A train rolling bearing state identification method and system

Abstract

The present invention provides a method and system for identifying the state of a train rolling bearing. The method includes acquiring monitoring data of a train rolling bearing, preprocessing the monitoring data to obtain processed data; performing data equalization on the processed data to obtain a balanced data set; dividing the balanced data set into a first data set and a second data set, performing modal interval processing on the second data set to obtain modal data; constructing a seed replacement deep migration initial model, inputting the first data set into the seed replacement deep migration initial model and optimizing the seed replacement deep migration initial model to output optimized data; determining the operating state of the train rolling bearing based on the interval between the optimized data and the modal data. The present invention improves the generalization performance of the model and solves the problem of large model data demand.

Description

Technical Field

[0001] The present invention belongs to the technical field of failure of key components of trains, and in particular relates to a method and system for identifying the state of a train rolling bearing. Background Art

[0002] The rapid development of rail transit has brought numerous conveniences to people's lives, but it has also garnered increasing attention for train safety. Rolling bearings are key components of bogies and axles, and are also crucial components of transmission systems. Promptly identifying their operating status is crucial for ensuring safe and stable train operation and the safety of train crew members.

[0003] During actual train operation, rolling bearings are subjected to heavy loads over extended periods of time and operate in a complex, harsh environment characterized by high speeds and high temperatures, making them susceptible to failure. Furthermore, rolling bearings are often integrated into other mechanical equipment, so the collected vibration signals contain not only the bearing's own vibration signals but also significant interference from impact noise and environmental noise. This makes it difficult to accurately identify the rolling bearing's operating status.

[0004] As for the train rolling bearing status recognition in the existing technology, it is generally recognized through a neural network, but it generally only relies on loss function learning, which is specifically a passive learning method. It has certain learning limitations and ultimately affects the accuracy of status recognition. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a train rolling bearing status identification method and system for solving the technical problems in the prior art.

[0006] In one aspect, the present invention provides the following technical solution: a method for identifying the state of a train rolling bearing, comprising:

[0007] Acquiring monitoring data of a train rolling bearing, and preprocessing the monitoring data to obtain processed data;

[0008] performing data balancing on the processed data to obtain a balanced data set;

[0009] Dividing the balanced data set into a first data set and a second data set, and performing modal interval processing on the second data set to obtain modal data;

[0010] Constructing a seed replacement deep migration initial model, inputting the first data set into the seed replacement deep migration initial model and optimizing the seed replacement deep migration initial model to output optimized data;

[0011] The operating state of the train rolling bearing is determined based on the interval between the optimization data and the modalization data.

[0012] Compared with the existing technology, the beneficial effects of the present invention are: the present invention is based on the deep migration regression method of seed replacement, constructs a balanced distribution adaptation loss function for regression tasks, realizes the full learning of domain knowledge, and realizes the active representation and learning of domain knowledge through the seed replacement strategy combined with clustering technology, thereby improving the generalization performance of the model and solving the problem of large model data demand.

[0013] Preferably, the step of preprocessing the monitoring data to obtain processed data includes:

[0014] Determining a window length, an overlap rate, and a sliding start position, and dividing the monitoring data into a plurality of windows of fixed lengths based on the window length and the sliding start position;

[0015] Overlapping two adjacent windows based on the overlap ratio to obtain expanded data;

[0016] A trained convolutional neural network is obtained, and the expanded data is input into the trained convolutional neural network for denoising to obtain processed data.

[0017] Preferably, the step of performing data balancing on the processed data to obtain a balanced data set includes:

[0018] Determine the source domain data D S :

[0019] ;

[0020] Where, For the i source domain sample feature vectors, for The corresponding label, m is the number of source domain samples;

[0021] Determine target domain data D T :

[0022] ;

[0023] Where, For the i a labeled target domain feature vector, for The corresponding label, n is the number of labeled target domain samples, For the i unlabeled target domain feature vectors,u is the number of unlabeled target domain samples;

[0024] Determine a distribution difference balancing function, and perform balancing processing on the processed data based on the distribution difference balancing function to obtain a balanced data set, wherein the distribution difference balancing function is:

[0025] ;

[0026] Where, λ is the balance factor, D MDA is the empirical mean value, D CDA is the conditional distribution adaptation, D BDA To balance the distribution, X S is the source domain sample feature vector, X T is the target domain sample feature vector.

[0027] Preferably, the step of performing modal interval processing on the second data set to obtain modal data includes:

[0028] The second data set is modally intervalized based on the interval feature quantization formula:

[0029] ;

[0030] Where, X k ( n ) indicates the k The modal components correspond to the modal interval form of the data set, x k ( n ), x k ( n -1) x k ( n +1) indicates the k The modal component corresponds to the n、n -1. n +1 data;

[0031] Convert the modal interval form of the second data set to obtain modal data:

[0032] X ( n ) =[ x ( n ), ( n )];

[0033] Where, x ( n ) represents the lower bound of the data set, ( n ) represents the upper bound of the dataset.

[0034] Preferably, the steps of constructing the seed replacement deep migration initial model, inputting the first data set into the seed replacement deep migration initial model, and optimizing the seed replacement deep migration initial model to output optimized data include:

[0035] Constructing an initial convolutional neural network model, freezing the shallow network to retain general feature extraction capabilities, migrating deep layer parameters, and using the source domain data of the initial convolutional neural network model as the initial parameters of the target domain;

[0036] Extracting high-dimensional features of the source domain data using the frozen shallow network, and performing clustering processing on the high-dimensional features of the source domain data to obtain a plurality of clusters;

[0037] Establishing a pairing relationship between the target domain label samples corresponding to the first data set and the source domain samples corresponding to the source domain data, and replacing the target domain label samples with the data in the cluster to obtain a first optimized data set;

[0038] Determine the fusion loss function L :

[0039] ;

[0040] Where, L r is the regression loss, L BAD is the marginal distribution difference loss and regularization, and are the first and second trade-off parameters, respectively. is the regularization coefficient, W Represents the model weight parameter;

[0041] By minimizing the fusion loss function L , and updating the first optimized data set to obtain a second optimized data set;

[0042] The parameters of the initial convolutional neural network model are updated to obtain an updated model, and the second optimized data set is updated based on the updated model to obtain optimized data.

[0043] Preferably, the steps of updating the parameters of the initial convolutional neural network model to obtain an updated model, and updating the second optimized data set based on the updated model to obtain optimized data include:

[0044] Calculating gradients g t :

[0045] ;

[0046] Where, Indicates the parameters of the model to be updated, Indicates the t The parameters to be updated. represents the gradient operator, represents the loss function;

[0047] Update the first-order moment estimate and the second-order moment estimate based on the gradient:

[0048] ;

[0049] Where, denote the first-order and second-order moment attenuation coefficients, respectively. Respectively represent t The first-order moment estimation and second-order moment estimation of the wheel, m t-1 、 v t-1 Respectively represent t -1 round of first-order moment estimation and second-order moment estimation;

[0050] Perform bias correction on the updated first-order moment estimate and second-order moment estimate:

[0051] ;

[0052] Where, 、 Respectively represent t Bias correction of first-order moment estimation and second-order moment estimation of the wheel;

[0053] The parameters of the initial convolutional neural network model are updated based on the bias-corrected first-order moment estimate and second-order moment estimate to obtain an updated model:

[0054] ;

[0055] Where, represents the parameters of the updated initial convolutional neural network model, is the learning rate, is a constant;

[0056] The second optimized data set is updated based on the updated model to obtain optimized data.

[0057] Preferably, the step of determining the operating state of the train rolling bearing based on the interval between the optimization data and the modal data includes:

[0058] The intervals of the optimized data and the modalized data are compared using a modal interval size comparison rule, and the operating state corresponding to the code corresponding to the maximum value of the modal interval is used as the operating state of the train rolling bearing.

[0059] In a second aspect, the present invention provides the following technical solution: a train rolling bearing status identification system, the system comprising:

[0060] a processing module, configured to obtain monitoring data of a train rolling bearing and pre-process the monitoring data to obtain processed data;

[0061] A balancing module, configured to perform data balancing on the processed data to obtain a balanced data set;

[0062] a modal module, configured to divide the balanced data set into a first data set and a second data set, and perform modal interval processing on the second data set to obtain modal data;

[0063] an optimization module, configured to construct an initial model of seed replacement deep migration, input the first data set into the initial model of seed replacement deep migration, and optimize the initial model of seed replacement deep migration to output optimized data;

[0064] A state module is used to determine the operating state of the train rolling bearing based on the interval between the optimization data and the modal data.

[0065] In a third aspect, the present invention provides the following technical solution: a computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor; the processor implements the above-mentioned method for identifying the state of a train rolling bearing when executing the computer program.

[0066] In a fourth aspect, the present invention provides the following technical solution: a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the train rolling bearing state identification method as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 any creative work.

[0068] Figure 1 This is a flow chart of a method for identifying the state of a train rolling bearing provided in Example 1 of the present invention;

[0069] Figure 2 This is a structural block diagram of a train rolling bearing status identification system provided in the second embodiment of the present invention;

[0070] Figure 3 A schematic diagram of the hardware structure of a computer provided in another embodiment of the present invention.

[0071] The embodiments of the present invention will be further described below with reference to the accompanying drawings. DETAILED DESCRIPTION

[0072] The following describes embodiments of the present invention in detail, 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 embodiments of the present invention, and should not be construed as limiting the present invention.

[0073] Example 1

[0074] In the first embodiment of the present invention, Figure 1 As shown, a method for identifying the state of a train rolling bearing includes:

[0075] S1. Acquire monitoring data of a train rolling bearing and pre-process the monitoring data to obtain processed data;

[0076] Specifically, the monitoring data acquisition process is as follows: A running gear test platform for train rolling bearings is constructed, consisting of two unidirectional accelerometers, a data acquisition card, a three-phase asynchronous motor, and a PC. The unidirectional accelerometers acquire operating data from the running gear rolling bearings. Two unidirectional accelerometers are installed horizontally and vertically on the end caps of the rolling bearings on the test bench. Time domain analysis is performed on the horizontal and vertical vibration signals captured by the two unidirectional accelerometers. The vibration signal with the largest amplitude is selected as the raw signal dataset for the train rolling bearing operation, from which the monitoring data is derived.

[0077] Wherein, step S1 includes:

[0078] S11 . Determine a window length, an overlap rate, and a sliding start position, and divide the monitoring data into a plurality of windows of fixed lengths based on the window length and the sliding start position.

[0079] S12, overlapping two adjacent windows based on the overlap ratio to obtain expanded data;

[0080] Specifically, when training convolutional neural networks, the effectiveness of the training is closely related to the number of samples. Insufficient samples can lead to overfitting of the model, resulting in a situation where the model performs well on the training set but poorly on the test set or new samples. Overlapping sampling is a common method for data preprocessing. Sliding window overlapping sampling expands the number of samples by dividing the original signal into fixed-length windows and overlapping adjacent windows.

[0081] S13, obtaining a trained convolutional neural network, and inputting the expanded data into the trained convolutional neural network for denoising to obtain processed data;

[0082] Specifically, the data can be denoised using a trained convolutional neural network to obtain processed data.

[0083] S2. performing data balancing on the processed data to obtain a balanced data set;

[0084] Specifically, there are differences in marginal distribution and conditional distribution between the target domain and the source domain, which is a common problem in the engineering field. In order to reduce the difference in marginal distribution and conditional distribution and break through the dependence of traditional loss functions, the present invention proposes a knowledge representation method based on clustering, which realizes the paradigm shift from passive optimization to active knowledge construction through clustering structure. This method does not focus on a single sample, but applies clustering to study the entire data set, realizes knowledge representation and storage through clustering-based global analysis and clustering structure, and obtains target domain characteristics through cluster center matching and local manifold self-learning;

[0085] Therefore, in order to achieve the simultaneous optimization of the distribution differences between the two categories, a weighted loss function suitable for regression tasks is proposed by referring to the domain adaptation strategy in classification tasks, and equalization is performed through this function.

[0086] Wherein, the step S2 includes:

[0087] S21. Determine source domain data D S :

[0088] ;

[0089] Where, For thei source domain sample feature vectors, for The corresponding label, m is the number of source domain samples;

[0090] Specifically, the source domain data can be understood as the training data of the model, all of which have corresponding labels and are sufficient in number.

[0091] S22. Determine target domain data D T :

[0092] ;

[0093] Where, For the i a labeled target domain feature vector, for The corresponding label, n is the number of labeled target domain samples, For the i unlabeled target domain feature vectors, u is the number of unlabeled target domain samples;

[0094] Specifically, the target domain data is the processed data obtained in the above steps in this application.

[0095] S23. Determine a distribution difference balancing function, and perform balancing processing on the processed data based on the distribution difference balancing function to obtain a balanced data set, wherein the distribution difference balancing function is:

[0096] ;

[0097] Where, λ is the balance factor, D MDA is the empirical mean value, D CDA is the conditional distribution adaptation, D BDA To balance the distribution, X S is the source domain sample feature vector, X T is the target domain sample feature vector;

[0098] Specifically, for the balance factor, it is used to dynamically adjust the weights of the two types of distribution differences: λ →1, the model focuses on the marginal distribution difference, which is suitable for the case where the overall distribution difference between the source domain and the target domain is significant. λ →0, the model mainly focuses on the conditional distribution difference, which is suitable for scenarios where the conditional distribution difference between categories dominates.

[0099] S3, dividing the balanced data set into a first data set and a second data set, and performing modal interval processing on the second data set to obtain modal data;

[0100] Specifically, the first data set and the second data set can be divided according to a ratio, the first data set is used for subsequent model output, and the second data set is used for modal interval processing to facilitate output of the final bearing state.

[0101] Wherein, the step S3 includes:

[0102] S31. Perform modal interval processing on the second data set based on the interval feature quantization formula:

[0103] ;

[0104] Where, X k ( n ) indicates the k The modal components correspond to the modal interval form of the data set, x k ( n ), x k ( n -1) x k ( n +1) indicates the k The modal component corresponds to the n、n -1. n +1 data;

[0105] Specifically, the interval feature quantization formula is the Teager Kaiser energy operator.

[0106] S32. Convert the modal interval form of the second data set to obtain modal data:

[0107] X ( n ) =[ x ( n ), ( n )];

[0108] Where, x ( n ) represents the lower bound of the data set, ( n ) represents the upper bound of the data set;

[0109] Specifically, since there are many uncertainties in the transmission path of the train rolling bearing vibration signal and the data preprocessing process, the second data set is converted into a modal interval form according to the error theory and modal interval theory to increase the reliability of the data set to be analyzed.

[0110] S4, constructing a seed replacement deep migration initial model, inputting the first data set into the seed replacement deep migration initial model, and optimizing the seed replacement deep migration initial model to output optimized data;

[0111] Wherein, the step S4 includes:

[0112] S41. Build an initial convolutional neural network model, freeze the shallow network to retain general feature extraction capabilities, migrate deep layer parameters, and use the source domain data of the initial convolutional neural network model as the initial parameters of the target domain;

[0113] Specifically, the initial model for deep migration with seed replacement is determined using a deep migration regression method based on seed replacement. In this framework, the shallow network corresponds to the convolutional blocks, and the deep network corresponds to the fully connected layers attached to the convolutional blocks. Seed samples are a small number of labeled samples in the target domain that guide the migration process. Seed replacement is part of the deep migration regression algorithm based on seed replacement. The main process is to use t-SNE to reduce the dimensionality of the 3D feature space. Through pre-training and fine-tuning strategies, knowledge can be transferred from the source domain to the target domain. Specifically, in the model pre-trained in the source domain, the parameters of the convolution block remain frozen and fixed during the model training process, and the source domain data of the initial convolutional neural network model is used as the initial parameters of the target domain. In the actual process, the intermediate output after the first fully connected layer is taken as the extracted feature data, and the seed replacement operation is applied to it. The pre-training and fine-tuning strategy is used to transfer knowledge from the source domain to the target domain. The rolling bearing categories are divided according to the operating status of the train rolling bearings. Combined with the modal intervalized train rolling bearing operating data set after denoising, the seed migration method is used to construct an initial seed replacement deep migration model, and the train rolling bearing operating status class is encoded to constitute the ideal output target of the seed replacement deep migration model.

[0114] S42. Using the frozen shallow network to extract high-dimensional features of the source domain data, and performing clustering processing on the high-dimensional features of the source domain data to obtain a plurality of clusters;

[0115] Specifically, the algorithm used in the clustering process here is the k-means++ clustering algorithm.

[0116] S43: establishing a pairing relationship between the target domain label samples corresponding to the first data set and the source domain samples corresponding to the source domain data, and replacing the target domain label samples with the data in the cluster to obtain a first optimized data set;

[0117] Specifically, the pairing relationship here can be established by determining the minimum distance between the data in the first data set and the cluster center, and establishing the pairing relationship through the minimum distance.

[0118] S44. Determine the fusion loss function L :

[0119] ;

[0120] Where, L r is the regression loss, L BAD is the marginal distribution difference loss and regularization, and are the first and second trade-off parameters, respectively. is the regularization coefficient, W Represents the model weight parameter;

[0121] Specifically, the first and second trade-off parameters are used to adjust the relationship between the regression loss and the balanced domain adaptation loss to prevent the model training from developing in the direction of excessively large values. The regression loss can enable the model to effectively integrate the features of the two domains and more fully inherit the knowledge of the source domain and target domain. W T W is the sum of squares of weights and is used to prevent the model from overfitting.

[0122] S45, by minimizing the fusion loss function L , and updating the first optimized data set to obtain a second optimized data set;

[0123] Specifically, the key to the model is to first keep the convolutional blocks fixed to avoid overfitting in small-sample training; secondly, feature reuse enables the fully connected layers to inherit the source model knowledge, accelerating the convergence of the target domain; and finally, balanced domain adaptation enables the fusion loss function to simultaneously optimize the difference between the marginal distribution and the conditional distribution.

[0124] At the same time, the deep migration regression method of seed replacement not only focuses on the adaptability of marginal distribution or conditional distribution, but also constructs a balanced distribution adaptation loss function for regression tasks to fully learn domain knowledge. By combining seed replacement with clustering technology, it realizes the active representation and learning of domain knowledge. Specifically, it has the following advantages:

[0125] 1) By balancing the domain adaptation loss function to simultaneously measure the difference between marginal distribution and conditional distribution, the full learning of domain knowledge is achieved;

[0126] 2) Using seed replacement technology, the source and target domain knowledge are integrated in the form of cluster structures and cluster centers to achieve active learning of domain knowledge;

[0127] 3) A seed replacement deep migration method is proposed to build a comprehensive and proactive regression task learning framework from three aspects: model architecture, data representation, and loss function.

[0128] S46. Update the parameters of the initial convolutional neural network model to obtain an updated model, and update the second optimized data set based on the updated model to obtain optimized data;

[0129] Wherein, the step S46 includes:

[0130] S461, calculate gradient g t :

[0131] ;

[0132] Where, Indicates the parameters of the model to be updated, Indicates the t The parameters to be updated. represents the gradient operator, represents the loss function.

[0133] S462: Update the first-order moment estimate and the second-order moment estimate based on the gradient:

[0134] ;

[0135] Where, denote the first-order and second-order moment attenuation coefficients, respectively. Respectively represent t The first-order moment estimation and second-order moment estimation of the wheel, m t-1 、 v t-1 Respectively represent t -1 round of first-order moment estimation and second-order moment estimation.

[0136] S463. Perform bias correction on the updated first-order moment estimate and second-order moment estimate:

[0137] ;

[0138] Where, 、 Respectively represent t Bias correction of the first-order moment estimation and second-order moment estimation of the wheel.

[0139] S464. Update the parameters of the initial convolutional neural network model based on the first-order moment estimate and the second-order moment estimate after deviation correction to obtain an updated model:

[0140] ;

[0141] Where, represents the parameters of the updated initial convolutional neural network model, is the learning rate, is a constant;

[0142] Specifically, the constant here is generally a very small constant to prevent division by zero errors, usually 10 -8 At the same time, the above update process is specifically the optimization process of the Adam optimizer. The Adam optimizer combines the advantages of momentum and adaptive learning rate, and dynamically adjusts the parameter update step size by calculating the first-order moment (mean) and second-order moment (variance) of the gradient.

[0143] S465: Update the second optimized data set based on the updated model to obtain optimized data.

[0144] S5. Determine the operating status of the train rolling bearing based on the interval between the optimization data and the modal data.

[0145] Specifically, step S5 is as follows:

[0146] Comparing the intervals of the optimized data and the modalized data using a modal interval size comparison rule, and taking the operating state corresponding to the code corresponding to the maximum value of the modal interval as the operating state of the train rolling bearing;

[0147] At the same time, in actual bearing fault diagnosis, the average accuracy of the deep seed replacement model on 12 migration tasks was 98.27%, which was significantly better than the comparative experiment with a small difference in accuracy. This shows that the model has good stability.

[0148] The train rolling bearing state identification method provided in Example 1 of the present invention is based on a deep migration regression method with seed replacement, constructs a balanced distribution adaptation loss function for regression tasks, realizes full learning of domain knowledge, and realizes active representation and learning of domain knowledge through a seed replacement strategy combined with clustering technology, thereby improving the generalization performance of the model and solving the problem of large model data demand.

[0149] Example 2

[0150] like Figure 2 As shown, in a second embodiment of the present invention, a train rolling bearing state identification system is provided, the system comprising:

[0151] Processing module 1, used to obtain monitoring data of train rolling bearings and pre-process the monitoring data to obtain processed data;

[0152] The balancing module 2 is used to perform data balancing on the processed data to obtain a balanced data set;

[0153] a modal module 3, configured to divide the balanced data set into a first data set and a second data set, and perform modal interval processing on the second data set to obtain modal data;

[0154] Optimization module 4 is used to construct an initial model of seed replacement deep migration, input the first data set into the initial model of seed replacement deep migration and optimize the initial model of seed replacement deep migration to output optimized data;

[0155] A state module 5 is configured to determine an operating state of a train rolling bearing based on an interval between the optimization data and the modalized data;

[0156] The processing module 1 includes:

[0157] A window submodule, configured to determine a window length, an overlap ratio, and a sliding start position, and to divide the monitoring data into a plurality of windows of fixed lengths based on the window length and the sliding start position;

[0158] an overlapping submodule, configured to overlap two adjacent windows based on the overlapping rate to obtain expanded data;

[0159] The denoising submodule is used to obtain a trained convolutional neural network and input the expanded data into the trained convolutional neural network for denoising to obtain processed data.

[0160] The balancing module 2 includes:

[0161] Source domain submodule, used to determine the source domain data D S :

[0162] ;

[0163] Where, For the i source domain sample feature vectors, for The corresponding label, m is the number of source domain samples;

[0164] Target domain submodule, used to determine the target domain data D T :

[0165] ;

[0166] Where, For the i a labeled target domain feature vector, for The corresponding label, n is the number of labeled target domain samples, For the i unlabeled target domain feature vectors, u is the number of unlabeled target domain samples;

[0167] The balancing submodule is configured to determine a distribution difference balancing function and perform balancing processing on the processed data based on the distribution difference balancing function to obtain a balanced data set, wherein the distribution difference balancing function is:

[0168] ;

[0169] Where, λ is the balance factor, D MDA is the empirical mean value, D CDA is the conditional distribution adaptation, D BDA To balance the distribution, X S is the source domain sample feature vector, X T is the target domain sample feature vector.

[0170] The modal module 3 includes:

[0171] The intervalization submodule is used to perform modal intervalization processing on the second data set based on the interval feature quantization formula:

[0172] ;

[0173] Where, X k ( n ) indicates the k The modal components correspond to the modal interval form of the data set, x k ( n ), x k ( n -1) x k ( n +1) indicates the k The modal component corresponds to the n、n -1. n +1 data;

[0174] The conversion submodule is used to convert the modal interval form of the second data set to obtain modal data:

[0175] X ( n ) =[ x ( n ), ( n )];

[0176] Where, x ( n ) represents the lower bound of the data set, ( n ) represents the upper bound of the dataset.

[0177] The optimization module 4 includes:

[0178] A freezing submodule is used to construct an initial convolutional neural network model, freeze the shallow network to retain the general feature extraction capability, migrate the deep layer parameters, and use the source domain data of the initial convolutional neural network model as the initial parameters of the target domain;

[0179] A clustering submodule, configured to extract high-dimensional features of the source domain data using a frozen shallow network, and perform clustering processing on the high-dimensional features of the source domain data to obtain a plurality of clusters;

[0180] a replacement submodule, configured to establish a pairing relationship between the target domain label samples corresponding to the first data set and the source domain samples corresponding to the source domain data, and replace the target domain label samples with the data in the cluster to obtain a first optimized data set;

[0181] Loss submodule, used to determine the fusion loss function L :

[0182] ;

[0183] Where, L r is the regression loss, L BAD is the marginal distribution difference loss and regularization, and are the first and second trade-off parameters, respectively. is the regularization coefficient, W Represents the model weight parameter;

[0184] The first update submodule is used to minimize the fusion loss function L , and updating the first optimized data set to obtain a second optimized data set;

[0185] The second updating submodule is used to update the parameters of the initial convolutional neural network model to obtain an updated model, and update the second optimized data set based on the updated model to obtain optimized data.

[0186] The second updating submodule includes:

[0187] Gradient unit, used to calculate gradient g t :

[0188] ;

[0189] Where, Indicates the parameters of the model to be updated, Indicates the t The parameters to be updated. represents the gradient operator, represents the loss function;

[0190] An estimation unit, configured to update a first-order moment estimate and a second-order moment estimate based on the gradient:

[0191] ;

[0192] Where, denote the first-order and second-order moment attenuation coefficients, respectively. Respectively represent t The first-order moment estimation and second-order moment estimation of the wheel, m t-1 、 v t-1 Respectively represent t -1 round of first-order moment estimation and second-order moment estimation;

[0193] The bias unit is used to correct the bias of the updated first-order moment estimate and second-order moment estimate:

[0194] ;

[0195] Where, 、 Respectively represent t Bias correction of first-order moment estimation and second-order moment estimation of the wheel;

[0196] A parameter unit is used to update the parameters of the initial convolutional neural network model based on the bias-corrected first-order moment estimate and the second-order moment estimate to obtain an updated model:

[0197] ;

[0198] Where, represents the parameters of the updated initial convolutional neural network model, is the learning rate, is a constant;

[0199] The second optimized data set is updated based on the updated model to obtain optimized data.

[0200] The state module 5 is specifically used for:

[0201] The intervals of the optimized data and the modalized data are compared using a modal interval size comparison rule, and the operating state corresponding to the code corresponding to the maximum value of the modal interval is used as the operating state of the train rolling bearing.

[0202] In some other embodiments of the present invention, the embodiments of the present invention provide the following technical solutions: a computer, comprising a memory 102, a processor 101, and a computer program stored in the memory 102 and executable on the processor 101; the processor 101 implements the train rolling bearing state identification method as described above when executing the computer program.

[0203] Specifically, the processor 101 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured as one or more integrated circuits for implementing the embodiments of the present invention.

[0204] Memory 102 may include a large-capacity memory for data or instructions. By way of example, and not limitation, memory 102 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 102 may include removable or non-removable (or fixed) media. Where appropriate, memory 102 may be internal or external to the data processing device. In certain embodiments, memory 102 is non-volatile memory. In certain embodiments, memory 102 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0205] The memory 102 may be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 101 .

[0206] The processor 101 implements the above-mentioned train rolling bearing state identification method by reading and executing the computer program instructions stored in the memory 102.

[0207] In some embodiments, the computer may further include a communication interface 103 and a bus 100. Figure 3 As shown, the processor 101 , the memory 102 , and the communication interface 103 are connected via a bus 100 and communicate with each other.

[0208] The communication interface 103 is used to implement communication between the various modules, devices, units, and / or equipment in the embodiments of the present invention. The communication interface 103 can also implement data communication with other components such as external devices, image / data acquisition equipment, databases, external storage, and image / data processing workstations.

[0209] Bus 100 includes hardware, software, or both, and couples components of a computer device to each other. Bus 100 includes, but is not limited to, at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 100 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Bus 100 may include one or more buses, where appropriate. Although embodiments of the present invention describe and illustrate a particular bus, the present invention contemplates any suitable bus or interconnect.

[0210] The computer can execute the train rolling bearing state identification method of the present invention based on the acquired train rolling bearing state identification system, thereby realizing train rolling bearing state identification.

[0211] In some further embodiments of the present invention, in combination with the above-mentioned train rolling bearing status identification method, the embodiments of the present invention provide the following technical solutions: a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned train rolling bearing status identification method is implemented.

[0212] Those skilled in the art will appreciate that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device), or in conjunction with such instruction execution system, apparatus, or device. For purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by an instruction execution system, apparatus, or device, or in conjunction with such instruction execution system, apparatus, or device.

[0213] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0214] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the aforementioned embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or a combination of the following technologies known in the art may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0215] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0216] The above-described embodiments merely represent several implementations of the present invention, and while their descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person of ordinary skill in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, and these variations and improvements fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A method for identifying the state of a train rolling bearing, characterized in that: include: Acquiring monitoring data of a train rolling bearing, and preprocessing the monitoring data to obtain processed data; performing data balancing on the processed data to obtain a balanced data set; Dividing the balanced data set into a first data set and a second data set, and performing modal interval processing on the second data set to obtain modal data; Constructing a seed replacement deep migration initial model, inputting the first data set into the seed replacement deep migration initial model and optimizing the seed replacement deep migration initial model to output optimized data; determining an operating state of a train rolling bearing based on an interval between the optimization data and the modalized data; The step of performing modal interval processing on the second data set to obtain modal data includes: The second data set is modally intervalized based on the interval feature quantization formula: ； Where, X k ( n ) indicates the k The modal components correspond to the modal interval form of the data set, x k ( n ), x k ( n -1) x k ( n +1) indicates the k The modal component corresponds to the n、n -1. n +1 data; Convert the modal interval form of the second data set to obtain modal data: X ( n ) =[ x ( n )， ( n )]； Where, x ( n ) represents the lower bound of the data set, ( n ) represents the upper bound of the data set; The steps of constructing a seed replacement deep migration initial model, inputting the first data set into the seed replacement deep migration initial model, and optimizing the seed replacement deep migration initial model to output optimized data include: Constructing an initial convolutional neural network model, freezing the shallow network to retain general feature extraction capabilities, migrating deep layer parameters, and using the source domain data of the initial convolutional neural network model as the initial parameters of the target domain; Extracting high-dimensional features of the source domain data using the frozen shallow network, and performing clustering processing on the high-dimensional features of the source domain data to obtain a plurality of clusters; Establishing a pairing relationship between the target domain label samples corresponding to the first data set and the source domain samples corresponding to the source domain data, and replacing the target domain label samples with the data in the cluster to obtain a first optimized data set; Determine the fusion loss function L : ； Where, L r is the regression loss, L BAD is the marginal distribution difference loss and regularization, and are the first and second trade-off parameters, respectively. is the regularization coefficient, W Represents the model weight parameter; By minimizing the fusion loss function L , and updating the first optimized data set to obtain a second optimized data set; Updating parameters of the initial convolutional neural network model to obtain an updated model, and updating the second optimized data set based on the updated model to obtain optimized data; The step of determining the operating state of the train rolling bearing based on the interval between the optimization data and the modal data includes: The intervals of the optimized data and the modalized data are compared using a modal interval size comparison rule, and the operating state corresponding to the code corresponding to the maximum value of the modal interval is used as the operating state of the train rolling bearing.

2. The train rolling bearing state identification method according to claim 1, characterized in that: The step of preprocessing the monitoring data to obtain processed data includes: Determining a window length, an overlap rate, and a sliding start position, and dividing the monitoring data into a plurality of windows of fixed lengths based on the window length and the sliding start position; Overlapping two adjacent windows based on the overlap ratio to obtain expanded data; A trained convolutional neural network is obtained, and the expanded data is input into the trained convolutional neural network for denoising to obtain processed data.

3. The train rolling bearing state identification method according to claim 1, characterized in that: The step of performing data balancing on the processed data to obtain a balanced data set includes: Determine the source domain data D S : ； Where, For the i source domain sample feature vectors, for The corresponding label, m is the number of source domain samples; Determine target domain data D T : ； Where, For the i a labeled target domain feature vector, for The corresponding label, n is the number of labeled target domain samples, For the i unlabeled target domain feature vectors, u is the number of unlabeled target domain samples; Determine a distribution difference balancing function, and perform balancing processing on the processed data based on the distribution difference balancing function to obtain a balanced data set, wherein the distribution difference balancing function is: ； Where, λ is the balance factor, D MDA is the empirical mean value, D CDA is the conditional distribution adaptation, D BDA To balance the distribution, X S is the source domain sample feature vector, X T is the target domain sample feature vector.

4. The train rolling bearing state identification method according to claim 1, characterized in that: The steps of updating the parameters of the initial convolutional neural network model to obtain an updated model, and updating the second optimized data set based on the updated model to obtain optimized data include: Calculating gradients g t : ； Where, Indicates the parameters of the model to be updated, Indicates the t The parameters to be updated. represents the gradient operator, represents the loss function; Update the first-order moment estimate and the second-order moment estimate based on the gradient: ； Where, denote the first-order and second-order moment attenuation coefficients, respectively. Respectively represent t The first-order moment estimation and second-order moment estimation of the wheel, m t-1 、 v t-1 Respectively represent t -1 round of first-order moment estimation and second-order moment estimation; Perform bias correction on the updated first-order moment estimate and second-order moment estimate: ； Where, 、 Respectively represent t Bias correction of first-order moment estimation and second-order moment estimation of the wheel; The parameters of the initial convolutional neural network model are updated based on the bias-corrected first-order moment estimate and second-order moment estimate to obtain an updated model: ； Where, represents the parameters of the updated initial convolutional neural network model, is the learning rate, is a constant; The second optimized data set is updated based on the updated model to obtain optimized data.

5. A train rolling bearing status identification system, the system adopting the train rolling bearing status identification method according to claim 1, characterized in that: The system comprises: a processing module, configured to obtain monitoring data of a train rolling bearing and pre-process the monitoring data to obtain processed data; A balancing module, configured to perform data balancing on the processed data to obtain a balanced data set; a modal module, configured to divide the balanced data set into a first data set and a second data set, and perform modal interval processing on the second data set to obtain modal data; an optimization module, configured to construct an initial model of seed replacement deep migration, input the first data set into the initial model of seed replacement deep migration, and optimize the initial model of seed replacement deep migration to output optimized data; A state module is used to determine the operating state of the train rolling bearing based on the interval between the optimization data and the modal data.

6. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the train rolling bearing state identification method according to any one of claims 1 to 4 is implemented.

7. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the train rolling bearing state identification method according to any one of claims 1 to 4.

Images