Ultrasonic welding pseudo soldering detection system and method

Abstract

The invention discloses an ultrasonic welding pseudo soldering detection system and method, and relates to the technical field of ultrasonic welding detection, the system comprises a hardware assembly and a visual software module, the hardware assembly comprises an upper camera, a lens, an annular light source, a PLC lower computer and the like, and the visual software module configures program instructions such as initialization, image acquisition, feature detection and the like; according to the detection method, images are collected through the upper camera, detection of the expanded material size, the grid edge width, the grid area and the peripheral overflow area is completed in sequence, and pseudo soldering is automatically judged in combination with a visual algorithm. Non-destructive online full detection is achieved, the detection precision reaches + / -0.01 mm, the single-station detection rhythm is smaller than or equal to 1 second, the problems that traditional detection products are large in waste and high in misjudgment risk are solved, the method is suitable for new energy power battery copper-aluminum bar metal ultrasonic welding products, and the detection efficiency and the quality control level are improved.

Description

Technical Field

[0001] This invention relates to the field of ultrasonic welding inspection technology, and in particular to an ultrasonic welding defect detection system and method. Background Technology

[0002] Ultrasonic welding of copper-aluminum alloys in new energy power batteries is a solid-state welding process. It removes the oxide film on the metal surface through ultrasonic vibration, allowing the metals to form a solid-state bond through interatomic attraction. Compared to traditional fusion welding, this reduces the heat-affected zone. However, the solid-state bonding characteristics of this weld joint make traditional non-destructive testing methods such as X-ray inspection and ultrasonic testing unsuitable.

[0003] Currently, the industry mainly uses two testing methods: one is destructive sampling inspection, which identifies faulty welds by observing residues after damaging the product, resulting in a large amount of product waste and is not conducive to mass production; the other is non-destructive testing sampling, where X-ray inspection requires manual interpretation and ultrasonic testing requires scanning with a medium, which not only affects subsequent processing but also poses a high risk of misjudgment.

[0004] Existing testing methods cannot achieve full online inspection of products, making it difficult to meet the quality control requirements of automated production. Therefore, there is an urgent need for a non-destructive, high-precision, and efficient ultrasonic welding defect detection technology. Summary of the Invention

[0005] To address the technical problems existing in the background art, the present invention proposes an ultrasonic welding defect detection system and method.

[0006] This invention proposes an ultrasonic welding defect detection method, comprising hardware components and a vision software module. The hardware components include a fixedly mounted upper camera, lens, ring light source, light source controller, communication connection cable, and dongle. The vision software module is configured with program instructions for system startup, hardware initialization, image acquisition, feature detection, data processing, and communication interaction. The hardware components and the vision software module work together to achieve defect detection of ultrasonic welding products.

[0007] Furthermore, the upper camera is an industrial camera with 12 megapixels and a resolution of 4024*3036. The installation height is not less than 400mm to avoid the material picking axis. The lens is 25mm. The long side of the field of view of the ring light source is 180±10mm. The wiring space reserved for the hardware components is not less than 60mm.

[0008] Furthermore, the ultrasonic welding defect detection system also includes a PLC lower-level machine. The vision software module interacts with the PLC via TCP / IP protocol. After data transmission, it needs to wait for the PLC's ACK confirmation signal. The structural accuracy of the system is 0.1mm in both the XY and Z directions.

[0009] Furthermore, an ultrasonic welding defect detection method includes the following steps: Step 1: Place the product to be tested inside the cavity of the carrier, and use a fixed upper camera to capture images of the product after welding from top to bottom; Step 2: Detect the bulging size of the product using a visual algorithm, identify the protruding points at the contact points of the aluminum product, calculate the distance D from the farthest protruding point to the baseline, and determine OK when D > standard setting value, and NG when D < standard setting value. Step 3: Check the edge width W of the 6*4 rectangular grid formed by the solder joints of the product. If W < standard setting value, it is judged as OK; if W > standard setting value, it is judged as NG. Step 4: Calculate the area value of the target region inside each of the 24 grids, and determine whether there are any anomalies such as the area being too small or multiple grids being connected based on the preset area boundary value; Step 5: Use a grayscale comparison algorithm to detect the overflow area around the rectangle composed of 24 grids to determine if there is any abnormality in the overflow.

[0010] Furthermore, in step 3, the uniformity of the grid edge is determined by capturing the area of ​​the region with similar grayscale to the product. When the proportion of black area exceeds the threshold, it is determined that the product has a poor solder joint.

[0011] Furthermore, the image acquisition cycle time in step 1 satisfies the following conditions: the total time for taking pictures and processing data at a single workstation does not exceed 1 second, and the detection accuracy of the vision system is ±0.01mm.

[0012] Furthermore, in step 4, a visual algorithm is used to identify the black area inside each cell. By calculating whether the area of ​​the black area is within a preset range, it is determined whether the cell has any broken or missing defects.

[0013] Furthermore, in step 5, the grayscale tool is used to compare the grayscale value difference between the outer area of ​​the grid rectangle and the product surface to identify whether the overflow area exceeds the allowable range.

[0014] Furthermore, the ultrasonic welding defect detection method also includes a risk response step: when irregular welding patterns cause the grid to connect with the unwelded parts, or when the vertical surface of the product blocks light and forms a gray-scale blind area, the consistency of the detection results determines whether to accept the product, and abnormal data is recorded for subsequent parameter optimization.

[0015] Furthermore, the vision software module is equipped with an AI computing model system, which includes a computing server and is configured with a training module and a computing module. The training module is used to extract ultrasonic welding product image data, historical inspection result data and defect annotation data stored in the storage server, and to iteratively train the computational model. The computing module is used to perform calculations on the collected product welding image data through the trained computing model, extract feature information of the bulging edge, grid structure and overflow area in the image, and then obtain the defect data related to the poor weld corresponding to the image data. The defect data is sent to the storage server and PLC lower computer via Ethernet. The computing model within the computing server includes: Convolutional neural networks extract depth features from image data through convolutional kernels to generate feature maps that include features such as bulge size, grid edges, grid area, and overflow area. The region proposal network is used to generate candidate boxes in the feature map for bulging protrusions, grid edge regions, grid interior regions and outer overflow regions, to preliminarily determine whether there are corresponding defects in the image; The interest pool is used to normalize candidate boxes of different sizes and convert them into feature maps of a fixed size to ensure consistency in subsequent calculations. Soft K-means clustering is used to perform cluster analysis on the defect features in the candidate box and identify the defect types, including substandard bulge size, abnormal grid edge width, excessive grid edge uniformity, abnormal grid area, and excessive perimeter overflow area. After training and processing a preset number of ultrasonic welding product images, the soft K-means clustering dynamically adjusts the number of clusters based on the elbow principle to maintain the optimal number of clusters. This allows it to identify new types of weld defects related to incomplete welds based on changes in the number of clusters during the training process.

[0016] The beneficial effects of this invention are as follows: Through deep collaboration between hardware components and vision software modules, it innovatively adopts an integrated design of "positioning and photography - multi-dimensional detection - intelligent judgment - risk response". On the hardware side, a stable detection foundation is built with a 12-megapixel high-resolution camera, an adaptable ring light source, and a Mitsubishi PLC, ensuring high precision of ±0.01mm and a high-efficiency cycle of ≤1 second per station. On the software side, targeted vision algorithms achieve comprehensive detection of material expansion size, grid edge width and uniformity, grid area, and peripheral overflow area. This not only completely eliminates product waste caused by traditional destructive inspection and effectively avoids the risk of misjudgment and subsequent process interference of non-destructive inspection methods such as X-ray inspection and ultrasonic inspection, but also flexibly handles special situations such as irregular welding edges and blind spots in lighting through risk response mechanisms. It realizes online full inspection of ultrasonic welding products of copper-aluminum metal in new energy power batteries, significantly reduces manual inspection costs, improves the quality control level and production efficiency of automated production lines, and provides the industry with a non-traditional vision inspection solution, helping to upgrade and iterate welding automation inspection technology, and has broad application prospects and promotional value. Attached Figure Description

[0017] Figure 1 This is a flowchart of the detection method of the present invention; Figure 2 This is a schematic diagram of step 2 in the detection method of the present invention; Figure 3 This is a schematic diagram of the qualified product in step 3 of the detection method of the present invention; Figure 4 This is a schematic diagram of the non-conforming products in step 3 of the detection method of the present invention; Figure 5 This is a schematic diagram of step 4 in the detection method of the present invention; Figure 6 This is a schematic diagram of step 5 in the detection method of the present invention.

[0018] Figure 7 This is a schematic diagram of the lens and detection process in the detection method of the present invention. Detailed Implementation

[0019] Reference Figure 1-7 This invention proposes an ultrasonic welding defect detection system and method. Its core is to achieve online full inspection of defective welds in ultrasonic welding products of copper and aluminum metals for new energy power batteries through the collaborative work of hardware components and vision software modules. The following is a detailed description of the specific structure, parameters and operation process.

[0020] I. Hardware Configuration and Installation of the Detection System The hardware components of this testing system are the foundation for achieving high-precision and rapid testing. The models, specifications, and installation requirements of each component are as follows: 1. Core testing components: 1.1 Camera: A Huazhou HZ13-GS120RC industrial camera with 12 megapixels and a resolution of 4024*3036 is used, which can clearly capture the subtle features of the product after welding. The camera is fixedly mounted on the machine platform at a height of no less than 400mm to avoid obstructing the material handling axis and ensure unobstructed photography. 1.2 Lens: Equipped with Huazhou FL-25-12MP-C lens, 25mm in size, which, when used with the camera, achieves a field of view coverage of 180±10mm along the long side, and can completely capture the 6*4 grid rectangular solder joint area and the surrounding overflow part; 1.3 Light source: Huazhou RLP-180B ring light source is selected to provide uniform and stable lighting conditions for image acquisition, avoiding feature recognition errors caused by uneven lighting. The installation position of the light source shall be strictly implemented in accordance with the lighting installation diagram to ensure clear imaging of solder joint grids, bulging protrusions and overflow areas.

[0021] 2. Control and connection components: 2.1 Light source controller: The Huazhou DKZ4-24V-4-AC model is used to adjust the brightness of the ring light source to adapt to the lighting needs of different products (long products, short products) and ensure sharp grayscale contrast in the image; 2.2 Lower-level PLC: A Mitsubishi PLC is selected as one of the control cores of the system. It is responsible for sending detection trigger signals, receiving detection results and performing subsequent action control. The system structure accuracy is 0.1mm in both the XY and Z directions, which matches the control accuracy of the PLC. 2.3 Communication and Power Supply Components: These include the Huazhou 6P-POWER-0500-FLEX power cord, the Huazhou STD-Z12K-120100C power adapter, the Huazhou HZ-Gigerj45 / rj455m network cable, and the Cognex VPRO-MAX-WB (vp9.6max) dongle. The default length of the power cord, network cable, and light source extension cable is 5 meters, which can be adjusted according to the actual wiring requirements of the device. All hardware components must have at least 60mm of clearance for wiring to ensure a neat layout and avoid affecting equipment operation.

[0022] II. Functional Implementation of the Vision Software Module The vision software module stores complete program instructions and is responsible for coordinating hardware components to complete the entire process from initialization to detection result output. Its core functions include: 1. System Initialization: After the system starts, the software automatically completes hardware initialization, camera module initialization and communication module initialization. At the same time, it loads calibration parameters and pre-trained AI computing models to ensure that all components are in a ready state and waiting for the trigger signal from the Mitsubishi PLC. 2. Image Acquisition and Preprocessing: After receiving the PLC trigger signal, the upper camera is controlled to acquire images of the welded product from top to bottom. The image acquisition cycle is strictly controlled to ensure that the total time for taking pictures and processing at a single workstation does not exceed 1 second, meeting the efficiency requirements of the automated production line. After acquisition, the images are preprocessed to optimize grayscale contrast and remove noise, laying the foundation for subsequent feature detection and AI model calculations. 3. Feature Detection and Data Processing: Through the collaborative work of preset visual algorithms and AI computing models, multi-dimensional detection data calculation and analysis are completed, specifically including expansion size calculation, grid edge width detection, grid area statistics, and overflow area identification. The detailed logic is as follows: 3.1 Basic Visual Algorithm Logic: 3.1 Basic Visual Algorithm Logic: 3.1.1. Identify the protruding points at the contact points of the aluminum product, select the farthest protruding point, and calculate its distance D from the baseline. 3.1.2 Identify the edge region of a 6*4 rectangular grid, calculate the edge width W between the grids, and simultaneously capture the area of ​​the grid edge that is similar in grayscale to the product, and judge the edge uniformity by the proportion of black area; 3.1.3 Locate the target area inside each of the 24 cells and calculate the area of ​​the black part inside each cell; 3.1.4 Use the grayscale tool to compare the grayscale difference between the outer area of ​​the grid rectangle and the product surface to identify the overflow area and calculate its area.

[0023] 3.2 Working Logic of the AI ​​Computation Model System: The AI ​​computing model system mounted on the vision software module includes a computing server, which is configured with a training module and a computing module, as specifically implemented as follows: Training module: Used to extract ultrasonic welding product image data, historical inspection result data and defect annotation data (covering various defect annotations such as non-compliant bulging size, abnormal grid edge, abnormal grid area, and excessive overflow) stored in the storage server, and to iteratively train the computational model to continuously optimize the model's detection accuracy; The computing module performs deep calculations on the pre-processed product welding image data through the trained computing model, accurately extracts the feature information of the bulging edge, grid structure and overflow area in the image, and then obtains the defect data related to the poor weld corresponding to the image data. The defect data is synchronously sent to the storage server and PLC lower computer via Ethernet for result judgment and data archiving. The computing model within the computing server consists of multiple modules that work together to achieve defect detection and identification. Convolutional Neural Network (CNN): It extracts features from the preprocessed image through a pre-defined convolutional kernel, generating a feature map containing features such as the size of the bulge, grid edges, grid area, and depth of the overflow area, providing a basis for subsequent defect candidate region localization; Region Proposal Network (RPN): In the feature map generated by CNN, it generates accurate candidate boxes for bulging protrusions, grid edge regions, grid interior regions and outer overflow regions, initially judges whether there are corresponding defects in each candidate box, and filters out high-probability defect regions. RoIPooling: Normalizes candidate boxes of different sizes output by RPN and converts them into feature maps of fixed size to ensure the consistency and accuracy of subsequent defect type identification calculations; SoftK-Means clustering: This method performs cluster analysis on the normalized candidate box features to accurately identify defect types, including substandard bulge size, abnormal grid edge width, excessive grid edge uniformity, abnormal grid area, and excessive perimeter overflow area. After training and processing with a preset number (e.g., 10,000) of ultrasonic welding product images, the number of clusters is dynamically adjusted based on the elbow principle to maintain an optimal number. This allows the system to identify new types of weld defects based on changes in the number of clusters during training, improving its adaptability to unknown defects.

[0024] Data Communication and Anomaly Handling: The detection results, jointly determined by the visual algorithm and AI model, are encapsulated in a data format and sent to the Mitsubishi PLC via TCP / IP protocol, awaiting an ACK confirmation signal from the PLC. If reception is successful, the error count is reset and the next detection cycle begins; if reception fails, the error count is incremented by 1. When three consecutive detection errors occur, the system triggers an emergency stop and sets an alarm flag to alert personnel to troubleshoot the fault.

[0025] 5. The testing system stores and traces product quality inspection data, allowing for real-time retrieval of early product testing data, thus facilitating the analysis of welding abnormalities and problem handling.

[0026] III. Specific Operating Procedures for the Detection Method Based on the above-mentioned detection system, the detection method of the present invention includes the following steps, realizing full automation from product loading to detection result determination: 1. Product loading and positioning: Place the ultrasonic welding product to be tested into the carrier cavity, ensuring that the product position is fixed and accurately aligned with the camera's field of view to avoid testing errors caused by product displacement.

[0027] 2. Inspection of material expansion dimensions ( Figure 1The system uses a visual algorithm to identify the protruding points at the contact points of aluminum products and calculates the distance D from the farthest protruding point to the baseline. A preset standard value (e.g., 0.2mm) is used. When D > the standard value, the product's expansion dimension is considered acceptable (OK), indicating that the extrusion force during welding meets the requirements. When D < the standard value, it is considered unacceptable (NG).

[0028] 3. Detection of grid edge width and uniformity ( Figure 2 , Figure 3 The detection method measures the edge width W of a 6x4 rectangular grid formed by the solder joints of the product. A preset standard value (e.g., 0.15mm) is used. When W < the standard value, it is considered OK, indicating a deep solder joint engagement. When W > the standard value, it is considered NG, indicating a shallow solder joint engagement. Simultaneously, uniformity is assessed by measuring the area of ​​the grid edge that is similar in grayscale to the product. If the proportion of black areas exceeds a threshold (e.g., 30%), it indicates that the grid edge is too thick, suggesting a weak solder joint, and the product is directly classified as NG.

[0029] 4. Grid internal area detection ( Figure 4 ): Calculate the area value of the black area inside each of the 24 grids. The preset area boundary value is (e.g., 5mm²-8mm²). When the area is <5mm², it is judged as a grid defect; when the area is >8mm², it is judged as multiple grids connected, and all are judged as NG; the area is within the preset range and is judged as OK.

[0030] 5. Detection of overflow area on the perimeter ( Figure 5 The grayscale tool is used to compare the grayscale values ​​of the outer area of ​​the grid rectangle with the product surface to identify the overflow area and calculate its area. A preset overflow area threshold (e.g., 10mm²) is set. When the overflow area is greater than the threshold, it is judged as abnormal (NG), otherwise it is OK.

[0031] 6. Risk Response and Handling: When irregular welding edges cause the grid to connect with unwelded parts, or when the vertical surface of the product blocks light and creates a gray-scale blind area, the consistency of the test results will be considered. If the consistency of the test results of 5 consecutive identical products is ≥90%, it will be considered acceptable with concessions, and the abnormal data will be recorded for subsequent optimization of standard settings and threshold parameters to improve the adaptability of the testing system.

[0032] IV. Ensuring Detection Accuracy and Efficiency This inspection system achieves a visual inspection accuracy of ±0.01mm and a structural accuracy of 0.1mm in both the XY and Z directions. Through multi-dimensional inspection (bulge size, grid edge, grid area, and overflow area), it ensures the accuracy of inspection results. Furthermore, the total time for photographing and processing at a single workstation is less than 1 second, adapting to the cycle time requirements of automated production lines. This enables online full inspection of products, avoiding product waste caused by traditional destructive sampling and the risk of misjudgment from manual inspection, significantly improving production efficiency and quality control.

[0033] In this embodiment, all standard settings and thresholds can be adjusted according to actual product specifications and welding process requirements, making it suitable for ultrasonic welding testing of copper-aluminum metal products for different models of new energy power batteries, and possessing wide applicability and adaptability.

[0034] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An ultrasonic welding defect detection system, characterized in that, The system includes hardware components and a vision software module. The hardware components include a fixed upper camera, lens, ring light source, light source controller, communication connection cable, and dongle. The vision software module is configured with program instructions for system startup, hardware initialization, image acquisition, feature detection, data processing, and communication interaction. The hardware components and the vision software module work together to detect incomplete welds in ultrasonic welding products.

2. The ultrasonic welding defect detection system according to claim 1, characterized in that, The upper camera is an industrial camera with 12 megapixels and a resolution of 4024*3036. The installation height is not less than 400mm to avoid the material picking axis. The lens is 25mm. The long side of the field of view of the ring light source is 180±10mm. The wiring space reserved for the hardware components is not less than 60mm.

3. The ultrasonic welding defect detection system according to claim 1, characterized in that, It also includes a PLC slave unit. The vision software module interacts with the PLC via TCP / IP protocol. After data transmission, it needs to wait for the PLC's ACK confirmation signal. The structural accuracy of the system is 0.1mm in both the XY and Z directions.

4. A method for detecting incomplete welds in ultrasonic welding, characterized in that, Includes the following steps: Step 1: Place the product to be tested inside the cavity of the carrier, and use a fixed upper camera to capture images of the product after welding from top to bottom; Step 2: Detect the bulging size of the product using a visual algorithm, identify the protruding points at the contact points of the aluminum product, calculate the distance D from the farthest protruding point to the baseline, and determine OK when D > standard setting value, and NG when D < standard setting value. Step 3: Check the edge width W of the 6*4 rectangular grid formed by the solder joints of the product. If W < standard setting value, it is judged as OK; if W > standard setting value, it is judged as NG. Step 4: Calculate the area value of the target region inside each of the 24 grids, and determine whether there are any anomalies such as the area being too small or multiple grids being connected based on the preset area boundary value; Step 5: Use a grayscale comparison algorithm to detect the overflow area around the rectangle composed of 24 grids to determine if there is any abnormality in the overflow.

5. The ultrasonic welding defect detection method according to claim 4, characterized in that, In step 3, the uniformity of the grid edge is judged by capturing the area of ​​the region with similar grayscale to the product. When the proportion of black area exceeds the threshold, it is determined that the product has a poor solder joint.

6. The ultrasonic welding defect detection method according to claim 4, characterized in that, The image acquisition cycle time in step 1 meets the following requirements: the total time for taking pictures and processing data at a single workstation does not exceed 1 second, and the detection accuracy of the vision system is ±0.01mm.

7. The ultrasonic welding defect detection method according to claim 4, characterized in that, In step 4, a visual algorithm is used to identify the black area inside each grid. By calculating whether the area of ​​the black area is within a preset range, it is determined whether the grid has any broken or missing defects.

8. The ultrasonic welding defect detection method according to claim 4, characterized in that, In step 5, the grayscale tool is used to compare the grayscale value difference between the outer area of ​​the grid rectangle and the product surface to identify whether the overflow area exceeds the allowable range.

9. The ultrasonic welding defect detection method according to claim 4, characterized in that, It also includes risk response steps: when irregular welding edges cause the grid to connect with the unwelded parts, or when the vertical surface of the product blocks light and forms a gray blind area, the consistency of the test results determines whether to accept the product, and abnormal data is recorded for subsequent parameter optimization.

10. The ultrasonic welding defect detection method according to claim 4, characterized in that, The vision software module is equipped with an AI computing model system, which includes a computing server and is configured with a training module and a computing module. The training module is used to extract ultrasonic welding product image data, historical inspection result data and defect annotation data stored in the storage server, and to iteratively train the computational model. The computing module is used to perform calculations on the collected product welding image data through the trained computing model, extract feature information of the bulging edge, grid structure and overflow area in the image, and then obtain the defect data related to the poor weld corresponding to the image data. The defect data is sent to the storage server and PLC lower computer via Ethernet. The computing model within the computing server includes: Convolutional neural networks extract depth features from image data through convolutional kernels to generate feature maps that include features such as bulge size, grid edges, grid area, and overflow area. The region proposal network is used to generate candidate boxes in the feature map for bulging protrusions, grid edge regions, grid interior regions and outer overflow regions, to preliminarily determine whether there are corresponding defects in the image; The interest pool is used to normalize candidate boxes of different sizes and convert them into feature maps of a fixed size to ensure consistency in subsequent calculations. Soft K-means clustering is used to perform cluster analysis on the defect features in the candidate box and identify the defect types, including substandard bulge size, abnormal grid edge width, excessive grid edge uniformity, abnormal grid area, and excessive perimeter overflow area. After training and processing a preset number of ultrasonic welding product images, the soft K-means clustering dynamically adjusts the number of clusters based on the elbow principle to maintain the optimal number of clusters. This allows it to identify new types of weld defects related to incomplete welds based on changes in the number of clusters during the training process.

Images