Train rolling bearing state identification method and system

Abstract

The invention provides a train rolling bearing state identification method and system, and the method comprises the steps: obtaining the monitoring data of a train rolling bearing, and carrying out the preprocessing of the monitoring data, so as to obtain the processing data; performing data equalization processing on the processed data to obtain an equalized 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 depth migration initial model, inputting the first data set into the seed replacement depth migration initial model, and optimizing the seed replacement depth migration initial model to output optimized data; and determining the running state of the train rolling bearing based on the interval of the optimization data and the modal data. According to the method, the generalization performance of the model is improved, and meanwhile the problem that the model data demand is large is solved.

Description

Technical Field

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

[0002] The rapid development of rail transit has brought many conveniences to people's lives, and at the same time, the safety of trains has received more attention. Rolling bearings are key components of the bogie components and axles, and are also one of the important components of the transmission system. Timely identification of their operating status is of great significance for ensuring the safe and stable operation of trains and the lives of train personnel.

[0003] When a train is actually running, its rolling bearings are under huge loads for a long time, and the bearings are in a complex and harsh operating environment of high speed and high temperature, which makes it easy for train bearings to fail. Moreover, rolling bearings are often integrated in other mechanical equipment, so in the collected vibration signals, in addition to the vibration signals of the bearings themselves, there is also a lot of interference from impact noise and environmental noise, which causes great difficulties in accurately identifying the operating status of rolling bearings.

[0004] For the identification of the state of train rolling bearings in the prior art, it generally uses a neural network for identification, but it generally only depends on loss function learning, which is a passive learning method and has certain learning limitations, ultimately affecting the accuracy of state identification. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a method and system for identifying the state of train rolling bearings to solve the technical problems in the prior art.

[0006] On the one hand, the present invention provides the following technical solution. A method for identifying the state of train rolling bearings includes: Obtaining monitoring data of train rolling bearings, and preprocessing the monitoring data to obtain processed data; Performing data equalization processing on the processed data to obtain an equalized data set; Dividing the equalized data set into a first data set and a second data set, and performing modal interval processing on the second data set to obtain modalized data; Constructing a seed replacement deep transfer initial model, inputting the first data set into the seed replacement deep transfer initial model and performing optimization processing on the seed replacement deep transfer initial model to output optimized data; Determining the operating state of the train rolling bearing based on the intervals of the optimized data and the modalized data.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows: Based on the deep transfer regression method of seed replacement, a balanced distribution adaptation loss function for regression tasks is constructed to achieve full learning of domain knowledge. Through the seed replacement strategy and combined with clustering technology, the active representation and learning of domain knowledge are realized, thereby improving the generalization performance of the model and solving the problem of large model data requirements.

[0008] Preferably, the step of preprocessing the monitoring data to obtain processed data includes: Determine the window length, overlap rate, and sliding start position, and divide the monitoring data into several windows of fixed length based on the window length and the sliding start position; Overlap two adjacent windows based on the overlap rate to obtain augmented data; Obtain a trained convolutional neural network, and input the augmented data into the trained convolutional neural network for denoising processing to obtain processed data.

[0009] Preferably, the step of performing data equalization processing on the processed data to obtain an equalized data set includes: Determine the source domain data D S : ; Wherein, is the i th source domain sample feature vector, is corresponding label, m is the number of source domain samples; Determine the target domain data D T : ; Wherein, is the i th labeled target domain feature vector, is corresponding label, n is the number of labeled target domain samples, is the i th unlabeled target domain feature vector, u is the number of unlabeled target domain samples; Determine the distribution difference equalization function, and perform equalization processing on the processed data based on the distribution difference equalization function to obtain an equalized data set, wherein the distribution difference equalization function is: ; Wherein, λ is the balance factor, DMDA is the empirical average value, D CDA is the conditional distribution adaptation, D BDA is the balanced distribution adaptation, X S is the source domain sample feature vector, X T is the target domain sample feature vector.

[0010] Preferably, the step of performing modal intervalization on the second data set to obtain modalized data includes: Performing modal intervalization on the second data set based on the interval feature quantization formula: ; In the formula, X k ( n ) represents the modal interval form of the data set corresponding to the k th modal component, x k ( n ), x k ( n -1), x k ( n +1) represent the k th and n, n -1, n +1 data in the data set corresponding to the th modal component; X ( n ) = x ( n ), ( n )]; In the formula, x ( n ) represents the lower bound of the data set, ( n ) represents the upper bound of the data set.

[0011] Preferably, the step of constructing a seed replacement deep transfer initial model, inputting the first data set into the seed replacement deep transfer initial model and optimizing the seed replacement deep transfer initial model to output optimized data includes: Construct an initial convolutional neural network model, freeze the shallow network to retain the general feature extraction ability, transfer the deep parameters, and use the source domain data of the initial convolutional neural network model as the initial parameters of the target domain; 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 several clustering clusters; 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 use the data within the clustering clusters to replace the target domain label samples to obtain a first optimized data set; Determine the fusion loss function L : ; In the formula, 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; 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.

[0012] Preferably, the step 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 includes: Calculate the gradient g t : ; In the formula, represents the parameter to be updated of the model, represents the parameter to be updated in the t th round, represents the gradient operator, represents the loss function; Update the first-order moment estimate and the second-order moment estimate based on the gradient: ; In the formula, represent the first-order and second-order moment decay coefficients respectively, represent the first-order moment estimate and the second-order moment estimate in the t th round respectively, m t-1 、 vt-1 respectively represent the first - order moment estimation and the second - order moment estimation of the t -1 - th round; Perform bias correction on the updated first - order moment estimation and second - order moment estimation: ; In the formula, and respectively represent the bias corrections of the first - order moment estimation and the second - order moment estimation of the t - th round; Update the parameters of the initial convolutional neural network model based on the bias - corrected first - order moment estimation and second - order moment estimation to obtain an updated model: ; In the formula, represents the parameters of the updated initial convolutional neural network model, is the learning rate, is a constant; Update the second optimization data set based on the updated model to obtain optimized data.

[0013] Preferably, the step of determining the running state of the train rolling bearing based on the interval between the optimized data and the modalized data includes: Compare the interval between the optimized data and the modalized data using the modal interval size comparison rule, and take the running state corresponding to the encoding of the maximum value of the modal interval as the running state of the train rolling bearing.

[0014] In a second aspect, the present invention provides the following technical solution. A train rolling bearing state recognition system, the system includes: A processing module, configured to obtain the monitoring data of the train rolling bearing, and pre - process the monitoring data to obtain processed data; An equalization module, configured to perform data equalization processing on the processed data to obtain an equalized data set; A modal module, configured to divide the equalized data set into a first data set and a second data set, and perform modal interval processing on the second data set to obtain modalized data; An optimization module, configured to construct a seed - replacement deep transfer initial model, input the first data set into the seed - replacement deep transfer initial model, and perform optimization processing on the seed - replacement deep transfer initial model to output optimized data; A state module, configured to determine the running state of the train rolling bearing based on the interval between the optimized data and the modalized data.

[0015] In a third aspect, the present invention provides the following technical solution: a computer, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the above-mentioned train rolling bearing state recognition method is implemented.

[0016] In a fourth aspect, the present invention provides the following technical solution: 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 state recognition method is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of the train rolling bearing state recognition method provided in Embodiment 1 of the present invention; Figure 2 It is a structural block diagram of the train rolling bearing state recognition system provided in Embodiment 2 of the present invention; Figure 3 It is a schematic diagram of the hardware structure of a computer provided in another embodiment of the present invention.

[0019] The following will further illustrate the embodiments of the present invention with reference to the drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The following will describe in detail the embodiments of the present invention. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the embodiments of the present invention and should not be construed as limiting the present invention.

[0021] Embodiment 1 In Embodiment 1 of the present invention, as Figure 1 shown, a train rolling bearing state recognition method includes: S1. Obtain the monitoring data of the train rolling bearing, and preprocess the monitoring data to obtain processed data; Specifically, the process of obtaining monitoring data is as follows: For the train rolling bearing, a train running gear experimental platform is built, including two unidirectional acceleration sensors, a data acquisition card, a three-phase asynchronous motor, a PC, etc. Among them, the unidirectional acceleration sensors obtain the operation data of the train running gear rolling bearing, and the two unidirectional acceleration sensors are respectively installed in the horizontal and vertical directions of the rolling bearing end cover of the test bench. Perform time-domain analysis on the vibration signals in the horizontal and vertical directions obtained by the two unidirectional acceleration sensors, and select the vibration signal with a large amplitude as the original signal dataset of the train rolling bearing operation, and thus the monitoring data can be obtained.

[0022] Among them, step S1 includes: S11. Determine the window length, overlap rate, and sliding start position, and divide the monitoring data into several windows with a fixed length based on the window length and the sliding start position.

[0023] S12. Overlap two adjacent windows based on the overlap rate to obtain extended data; Specifically, when training a convolutional neural network, the training effect is closely related to the number of samples. If the number of samples is insufficient, it may cause overfitting in model training, resulting in a good performance of the model on the training set but poor performance on the test set or new samples. Overlapping sampling is one of the common processing methods in data preprocessing. Among them, sliding window overlapping sampling expands the number of samples by dividing the original signal into windows with a fixed length and overlapping between each adjacent window.

[0024] S13. Obtain a trained convolutional neural network, and input the extended data into the trained convolutional neural network for denoising processing to obtain processed data; Specifically, the trained convolutional neural network can perform denoising processing on the data to obtain processed data.

[0025] S2. Perform data equalization processing on the processed data to obtain an equalized data set; Specifically, there are differences between the target domain and the source domain in terms of marginal distribution and conditional distribution, which is a common problem in the engineering field. To reduce the marginal distribution difference and conditional distribution difference and break through the dependence on traditional loss functions, the present invention proposes a clustering-based knowledge representation method, which realizes a paradigm shift from passive optimization to active knowledge construction through a clustering structure. This method does not focus on individual samples, 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 features through clustering center matching and local manifold self-learning; Therefore, to achieve synchronous optimization of the distribution differences between the two types, a domain adaptation strategy in classification tasks is referred to, and a weighted loss function applicable to regression tasks is proposed, and the processing data is equalized through this function.

[0026] Among them, the step S2 includes: S21. Determine the source domain data D S : ; In the formula, is the i th source domain sample feature vector, is corresponding label, m is the number of source domain samples; Specifically, the source domain data can be understood as the training data of the model, all of its samples are with corresponding labels, and the quantity is sufficient.

[0027] S22. Determine the target domain data D T : ; In the formula, is the i th labeled target domain feature vector, is corresponding label, n is the number of labeled target domain samples, is the i th unlabeled target domain feature vector, u is the number of unlabeled target domain samples; Specifically, the target domain data is the processed data obtained in the above steps of this application.

[0028] S23. Determine the distribution difference equalization function, and equalize the processed data based on the distribution difference equalization function to obtain an equalized data set. Among them, the distribution difference equalization function is: ; In the formula, λ is the balance factor, D MDA is the empirical average value, D CDA is the conditional distribution adaptation, D BDA is the balanced distribution adaptation, X S is the source domain sample feature vector, X T is the target domain sample feature vector; Specifically, for the balance factor, it is used to dynamically adjust the weights of the differences between the two types of distributions: When λ → 1, the model mainly focuses on the marginal distribution difference, which is applicable to the case where the overall distribution difference between the source domain and the target domain is significant. When λ → 0, the model mainly focuses on the conditional distribution difference, which is applicable to the scenario where the conditional distribution difference between categories dominates.

[0029] S3. Divide the balanced data set into a first data set and a second data set, and perform modal intervalization on the second data set to obtain modalized data; 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 intervalization to facilitate the output of the final bearing state.

[0030] Among them, the step S3 includes: S31. Perform modal intervalization on the second data set based on the interval feature quantization formula: ; In the formula, X k ( n ) represents the modal interval form of the data set corresponding to the k th modal component, x k ( n ), x k ( n - 1), x k ( n + 1) represent the k th and n, n - 1, n + 1 data in the data set corresponding to the th modal component; Specifically, the interval feature quantization formula is the Teager Kaiser energy operator.

[0031] S32. Convert the modal interval form of the second data set to obtain modalized data: X ( n ) = x ( n ), ( n )]; In the formula, x ( n ) represents the lower bound of the data set, ( n ) represents the upper bound of the data set; Specifically, due to many uncertainties in the vibration signal transmission path of the train rolling bearing and the data preprocessing process, according to the error theory and the modal interval theory, the second data set is converted into the modal interval form to increase the reliability of the data set to be analyzed.

[0032] S4. Construct an initial model for seed replacement-based deep transfer, input the first data set into the initial model for seed replacement-based deep transfer, and perform optimization processing on the initial model for seed replacement-based deep transfer to output optimized data; Among them, the step S4 includes: S41. Construct an initial convolutional neural network model, freeze the shallow network to retain the general feature extraction ability, transfer the deep parameters, and use the source domain data of the initial convolutional neural network model as the initial parameters of the target domain; Specifically, for the initial model for seed replacement-based deep transfer, it is determined by the deep transfer regression method based on seed replacement. In the framework of the deep transfer regression method based on seed replacement, the shallow network corresponds to the convolutional block, and the deep network corresponds to the fully connected layer attached to the convolutional block. The seed samples are a small number of labeled samples in the target domain, which are used to guide the transfer process; seed replacement is a part of the algorithm of the deep transfer regression method based on seed replacement. The main process is to use t-SNE to reduce the dimension in the 3D feature space. Through the 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 convolutional block remain frozen and unchanged 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 strategies are used to transfer knowledge from the source domain to the target domain. According to the operating state of the train rolling bearing, the rolling bearing categories are divided. Combining with the modal intervalized train rolling bearing operating data set after noise reduction, using the seed transfer method, an initial model for seed replacement-based deep transfer is constructed, and the operating state categories of the train rolling bearing are encoded to form the ideal output target of the model for seed replacement-based deep transfer.

[0033] S42. Use the frozen shallow network to extract the high-dimensional features of the source domain data, and perform clustering processing on the high-dimensional features of the source domain data to obtain several clustering clusters; Specifically, the algorithm used for the clustering processing here is the k-means++ clustering algorithm.

[0034] S43. 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 use the data within the clustering cluster to replace the target domain label samples to obtain a first optimized data set; Specifically, the pairing relationship here can be established by determining the minimum distance between the data in the first dataset and the cluster center, and establishing the pairing relationship through the minimum distance.

[0035] S44. Determine the fusion loss function L : ; In the formula, 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; Specifically, the first and second trade-off parameters are used to adjust the magnitude relationship between the regression loss and the balanced domain adaptation loss to prevent the model training from developing in the direction of too large a value. The regression loss can enable the model to effectively fuse the features of the two domains and inherit the knowledge of the source domain and the target domain more fully. W T W is the sum of the squares of the weights, which is used to prevent the model from overfitting.

[0036] S45. By minimizing the fusion loss function L , and updating the first optimized dataset to obtain a second optimized dataset; Specifically, the key of the model is that first, the convolutional block remains fixed to avoid overfitting in small-sample training. Second, feature reuse enables the fully connected layer to inherit the knowledge of the source model and accelerate the convergence of the target domain. Finally, balanced domain adaptation synchronously optimizes the differences in the marginal distribution and the conditional distribution of the fusion loss function; At the same time, the deep transfer regression method with seed replacement not only pays attention to the adaptability of the marginal distribution or the conditional distribution, but also constructs a balanced distribution adaptation loss function for the regression task to achieve full learning of domain knowledge, and realizes the active representation and learning of domain knowledge through the combination of seed replacement and clustering technology. Specifically, it has the following advantages: 1) Synchronously measure the differences in the marginal distribution and the conditional distribution through the balanced domain adaptation loss function to achieve full learning of domain knowledge; 2) Adopt the seed replacement technology to fuse the source domain - target domain knowledge in the form of a clustering structure and a cluster center to achieve active learning of domain knowledge; 3) Propose the deep transfer method with seed replacement to construct a sufficient and active regression task learning framework from three aspects: model architecture, data representation, and loss function.

[0037] 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; Among them, step S46 includes: S461. Calculate the gradient g t : ; In the formula, represents the parameter to be updated in the model, represents the parameter to be updated in the t round, represents the gradient operator, represents the loss function.

[0038] S462. Update the first-order moment estimate and the second-order moment estimate based on the gradient: ; In the formula, respectively represent the first-order and second-order moment decay coefficients, respectively represent the first-order moment estimate and the second-order moment estimate in the t round, m t-1 , v t-1 respectively represent the first-order moment estimate and the second-order moment estimate in the t -1 round.

[0039] S463. Perform bias correction on the updated first-order moment estimate and second-order moment estimate: ; In the formula, , respectively represent the bias corrections of the first-order moment estimate and the second-order moment estimate in the t round.

[0040] S464. Update the parameters of the initial convolutional neural network model based on the bias-corrected first-order moment estimate and second-order moment estimate to obtain an updated model: ; In the formula, represents the parameter of the updated initial convolutional neural network model, is the learning rate, is a constant; Specifically, the constant here is generally a very small constant used to prevent division-by-zero errors, and generally takes 10 -8, meanwhile, the above update process is specifically the process optimized by 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 moment (mean) and second moment (variance) of the gradient.

[0041] S465. Update the second optimized data set based on the updated model to obtain optimized data.

[0042] S5. Determine the running state of the train rolling bearing based on the interval between the optimized data and the modalized data.

[0043] Specifically, step S5 is specifically as follows: Compare the interval between the optimized data and the modalized data using the modal interval size comparison rule, and take the running state corresponding to the encoding of the maximum value of the modal interval as the running state of the train rolling bearing; Meanwhile, in actual bearing fault diagnosis, the average accuracy of the deep seed replacement model on 12 migration tasks is 98.27%, which is significantly better than the comparative experiment and the accuracy difference is small. Thus, it can be seen that the model has good stability.

[0044] The train rolling bearing state recognition method provided in Embodiment 1 of the present invention. Based on the deep transfer regression method of seed replacement, a balanced distribution adaptation loss function for regression tasks is constructed to achieve full learning of domain knowledge. Through the seed replacement strategy and combined with clustering technology, active representation and learning of domain knowledge are realized, thereby improving the generalization performance of the model and solving the problem of large data demand of the model.

[0045] Embodiment 2 As Figure 2 shown, Embodiment 2 of the present invention provides a train rolling bearing state recognition system, and the system includes: Processing module 1, configured to obtain monitoring data of the train rolling bearing, and preprocess the monitoring data to obtain processed data; Balancing module 2, configured to perform data balancing processing on the processed data to obtain a balanced data set; 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 modalized data; Optimization module 4, configured to construct an initial model of deep transfer with seed replacement, input the first data set into the initial model of deep transfer with seed replacement, and perform optimization processing on the initial model of deep transfer with seed replacement to output optimized data; State module 5, configured to determine the running state of the train rolling bearing based on the interval between the optimized data and the modalized data; The processing module 1 includes: A window sub-module, configured to determine a window length, an overlap rate, and a sliding start position, and divide the monitoring data into a plurality of windows with a fixed length based on the window length and the sliding start position; An overlap sub-module, configured to overlap two adjacent windows based on the overlap rate to obtain extended data; A denoising sub-module, configured to obtain a trained convolutional neural network, and input the extended data into the trained convolutional neural network for denoising processing to obtain processed data.

[0046] The balancing module 2 includes: A source domain sub-module, configured to determine source domain data D S : ; Wherein, is the i th source domain sample feature vector, is corresponding label, m is the number of source domain samples; A target domain sub-module, configured to determine target domain data D T : ; Wherein, is the i th labeled target domain feature vector, is corresponding label, n is the number of labeled target domain samples, is the i th unlabeled target domain feature vector, u is the number of unlabeled target domain samples; A balancing sub-module, 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: ; Wherein, λ is a balance factor, D MDA is the empirical average value, D CDA is the conditional distribution adaptation, D BDA is the balanced distribution adaptation, X S is the source domain sample feature vector, X TIs the feature vector of the target domain sample.

[0047] The modality module 3 includes: An intervalization sub-module, configured to perform modality intervalization processing on the second data set based on an interval feature quantization formula: ; In the formula, X k ( n ) represents the modality interval form of the data set corresponding to the k th modality component, x k ( n ), x k ( n -1), x k ( n +1) represents the k th data in the data set corresponding to the modality component, the n, n -1, n +1 data; A conversion sub-module, configured to convert the modality interval form of the second data set to obtain modality data: X ( n ) = x ( n ), ( n )]; In the formula, x ( n ) represents the lower bound of the data set, ( n ) represents the upper bound of the data set.

[0048] The optimization module 4 includes: A freezing sub-module, configured to build an initial convolutional neural network model, freeze the shallow network to retain the general feature extraction ability, transfer the deep parameters, and use the source domain data of the initial convolutional neural network model as the initial parameters of the target domain; A clustering sub-module, configured to extract high-dimensional features of the source domain data using the frozen shallow network, and perform clustering processing on the high-dimensional features of the source domain data to obtain several clustering clusters; A substitution sub-module, 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 use the data within the clustering clusters to replace the target domain label samples to obtain a first optimized data set; A loss sub-module, configured to determine a fusion loss function L : ; In the formula, 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; The first update sub-module is used to update the first optimization dataset by minimizing the fusion loss function L , and to obtain a second optimization dataset; The second update sub-module is used to update the parameters of the initial convolutional neural network model to obtain an updated model, and to update the second optimization dataset based on the updated model to obtain optimized data.

[0049] The second update sub-module includes: The gradient unit is used to calculate the gradient g t : ; In the formula, represents the parameter to be updated by the model, represents the t round of parameter to be updated, represents the gradient operator, represents the loss function; The estimation unit is used to update the first-order moment estimation and second-order moment estimation based on the gradient: ; In the formula, represent the first-order and second-order moment decay coefficients respectively, represent the first-order moment estimation and second-order moment estimation of the t round respectively, m t-1 , v t-1 represent the first-order moment estimation and second-order moment estimation of the t -1 round respectively; The bias unit is used to correct the bias of the updated first-order moment estimation and second-order moment estimation: ; In the formula, , represent the bias corrections of the first-order moment estimation and second-order moment estimation of the t round respectively; A parameter unit, configured to update parameters of the initial convolutional neural network model based on the first-order moment estimation and the second-order moment estimation after deviation correction, so as to obtain an updated model: ; In the formula, represents the parameters of the updated initial convolutional neural network model, is the learning rate, is a constant; Update the second optimization data set based on the updated model to obtain optimized data.

[0050] The state module 5 is specifically configured to: Compare the optimized data with the interval of the modalized data by using the modal interval size comparison rule, and use the operating state corresponding to the encoding corresponding to the maximum value of the modal interval as the operating state of the train rolling bearing.

[0051] In some other embodiments of the present invention, the present invention provides the following technical solution. A computer includes a memory 102, a processor 101, and a computer program stored on the memory 102 and executable on the processor 101. When the processor 101 executes the computer program, the method for identifying the state of the train rolling bearing as described above is implemented.

[0052] Specifically, the above-mentioned processor 101 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC for short), or one or more integrated circuits configured to implement the embodiments of the present invention.

[0053] Among them, the memory 102 may include a mass storage for data or instructions. By way of example and not limitation, the memory 102 may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disc, a magneto-optical disc, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. In suitable cases, the memory 102 may include removable or non-removable (or fixed) media. In suitable cases, the memory 102 may be internal or external to the data processing device. In a particular embodiment, the memory 102 is a non-volatile memory. In a particular embodiment, the memory 102 includes a read-only memory (ROM) and a random access memory (RAM). In suitable cases, 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, or a combination of two or more of these. In suitable cases, the RAM may be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM may be a fast page mode dynamic random access memory (FPMDRAM), an extended data out dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0054] The memory 102 can be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 101.

[0055] The processor 101 reads and executes the computer program instructions stored in the memory 102 to implement the above-mentioned train rolling bearing state recognition method.

[0056] In some embodiments, the computer may further include a communication interface 103 and a bus 100. Among them, as Figure 3 shown, the processor 101, the memory 102, and the communication interface 103 are connected through the bus 100 and complete communication with each other.

[0057] The communication interface 103 is used to implement communication between various modules, devices, units, and / or devices 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 devices, databases, external storage, and image / data processing workstations.

[0058] 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: Data Bus, Address Bus, Control Bus, Expansion Bus, 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 bus or a combination of two or more of these. In suitable cases, Bus 100 may include one or more buses. Although embodiments of the present invention describe and illustrate specific buses, the present invention contemplates any suitable bus or interconnect.

[0059] The computer can execute the train rolling bearing state recognition method of the present invention based on the obtained train rolling bearing state recognition system, thereby realizing the train rolling bearing state recognition.

[0060] In some further embodiments of the present invention, in combination with the above train rolling bearing state recognition method, embodiments of the present invention provide the following technical solution: a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above train rolling bearing state recognition method is realized.

[0061] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in combination with an instruction execution system, apparatus, or device.

[0062] More specific examples (non-exhaustive list) of readable media include the following: an electrical connection part with one or more wirings (electronic device), a portable computer disk cartridge (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or other suitable processing if necessary, and then storing it in a computer memory.

[0063] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well-known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application-specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0064] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as within the scope described in this specification.

[0065] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A method for identifying the state of a train rolling bearing, characterized in that, Including: Obtain the monitoring data of the train rolling bearing, and preprocess the monitoring data to obtain processed data; Perform data equalization processing on the processed data to obtain an equalized data set; Divide the equalized data set into a first data set and a second data set, and perform modal interval processing on the second data set to obtain modalized data; Construct a seed replacement deep transfer initial model, input the first data set into the seed replacement deep transfer initial model, and perform optimization processing on the seed replacement deep transfer initial model to output optimized data; Determine the operating state of the train rolling bearing based on the interval between the optimized data and the modalized data.

2. The method for identifying the state of a train rolling bearing according to claim 1, characterized in that, The step of preprocessing the monitoring data to obtain processed data includes: Determine the window length, overlap rate, and sliding start position, and divide the monitoring data into several windows of fixed length based on the window length and the sliding start position; Overlap two adjacent windows based on the overlap rate to obtain augmented data; Obtain a trained convolutional neural network, input the augmented data into the trained convolutional neural network for denoising processing to obtain processed data.

3. The method for identifying the state of a train rolling bearing according to claim 1, characterized in that, The step of performing data equalization processing on the processed data to obtain an equalized data set includes: Determine the source domain data D S : ； In the formula, is the i th source domain sample feature vector, is the corresponding label, m is the number of source domain samples; Determine the target domain data D T : ； Wherein, is the i th labeled target domain feature vector, is the corresponding label, n is the number of labeled target domain samples, is the i th unlabeled target domain feature vector, u is the number of unlabeled target domain samples; Determine a distribution difference equalization function, and perform equalization processing on the processed data based on the distribution difference equalization function to obtain an equalized data set, where the distribution difference equalization function is: ； In the formula, λ is the balance factor, D MDA is the empirical average value, D CDA is the conditional distribution adaptation, D BDA is the balance distribution adaptation, X S is the source domain sample feature vector, X T is the target domain sample feature vector.

4. The method for identifying the state of a train rolling bearing according to claim 1, characterized in that, The step of performing modal interval processing on the second data set to obtain modalized data includes: Perform modal interval processing on the second data set based on the interval feature quantization formula: ； In the formula, X k ( n ) represents the modal interval form of the dataset corresponding to the k th modal component, x k ( n ), x k ( n -1), x k ( n +1) represent the k th data in the dataset corresponding to the n, n -1, n +1 data of the modal component; Convert the modal interval form of the second data set to obtain modalized data: X ( n ) =[ x ( n )， ( n )]； In the formula, x ( n ) represents the lower bound of the data set, ( n ) represents the upper bound of the data set.

5. The method for identifying the state of a train rolling bearing according to claim 1, wherein The step of constructing a seed replacement deep transfer initial model, inputting the first data set into the seed replacement deep transfer initial model, and performing optimization processing on the seed replacement deep transfer initial model to output optimized data includes: Construct an initial convolutional neural network model, freeze the shallow network to retain the general feature extraction ability, transfer the deep parameters, and use the source domain data of the initial convolutional neural network model as the initial parameters of the target domain; Use the frozen shallow network to extract the high-dimensional features of the source domain data, and perform clustering processing on the high-dimensional features of the source domain data to obtain several clustering clusters; 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 use the data within the clustering clusters to replace the target domain label samples to obtain a first optimized data set; Determine the fusion loss function L : ； Wherein, 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; 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.

6. The method for identifying the state of a train rolling bearing according to claim 5, characterized in that, The step 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 includes: Calculate the gradient g t : ； In the formula, represents the parameter to be updated by the model, represents the t parameter to be updated in the round, and represents the loss function; Update the first-order moment estimate and the second-order moment estimate based on the gradient: ； wherein, respectively represent the first-order and second-order moment decay coefficients, respectively represent the first-order moment estimate and the second-order moment estimate of the t th round, m t-1 and v t-1 respectively represent the first-order moment estimate and the second-order moment estimate of the t (t-1)th round; Perform bias correction on the updated first - moment estimate and second - moment estimate: ； In the formula, and respectively represent the bias correction of the first moment estimate and the second moment estimate in the t round. Update the parameters of the initial convolutional neural network model based on the bias - corrected first - moment estimate and second - moment estimate to obtain an updated model: ； In the formula, represents the parameters of the updated initial convolutional neural network model, is the learning rate, is a constant; Update the second optimization dataset based on the updated model to obtain optimized data.

7. The method for identifying the state of a train rolling bearing according to claim 1, characterized in that The step of determining the operating state of the train rolling bearing based on the interval between the optimized data and the modalized data includes: Compare the interval between the optimized data and the modalized data using the modal interval size comparison rule, and take the operating state corresponding to the encoding of the maximum value of the modal interval as the operating state of the train rolling bearing.

8. A train rolling bearing state recognition system, characterized in that, The system includes: A processing module for acquiring monitoring data of the train rolling bearing and pre - processing the monitoring data to obtain processed data; An equalization module for performing data equalization processing on the processed data to obtain an equalized dataset; A modal module for dividing the equalized dataset into a first dataset and a second dataset, and performing modal interval processing on the second dataset to obtain modalized data; An optimization module for constructing a seed - replacement deep transfer initial model, inputting the first dataset into the seed - replacement deep transfer initial model, and performing optimization processing on the seed - replacement deep transfer initial model to output optimized data; A state module for determining the operating state of the train rolling bearing based on the interval between the optimized data and the modalized data.

9. A computer, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the train rolling bearing state recognition method according to any one of claims 1 to 7.

10. A storage medium, characterized in that, A computer program is stored on the storage medium, and when the computer program is executed by the processor, it implements the train rolling bearing state recognition method according to any one of claims 1 to 7.

Images