sample in the support set is calculated as follows:

[0065]

[0066] Among them, p c is the prototype of category c in the support set, N and K represent the number of categories in the support set and the number of samples in each category, respectively, g θ is a feature extractor with parameters θ, x i represents the i-th sample of the input;

[0067] Calculate the Euclidean distance between each sample in the query set and the prototype, and calculate the predicted probability of the query sample in the query set. The expression is as follows:

[0068]

[0069] Among them, F d is the Euclidean distance function from a to b, N is the number of categories in each episode; x j ,y j Represent the samples and labels in the query set respectively.

[0070] S42 constructs a novel adversarial domain adaptation meta-learning network based on the feature extraction network, the metric classifier and the domain identifier.

[0071] Specifically, a complete NADMN model is built, and the NADMN model structure is shown in the figure.

[0072] S5: using the support set to train and optimize the novel adversarial domain adaptation meta-learning network, and then inputting the query set into the trained novel adversarial domain adaptation meta-learning network to obtain a diagnostic model;

[0073] Specifically, the preprocessed data is input into the NADMN model for training and testing. First, the parameters of the NADMN model are updated by training the support set. At the same time, the model parameters are adjusted according to the iteration of the model loss value and accuracy on the support set. Once the target diagnostic accuracy is achieved, the query set can be used to evaluate the NADMN model to evaluate its recognition ability on different bearing fault types. If the model can effectively realize the fault diagnosis of electric excavator bearings, it can be applied to actual scenarios.

[0074] The NADMN learning strategy consists of three parts:

[0075] (1) Optimize the metric-based classifier to minimize the classification error L of the classifier c ,in order to better identify the features extracted from the source domain;

[0076] (2) Optimize the domain discriminator to minimize the discrimination error L between source domain data and target domain data d , thereby improving the domain discriminator's ability to distinguish between source domain data and target domain data.

[0077] (3) Optimize the feature extraction network to maximize the discrimination error L between the source domain data and the target domain data d , thus enabling the feature extraction network to generate more domain-invariant features and further deceive the domain discriminator.

[0078] Metric classifier classification error L c and domain discriminator error L d As shown in the following expression:

[0079]

[0080] Among them, S and Q represent the support set and query set respectively, X S ∈S∪Q,n q 、n s 、n t Respectively represent the total number of query set tasks, the total number of source domain tasks, and the total number of target domain tasks. In the process of training and optimizing the parameters of the NADMN model, a joint loss function is used to measure the gap between the predicted value and the actual value of the NADMN model. The joint loss function is shown in the following expression:

[0081] L=L c +λL d

[0082] In the formula, λ∈[0,1], the smaller the loss value L is, the better the performance of the method is.

[0083] If the fault diagnosis accuracy of the constructed NADMN model exceeds the target accuracy and the loss value is low, it can be inferred that the model has excellent fault diagnosis capabilities.

[0084] S6 uses the diagnosis model to classify and diagnose the faults of the excavator bearing.

[0085] Specifically, the faults of the excavator bearings are classified and diagnosed by using the diagnostic model in the aforementioned steps.

[0086] See also Figure 4 In the second aspect, a bearing fault diagnosis device for an electric excavator is used in the bearing fault diagnosis method for an electric excavator according to the first aspect.

[0087] It includes a bearing vibration signal acquisition module 1, a bearing signal preprocessing module 2, a bearing signal feature extraction module 3 and a bearing fault diagnosis module 4, wherein the bearing vibration signal acquisition module 1, the bearing signal preprocessing module 2, the bearing signal feature extraction module 3 and the bearing fault diagnosis module 4 are electrically connected;

[0088] The bearing vibration signal acquisition module 1 is used to acquire the bearing vibration signal;

[0089] The bearing signal preprocessing module 2 is used to perform normalization preprocessing on the bearing vibration signal, and divide the processed data into source domain data and target domain data, including a support set and a query set;

[0090] The bearing signal feature extraction module 3 is used to input the vibration data of the electric excavator motor bearing into the feature extraction network to extract features, and output the extracted features to the high-dimensional quantity space, and calculate the distance between the predicted sample and the initial prototype through the distance function;

[0091] The bearing fault diagnosis module 4 is used to perform fault diagnosis on the motor bearing data of the electric excavator to be tested using a novel adversarial domain adaptive meta-learning network.

[0092] In this embodiment, the bearing vibration signal acquisition module 1 is used to collect bearing vibration signals; the bearing signal preprocessing module 2 is used to perform normalization preprocessing on the bearing vibration signals and divide the processed data into source domain data and target domain data, including a support set and a query set; the bearing signal feature extraction module 3 is used to input the electric excavator motor bearing vibration data into a feature extraction network to extract features and output the extracted features to a high-dimensional quantity space to calculate the distance between the predicted sample and the initial prototype through a distance function; the bearing fault diagnosis module 4 is used to use a new adversarial domain adaptive meta-learning network to perform fault diagnosis on the electric excavator motor bearing data to be tested.

[0093] In order to verify the effectiveness of the electric excavator bearing fault diagnosis method based on meta-learning algorithm proposed in this paper, this method is compared with traditional deep learning methods and other meta-learning methods, including fine-tuned convolutional neural network (CNN-FT), relational network (RelaNet), prototype network (ProtoNet) and semi-supervised meta-learning network (SSMN). Through a set of examples containing 7 types of electric excavator motor bearing fault data under different working conditions, the experimental results of each method under this experiment are shown in Table 1.

[0094] Table 1. Experimental results of each method

[0095]

[0096] According to the data in Table 1, the NADMN method is superior to other comparison methods in terms of test accuracy, training loss value and test time. Specifically, the average accuracy of the NADMN method reached 98.17%, which is 23.24% higher than that of CNN-FT, 12.72% higher than that of RelaNet, 10.43% higher than that of ProtoNet, and 8.51% higher than that of SSMN. In addition, the NADMN method not only has the lowest training loss value, but also can complete the test in the shortest time. Experiments have shown that the electric excavator bearing fault diagnosis method and device based on the meta-learning algorithm designed by the present invention can obtain bearing vibration signal data in real time, and effectively perform state detection and fault diagnosis.

[0097] The present invention has the following beneficial technical effects:

[0098] (1) The present invention can comprehensively and accurately perform cross-domain fault diagnosis of electric excavator motor bearings through a meta-learning method, thereby improving the diagnostic accuracy and fault diagnosis speed;

[0099] (2) The present invention uses a normalization function to pre-process the vibration data of the electric excavator motor bearing, thereby significantly reducing the amount of model calculation and improving the model operation performance;

[0100] (3) The present invention proposes a LECA module and adds it to the feature extraction network, which improves the model feature extraction capability and can obtain more comprehensive feature information;

[0101] (4) In the process of building NADMN, the present invention incorporates the convolution module and the coordinate attention module. The convolution module effectively reduces the computational burden, making the entire model more lightweight and efficient; while the coordinate attention module enhances the feature extraction capability, thereby significantly improving the fault diagnosis accuracy;

[0102] (5) The electric excavator motor bearing fault diagnosis device designed in the present invention has the characteristics of good cross-domain fault diagnosis performance, fast diagnosis speed, high accuracy, etc., which can meet the practical application needs of electric excavator motor bearing fault diagnosis, and also has great application potential in other application fields.

[0103] What is disclosed above is only a preferred embodiment of the electric excavator bearing fault diagnosis method and device of the present invention. Of course, this cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention are still within the scope of the invention.

Claims

1. A method for diagnosing bearing faults of an electric excavator, characterized in that: The following steps are involved: Using accelerometer to obtain vibration signal of electric excavator motor bearing; Performing standardized preprocessing on the bearing vibration signal, and dividing the data into a source domain data set and a target domain data set; Extracting features of the source domain dataset and the target domain dataset to obtain a query set and a support set; Construct a new adversarial domain adaptation meta-learning network model; Using the support set to train and optimize the novel adversarial domain adaptation meta-learning network, and then inputting the query set into the trained novel adversarial domain adaptation meta-learning network to obtain a diagnostic model; The diagnostic model is used to classify and diagnose the faults of excavator bearings.

2. The electric excavator bearing fault diagnosis method according to claim 1, characterized in that: In "performing standard preprocessing on the bearing vibration signal, dividing the data into a source domain data set and a target domain data set, and including a support set and a query set", it includes: Using a normalization function to normalize the bearing vibration signal data to obtain normalized data; Number the corresponding bearing fault types; The normalized data is divided into a source domain dataset and a target domain dataset using an N-way K-shot meta-learning training strategy.

3. The electric excavator bearing fault diagnosis method according to claim 2, characterized in that: In "Constructing a Novel Adversarial Domain Adaptation Meta-Learning Network", including Build feature extraction networks, metric classifiers, and domain identifiers; A novel adversarial domain adaptation meta-learning network is constructed based on the feature extraction network, the metric classifier and the domain identifier.

4. The electric excavator bearing fault diagnosis method according to claim 3, characterized in that: In "Building Feature Extraction Networks, Metric Classifiers, and Domain Identifiers", The feature extraction network consists of 6 convolution blocks and 1 lightweight and efficient channel attention module, wherein each convolution block consists of a one-dimensional convolution layer, a batch normalization layer, an activation function layer and a pooling layer; The domain discriminator consists of a batch normalization layer, a ReLU activation function layer and a linear layer.

5. An electric excavator bearing fault diagnosis device, used in the electric excavator bearing fault diagnosis method according to any one of claims 1 to 4, characterized in that: It includes a bearing vibration signal acquisition module, a bearing signal preprocessing module, a bearing signal feature extraction module and a bearing fault diagnosis module, wherein the bearing vibration signal acquisition module, the bearing signal preprocessing module, the bearing signal feature extraction module and the bearing fault diagnosis module are electrically connected; The bearing vibration signal acquisition module is used to acquire the bearing vibration signal; The bearing signal preprocessing module is used to perform normalization preprocessing on the bearing vibration signal, and divide the processed data into source domain data and target domain data, including a support set and a query set; The bearing signal feature extraction module is used to input the vibration data of the electric excavator motor bearing into the feature extraction network to extract features, and output the extracted features to the high-dimensional quantity space, and calculate the distance between the predicted sample and the initial prototype through the distance function; The bearing fault diagnosis module is used to perform fault diagnosis on the motor bearing data of the electric excavator to be tested by using a novel adversarial domain adaptive meta-learning network.