Intelligent monitoring and fault diagnosis integrated RV reducer control system and method

Abstract

The invention relates to the technical field of RV speed reducers, in particular to an intelligent monitoring and fault diagnosis integrated RV speed reducer control system and method. Collecting multi-mode operation data of the RV speed reducer, wherein the multi-mode operation data comprises vibration, temperature, sound and current data; feature extraction is carried out on the multi-modal operation data, an adaptive feature fusion method based on an attention mechanism is designed, weights of different features are dynamically adjusted, and comprehensive and intelligent state monitoring is achieved; constructing a fault diagnosis model based on a multi-branch convolutional neural network, and performing supervised training on the model by using historical operation data and fault labels to realize high-precision fault classification; a knowledge distillation technology is adopted to compress the trained model, the model is deployed into the RV reducer, and online real-time diagnosis is achieved; according to the method, the value of multi-source heterogeneous data is fully mined, and the intelligent level of RV reducer fault diagnosis is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of RV reducers, and particularly to an RV reducer control system and method integrating intelligent monitoring and fault diagnosis. Background Art

[0002] An RV reducer is a high-precision transmission device applied in fields such as industrial robots, numerically controlled machine tools, and automated production lines. Through a special internal gear structure and the principle of multi-tooth cross meshing, it achieves excellent performance such as high transmission ratio, high rigidity, high efficiency, low noise, and low vibration, and has an irreplaceable position in the field of precision transmission. The reliable operation of the RV reducer is directly related to the performance and efficiency of the entire electromechanical system. Therefore, effective condition monitoring and fault diagnosis of it are crucial.

[0003] With the rapid development of industrial automation and intelligent manufacturing, the manufacturing industry has put forward higher requirements for the reliability and operating efficiency of equipment. As a key functional component, the health status of the RV reducer directly affects the stability of the entire production line and the product quality. Traditional RV reducer monitoring methods mainly rely on regular manual inspections and empirical judgments, and there are many deficiencies. First of all, this method usually only focuses on a single parameter, such as the vibration signal, and it is difficult to comprehensively reflect the overall health status of the reducer. Although the vibration signal is an important basis for diagnosing the faults of the reducer, it is impossible to accurately identify the fault type and severity based on a single parameter alone. The changes in other parameters such as temperature, noise, and current also contain rich state information and need to be comprehensively analyzed.

[0004] Moreover, due to the complex structure and variable working conditions of the RV reducer, the traditional method adopts an offline analysis method and mainly relies on expert experience to judge the cause of the fault and take maintenance measures. This diagnosis mode has problems such as strong subjectivity, low accuracy, and long cycle. Especially when facing complex faults, non-expert personnel are difficult to make correct judgments in a timely manner, and the time lag for problem discovery and handling is relatively large, making it difficult to avoid unexpected shutdowns of the RV reducer and production losses. With the improvement of equipment complexity and the emergence of new types of faults, the diagnosis method relying solely on experience has been difficult to meet the actual needs.

[0005] The patent with the publication number CN113405795A discloses a method for identifying weak faults in a joint RV reducer. This method synchronously collects the servo motor current signal of the robot joint and the vibration signal of the operating state of the joint RV reducer; uses the time point corresponding to the maximum rotation frequency in the time-frequency diagram of the joint current signal as the starting time, finds and intercepts the vibration signal in the peak steady state stage in the filtered vibration signal, and through cepstrum analysis, when there are spectral lines with a sharp increase in amplitude in the cepstrum, it is determined that there is a fault in the joint RV reducer. After determining whether there is a fault, variational mode decomposition analysis is performed on the vibration signal with a fault to obtain the component signals of the vibration signal. Using information entropy as the selection index, a reconstructed signal is selected from the component signals, and Fourier transform is performed on the reconstructed signal to obtain the fault signal frequency spectrum diagram, and the fault characteristics are extracted from the diagram to identify the fault type of the weak fault. The method of the prior art only uses two signals, vibration and current, the diagnosis process is complex, lacks the comprehensive utilization of multi-modal data, is difficult to fully excavate the fault characteristics, and only targets the early fault diagnosis of the RV reducer.

[0006] In view of this, the present invention proposes an RV reducer control system and method integrating intelligent monitoring and fault diagnosis. Summary of the Invention

[0007] To achieve the above object, the present invention provides an RV reducer control system and method integrating intelligent monitoring and fault diagnosis, and an RV reducer control method integrating intelligent monitoring and fault diagnosis. The technical solution is as follows:

[0008] Collect multi-modal operation data of the RV reducer, and the operation data includes vibration data, temperature data, sound data, and current data;

[0009] Extract features from the collected operation data of the RV reducer, and extract vibration data features, temperature data features, sound data features, and current data features in the operation data;

[0010] For heterogeneous data collected by different sensors, after extracting the features of different types of operation data, design an adaptive feature data fusion method based on the attention mechanism to dynamically adjust the weights of different types of feature data to achieve intelligent monitoring;

[0011] Construct a fault diagnosis model based on a convolutional neural network. The convolutional neural network adopts a multi-branch neural network structure, and each branch corresponds to a type of sensor data. The fault diagnosis model is used for data fusion and fault classification of the RV reducer;

[0012] Train the fault diagnosis model in a supervised manner, train the fault diagnosis model based on the historical operation data of the RV reducer and the set fault labels, perform knowledge distillation on the trained deep learning model to compress the model volume, and deploy it to the RV reducer.

[0013] Preferably, vibration, temperature, sound and current sensors are installed on the RV reducer to collect the operation data of the RV reducer in real time;

[0014] Preprocess the collected operation data, and the preprocessing includes removing outliers, data denoising and data normalization; the outlier removal includes: using criteria to remove outliers; the data denoising includes: using wavelet transform for data denoising; the data normalization includes: using the maximum and minimum values for data normalization;

[0015] Through the time synchronization algorithm, align the operation data collected by different sensors to a unified timestamp to form a synchronized multi-modal data stream; the time synchronization algorithm includes timestamp alignment and data splicing.

[0016] Preferably, extract the vibration data characteristics, temperature data characteristics, sound data characteristics and current data characteristics from the operation data; let the vibration data be ; the temperature data is ; the sound data is ; the current data is ;

[0017] Extract the vibration data characteristics, and the vibration data characteristics include the time-domain characteristics, frequency-domain characteristics and time-frequency domain characteristics of the vibration data; the time-domain characteristics of the vibration data include the root mean square value of the vibration data ; the frequency-domain characteristics of the vibration data include: performing a fast Fourier transform on the vibration signal to obtain the frequency spectrum of the vibration data ; extract the amplitude of the vibration data spectrum ; the time-frequency domain characteristics of the vibration data are ; extract the temperature data characteristics, and the temperature data characteristics include the temperature mean , temperature gradient and temperature fluctuation ; extract the sound data characteristics, and the sound data characteristics include the time-domain characteristics and frequency-domain characteristics of the sound data, and the time-domain characteristics of the sound data include the root mean square value of the sound data ; the frequency-domain characteristics of the sound data include: performing a fast Fourier transform on the sound signal to obtain the frequency spectrum of the sound data ; extract the main frequency component of the sound data ;

[0018] Extract the current data characteristics, and the current data characteristics include the time-domain characteristics and frequency-domain characteristics of the current data characteristics; the time-domain characteristics of the current data characteristics include the current mean , root mean square value of current ; The frequency domain features of the current data features include: performing a fast Fourier transform on the current data to obtain the frequency spectrum of the current data , and extracting the fundamental wave component ;

[0019] Fuse the extracted feature data:

[0020] ;

[0021] Among them, is the feature vector after feature data fusion.

[0022] Preferably, after extracting the data features of different types of operating data, adaptively allocate the weights of different modal features by learning the correlation between the data; use the fused feature vector as the input of the attention mechanism;

[0023] Let , , represent the query matrix, key matrix, and value matrix respectively, where is the dimension of the fused feature vector , , , are the dimensions of the query, key, and value respectively; R （*） represents the set of real numbers; the calculation process of the attention mechanism is as follows:

[0024] ;

[0025] Among them, , , are learnable weight matrices; calculate the similarity score between the query and the key: ; Among them, is the similarity matrix; is the transpose of the key matrix; perform softmax normalization on the similarity matrix to obtain the attention weights: ; Among them, is the attention weight matrix;

[0026] Multiply the attention weights by the value matrix to obtain the weighted fused feature: ; Among them, is the fused feature matrix.

[0027] Preferably, construct a fault diagnosis model based on a convolutional neural network. The fault diagnosis model is constructed based on a convolutional neural network with 4 parallel branches, and each branch corresponds to a type of sensor data; use the feature matrix after fusing the attention mechanism Divided into 4 sub - matrices , , , , representing the eigen - sub - matrices corresponding to vibration, temperature, sound, and current respectively; each sub - matrix serves as the input for the corresponding branch;

[0028] For the th branch, , let represent the corresponding eigen - sub - matrix, where is the eigen - dimension, is the time step; reshape into the form of , where and are the height and width respectively, satisfying ; perform convolution and pooling operations on :

[0029] ;

[0030] ;

[0031] where, and are the weights and biases of the th branch, the th convolutional layer, represents the convolution operation, is the number of convolutional layers of the th branch, is the activation function, is the max - pooling operation; is the convolution of the th branch, the th convolutional layer; is the pooling of the th branch, the th convolutional layer;

[0032] Flatten the pooling result of each branch into a vector , and concatenate them to form a feature vector ; input into the fully - connected layer for feature fusion and classification:

[0033] ;

[0034] ;

[0035] where, , , , are the weights and biases of the fully connected layer, , is the output probability vector, is the number of fault categories; The loss function of the fault diagnosis model adopts the cross-entropy function:

[0036] ;

[0037] where, is the one-hot encoding of the true label, is the predicted probability, and c is the fault category index number.

[0038] Preferably, the fault diagnosis model is trained as follows:

[0039] Collect historical operation data of the RV reducer, including vibration, temperature, sound and current data, as well as the corresponding fault labels; Divide the historical operation data into training set, validation set and test set, and preprocess the historical operation data, including removing outliers, data denoising and data normalization; Extract the features of various historical operation data and perform adaptive feature fusion through the attention mechanism;

[0040] Construct a fault diagnosis model based on a convolutional neural network, with each branch of the convolutional neural network corresponding to a type of sensor data, and the input is the feature submatrix fused by the attention mechanism; In each branch of the convolutional network, perform convolution, activation and pooling operations on the input data to extract high-level features; Flatten and splice the features of each branch, and then perform feature fusion and fault classification through the fully connected layer;

[0041] Define the cross-entropy loss function and optimizer of the fault diagnosis model, and set the number of training epochs, batch size and learning rate; Perform model training on the training set, update the model parameters through the backpropagation algorithm, and minimize the loss function; After each training cycle ends, evaluate the model performance on the validation set. If the current loss function value is less than the loss function value of the previous training cycle, save the current model;

[0042] If the model performance on the validation set does not improve for several consecutive training cycles, reduce the learning rate and trigger the early stopping strategy.

[0043] Preferably, construct a teacher model and a student model, use the trained deep learning model as the teacher model, and define the teacher model parameters as , and the output is the soft label ; Construct a student model, and define the student model parameters as , and the output is ;

[0044] Define the distillation loss function. Let represent the true label of the -th sample, represent the soft label output of the teacher model, represent the output of the student model on the sample; the distillation loss function is defined as:

[0045] ;

[0046] where is the weight coefficient of the distillation process; is the temperature parameter of the soft label, which is used to control the smoothness of the soft label; is the cross-entropy loss function;

[0047] Train the student model, fix the parameters of the teacher model , and train the student model to minimize the distillation loss function; each training step is as follows:

[0048] ;

[0049] where is the learning rate, is the training batch size; repeat the above training steps until the distillation loss function converges;

[0050] Integrate the student model after knowledge distillation into the embedded system of the RV reducer for operation monitoring and fault diagnosis of the RV reducer.

[0051] An integrated intelligent monitoring and fault diagnosis RV reducer control system, which is used to implement the integrated intelligent monitoring and fault diagnosis RV reducer control method described above, includes: a data acquisition module, a feature extraction module, a data fusion module, a fault diagnosis module, and a distillation deployment module;

[0052] The data acquisition module is used to collect multi-modal operation data of the RV reducer, and the operation data includes vibration data, temperature data, sound data, and current data;

[0053] The feature extraction module extracts features from the collected operation data of the RV reducer, and extracts vibration data features, temperature data features, sound data features, and current data features in the operation data;

[0054] The data fusion module designs an adaptive feature data fusion method based on the attention mechanism for heterogeneous data collected by different sensors. After extracting the features of different types of operation data, it dynamically adjusts the weights of different types of feature data to achieve intelligent monitoring;

[0055] The fault diagnosis module is used to construct a fault diagnosis model based on a convolutional neural network. The convolutional neural network adopts a multi-branch neural network structure, and each branch corresponds to a type of sensor data. The fault diagnosis model is used to perform data fusion and fault classification on the RV reducer.

[0056] The distillation and deployment module trains the fault diagnosis model in a supervised manner. Based on the historical operation data of the RV reducer and the set fault labels, the fault diagnosis model is trained. Knowledge distillation is performed on the trained deep learning model to compress the model volume and deploy it into the RV reducer.

[0057] An electronic device includes: a processor and a memory. Among them, a computer program that can be called by the processor is stored in the memory. The processor executes the RV reducer control method for integrated intelligent monitoring and fault diagnosis by calling the computer program stored in the memory.

[0058] A computer-readable storage medium stores instructions. When the instructions run on a computer, the computer is made to execute the RV reducer control method for integrated intelligent monitoring and fault diagnosis.

[0059] The beneficial effects of the present invention:

[0060] The present invention comprehensively obtains the device operation state information by collecting vibration, temperature, sound, and current data of the RV reducer through multi-modal, providing a data basis for subsequent intelligent monitoring and fault diagnosis.

[0061] The present invention extracts features from the collected multi-source heterogeneous data, mines the fault-related information contained in the data, reduces data redundancy, and improves data quality.

[0062] The present invention designs an adaptive feature data fusion method based on the attention mechanism, automatically learns the importance weights of different features, dynamically adjusts the fusion strategy, improves the representation ability of multi-modal data, and realizes device intelligent monitoring.

[0063] The present invention constructs a multi-branch fault diagnosis model based on a convolutional neural network, makes full use of the complementary information of multi-sensor data, and realizes feature fusion and fault classification through end-to-end learning, improving the diagnosis accuracy.

[0064] The present invention trains the fault diagnosis model in a supervised manner, uses the historical operation data and fault labels of the RV reducer to learn the internal pattern of fault data, and enables the model to have the ability of fault diagnosis.

[0065] The present invention performs knowledge distillation on the trained deep learning model, compresses the model volume while maintaining the diagnostic performance, and facilitates deployment into the resource-constrained RV reducer, realizing intelligent monitoring and fault diagnosis at the device end of the RV reducer. Brief Description of the Drawings

[0066] Figure 1 It is a flow chart of the control method for the intelligent monitoring and fault diagnosis integrated RV reducer provided by the present invention;

[0067] Figure 2 It is a structural diagram of the intelligent monitoring and fault diagnosis integrated RV reducer control system provided by the present invention. Detailed Embodiments

[0068] For a better understanding of the present invention, more detailed descriptions of various aspects of the present invention will be made with reference to the accompanying drawings. It should be understood that these detailed descriptions are only descriptions of the exemplary embodiments of the present invention, and do not limit the scope of the present invention in any way. Throughout the specification, the same reference numerals refer to the same elements. The expression "and / or" includes any and all combinations of one or more of the associated listed items.

[0069] In the drawings, for ease of illustration, the sizes, dimensions, and shapes of the elements have been slightly adjusted. The drawings are only examples and are not drawn to an exact scale. As used herein, terms such as "substantially", "about", and similar terms are used as terms indicating approximation, rather than terms indicating degree, and are intended to account for the inherent deviations in measured or calculated values that would be recognized by a person of ordinary skill in the art. Additionally, in the present invention, the order of description of the various step processes does not necessarily represent the order in which these processes occur in actual operation, unless there is a clear other limitation or can be deduced from the context.

[0070] It should also be understood that expressions such as "including", "comprising", "having", "containing", and / or "comprising of" are open-ended rather than closed-ended expressions in this specification, which means that there are the stated features, elements, and / or components, but do not exclude the existence of one or more other features, elements, components, and / or their combinations. In addition, when an expression such as "at least one of..." appears after a list of listed features, it modifies the entire list of features, rather than just individual elements in the list. In addition, when describing the embodiments of the present invention, the use of "may" means "one or more embodiments of the present invention". And the term "exemplary" is intended to refer to an example or illustration.

[0071] Unless otherwise defined, all terms used herein (including engineering terms and scientific and technical terms) have the same meaning as commonly understood by a person of ordinary skill in the art to which the present invention pertains. It should also be understood that, unless there is a clear statement in the present invention, words defined in a common dictionary should be interpreted as having a meaning consistent with their meaning in the context of the related art, and should not be interpreted in an idealized or overly formal sense.

[0072] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.

[0073] Embodiment 1

[0074] Refer to Figure 1 , which is the first embodiment of the present invention, providing a control method for an RV reducer integrating intelligent monitoring and fault diagnosis.

[0075] S1: Collect multi-modal operation data of the RV reducer, where the operation data includes vibration data, temperature data, sound data, and current data, and preprocess the collected data.

[0076] Install vibration, temperature, sound, and current sensors on the RV reducer to collect the operation data of the RV reducer in real time.

[0077] Preprocess the collected operation data, and the preprocessing includes removing outliers, data denoising, and data normalization; the removing of outliers includes: using the criterion to remove outliers; the data denoising includes: using wavelet transform for data denoising; the data normalization includes: using the maximum and minimum values for data normalization.

[0078] Through a time synchronization algorithm, align the operation data collected by different sensors to a unified timestamp to form a synchronized multi-modal data stream; the time synchronization algorithm includes timestamp alignment and data splicing.

[0079] By installing a variety of sensors on the RV reducer and performing data collection and preprocessing, high-quality multi-modal monitoring data is obtained. Methods such as outlier removal, data denoising, and normalization are used to improve the data quality. The use of the time synchronization algorithm realizes the synchronization of data from different sensors, laying a foundation for subsequent feature extraction and fusion.

[0080] S2: Extract features from the collected operation data of the RV reducer, and extract vibration data features, temperature data features, sound data features, and current data features in the operation data.

[0081] Extract vibration data features, temperature data features, sound data features, and current data features in the operation data; let the vibration data be ; the temperature data be ; the sound data be ; the current data be , where n is a parameter.

[0082] Extract the vibration data features, where the vibration data features include the time-domain features, frequency-domain features, and time-frequency domain features of the vibration data; the time-domain features of the vibration data include the root mean square value of the vibration data : . Among them, represents the number of sampling points of the vibration data;

[0083] The frequency-domain features of the vibration data include: performing a fast Fourier transform on the vibration signal to obtain the frequency spectrum of the vibration data : ; extracting the amplitude of the vibration data frequency spectrum : , frequency , where is the sampling frequency of the vibration data.

[0084] The time-frequency domain features of the vibration data are including: using the short-time Fourier transform to segment the vibration signal and perform a fast Fourier transform on each segment to obtain the time-frequency domain features of the vibration signal , is the integration variable, representing the time offset of the time-domain signal; t represents the time variable, represents the time window function, f represents the frequency variable, and j is the imaginary unit.

[0085] Extract the temperature data features, where the temperature data features include the temperature mean, temperature gradient, and temperature fluctuation; the temperature mean : ; represents the number of sampling points of the temperature data; the temperature gradient : , where is the time interval between adjacent temperature sampling points; the temperature fluctuation : .

[0086] Extract the sound data features, where the sound data features include the time-domain features and frequency-domain features of the sound data, and the time-domain features of the sound data include the root mean square value of the sound data : ; represents the number of sampling points of the sound data; the frequency-domain features of the sound data include: performing a fast Fourier transform on the sound signal to obtain the frequency spectrum of the sound data : ; extracting the main frequency component , that is, the frequency corresponding to the maximum amplitude of the sound data frequency spectrum.

[0087] Extract the current data features, where the current data features include the time-domain features of the current data features and the frequency-domain features of the current signal; the time-domain features of the current data features include the current mean value : ; represents the number of sampling points of the current data; the root mean square value of the current : ; the frequency-domain features of the current data features include performing a fast Fourier transform on the current signal to obtain the frequency spectrum of the current signal : ; extract the fundamental wave component , where corresponds to the fundamental wave frequency.

[0088] Fuse the extracted feature data:

[0089] ;

[0090] where is the feature vector after feature data fusion.

[0091] Extract the features of the RV reducer operation data, and mine the fault feature information in the multi-modal data. Extract the time-domain, frequency-domain, and time-frequency domain features of the vibration data, the mean value, gradient, and fluctuation features of the temperature data, the time-domain and frequency-domain features of the sound data, and the time-domain and frequency-domain features of the current data. Convert the original data into a more discriminative feature representation.

[0092] S3: For the heterogeneous data collected by different sensors, after extracting the features of different types of operation data, design an adaptive feature data fusion method based on the attention mechanism to dynamically adjust the weights of different types of feature data and achieve intelligent monitoring.

[0093] After extracting the data features of different types of operation data, adaptively allocate the weights of different modal features by learning the correlation between the data; use the fused feature vector as the input of the attention mechanism.

[0094] Let , , represent the query matrix, key matrix, and value matrix respectively, where is the dimension of the fused feature vector , and R （*） represents the set of real numbers; , , are the dimensions of the query, key, and value respectively; the calculation process of the attention mechanism is as follows:

[0095] ;

[0096] Among them, 、 、 are learnable weight matrices; calculate the similarity score between the query and the key: ; among them, is the similarity matrix; is the transpose of the key matrix; perform softmax normalization on the similarity matrix to obtain the attention weights: ; among them, is the attention weight matrix.

[0097] Multiply the attention weights by the value matrix to obtain the weighted and fused features: ; among them, is the fused feature matrix.

[0098] An adaptive feature fusion method based on the attention mechanism is designed, which effectively fuses the feature information of heterogeneous data. The attention mechanism adaptively learns the importance weights of features and dynamically adjusts the fusion strategy. The attention mechanism calculates the similarity score, generates the attention weight matrix, and obtains the weighted and fused feature representation, improving the feature discriminability and diagnostic accuracy.

[0099] S4: Construct a fault diagnosis model based on a convolutional neural network. The convolutional neural network adopts a multi-branch neural network structure, and each branch corresponds to a type of sensor data. The fault diagnosis model is used for data fusion and fault classification of the RV reducer.

[0100] Construct a fault diagnosis model based on a convolutional neural network. The fault diagnosis model is constructed based on a convolutional neural network with 4 parallel branches, and each branch corresponds to a type of sensor data; the feature matrix after fusion by the attention mechanism is divided into 4 sub-matrices , , , , respectively representing the feature sub-matrices corresponding to vibration, temperature, sound, and current; each sub-matrix is used as the input of the corresponding branch.

[0101] For the th branch, , let represent the corresponding feature sub-matrix, where is the feature dimension, is the time step; reshape into the form of , where and are the height and width respectively, satisfying ; for Perform convolution and pooling operations:

[0102] ;

[0103] ;

[0104] Among them, and are the weights and biases of the th branch and the th convolutional layer respectively, represents the convolution operation, is the number of convolutional layers of the th branch, is the activation function, is the max pooling operation; is the convolution of the th branch and the th convolutional layer; is the pooling of the th branch and the th convolutional layer.

[0105] Flatten the pooling result of each branch into a vector , and splice them to form a feature vector ; are the outputs of the th pooling layer of the vibration branch, the outputs of the th pooling layer of the temperature branch, the outputs of the th pooling layer of the sound branch, and the outputs of the th pooling layer of the current branch respectively; Input into the fully connected layer for feature fusion and classification:

[0106] ;

[0107] ;

[0108] Among them, , , , are the weights and biases of the fully connected layer respectively, , is the output probability vector, is the number of fault categories; The loss function of the fault diagnosis model uses the cross-entropy function:

[0109] ;

[0110] Among them, is the one-hot encoding of the true label, where \(p\) is the predicted probability and \(c\) is the index number of the fault category.

[0111] By constructing a multi-branch fault diagnosis model based on a convolutional neural network, each branch corresponds to a type of sensor data. High-level features are extracted through convolutional, activation, and pooling operations to capture key information. The features of each branch are fused and classified through a fully connected layer. The complementary information of multi-sensor data is fully utilized, improving the diagnostic accuracy and efficiency.

[0112] S5: Train the fault diagnosis model in a supervised manner. Based on the historical operation data of the RV reducer and the set fault labels, train the fault diagnosis model.

[0113] Train the constructed fault diagnosis model. The training steps are as follows:

[0114] Collect the historical operation data of the RV reducer, including vibration, temperature, sound, and current data, as well as the corresponding fault labels; divide the historical operation data into a training set, a validation set, and a test set, and preprocess the historical operation data, including removing outliers, data denoising, and data normalization; extract the features of various historical operation data and perform adaptive feature fusion through an attention mechanism;

[0115] Construct a fault diagnosis model based on a convolutional neural network, where each branch of the convolutional neural network corresponds to a type of sensor data, and the input is the feature submatrix fused by the attention mechanism; in each branch of the convolutional network, perform convolutional, activation, and pooling operations on the input data to extract high-level features; flatten and concatenate the features of each branch, and then perform feature fusion and fault classification through a fully connected layer.

[0116] Define the cross-entropy loss function and optimizer of the fault diagnosis model, set the number of training epochs, batch size, and learning rate; train the model on the training set, update the model parameters through the backpropagation algorithm, and minimize the loss function; after each training cycle ends, evaluate the model performance on the validation set. If the current loss function value is less than the loss function value of the previous training cycle, save the current model.

[0117] If the model performance on the validation set does not improve for several consecutive training cycles, then reduce the learning rate and trigger the early stopping strategy.

[0118] By training the fault diagnosis model in a supervised manner and using the historical operation data and fault labels, the model learns the feature representation and classification boundary of the fault data and has the ability to accurately diagnose.

[0119] S6: Perform knowledge distillation on the trained deep learning model to compress the model volume and deploy it to the RV reducer.

[0120] Build a teacher model and a student model. Use the trained deep learning model as the teacher model, and define the parameters of the teacher model as , and the output is the soft label ; Build a student model, and define the parameters of the student model as , and the output is .

[0121] Define the distillation loss function, and let represent the true label of the th sample, represent the soft label output of the teacher model, represent the output of the student model on the sample; the distillation loss function is defined as:

[0122] ;

[0123] Among them, is the weight coefficient of the distillation process; is the temperature parameter of the soft label, which is used to control the smoothness of the soft label; is the cross-entropy loss function.

[0124] Train the student model, fix the parameters of the teacher model , and train the student model to minimize the distillation loss function; each training step is as follows:

[0125] ;

[0126] Among them, is the learning rate, is the training batch size; repeat the above training steps until the distillation loss function converges.

[0127] Integrate the student model after knowledge distillation into the embedded system of the RV reducer for operation monitoring and fault diagnosis of the RV reducer.

[0128] Perform knowledge distillation on the trained deep learning model to compress it into a student model with a smaller volume and more efficient calculation. Use the soft label of the teacher model to guide the learning of the student model. By minimizing the distillation loss function, the output of the student model is close to that of the teacher model; the student model inherits the knowledge of the teacher model and maintains high diagnostic performance while compressing the model. Integrate the distilled model into the RV reducer to achieve efficient and real-time monitoring and diagnosis.

[0129] Embodiment 2

[0130] Refer to Figure 2 , which is the second embodiment of the present invention, and provides an RV reducer control system integrating intelligent monitoring and fault diagnosis.

[0131] The system includes a data acquisition module, a feature extraction module, a data fusion module, a fault diagnosis module, and a distillation and deployment module.

[0132] The data acquisition module is used to collect multi-modal operation data of the RV reducer, and the operation data includes vibration data, temperature data, sound data, and current data.

[0133] The feature extraction module extracts features from the collected operation data of the RV reducer, and extracts vibration data features, temperature data features, sound data features, and current data features in the operation data.

[0134] For the heterogeneous data collected by different sensors, after extracting the features of different types of operation data, the data fusion module designs an adaptive feature data fusion method based on the attention mechanism, dynamically adjusts the weights of different types of feature data, and realizes intelligent monitoring.

[0135] The fault diagnosis module is used to construct a fault diagnosis model based on a convolutional neural network. The convolutional neural network adopts a multi-branch neural network structure, and each branch corresponds to a type of sensor data. The fault diagnosis model is used to perform data fusion and fault classification on the RV reducer.

[0136] The distillation and deployment module trains the fault diagnosis model in a supervised manner, trains the fault diagnosis model based on the historical operation data of the RV reducer and the set fault labels, performs knowledge distillation on the trained deep learning model, compresses the model volume, and deploys it to the RV reducer.

[0137] Embodiment 3

[0138] The present invention also provides an electronic device. The electronic device may include one or more processors and one or more memories. Among them, computer-readable code is stored in the memory, and when the computer-readable code is run by one or more processors, it can execute the above-mentioned integrated intelligent monitoring and fault diagnosis control method for the RV reducer.

[0139] The method or system according to the embodiment of the present invention can also be implemented by means of the architecture of the electronic device of the present invention.

[0140] The electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, input / output components, a hard disk, etc.

[0141] The storage device in the electronic device, such as ROM or hard disk, can store the integrated intelligent monitoring and fault diagnosis control method for the RV reducer provided by the present invention.

[0142] Intelligent monitoring and fault diagnosis integrated RV reducer control method, including: collecting RV reducer operation data in multiple modalities, where the operation data includes vibration data, temperature data, sound data, and current data; extracting features from the collected RV reducer operation data, extracting vibration data features, temperature data features, sound data features, and current data features in the operation data; for heterogeneous data collected by different sensors, after extracting features of different types of data, designing an adaptive feature data fusion method based on the attention mechanism to dynamically adjust the weights of different types of feature data to achieve intelligent monitoring; constructing a fault diagnosis model based on a convolutional neural network, the convolutional neural network adopts a multi-branch neural network structure, each branch corresponds to a type of sensor data, and the fault diagnosis model is used to perform data fusion and fault classification on the RV reducer; training the diagnosis model in a supervised manner, training the fault diagnosis model based on the RV reducer historical operation data and the set fault labels; performing knowledge distillation on the trained deep learning model to compress the model volume and deploy it to the RV reducer.

[0143] Further, the electronic device may further include a user interface. Of course, the architecture of the present invention is only exemplary. When implementing different devices, one or more components in the electronic device of the present invention may be omitted according to actual needs.

[0144] Embodiment 4

[0145] An embodiment of the present invention discloses a computer-readable storage medium.

[0146] Computer-readable instructions are stored on the computer-readable storage medium.

[0147] When the computer-readable instructions are run by a processor, the intelligent monitoring and fault diagnosis integrated RV reducer control method according to the embodiments of the present invention described with reference to the above drawings can be executed.

[0148] The storage medium includes but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory, etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. Additionally, according to the embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs.

[0149] For example, the present invention provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be run by a processor to execute instructions corresponding to the method steps provided by the present invention. For example: multi-modal collection of RV reducer operation data, where the operation data includes vibration data, temperature data, sound data, and current data; feature extraction of the collected RV reducer operation data, extracting vibration data features, temperature data features, sound data features, and current data features in the operation data; for heterogeneous data collected by different sensors, after extracting the features of different types of data, designing an adaptive feature data fusion method based on the attention mechanism to dynamically adjust the weights of different types of feature data to achieve intelligent monitoring; constructing a fault diagnosis model based on a convolutional neural network, where the convolutional neural network adopts a multi-branch neural network structure, and each branch corresponds to a type of sensor data, and the fault diagnosis model is used for data fusion and fault classification of the RV reducer; training the diagnosis model in a supervised manner, training the fault diagnosis model based on the historical operation data of the RV reducer and the set fault labels; performing knowledge distillation on the trained deep learning model to compress the model volume and deploy it to the RV reducer.

[0150] When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present invention are executed. The method and apparatus, device of the present invention may be implemented in many ways. For example, the method and apparatus, device of the present invention may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware.

[0151] The above order of steps for the method is only for illustration, and the steps of the method of the present invention are not limited to the above specifically described order unless otherwise specifically stated.

[0152] In addition, in some embodiments, the present invention may also be implemented as a program recorded in a recording medium, and these programs include machine-readable instructions for implementing the method according to the present invention. Thus, the present invention also covers a recording medium storing a program for executing the method according to the present invention.

[0153] In addition, parts of the above technical solutions provided in the embodiments of the present invention that are consistent with the corresponding technical solutions in the prior art in terms of implementation principles are not described in detail to avoid excessive elaboration.

[0154] As described above in the specific embodiments, the purpose, technical solutions, and beneficial effects of the present invention are further described in detail. It should be understood that the above is only the specific embodiments of the present invention and is not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A control method for an RV reducer integrating intelligent monitoring and fault diagnosis, characterized in that, Including: Collecting multi-modal operation data of the RV reducer, where the operation data includes vibration data, temperature data, sound data, and current data; Extracting features from the collected operation data of the RV reducer, extracting vibration data features, temperature data features, sound data features, and current data features in the operation data; For heterogeneous data collected by different sensors, after extracting the features of different types of operation data, designing an adaptive feature data fusion method based on the attention mechanism to dynamically adjust the weights of different types of feature data and achieve intelligent monitoring; Constructing a fault diagnosis model based on a convolutional neural network. The convolutional neural network adopts a multi-branch neural network structure, and each branch corresponds to a type of sensor data. The fault diagnosis model is used for data fusion and fault classification of the RV reducer; The fault diagnosis model is constructed based on a convolutional neural network with 4 parallel branches, and each branch corresponds to a type of sensor data; the feature matrix after fusing the attention mechanism is divided into 4 sub-matrices , , , , which respectively represent the feature sub-matrices corresponding to vibration, temperature, sound, and current; each sub-matrix serves as the input of the corresponding branch; Training the fault diagnosis model in a supervised manner. Based on the historical operation data of the RV reducer and the set fault labels, training the fault diagnosis model, performing knowledge distillation on the trained deep learning model to compress the model volume, and deploying it to the RV reducer; 2. The intelligent monitoring and fault diagnosis integrated RV reducer control method according to claim 1, wherein Installing vibration, temperature, sound, and current sensors on the RV reducer to collect the operation data of the RV reducer in real time; Preprocessing the collected operation data, where the preprocessing includes removing outliers, data denoising, and data normalization; The outlier removal includes: using criteria to remove outliers; the data denoising includes: using wavelet transform for data denoising; the data normalization includes: using the maximum and minimum values for data normalization; Through a time synchronization algorithm, aligning the operation data collected by different sensors to a unified timestamp to form a synchronized multi-modal data stream; the time synchronization algorithm includes timestamp alignment and data splicing; 3. The control method of the RV reducer for integrated intelligent monitoring and fault diagnosis according to claim 2, characterized in that, Extract the vibration data features, temperature data features, sound data features, and current data features from the operating data; let the vibration data be ; let the temperature data be ; let the sound data be ; let the current data be ; Extract the vibration data features, where the vibration data features include the time-domain features of the vibration data, the frequency-domain features of the vibration data, and the time-frequency domain features of the vibration data; the time-domain features of the vibration data include the root mean square value of the vibration data ; The frequency-domain features of the vibration data include: performing a fast Fourier transform on the vibration signal to obtain the frequency spectrum of the vibration data ; extracting the amplitude of the vibration data frequency spectrum ; the time-frequency domain features of the vibration data are ; extracting temperature data features, where the temperature data features include the temperature mean , the temperature gradient and the temperature fluctuation ; extracting sound data features, where the sound data features include the time-domain features and the frequency-domain features of the sound data, and the time-domain features of the sound data include the root mean square value of the sound data ; the frequency-domain features of the sound data include: performing a fast Fourier transform on the sound signal to obtain the frequency spectrum of the sound data ; extracting the main frequency component of the sound data ; Extract the current data features, where the current data features include the time-domain features and the frequency-domain features of the current data features; the time-domain features of the current data features include the current mean value , the root mean square value of the current ; the frequency-domain features of the current data features include: performing a fast Fourier transform on the current data to obtain the frequency spectrum of the current data , and extracting the fundamental wave component ; Fusing the extracted feature data: ； Among them, is the feature vector after feature data fusion.

4. The control method of the RV reducer for integrated intelligent monitoring and fault diagnosis according to claim 3, wherein After extracting the data features of different types of operation data, the weights of different modal features are adaptively allocated by learning the correlation between the data; using the fused feature vector as the input of the attention mechanism; Let , , represent the query matrix, the key matrix, and the value matrix respectively, where is the dimension of the fused feature vector . , , are the dimensions of the query, key, and value respectively; R （*） represents the set of real numbers; the calculation process of the attention mechanism is as follows: ； Among them, , , are learnable weight matrices; calculate the similarity score between the query and the key: ; among them, is the similarity matrix; is the transpose of the key matrix; perform softmax normalization on the similarity matrix to obtain the attention weights: ; among them, is the attention weight matrix; Multiply the attention weights with the value matrix to obtain the weighted and fused features: ; where is the fused feature matrix.

5. The intelligent monitoring and fault diagnosis integrated RV reducer control method according to claim 4, characterized in that, Constructing a fault diagnosis model based on a convolutional neural network, including: For the th branch, , let denote the corresponding eigen submatrix, where is the eigen dimension, is the time step; reshape into the form of , where and are the height and width respectively, satisfying ; perform convolution and pooling operations on : ； ； Among them, and are the weights and biases of the -th branch and the -th convolutional layer respectively. represents the convolution operation. is the number of convolutional layers of the -th branch. is the activation function. is the max pooling operation. is the convolution of the -th branch and the -th convolutional layer. is the pooling of the -th branch and the -th convolutional layer. Pooling results of each branch are flattened into vectors , and concatenated to form a feature vector ; The is input into a fully connected layer for feature fusion and classification: ； ； Among them, , , , are the weights and biases of the fully connected layer respectively, , is the output probability vector, is the number of fault categories; the loss function of the fault diagnosis model adopts the cross-entropy function: ； Among them, is the one-hot encoding of the true label, is the predicted probability, and c is the fault category index number.

6. The control method of the RV reducer for integrated intelligent monitoring and fault diagnosis according to claim 5, characterized in that, Training the fault diagnosis model, and the training steps are as follows: Collecting the historical operation data of the RV reducer, including vibration, temperature, sound, and current data, and the corresponding fault labels; dividing the historical operation data into a training set, a validation set, and a test set, preprocessing the historical operation data, including removing outliers, data denoising, and data normalization; extracting the features of various historical operation data and performing adaptive feature fusion through the attention mechanism; Constructing a fault diagnosis model based on a convolutional neural network, with each branch of the convolutional neural network corresponding to a type of sensor data, and the input being the feature sub-matrix fused by the attention mechanism; in each branch of the convolutional network, performing convolution, activation, and pooling operations on the input data to extract high-level features; flattening and splicing the features of each branch, and then performing feature fusion and fault classification through a fully connected layer; Defining the cross-entropy loss function and optimizer of the fault diagnosis model, setting the number of training epochs, batch size, and learning rate; training the model on the training set, updating the model parameters through the backpropagation algorithm to minimize the loss function; after each training cycle ends, evaluating the model performance on the validation set. If the current loss function value is less than the loss function value of the previous training cycle, then save the current model; If the model performance on the validation set does not improve for several consecutive training cycles, then reduce the learning rate and trigger an early stopping strategy.

7. The control method of the RV reducer for integrated intelligent monitoring and fault diagnosis according to claim 6, characterized in that, Construct a teacher model and a student model. Use the trained deep learning model as the teacher model, and define the parameters of the teacher model as , and the output is the soft label ; Construct the student model, and define the parameters of the student model as , and the output is ; Define the distillation loss function and let represent the true label of the th sample, represent the soft label output of the teacher model, represent the output of the student model on the sample; the distillation loss function is defined as: ； Among them, is the weight coefficient of the distillation process; is the temperature parameter of the soft label, which is used to control the smoothness of the soft label; is the cross-entropy loss function; Train the student model while fixing the parameters of the teacher model , train the student model to minimize the distillation loss function; each training step is as follows: ； Among them, is the learning rate, is the training batch size; repeat the above training steps until the distillation loss function converges; Integrating the student model after knowledge distillation into the embedded system of the RV reducer for operation monitoring and fault diagnosis of the RV reducer.

8. The control system of the RV reducer integrated with intelligent monitoring and fault diagnosis is used to implement the control method of the RV reducer integrated with intelligent monitoring and fault diagnosis according to any one of claims 1 to 7, and is characterized in that, Including: A data acquisition module, a feature extraction module, a data fusion module, a fault diagnosis module, and a distillation and deployment module; The data acquisition module is used to acquire the operation data of the RV reducer in a multi-modal manner, and the operation data includes vibration data, temperature data, sound data, and current data; The feature extraction module extracts features from the acquired operation data of the RV reducer, and extracts vibration data features, temperature data features, sound data features, and current data features in the operation data; For the heterogeneous data collected by different sensors, after extracting the features of different types of operation data, the data fusion module designs an adaptive feature data fusion method based on the attention mechanism, dynamically adjusts the weights of different types of feature data, and realizes intelligent monitoring; The fault diagnosis module is used to construct a fault diagnosis model based on a convolutional neural network. The convolutional neural network adopts a multi-branch neural network structure, and each branch corresponds to a type of sensor data. The fault diagnosis model is used to perform data fusion and fault classification on the RV reducer; The distillation and deployment module trains the fault diagnosis model in a supervised manner, trains the fault diagnosis model based on the historical operation data of the RV reducer and the set fault labels; performs knowledge distillation on the trained deep learning model, compresses the model volume, and deploys it to the RV reducer.

9. An electronic device, characterized in that, Including: A processor and a memory. Among them, the memory stores a computer program that can be called by the processor; the processor executes the RV reducer control method for integrated intelligent monitoring and fault diagnosis according to any one of claims 1 to 7 by calling the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that: Stores instructions that, when the instructions run on a computer, cause the computer to execute the RV reducer control method for integrated intelligent monitoring and fault diagnosis according to any one of claims 1 to 7.

Images