Workpiece welding quality prediction device and method based on opc ua architecture

Abstract

The application discloses a workpiece welding quality prediction device and method based on an OPC UA architecture, which comprises a data acquisition and storage layer, an application layer and a device layer for acquiring production data of a workpiece welding process, the data acquisition and storage layer comprises an OPC UA server, the OPC UA server is in communication connection with the device layer, the application layer comprises an OPC UA client, a welding quality prediction system and an MES system, the OPC UA client is in communication connection with the OPC UA server, displays the production data and compares the production data with a set range, the welding quality prediction system is in communication connection with the data acquisition and storage layer, and the production data is submitted to the welding quality prediction system for welding quality prediction, and the MES system is used for marking workpieces whose production data is not in the set range or workpieces whose welding quality prediction is unqualified. The application can well solve the inconvenience caused by different device communication protocols, can quickly and orderly acquire production data of the workpieces in a welding workshop production process, and provides a basis for subsequent quality prediction and quality tracing.

Description

Technical Field

[0001] This invention relates to workpiece welding quality prediction, and more specifically to a workpiece welding quality prediction device and method based on the OPC UA architecture. Background Technology

[0002] Automotive production welding workshops employ various types of welding machines. During operation, it's crucial to record parameters such as voltage, current, and temperature, as this data is vital for quality prediction and problem tracing. Previously, collecting and accessing data from field equipment required each application software developer to write dedicated interface functions. Due to the diverse range of field equipment and continuous product upgrades, this often placed a significant workload on both users and software developers. This approach typically fails to meet actual needs, and system integrators and developers urgently require a plug-and-play device driver that is efficient, reliable, open, and interoperable. The issue of semantic interoperability between different CNC equipment remains unresolved. Currently, most research and applications focus on CNC equipment data acquisition, manufacturing workshop data acquisition, and monitoring systems, achieving basic network interconnection. However, due to data heterogeneity between the proprietary data acquisition protocols of different CNC equipment, semantic interoperability is impossible, hindering further development of workshop digitization. The difficulty in data processing further complicates welding quality prediction and traceability.

[0003] In terms of welding quality prediction, the lack of rapid real-time data acquisition and transmission methods makes real-time prediction difficult. Furthermore, many manufacturers treat this as a separate component, using the same predictive model for extended periods after its initial setup, leading to delays in model updates and increasing the probability of prediction errors. Regarding welding quality traceability, the car body enters a quality inspection phase after welding. If the quality inspection fails, the car body needs to be reprocessed, requiring the acquisition of various welding process data to determine the distribution of defective products and identify the causes of defects. Currently, after obtaining production data on defective car bodies, many manufacturers simply store the data or only submit it to the quality prediction system after a certain amount has been stored, resulting in untimely updates to the quality prediction system.

[0004] The diverse communication protocols used by various welding machines and sensors necessitate significant costly integration of collected data in welding workshops. The inability to monitor and record critical parameters such as voltage and current in real time during welding compromises product quality. The lack of a unified communication protocol forces Manufacturing Execution Systems (MES) to passively integrate various protocols and adapt them to different devices during design, increasing development complexity and hindering overall workshop informatization. Furthermore, the absence of effective data processing methods significantly impedes subsequent quality prediction models for real-time welding quality forecasting and quality traceability.

[0005] The lack of a rapid real-time data acquisition and transmission method makes it difficult to achieve real-time prediction. With welding quality prediction and welding quality traceability being separated, the quality prediction system cannot quickly utilize the latest samples to update the structure, which is not conducive to actual production and has room for improvement. Summary of the Invention

[0006] The purpose of this invention is to provide a workpiece welding quality prediction device and method based on OPC UA architecture, which can effectively solve the inconvenience caused by different equipment communication protocols, and can quickly and orderly collect workpiece production data during the welding workshop production process, providing a foundation for subsequent quality prediction and quality traceability.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A workpiece welding quality prediction device based on OPC UA architecture includes a device layer, a data acquisition and storage layer, and an application layer. The device layer acquires production data from the workpiece welding process. The data acquisition and storage layer includes an OPC UA server, which is communicatively connected to the device layer and serves as both a data acquisition endpoint and a data transmission endpoint. The application layer includes an OPC UA client, a welding quality prediction system, and a MES system. The OPC UA client is communicatively connected to the OPC UA server and serves as a data transmission receiver, used to display production data and compare it with a set range. The welding quality prediction system is communicatively connected to the data acquisition and storage layer and submits the workpiece welding process production data to the welding quality prediction system for welding quality prediction. The MES system is used to mark workpieces whose production data is outside the set range or whose welding quality prediction is unqualified.

[0009] Furthermore, the data acquisition and storage layer also includes a database for storing production data.

[0010] Furthermore, the welding quality prediction system integrates a welding quality prediction model based on a convolutional neural network. The welding quality prediction model takes the production data of the workpiece welding process as input and the welding quality level as output.

[0011] Furthermore, if a welded workpiece is deemed unqualified during subsequent quality inspection during the production process, the production data of the unqualified workpiece is extracted and used to train the welding quality prediction model, thereby achieving real-time updates and improvements to the welding quality prediction model.

[0012] Furthermore, the production data includes welding process parameters and workpiece size data.

[0013] A method for predicting workpiece welding quality based on OPC UA architecture includes the following steps:

[0014] S1, the workpiece begins welding processing, and the data acquisition and storage layer collects the production data of the current workpiece welding process through the equipment layer;

[0015] S2, the collected production data is sent to the OPC UA client in the application layer for display, and the production data is compared with the set range. If the production data of a certain workpiece is within the set range, then proceed to S3. If the production data of a certain workpiece is not within the set range, then the workpiece is marked by the MES system and key quality inspection is carried out after the workpiece welding is completed.

[0016] S3 submits the collected production data to the welding quality prediction system for quality prediction to determine whether the welding quality of the workpiece is qualified.

[0017] Furthermore, if the prediction result in S3 indicates that the workpiece is unqualified, the workpiece will be marked through the MES system, and key quality inspection will be carried out after the workpiece welding is completed.

[0018] Furthermore, it also includes S4, where a quality inspection is performed after the workpiece welding is completed. If the workpiece fails the quality inspection, the production data of the workpiece is extracted and used to update the welding quality prediction model in the welding quality prediction system.

[0019] The beneficial effects of this invention are:

[0020] 1. This invention adopts an OPC UA architecture to solve the inconvenience caused by different device communication protocols, and can quickly and orderly collect production data in the welding workshop production process, providing a foundation for subsequent quality prediction and quality traceability.

[0021] 2. This invention utilizes a welding quality prediction model based on convolutional neural networks to predict the welding quality of workpieces in real time. If the prediction result indicates that the workpiece is unqualified, the workpiece is marked through the MES system, and key quality inspection is carried out after the workpiece welding is completed.

[0022] 3. This invention combines quality traceability to extract production data of defective workpieces for updating the welding quality prediction model in the welding quality prediction system, thereby maintaining the growth and improvement of the welding quality prediction model. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the workpiece welding quality prediction device based on the OPC UA architecture described in this invention.

[0024] Figure 2 This is a flowchart of the workpiece welding quality prediction method based on the OPC UA architecture described in this invention;

[0025] Figure 3 This is a schematic diagram of the working mode of the OPC UA architecture;

[0026] Figure 4 This is a schematic diagram of the OPC UA architecture;

[0027] Figure 5 This is a diagram illustrating the connection between an OPC UA client and an OPC UA server.

[0028] Figure 6 This is a schematic diagram of the structure of a convolutional neural network;

[0029] Figure 7 This is a schematic diagram of the convolutional neural network structure used in this invention to train a welding quality prediction model. Detailed Implementation

[0030] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0031] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0032] See Figure 1The workpiece welding quality prediction device based on the OPC UA architecture shown includes an equipment layer, a data acquisition and storage layer, and an application layer. The equipment layer includes welding equipment and various sensors for acquiring production data of the workpiece welding process, including welding process parameters and workpiece dimension data. The data acquisition and storage layer includes an OPC UA server and a database for storing production data. The OPC UA server is communicatively connected to the equipment layer, serving as both the data acquisition end and the data transmission end. The application layer includes an OPC UA client, a welding quality prediction system, and a MES system. The OPC UA client is communicatively connected to the OPC UA server, serving as the data transmission receiver, displaying production data, and comparing it with a set range. The welding quality prediction system is communicatively connected to the data acquisition and storage layer, submitting the workpiece welding process production data to the welding quality prediction system for welding quality prediction. The MES system is used to mark workpieces whose production data is outside the set range or whose welding quality prediction is unqualified.

[0033] The OPC UA architecture consists of an OPC UA client and an OPC UA server. The OPC UA server is responsible for data acquisition from the underlying devices and responding to client requests. The OPC UA client sends requests to the OPC UA server, and after responding, the OPC UA server sends the received information to the user. For the working modes of the OPC UA client and OPC UA server, please refer to [link to documentation]. Figure 3 .

[0034] See Figure 4 The specific steps for OPC UA client to interact with server are as follows:

[0035] ① The OPC UA client sends a service request;

[0036] ②The service request is sent to the server after passing through the OPC UA communication stack;

[0037] ③After receiving a request, the OPC UA server calls the corresponding service set to execute the task on the node in the address space or in modules such as monitoring and subscription;

[0038] ④ The OPC UA server returns the corresponding request response to the OPC UA client;

[0039] ⑤ The response is sent to the OPC UA client via the OPC UA communication stack;

[0040] ⑥ The OPC UA client processes the information returned by the server;

[0041] ⑦ The OPC UA client returns node information and other data obtained from the OPC UA server's address space to the user.

[0042] The principle of applying the OPC UA protocol in the workshop is to embed the OPC UA server into various underlying production equipment in the workshop, and map the data collected by various CNC systems to the address space of the OPC UA server. After the OPC UA client connects to the server, data communication begins.

[0043] Data transmission is performed using an OPC UA architecture; see [link / reference]. Figure 5 The specific steps include:

[0044] ①Find the server

[0045] The OPC UA client first needs to establish a connection with the discovery server, and then obtain the list of registered OPC UA servers through the discovery server, thereby enabling data communication.

[0046] The OPC UA server registers itself by calling the RegisterServer() method. After the OPC UA client establishes a connection with the discovery server, the client uses the FindServers() method provided by the discovery server to find all registered OPC UA servers, thereby obtaining the necessary descriptive information about the OPC UA server, including all information related to the connection process. Finally, the client calls the GetEndpoints() method to obtain the OPC UA server's endpoint information, which includes the IP address and security settings required to establish a connection between the OPC UA client and the OPC UA server.

[0047] ② Browse the address space

[0048] The address space is a collection of information with multiple meanings within the OPC UA server. OPC UA unifies the address space, services, and security model, providing a unified data interface to achieve unified access to information and improve data communication speed.

[0049] The address space of OPC UA consists of nodes within the OPC UA server. These nodes are also known as the basic units of the address space. Data within the address space is represented through these nodes. OPC UA nodes come in various types, each possessing both common and type-specific attributes. The three most important node types are: object nodes, variable nodes, and method nodes.

[0050] Nodes of object type in the address space manage the object itself, as well as its contained variables, methods, and bound events. OPC UA clients can obtain the variable values ​​of object nodes in the OPC UA server's address space by reading and subscribing. Variable type nodes store low-level information about the workshop equipment obtained by the OPC UA server. The values ​​of variable nodes in the address space can be read or modified by the OPC UA client, and changes to variable node data can be monitored by subscribing to them. Method nodes are configured by the OPC UA server, including the setting of OPC UA client input parameters and the format of the returned output results, generally involving simple data processing.

[0051] Data generated by workshop equipment or sensors is bound to the server's address space in the form of nodes, with each node object representing a data item. OPC UA clients access the address space of workshop equipment, sensors, etc., to obtain data through browsing, reading, and subscription.

[0052] The OPC UA client uses the Browse service to submit the NodeId of the initial node and the filtering conditions to the OPC UA server, and then searches for a list of all nodes in the server's address space.

[0053] After obtaining the list of nodes, the client uses nodesToRead as the input parameter and reads the attribute values ​​of the nodes to be accessed from the address space using the Read service, and displays the reading results using the updateAttributeList() function.

[0054] ③ Subscribe to data

[0055] Real-time data acquisition is achieved through OPC UA's periodic data retrieval using a publish / subscribe model. First, the OPC UA client initializes a subscription service and creates monitoring items. Then, the OPC UA server samples the monitoring items at equal intervals and returns the data to the OPC UA client.

[0056] The subscription mechanism primarily relies on the methods CreateSubscription, Publish, DeleteSubscription, CreateMonitoredItems, DeleteMonitoredItems, and ModifyMonitoredItems, as shown in Table 1:

[0057] Table 1 Subscription Implementation Functions and Their Meanings

[0058] function Function meaning CreateSubscription() Create a new subscription item Publish() Publish subscription items DeleteSubscription() Delete subscription CreateMonitoredItems() Create a new monitoring item ModifyMonitoredItems() Modify monitoring items DeleteMonitoredItems() Delete monitoring item

[0059] When creating a subscription item using `CreateSubscription()`, you can set the subscription interval, modify the subscription interval, request the usage period, and modify the usage period. When creating a monitoring item using `CreateMonitoredItems()`, you need to specify the list of results to be returned.

[0060] When inspecting the welding quality of workpieces, car manufacturers have their own grading methods for workpiece welding quality. Based on different production data and corresponding welding quality levels, they train welding quality prediction models using convolutional neural network models.

[0061] The welding quality prediction system integrates a welding quality prediction model based on a convolutional neural network. The welding quality prediction model takes the production data of the workpiece welding process as input and the welding quality level as output.

[0062] See Figure 6 A Convolutional Neural Network (CNN) is a deep neural network with a convolutional structure, including an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. Furthermore, CNNs employ local connectivity and weight sharing, reducing network complexity and increasing the network's learning rate.

[0063] Convolutional layers are primarily used to learn features. The input data for a convolutional layer is the output data from the input layer or pooling layer. Each convolutional kernel is convolved with the feature map from the previous layer, and after passing through an activation function, the output feature map is obtained. The formula for the convolution process is: In the formula: l is the number of layers in the convolutional neural network structure, and j is the j-th channel. Let f be the output of the j-th channel of convolutional layer l, and f be the activation function; N j A subset of feature maps; * represents the convolution kernel; * represents the convolution operation. For additional bias.

[0064] Pooling layers, also called downsampling layers, reduce the dimensionality of feature maps extracted by convolutional layers, effectively reducing the computational cost of the network. Common pooling methods include average pooling and max pooling. Max pooling calculates the maximum value of the feature map input into the pooling window. The pooling layer calculation formula is: In the formula: l represents the number of layers in the convolutional neural network structure, down is the pooling operation, and w and These represent the pooling weights and additional biases, respectively.

[0065] In a fully connected network, the feature maps from the previous layer are concatenated into a one-dimensional feature map, which is then used as input to the fully connected network for computation. The computation formula is: In the formula: w j For fully connected network weights.

[0066] The training process of a convolutional neural network (CNN) is similar to that of a traditional neural network, consisting of two stages: forward propagation and backward propagation. In the forward propagation stage, sample data from the training set is input into the CNN to calculate the predicted values ​​of the samples and determine the error between the predicted and actual values. If the error exceeds a set threshold, the backward propagation stage begins. Backpropagation involves propagating the error between the predicted and actual values ​​through the hidden layers to the input layer in a specific form. During backpropagation, the error signal acts on neurons in each layer, continuously adjusting the weights and biases until the network's error falls below a threshold, at which point training stops.

[0067] The convolutional neural network designed in this paper consists of one input layer, two convolutional layers, two pooling layers, one fully connected layer, and one output layer.

[0068] The model takes production data such as welding process voltage, current, and temperature as input and welding quality level as output. It is trained using actual production data and can be used as a welding quality prediction model after training.

[0069] See Figure 7 The invention shown employs a convolutional neural network structure to train a welding quality prediction model. After training, the model is applied to actual production to achieve real-time welding quality prediction. During the welding process of the current workpiece, the collected production data is input into the welding quality prediction model to obtain the predicted welding quality level. If the predicted quality is unqualified, the current workpiece is marked for subsequent focused inspection.

[0070] If a welded workpiece is deemed unqualified during subsequent quality inspection, the production data of the unqualified workpiece is extracted and used to train the welding quality prediction model. This enables real-time updates and improvements to the welding quality prediction model, reduces the time that unqualified product data is left unused, and provides more accurate services for subsequent production.

[0071] See Figure 2 The workpiece welding quality prediction method based on the OPC UA architecture shown includes the following steps:

[0072] S1, Production begins. The workpiece begins welding. The data acquisition and storage layer collects production data of the current workpiece welding process through the equipment layer. This production data includes welding process parameters and workpiece dimension data obtained by the welding equipment and sensors in the equipment layer. The OPC UA server collects the production data of the current workpiece welding process and stores the production data in the database.

[0073] S2, the collected production data is sent to the OPC UA client in the application layer for display. The OPC UA client compares the production data with a set range. If the production data of a certain workpiece is within the set range, then proceed to S3. If the production data of a certain workpiece is not within the set range, then the workpiece is marked by the MES system. After the workpiece welding is completed, the workpiece is subject to key quality inspection.

[0074] S3. The collected production data is submitted to the welding quality prediction system for quality prediction. The production data is input into the welding quality prediction model in the welding quality prediction system, which outputs the welding quality level to determine whether the workpiece welding quality is qualified. If the prediction result is that the workpiece is unqualified, the workpiece is marked through the MES system, and a key quality inspection is carried out on the workpiece after welding is completed.

[0075] S4. After the workpiece is welded, a quality inspection is performed. If the workpiece fails the quality inspection, the production data of the workpiece is extracted and used to update the welding quality prediction model in the welding quality prediction system.

[0076] Repeat steps S1 to S4 until production is complete.

[0077] When processing begins in the welding workshop, the OPC UA server, based on the OPC UA architecture, can conveniently collect production data of the workpiece welding process and display it intuitively through the OPC UA client. Thanks to the uniformity and standardization of the OPC UA architecture, it efficiently collects data from different welding machines, equipment, and sensors, converts it into a unified format, and stores the collected production data in a database for easy subsequent use. When the OPC UA client detects significant errors in the current body's processing data, it submits the error result to the MES management system, marking the workpiece. Marked problematic workpieces will be subject to priority inspection. The current workpiece's production data is submitted to the welding quality prediction model for quality prediction. If the prediction result is unqualified, the unqualified result is submitted to the MES management system, marking the workpiece again. Marked problematic workpieces will be subject to priority inspection. If a problematic workpiece is found during the body quality inspection process, its production data is retrieved from the database and used to update and train the welding quality prediction model. This process is repeated until production is completed.

[0078] The above embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention.

Claims

1. A workpiece welding quality prediction device based on OPC UA architecture, characterized in that: It includes the device layer, data acquisition and storage layer, and application layer; The equipment layer is used to acquire production data of the workpiece welding process; The data acquisition and storage layer includes an OPC UA server, which is communicatively connected to the device layer and serves as both the data acquisition end and the data transmission end. The application layer includes an OPC UA client, a welding quality prediction system, and an MES system. The OPC UA client communicates with the OPC UA server and acts as the receiving end for data transmission, displaying production data and comparing it with a set range. The welding quality prediction system is connected to the data acquisition and storage layer to submit the production data of the workpiece welding process to the welding quality prediction system for welding quality prediction; the MES system is used to mark workpieces whose production data is not within the set range or whose welding quality prediction is unqualified. Production data is bound to the address space of the OPC UA server in the form of nodes, with each node object representing a data item; OPC UA clients access the address space of the OPC UA server to obtain production data by browsing, reading, and subscribing. The welding quality prediction system integrates a welding quality prediction model based on a convolutional neural network. The welding quality prediction model takes the production data of the workpiece welding process as input and the welding quality level as output. If a welded workpiece is deemed unqualified during subsequent quality inspection during the production process, the production data of the unqualified workpiece is extracted and used to train the welding quality prediction model, thereby achieving real-time updates and improvements to the welding quality prediction model.

2. The workpiece welding quality prediction device based on OPC UA architecture according to claim 1, characterized in that: The data acquisition and storage layer also includes a database for storing production data.

3. The workpiece welding quality prediction device based on OPC UA architecture according to claim 1 or 2, characterized in that: The production data includes welding process parameters and workpiece size data.

4. A method for predicting workpiece welding quality based on OPC UA architecture, characterized in that, Includes the following steps: S1, the workpiece begins welding processing, and the data acquisition and storage layer collects the production data of the current workpiece welding process through the equipment layer; S2, the collected production data is sent to the OPC UA client in the application layer for display, and the production data is compared with the set range. If the production data of a certain workpiece is within the set range, then proceed to S3. If the production data of a certain workpiece is not within the set range, then the workpiece is marked by the MES system and key quality inspection is carried out after the workpiece welding is completed. S3. The collected production data is submitted to the welding quality prediction system for quality prediction to determine whether the welding quality of the workpiece is qualified. If the prediction result is that the workpiece is unqualified, the workpiece is marked through the MES system and key quality inspection is carried out after the workpiece welding is completed. S4. After the workpiece is welded, a quality inspection is performed. If the workpiece fails the quality inspection, the production data of the workpiece is extracted and used to update the welding quality prediction model in the welding quality prediction system.

Images