Multi-process collaborative control method and system for processing electric vehicle shock absorption cylinder
By using a convolutional neural network model and anomaly cause mapping table, the problems of low detection efficiency, uncoordinated repair, and difficulty in tracing the root cause in the processing of electric vehicle shock absorbers were solved, and efficient and accurate multi-process collaborative control was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANJIN HAIDE TECHNOLOGY CO LTD
- Filing Date
- 2025-07-22
- Publication Date
- 2026-05-19
AI Technical Summary
Traditional electric vehicle shock absorber manufacturing processes suffer from insufficient inspection efficiency and accuracy, lack of coordination in anomaly repair, difficulty in tracing the root cause of anomalies, and difficulty in effectively controlling quality in multi-process collaborative operations.
A convolutional neural network model is used to evaluate multi-dimensional quality features, construct an anomaly label-feature deviation matrix and process correction matrix, analyze suspected cause parameters, construct an anomaly cause mapping table, and achieve accurate repair and anomaly root cause tracing.
This improves detection efficiency and accuracy, ensuring that defective products are repaired promptly and effectively in subsequent processes, accurately pinpointing the root cause of the defect, and enhancing production efficiency and quality control.
Smart Images
Figure CN120893758B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of electric vehicle shock absorber processing technology, specifically relating to a multi-process collaborative control method and system for electric vehicle shock absorber processing. Background Technology
[0002] In the field of electric vehicle shock absorber manufacturing, traditional production quality control methods have many shortcomings; existing methods have the following technical problems:
[0003] Insufficient detection efficiency and accuracy: Traditional detection relies on manual labor and simple automated equipment, which is inefficient and easily affected by subjective factors. It is difficult to comprehensively evaluate multi-dimensional quality characteristics, leading to missed detections and misjudgments.
[0004] Lack of coordination in anomaly repair: Existing repair measures are insufficient in multi-process coordination, making it difficult for subsequent processes to provide targeted repairs for abnormal products, which affects production efficiency and quality control;
[0005] Tracing the root cause of anomalies is difficult: Traditional tracing methods rely on human experience and subjective speculation based on limited data fragments, making it difficult to accurately pinpoint the cause of anomalies. Especially when multiple processes are working together, it is difficult to prevent the recurrence of anomalies from the root cause. To address this, we propose a multi-process collaborative control method and system for electric vehicle shock absorber manufacturing. Summary of the Invention
[0006] The purpose of this invention is to provide a multi-process collaborative control method and system for the processing of electric vehicle shock absorbers, so as to solve the problems mentioned in the background art.
[0007] To achieve the above objectives, the present invention provides the following technical solution: a multi-process collaborative control method for the processing of electric vehicle shock absorbers, comprising the following steps:
[0008] Step S1: For each process in the manufacturing of electric vehicle shock absorbers, construct a convolutional neural network evaluation model to label the products produced in each process as high-quality and abnormal.
[0009] Step S2: Mark abnormal processes, extract abnormal evaluation labels and calculate feature deviation values, and construct an abnormal label-feature deviation matrix; analyze the repairable feature types and dynamic correction range of subsequent processes, and construct a process correction matrix; based on the abnormal label-feature deviation matrix and the process correction matrix, determine the candidate repair processes corresponding to each abnormal evaluation label, and analyze the comprehensive matching value to determine the selected repair processes.
[0010] Step S3: For each abnormal process, organize the abnormal evaluation tags to build an abnormal evaluation tag set; analyze the suspected cause parameters of each abnormal evaluation tag; calculate the abnormal evaluation value to determine the abnormal induction parameters, build an abnormal cause mapping table and send it to the operation and maintenance personnel.
[0011] Preferably, the construction process of the convolutional neural network model is as follows:
[0012] Step S101: For each process in the manufacturing of electric vehicle shock absorbers, establish a product quality evaluation label; the product quality evaluation label contains m corresponding process evaluation features;
[0013] Step S102: Establish a 6-layer convolutional neural network model, including an input layer, three convolutional layers, and two fully connected layers. The convolutional layers contain convolution, activation, and pooling operations for feature extraction; the fully connected layers are used for feature integration and classification.
[0014] Step S103: Collect photos related to the established m quality assessment label features. The number of photos corresponding to each label feature is k, for a total of d photos; number each photo with the number q, q = 1, 2, ..., d; d = mk;
[0015] Each of the d photos is evaluated and labeled; for a photo containing a certain evaluation label feature, the corresponding evaluation label Ep = 1; otherwise, Ep = 0; where p = 1, 2, ..., m;
[0016] Step S104: Input the d photos marked with evaluation labels as the original images into the 6-layer convolutional neural network model, and make the specific value of the output value Fp of the 6-layer convolutional neural network model corresponding to each original image equal to the evaluation label Ep with the same number p.
[0017] Perform training operations on the convolutional neural network, and record all weight factors and bias factors obtained after training.
[0018] Test the convolutional neural network: Repeat step S103 to obtain d new photos as the original images and input them into the 6-layer convolutional neural network after training; then record the output values Fp = F1, F2, ..., Fm.
[0019] Through the formula: Calculate the test success rate S; if the test success rate S is greater than the preset threshold, the training is determined to be successful, and all weight factors and bias factors obtained from the training are output and the process proceeds to step S105.
[0020] Otherwise, the training is deemed to have failed, and step S104 is repeated until the condition is met: the test success rate S is greater than the preset threshold.
[0021] Preferably, the specific process of step S105 is as follows:
[0022] Step S105: Take G images of the shock absorber cylinder product after the current process using a high-resolution camera and supporting image acquisition equipment;
[0023] The collected G product images are used as raw images and input into a 6-layer neural network. The output is a set of output values F1, F2, F3, ..., Fm corresponding to m product evaluation labels for each product image.
[0024] For each output value, compare it with a preset threshold. If Fp is greater than the preset threshold, add a field called "Quality Rating" to the metadata of the product image corresponding to that output value. The initial value is 0. When the conditions for evaluating a high-quality product are met, the value of this field is increased by 1.
[0025] For each feature label, count the number of times 0 appears in G photos, and record it as the anomaly frequency. If the anomaly frequency of the feature label in G photos is greater than the corresponding threshold, then mark the feature label as an anomaly evaluation label. Perform the above statistical and marking operations on all feature labels, and finally output all anomaly evaluation labels; and mark the product as an anomaly product.
[0026] Preferably, in step S2, the specific process of constructing the anomaly label-feature deviation matrix is as follows:
[0027] Step S201: Mark the process in which abnormal products occur as an abnormal process, extract features from the abnormal product image, and obtain various abnormal evaluation labels for the abnormal products.
[0028] The abnormal features corresponding to each abnormal evaluation label are compared with the corresponding standard features, and the feature deviation value is calculated.
[0029] All anomaly evaluation labels and their corresponding feature deviation values are organized to construct an anomaly label-feature deviation matrix.
[0030] Preferably, the specific process for constructing the process correction matrix is as follows:
[0031] Step S202: For each abnormal process, mark the subsequent processing process as the next process; determine the repairable feature type and limit correction range for each process;
[0032] For each subsequent process, a table of influencing factors for repair features is established. In the table of influencing factors for repair features, each repair feature corresponds to a set of influencing factors.
[0033] Match the repair features with the corresponding process's repair feature influencing factor table to output the corresponding influencing factor set; for each influencing factor in the influencing factor set, calculate the influencing factor deviation value;
[0034] The deviation values corresponding to each influencing factor are multiplied by their corresponding weighting coefficients and then summed to obtain the comprehensive impact value.
[0035] Several comprehensive impact value ranges are set, and each range corresponds to a correction coefficient. By matching the comprehensive impact value corresponding to the repair feature with all the corresponding comprehensive impact value ranges, the corresponding correction coefficient is output.
[0036] The dynamic correction range is obtained by multiplying the correction coefficient corresponding to each repair feature in the subsequent process by the upper limit value in the corresponding limit correction range;
[0037] Record the correction feature type and corresponding dynamic correction range corresponding to the current moment of each subsequent process, and combine them to construct a process correction matrix.
[0038] Preferably, the specific process for determining the selected repair procedure is as follows:
[0039] Step S203: Compare each abnormal feature in the abnormal label-feature deviation matrix with the process correction matrix, and select subsequent processes that can cover the abnormal feature correction requirements as candidate repair processes.
[0040] If all abnormal features of the abnormal product generated in the abnormal process have corresponding candidate repair processes, then the abnormal process is repaired without stopping the machine. Otherwise, if any abnormal feature has no corresponding candidate repair process, then the machine is repaired without stopping the machine.
[0041] Step S204: For each candidate repair procedure corresponding to an abnormal feature, calculate the comprehensive matching value; from the candidate repair procedures corresponding to the abnormal feature, select the candidate repair procedure with the largest comprehensive matching value and designate it as the repair selection procedure for repairing the abnormal feature.
[0042] Preferably, in step S3, the specific process of analyzing the suspected cause parameters of each abnormal evaluation label is as follows:
[0043] Step S301: For each abnormal process, organize the abnormal evaluation tags that appear in the process and construct an abnormal evaluation tag set;
[0044] For each abnormal evaluation label in the abnormal evaluation label set, analyze the influence parameters that are more than the threshold in relation to the abnormal evaluation label, and mark them as suspected cause parameters.
[0045] Preferably, the specific process for determining the anomaly induction parameters is as follows:
[0046] Step S302: For each suspected cause parameter, establish a two-dimensional rectangular coordinate system, with the horizontal axis representing time and the vertical axis representing the value of the suspected cause parameter; by marking the values of the suspected cause parameter collected from each collection point in the recent log of the suspected cause parameter in the rectangular coordinate system, several value points are obtained; by connecting adjacent value points with curves, a curve of the change of the suspected cause parameter is obtained.
[0047] Step S303: Draw two threshold lines on the suspected cause parameter change curve graph, namely the upper threshold line and the lower threshold line. Mark the area enclosed by the suspected cause parameter change curve above the upper threshold line as the threshold overshoot area; mark the area enclosed by the suspected cause parameter change curve below the lower threshold line as the threshold descent area; sum all the threshold overshoot areas and threshold descent areas to obtain the threshold overshoot product YC and threshold descent product YJ; calculate the standard deviation of the values corresponding to each collection point of the suspected cause parameter change curve to obtain the curve fluctuation value QB.
[0048] Step S304: After normalizing the threshold over-limit product YC, threshold descent product YJ, and curve fluctuation value QB corresponding to the suspected cause parameters, the abnormal evaluation value YPZ is obtained using the formula: YPZ=YC×d1+YJ×d2+QB×d3, where d1, d2, and d3 are preset weight coefficients.
[0049] A preset threshold for abnormal evaluation values is set. The abnormal evaluation values of suspected trigger parameters are compared with the corresponding threshold. If the abnormal evaluation value is greater than or equal to the corresponding threshold, the suspected trigger parameter is marked as an abnormal trigger parameter.
[0050] Preferably, the specific process for constructing the abnormal cause mapping table is as follows:
[0051] Step S305: Organize each anomaly evaluation label and its corresponding anomaly cause parameters to construct an anomaly cause mapping table; and send it to the operation and maintenance personnel. The operation and maintenance personnel adjust the anomaly cause parameters that lead to the anomaly evaluation label according to the anomaly cause mapping table.
[0052] Compared with the prior art, the beneficial effects of the present invention are:
[0053] (1) The multi-process collaborative control method and system for the processing of electric vehicle shock absorbers realizes automatic evaluation of multi-dimensional quality characteristics by using a convolutional neural network model; specifically, by constructing product quality evaluation labels that include features such as appearance size, surface defects and structural integrity, and training a convolutional neural network model, the system accurately identifies whether the product meets the high-quality standard; at the same time, it locates abnormal evaluation labels to provide a clear direction for subsequent repair and adjustment.
[0054] (2) The multi-process collaborative control method and system for the processing of electric vehicle shock absorbers achieves targeted repair of abnormal products in subsequent processes by constructing an anomaly label-feature deviation matrix and a process correction matrix. It clarifies the repairable feature types and dynamic correction range of subsequent processes and accurately matches abnormal features with repair processes to ensure that abnormal products are repaired in a timely and effective manner in subsequent processes.
[0055] (3) The multi-process collaborative control method and system for the processing of electric vehicle shock absorbers, by constructing an abnormal cause mapping table, for each abnormal process, organizes abnormal evaluation tags to construct an abnormal evaluation tag set, analyzes suspected cause parameters and determines abnormal inducement parameters; further, organizes these parameters and their corresponding abnormal evaluation tags to construct an abnormal cause mapping table and sends it to the operation and maintenance personnel end, so as to realize accurate tracing of the root cause of the abnormality. Attached Figure Description
[0056] Figure 1 This is a flowchart of the present invention;
[0057] Figure 2 This is a graph showing the variation of parameters that may be the cause of the present invention. Detailed Implementation
[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0059] Example 1
[0060] Please see Figures 1-2 This invention provides a multi-process collaborative control method for the processing of electric vehicle shock absorbers, comprising the following steps:
[0061] Step S1: For each process in the manufacturing of electric vehicle shock absorbers, a convolutional neural network evaluation model is constructed to label the products produced in each process as high-quality or abnormal. The specific process is as follows:
[0062] Step S101: For each process in the manufacturing of electric vehicle shock absorbers, establish a product quality evaluation label; the product quality evaluation label contains m corresponding process evaluation features; the evaluation features include: appearance dimensions (such as diameter, length and tolerance range, etc.), surface defects (such as scratches, pits, cracks and their characteristics and grading standards, etc.), and structural integrity (such as weld quality, connection firmness inspection points and judgment criteria, etc.);
[0063] Step S102: Establish a 6-layer convolutional neural network model, including an input layer, a first convolutional layer, a second convolutional layer, a third convolutional layer, a first fully connected layer, and a second fully connected layer;
[0064] The input layer contains a compressor and a preprocessor;
[0065] The compressor converts the image height and width to 512×512 pixels by compressing and cropping the original image;
[0066] The preprocessor in the first convolutional layer filters the input image through compression cropping and Gaussian transform. The Gaussian transform formula is: Where σ is the Gaussian standard deviation, * is the convolution operator, μ(x,y) is the Gaussian convolution kernel; f(x,y) is the input original image, where y0(x,y) is the output image of the input layer;
[0067] The first, second, and third convolutional layers all contain convolutional layers, activation layers, and pooling layers;
[0068] The first convolutional layer contains n1 nodes, and its output formula is: Where j1 = 1, 2, 3, ..., n1; where * is the convolution operator; where The weights of the first convolutional layer. is the bias factor of the first convolutional layer; where F is the activation function of the activation layer;
[0069] The second convolutional layer contains n² nodes, and the output formula of the second convolutional layer is: Where j2 = 1, 2, 3, ..., n2; where The weights of the second convolutional layer. This is the bias factor for the second convolutional layer;
[0070] The third convolutional layer contains n3 nodes, and the output formula of the third convolutional layer is: Where j3 = 1, 2, 3, ..., n3; where The weights of the third convolutional layer. This is the bias factor for the second convolutional layer;
[0071] The pooling layers in the first, second, and third convolutional layers perform noise reduction and redundant data removal through average pooling operations. The average pooling is the average value of the output results of the pixels in the pooling region.
[0072] The first fully connected layer contains n4 nodes, and its output formula is: Where j4 = 1, 2, 3, ..., n4; where As the weight factor of the first fully connected layer, The bias factor for the first fully connected layer;
[0073] The second fully connected layer contains 7 nodes, and its output formula is: Where p = 1, 2, ..., m; m is the number of process evaluation features, where w P b is the weight factor for the second fully connected layer. P This is the bias factor for the second fully connected layer;
[0074] Step S103: Collect photos related to the established m quality assessment label features. The number of photos corresponding to each label feature is k, for a total of d photos; number each photo with the number q, q = 1, 2, ..., d; d = mk;
[0075] Each of the d photos is evaluated and labeled; for a photo containing a certain evaluation label feature, the corresponding evaluation label Ep = 1; otherwise, Ep = 0; where p = 1, 2, ..., m;
[0076] Step S104: Input the d photos labeled with evaluation tags as the original images into the 6-layer convolutional neural network model, and set the output value of the 6-layer convolutional neural network model corresponding to each original image. The specific value is equal to the evaluation label Ep that has the same number p;
[0077] Perform training operations on the convolutional neural network, and record all weight factors and bias factors obtained after training.
[0078] Test the convolutional neural network: Repeat step S103 to obtain d new photos as the original images and input them into the 6-layer convolutional neural network after training; then record the output values Fp = F1, F2, ..., Fm.
[0079] Through the formula: Calculate the test success rate; if the test success rate S is greater than the preset threshold, the training is determined to be successful, and all weight factors and bias factors obtained from the training are output and the process proceeds to step S105.
[0080] Otherwise, the training is deemed to have failed, and step S104 is repeated until the condition is met: the test success rate S is greater than the preset threshold.
[0081] Step S105: Take G images of the shock absorber cylinder product after the current process using a high-resolution camera and supporting image acquisition equipment;
[0082] The collected G product images are used as raw images and input into a 6-layer neural network. The output is a set of output values F1, F2, F3, ..., Fm corresponding to m product evaluation labels for each product image.
[0083] For each output value, it is compared with a preset threshold. If Fp is greater than the preset threshold, it indicates that the shock absorber meets the evaluation criteria for a high-quality product in terms of the features described by the corresponding feature label. At this time, a field named "high-quality score" is added to the metadata of the product image corresponding to the output value. The initial value is 0. When the evaluation criteria for a high-quality product are met, the value of this field is increased by 1.
[0084] For each feature tag, count the number of times 0 appears in G photos, and record it as the anomaly frequency. Set an anomaly frequency threshold. If the anomaly frequency of the feature tag in G photos is greater than the corresponding threshold, then mark the feature tag as an anomaly evaluation tag. Perform the above statistical and marking operations on all feature tags, and finally output all anomaly evaluation tags; and mark the product as an abnormal product.
[0085] It should be noted that by establishing product quality evaluation labels and convolutional neural network models, automatic evaluation of multiple dimensions of shock absorber products, such as appearance size, surface defects and structural integrity, has been achieved, replacing the traditional manual evaluation method and improving evaluation efficiency and accuracy.
[0086] For each feature tag, the frequency of anomalies is counted to accurately locate the anomaly evaluation tags, providing a clear direction for subsequent repair and adjustment.
[0087] Step S2: Mark abnormal processes, extract abnormal evaluation labels and calculate feature deviation values to construct an abnormal label-feature deviation matrix; analyze the repairable feature types and dynamic correction range of subsequent processes to construct a process correction matrix; based on the abnormal label-feature deviation matrix and the process correction matrix, determine the candidate repair processes corresponding to each abnormal evaluation label, and analyze the comprehensive matching value to determine the selected repair processes. The specific process is as follows:
[0088] Step S201: Mark the process in which abnormal products occur as abnormal process, and use deep learning algorithm to extract features from the abnormal product image to obtain various abnormal evaluation labels for the abnormal product; including but not limited to: appearance size deviation, surface defect features, structural integrity issues, etc.
[0089] The abnormal features corresponding to each anomaly evaluation label are compared with the corresponding standard features, and the feature deviation value is accurately calculated using computer vision technology; such as the specific value of the size deviation, the area and depth of the surface defect, etc., and the direction and location information of the deviation are recorded.
[0090] All anomaly evaluation labels and their corresponding feature deviation values are organized to construct an anomaly label-feature deviation matrix. This matrix not only records each anomaly label and its corresponding feature deviation value, but also contains the feature vector description of the deviation, so as to facilitate subsequent feature matching and repair strategy formulation.
[0091] Step S202: For each abnormal process, mark the subsequent processing process as the follow-up process; based on the equipment specifications and process parameter range of the follow-up process, determine the repairable feature type and limit correction range of each process;
[0092] For each subsequent process, a table of influencing factors for repair features is established. In the table of influencing factors for repair features, each repair feature corresponds to a set of influencing factors.
[0093] Match the repair features with the corresponding process's repair feature influencing factor table to output the corresponding influencing factor set; for each influencing factor in the influencing factor set, subtract the current value of the influencing factor from the corresponding preset benchmark value to obtain the influencing factor deviation value;
[0094] Based on the absolute value of the correlation between each influencing factor and the corresponding repair feature, a weight coefficient is assigned to the deviation value of each influencing factor. Then, the deviation values of each influencing factor are multiplied by the corresponding weight coefficient and summed to obtain the comprehensive influence value.
[0095] Several comprehensive impact value ranges are set, and each range corresponds to a correction coefficient. The correction coefficient is output by matching the comprehensive impact value corresponding to the repair feature with all the corresponding comprehensive impact value ranges; where the correction coefficient is less than or equal to one.
[0096] The dynamic correction range is obtained by multiplying the correction coefficient corresponding to each repair feature in the subsequent process by the upper limit value in the corresponding limit correction range;
[0097] Record the correction feature type and corresponding dynamic correction range corresponding to the current moment of each subsequent process, and combine them to construct a process correction matrix;
[0098] Step S203: Compare each abnormal feature in the abnormal label-feature deviation matrix with the process correction matrix, and select subsequent processes that can cover the abnormal feature correction requirements as candidate repair processes.
[0099] If all abnormal features of the abnormal product generated in the abnormal process have corresponding candidate repair processes, then the abnormal process is repaired without stopping the machine. Otherwise, if any abnormal feature has no corresponding candidate repair process, then the machine is repaired without stopping the machine.
[0100] S204: For each abnormal feature, the candidate repair procedure is determined using the formula: The matching accuracy value P is obtained; where Da is the abnormal feature deviation value, Ds is the center value of the candidate process correction capability range, and Dr is the half length of the candidate process correction capability range.
[0101] The total number of times the abnormal feature was repaired and the number of times it was successfully repaired in the historical candidate repair process are retrieved. The matching stability value W is obtained by dividing the number of successfully repaired features by the total number of abnormal features.
[0102] After normalizing the matching accuracy value P and the matching stability value W, the comprehensive matching value ZP is obtained using the formula: ZP=P×f1+W×f2; where f1 and f2 are preset weight coefficients.
[0103] From the candidate repair processes corresponding to the abnormal features, the candidate repair process with the highest comprehensive matching value is selected and designated as the repair confirmation process for repairing the abnormal feature. When the abnormal product enters the repair confirmation process corresponding to the abnormal feature, the repair work for the abnormal feature is carried out.
[0104] It should be noted that, based on the equipment specifications and process parameter ranges of subsequent processes, the types of repairable features and the limit correction range are determined, and the dynamic correction range is obtained by adjusting the correction coefficient, which enhances the flexibility and adaptability of the repair scheme.
[0105] Record the correction feature type and corresponding dynamic correction range of each subsequent process, and combine them to construct a process correction matrix, providing comprehensive data support for matching abnormal features with repair processes;
[0106] By comparing the anomaly label-feature deviation matrix with the process correction matrix, subsequent processes that can cover the anomaly feature correction requirements are selected as candidate repair processes, ensuring the feasibility of the repair plan.
[0107] By calculating the matching accuracy value and the matching stability value, a comprehensive matching value analysis is performed on the candidate repair procedures to select the optimal repair procedure, thereby improving the success rate and reliability of the repair.
[0108] Based on whether each abnormal feature has a corresponding candidate repair procedure, the non-stop repair or shutdown repair mode is selected, which reduces production interruption and improves production efficiency.
[0109] Step S3: For each abnormal process, organize abnormal evaluation tags to construct an abnormal evaluation tag set; analyze the suspected cause parameters of each abnormal evaluation tag; calculate the abnormal evaluation value to determine the abnormal induction parameters, construct an abnormal cause mapping table and send it to the operation and maintenance personnel. The specific process is as follows:
[0110] Step S301: For each abnormal process, organize the abnormal evaluation tags that appear in the process and construct an abnormal evaluation tag set;
[0111] For each anomaly assessment label in the set of anomaly assessment labels, the influence parameters that are more than the threshold in correlation with the anomaly assessment label are analyzed based on the Pearson correlation coefficient method and marked as suspected cause parameters.
[0112] Step S302: For each suspected trigger parameter, retrieve the recent log records of the suspected trigger parameter, establish a two-dimensional rectangular coordinate system, with the horizontal axis representing time and the vertical axis representing the value of the suspected trigger parameter; by labeling the values of the suspected trigger parameter collected from each collection point in the recent log records of the suspected trigger parameter on the rectangular coordinate system, several value points are obtained; by connecting adjacent value points with curves, a curve graph of the change of the suspected trigger parameter is obtained; for example... Figure 2 ;
[0113] Step S303: Draw two threshold lines on the suspected cause parameter change curve graph, namely the upper threshold line and the lower threshold line. Mark the area enclosed by the suspected cause parameter change curve above the upper threshold line as the threshold overshoot area; mark the area enclosed by the suspected cause parameter change curve below the lower threshold line as the threshold descent area; sum all the threshold overshoot areas and threshold descent areas to obtain the threshold overshoot product YC and threshold descent product YJ; calculate the standard deviation of the values corresponding to each collection point of the suspected cause parameter change curve to obtain the curve fluctuation value QB.
[0114] Step S304: After normalizing the threshold over-limit product YC, threshold descent product YJ, and curve fluctuation value QB corresponding to the suspected cause parameters, the abnormal evaluation value YPZ is obtained using the formula: YPZ=YC×d1+YJ×d2+QB×d3, where d1, d2, and d3 are preset weight coefficients.
[0115] A preset threshold for abnormal evaluation values is set. The abnormal evaluation values of suspected trigger parameters are compared with the corresponding threshold. If the abnormal evaluation value is greater than or equal to the corresponding threshold, the suspected trigger parameter is marked as an abnormal trigger parameter.
[0116] Step S305: Organize each anomaly evaluation label and its corresponding anomaly cause parameters to construct an anomaly cause mapping table; and send it to the operation and maintenance personnel. The operation and maintenance personnel adjust the anomaly cause parameters that cause the anomaly evaluation label according to the anomaly cause mapping table to restore the stable operation of the process and ensure that the product quality meets the standards.
[0117] A multi-process collaborative control system for electric vehicle shock absorber manufacturing includes:
[0118] Quality assessment module: For each process in the manufacturing of electric vehicle shock absorbers, a convolutional neural network model is constructed to label the products produced in each process as high-quality or abnormal.
[0119] Process Repair Matching Module: This module marks abnormal processes, extracts abnormal evaluation labels, calculates feature deviation values, and constructs an abnormal label-feature deviation matrix. It analyzes the repairable feature types and dynamic correction ranges of subsequent processes, constructing a process correction matrix. Based on the abnormal label-feature deviation matrix and the process correction matrix, it determines the candidate repair processes corresponding to each abnormal evaluation label and analyzes the comprehensive matching value to determine the selected repair processes.
[0120] Anomaly Cause Analysis Module: For each abnormal process, it organizes anomaly evaluation tags to build an anomaly evaluation tag set; analyzes the suspected cause parameters of each anomaly evaluation tag; calculates the anomaly evaluation value to determine the anomaly induction parameters; builds an anomaly cause mapping table and sends it to the operation and maintenance personnel.
[0121] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A multi-process collaborative control method for the processing of electric vehicle shock absorbers, characterized in that: Includes the following steps: Step S1: For each process in the manufacturing of electric vehicle shock absorbers, construct a convolutional neural network model to label the products produced in each process as high-quality and abnormal. Step S2: Mark abnormal processes, extract abnormal evaluation labels and calculate feature deviation values, and construct an abnormal label-feature deviation matrix; analyze the repairable feature types and dynamic correction range of subsequent processes, and construct a process correction matrix; based on the abnormal label-feature deviation matrix and the process correction matrix, determine the candidate repair processes corresponding to each abnormal evaluation label, and analyze the comprehensive matching value to determine the selected repair processes. In step S2, the specific process of constructing the anomaly label-feature deviation matrix is as follows: Step S201: Mark the process in which abnormal products occur as an abnormal process, extract features from the abnormal product image, and obtain various abnormal evaluation labels for the abnormal products. The abnormal features corresponding to each abnormal evaluation label are compared with the corresponding standard features, and the feature deviation value is calculated. All anomaly evaluation labels and their corresponding feature deviation values are organized to construct an anomaly label-feature deviation matrix; The specific process for constructing the process correction matrix is as follows: Step S202: For each abnormal process, mark the subsequent processing process as the next process; determine the repairable feature type and limit correction range for each process; For each subsequent process, a table of influencing factors for repair features is established. In the table of influencing factors for repair features, each repair feature corresponds to a set of influencing factors. Match the repair features with the corresponding process's repair feature influencing factor table to output the corresponding influencing factor set; for each influencing factor in the influencing factor set, calculate the influencing factor deviation value; The deviation values corresponding to each influencing factor are multiplied by their corresponding weighting coefficients and then summed to obtain the comprehensive impact value. Several comprehensive impact value ranges are set, and each range corresponds to a correction coefficient. By matching the comprehensive impact value corresponding to the repair feature with all the corresponding comprehensive impact value ranges, the corresponding correction coefficient is output. The dynamic correction range is obtained by multiplying the correction coefficient corresponding to each repair feature in the subsequent process by the upper limit value in the corresponding limit correction range; Record the correction feature type and corresponding dynamic correction range corresponding to the current moment of each subsequent process, and combine them to construct a process correction matrix; The specific process for determining the selected repair procedure is as follows: Step S203: Compare each abnormal feature in the abnormal label-feature deviation matrix with the process correction matrix, and select subsequent processes that can cover the abnormal feature correction requirements as candidate repair processes. If all abnormal features of the abnormal product generated in the abnormal process have corresponding candidate repair processes, then the abnormal process is repaired without stopping the machine. Otherwise, if any abnormal feature has no corresponding candidate repair process, then the machine is repaired without stopping the machine. Step S204: For each candidate repair procedure corresponding to an abnormal feature, calculate the comprehensive matching value; from the candidate repair procedures corresponding to the abnormal feature, select the candidate repair procedure with the largest comprehensive matching value and designate it as the repair selection procedure for repairing the abnormal feature. Step S3: For each abnormal process, organize the abnormal evaluation tags to build an abnormal evaluation tag set; analyze the suspected cause parameters of each abnormal evaluation tag; calculate the abnormal evaluation value to determine the abnormal cause parameters, build an abnormal cause mapping table and send it to the operation and maintenance personnel.
2. The multi-process collaborative control method for processing electric vehicle shock absorbers according to claim 1, characterized in that: The process of building a convolutional neural network model is as follows: Step S101: Establish product quality evaluation labels for each process in the processing of electric vehicle shock absorbers; each product quality evaluation label contains m corresponding process evaluation features; Step S102: Establish a 6-layer convolutional neural network model, including an input layer, three convolutional layers, and two fully connected layers. The convolutional layers contain convolution, activation, and pooling operations for feature extraction; the fully connected layers are used for feature integration and classification. Step S103: Collect photos related to the established m quality assessment label features. The number of photos corresponding to each label feature is k, for a total of d photos; number each photo with the number q, q=1,2,...,d; d=mk; Each of the d photos is evaluated and labeled; for a photo containing a certain evaluation label feature, the corresponding evaluation label Ep = 1; otherwise, Ep = 0; where p = 1, 2, ..., m; Step S104: Input the d photos marked with evaluation labels as the original images into the 6-layer convolutional neural network model, and make the specific value of the output value Fp of the 6-layer convolutional neural network model corresponding to each original image equal to the evaluation label Ep with the same number p. The convolutional neural network model is trained, and all weight factors and bias factors obtained during training are recorded after training. Test the convolutional neural network model: Repeat step S103 to obtain d new photos as the original images and input them into the 6-layer convolutional neural network model after training; then record the output values Fp=F1, F2, ..., Fm. Through the formula: Calculate the test success rate S; If the test success rate S is greater than the preset threshold, the training is determined to be successful, all weight factors and bias factors obtained from the training are output and the process proceeds to step S105. Otherwise, the training is deemed to have failed, and step S104 is repeated until the condition is met: the test success rate S is greater than the preset threshold.
3. The multi-process collaborative control method for processing electric vehicle shock absorbers according to claim 2, characterized in that: The specific process of step S105 is as follows: Step S105: Take G images of the shock absorber cylinder product after the current process using a high-resolution camera and supporting image acquisition equipment; The collected G product images are used as raw images and input into a 6-layer neural network model. The output is a set of output values F1, F2, F3, ..., Fm corresponding to m product evaluation labels for each product image. For each output value, compare it with a preset threshold. If Fp is greater than the preset threshold, add a field called "Quality Rating" to the metadata of the product image corresponding to that output value. The initial value is 0. When the conditions for evaluating a high-quality product are met, the value of this field is increased by 1. For each feature label, count the number of times 0 appears in G product images, and record it as the anomaly frequency. If the anomaly frequency of the feature label in G product images is greater than the corresponding threshold, then mark the feature label as an anomaly evaluation label. Perform the above statistical and marking operations on all feature labels, and finally output all anomaly evaluation labels; and mark the product as an abnormal product.
4. The multi-process collaborative control method for processing electric vehicle shock absorbers according to claim 3, characterized in that: In step S3, the specific process of analyzing the suspected cause parameters for each anomaly evaluation label is as follows: Step S301: For each abnormal process, organize the abnormal evaluation tags that appear in the process and construct an abnormal evaluation tag set; For each abnormal evaluation label in the abnormal evaluation label set, analyze the influence parameters that are more than the threshold in relation to the abnormal evaluation label, and mark them as suspected cause parameters.
5. The multi-process collaborative control method for processing electric vehicle shock absorbers according to claim 4, characterized in that: The specific process for determining the parameters of abnormal causes is as follows: Step S302: For each suspected cause parameter, establish a two-dimensional rectangular coordinate system, with the horizontal axis representing time and the vertical axis representing the value of the suspected cause parameter; by marking the values of the suspected cause parameter collected from each collection point in the recent log of the suspected cause parameter in the rectangular coordinate system, several value points are obtained; by connecting adjacent value points with curves, a curve of the change of the suspected cause parameter is obtained. Step S303: Draw two threshold lines on the suspected cause parameter change curve graph, namely the upper threshold line and the lower threshold line. Mark the area enclosed by the suspected cause parameter change curve above the upper threshold line as the threshold overshoot area; mark the area enclosed by the suspected cause parameter change curve below the lower threshold line as the threshold descent area; sum all the threshold overshoot areas and threshold descent areas to obtain the threshold overshoot product YC and threshold descent product YJ; calculate the standard deviation of the values corresponding to each collection point of the suspected cause parameter change curve to obtain the curve fluctuation value QB. Step S304: After normalizing the threshold overshoot product YC, threshold descent product YJ, and curve fluctuation value QB corresponding to the suspected cause parameters, the abnormal evaluation value YPZ is obtained using the formula: YPZ=YC×d1+YJ×d2+QB×d3, where d1, d2, and d3 are preset weight coefficients. A preset threshold for abnormal assessment values is set. The abnormal assessment values of suspected cause parameters are compared with the corresponding threshold. If the abnormal assessment value is greater than or equal to the corresponding threshold, the suspected cause parameter is marked as an abnormal cause parameter.
6. The multi-process collaborative control method for processing electric vehicle shock absorbers according to claim 5, characterized in that: The specific process of constructing the anomaly cause mapping table is as follows: Step S305: Organize each anomaly evaluation label and its corresponding anomaly cause parameters to construct an anomaly cause mapping table; and send it to the operation and maintenance personnel. The operation and maintenance personnel adjust the anomaly cause parameters that lead to the anomaly evaluation label according to the anomaly cause mapping table.
7. A multi-process collaborative control system for electric vehicle shock absorber machining, applied to the multi-process collaborative control method for electric vehicle shock absorber machining proposed in any one of claims 1-6, characterized in that: include: Quality assessment module: For each process in the manufacturing of electric vehicle shock absorbers, a convolutional neural network model is constructed to label the products produced in each process as high-quality or abnormal. Process repair and matching module: Mark abnormal processes, extract abnormal evaluation labels and calculate feature deviation values, and construct an abnormal label-feature deviation matrix; Analyze the repairable feature types and dynamic correction range of subsequent processes, and construct a process correction matrix; based on the anomaly label-feature deviation matrix and the process correction matrix, determine the candidate repair processes corresponding to each anomaly evaluation label, and analyze the comprehensive matching value to determine the selected repair processes; Anomaly Cause Analysis Module: For each abnormal process, it organizes anomaly evaluation tags to build an anomaly evaluation tag set; analyzes the suspected cause parameters of each anomaly evaluation tag; calculates the anomaly evaluation value to determine the anomaly cause parameters; builds an anomaly cause mapping table and sends it to the operation and maintenance personnel.