Method for monitoring pass category of radial forging machine

Abstract

The invention discloses a pass category monitoring method for a radial forging machine, and relates to the technical field of radial forging. Obtaining a target function for clustering; the objective function is constructed according to a kernel space distance function between the data point and a clustering center of the pass category and a membership function of the data point to the pass category; performing feature extraction on the forging process sample data and the equipment state sample data at each time point, and determining features of a plurality of data points; initializing a clustering center of each pass category; iteratively updating the clustering center of each pass category by taking the minimization of an objective function value as an objective until iteration meets a convergence condition, and determining the clustering center meeting the convergence condition as a target clustering center of each pass category; and the forging pass category of the radial forging machine is monitored in real time through the target clustering center. According to the method, the pass type of the radial forging machine can be accurately monitored.

Description

Technical Field

[0001] The present invention relates to the technical field of radial forging, and particularly relates to a method for monitoring the pass categories of a radial forging machine. Background Art

[0002] Currently, radial forging often requires multiple passes to machine a workpiece from its initial state to the final desired shape and size, which describes the development process of the workpiece's geometric shape from the initial state to the final state.

[0003] For example, when forging a thick shaft blank into a precision shaft-like part, it may require 4 to 5 passes. Each pass has specific deformation amounts and process parameters, gradually changing the dimensions such as the diameter and length of the workpiece, and simultaneously improving its internal tissue properties. Therefore, it is extremely necessary to monitor the pass categories.

[0004] However, the prior art lacks a method that can accurately monitor the pass categories of a radial forging machine. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide a method for monitoring the pass categories of a radial forging machine, which can accurately monitor the pass categories of a radial forging machine.

[0006] The present invention adopts the following technical solutions: The present invention provides a method for monitoring the pass categories of a radial forging machine, including: Obtaining an objective function for clustering; the objective function is constructed based on the kernel space distance function between data points and the clustering centers of pass categories, and the membership function of data points to pass categories; the kernel space distance function includes the kernel function of the data points themselves, the kernel function of the clustering centers themselves, and the kernel function between the data points and the clustering centers; the kernel function includes a univariate Gaussian kernel term for representing the similarity between two univariate data points and a bivariate interaction kernel term for representing the interaction effect between two variables; Obtaining the forging process sample data and equipment state sample data at multiple time points of the radial forging machine, and respectively performing feature extraction on the forging process sample data and equipment state sample data at each time point to determine the features of multiple data points; Initializing the clustering centers of each pass category; Taking the minimization of the objective function value as the goal, and iteratively updating the clustering centers of each pass category according to the features of multiple data points until the iteration meets the convergence condition, and determining the clustering centers that meet the convergence condition as the target clustering centers of each pass category; Extract the target features of the forging process data and equipment status data of the radial forging machine at the current moment, calculate the membership degrees of the target features to the target cluster centers of each pass category respectively, and determine the pass category forged by the radial forging machine at the current moment as the pass category corresponding to the maximum membership degree.

[0007] Optionally, the kernel function is: ; Where is the kernel function between data point and data point , are the univariate Gaussian kernel term and the bivariate interaction term respectively, is the feature dimension of data point and data point , is the feature weight, , is the kernel sensitivity factor, is the feature of data point in the th dimension, is the feature of data point in the th dimension, represents the L2 norm, , represents the non-linear interaction effect weight between the feature of the th dimension and the feature of the th dimension of the data point, represents the interaction term sensitivity factor between the feature of the th dimension and the feature of the th dimension of the data point, is the feature of data point in the th dimension, is the feature of data point in the th dimension, is the feature of data point in the th dimension, is the feature of data point in the th dimension.

[0008] Optionally, the kernel space distance function between data point and the cluster center of the pass category is: ; Where is the cluster center of the pass category, represents data point Its own kernel function, is the clustering center Its own kernel function, is the data point The kernel function between and the clustering center; The th data point The membership function of the th pass category is: ; Among them, is the number of pass categories, The th data point and the th clustering center of the pass category The kernel function between The th data point and the th clustering center of the pass category The kernel function between is the reciprocal of; The objective function is: ; Among them, is the total number of data points, The th data point The membership function for the th pass category, The th data point Its own kernel function, The th clustering center of the pass category Its own kernel function, The th data point and the th clustering center of the pass category The kernel function between is the first regularization coefficient, is the second regularization coefficient, is the feature weight, , is 's L1 norm, is 's L2 norm, is the fuzzy index.

[0009] Optionally, the feature weights are iteratively updated by a soft threshold algorithm during the clustering process; the update formula is: ; where is the feature weight at the -th iteration, , is the -th iteration of the -th data point to the -th pass category's clustering center the kernel space distance function between, is the feature weight at the -th iteration, is the -th iteration of the -th data point to the -th pass category's membership function, represents the soft threshold operator, is the input of the soft threshold operator, is the threshold, is 's L2 norm.

[0010] Optionally, a momentum mechanism is introduced in the clustering center update process; the update formula for the clustering center is: ; where is the -th iteration of the -th pass category's clustering center, is the -th iteration of the -th pass category's clustering center, is the momentum coefficient.

[0011] Optionally, the convergence condition is: The value of the objective function is less than the first preset threshold, and the displacement of the clustering center of each pass category between the current iteration and the previous iteration is less than the second preset threshold.

[0012] Optionally, feature extraction is performed on the forging process sample data and equipment status sample data at each time point respectively to determine the features of multiple data points, including: Filter out abnormal data points beyond the range of forging process sample data and equipment status sample data through the three-standard deviation principle, and use the mean filling method to complete the missing values after filtering out abnormal data points to obtain complete data points; Perform denoising processing on the complete data points to obtain denoised data points; Extract time-domain features and frequency-domain features from the forging process sample data and equipment status sample data of the denoised data points respectively to obtain multiple time-domain features and multiple frequency-domain features of each data point; Calculate the importance scores of all features, and select the top K features with the largest importance scores as the features of each data point.

[0013] Optionally, extract the target features of the forging process data and equipment status data of the radial forging machine at the current moment, including: Extract the key features of the forging process data and equipment status data of the radial forging machine at the current moment, and determine the key features as the target features.

[0014] The present invention provides a monitoring device for the pass category of a radial forging machine, including: A first acquisition module for acquiring an objective function for clustering; the objective function is constructed based on the kernel space distance function between the data points and the clustering centers of the pass categories, and the membership function of the data points to the pass categories; the kernel space distance function includes the kernel function of the data points themselves, the kernel function of the clustering centers themselves, and the kernel function between the data points and the clustering centers; the kernel function includes a univariate Gaussian kernel term for representing the similarity between two univariate data points and a bivariate interaction kernel term for representing the interaction effect between two variables; A second acquisition module for acquiring the forging process sample data and equipment status sample data of the radial forging machine at multiple time points, and respectively performing feature extraction on the forging process sample data and equipment status sample data of each time point to determine the features of multiple data points; An initialization module for initializing the clustering centers of each pass category; An update module for minimizing the objective function value, and iteratively updating the clustering centers of each pass category according to the features of multiple data points until the iteration meets the convergence condition, and determining the clustering centers that meet the convergence condition as the target clustering centers of each pass category; A monitoring module for extracting the target features of the forging process data and equipment status data of the radial forging machine at the current moment, and calculating the membership degrees of the target features to the target clustering centers of each pass category respectively, and determining the pass category corresponding to the maximum membership degree as the pass category forged by the radial forging machine at the current moment.

[0015] The present invention provides a computer-readable storage medium storing a computer program, which when executed by a processor implements the above-mentioned pass category monitoring method for a radial forging machine.

[0016] The present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the program, it implements the above-mentioned pass category monitoring method for a radial forging machine.

[0017] At least one of the above technical solutions adopted by the present invention can achieve the following beneficial effects: In the present invention, the univariate Gaussian kernel term can capture the contribution of each feature to the sample similarity alone, reflecting the characteristics of a single feature; based on the bivariate interaction kernel term, the non-linear interaction information between different features is mined. In this way, the two are combined to construct a kernel function, comprehensively considering the feature itself and the interaction relationship between features, so that the kernel function can more accurately describe the similarity between data points, laying a good foundation for subsequent analysis. Further, by synthesizing the kernel function of the data point itself, the kernel function of the clustering center itself, and the kernel function of the data point and the clustering center, a kernel space distance function is constructed to comprehensively measure the relationship between the data point and the clustering center, and a target function is constructed in combination with the membership degree, taking into account the belonging degree of the data point to different pass categories, so that the target function can effectively reflect the degree of fit between the data and the pass category clustering, guiding the clustering process to be more accurate; and when extracting data features, the equipment status and process data of the radial forging machine are collected simultaneously, enriching the feature dimension, so as to learn the pass features from multiple angles, improving the discrimination ability for different pass categories, and thus improving the monitoring accuracy of the pass category of the radial forging machine. Description of the Drawings

[0018] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention, and do not constitute an improper limitation to the present invention. In the drawings: Figure 1 It is a schematic flow chart of a pass category monitoring method for a radial forging machine provided by the present invention; Figure 2 It is a schematic diagram of the result of outlier cleaning of a simulation signal provided by the present invention; Figure 3 It is a schematic diagram of the result of missing item filling of a simulation signal provided by the present invention; Figure 4 It is a schematic diagram of the result of noise removal of a simulation signal provided by the present invention; Figure 5 It is a schematic diagram of the result of data scale unification of a simulation signal provided by the present invention; Figure 6Schematic flow chart of another pass category monitoring method for the radial forging machine provided by the present invention; Figure 7 Schematic diagram of a computer device for implementing the pass category monitoring method of the radial forging machine provided by the present invention. Detailed implementation manners

[0019] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments and corresponding drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] Currently, during the radial forging process of rotary parts (such as hydraulic cylinder bodies, shaft parts, etc.), their dimensional accuracy and surface quality highly depend on the precise control of forging pass parameters. Pass monitoring is crucial in radial forging and is an important part of realizing a fully automatic intelligent production line. Achieving precise pass monitoring is the premise for the intelligent monitoring of the production line state. Its functions are as follows: in the scenario of manual feeding, it can prevent waste products caused by incorrect manual feeding; in the fully automatic scenario, the abnormal recognition accuracy can be improved by dynamically switching the dedicated detection model through pass monitoring.

[0021] Traditional pass monitoring methods mainly rely on off-line measurement or fixed rules for judgment, and it is difficult to adapt to the complex and changeable process dynamic characteristics in the production mode of multiple varieties and small batches. Especially in the scenario of precision forging, where the material deformation is non-linear and the process parameters are strongly coupled, the existing technologies have the following prominent problems: (1) Disconnection between off-line prediction and real-time monitoring: Traditional methods rely on static modeling of historical data and cannot capture the dynamic changes (such as sudden changes in material rheology and cumulative thermal deformation of the die) during the forging process, resulting in a lag in parameter adjustment.

[0022] (2) Insufficient cooperation of multi-source data: Process parameters, equipment states (vibration, oil pressure) and part features are scattered in independent systems, and the existing threshold method or single model is difficult to characterize the coupling effect of multiple factors, and the false alarm rate reaches 15%-20%.

[0023] (3) Poor adaptability to small samples: In the mode of multiple varieties and small batches, deep learning methods are prone to overfitting due to sparse data, and traditional clustering algorithms such as K-means are sensitive to noise and have a high feature redundancy.

[0024] (4) Bottleneck in computing efficiency: The processing delay of traditional machine learning methods exceeds 500 ms, which is difficult to meet the requirements of real-time monitoring.

[0025] Based on this, the present invention provides a method for monitoring pass categories of a radial forging machine, aiming to realize real-time perception and adaptive adjustment of forging process parameters of rotary parts through multi-source data fusion and improved sparse kernel fuzzy clustering, break through the dependence of the existing technology on static rules and large amounts of data, and improve the monitoring accuracy and process stability under complex working conditions.

[0026] The execution subject of the method in the present invention can be a server set up in a business platform, or devices such as desktop computers and laptop computers that can execute the solution of the present invention. The following will detail the technical solutions provided by each embodiment of the present invention with reference to the accompanying drawings.

[0027] Figure 1 FIG. is a schematic flow chart of a method for monitoring pass categories of a radial forging machine in the present invention, which specifically includes the following steps: S101, obtain an objective function for clustering; the objective function is constructed according to the kernel space distance function between data points and the clustering center of pass categories, and the membership function of data points to pass categories; the kernel space distance function includes the kernel function of data points themselves, the kernel function of clustering centers themselves, and the kernel function between data points and clustering centers; the kernel function includes a univariate Gaussian kernel term for representing the similarity between two univariate data points and a bivariate interaction kernel term for representing the interaction effect between two variables.

[0028] The kernel function is a structured ANOVA kernel function, and the kernel function is: (1); Among them, is the kernel function between data point and data point , are respectively the univariate Gaussian kernel term and the bivariate interaction term, is the feature dimension of data point and data point , is the feature weight, , screening out dominant classification features through sparse constraints; is the kernel sensitivity factor, is the feature of data point in the rd dimension, is the feature of data point in the th dimension, represents the L2 norm, , represents the non-linear interaction effect weight between the feature of the data point in the th dimension and the feature in the th dimension, represents the data point in the The interaction term sensitivity factor between the feature of the dimension and the feature of the For the data point the -dimensional feature, For the data point the -dimensional feature, For the data point the -dimensional feature, For the data point the -dimensional feature.

[0029] Among them, , For the data point and the data point the standard deviation of the -dimensional feature, , For all data points, the standard deviation of the interaction feature between the -dimensional and the -dimensional.

[0030] Among them, , .

[0031] Optionally, the kernel space distance function between the data point and the clustering center of the pass category is: (2); Among them, is the clustering center of the pass category, represents the kernel function of the data point itself, is the kernel function of the clustering center itself, is the kernel function between the data point and the clustering center .

[0032] According to the kernel space distance function, determine the kernel space weighted distance term. The calculation form of the kernel space weighted distance term is: (3); Among them, represents the membership function of the -th data point to the -th pass category, is the fuzzy index, is the number of pass categories, is the total number of data points, denotes the th data point its own kernel function, denotes the th clustering center of the pass category its own kernel function, is expressed as the th data point and the th clustering center of the pass category the kernel function between them.

[0033] Optionally, the th data point the membership function for the th pass category is: (4); wherein, is the number of pass categories, is the th data point and the th clustering center of the pass category the kernel function between them, is the th data point and the th clustering center of the pass category the kernel function between them, is the reciprocal of.

[0034] The objective function is defined as the combination of the kernel space weighted distance term and the regularization term, that is, the objective function is: (5); wherein, is the total number of data points, is the th data point the membership function for the th pass category, satisfying , and ; is the th data point its own kernel function, is the th clustering center of the pass category its own kernel function, is the th data point With the cluster center of the th pass category is the first regularization coefficient, is the second regularization coefficient, is the feature weight, , is 's L1 norm, is 's L2 norm, is the fuzzy exponent, the first regularization coefficient , forcing sparsity to eliminate redundant features; the second regularization coefficient , preventing weight overfitting.

[0035] is the fuzzy exponent, controlling the sharpness of the membership degree distribution (the larger the value, the more blurred the classification boundary).

[0036] S102. Obtain the forging process sample data and equipment status sample data at multiple time points of the radial forging machine, and perform feature extraction on the forging process sample data and equipment status sample data at each time point respectively to determine the features of multiple data points.

[0037] Optionally, perform feature extraction on the forging process sample data and equipment status sample data at each time point respectively to determine the features of multiple data points, including: filtering out abnormal data points beyond the range of the forging process sample data and equipment status sample data through the 3-standard deviation principle, and using the mean filling method to complete the missing values after filtering out the abnormal data points to obtain complete data points; performing denoising processing on the complete data points to obtain denoised data points; performing time-domain feature extraction and frequency-domain feature extraction on the forging process sample data and equipment status sample data of the denoised data points respectively to obtain multiple time-domain features and multiple frequency-domain features of each data point; calculating the importance scores of all features, and selecting the K features with the largest importance scores as the features of each data point; optionally, according to all features, obtain multiple feature sets, traverse each feature set to score the clustering index (mainly referring to the classification coefficient), and select the set with the best clustering effect. The features corresponding to the best set are the K features with the largest importance scores.

[0038] Specifically, assuming that the total number of features is W , it is necessary to select K features and generate kinds of feature sets. For example, if , , then the number of feature sets is .

[0039] Taking the silhouette coefficient as an example of the classification coefficient, for each subset, a clustering model (such as K-means) is trained with the characteristic data of the subset. Based on the clustering results, the silhouette coefficient of each sample data in the subset is calculated, and the mean value of the silhouette coefficients in the subset is calculated. The subset corresponding to the largest mean silhouette coefficient is determined as the one with the largest selected importance score. K features.

[0040] Specifically, the forging process sample data includes forging frequency and reduction; the equipment status sample data includes vibration acceleration (0 - 20 kHz), oil pressure fluctuation (0 - 100 MPa), and die temperature (20 - 800 °C).

[0041] Outlier cleaning: Adopting the 3-standard deviation criterion, by calculating the mean value ( ) and variance ( ), setting the threshold , filtering out the outlier data points beyond the range. As shown in Figure 2 , Figure 2 Figure (a) in Figure 2 is the signal schematic diagram before outlier cleaning, and Figure 2 Figure (b) in

[0042] Missing value filling: Using the mean filling method to complete the missing values to ensure data integrity. As shown in Figure 3 , Figure 3 Figure (a) in Figure 3 is the signal schematic diagram before missing value filling, and Figure 3 Figure (b) in

[0043] Noise removal: Applying the Savitzky-Golay filter for signal smoothing, with a window size of 7 and a polynomial order of 2, suppressing high-frequency noise while retaining the signal characteristics. As shown in Figure 4 , Figure 4 Figure (a) in Figure 4 is the signal schematic diagram before noise removal, and Figure 4 Figure (b) in

[0044] Data scale unification: Adopting the normalization algorithm , eliminating the dimension difference. As shown in Figure 5 , Figure 5 Figure (a) in Figure 5 is the signal schematic diagram before data scale unification, and Figure 5Figure (b) in it is a schematic diagram of the signal after data scale unification.

[0045] After unifying the data scale, the forging process sample data and equipment status sample data of the data points are respectively subjected to time-domain feature extraction and frequency-domain feature extraction. The forging process sample data includes forging frequency and reduction; the equipment status sample data includes vibration acceleration, oil pressure fluctuation, and die temperature.

[0046] Among them, the time-domain feature extraction includes: extracting parameters such as mean, standard deviation, maximum value, minimum value, root mean square, waveform factor (with dimension), and kurtosis, skewness (dimensionless), etc.

[0047] The specific calculation method is: ; ; ; ; ; ; ; . Among them, is the number of data points of the time series data corresponding to a single sample data in the forging process sample data and equipment status sample data, represents the value of the corresponding sample data at time series.

[0048] The frequency-domain feature extraction includes: extracting spectral mean, spectral maximum value, spectral energy, spectral entropy (with dimension), and the specific method is: ; ; ; , where, ; where, the spectral of the corresponding sample data at time series.

[0049] According to the above calculation method, calculate the time-domain features and frequency-domain features of each time point respectively, and then evaluate the feature importance based on the classification coefficient, and screen the most important K key features (such as the classification coefficients of standard deviation, minimum value, root mean square, waveform factor, and spectral mean > 0.85, which can be considered as key features).

[0050] Take the obtained K key features as the features of each time point. A time point can be regarded as a data point, that is, the obtained are multiple features of each data point.

[0051] S103. Initialize the clustering centers for each pass category, and with the goal of minimizing the objective function value, update the clustering centers for each pass category iteratively according to the characteristics of multiple data points until the iteration meets the convergence condition, and determine the clustering centers that meet the convergence condition as the target clustering centers for each pass category.

[0052] Among them, the pass categories of the radial forging machine can include four categories, as shown in Table 1.

[0053] Table 1

[0054] Optionally, the feature weights are iteratively updated through a soft threshold algorithm during the clustering process; the update formula is: (6); Among them, is the feature weight for the th iteration , is the th data point in the th iteration For the th clustering center of the pass category the kernel space distance function between is the feature weight for the th iteration, is the th data point at the th iteration For the th membership function of the pass category represents the soft threshold operator, is the input of the soft threshold operator, is the threshold, is the L2 norm of , represents at the th iteration of , represents at the th iteration of , represents at the th iteration of .

[0055] Determines the weight contraction amplitude ( ); : Sparsity parameter, is the feature dimension of the data point, controlling the number of retained features (such as When retaining approximately 4 dimensions).

[0056] Optionally, a momentum mechanism is introduced into the clustering center update process; the update formula for the clustering center is: (7); Wherein, is the clustering center of the th iteration of the th pass category, is the clustering center of the th iteration of the th pass category, is the momentum coefficient, .

[0057] Optionally, the convergence condition is: the value of the objective function is less than the first preset threshold, and the displacement of the clustering center of each pass category between the current iteration and the previous iteration is less than the second preset threshold.

[0058] The first preset threshold and the second preset threshold can be set according to actual requirements.

[0059] S104, extract the target features of the forging process data and equipment status data of the radial forging machine at the current moment, calculate the membership degrees of the target features to the target clustering centers of each pass category respectively, and determine the pass category forged by the radial forging machine at the current moment as the pass category corresponding to the maximum membership degree.

[0060] Optionally, extracting the target features of the forging process data and equipment status data of the radial forging machine at the current moment includes: extracting the key features of the forging process data and equipment status data of the radial forging machine at the current moment, and determining the key features as the target features.

[0061] Based on the target features, calculate the membership degrees of the data points at the current moment to each pass category according to formula (4), and determine the pass category corresponding to the maximum membership degree as the pass category forged by the radial forging machine at the current moment.

[0062] In one embodiment, the present invention also provides a method for monitoring the pass category of a radial forging machine, as Figure 6 shown, this embodiment specifically includes: S601, the acquired signal group.

[0063] Wherein, the signal group includes forging process sample data and equipment status sample data.

[0064] S602, data preprocessing.

[0065] S603, training data.

[0066] Use the preprocessed data as training data.

[0067] S604, Feature extraction.

[0068] Extract features from the training data.

[0069] S605, Cluster the extracted features.

[0070] Cluster the extracted features by an improved sparse kernel fuzzy clustering algorithm. The specific clustering method is the same as that in the above embodiment, and this embodiment will not limit it here.

[0071] S606, Determine whether the clustering effect is optimal.

[0072] Determine whether the clustering effect is optimal: Determine the accuracy of clustering through the target cluster centers after clustering. If the accuracy is less than the accuracy threshold, the clustering effect is not optimal; if not, re - execute S604 (adjust the selected features), otherwise, execute step S608.

[0073] S607, Test data.

[0074] S608, Pass classification.

[0075] S609, Output the classification result.

[0076] The present invention proposes a method for monitoring the pass categories of a radial forging machine. Through multi - source data fusion, ANOVA kernel decomposition and double sparse constraints, and through a structured kernel function and an adaptive optimization mechanism, the classification performance under complex working conditions is significantly improved. The core of this algorithm is to map the equipment state and process parameters to a high - dimensional feature space, use this method to enhance the feature separability, and dynamically screen key features through sparse constraints, breaking through the limitations of traditional offline prediction and static models, and realizing real - time monitoring of forging pass categories and process adaptive optimization. Technical effects: Real - time performance: Processing delay ≤ 200ms (processor i9 - 14900k), a 60% improvement compared with traditional methods; Accuracy: The recognition accuracy of pass categories ≥ 98.1%, and the false alarm rate ≤ 4.7%; Fast training: The training cycle is shortened to 2 - 3 batches.

[0077] When applying the method for monitoring the pass categories of the radial forging machine provided by the present invention, it is not necessary to execute according to Figure 1 the order of the steps shown. The specific execution order of each step can be determined as needed, and the present invention does not limit this.

[0078] The above is the method for monitoring the pass categories of a radial forging machine provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding device for monitoring the pass categories of a radial forging machine. The device includes: The first acquisition module is configured to acquire an objective function for clustering; the objective function is constructed based on a kernel space distance function between a data point and a clustering center of a pass category, and a membership function of the data point to the pass category; the kernel space distance function includes a kernel function of the data point itself, a kernel function of the clustering center itself, and a kernel function between the data point and the clustering center; the kernel function includes a univariate Gaussian kernel term for representing the similarity between two univariate data points and a bivariate interaction kernel term for representing the interaction effect between two variables. The second acquisition module is configured to acquire forging process sample data and equipment status sample data of a radial forging machine at multiple time points, and respectively perform feature extraction on the forging process sample data and the equipment status sample data at each time point to determine the features of multiple data points. The initialization module is configured to initialize the clustering center of each pass category. The update module is configured to aim at minimizing the objective function value, and iteratively update the clustering center of each pass category according to the features of multiple data points until the iteration meets the convergence condition, and determine the clustering center that meets the convergence condition as the target clustering center of each pass category; a momentum mechanism is introduced in the clustering center update process. The monitoring module is configured to extract the target features of the forging process data and the equipment status data of the radial forging machine at the current moment, calculate the membership degrees of the target features to the target clustering centers of each pass category respectively, and determine the pass category corresponding to the maximum membership degree as the pass category forged by the radial forging machine at the current moment.

[0079] For the specific limitations of the pass category monitoring device of the radial forging machine, reference can be made to the limitations of the pass category monitoring method of the radial forging machine in the foregoing text, which will not be elaborated here. Each module in the above-mentioned pass category monitoring device of the radial forging machine can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor of the computer device in the form of hardware or be independent of it, or can be stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to the above-mentioned modules.

[0080] The present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program can be used to execute the above-mentioned Figure 1 Provided pass category monitoring method of the radial forging machine.

[0081] The present invention also provides Figure 7 The structural schematic diagram of the computer device shown in, as Figure 7As shown, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above Figure 1 Monitoring method for pass categories of the rotary forging machine provided.

[0082] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memories. The non-volatile memory can include a read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. The volatile memory can include a random access memory (RAM) or an external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as a static random access memory (SRAM) or a dynamic random access memory (DRAM), etc.

[0083] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the 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 the scope recorded in the present invention.

Claims

1. A method for monitoring pass categories of a radial forging machine, characterized in that, Including: Obtain the objective function for clustering; The objective function is constructed based on the kernel space distance function between the data points and the cluster centers of the pass categories, and the membership function of the data points to the pass categories; the kernel space distance function includes the kernel function of the data points themselves, the kernel function of the cluster centers themselves, and the kernel function between the data points and the cluster centers; the kernel function includes a univariate Gaussian kernel term for representing the similarity between two univariate data points and a bivariate interaction kernel term for representing the interaction effect between two variables; Obtain the forging process sample data and equipment status sample data at multiple time points of the radial forging machine, and perform feature extraction on the forging process sample data and equipment status sample data at each time point respectively to determine the features of multiple data points; Initialize the cluster centers of each pass category; Aiming at minimizing the objective function value, update the cluster centers of each pass category iteratively according to the features of multiple data points until the iteration meets the convergence condition, and determine the cluster centers that meet the convergence condition as the target cluster centers of each pass category; Extract the target features of the forging process data and equipment status data of the radial forging machine at the current moment, and calculate the membership degrees of the target features to the target cluster centers of each pass category respectively, and determine the pass category corresponding to the maximum membership degree as the pass category forged by the radial forging machine at the current moment.

2. The method according to claim 1, characterized in that, The kernel function is: ； Among them, is the kernel function between data point and data point which are the univariate Gaussian kernel term and the bivariate interaction term respectively, is the feature dimension of data point and data point is the feature weight, , is the kernel sensitivity factor, is the feature of the th dimension of data point the th dimension represents the L2 norm, , represents the weight of the non - linear interaction effect between the feature of the th dimension and the feature of the th dimension of the data point, is the feature of the th dimension of data point the th dimension is the feature of the th dimension of data point the th dimension 3. The method according to claim 2, wherein Data point The clustering center of the pass category The kernel space distance function between them is as follows: ； Among them, is the clustering center of the pass category, represents the kernel function of the data point itself, is the kernel function of the clustering center itself, is the kernel function between the data point and the clustering center ; The th data point for the membership function of the th pass category is: ； Among them, is the number of categories of passes, is the th data point and the th clustering center of the pass category the kernel function between them, is the th data point and the th clustering center of the pass category the kernel function between them, is the reciprocal of; Objective function is as follows: ； wherein, is the total number of data points, is the th data point 's membership function for the th pass category, is the th data point 's own kernel function, is the th pass category's clustering center 's own kernel function, is the th data point 's kernel function with the th pass category's clustering center in between, is the first regularization coefficient, is the second regularization coefficient, is the feature weight, , is 's L1 norm, is 's L2 norm, is the fuzzy exponent.

4. The method according to claim 3, wherein Feature weight Iteratively updated through a soft threshold algorithm during the clustering process; the update formula is: ； Among them, is the feature weight of the th iteration, , is the th data point of the th iteration, for the th clustering center of the th pass category, is the feature weight of the th iteration, is the th membership function of the th data point at the th iteration for the th pass category, represents the soft threshold operator, is the input of the soft threshold operator, is the threshold, is the L2 norm of.

5. The method according to claim 4, wherein The momentum mechanism is introduced in the cluster center update process; the update formula of the cluster center is: ； Among them, is the clustering center of the th pass category in the th iteration, is the clustering center of the th pass category in the th iteration, is the momentum coefficient.

6. The method according to claim 1, wherein The convergence condition is: The objective function value is less than the first preset threshold, and the displacement of the cluster center of each pass category between the current iteration and the previous iteration is less than the second preset threshold.

7. The method according to claim 1, characterized in that, The performing feature extraction on the forging process sample data and equipment status sample data at each time point respectively to determine the features of multiple data points includes: Filter out the abnormal data points beyond the ranges of the forging process sample data and equipment status sample data through the 3-standard deviation principle, and use the mean filling method to complete the missing values after filtering out the abnormal data points to obtain complete data points; Perform denoising processing on the complete data points to obtain the denoised data points; Perform time-domain feature extraction and frequency-domain feature extraction on the forging process sample data and equipment status sample data of the denoised data points respectively to obtain multiple time-domain features and multiple frequency-domain features of each data point; Calculate the importance scores of all features, and select the K features with the largest importance scores as the features of each data point.

8. The method according to claim 7, wherein The extracting the target features of the forging process data and equipment status data of the radial forging machine at the current moment includes: Extract the key features of the forging process data and equipment status data of the radial forging machine at the current moment, and determine the key features as the target features.

Images