Bridge crane transmission system bearing fault diagnosis method based on domain adaptive technology

Abstract

The invention relates to the technical field of crane fault diagnosis, in particular to a bridge crane transmission system bearing fault diagnosis method based on a domain self-adaption technology. Comprising the following steps: acquiring bearing data of the bridge crane running in a laboratory as source domain data; obtaining bearing data of the bridge crane running in an actual working condition as target domain data; performing data cleaning processing on the source domain data and the target domain data; training a domain shared feature extraction network by using the cleaned source domain data and target domain data; and the trained domain sharing feature extraction network is used to diagnose the fault type of the bridge crane transmission system bearing. According to the method, the problems that in the real environment, the bridge crane has few fault data, the collected data is lack of labels, and the fault diagnosis precision is influenced by a large amount of noise can be solved, the data collection cost can be remarkably reduced, and the operation safety and reliability of the bridge crane are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of crane fault diagnosis, and particularly relates to a bearing fault diagnosis method for a bridge crane drive system based on domain adaptation technology. Background Art

[0002] Bridge cranes are widely used in factory workshops, construction sites, port terminals and other places, undertaking tasks such as handling heavy materials or finished products, installing or hoisting large structural components. Among them, the drive system, as one of the core components of the bridge crane, is responsible for transmitting power from the motor to each working mechanism to achieve the hoisting of materials. Once the drive system fails, it will affect production efficiency and quality to a lesser extent, and in severe cases, it will cause further damage to the equipment or even casualties. At present, the maintenance of bridge cranes still adopts the method of regular inspection. However, for the components of the drive system, taking bearings as an example, even for bearings of the same model produced by the same manufacturer using the same material, there are significant differences in the remaining service life under different working conditions, different operating habits and long-term operation. If the drive system of a bridge crane with a remaining service life that has not expired is maintained, it will consume a large amount of manpower and material resources. At the same time, because most faults are not easily detected in the early stage, misjudgment and missed judgment are likely to occur, further affecting production safety. Secondly, if the drive system has reached the remaining service life and is not maintained, more serious consequences will occur. Therefore, it is necessary to conduct accurate intelligent fault diagnosis for the drive system of bridge cranes, especially bearings, to improve the reliability and safety of the equipment, while reducing labor costs and achieving the effect of cost reduction and efficiency increase.

[0003] Traditional fault diagnosis of bridge crane bearings mainly relies on signal processing methods. These methods require engineers to have extremely high theoretical knowledge and mathematical analysis capabilities. However, their processing efficiency and diagnostic accuracy cannot meet the requirements of customers for the operating reliability. With the development of deep learning, its advantage of being able to give accurate diagnostic results only relying on data, without the need for too much prior knowledge and being able to process a large amount of data, has gradually replaced the traditional fault diagnosis methods for bridge crane bearings. However, the fault diagnosis technology for bridge crane bearings based on deep learning has not been widely popularized, and there are two important reasons: 1. The class imbalance problem, that is, during the service process of the bridge crane, most of the collected data is normal data, lacking fault sample data. This makes the existing models unable to learn the boundaries of fault classification, resulting in overfitting and affecting the generalization performance of intelligent fault diagnosis. 2. The data drift problem, that is, during the service process of the bridge crane, due to different goods, there are large differences in the power, speed, etc. output by its transmission system, so there are also large differences in its operating signals. If only relying on the operating signals collected under a certain speed or load condition to train the neural network, the effect is often poor when performing fault diagnosis under other speeds or load conditions. And collecting operating signals separately for each task and classifying the signals through manual annotation and then retraining the neural network will undoubtedly increase a large amount of labor costs.

[0004] Based on this, it is still necessary to find new algorithms for the bearing health status of bridge cranes to solve the impacts brought by the sample class imbalance problem and the data drift problem. However, the existing technologies have the following defects: 1. The existing technologies cannot effectively process large and complex data, resulting in low model diagnosis efficiency and success rate. 2. A large amount of prior knowledge is required, which is difficult for non-professionals in this industry to understand, restricting the development of data science and failing to give full play to the advantages of the big data era. 3. In the existing technologies, the problems of insufficient samples, class imbalance, and data drift have not been effectively solved, affecting the generalization ability and accuracy of the model. 4. The existing technologies rely on a large number of labeled high-quality data sets, and the cost of data collection is too high. The collected data usually does not contain a large number of labels, resulting in low diagnostic accuracy of the existing methods. 5. Traditional transfer learning methods only rely on the way of feature difference measurement to extract domain-invariant features, ignoring the extraction of discriminative features, and the effective features that can be extracted by the existing feature difference measurement methods are less, often unable to cover all the features of the operating signals of bridge crane bearings. Summary of the Invention

[0005] The present invention discloses a fault diagnosis method for the bearings of the drive system of a bridge crane based on domain adaptation technology. The specific method is as follows:

[0006] Obtain the bearing data of the bridge crane operating in the laboratory as the source domain data;

[0007] Obtain the bearing data of the bridge crane during actual operation as the target domain data;

[0008] Perform data cleaning on the source domain data and the target domain data;

[0009] Train a domain - shared feature extraction network with the cleaned source domain data and target domain data;

[0010] Use the trained domain - shared feature extraction network to diagnose the bearing fault categories of the bridge crane drive system.

[0011] Furthermore, the domain - shared feature extraction network includes a feature extraction model, a domain discriminator, and a diagnostic model;

[0012] The feature extraction model extracts the cleaned source domain data and target domain data;

[0013] The domain discriminator discriminates the differences between the extracted source domain data and target domain data;

[0014] The diagnostic model completes model training with the extracted source domain data and target domain data, and the trained diagnostic model diagnoses the bearing fault categories of the bridge crane drive system according to the target domain data.

[0015] Furthermore, the training methods of the feature extraction model, the domain discriminator, and the diagnostic model are as follows:

[0016] Construct the total loss function of the feature extraction model, the domain discriminator, and the diagnostic model;

[0017] According to the total loss function, through backpropagation, adjust the neuron parameters of the feature extraction model, the domain discriminator, and the diagnostic model;

[0018] Until the consistency of the source domain data features and target data features extracted by the feature extraction model, the difficulty of the discriminator in distinguishing source domain data features and target data features, and the ease of recognition between various fault diagnosis regions all reach the preset goals.

[0019] Furthermore, the consistency of the source domain data features and target data features extracted by the feature extraction model is quantified by the source - domain target - domain feature extraction consistency evaluation function, and the specific formula is as follows:

[0020]

[0021] Among them, n and m are the number of samples in the source domain and the target domain respectively, and K(·,·) represents the kernel function obtained by weighted summation of Gaussian kernels with different bandwidths through the method of dynamically calculating the kernel bandwidth;

[0022] They respectively represent the sum of the kernel function values of all sample pairs within the source domain, within the target domain, and between the source domain and the target domain.

[0023] Furthermore, the difficulty for the discriminator to distinguish the source domain data features and the target data features is quantified by the domain discriminator evaluation function, and the specific formula is as follows:

[0024]

[0025] where g i is the label of the source domain, the target domain label is set to 0, d(x i ) is the probability estimation value that the i-th sample output by the domain classifier belongs to the source domain, and m represents the number of samples.

[0026] Furthermore, the ease of recognition between various fault diagnosis regions is quantified by the sample health classification evaluation function, and the specific formula is as follows:

[0027] L csoft = L arc + L soft

[0028] where

[0029]

[0030]

[0031] In the formula, L csoft is the sample health classification evaluation function, L arc represents the magnitude of the loss value, N is the number of samples, yi is the true label of the i-th sample; using the method of pseudo-labels to replace the true labels, m is a hyperparameter used to increase the distance of the classification boundary, making different samples farther apart, is the cosine value after increasing the distance, increasing the decision boundary of the correct class, n is the category of the health state of the bridge crane bearing, cosθ j is the cosine similarity between the feature vector of each category j and the weight vector;

[0032] L softmax is the conventional cross-entropy loss, N is the number of samples, C is the type of the health state of the bridge crane bearing, y ij is the indicator variable that the i-th sample belongs to the j-th class, p ij is the probability that the model predicts the i-th sample belongs to the j-th class.

[0033] Furthermore, the specific formula of the total loss function is as follows:

[0034] L Total = L csoft + nL d + LMSD

[0035] That is

[0036]

[0037] Among them, L Total is the total loss function, L csoft is the sample health classification evaluation function, L d is the domain discriminator evaluation function, L MSD is the source domain and target domain feature extraction consistency evaluation function; n is a hyperparameter for determining the degree of domain adaptation, and θ f , θ c , θ d respectively represent the neuron parameters of the feature extraction model, the diagnosis model, and the domain discriminator.

[0038] Furthermore, through backpropagation, adjust the neuron parameters of the feature extraction model, the domain discriminator, and the diagnosis model. The specific formulas are as follows:

[0039]

[0040]

[0041]

[0042] Among them, ε is the learning rate.

[0043] Furthermore, the bearing fault categories of the bridge crane drive system include: inner ring fault, outer ring fault, and ball fault.

[0044] Due to the adoption of the above technical solutions, the present invention has the following beneficial effects:

[0045] 1. The proposed intelligent fault diagnosis method for bridge cranes based on domain adaptation technology can solve the problems of few fault data of bridge cranes in the real environment, lack of labels for the collected data, and a large amount of noise affecting the fault diagnosis accuracy. The method can significantly reduce the data collection cost and greatly improve the operation safety and reliability of bridge cranes.

[0046] 2. The SCAD algorithm based on the maximum mean square deviation difference loss and the improved Softmax loss is more robust than conventional domain adaptation methods based on KL divergence, maximum mean difference, deep correlation contrast, etc., and is more in line with the actual operation logic of bridge cranes, with high diagnosis accuracy.

[0047] Other advantages, objects, and features of the present invention will be set forth in part in the following description, and in part will be obvious to those skilled in the art upon examination of the following, or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained by the following description. Brief Description of the Drawings

[0048] The drawings of the present invention are described as follows.

[0049] Figure 1 It is a schematic diagram of the overall process.

[0050] Figure 2 It is a diagram showing the scale of bearing faults.

[0051] Figure 3 It is a schematic diagram of bearing vibration signals.

[0052] Figure 4 It is a comparison diagram of the classification boundaries of the original Softmax function and the C-softmax function. Detailed Description of the Preferred Embodiments

[0053] The present invention will be further described below with reference to the drawings and embodiments.

[0054] A method for diagnosing bearing faults in the drive system of a bridge crane based on domain adaptation technology, as Figure 1 shown, the specific steps are as follows:

[0055] S1. Obtain the bearing data of the bridge crane operating in the laboratory as the source domain data.

[0056] S2. Obtain the bearing data of the bridge crane operating under actual working conditions as the target domain data.

[0057] In steps S1 and S2, the source domain dataset for training the subsequent model is derived from the sensor acquisition records of the laboratory bearing operation, and the appearance of bearing faults is as Figure 2 shown. These data are labeled standard data for characterizing bearing health. It includes inner ring faults IF1 - IF4, outer ring faults OF1 - OF4, rolling element faults RF1 - RF4, and the normal state. The specific data format is as follows:

[0058] IF1: Inner ring fault scale 1, 0.43 (mm)

[0059] IF2: Inner ring fault scale 2, 1.01 (mm)

[0060] IF3: Inner ring fault scale 3, 1.56 (mm)

[0061] IF4: Inner ring fault scale 4, 2.03 (mm)

[0062] OF1: Outer ring fault scale 1, 0.42 (mm)

[0063] OF2: Outer ring fault scale 2, 0.86 (mm)

[0064] OF3: Outer ring fault scale 3, 1.55 (mm)

[0065] OF4: Outer ring fault scale 4, 1.97 (mm)

[0066] RF1: Ball fault scale 1, 0.49 (mm)

[0067] RF2: Ball fault scale 2, 1.16 (mm)

[0068] RF3: Ball fault scale 3, 1.73 (mm)

[0069] RF4: Ball fault scale 4, 2.12 (mm)

[0070] S3. Perform data cleaning on the source domain data and the target domain data.

[0071] In step S3, the target domain data is automatically collected through the overhead crane dynamic equipment monitoring system. Since the dynamic equipment monitoring system may be affected by wind disturbance, abnormal operations of on-site operators, abnormal bumps in the working environment, etc. when collecting operation data, the sensing equipment may collect noise signals of different degrees. Or it may be affected by electromagnetic signal interference or data loss may occur during data transmission and storage, which affects the performance of the intelligent fault diagnosis model. Therefore, before inputting the data into the model, the bearing vibration signal is denoised and missing values are filled through data cleaning.

[0072] The specific method is to input the data into the data cleaning network. The network will adopt different data cleaning strategies according to the data volume. If the data volume meets the training requirements, the columns where obvious abnormal data or missing data are located will be deleted as a whole. If the data volume is small, adopting the deletion strategy will affect the network generalization performance. Therefore, the adjacent value mean interpolation method is used for data filling.

[0073] S4. Train the domain-shared feature extraction network with the cleaned source domain data and target domain data.

[0074] In step S4, the domain sharing feature extraction network includes a feature extraction model, a domain discriminator, and a diagnostic model; the feature extraction model extracts the cleaned source domain data and target domain data; the domain discriminator discriminates the differences between the extracted source domain data and target domain data; the diagnostic model uses the extracted source domain data and target domain data to complete model training, and the trained diagnostic model diagnoses the bearing fault categories of the bridge crane drive system according to the target domain data.

[0075] The training of the feature extraction model for extracting the common features of the source domain and the target domain, and the parameters of the network structure are shown in Table 1 below:

[0076]

[0077] The features extracted by the convolutional neural network will use the kernel method to map the features to a high-dimensional Hilbert space, and through network optimization, the aim is to obtain domain-invariant features. At the same time, it is expected that these domain-invariant features can achieve a high accuracy rate on the health state classifier.

[0078] The training steps of the feature extraction model, domain discriminator, and diagnostic model are as follows:

[0079] S41. Construct the total loss function of the feature extraction model, domain discriminator, and diagnostic model.

[0080] In step S41, the total loss function includes the following three parts:

[0081] S411. The source domain target domain feature extraction consistency evaluation function L MSD , which is used to measure the difference between the features extracted from the source domain data and the features extracted from the target domain.

[0082] Let the features extracted from the source domain be: S(f) = {G f (x; θ f ) | x ~ S(x)} represents the source domain data distribution, T(f) = {G f (x; θ f ) | x ~ T(x)} represents the target domain data distribution, G f (x; θ f ) represents the features after passing through the domain sharing feature extraction network, θ f represents the parameters of the domain sharing feature extractor, x represents the sample, S represents the source domain, T represents the target domain, x ~ S(x) represents the probability distribution that the sample follows the source domain, and x ~ T(x) represents the probability distribution that the sample follows the target domain. Among them, since the source domain data is collected from the laboratory and the target domain data is obtained from the actual working conditions of the bridge crane, there is a domain drift phenomenon, that is, S(x) ≠ T(x).

[0083] To more conveniently calculate the similarity between the feature sets of two samples, a Gaussian kernel matrix is constructed by dynamically calculating the bandwidth. Then, sub-matrices within the source domain, within the target domain, and across domains (i.e., between the target and the source) are extracted from the joint kernel matrix, and the mean square error formed between these sub-matrices is calculated as the difference, which is returned as the loss value to guide the training of the model.

[0084] The formula for the maximum mean square error loss function is as follows:

[0085]

[0086] Where n and m are the number of samples in the source domain and the target domain respectively, and K(·,·) represents the kernel function obtained by weighted summation of Gaussian kernels with different bandwidths through the method of dynamically calculating the kernel bandwidth. respectively represent the sum of the kernel function values of all sample pairs within the source domain, within the target domain, and between the source domain and the target domain.

[0087] Where S and T represent the source domain feature distribution and the target domain feature distribution respectively. The significance of this loss function is that when the network performs backpropagation, it can adjust the network parameters for domain-shared feature extraction according to the loss function, so that the maximum mean square error difference of the extracted features in the reproducing Hilbert space is minimized, that is, features invariant across domains can be extracted.

[0088] Figure 3 It is a schematic diagram of the bearing vibration signal. It can be seen that its mean value is near 0. Therefore, a large amount of information will be lost when using the traditional maximum mean discrepancy to extract domain-invariant features. However, using the maximum mean square error loss function can effectively improve this problem and enhance the generalization performance of the model.

[0089] S412, Sample health classification evaluation function L csoft , which is used to measure the accuracy of the features of the source domain data extracted by the same feature extractor in the fault classification task.

[0090] This step of improving Softmax is named C-Softmax. The reason for proposing the C-Softmax function is that the objective of conventional Softmax optimization is only to correctly classify known samples, without requiring the features to have a small intra-class distance and a large inter-class distance. In this way, during the knowledge transfer process, samples not seen in the target domain may be near the classification boundary, resulting in misclassification when diagnosing faults in bridge crane bearings. Taking the binary classification problem as an example, as Figure 4 shown.

[0091] Sample health classification evaluation function L csoft , is obtained by weighted summation of L arc and L softmax .

[0092]

[0093] Among them, L arc represents the magnitude of the loss value, N is the number of samples, yi is the true label of the i-th sample. In the present invention, the pseudo-label method is used to replace the true label. m is a hyperparameter used to increase the distance of the classification boundary, making different samples farther apart. is the cosine value after the distance is increased, used to increase the decision boundary of the correct class. n is the category of the health state of the bridge crane bearing, cosθ j is the cosine similarity between the feature vector and the weight vector for each category j.

[0094]

[0095] Among them, L softmax is the conventional cross-entropy loss, N is the number of samples, C is the type of the health state of the bridge crane bearing, y ij is the indicator variable that the i-th sample belongs to the j-th class, p ij is the probability that the model predicts that the i-th sample belongs to the j-th class. In the present invention, using only L arc loss will affect the convergence of the network due to the excessive initial classification error. Therefore, in combination with L softmax loss, on the basis that the model has a certain degree of accuracy, then using L arc loss to adjust the distribution of features, so the improved classification loss formula is as follows:

[0096] L csoft = L arc + L soft

[0097] Among them, L csoft is the improved classification loss.

[0098] S413. Domain discriminator evaluation function L d , making it difficult for the domain discriminator to distinguish between the two domains.

[0099]

[0100] Among them, g i is the label of the source domain. In the present invention, the source domain label is set to 1, and the target domain label is set to 0. d(x i ) is the probability estimation value that the i-th sample output by the domain classifier belongs to the source domain. m represents the number of samples. During the training process, we use gradient reversal to maximize this loss, so that the domain classifier cannot judge the true labels of the source domain and the target domain, in order to achieve the purpose of extracting domain-invariant features.

[0101] S42. According to the total loss function, through backpropagation, adjust the neuron parameters of the feature extraction model, domain discriminator, and diagnosis model.

[0102] In step S42, the overall optimization goal of the network is that the network can extract the common features of the laboratory bearing data and the actual operation data of the bridge crane. At the same time, the common features have the characteristics of small intra-class distance and large inter-class distance, and have sufficient discriminability. Therefore, the overall loss of the network can be characterized by the following formula:

[0103] L Total = L csoft + nL d + L MSD

[0104] Among them, n is a hyperparameter for determining the degree of domain adaptation. After clarifying the above loss function, that is, the optimization goal, let θ f , θ c , θ d represent the parameters of the domain-shared feature extractor, health status classifier, and domain classifier respectively. Then the total loss function can be re-given by the following formula:

[0105]

[0106] This embodiment trains the proposed network by means of stochastic gradient descent, and updates the parameters of the network through backpropagation. Using the stochastic gradient descent method, the process of updating θ f , θ c , θ d can be carried out by the formula:

[0107]

[0108]

[0109]

[0110] Among them, ε is the learning rate.

[0111] S43. Until the consistency of the source domain data features and target data features extracted by the feature extraction model, the difficulty of the discriminator in distinguishing the source domain data features and target data features, and the ease of recognition between various fault diagnosis regions all reach the preset goals.

[0112] S5. Use the trained domain-shared feature extraction network to diagnose the bearing fault categories of the bridge crane drive system.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: it is still possible to modify the specific implementation manners of the present invention or make equivalent substitutions, and any modification or equivalent substitution that does not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology, characterized in that: The specific method is as follows: Obtain bearing data of a bridge crane running in a laboratory as source domain data; Obtain the bearing data of the bridge crane in actual working conditions as the target domain data; Perform data cleaning on source domain data and target domain data; The domain-shared feature extraction network is trained with the cleaned source domain data and target domain data; Diagnose bearing fault categories of bridge crane transmission system using trained domain-shared feature extraction network.

2. The bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology according to claim 1 is characterized in that: The domain-shared feature extraction network includes a feature extraction model, a domain discriminator, and a diagnosis model; The feature extraction model extracts cleaned source domain data and target domain data; The domain discriminator discriminates the difference between the extracted source domain data and the target domain data; The diagnostic model is trained using the extracted source domain data and target domain data. The trained diagnostic model diagnoses the bearing fault category of the bridge crane transmission system based on the target domain data.

3. The bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology according to claim 2 is characterized in that: The training methods of the feature extraction model, domain discriminator and diagnosis model are as follows: Construct the total loss function of the feature extraction model, domain discriminator and diagnosis model; According to the total loss function, the neuron parameters of the feature extraction model, domain discriminator and diagnosis model are adjusted through back propagation; Until the consistency of the source domain data features extracted by the feature extraction model and the target data features, the difficulty of the discriminator in distinguishing the source domain data features from the target data features, and the degree of identification between various fault diagnosis areas, all reach the preset goals.

4. The bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology according to claim 3 is characterized in that: The consistency between the source domain data features extracted by the feature extraction model and the target data features is quantified by the source domain and target domain feature extraction consistency evaluation function. The specific formula is as follows: Where n and m are the number of samples in the source domain and the target domain, respectively. K(·,·) represents the kernel function obtained by weighted summation of Gaussian kernels of different bandwidths obtained by the dynamic kernel bandwidth calculation method. They represent the sum of the kernel function values ​​of all sample pairs in the source domain, the target domain, and between the source domain and the target domain.

5. The bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology according to claim 3 is characterized in that: The difficulty of the discriminator in distinguishing the characteristics of source domain data from the characteristics of target data is quantified by the domain discriminator evaluation function. The specific formula is as follows: where g i is the label of the source domain, the label of the target domain is set to 0, d(x i ) is the probability estimate of the i-th sample output by the domain classifier belonging to the source domain, and m represents the number of samples.

6. The bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology according to claim 3 is characterized in that: The degree of identification between various fault diagnosis areas is quantified by the sample health classification evaluation function. The specific formula is as follows: L csoft =L arc +L soft in Where, L csoft is the sample health classification evaluation function, L arc Indicates the size of the loss value, N is the number of samples, yi is the true label of the i-th sample; pseudo labels are used instead of true labels, and m is a hyperparameter used to increase the distance between classification boundaries, making different samples farther apart. To increase the cosine value after the distance, increase the decision boundary of the correct category, n is the category of the health status of the bridge crane bearing, cosθ j is the cosine similarity between the feature vector and the weight vector for each category j; L softmax is the conventional cross entropy loss, N is the number of samples, C is the type of bridge crane bearing health status, y ij is the indicator variable that the i-th sample belongs to the j-th class, p ij The model predicts the probability that the i-th sample belongs to the j-th class.

7. The bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology according to claim 3 is characterized in that: The specific formula of the total loss function is as follows: L Total =L csoft +nL d +L MSD Right now Among them, L Total is the total loss function, L csoft is the sample health classification evaluation function, L d is the domain discriminator evaluation function, L MSD is the consistency evaluation function for extracting features from the source domain and the target domain; n is the hyperparameter for determining domain adaptation, θ f ,θ c ,θ d Represent the neuron parameters of feature extraction model, diagnosis model and domain discriminator respectively.

8. The bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology according to claim 7, characterized in that: Through back propagation, the neuron parameters of the feature extraction model, domain discriminator and diagnosis model are adjusted. The specific formula is as follows: Among them, ε is the learning rate.

9. The bearing fault diagnosis method for a bridge crane transmission system based on domain adaptive technology according to claim 1, characterized in that: The types of bearing failures in the transmission system of bridge cranes include inner ring failure, outer ring failure and ball failure.

Images