A method, device and system for detecting quality of a connection process

Abstract

The application discloses a connection process quality detection method, device and system. The pressure data is obtained, the pressure data is used for reflecting the pressure between the sheet metal and the rivet when the riveting operation is performed on the sheet metal, the qualified probability of the riveting operation as a qualified operation is determined according to the pressure data and the neural network model, the operation quality of the riveting operation is determined based on the qualified probability, and then the qualified probability of the riveting operation can be determined through the neural network model, the manual sampling inspection condition is reduced, and the connection defect missed detection and the like are reduced, and the quality of the self-piercing riveting is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of whole vehicle manufacturing, and particularly relates to a detection method, device and system for connection process quality. BACKGROUND

[0002] Self-piercing riveting is a cold connection technology. A special rivet (π-shaped rivet) penetrates the top plate, and under the action of a riveting die, the hollow structure at the tail of the rivet expands and penetrates into the bottom plate, forming a firm riveting point, thereby connecting two or more metal plates together.

[0003] At present, in order to ensure the quality of self-piercing riveting, the connection quality of self-piercing riveting needs to be detected by manual sampling inspection, so as to recycle the plate with connection defects. However, due to the low frequency of manual sampling inspection and the dependence on the professional level of the sampling inspector in evaluating the connection defects, there is a situation of missing detection of connection defects, which reduces the quality of self-piercing riveting. SUMMARY

[0004] Embodiments of the application provide a detection method, device and system for connection process quality, aiming to solve the problem of missing detection of connection defects when the quality of self-piercing riveting is detected by the existing manual sampling inspection, thereby reducing the quality of self-piercing riveting.

[0005] In a first aspect, embodiments of the application provide a detection method for connection process quality, which comprises:

[0006] obtaining pressure data, wherein the pressure data is used to reflect the pressure between the plate and the rivet when the riveting operation is performed on the plate;

[0007] determining a qualified probability of the riveting operation being a qualified operation according to the pressure data and a neural network model, to determine the operation quality of the riveting operation based on the qualified probability; the neural network model is obtained by training a plurality of sample data, and the sample data includes positive sample data and negative sample data, the positive sample data includes pressure data between the plate and the rivet when the riveting operation performed on the plate is a qualified operation, and the negative sample data includes pressure data between the plate and the rivet when the riveting operation performed on the plate is an unqualified operation.

[0008] In a possible implementation manner of the above first aspect, the method further comprises:

[0009] determining a target operation station where the plate is located when the riveting operation is performed on the plate;

[0010] The method further comprises: determining a target operation station where the plate is located when the riveting operation is performed on the plate; and determining the qualified probability of the riveting operation being a qualified operation according to the pressure data and the neural network model, comprising:

[0011] inputting the pressure data into the neural network model corresponding to the target operation station to obtain a qualified probability of the riveting operation being a qualified operation; different operation stations correspond to different neural network models, and the neural network model corresponding to any operation station is used to determine a qualified probability of the riveting operation performed on a plate at the operation station being a qualified operation.

[0012] In a possible implementation of the first aspect, before the step of inputting the pressure data into the neural network model corresponding to the target operation station to obtain a qualified probability of the riveting operation being a qualified operation, the method further includes:

[0013] determining a target event topic corresponding to the riveting operation, and recording the pressure data of the riveting operation in the target event topic;

[0014] The step of inputting the pressure data into the neural network model corresponding to the target operation station to obtain a qualified probability of the riveting operation being a qualified operation includes:

[0015] determining a target thread subscribed to the target event topic; the target thread is a thread for transmitting data to the neural network model corresponding to the target operation station, and different threads are used to transmit different data to corresponding neural network models;

[0016] transmitting, by the target thread, the pressure data in the target event topic to the neural network model corresponding to the target operation station to obtain a qualified probability of the riveting operation being a qualified operation.

[0017] In a possible implementation of the first aspect, before the step of determining a qualified probability of the riveting operation being a qualified operation according to the pressure data and the neural network model, the method further includes:

[0018] generating a data sequence corresponding to the unqualified operation according to the pressure data of the unqualified operation;

[0019] processing a first data segment in the data sequence based on a first random number to obtain a processed first data segment; the first random number belongs to a first random number range,

[0020] dividing a second data segment in the data sequence into at least two data segments; the average value of the second data segment is greater than the average value of the first data segment,

[0021] processing each of the at least two data segments based on a corresponding second random number of the data segment, to obtain a processed second data segment; the second random number belongs to a second random number range, and the second random number range is larger than the first random number range;

[0022] generating processed negative sample data based on the processed first data segment and the processed second data segment, and training the neural network model based on the processed negative sample data.

[0023] In a possible implementation of the first aspect, before the processing of the first data segment in the data sequence based on the first random number to obtain the processed first data segment, the method further includes:

[0024] determining a target abnormal period corresponding to the unqualified operation; the target abnormal period is a concentrated period of abnormality in the unqualified operation;

[0025] dividing the data sequence into a first data segment and a second data segment based on the target abnormal period; the second data segment is the concentrated period of abnormality.

[0026] In a possible implementation of the first aspect, the method further includes:

[0027] determining a pressure change trend of the riveting operation according to the pressure data; the pressure change trend reflects a pressure change between the sheet metal and the rivet when the riveting operation is performed on the sheet metal in a time period;

[0028] The operation quality of the riveting operation is determined based on the qualified probability, including:

[0029] determining the operation quality of the riveting operation according to the pressure change trend and the qualified probability.

[0030] In a possible implementation of the first aspect, the method further includes:

[0031] determining whether the riveting operation is a target riveting operation according to an execution position of the riveting operation on the sheet metal; the target riveting operation is a riveting operation with a preset change feature in the pressure change trend;

[0032] In the case where the riveting operation is the target riveting operation, the neural network model is used to identify a target feature in the pressure change trend;

[0033] The operation quality of the riveting operation is determined according to the pressure change trend and the probability, including:

[0034] determine the operation quality of the riveting operation according to the target feature, the pressure change trend and the qualified probability.

[0035] In a possible implementation of the first aspect, the method further includes:

[0036] displaying the pressure change trend;

[0037] The determining of the operation quality of the riveting operation according to the qualified probability includes:

[0038] detecting an evaluation operation on the pressure change trend;

[0039] determining quality evaluation information of the riveting operation in response to the evaluation operation;

[0040] determining the operation quality of the riveting operation according to the quality evaluation information and the qualified probability.

[0041] In a second aspect, a device for detecting quality of a connection process includes:

[0042] an acquisition module configured to acquire pressure data, wherein the pressure data is used to reflect pressure between a sheet metal and a rivet when a riveting operation is performed on the sheet metal;

[0043] a determination module configured to determine a qualified probability that the riveting operation is a qualified operation according to the pressure data and a neural network model, and determine operation quality of the riveting operation based on the qualified probability, wherein the neural network model is obtained by training a plurality of sample data, and the sample data includes positive sample data and negative sample data, the positive sample data includes pressure data between the sheet metal and the rivet when the riveting operation performed on the sheet metal is a qualified operation, and the negative sample data includes pressure data between the sheet metal and the rivet when the riveting operation performed on the sheet metal is an unqualified operation.

[0044] In a third aspect, an embodiment of the present application provides a system for detecting quality of a connection process, the system including:

[0045] a data acquisition module configured to acquire pressure data between a rivet and a sheet metal in a process of performing a riveting operation at each operation station, wherein the riveting operation is an operation of riveting the sheet metal;

[0046] a quality detection module configured to distribute the pressure data to a neural network model corresponding to the operation station, obtain an output result of the neural network model, and generate operation quality of the riveting operation performed at the operation station based on the output result.

[0047] In a fourth aspect, an embodiment of the present application provides a terminal device, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the connection process quality detection method provided in the first aspect when executing the computer program.

[0048] In a fifth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the connection process quality detection method provided in the first aspect.

[0049] In a sixth aspect, an embodiment of the present application provides a computer program product, comprising a computer program, which, when running on a computer, causes the computer to execute the connection process quality detection method provided in the first aspect.

[0050] It can be understood that the beneficial effects of the second aspect to the sixth aspect can be referred to the related description in the first aspect, which will not be repeated here.

[0051] Compared with the prior art, the beneficial effects of the embodiments of the present application are as follows:

[0052] In the embodiments of the present application, by acquiring pressure data, the pressure data is used to reflect the pressure between the sheet metal and the rivet when the riveting operation is performed on the sheet metal, and according to the pressure data and the neural network model, the qualified probability of the riveting operation as a qualified operation is determined, so as to determine the operation quality of the riveting operation based on the qualified probability, and then the qualified probability of the riveting operation can be determined through the neural network model, the manual sampling inspection is reduced, and the connection defect missed detection and the like are reduced, thereby improving the quality of the self-piercing riveting. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 is a step flow chart of a connection process quality detection method provided by an embodiment of the present application;

[0054] Figure 2 is a structural schematic diagram of a neural network model provided by an embodiment of the present application;

[0055] Figure 3 is a step flow chart of another connection process quality detection method provided by an embodiment of the present application;

[0056] Figure 4 is a schematic diagram of a pressure change trend provided by an embodiment of the present application;

[0057] Figure 5 is a structural schematic diagram of a connection process quality detection system provided by an embodiment of the present application;

[0058] Figure 6 is a flowchart of a connection process quality detection method provided by an embodiment of the present application;

[0059] Figure 7 is a structural diagram of a connection process quality detection device provided by an embodiment of the present application;

[0060] Figure 8 is a structural block diagram of a terminal device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0061] In order to make the technical problems, technical solutions and beneficial effects of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0062] Self-piercing riveting is a cold connection technology. A special rivet (π-shaped rivet) is inserted and penetrates through the top plate, and under the action of a riveting die, the hollow structure of the rivet tail expands and penetrates into the bottom plate, forming a firm riveting point, thereby connecting two or more metal plates together. It is widely used in the field of vehicle manufacturing, especially in the connection process of automobile body, such as the connection of automobile body frame, door, engine cover, trunk, seat and other parts.

[0063] However, in the actual riveting process, it will be affected by the riveting process, riveting equipment and the quality of the rivet, etc., resulting in connection defects, such as rivet deviation from the riveting position, rivet distortion, rivet crack, etc. Specifically, the connection defects can also include defects affecting the connection strength and defects affecting the connection appearance. The defects affecting the connection strength can include cracks in the plate and / or rivet after riveting operation, deformation in the plate and / or rivet after riveting operation and the size change after deformation is greater than a preset proportion, and the distance of the rivet from the riveting position is greater than a preset distance. The defects affecting the connection appearance can include scratches on the rivet or plate, deformation in the plate and / or rivet after riveting operation and the size change after deformation is less than or equal to a preset proportion, and the distance of the rivet from the riveting position is less than or equal to a preset distance.

[0064] The defects affecting the connection strength affect the connection strength between each sheet material, so that the connected sheet material is easily disconnected by external force, that is, after two or more sheet materials are connected, the riveting part between each sheet material is easily loosened or broken by external force, so that the connection between each sheet material is disconnected. The defects affecting the appearance of the connection do not affect the connection strength between each sheet material or have less effect than the predetermined range, so that the connected sheet material is difficult to be disconnected by external force, and only the appearance of the connected sheet material is affected.

[0065] At present, in order to ensure the quality of self-punching riveting, the connection quality of self-punching riveting needs to be detected by artificial sampling to recycle the sheet material with connection defects. Generally, the artificial sampling method is to extract one or more sheet materials from all sheet materials subjected to self-punching riveting, and to artificially chisel the riveting part of the extracted sheet material, so that the chiseled part can be observed by the sampling personnel to evaluate the riveting connection quality of the riveting part, so that in the case that the riveting connection quality of the sheet material is evaluated as unqualified, it is determined that the riveting sheet material of the batch has riveting quality problem, and the sheet material of the batch is recycled for further detection and processing to ensure the connection quality of self-punching riveting.

[0066] However, due to the low frequency of artificial sampling and the dependence on the professional level of sampling personnel in evaluating the riveting connection quality, there is a situation of missing connection defects, which reduces the quality of self-punching riveting.

[0067] Therefore, the present application provides a connection process quality detection method, which acquires pressure data reflecting the pressure between the sheet material and the rivet during the riveting operation of the sheet material, determines the qualified probability of the riveting operation as a qualified operation according to the pressure data and a neural network model, and determines the operation quality of the riveting operation based on the qualified probability, so that the qualified probability of the riveting operation can be determined by the neural network model, the situation of artificial sampling is reduced, and the situation of missing connection defects is reduced, thereby improving the quality of self-punching riveting.

[0068] Referring to Figure 1 , Figure 1 A step flowchart of a connection process quality detection method provided by an embodiment of the present application is shown, which can specifically include the following steps:

[0069] Step 101, acquiring pressure data.

[0070] The pressure data can be used to reflect the pressure between the rivet and the sheet metal during the riveting operation on the sheet metal, the riveting operation can be an operation of riveting the sheet metal, and the pressure between the rivet and the sheet metal corresponds to the displacement of the rivet inserted into the sheet metal, different displacements can correspond to different pressures, and the sheet metal can be a sheet material that needs to be riveted.

[0071] In actual application, different sheet metals can be riveted by using a riveting device. Specifically, the riveting device can be used to apply pressure to the rivet so that the rivet penetrates the top sheet metal, and under the action of the rivet die, the hollow structure of the rivet tail expands and penetrates the sheet metal to form a firm riveting point, thereby completing the riveting operation on the sheet metal. The riveting device can record the pressure between the rivet and the sheet metal during riveting, and record the displacement of the rivet inserted into the sheet metal, and thus the corresponding relationship between the pressure between the rivet and the sheet metal and the displacement of the rivet inserted into the sheet metal can be established, so that the pressure data of each riveting operation can be recorded.

[0072] In specific implementation, the terminal device can obtain the pressure data recorded in the riveting device.

[0073] In step 102, the qualified probability of the riveting operation being a qualified operation is determined according to the pressure data and the neural network model, and the operation quality of the riveting operation is determined based on the qualified probability.

[0074] The neural network model can be used to determine the qualified probability of the riveting operation being a qualified operation according to the pressure data, and the riveting connection quality being qualified can indicate that there is no connection defect affecting the connection strength at the riveting position. The neural network model can be obtained by training a plurality of sample data, and specifically can be obtained by training sample data of each operation station. The operation station can be a station for riveting the sheet metal, and each operation station can deploy a riveting device for performing the riveting operation. The sample data can be the pressure data between the rivet and the sheet metal during the riveting operation, and the sample data can include positive sample data and negative sample data. The positive sample data can include the pressure data between the sheet metal and the rivet when the riveting operation on the sheet metal is a qualified operation, and specifically can include the pressure data of at least one qualified operation. The negative sample data can include the pressure data between the sheet metal and the rivet when the riveting operation on the sheet metal is an unqualified operation, and specifically can include the pressure data corresponding to at least one unqualified operation. The qualified operation can be a riveting operation with qualified riveting connection quality, and the unqualified operation can be a riveting operation with unqualified riveting connection quality. The riveting connection quality being qualified can indicate that there is no connection defect affecting the connection strength at the riveting position, and the riveting connection quality being unqualified can indicate that there is a connection defect affecting the connection strength at the riveting position. The operation quality can be information for evaluating the quality of the riveting operation, and specifically can be information for evaluating whether the riveting operation will produce a connection defect affecting the connection strength.

[0075] After obtaining the pressure data of the riveting operation, the operating station for performing the riveting operation can be determined, as well as the neural network model corresponding to the operating station. The pressure data of the riveting operation is then input into the neural network model, which can then output the corresponding pass probability based on the input pressure data. This pass probability indicates that the riveting operation is a qualified operation, and the operation quality of the riveting operation can be evaluated based on the pass probability of the riveting operation.

[0076] In practical applications, the pass probability output by the neural network model can be compared with a pre-set pass probability threshold. If the pass probability output by the neural network model is greater than or equal to the pass probability threshold, the riveting operation can be determined to be a pass operation, and the operation quality of the riveting operation can be determined to be that it will not produce connection defects that affect the connection strength, that is, the riveting operation has no riveting quality problem. However, if the pass probability output by the neural network model is less than the pass probability threshold, the riveting operation can be determined to be a fail operation, and the operation quality of the riveting operation can be determined to be that it will produce connection defects that affect the connection strength, that is, the riveting operation has a riveting quality problem.

[0077] In practice, before determining the neural network model, a neural network model can be pre-constructed to determine the probability of a successful riveting operation. Specifically, the constructed neural network model can be a convolutional neural network model.

[0078] See Figure 2 , Figure 2 This application provides a schematic diagram of the structure of a neural network model according to an embodiment of the present application. Figure 2 As shown, the neural network model 2 consists of two convolutional layers 21, two max-pooling layers 22, and one fully connected layer 23, with each convolutional layer 21 followed by a max-pooling layer 22. The convolutional layers 21 are used to perform convolution operations on the input data to identify features. The max-pooling layers 22 are used to remove features with low relevance from the input data. The fully connected layer 23 outputs corresponding probability information based on the features with high relevance, i.e., the probability that the riveting operation is a successful operation.

[0079] After constructing the neural network model, positive and negative sample data can be collected. The neural network model can then be trained using the positive and negative sample data, and the model parameters can be updated based on the training results. Based on the trained neural network model, the operation of determining the success probability of riveting operation as a qualified operation can be performed.

[0080] In actual application, after the sample data is collected, a pretreatment operation can be performed on the sample data, that is, each sample data is sorted in ascending order according to the displacement corresponding to the pressure data, to obtain a data sequence. It should be understood that, since the riveting equipment records the pressure between the rivet and the plate and the displacement of the rivet inserted into the plate when performing the riveting operation, a corresponding relationship between the pressure and the displacement of the rivet inserted into the plate can be established, that is, each pressure data can correspond to the displacement of the rivet inserted into the plate. For the sample data, the pressure data corresponding to each displacement of the sample data can be sorted in ascending order to obtain the data sequence of each sample data.

[0081] After obtaining the data sequence, the data sequence can be divided into at least two data segments, and a sample matrix can be formed based on the divided data segments, that is, a matrix composed of each data segment is obtained, and then the sample matrix can be input into the neural network model to train the neural network model.

[0082] In actual application, after obtaining the data sequence of each sample data, that is, obtaining the data sequence of each positive sample data and obtaining the data sequence of each negative sample data, for each data sequence, the data sequence can be divided according to a predetermined data number to obtain at least two data segments, and a sample matrix can be generated based on each data segment, so as to obtain the sample matrix corresponding to each data sequence.

[0083] In specific implementation, since the sample data can be the pressure data between the rivet and the plate during the riveting operation, and the sample matrix corresponding to each sample data can represent the sample matrix corresponding to each riveting operation.

[0084] For example, the data sequence a can be composed of 256 pressure data, and the predetermined data number can be 16, and then the data sequence a can be divided according to the data number of 16 to obtain 16 data segments, and each data segment can include 16 pressure data. The 16 data segments can be stacked to generate a 16*16 sample matrix.

[0085] After obtaining the sample matrix corresponding to each data sequence, for each sample matrix, the sample matrix can be input into the neural network model, and the convolution layer, the maximum pooling layer and the full connection layer in the neural network model are operated in turn to output the probability that the riveting operation corresponding to the sample matrix is a qualified operation, or output the probability that the riveting operation corresponding to the sample matrix is an unqualified operation, and then the model parameters of the neural network model can be adjusted based on the output probability, so that the trained neural network model can determine the qualified probability of the riveting operation as a qualified operation.

[0086] In actual application, the riveting operation is performed at different positions, and the thickness, material and structure of the sheet metal at different positions are different, and thus the pressure between the rivet and the sheet metal at different positions is different, and the pressure of qualified operation corresponding to different operation stations is different. In addition, the riveting operation can include positioning, holding, piercing, extruding and forming operations, and the pressure between the rivet and the sheet metal in each operation of the riveting operation performed at different positions is different. If the same neural network model is used to determine the qualified probability of the riveting operation of all operation stations, the accuracy will be low or even unable to be determined. Therefore, a corresponding neural network model can be constructed for each operation station, and different operation stations can correspond to different neural network models. For each operation station, positive sample data and negative sample data in the operation station can be collected to train the neural network model corresponding to the operation station based on the positive sample data and the negative sample data of the operation station, so as to obtain the trained neural network model corresponding to each operation station.

[0087] In an embodiment of the present application, before step 102, the following steps can also be included:

[0088] According to the unqualified operation pressure data, the data sequence corresponding to the unqualified operation is generated, the first data segment in the data sequence is processed based on the first random number to obtain the processed first data segment, the second data segment in the data sequence is divided into at least two data segments, each data segment in the at least two data segments is processed based on the second random number corresponding to the data segment to obtain the processed second data segment, and the processed negative sample data is generated according to the processed first data segment and the processed second data segment, so as to train the neural network model based on the processed negative sample data.

[0089] The first random number can belong to a first random number range, and the second random number can belong to a second random number range. The first random number range and the second random number range can be the range of generated random numbers, and the first random number range and the second random number range are different. The first random number range and the second random number range can be determined based on the value of the sample data in the operation station. The second random number range is greater than the first random number range, and the average value of the second data segment is greater than the average value of the first data segment.

[0090] In actual application, after the sample data is preprocessed, the data enhancement operation can also be performed on the negative sample data.

[0091] Since the sample data can be obtained from actual production, i.e., the pressure data in each riveting operation is recorded, and it is determined whether each riveting operation is a qualified operation, and then in the case of a qualified operation, the pressure data of the riveting operation is determined as positive sample data, and in the case of an unqualified operation, the pressure data of the riveting operation is determined as negative sample data. It should be understood that since the number of negative sample data is small in actual production, it usually accounts for one thousandth of the total sample number, and then in order to improve the training efficiency of the neural network model and improve the accuracy of the neural network model after training, the data enhancement operation needs to be performed on the negative sample data.

[0092] Specifically, since the negative sample data can include pressure data corresponding to at least one unqualified operation, for each unqualified operation, the pressure data in the unqualified operation can be sorted according to the displacement size corresponding to the pressure data to generate a data sequence corresponding to the unqualified operation, and then the data sequence can be divided into a first data segment and a second data segment.

[0093] In an embodiment of the present application, before processing the first data segment in the data sequence based on the first random number, the following steps can also be included:

[0094] Determine the target abnormal period corresponding to the unqualified operation, and divide the data sequence into a first data segment and a second data segment based on the target abnormal period.

[0095] Among them, the second data segment can be an abnormal concentrated period, the first data segment can be a data segment in the data sequence except the second data segment, and the target abnormal period can be an abnormal concentrated period in the unqualified operation.

[0096] After obtaining the data sequence, the target abnormal period in the unqualified operation can be determined.

[0097] In actual application, the pressure data corresponding to the unqualified operation of each operation station in the past period of time can be obtained in advance, and for each operation station, the abnormal period in each unqualified operation corresponding to the operation station can be determined and counted, and then according to the abnormal period in each unqualified operation, the concentrated period of abnormality in the riveting operation of the operation station can be determined, i.e., the period with the largest abnormality proportion in all unqualified operations is the target abnormal period, and then the target abnormal period corresponding to each operation station can be determined.

[0098] In a specific implementation, the operation station corresponding to each negative sample data can be determined, and a target abnormal period corresponding to the operation station is determined, i.e., a target abnormal period corresponding to unqualified operation, and then the data segment in the target abnormal period from the unqualified operation is determined as the second data segment, and the data segment not in the target abnormal period is determined as the first data segment.

[0099] For example, the riveting operation can include positioning, holding, piercing, extruding, forming, etc., i.e., the riveting operation can include pressure data between the rivet and the sheet metal in each operation, so for each unqualified operation performed by the operation station in the past period, the operation in which an abnormality occurs in each unqualified operation of the operation station can be determined, and then the abnormal operation can be counted from the operation in which an abnormality occurs in each unqualified operation, i.e., the operation in which the number of abnormal occurrences is the most is determined as the abnormal operation, and the period in which the abnormal operation occurs is determined as the concentrated period in which the riveting operation occurs abnormally in the operation station.

[0100] For example, the operation station a performs 100 unqualified operations in the past 30 days, among which the piercing operation occurs abnormally 90 times, the extruding operation occurs abnormally 8 times, and the forming operation occurs abnormally 2 times, so the operation in which the number of abnormal occurrences is the most is determined as the piercing operation, the piercing operation is determined as the abnormal operation of the operation station a, and the period in which the piercing operation occurs is determined as the target abnormal period corresponding to the operation station a.

[0101] After the first data segment and the second data segment are determined, a first random number range corresponding to the first data segment and a second random number range corresponding to the second data segment can be determined.

[0102] In actual application, the first random number range can be determined based on the maximum value in all data of the first data segment, and similarly, the second random number range can be determined based on the maximum value in the second data segment. Generally, at least two first random number ranges can be defined in advance, and the correspondence between each first random number range and any numerical value is determined, so that after the maximum value in all data of the first data segment is determined, the corresponding first random number range of the maximum value in all data of the first data segment can be determined based on the maximum value in all data of the first data segment, and the second random number is the same.

[0103] After the first random number range and the second random number range are determined, a first random number can be randomly determined from the first random number range, and the first data segment is processed based on the first random number to obtain a processed first data segment.

[0104] In actual application, the first random number can be added to each data in the first data segment to obtain the processed first data segment.

[0105] Exemplarily, the first random number range can be [-2, 2], the first random number can be randomly selected as 1 from the first random number range, and then each data in the first data segment can be added based on the first random number, i.e., each data in the first data segment is added by 1, to obtain the processed first data segment.

[0106] After the first random number range and the second random number range are determined, the second data segment can be divided into at least two data segments.

[0107] In actual application, the second data segment can be divided according to a preset data quantity to obtain at least two data segments. Generally, the data quantity can be set as 20, i.e., the second data segment is divided according to the data quantity of 20 to obtain at least two data segments, and each data segment includes 20 data.

[0108] After the at least two data segments are determined, the second random number corresponding to each data segment can be randomly determined from the second random number range, and the corresponding data segment can be processed based on each second random number to obtain the processed second data segment.

[0109] In specific implementation, each data in the corresponding data segment can be added by the second random number to obtain the processed second data segment.

[0110] In an embodiment of the present application, the second random number corresponding to each data segment can be determined according to a predetermined order. For example, for the first data segment, the second random number corresponding to the first data segment can be randomly determined from the second random number range, and for the second data segment, the second random number range corresponding to the second data segment can be determined based on the second random number corresponding to the first data segment and the second random number range, and the second random number corresponding to the second data segment can be determined from the second random number range corresponding to the second data segment, and the second random number corresponding to the second data segment to the last data segment can be obtained in turn.

[0111] Exemplarily, the second random number range can be [a, b], the second random number corresponding to the first data segment can be k1, the second random number range corresponding to the second data segment can be [k1+a, k1+b], the second random number corresponding to the second data segment can be k2, the second random number range corresponding to the third data segment can be [k2+a, k2+b], and the second random number range corresponding to the third data segment to the last data segment and the second random number corresponding to the third data segment to the last data segment can be obtained in the same way.

[0112] After obtaining the processed first data segment and the processed second data segment, a processed negative sample data can be generated according to the processed first data segment and the processed second data segment, and the neural network model can be trained based on the processed negative sample data.

[0113] In an embodiment of the present application, the method can further include the following steps:

[0114] The target operation station is determined to be the operation station at which the riveting operation is performed on the sheet metal.

[0115] The target operation station can be an operation station at which the riveting operation is performed on the sheet metal.

[0116] After obtaining the pressure data of the riveting operation, the operation station at which the riveting operation is performed is determined as the target operation station.

[0117] In an embodiment of the present application, step 102 can include the following steps:

[0118] The pressure data is input into the neural network model corresponding to the target operation station to obtain a qualified probability of the riveting operation being a qualified operation.

[0119] Different operation stations can correspond to different neural network models.

[0120] After the target operation station is determined, the neural network model corresponding to the target operation station can be determined, i.e., the neural network model corresponding to the target operation station is determined from all neural network models.

[0121] In actual applications, since a neural network model corresponding to each operation station can be constructed in advance, the neural network model corresponding to the target operation station can be determined after the target operation station is determined.

[0122] For example, after constructing the neural network model corresponding to each operation station, a mapping relationship between each station identifier and the neural network model can be established according to the station identifier of the operation station, so that after the target operation station of the riveting operation is determined, the neural network model corresponding to the target operation station can be determined and called based on the station identifier of the target operation station and the mapping relationship between each station identifier and the neural network model.

[0123] After the corresponding neural network model is determined, the pressure data of the riveting operation can be input into the corresponding neural network model, and a qualified probability of the riveting operation being a qualified operation can be obtained after the operation of the corresponding neural network model, so as to generate the operation quality of the riveting operation based on the qualified probability.

[0124] In an embodiment of the present application, before step 102, the method can further include the following steps:

[0125] determine a target event topic corresponding to the riveting operation, and record the pressure data of the riveting operation in the target event topic.

[0126] The target event topic can be an event topic corresponding to the riveting operation among all event topics. The event topic can be related to a message type of a published message in a subscription and publication mode. Different message types can correspond to different event topics. The event topic can be used to receive a message related to the event topic published by a publisher and distribute the received message to a subscriber subscribing to the event topic.

[0127] After the pressure data of the riveting operation is obtained, the target event topic corresponding to the riveting operation can be determined.

[0128] In actual application, since each operation station can correspond to a neural network model, in order to input the pressure data of the riveting operation performed by each operation station into the corresponding neural network model, a corresponding event topic can be created for each operation station in advance, and a corresponding message queue can be created for different event topics. Different operation stations correspond to different event topics. Then, after the pressure data of the riveting operation is obtained, the pressure data can be published to a target event topic corresponding to a target operation station corresponding to the riveting operation, and the pressure data of the riveting operation is recorded in the target event topic, that is, the pressure data of the riveting operation is recorded in the message queue corresponding to the target event topic, so as to distribute the pressure data of the riveting operation to a subscriber subscribing to the target event topic through the target event topic.

[0129] In specific implementation, the riveting device can be a publisher in the subscription and publication mode, and then the pressure data of the riveting operation can be transmitted to the corresponding event topic after the pressure data of the riveting operation is obtained.

[0130] In an embodiment of the present application, step 102 can be implemented in the following manner:

[0131] A target thread subscribing to the target event topic is determined, the pressure data of the riveting operation is transmitted to the neural network model corresponding to the target operation station through the target thread, and a qualified probability of the riveting operation being a qualified operation is obtained.

[0132] The target thread can be a thread for transmitting data to the neural network model corresponding to the target operation station. Different threads can be used to transmit different data to corresponding neural network models.

[0133] After the target event topic corresponding to the riveting operation and the neural network model corresponding to the target operation station are determined, a target thread subscribing to the target event topic can be determined.

[0134] In actual application, a thread for transmitting data to a neural network model can be created in advance for each neural network model, and the thread can be used as a subscriber in a subscription publishing mode, different threads can correspond to different neural network models, and different threads can subscribe to different event topics, so that after determining a target event topic, a target thread subscribing to the target event topic can be determined.

[0135] After determining the target thread, the data in the target event topic can be obtained through the target thread, and the data can be transmitted to the neural network model corresponding to the target operation station to obtain the qualified probability of the riveting operation being a qualified operation.

[0136] In the embodiment of the present application, by obtaining pressure data reflecting the pressure between the sheet metal and the rivet during the riveting operation on the sheet metal, the qualified probability of the riveting operation being a qualified operation is determined according to the pressure data and the neural network model, and the operation quality of the riveting operation is determined based on the qualified probability, so that the qualified probability of the riveting operation can be determined through the neural network model, the manual sampling inspection is reduced, and the connection defect missed detection and other situations are reduced, thereby improving the quality of self-piercing riveting.

[0137] Referring to Figure 3 , Figure 3 A step flowchart of another connection process quality detection method provided by an embodiment of the present application is shown, which can specifically include the following steps:

[0138] Step 301, obtaining pressure data.

[0139] For related description of step 301, please refer to step 101, which will not be repeated here.

[0140] Step 302, determining the pressure change trend of the riveting operation according to the pressure data.

[0141] The pressure change trend can be used to reflect the pressure change between the sheet metal and the rivet during the riveting operation on the sheet metal in a time period.

[0142] In actual application, after obtaining the pressure data, the displacement corresponding to each pressure data can be determined, and then based on the value of each pressure data and the displacement corresponding to each pressure data, the relationship between the pressure data and the displacement in the riveting operation, i.e. the pressure change trend that the pressure data changes with the increase of displacement, can be determined.

[0143] In an embodiment of the present application, the pressure change trend can also be a pressure change trend that the pressure data changes with an increase of the execution time, and when the pressure change trend is the pressure change trend that the pressure data changes with the increase of the execution time, the acquisition time of each pressure data can be determined after the pressure data is obtained, and then the pressure change trend that the pressure data changes with the increase of the execution time in the riveting operation can be determined based on the acquisition time of each pressure data and the value of each pressure data.

[0144] In step 303, the pressure data is input into the neural network model to obtain the qualified probability of the riveting operation.

[0145] After the pressure data is obtained, the pressure data can be input into the neural network model, and then the qualified probability of the riveting operation can be obtained through the processing of the neural network model.

[0146] In step 304, the operation quality of the riveting operation is generated according to the pressure change trend and the qualified probability.

[0147] After the pressure change trend of the riveting operation and the qualified probability of the riveting operation are obtained, the operation quality of the riveting operation can be evaluated according to the pressure change trend and the qualified probability.

[0148] In actual application, the corresponding qualified operation and unqualified operation in each operation station can be determined in advance, and the qualified pressure change trend corresponding to each qualified operation and the error pressure change trend corresponding to each unqualified operation can be determined, and then after the pressure data of the riveting operation is obtained, the target operation station corresponding to the riveting operation can be determined, and the qualified pressure change trend and the error pressure change trend corresponding to the target operation station can be determined.

[0149] In specific implementation, the qualified probability output by the neural network model can be compared with the pre-set qualified probability threshold, and the pressure change trend of the riveting operation can be compared with the qualified pressure change trend, and then in the case that the qualified probability output by the neural network model is greater than or equal to the qualified probability threshold, and the similarity between the pressure change trend of the riveting operation and the qualified pressure change trend is greater than or equal to the similarity threshold, it can be determined that the riveting operation is a qualified operation, and the operation quality of the riveting operation is a connection defect that does not affect the connection strength, that is, the riveting operation does not have riveting quality problems.

[0150] And in the case that the qualified probability output by the neural network model is less than the qualified probability threshold, and / or the similarity between the pressure change trend of the riveting operation and the qualified pressure change trend is less than the similarity threshold, it can be determined that the riveting operation is an unqualified operation, and the operation quality of the riveting operation is a connection defect that affects the connection strength, that is, the riveting operation has riveting quality problems.

[0151] In an embodiment of the present application, the method further comprises the following steps:

[0152] According to the execution position of the riveting operation on the sheet metal, it is determined whether the riveting operation is a target riveting operation, and in the case where the riveting operation is a target riveting operation, a neural network model is used to identify a target feature in the pressure change trend.

[0153] The execution position can be a position at which the riveting operation is performed on the sheet metal, the target riveting operation can be a riveting operation in which a preset change feature exists in the pressure change trend, the preset change feature can represent a change that is obvious and related to the process of the riveting operation, the preset change feature can be determined based on the process of the riveting operation, and the target feature can be a feature in which the pressure change in a preset time period exceeds a preset change range.

[0154] In actual application, the execution position of the riveting operation can be determined, and then whether the riveting operation is a target riveting operation can be determined according to the execution position of the riveting operation.

[0155] In specific implementation, for each operation station, it can be determined in advance from the riveting operations of the operation station in the past period of time whether a target riveting operation exists, that is, whether a preset change feature exists in the pressure change trend of each riveting operation in the past period of time, if the preset change feature exists in the pressure change trend of one riveting operation, it can be determined that the riveting operation is a target riveting operation, and it can be determined that the operation station is an operation station in which a target riveting operation exists, and then it can be determined that the riveting operation in the target operation station is a target riveting operation in the case where the target operation station is an operation station in which a target riveting operation exists, and it can be determined that the riveting operation in the target operation station is a non-target riveting operation in the case where the target operation station is an operation station in which a target riveting operation does not exist.

[0156] In the case where the riveting operation is a target riveting operation, a neural network model can be used to identify a target feature in the pressure change trend of the riveting operation.

[0157] It needs to be understood that because the processes of the riveting operations performed at different positions can be different, the pressure change trends of each riveting operation will also be different, and the pressure change trend of the target riveting operation can have a change feature in which the pressure change exceeds a preset change range in a preset time period, that is, the pressure change trend of the target riveting operation can have a preset change feature, and for operation stations at different positions, the corresponding neural network model can be trained through sample data, and then it can be identified whether the pressure change trend of the riveting operation of the operation station has a change feature in which the pressure change exceeds a preset change range in a preset time period.

[0158] In actual application, when the riveting operation is determined as the target riveting operation, the neural network model corresponding to the target operation station can be called to identify the target feature in the pressure change trend of the riveting operation.

[0159] In an embodiment of the present application, step 304 can also be implemented in the following manner:

[0160] According to the target feature, the pressure change trend and the qualified probability, the operation quality of the riveting operation is generated.

[0161] After obtaining the target feature, the pressure change trend and the qualified probability, the operation quality of the riveting operation can be evaluated according to the target feature, the pressure change trend and the qualified probability.

[0162] In actual application, the target feature can be compared with the preset change feature, the qualified probability output by the neural network model can be compared with the preset qualified probability threshold, and the pressure change trend of the riveting operation can be compared with the qualified pressure change trend. Thus, when the target feature matches the preset change feature, the qualified probability output by the neural network model is greater than or equal to the qualified probability threshold, and the similarity between the pressure change trend of the riveting operation and the qualified pressure change trend is greater than or equal to the similarity threshold, it can be determined that the riveting operation is a qualified operation, and the operation quality of the riveting operation is that no connection defects affecting the connection strength are generated, i.e., the riveting operation has no riveting quality problem.

[0163] When the target feature does not match the preset change feature, or the qualified probability output by the neural network model is less than the qualified probability threshold, or the similarity between the pressure change trend of the riveting operation and the qualified pressure change trend is less than the similarity threshold, it can be determined that the riveting operation is an unqualified operation, and the operation quality of the riveting operation is that connection defects affecting the connection strength are generated, i.e., the riveting operation has a riveting quality problem.

[0164] In an embodiment of the present application, the following steps can also be included:

[0165] The pressure change trend is displayed to the user.

[0166] In actual application, after obtaining the pressure change trend corresponding to the riveting operation, the pressure change trend can be displayed through a display device. Specifically, the display device can be a device for displaying images. For example, a display, a mobile terminal with a display screen, and the like.

[0167] Referring to Figure 4 , Figure 4 A schematic diagram of a pressure change trend is shown, as Figure 4As shown, the abscissa can represent the displacement of the rivet inserted into the sheet metal, the ordinate can be the pressure between the rivet and the sheet metal, and the curve a can represent the corresponding pressure change trend of the riveting operation, i.e., the relationship between the displacement of the rivet inserted into the sheet metal and the pressure between the rivet and the sheet metal in the riveting operation.

[0168] It needs to be understood that, since at least one riveting operation can be performed on the same part of the sheet metal in the same operation station, i.e., at least one rivet can be inserted into the part to be connected, the pressure change trend corresponding to all riveting operations in the operation station can be displayed, i.e., the pressure change trend of the pressure between each rivet and the sheet metal is displayed.

[0169] In an embodiment of the present application, step 304 can also be performed in the following manner:

[0170] In response to the evaluation operation of the user on the pressure change trend, the quality evaluation information of the riveting operation evaluated by the user is generated, and the operation quality of the riveting operation is generated according to the quality evaluation information and the qualified probability.

[0171] The evaluation operation can be an operation in which the user evaluates whether the riveting operation is a qualified operation based on the pressure change trend, and the quality evaluation information can be information in which the user evaluates whether the riveting operation is a qualified operation. The quality evaluation information can include information in which the user evaluates that the riveting operation is a qualified operation and information in which the user evaluates that the riveting operation is an unqualified operation.

[0172] After the pressure change trend is displayed to the user, the user can evaluate whether the riveting operation is a qualified operation based on the displayed pressure change trend, and then the user can output the result of the evaluation and perform a corresponding evaluation operation, so that the quality evaluation information corresponding to the evaluation operation of the user can be generated in response to the evaluation operation of the user on the pressure change trend after the evaluation operation of the user is detected, and the operation quality of the riveting operation can be generated according to the quality evaluation information and the qualified probability.

[0173] In actual applications, the qualified probability output by the neural network model can be compared with a pre-set qualified probability threshold, and the pressure change trend of the riveting operation can be compared with a qualified pressure change trend, so that in the case that the quality evaluation information is information in which the user evaluates that the riveting operation is a qualified operation, the qualified probability output by the neural network model is greater than or equal to the qualified probability threshold, and the similarity between the pressure change trend of the riveting operation and the qualified pressure change trend is greater than or equal to a similarity threshold, it can be determined that the riveting operation is a qualified operation, and the operation quality of the riveting operation is determined to be a connection defect that does not affect the connection strength, i.e., the riveting operation does not have a riveting quality problem.

[0174] When the quality evaluation information is information that the user evaluates the riveting operation as an unqualified operation, or the qualified probability output by the neural network model is less than the qualified probability threshold, or the similarity between the pressure change trend of the riveting operation and the qualified pressure change trend is less than the similarity threshold, it can be determined that the riveting operation is an unqualified operation, and the operation quality of the riveting operation is a connection defect that affects the connection strength, that is, the riveting operation has a riveting quality problem.

[0175] In an embodiment of the present application, after obtaining the operation quality of the riveting operation, the operation quality of the riveting operation and the related information of the riveting operation can be recorded in the data warehouse, and the related information of the riveting operation can include the station identifier of the operation station, the plate identifier of the plate on which the riveting operation is performed, the position in the plate on which the riveting operation is performed, the time when the riveting operation is performed, the pressure change trend corresponding to the riveting operation, and the like. Furthermore, the data warehouse can record the operation quality of all riveting operations in the past period of time and the related information of each riveting operation, so as to provide data query, data analysis and the like for the user.

[0176] Specifically, the data query can be to query the pressure change trend of unqualified operations in the past period of time, or to query a specific unqualified operation in the past period of time, and the data analysis can be to analyze the pressure change trend of the riveting operation, or to statistically analyze the abnormal operation of unqualified operations in the past period of time.

[0177] In an embodiment of the present application, by obtaining the pressure data of the riveting operation, the pressure data being the pressure between the rivet and the plate, and determining the pressure change trend of the riveting operation according to the pressure data, the pressure data is input into the neural network model, the neural network model being used to determine the qualified probability of the riveting operation according to the pressure data, so as to obtain the qualified probability of the riveting operation, and generate the operation quality of the riveting operation according to the pressure change trend and the qualified probability, thereby reducing the manual sampling inspection, and reducing the connection defect missed detection and the like, and improving the quality of the self-piercing riveting.

[0178] Referring to Figure 5 , Figure 5 Fig. 1 shows a structural schematic diagram of a connection process quality detection system provided by an embodiment of the present application, as shown in the figure, the detection system can include: Figure 5

[0179] The data acquisition module 51 is configured to acquire pressure data between the rivet and the plate during the execution of the riveting operation by each operation station. The riveting operation can be an operation of riveting the plate.

[0180] ​Specifically, the data acquisition module 51 can be deployed in the riveting equipment of each operation station, and then when the riveting equipment performs the riveting operation, the pressure data between the rivet and the plate during the riveting operation can be acquired through the data acquisition module 51 of each operation station.

[0181] The quality detection module 52 is configured to distribute the pressure data to the neural network model corresponding to the operation station, obtain an output result of the neural network model, and generate an operation quality of the riveting operation performed at the operation station based on the output result.

[0182] The neural network model is configured to determine a qualified probability of the riveting operation being a qualified operation according to the pressure data, and generate the operation quality of the riveting operation based on the qualified probability, that is, the output result of the neural network model.

[0183] In an embodiment of the present application, the detection system can further include:

[0184] The data warehousing module 53 is configured to transmit the pressure data acquired by the data acquisition module 51 to a pre-created data warehouse, and the data warehouse can be a data collection integrating at least one data source and at least one time point.

[0185] The change data capture module 54 is configured to identify changed data from the data warehouse by using a change data capture technology, and transmit the changed data to the quality detection module 52.

[0186] In actual application, when the pressure data acquired by the data acquisition module 51 is stored in the data warehouse, the data in the data warehouse will change, so that the change data capture module 54 can identify the pressure data and transmit the pressure data to the quality detection module 52.

[0187] In an embodiment of the present application, the quality detection module 52 can deploy a corresponding neural network model for each operation station, and then after receiving the pressure data transmitted by the change data capture module 54, the quality detection module 52 can distribute the received pressure data to the corresponding neural network model.

[0188] For example, the quality detection module 52 can distribute the received pressure data to the corresponding neural network model by using a subscription and publication mode.

[0189] In actual application, at least one event topic can be pre-created, and a message queue can be allocated for each event topic, and each event topic can be related to a type of message, and the type of message can be defined as the type of operation station, that is, different operation stations can correspond to different event topics, and the quality detection module 52 can deploy the message queue corresponding to each event topic.

[0190] In a specific implementation, after receiving the pressure data transmitted by the change data capturing module 54, the quality detection module 52 can determine the operation station corresponding to the pressure data, i.e., determine the riveting operation that produces the pressure data and the operation station that performs the riveting operation, and further determine the event topic corresponding to the operation station and determine the message queue corresponding to the event topic, and transmit the pressure data to the message queue.

[0191] For each neural network model in the quality detection module 52, a thread for transmitting the pressure data in the message queue to the neural network model can be created in advance for each neural network model, and different neural network models can correspond to one of the threads. Since each neural network model can correspond to a different operation station, and different operation stations can correspond to different event topics, different neural network models can correspond to different event topics. For each neural network model, the thread corresponding to the neural network model can subscribe to the event topic corresponding to the neural network model, the thread corresponding to the neural network model can transmit the pressure data in the message queue corresponding to the event topic to the neural network model, obtain the output result of the neural network model, and generate the operation quality of the riveting operation performed by the operation station based on the output result.

[0192] In an embodiment of the present application, the detection system further comprises:

[0193] The visualization module 55 is configured to display the pressure data to the user in the case that the operation quality indicates that the riveting operation is a qualified operation.

[0194] For example, the visualization module 55 can be further configured to display the pressure change trend of the riveting operation to the user in the case that the riveting operation is a target riveting operation, receive and respond to the evaluation operation of the user on the pressure change trend of the riveting operation, and generate quality evaluation information to transmit the quality evaluation information to the quality detection module 52 for quality evaluation.

[0195] In an embodiment of the present application, the visualization module 55 can be further configured to display the pressure change trend of the riveting operation of each operation station in the past period of time to the user.

[0196] In an embodiment of the present application, the quality detection module 52 is further configured to generate alarm information and push the alarm information to the user to alarm the user in the case that the operation quality indicates that the riveting operation is an unqualified operation, and transmit the pressure data to the visualization module 55 to display the pressure data of the riveting operation to the user in the case that the operation quality indicates that the riveting operation is a qualified operation.

[0197] Exemplarily, the quality detection module 52 is further configured to determine whether the riveting operation is a target riveting operation, and transmit, in a case where the riveting operation is the target riveting operation, a pressure change trend of the riveting operation to the visualization module 55 for display by the visualization module 55, and configured to, after receiving quality assessment information transmitted by the visualization module 55, distribute the pressure data to the neural network model corresponding to the operation station to obtain a qualified probability that the riveting operation is a qualified operation, and generate operation quality of the riveting operation based on the quality assessment information and the qualified probability.

[0198] In an embodiment of the present application, the detection system further comprises:

[0199] The training module 56 is configured to obtain sample data of each operation station, and train the neural network model based on the sample data to obtain a trained neural network model, and deploy the trained neural network model in the quality detection module 52.

[0200] In actual application, the training module can be further configured to, after obtaining the sample data of each operation station, perform a preprocessing operation on the sample data and a data enhancement operation on negative sample data in the sample data, and train the neural network model based on the processed sample data.

[0201] In an embodiment of the present application, by obtaining the pressure data, the pressure data is used to reflect the pressure between the sheet metal and the rivet when the riveting operation is performed on the sheet metal, and according to the pressure data and the neural network model, a qualified probability that the riveting operation is a qualified operation is determined to determine the operation quality of the riveting operation based on the qualified probability, so as to determine the qualified probability of the riveting operation through the neural network model, reduce the manual sampling inspection, and thus reduce the connection defect missed detection and improve the quality of the self-piercing riveting.

[0202] Referring to Figure 6 , Figure 6 A flowchart of a connection process quality detection method provided by an embodiment of the present application is shown, which can specifically include the following steps:

[0203] Step 601: The riveting equipment collects pressure data of a riveting operation;

[0204] Step 602: The riveting equipment stores the pressure data in a data warehouse;

[0205] Step 603: The change data capture module obtains the pressure data stored in the data warehouse, and publishes the pressure data to an event topic corresponding to a target operation station of the riveting operation;

[0206] Specifically, the change data capture module can use the change data capture technology to obtain the data changed in the data warehouse;

[0207] Step 604, determine the target thread subscribing to the event topic corresponding to the riveting operation, and distribute the pressure data published in the event topic corresponding to the riveting operation to the neural network model corresponding to the target operation station through the target thread to obtain the qualified probability output by the neural network model;

[0208] Step 605, determine whether the riveting operation is a qualified operation according to the qualified probability, if yes, the process ends, if not, execute step 606;

[0209] Step 606, generate alarm information for the riveting operation, and push the alarm information to the user to alarm the user.

[0210] In the embodiment of the application, the pressure data of the riveting operation is collected by the riveting equipment and stored in the data warehouse, and then the change data capture module obtains the pressure data stored in the data warehouse and publishes the pressure data to the event topic corresponding to the target operation station of the riveting operation. The target thread subscribing to the event topic corresponding to the riveting operation is determined, and the pressure data published in the event topic corresponding to the riveting operation is distributed to the neural network model corresponding to the target operation station through the target thread to obtain the qualified probability output by the neural network model. The situation of manual sampling inspection is reduced, thereby reducing the situation of connection defect missed detection and improving the quality of self-punching riveting. According to the qualified probability, it is determined whether the riveting operation is a qualified operation, and in the case that the riveting operation is an unqualified operation, alarm information for the riveting operation is generated and pushed to the user to alarm the user.

[0211] Referring to Figure 7 , Figure 7 The structure of a connection process quality detection device provided by an embodiment of the application is shown, which can specifically include the following modules:

[0212] The acquisition module 701 is configured to acquire pressure data, wherein the pressure data is used to reflect the pressure between the sheet metal and the rivet when the riveting operation is performed on the sheet metal.

[0213] The determination module 702 is configured to determine a qualified probability of the riveting operation being a qualified operation according to the pressure data and a neural network model, so as to determine the operation quality of the riveting operation based on the qualified probability. The neural network model is obtained by training a plurality of sample data, and the sample data includes positive sample data and negative sample data. The positive sample data includes pressure data between the sheet metal and the rivet when the riveting operation performed on the sheet metal is a qualified operation, and the negative sample data includes pressure data between the sheet metal and the rivet when the riveting operation performed on the sheet metal is an unqualified operation.

[0214] In an implementation manner, the determining module 702 can be further configured to:

[0215] determine a target operation station at which the sheet metal is located when the riveting operation is performed on the sheet metal;

[0216] input the pressure data into a neural network model corresponding to the target operation station to obtain a qualified probability that the riveting operation is a qualified operation; different operation stations correspond to different neural network models, and a neural network model corresponding to any operation station is used to determine a qualified probability that the riveting operation performed on the sheet metal at the operation station is a qualified operation.

[0217] In an implementation manner, the apparatus further includes the following modules:

[0218] a recording module configured to determine a target event theme corresponding to the riveting operation before inputting the pressure data into the neural network model corresponding to the target operation station to obtain the qualified probability that the riveting operation is a qualified operation, and record the pressure data of the riveting operation in the target event theme;

[0219] In an implementation manner, the determining module 702 can be further configured to:

[0220] determine a target thread that subscribes to the target event theme; the target thread is a thread that transmits data to the neural network model corresponding to the target operation station, and different threads are used to transmit different data to corresponding neural network models;

[0221] transmit the pressure data in the target event theme to the neural network model corresponding to the target operation station through the target thread to obtain the qualified probability that the riveting operation is a qualified operation.

[0222] In an implementation manner, the determining module 702 can be further configured to:

[0223] generate a data sequence corresponding to the unqualified operation according to pressure data of the unqualified operation before determining the qualified probability that the riveting operation is a qualified operation according to the pressure data and the neural network model;

[0224] process a first data segment in the data sequence based on a first random number to obtain a processed first data segment; the first random number belongs to a first random number range,

[0225] divide a second data segment in the data sequence into at least two data segments; an average value of the second data segment is greater than an average value of the first data segment,

[0226] process the first data segment in the data sequence based on the first random number to obtain a processed first data segment;

[0227] generate processed negative sample data based on the processed first data segment and the processed second data segment, and train the neural network model based on the processed negative sample data.

[0228] In an implementation manner, the determination module 702 can be further configured to:

[0229] determine a target abnormal time period corresponding to the unqualified operation before processing the first data segment in the data sequence based on the first random number to obtain a processed first data segment; the target abnormal time period is a concentrated time period in which an anomaly occurs in the unqualified operation;

[0230] divide the data sequence into the first data segment and a second data segment based on the target abnormal time period; the second data segment is the concentrated time period in which the anomaly occurs.

[0231] In an implementation manner, the determination module 702 can be further configured to:

[0232] determine a pressure change trend of the riveting operation according to the pressure data; the pressure change trend is used to reflect a pressure change between the sheet metal and the rivet when the riveting operation is performed on the sheet metal in a time period;

[0233] determine an operation quality of the riveting operation according to the pressure change trend and the qualified probability.

[0234] In an implementation manner, the determination module 702 can be further configured to:

[0235] determine whether the riveting operation is a target riveting operation according to an execution position of the riveting operation on the sheet metal; the target riveting operation is a riveting operation in which a preset change feature exists in the pressure change trend;

[0236] when the riveting operation is the target riveting operation, identify a target feature in the pressure change trend by using the neural network model;

[0237] determine the operation quality of the riveting operation according to the target feature, the pressure change trend, and the qualified probability.

[0238] In an implementation manner, the apparatus further includes the following modules:

[0239] The display module is configured to display the pressure change trend to a user.

[0240] In an implementation manner, the determination module 702 can be further configured to:

[0241] detecting an evaluation operation on the pressure change trend;

[0242] in response to the evaluation operation, determining quality evaluation information for the riveting operation;

[0243] determining the operation quality of the riveting operation according to the quality evaluation information and the qualified probability.

[0244] In the embodiments of the present application, by acquiring pressure data, the pressure data is used to reflect the pressure between the sheet metal and the rivet when the riveting operation is performed on the sheet metal, according to the pressure data and the neural network model, the qualified probability of the riveting operation is determined, and the operation quality of the riveting operation is determined based on the qualified probability, so as to determine the qualified probability of the riveting operation through the neural network model, reduce the manual sampling inspection, and thus reduce the connection defect missed detection and improve the quality of the self-piercing riveting.

[0245] It should be noted that the information interaction, execution process and the like between the above devices, since based on the same concept as the method embodiments of the present application, the specific functions and the technical effects brought about can be referred to the method embodiments part, and will not be repeated here.

[0246] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of each functional unit and module is exemplified, and in actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiments can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The above integrated unit can be realized in the form of hardware or software. In addition, the specific names of each functional unit and module are only for easy distinction, and do not limit the protection scope of the present application. The specific working process of the unit and module in the above system can be referred to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0247] Referring to Figure 8 , Figure 8 shows a structural block diagram of a terminal device provided by an embodiment of the present application, as Figure 8 shown, the present embodiment provides a terminal device 81, which comprises at least one processor 811, a memory 812, and a computer program 8121 stored in the memory 812 and executable on the at least one processor 811. When the processor 811 executes the computer program 8121, the steps in any of the above method embodiments are implemented.

[0248] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the steps in any one of the method embodiments.

[0249] The embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device is enabled to implement the steps in each of the method embodiments.

[0250] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the embodiment of the present application can implement all or part of the processes in the above method embodiments by a computer program to instruct related hardware to complete. The computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each of the method embodiments can be implemented. The computer program includes computer program code. The computer program code can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium at least includes any entity or device capable of carrying the computer program code to the terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium.

[0251] The above embodiments are only used to illustrate the technical solutions of the present application, rather than limit the same. Although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent ones. The modification or replacement does not change the essence of the corresponding technical solutions, and should be included in the protection scope of the present application.

Claims

1. A method for detecting the quality of a joining process, characterized in that, The method includes: Acquire pressure data; wherein the pressure data is used to reflect the pressure between the sheet metal and the rivet when a riveting operation is performed on the sheet metal; Based on the pressure data and the neural network model, the probability of the riveting operation being a qualified operation is determined, and the operation quality of the riveting operation is determined based on the qualified probability. The neural network model is trained from multiple sample data, which includes positive sample data and negative sample data. The positive sample data includes the pressure data between the sheet metal and the rivet when the riveting operation performed on the sheet metal is a qualified operation, and the negative sample data includes the pressure data between the sheet metal and the rivet when the riveting operation performed on the sheet metal is a unqualified operation.

2. The method for detecting the quality of the connection process as described in claim 1, characterized in that, The method further includes: Determine the target operating station where the sheet metal is located when the riveting operation is performed on the sheet metal; The step of determining the success probability of the riveting operation being a qualified operation based on the pressure data and the neural network model includes: The pressure data is input into the neural network model corresponding to the target operation station to obtain the pass probability of the riveting operation being a qualified operation; wherein, different operation stations correspond to different neural network models, and the neural network model corresponding to any operation station is used to determine the pass probability of the riveting operation performed on the sheet metal at the operation station being a qualified operation.

3. The method for detecting the quality of the connection process as described in claim 2, characterized in that, Before inputting the pressure data into the neural network model corresponding to the target operation station to obtain the probability that the riveting operation is a qualified operation, the method further includes: Determine the target event topic corresponding to the riveting operation, and record the pressure data of the riveting operation in the target event topic; The step of inputting the pressure data into the neural network model corresponding to the target operation station to obtain the pass probability of the riveting operation being a qualified operation includes: Determine the target thread that subscribes to the target event topic; wherein, the target thread is the thread that transmits data to the neural network model corresponding to the target operation station, and different threads are used to transmit different data to the corresponding neural network model; The pressure data in the target event topic is transmitted to the neural network model corresponding to the target operation station through the target thread to obtain the pass probability of the riveting operation being a qualified operation.

4. The method for detecting the quality of the connection process as described in any one of claims 1 to 3, characterized in that, Before determining the probability of the riveting operation being a qualified operation based on the pressure data and the neural network model, the method further includes: Based on the pressure data of the non-compliant operation, generate a data sequence corresponding to the non-compliant operation; The first data segment in the data sequence is processed based on the first random number to obtain the processed first data segment; wherein the first random number belongs to the range of the first random number. The second data segment in the data sequence is divided into at least two data segments; wherein the average value of the second data segment is greater than the average value of the first data segment; The corresponding data segments are processed based on the second random number corresponding to each of the at least two data segments to obtain the processed second data segment; wherein the second random number belongs to the range of the second random number, and the range of the second random number is greater than the range of the first random number. Based on the processed first data segment and the processed second data segment, processed negative sample data is generated to train the neural network model.

5. The method for detecting the quality of the connection process as described in claim 4, characterized in that, Before processing the first data segment in the data sequence based on the first random number to obtain the processed first data segment, the method further includes: Determine the target abnormal period corresponding to the non-conforming operation; wherein, the target abnormal period is the concentrated period during which abnormalities occur in the non-conforming operation; Based on the target abnormal period, the data sequence is divided into a first data segment and a second data segment; wherein, the second data segment is the concentrated period in which the abnormality occurs.

6. The method for detecting the quality of the connection process as described in any one of claims 1 to 3 or 5, characterized in that, The method further includes: Based on the pressure data, the pressure change trend of the riveting operation is determined; wherein, the pressure change trend is used to reflect the pressure change between the sheet metal and the rivet when the riveting operation is performed on the sheet metal within a time period; The step of determining the operational quality of the riveting operation based on the pass probability includes: The operational quality of the riveting operation is determined based on the pressure change trend and the pass rate.

7. The method for detecting the quality of the connection process as described in claim 6, characterized in that, The method further includes: Based on the location of the riveting operation on the sheet metal, determine whether the riveting operation is a target riveting operation; wherein, the target riveting operation is a riveting operation that exhibits a preset change characteristic in the pressure change trend; When the riveting operation is a target riveting operation, the neural network model is used to identify the target features in the pressure change trend; Determining the operational quality of the riveting operation based on the pressure change trend and the probability includes: The operational quality of the riveting operation is determined based on the target characteristics, the pressure change trend, and the pass probability.

8. The method for detecting the quality of the connection process as described in claim 6, characterized in that, The method further includes: This demonstrates the trend of pressure changes; The step of determining the operational quality of the riveting operation based on the pass probability includes: The assessment operation is performed to detect the pressure change trend; In response to the evaluation operation, quality evaluation information for the riveting operation is determined; The operational quality of the riveting operation is determined based on the quality assessment information and the pass rate.

9. A device for detecting the quality of a connection process, characterized in that, The device includes: An acquisition module is used to acquire pressure data; wherein the pressure data is used to reflect the pressure between the sheet metal and the rivet when a riveting operation is performed on the sheet metal; A determination module is used to determine the probability of the riveting operation being a qualified operation based on the pressure data and the neural network model, so as to determine the operation quality of the riveting operation based on the qualification probability; the neural network model is trained by multiple sample data, the sample data including positive sample data and negative sample data, the positive sample data including the pressure data between the sheet metal and the rivet when the riveting operation performed on the sheet metal is a qualified operation, and the negative sample data including the pressure data between the sheet metal and the rivet when the riveting operation performed on the sheet metal is a unqualified operation.

10. A system for detecting the quality of a connection process, characterized in that, The system includes: The data acquisition module is used to collect the pressure data between the rivet and the sheet metal during the riveting operation at each workstation; wherein, the riveting operation is the operation of riveting the sheet metal. The quality inspection module is used to distribute the pressure data to the neural network model corresponding to the operation station, obtain the output result of the neural network model, and generate the operation quality of performing the riveting operation at the operation station based on the output result.

Images