A Quality Inspection Method and System for Friction Welding of Mechanical Parts Based on Multimodal Information

Abstract

This application discloses a method and system for quality inspection of friction welding of mechanical parts based on multi-mode information. The technical solution provided in this application acquires process information of the release agent in the die-casting process, parameters for residual release agent treatment in the die-casting post-processing stage, and images of the welded end face before the welding process. Combined with real-time sensor data during the friction welding process, anomaly determination and root cause analysis are performed. This allows for the correlation and location of welding quality anomalies with specific preceding process steps, improving the accuracy and traceability of welding quality inspection and meeting the precision requirements for quality inspection of friction welding of die-cast parts.

Description

Technical Field

[0001] This application relates to the field of friction welding technology for die-cast parts, and in particular to a method and system for quality inspection of friction welding of mechanical parts based on multi-mode information. Background Technology

[0002] Currently, in the quality inspection of friction welding of mechanical parts, multimodal sensor data is typically collected during the welding process to assess weld quality. Specifically, existing friction welding quality inspection solutions deploy various sensors at the welding station, including pressure sensors, torque sensors, infrared thermal imagers, acoustic emission sensors, and vibration sensors, to acquire mechanical signals, temperature field distribution, acoustic emission signals, and vibration spectra in real time during the welding process. The collected multimodal data is input into a pre-built deep learning model, which extracts features related to welding quality from each modal signal, and then outputs the quality level or defect type of the weld joint. Thus, through the fusion analysis of multi-source information, online monitoring and evaluation of friction welding quality are achieved to a certain extent.

[0003] However, existing friction welding quality inspection solutions only focus on various sensor signals generated during the welding process, and their inspection results can only reflect the operational quality of the welded parts during the welding process. The inherent properties of the welded parts formed in the preceding manufacturing processes also have a potential impact on welding quality. For mechanical parts formed by die casting, the release agent sprayed on the mold cavity surface during die casting leaves residue on the part surface. This residue decomposes and vaporizes under the high temperature and pressure environment of subsequent friction welding, leading to defects such as weld interface contamination, abnormal friction coefficient, and weld porosity, directly affecting the quality of the weld joint. Simply monitoring the welding process cannot identify welding quality problems caused by release agent residue, and when abnormal welding quality occurs, it is difficult to trace the specific cause in the die casting process, resulting in low accuracy and reliability of friction welding quality inspection. Summary of the Invention

[0004] This application provides a method and system for quality inspection of friction welding of mechanical parts based on multi-mode information. It enables the correlation and location of welding quality abnormalities with specific preceding process steps, improves the accuracy and traceability of welding quality inspection, and solves the technical problem that friction welding quality inspection ignores the characteristics of preceding processes such as release agent residue and cannot identify the root cause of welding failure.

[0005] In a first aspect, embodiments of this application provide a method for quality inspection of friction welding of mechanical parts based on multi-modal information, including: The process information of the release agent in the die casting process of the target mechanical part, the release agent residue treatment parameters in the die casting post-processing process, and the welding end face image in the welding process are obtained. The release agent process information includes release agent spraying parameters and mold temperature distribution information. The release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters. Real-time sensing data of the target mechanical part during the friction welding process is collected, and the real-time sensing data is compared with the standard sensing data of the preset normal welding mode. When the real-time sensing data is in a set deviation state relative to the standard sensing data, it is determined that the target mechanical part has a welding quality abnormality. The real-time sensing data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal and vibration signal of the target mechanical part during the welding process. Based on the release agent process information, release agent residue treatment parameters, and welding end face images, the welding quality anomaly analysis of the target mechanical parts is performed to determine the root cause of the current welding quality anomaly, and the friction welding quality inspection results of the target mechanical parts are output according to the root cause.

[0006] Furthermore, based on the release agent process information, release agent residue treatment parameters, and weld end face images, an analysis of welding quality anomalies in the target mechanical parts is conducted to determine the root causes of the current welding quality anomalies, including: The process information of the release agent is compared with the preset standard process parameter range of the release agent to obtain the first comparison result. The release agent residue treatment parameters are compared with the preset standard post-treatment parameter range to obtain the second comparison result. The weld end face image is compared with the preset standard end face image to obtain the third comparison result. Anomalies are identified from the first, second, and third comparison results, and the root cause of the current welding quality anomaly is determined based on these anomalies.

[0007] Furthermore, the release agent process information is compared with the preset standard release agent process parameter range to obtain the first comparison result, including: The release agent spraying parameters are compared with the preset standard spraying parameter range, the mold temperature distribution information is compared with the preset standard temperature distribution information, and the temperature deviation value of each temperature measuring point is calculated. When the release agent spraying parameters exceed the standard spraying parameter range or the temperature deviation value of any temperature measurement point exceeds the preset temperature deviation threshold, the first comparison result is marked as abnormal, and the abnormality type is recorded as the corresponding release agent spraying abnormality or mold temperature abnormality.

[0008] Furthermore, abnormal comparison results are identified from the first, second, and third comparison results. Based on these abnormal comparison results, the root cause of the current welding quality abnormality is determined, including: When there is an abnormal comparison result among the first comparison result, the second comparison result, and the third comparison result, the abnormal type corresponding to the current abnormal comparison result is determined as the abnormal root cause of the welding quality abnormality. When there are multiple abnormal comparison results among the first, second, and third comparison results, the abnormal type corresponding to the abnormal comparison result with the highest priority is selected as the abnormal root cause of the welding quality abnormality according to the preset priority order.

[0009] Furthermore, based on the release agent process information, release agent residue treatment parameters, and weld end face images, an analysis of welding quality anomalies in the target mechanical parts is conducted to determine the root causes of the current welding quality anomalies, including: The release agent process information, release agent residue treatment parameters, weld end face images, and real-time sensor data are input into the pre-built multimodal fusion model; Visual features are extracted from the weld end face image based on the multimodal fusion model, and a visual feature vector representing the distribution state of release agent residue is output. Temporal modeling is performed on real-time sensing data based on the multimodal fusion model to determine the temporal change feature vector of the target mechanical part during the welding process. Release agent process information and release agent residue treatment parameters are encoded based on the multimodal fusion model, and a statistical feature vector representing the pre-welding process state of the target mechanical part is output. The visual feature vector, temporal variation feature vector, and statistical feature vector are fused to obtain multimodal fusion features. The multimodal fusion features are then input into a classification network to perform welding quality anomaly analysis on the target mechanical parts. The root cause of the current welding quality anomaly is output. The root cause of the anomaly includes at least one of the following: abnormal release agent spraying, abnormal mold temperature, insufficient shot blasting, excessive cleaning residue, and poor machining of the weld front end face.

[0010] Furthermore, the training process for the multimodal fusion model includes: Obtain the training dataset, which includes samples of mold release agent process information, mold release agent residue treatment parameters, welding end face images, welding process sensor data, and pre-labeled anomaly root cause type labels for each of the multiple sample mechanical parts. An initial multimodal fusion model is constructed, which includes a visual feature extraction network, a temporal feature extraction network, a statistical feature encoding network, a feature fusion layer, and a classification network. Samples of mold release agent process information and mold release agent residue treatment parameters are input into the statistical feature encoding network to output statistical feature vector samples. Samples of weld end face images are input into the visual feature extraction network to output visual feature vector samples. Samples of welding process sensor data are input into the temporal feature extraction network to output temporal feature vector samples. Statistical feature vector samples, visual feature vector samples, and temporal feature vector samples are input into the feature fusion layer for fusion to obtain multimodal fused feature samples. The multimodal fused feature samples are then input into the classification network to output the predicted anomaly root cause category. Based on the difference between the predicted root cause category and the corresponding root cause type label, the loss function value is calculated, and the network parameters of the multimodal fusion model are iteratively updated according to the loss function value until the loss function value converges, thus obtaining the trained multimodal fusion model.

[0011] Furthermore, the real-time sensor data is compared with the standard sensor data under the preset normal welding mode. When the real-time sensor data deviates from the standard sensor data by a set margin, it is determined that the target mechanical part has an abnormal welding quality, including: The torque signal, thermal imaging signal, acoustic emission signal, and vibration signal in the real-time sensing data are compared with their respective preset standard sensing data. When any signal comparison result exceeds the corresponding preset threshold range, it is determined that the real-time sensing data is in a set deviation state relative to the standard sensing data, and the welding quality of the target mechanical part is judged to be abnormal.

[0012] In a second aspect, embodiments of this application provide a quality inspection system for friction welding of mechanical parts based on multi-modal information, comprising: The information acquisition module is used to acquire the mold release agent process information of the target mechanical part in the die casting process, the mold release agent residue treatment parameters in the die casting post-processing process, and the welding end face image in the welding process. The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters. The comparison module is used to collect real-time sensing data of the target mechanical part during the friction welding process, compare the real-time sensing data with the standard sensing data of the preset normal welding mode, and determine that the target mechanical part has a welding quality abnormality when the real-time sensing data is in a set deviation state relative to the standard sensing data. The real-time sensing data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal and vibration signal of the target mechanical part during the welding process. The quality inspection module is used to perform welding quality anomaly analysis on the target mechanical part based on the release agent process information, the release agent residue treatment parameters, and the welding end face image, determine the root cause of the current welding quality anomaly, and output the friction welding quality inspection result of the target mechanical part according to the root cause.

[0013] In a third aspect, embodiments of this application provide an electronic device, including: Memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the friction welding quality inspection method for mechanical parts based on multi-mode information as described in the first aspect.

[0014] In a fourth aspect, embodiments of this application provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the friction welding quality inspection method for mechanical parts based on multi-mode information as described in the first aspect.

[0015] This application embodiment acquires the mold release agent process information of the target mechanical part in the die casting process, the mold release agent residue treatment parameters in the die casting post-processing process, and the weld end face image in the welding process. The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters. Real-time sensor data of the target mechanical part in the friction welding process is collected and compared with the standard sensor data in the preset normal welding mode. When the real-time sensor data deviates from the standard sensor data in a set state, it is determined that the target mechanical part has a welding quality abnormality. The real-time sensor data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal, and vibration signal of the target mechanical part during the welding process. Based on the mold release agent process information, mold release agent residue treatment parameters, and weld end face image, the welding quality abnormality analysis of the target mechanical part is performed to determine the root cause of the current welding quality abnormality. The friction welding quality inspection result of the target mechanical part is output according to the root cause of the abnormality. By employing the aforementioned technical means, information on the release agent process in the die casting process, parameters for residual release agent treatment in the die casting post-processing process, and images of the welded end face before the welding process are obtained. Combined with real-time sensor data during the friction welding process, anomaly judgment and root cause analysis are performed. This allows welding quality anomalies to be correlated with specific preceding process steps, improving the accuracy and traceability of welding quality inspection and meeting the precision requirements for quality inspection of friction welding of die castings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1This is a flowchart of a friction welding quality inspection method for mechanical parts based on multi-mode information, provided in Embodiment 1 of this application; Figure 2 This is a flowchart illustrating the determination of anomaly root causes in Embodiment 1 of this application; Figure 3 This is another flowchart for determining the abnormal root cause in Embodiment 1 of this application; Figure 4 This is a flowchart of the training process of the multimodal fusion model in Embodiment 1 of this application; Figure 5 This is a schematic diagram of the structure of a mechanical parts friction welding quality inspection system based on multi-mode information provided in Embodiment 2 of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0019] Example 1: Figure 1 A flowchart of a multi-mode information-based friction welding quality inspection method for mechanical parts is provided in Embodiment 1 of this application. This method can be executed by a multi-mode information-based friction welding quality inspection device. This device can be implemented through software and / or hardware. The device can consist of two or more physical entities, or it can consist of a single physical entity. Generally, this multi-mode information-based friction welding quality inspection device can be a die-casting system controller, a quality inspection system controller, or other processing equipment.

[0020] The following description uses the system controller as the main body for implementing a multi-mode information-based quality inspection method for friction welding of mechanical parts. (Refer to...) Figure 1 The quality inspection method for friction welding of mechanical parts based on multi-mode information specifically includes: S110. Obtain the mold release agent process information of the target mechanical part in the die casting process, the mold release agent residue treatment parameters in the die casting post-processing process, and the welding end face image in the welding process. The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters.

[0021] In the process of friction welding quality inspection of die-cast mechanical parts, this application achieves precise location of welding quality anomalies and specific preceding process steps by integrating and analyzing the usage status of the release agent in the die-casting process, the removal of release agent residue in the post-processing process, and the end face image information before the welding process with multimodal real-time sensing data collected during the welding process.

[0022] The mechanical parts currently undergoing friction welding are defined as the target mechanical parts. The system controller first acquires the mold release agent process information for the target mechanical parts during the die-casting process. This information refers to the process parameters related to the mold release agent sprayed onto the mold cavity surface to facilitate the smooth demolding of the formed parts during die-casting. Specifically, this includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent spraying parameters record the flow rate, spraying duration, and spraying coverage area of ​​the spraying equipment, used to quantify the initial deposition amount of mold release agent on the cavity surface. The mold temperature distribution information is collected in real-time by temperature sensors distributed in designated areas of the mold, used to characterize the drying and film-forming state of the mold release agent under high-temperature conditions. Excessively high mold temperatures can cause excessive volatilization and failure of the mold release agent, while excessively low temperatures can lead to increased residual mold release agent.

[0023] Furthermore, this application also obtains the release agent residue treatment parameters for the target mechanical parts in the die-casting post-processing stage. These release agent residue treatment parameters refer to the process parameters of various treatment steps performed after the parts are removed from the mold to remove or reduce surface release agent residue. Specifically, these include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters. The shot blasting parameters record information such as shot flow rate, shot velocity, and shot blasting time during shot blasting operations, used to characterize the degree of physical impact removal of residue. The cleaning parameters record the cleaning solution concentration, cleaning temperature, and cleaning time during release agent residue cleaning operations, used to characterize the effect of chemical dissolution in removing residue. The welding end face machining parameters record the spindle load current fluctuation curve during the CNC machine tool's turning of the target mechanical parts' welding end face. This curve reflects whether abnormal cutting resistance is encountered during the cutting process due to hardening or contamination caused by release agent residue.

[0024] Based on this, this application also obtains images of the welded end face of the target mechanical part during the welding process. These images are acquired by an online optical image measuring instrument after the aforementioned post-processing steps are completed and before the part enters the friction welding equipment. These images are used to identify surface contamination features such as dark spots, oil stains, or abnormal localized reflections caused by residual release agent on the end face.

[0025] By acquiring the above information, we can establish a data-level correlation between the various pre-welding processes and the final state of the mechanical parts, providing basic information for subsequent welding quality anomaly analysis.

[0026] S120. Collect real-time sensing data of the target mechanical part during the friction welding process, compare the real-time sensing data with the standard sensing data of the preset normal welding mode, and determine that the target mechanical part has a welding quality abnormality when the real-time sensing data is in a set deviation state relative to the standard sensing data. The real-time sensing data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal and vibration signal of the target mechanical part during the welding process.

[0027] Furthermore, during the friction welding process on the target mechanical parts, multiple sensors are deployed on the friction welding equipment to continuously collect real-time sensing data of the target mechanical parts during the welding process. This real-time sensing data includes at least one of torque signals, thermal imaging signals, acoustic emission signals, and vibration signals. Specifically, the torque signal is collected by a torque sensor mounted on the rotating spindle of the friction welding equipment, reflecting the change in frictional resistance between the two end faces during welding; the thermal imaging signal is collected by an infrared thermal imager, recording the dynamic distribution and evolution of the temperature field in the welding area; the acoustic emission signal is collected by an acoustic emission sensor, recording high-frequency stress waves generated during welding due to material cracking, gas escape, or interface slippage; and the vibration signal is collected by an acceleration sensor, used to characterize the mechanical stability of the welding equipment and the smoothness of the welding process.

[0028] After acquiring the aforementioned real-time sensor data, the system controller compares the real-time sensor data with the preset standard sensor data under normal welding mode, item by item. The preset standard sensor data under normal welding mode refers to the standard range of each sensor parameter determined by statistically analyzing the sensor data of a large number of sample parts with qualified welding quality during the welding process. When the real-time sensor data deviates from the set standard sensor data, it is determined that the target mechanical part has an abnormal welding quality.

[0029] Specifically, real-time sensor data is compared with standard sensor data under a preset normal welding mode. When the real-time sensor data deviates from the standard sensor data by a set margin, it is determined that the target mechanical part has an abnormal welding quality, including: The torque signal, thermal imaging signal, acoustic emission signal, and vibration signal in the real-time sensing data are compared with their respective preset standard sensing data. When any signal comparison result exceeds the corresponding preset threshold range, it is determined that the real-time sensing data is in a set deviation state relative to the standard sensing data, and the welding quality of the target mechanical part is judged to be abnormal.

[0030] The set deviation state can be specifically defined by comparing the torque signal, thermal imaging signal, acoustic emission signal, and vibration signal in the real-time sensing data with their respective preset standard signals. If the real-time value of any signal exceeds the corresponding preset threshold range, a welding quality abnormality is determined. Alternatively, a signal waveform comparison method can be used; if the difference between the waveform of any real-time sensor data and the waveform of the standard signal exceeds a preset similarity threshold, a welding quality abnormality is determined. This application does not impose fixed limitations on the specific signal comparison method, and will not elaborate further here.

[0031] Optionally, considering that preset standard sensor data may become invalid due to equipment aging or operating condition drift, this application collects a large amount of sensor data from qualified welded parts in the initial stage, and determines the preset threshold range of the initial standard sensor data through statistical analysis. In subsequent production processes, sensor data from qualified welded parts is continuously collected, and the newly collected data is grouped according to time windows. The median and interquartile range of each sensor parameter within each time window are calculated, and the threshold boundary is dynamically updated using an exponentially weighted moving average. When the welding quality of a batch of parts is subsequently confirmed as qualified, the sensor data corresponding to that batch is included in the threshold update sample set, and the updated threshold range is recalculated. Simultaneously, the long-term trend of the threshold range is monitored. When the threshold boundary of a certain parameter continuously drifts beyond a preset range, an equipment status warning is issued, prompting operators to check for mechanical wear or sensor calibration misalignment in the friction welding equipment. This allows the standard sensor data to adaptively adjust with the equipment status, maintaining the accuracy and reliability of anomaly detection.

[0032] S130. Based on the release agent process information, release agent residue treatment parameters, and welding end face image, perform welding quality anomaly analysis on the target mechanical part, determine the root cause of the current welding quality anomaly, and output the friction welding quality inspection results of the target mechanical part according to the root cause.

[0033] After determining that the welding quality is abnormal, the system controller uses previously acquired information on the release agent process, release agent residue treatment parameters, and weld end face images to analyze and locate the root cause of the abnormality. Specifically, by comprehensively analyzing the aforementioned pre-welding process information and real-time sensor data during the welding process, the system aims to determine the specific preceding process step that led to the current welding quality abnormality. Thus, by performing cross-process analysis of multi-mold information from the die-casting, post-processing, and welding processes, the system achieves the correlation and location of welding quality abnormalities with specific preceding process steps, improving the accuracy and traceability of welding quality detection and avoiding situations where the root cause of welding failure cannot be identified due to neglecting pre-process characteristics such as release agent residue.

[0034] Optionally, refer to Figure 2 Based on the release agent process information, release agent residue treatment parameters, and weld end face images, anomaly analysis of the welding quality of the target mechanical parts is performed to determine the root cause of the current welding quality anomaly, including: S1301. Compare the release agent process information with the preset standard release agent process parameter range to obtain the first comparison result; compare the release agent residue treatment parameters with the preset standard post-treatment parameter range to obtain the second comparison result; compare the weld end face image with the preset standard end face image to obtain the third comparison result. S1302. Determine the abnormal comparison results from the first comparison results, the second comparison results, and the third comparison results, and determine the root cause of the current welding quality abnormality based on the abnormal comparison results.

[0035] By comparing the release agent process information with the preset standard release agent process parameter range, it can be determined whether the release agent spraying parameters exceed the standard or whether the mold temperature deviates from the normal range. The release agent residue treatment parameters are compared with the preset standard post-treatment parameter range to determine if there are parameter deviations in the shot blasting, cleaning, or machining processes. The weld end face image is compared with the preset standard end face image to determine if there are visible characteristics of release agent residue on the end face. When the above comparison results show anomalies, the root cause of the welding quality abnormality can be determined based on the corresponding process steps.

[0036] The standard release agent process parameter range refers to the threshold range of process parameters used to determine whether the release agent is functioning normally in the die-casting process, based on statistical data of release agent spraying flow rate, spraying duration, spraying coverage area, and temperature distribution at various temperature measurement points of the mold corresponding to historical qualified batches of die-casting parts. The standard post-processing parameter range refers to the threshold range of post-processing process parameters used to determine whether the release agent residue removal effect meets the standards, based on statistical data of shot blasting intensity, cleaning time, cleaning fluid concentration, and spindle load fluctuation during the machining of the welded end face corresponding to historical qualified batches of die-casting parts. The standard end face image refers to a standard surface state template image representing the absence of release agent residue contamination, formed by image processing and feature extraction from welded end face images collected from historically qualified welded mechanical parts.

[0037] Specifically, the release agent process information is compared with a preset standard release agent process parameter range to obtain the first comparison result, including: The release agent spraying parameters are compared with the preset standard spraying parameter range, the mold temperature distribution information is compared with the preset standard temperature distribution information, and the temperature deviation value of each temperature measuring point is calculated. When the release agent spraying parameters exceed the standard spraying parameter range or the temperature deviation value of any temperature measurement point exceeds the preset temperature deviation threshold, the first comparison result is marked as abnormal, and the abnormality type is recorded as the corresponding release agent spraying abnormality or mold temperature abnormality.

[0038] The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The spraying parameters include spraying flow rate, spraying duration, and spraying coverage area. These parameters are compared item by item with preset standard spraying parameter ranges to determine if any exceed the upper or lower limits are met. Simultaneously, the system controller acquires temperature distribution data from each temperature measuring point on the mold and compares it with preset standard temperature distribution information, calculating the deviation between the actual temperature and the standard temperature at each measuring point. When any indicator of the mold release agent spraying parameters exceeds the standard spraying parameter range, or when the temperature deviation at any measuring point exceeds a preset temperature deviation threshold, the first comparison result is marked as abnormal, and the abnormality type is recorded.

[0039] Optionally, the controller can record the anomaly type based on the triggering conditions. If triggered by exceeding the spraying parameters, it is recorded as an anomaly in the release agent spraying; if triggered by temperature deviation, it is recorded as an anomaly in the mold temperature. This allows the first comparison result to more clearly point to the specific problematic link in the die-casting process, providing a clear basis for subsequent root cause localization.

[0040] In addition, the parameters for treating residual release agent include shot blasting parameters, cleaning parameters, and machining parameters of the weld front face. The system compares these parameters with the preset standard post-treatment parameter range to determine whether the shot blasting intensity, cleaning time, or machining load fluctuations deviate from the normal range. If a deviation from the normal range occurs, the second comparison result is marked as abnormal.

[0041] The weld end face image is compared with a preset standard end face image to extract the specified residue area, surface texture and grayscale distribution features. When the area features reflecting dark spots, oil stains or abnormal reflection are detected, the third comparison result is marked as abnormal.

[0042] Prior to this, the system controller pre-constructed an image comparison model based on a deep convolutional neural network. This model contains two structurally identical feature extraction sub-networks. Each sub-network consists of five convolutional layers, three pooling layers, and one fully connected layer stacked sequentially. The convolutional layers use 3×3 convolutional kernels with the ReLU activation function, the pooling layers use 2×2 max pooling, and the fully connected layer reduces the extracted high-dimensional features to a 512-dimensional feature vector. During the model training phase, a training dataset was first constructed. The dataset contains two types of samples: positive sample pairs consist of welded end face images of the same qualified part from different angles or under different lighting conditions, and negative sample pairs consist of standard end face images of qualified parts and welded end face images with abnormal release agent residue. The abnormal release agent residue includes three pre-labeled defect types: dark spots, oil stains, and abnormal reflection. The model training employs a contrastive loss function, which calculates the Euclidean distance between the output feature vectors of the two branches. This distance is minimized when the input consists of positive sample pairs and maximized when the input consists of negative sample pairs. A distance threshold is also set, enabling the model to learn discriminative features that differentiate between normal and abnormal end faces. The model parameters are continuously optimized during training until the loss function converges.

[0043] In practical applications, the system inputs the acquired image of the welded end face of the target mechanical part into a trained feature extraction sub-network, and inputs a preset standard end face image into another feature extraction sub-network. Each branch outputs a corresponding 512-dimensional feature vector. The cosine similarity between the two feature vectors is then calculated, with similarity values ​​ranging from 0 to 1. When the similarity is below a preset threshold of 0.75, the system determines that the current welded end face image differs significantly from the standard end face image, and then performs abnormal region localization. A heatmap is generated by weighted fusion of the feature map from the last convolutional layer of the network and the classification gradient. The highlighted areas in the heatmap are the main regions causing the decrease in similarity. Further, the system extracts the contours of the connected components in these highlighted areas, calculates the area, shape factor, and the difference between the area's area, shape factor, and grayscale mean and the corresponding area's grayscale mean in the standard end face image. When the area exceeds a preset area threshold and the grayscale difference exceeds a preset grayscale threshold, the area is identified as an abnormal region with residual mold release agent. For the identified abnormal areas, the abnormality type is also classified according to their gray-scale distribution characteristics. When an abnormal area of ​​any of the above pre-labeled defect types is detected, the third comparison result is marked as abnormal and the abnormality type is recorded.

[0044] After completing the three sets of comparisons, the system controller filters out the results marked as abnormal from the first, second and third comparison results, and determines the root cause of the current welding quality abnormality based on the abnormal comparison results.

[0045] Furthermore, abnormal comparison results are identified from the first, second, and third comparison results. Based on these abnormal comparison results, the root cause of the current welding quality abnormality is determined, including: When there is an abnormal comparison result among the first comparison result, the second comparison result, and the third comparison result, the abnormal type corresponding to the current abnormal comparison result is determined as the abnormal root cause of the welding quality abnormality. When there are multiple abnormal comparison results among the first, second, and third comparison results, the abnormal type corresponding to the abnormal comparison result with the highest priority is selected as the abnormal root cause of the welding quality abnormality according to the preset priority order.

[0046] After marking the first, second, and third comparison results, the system first performs anomaly counting on the three sets of comparison results. When only one abnormal comparison result is detected, it indicates that the process parameters corresponding to that abnormal comparison result have deviated, and this deviation is unrelated to the state of other process steps. In this case, the system directly identifies the anomaly type corresponding to that abnormal comparison result as the root cause of the welding quality anomaly. For example, if only the first comparison result is abnormal, the root cause is determined to be from the mold release agent spraying or mold temperature control process in the die casting process. When the system detects multiple abnormal comparison results, it indicates that the welding quality anomaly may be caused by the combined influence of multiple preceding process steps. In this case, the system calls the preset priority order to screen for the root cause. The preset priority order is based on the direct impact of each process step on welding quality. The weld end face image reflects the final surface state of the part before entering the welding equipment, directly determining the initial cleanliness and contact conditions of the welding interface; therefore, the third comparison result has the highest priority. The mold release agent residue treatment parameters reflect the removal effect of post-processing on mold release agent residue; anomalies in these parameters indicate that previous removal methods failed to achieve the expected results, thus the second comparison result has a lower priority than the third. The mold release agent process information reflects the usage status of the mold release agent at the die-casting source; while anomalies in this information may affect the residue amount, they can potentially be compensated for through subsequent post-processing steps, therefore the first comparison result has the lowest priority. The system selects the highest priority among multiple anomaly comparison results according to this priority order and outputs the anomaly type corresponding to that result as the root cause of the welding quality anomaly.

[0047] By combining direct location of a single anomaly with priority ranking of multiple anomalies, this application can provide clear and operable root cause indications while ensuring location accuracy, thus avoiding root cause tracing errors caused by the simultaneous existence of multiple anomalies.

[0048] Furthermore, this application may also employ a severity-weighted scoring method, which involves pre-setting severity weights for each anomaly type and calculating a weighted score based on the deviation from the threshold in the actual comparison results, selecting the anomaly type with the highest score as the root cause. This application does not impose fixed restrictions on the method for determining the root cause of anomalies based on anomaly comparison results, and will not elaborate further here.

[0049] Based on this, this application can also employ a pre-constructed multimodal fusion model for anomaly root cause analysis. Optionally, refer to Figure 3 Based on the release agent process information, release agent residue treatment parameters, and weld end face images, anomaly analysis of the welding quality of the target mechanical parts is performed to determine the root cause of the current welding quality anomaly, including: S1303. Input the release agent process information, release agent residue treatment parameters, welding end face image and real-time sensor data into the pre-built multimodal fusion model; S1304. Based on the multimodal fusion model, visual features are extracted from the weld end face image, and a visual feature vector representing the distribution state of the release agent residue is output. Based on the multimodal fusion model, time-series modeling is performed on real-time sensing data to determine the time-series change feature vector of the target mechanical part during the welding process. Based on the multimodal fusion model, the release agent process information and release agent residue treatment parameters are encoded, and a statistical feature vector representing the pre-welding process state of the target mechanical part is output. S1305. The visual feature vector, temporal change feature vector and statistical feature vector are fused to obtain multimodal fusion features. The multimodal fusion features are then input into the classification network to perform welding quality anomaly analysis of the target mechanical parts. The root cause of the current welding quality anomaly is output. The root cause of the anomaly includes at least one of the following: abnormal release agent spraying, abnormal mold temperature, insufficient shot blasting, excessive cleaning residue, and poor machining of the weld front end face.

[0050] Before proceeding, a training dataset for the multimodal fusion model is first obtained. Each sample in the training dataset contains samples of mold release agent process information, mold release agent residue treatment parameters, weld end face images, welding process sensor data, and manually labeled anomaly root cause types. An initial model is then constructed, comprising a visual feature extraction network, a temporal feature extraction network, a statistical feature encoding network, a feature fusion layer, and a classification network. During initial model training, statistical feature samples are input into the statistical feature encoding network to obtain statistical feature vectors, image samples are input into the visual feature extraction network to obtain visual feature vectors, and sensor data samples are input into the temporal feature extraction network to obtain temporal feature vectors. These three vectors are concatenated by the feature fusion layer and then input into the classification network, outputting the predicted anomaly root cause category. The loss function value between the prediction result and the labeled tag is calculated, and the network parameters are iteratively updated based on the loss function value until the model converges.

[0051] In the subsequent model application phase, the mold release agent process information, mold release agent residue treatment parameters, weld end face images, and real-time sensor data of the target mechanical part are input into the trained multimodal fusion model. The model extracts corresponding feature vectors through a visual feature extraction network, a temporal feature extraction network, and a statistical feature encoding network, respectively. These feature vectors are then fused to obtain multimodal fusion features, which are input into a classification network for welding quality anomaly analysis of the target mechanical part. The model outputs anomaly root cause categories, including at least one of the following: abnormal mold release agent spraying, abnormal mold temperature, insufficient shot blasting, excessive cleaning residue, and poor weld end face machining. Based on the identified root causes, the model outputs the friction welding quality inspection results for the target mechanical part. These results not only determine whether the welding quality is acceptable but also identify the specific preceding process steps that led to the quality anomaly, providing quality inspectors with precise adjustment guidelines.

[0052] Specifically, refer to Figure 4 The training process for a multimodal fusion model includes: S1001. Obtain the training dataset, which includes multiple sample mechanical parts, each corresponding to a sample of mold release agent process information, mold release agent residue treatment parameter samples, welding end face image samples, welding process sensor data samples, and pre-labeled abnormal root cause type labels. S1002. Construct an initial multimodal fusion model, which includes a visual feature extraction network, a temporal feature extraction network, a statistical feature encoding network, a feature fusion layer, and a classification network. S1003. Input the mold release agent process information sample and the mold release agent residual treatment parameter sample into the statistical feature encoding network and output the statistical feature vector sample. Input the welding end face image sample into the visual feature extraction network and output the visual feature vector sample. Input the welding process sensor data sample into the temporal feature extraction network and output the temporal feature vector sample. S1004. Input the statistical feature vector samples, visual feature vector samples, and temporal feature vector samples into the feature fusion layer for fusion to obtain multimodal fusion feature samples. Input the multimodal fusion feature samples into the classification network and output the predicted abnormal root cause category. S1005. Based on the difference between the predicted abnormal root cause category and the corresponding abnormal root cause type label, calculate the loss function value, and iteratively update the network parameters of the multimodal fusion model according to the loss function value until the loss function value converges, thus obtaining the trained multimodal fusion model.

[0053] In the multimodal fusion model training process, the training dataset comes from multiple sample mechanical parts collected from historical production batches. Each sample mechanical part corresponds to a complete set of multi-source data, specifically including samples of mold release agent process information recorded in the die casting process, samples of mold release agent residue treatment parameters recorded in the die casting post-processing process, samples of weld end face images collected by an optical image measuring instrument before welding, samples of welding process sensor data collected by multiple types of sensors during friction welding, and abnormal root cause type labels pre-labeled by quality inspectors based on subsequent inspection results. The abnormal root cause type labels include at least categories such as abnormal mold release agent spraying, abnormal mold temperature, insufficient shot blasting removal, excessive cleaning residue, poor welding end face processing, and normal welding quality, forming a labeled dataset covering various typical defects.

[0054] After the dataset is constructed, the system builds an initial multimodal fusion model. This initial model adopts a multi-branch heterogeneous network architecture, specifically including a visual feature extraction network, a temporal feature extraction network, a statistical feature encoding network, a feature fusion layer, and a classification network. The visual feature extraction network uses a convolutional neural network structure, consisting of multiple convolutional layers, pooling layers, and fully connected layers stacked together, to extract visual feature vectors representing the distribution morphology, surface texture, and contamination level of release agent residues from the weld end face image. The temporal feature extraction network uses a long short-term memory network to perform temporal modeling on the torque signal, thermal imaging signal, acoustic emission signal, and vibration signal continuously acquired during the welding process, extracting temporal feature vectors reflecting the dynamic stability of the welding process. The statistical feature encoding network uses a multi-layer fully connected network structure to encode the release agent process information samples and release agent residue treatment parameter samples, converting them into statistical feature vectors representing the previous process state.

[0055] During the model training phase, samples of mold release agent process information and mold release agent residue treatment parameters are input into the statistical feature encoding network. The network outputs fixed-dimensional statistical feature vector samples through layer-by-layer linear transformation and nonlinear activation. Samples of welded end face images are input into the visual feature extraction network. The network extracts multi-level abstract features of the images through convolution operations and outputs visual feature vector samples after global average pooling. Samples of welding process sensor data are input into the temporal feature extraction network. The network processes the temporal sequence frame by frame through recurrent units and performs average pooling on the output of all time steps to obtain temporal feature vector samples.

[0056] Subsequently, the statistical feature vector samples, visual feature vector samples, and temporal feature vector samples output from the three branches are input into the feature fusion layer for fusion. The feature fusion layer concatenates the three vectors in a dimensional manner to form a unified multimodal fusion feature sample. This multimodal fusion feature sample is then input into a classification network, which consists of several fully connected layers and an output layer. Through layer-by-layer mapping, the fusion features are mapped to the anomaly root cause category space, outputting the predicted anomaly root cause category. Based on the difference between the predicted anomaly root cause category and the corresponding anomaly root cause type label, the cross-entropy loss function is used to calculate the loss function value. Using this loss function value as the optimization objective, the backpropagation algorithm is used to calculate the parameter gradient of each network layer, and the Adam optimizer is used to iteratively update the weight parameters of each network layer with a preset learning rate. Training is stopped and the current network parameters are saved until the model converges, resulting in the trained multimodal fusion model. This training process constructs a multi-source heterogeneous dataset covering the entire process of die casting, post-processing, and welding. It then uses branch networks to extract deep features from different modal data and performs fusion learning, enabling the model to fully learn the complex mapping relationship between the characteristics of release agent residue and the root causes of welding quality abnormalities.

[0057] Once the model is trained, the root cause of the anomaly can be determined based on the input mold release agent process information, mold release agent residue treatment parameters, weld end face image, and real-time sensor data. Based on the determined root cause of the anomaly, the friction welding quality inspection results of the target mechanical part can be output.

[0058] Optionally, this application further optimizes process parameters based on root cause localization results, feeding back quality inspection results to preceding processes to achieve automatic adjustment of process parameters. When the system outputs the root cause of welding quality anomalies, it generates corresponding process parameter adjustment instructions based on the type of root cause. Specifically, if the root cause is abnormal release agent spraying, an adjustment instruction is sent to the release agent spraying control system of the die-casting equipment to reduce the spraying flow rate or adjust the spraying duration by a preset step size; if the root cause is abnormal mold temperature, an adjustment instruction is sent to the mold temperature control system to correct the cooling water flow rate or heating rod power; if the root cause is insufficient shot blasting or excessive cleaning residue, an adjustment instruction is sent to the post-processing equipment to increase the shot blasting time or increase the cleaning fluid concentration; if the root cause is poor machining of the weld front face, an adjustment instruction is sent to the CNC machine tool to correct the cutting parameters. After the adjustment instructions are executed, the system continuously monitors the welding quality of subsequent batches of parts. If the anomaly rate drops below a preset threshold, the current process parameters are fixed; if the anomaly rate does not significantly improve, iterative adjustments continue until the optimization target is achieved. This achieves process optimization and improves the automation level of the production process.

[0059] The above-mentioned process involves acquiring the mold release agent process information of the target mechanical part in the die casting process, the mold release agent residue treatment parameters in the die casting post-processing process, and the weld end face image in the welding process. The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters. Real-time sensor data of the target mechanical part in the friction welding process is collected and compared with the standard sensor data under the preset normal welding mode. When the real-time sensor data deviates from the standard sensor data by a set value, it is determined that the target mechanical part has a welding quality abnormality. The real-time sensor data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal, and vibration signal of the target mechanical part during the welding process. Based on the mold release agent process information, mold release agent residue treatment parameters, and weld end face image, the welding quality abnormality analysis of the target mechanical part is performed to determine the root cause of the current welding quality abnormality. The friction welding quality inspection result of the target mechanical part is output according to the root cause of the abnormality. By employing the aforementioned technical means, information on the release agent process in the die casting process, parameters for residual release agent treatment in the die casting post-processing process, and images of the welded end face before the welding process are obtained. Combined with real-time sensor data during the friction welding process, anomaly judgment and root cause analysis are performed. This allows welding quality anomalies to be correlated with specific preceding process steps, improving the accuracy and traceability of welding quality inspection and meeting the precision requirements for quality inspection of friction welding of die castings.

[0060] Example 2: Based on the above embodiments, Figure 5 This is a schematic diagram of a quality inspection system for friction welding of mechanical parts based on multi-mode information, provided in Embodiment 2 of this application. (Reference) Figure 5 The mechanical parts friction welding quality inspection system based on multi-mode information provided in this embodiment specifically includes: Information acquisition module 21 is used to acquire mold release agent process information of target mechanical parts in the die casting process, mold release agent residue treatment parameters in the die casting post-processing process, and weld end face image in the welding process. The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and weld end face machining parameters. The comparison module 22 is used to collect real-time sensing data of the target mechanical part during the friction welding process, compare the real-time sensing data with the standard sensing data of the preset normal welding mode, and determine that the target mechanical part has a welding quality abnormality when the real-time sensing data is in a set deviation state relative to the standard sensing data. The real-time sensing data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal and vibration signal of the target mechanical part during the welding process. The quality inspection module 23 is used to perform welding quality anomaly analysis on the target mechanical part based on the release agent process information, the release agent residue treatment parameters and the welding end face image, determine the root cause of the current welding quality anomaly, and output the friction welding quality inspection result of the target mechanical part according to the root cause.

[0061] Specifically, based on the release agent process information, release agent residue treatment parameters, and weld end face images, welding quality anomaly analysis of the target mechanical parts is performed to determine the root cause of the current welding quality anomaly, including: The process information of the release agent is compared with the preset standard process parameter range of the release agent to obtain the first comparison result. The release agent residue treatment parameters are compared with the preset standard post-treatment parameter range to obtain the second comparison result. The weld end face image is compared with the preset standard end face image to obtain the third comparison result. Anomalies are identified from the first, second, and third comparison results, and the root cause of the current welding quality anomaly is determined based on these anomalies.

[0062] Specifically, the release agent process information is compared with a preset standard release agent process parameter range to obtain the first comparison result, including: The release agent spraying parameters are compared with the preset standard spraying parameter range, the mold temperature distribution information is compared with the preset standard temperature distribution information, and the temperature deviation value of each temperature measuring point is calculated. When the release agent spraying parameters exceed the standard spraying parameter range or the temperature deviation value of any temperature measurement point exceeds the preset temperature deviation threshold, the first comparison result is marked as abnormal, and the abnormality type is recorded as the corresponding release agent spraying abnormality or mold temperature abnormality.

[0063] Specifically, abnormal comparison results are determined from the first, second, and third comparison results. Based on these abnormal comparison results, the root cause of the current welding quality abnormality is determined, including: When there is an abnormal comparison result among the first comparison result, the second comparison result, and the third comparison result, the abnormal type corresponding to the current abnormal comparison result is determined as the abnormal root cause of the welding quality abnormality. When there are multiple abnormal comparison results among the first, second, and third comparison results, the abnormal type corresponding to the abnormal comparison result with the highest priority is selected as the abnormal root cause of the welding quality abnormality according to the preset priority order.

[0064] Specifically, based on the release agent process information, release agent residue treatment parameters, and welding end face images, the welding quality anomaly analysis of the target mechanical parts is performed to determine the root cause of the current welding quality anomaly, including: inputting the release agent process information, release agent residue treatment parameters, welding end face images, and real-time sensor data into a pre-constructed multimodal fusion model; Visual features are extracted from the weld end face image based on the multimodal fusion model, and a visual feature vector representing the distribution state of release agent residue is output. Temporal modeling is performed on real-time sensing data based on the multimodal fusion model to determine the temporal change feature vector of the target mechanical part during the welding process. Release agent process information and release agent residue treatment parameters are encoded based on the multimodal fusion model, and a statistical feature vector representing the pre-welding process state of the target mechanical part is output. The visual feature vector, temporal variation feature vector, and statistical feature vector are fused to obtain multimodal fusion features. The multimodal fusion features are then input into a classification network to perform welding quality anomaly analysis on the target mechanical parts. The root cause of the current welding quality anomaly is output. The root cause of the anomaly includes at least one of the following: abnormal release agent spraying, abnormal mold temperature, insufficient shot blasting, excessive cleaning residue, and poor machining of the weld front end face.

[0065] Specifically, the training process for the multimodal fusion model includes: Obtain the training dataset, which includes samples of mold release agent process information, mold release agent residue treatment parameters, welding end face images, welding process sensor data, and pre-labeled anomaly root cause type labels for each of the multiple sample mechanical parts. An initial multimodal fusion model is constructed, which includes a visual feature extraction network, a temporal feature extraction network, a statistical feature encoding network, a feature fusion layer, and a classification network. Samples of mold release agent process information and mold release agent residue treatment parameters are input into the statistical feature encoding network to output statistical feature vector samples. Samples of weld end face images are input into the visual feature extraction network to output visual feature vector samples. Samples of welding process sensor data are input into the temporal feature extraction network to output temporal feature vector samples. Statistical feature vector samples, visual feature vector samples, and temporal feature vector samples are input into the feature fusion layer for fusion to obtain multimodal fused feature samples. The multimodal fused feature samples are then input into the classification network to output the predicted anomaly root cause category. Based on the difference between the predicted root cause category and the corresponding root cause type label, the loss function value is calculated, and the network parameters of the multimodal fusion model are iteratively updated according to the loss function value until the loss function value converges, thus obtaining the trained multimodal fusion model.

[0066] Specifically, real-time sensor data is compared with standard sensor data under a preset normal welding mode. When the real-time sensor data deviates from the standard sensor data by a set margin, it is determined that the target mechanical part has an abnormal welding quality, including: The torque signal, thermal imaging signal, acoustic emission signal, and vibration signal in the real-time sensing data are compared with their respective preset standard sensing data. When any signal comparison result exceeds the corresponding preset threshold range, it is determined that the real-time sensing data is in a set deviation state relative to the standard sensing data, and the welding quality of the target mechanical part is judged to be abnormal.

[0067] The above-mentioned process involves acquiring the mold release agent process information of the target mechanical part in the die casting process, the mold release agent residue treatment parameters in the die casting post-processing process, and the weld end face image in the welding process. The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters. Real-time sensor data of the target mechanical part in the friction welding process is collected and compared with the standard sensor data under the preset normal welding mode. When the real-time sensor data deviates from the standard sensor data by a set value, it is determined that the target mechanical part has a welding quality abnormality. The real-time sensor data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal, and vibration signal of the target mechanical part during the welding process. Based on the mold release agent process information, mold release agent residue treatment parameters, and weld end face image, the welding quality abnormality analysis of the target mechanical part is performed to determine the root cause of the current welding quality abnormality. The friction welding quality inspection result of the target mechanical part is output according to the root cause of the abnormality. By employing the aforementioned technical means, information on the release agent process in the die casting process, parameters for residual release agent treatment in the die casting post-processing process, and images of the welded end face before the welding process are obtained. Combined with real-time sensor data during the friction welding process, anomaly judgment and root cause analysis are performed. This allows welding quality anomalies to be correlated with specific preceding process steps, improving the accuracy and traceability of welding quality inspection and meeting the precision requirements for quality inspection of friction welding of die castings.

[0068] The friction welding quality inspection system for mechanical parts based on multi-mode information provided in Embodiment 2 of this application can be used to execute the friction welding quality inspection method for mechanical parts based on multi-mode information provided in Embodiment 1 above, and has corresponding functions and beneficial effects.

[0069] Example 3: This application provides an electronic device in embodiment three, referring to... Figure 6The electronic device includes a processor 31, a memory 32, a communication module 33, an input device 34, and an output device 35. The electronic device may have one or more processors and one or more memories. The processor, memory, communication module, input device, and output device of the electronic device can be connected via a bus or other means.

[0070] Memory, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the multi-mode information-based friction welding quality inspection method for mechanical parts described in any embodiment of this application (e.g., the information acquisition module, comparison module, and quality inspection module in a multi-mode information-based friction welding quality inspection system for mechanical parts). Memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the device, etc. Furthermore, memory may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0071] The communication module is used for data transmission.

[0072] The processor executes various functional applications and data processing of the device by running software programs, instructions, and modules stored in memory, thereby realizing the above-mentioned quality inspection method for friction welding of mechanical parts based on multi-mode information.

[0073] Input devices can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of the device. Output devices may include display devices such as displays.

[0074] The electronic device provided above can be used to perform the friction welding quality inspection method for mechanical parts based on multi-mode information provided in Embodiment 1 above, and has the corresponding functions and beneficial effects.

[0075] Example 4: This application embodiment also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a multi-mode information-based friction welding quality inspection method for mechanical parts. This multi-mode information-based friction welding quality inspection method for mechanical parts includes: acquiring mold release agent process information of the target mechanical part in the die-casting process, mold release agent residue treatment parameters in the die-casting post-processing process, and weld end face images in the welding process. The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and weld end face machining parameters; collecting data on the target mechanical part... The system uses real-time sensing data during the friction welding process. This data is compared with standard sensing data under a preset normal welding mode. When the real-time sensing data deviates from the standard sensing data by a set value, the system determines that the target mechanical part has a welding quality abnormality. The real-time sensing data includes at least one of the following: torque signal, thermal imaging signal, acoustic emission signal, and vibration signal of the target mechanical part during the welding process. Based on the release agent process information, release agent residue treatment parameters, and welding end face image, the system analyzes the welding quality abnormality of the target mechanical part, determines the root cause of the current welding quality abnormality, and outputs the friction welding quality inspection result of the target mechanical part based on the root cause.

[0076] Storage medium – any type of memory device or storage device. The term “storage medium” is intended to include: mounting media, such as CD-ROM, floppy disk, or magnetic tape devices; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements, etc. Storage medium may also include other types of memory or combinations thereof. Furthermore, storage medium may reside in a first computer system in which the program is executed, or it may reside in a different second computer system connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term “storage medium” can include two or more storage media residing in different locations (e.g., in different computer systems connected via a network). Storage medium may store program instructions (e.g., specifically implemented as a computer program) executable by one or more processors.

[0077] Of course, the computer-executable instructions provided in the embodiments of this application are not limited to the multi-mode information-based friction welding quality inspection method for mechanical parts as described above, but can also perform related operations in the multi-mode information-based friction welding quality inspection method for mechanical parts provided in any embodiment of this application.

[0078] The mechanical parts friction welding quality inspection system, storage medium and electronic device based on multi-mode information provided in the above embodiments can execute the mechanical parts friction welding quality inspection method based on multi-mode information provided in any embodiment of this application. For technical details not described in detail in the above embodiments, please refer to the mechanical parts friction welding quality inspection method based on multi-mode information provided in any embodiment of this application.

[0079] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A quality inspection method for friction welding of mechanical parts based on multi-modal information, characterized in that, include: The process information of the release agent in the die casting process of the target mechanical part, the release agent residue treatment parameters in the die casting post-processing process, and the welding end face image in the welding process are obtained. The release agent process information includes release agent spraying parameters and mold temperature distribution information. The release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters. Real-time sensing data of the target mechanical part during the friction welding process is collected, and the real-time sensing data is compared with the standard sensing data of the preset normal welding mode. When the real-time sensing data is in a set deviation state relative to the standard sensing data, it is determined that the target mechanical part has a welding quality abnormality. The real-time sensing data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal and vibration signal of the target mechanical part during the welding process. Based on the release agent process information, the release agent residue treatment parameters, and the welding end face image, the welding quality anomaly analysis of the target mechanical part is performed to determine the root cause of the current welding quality anomaly, and the friction welding quality inspection result of the target mechanical part is output according to the root cause.

2. The method for quality inspection of friction welding of mechanical parts based on multi-mode information according to claim 1, characterized in that, The analysis of welding quality anomalies in the target mechanical part based on the release agent process information, the release agent residue treatment parameters, and the weld end face image, to determine the root cause of the current welding quality anomaly, includes: The process information of the release agent is compared with the preset standard process parameter range of the release agent to obtain a first comparison result; the residual treatment parameters of the release agent are compared with the preset standard post-treatment parameter range to obtain a second comparison result; and the welded end face image is compared with the preset standard end face image to obtain a third comparison result. An abnormal comparison result is determined from the first comparison result, the second comparison result, and the third comparison result, and the root cause of the current welding quality abnormality is determined based on the abnormal comparison result.

3. The method for quality inspection of friction welding of mechanical parts based on multi-mode information according to claim 2, characterized in that, The step of comparing the release agent process information with a preset standard release agent process parameter range to obtain a first comparison result includes: The release agent spraying parameters are compared with the preset standard spraying parameter range, and the mold temperature distribution information is compared with the preset standard temperature distribution information to calculate the temperature deviation value of each temperature measuring point. When the release agent spraying parameters exceed the standard spraying parameter range or the temperature deviation value of any temperature measurement point exceeds the preset temperature deviation threshold, the first comparison result is marked as abnormal, and the abnormality type is recorded as the corresponding release agent spraying abnormality or mold temperature abnormality.

4. The method for quality inspection of friction welding of mechanical parts based on multi-mode information according to claim 2, characterized in that, The step of determining abnormal comparison results from the first comparison result, the second comparison result, and the third comparison result, and determining the root cause of the current welding quality abnormality based on the abnormal comparison results, includes: When there is an abnormal comparison result among the first comparison result, the second comparison result and the third comparison result, the abnormal type corresponding to the current abnormal comparison result is determined as the abnormal root cause of the welding quality abnormality; When there are multiple abnormal comparison results among the first comparison result, the second comparison result, and the third comparison result, the abnormal type corresponding to the abnormal comparison result with the highest priority is selected as the abnormal root cause of the welding quality abnormality according to the preset priority order.

5. The method for quality inspection of friction welding of mechanical parts based on multi-mode information according to claim 1, characterized in that, The analysis of welding quality anomalies in the target mechanical part based on the release agent process information, the release agent residue treatment parameters, and the weld end face image, to determine the root cause of the current welding quality anomaly, includes: The release agent process information, the release agent residue treatment parameters, the weld end face image, and the real-time sensor data are input into the pre-constructed multimodal fusion model; Based on the multimodal fusion model, visual features are extracted from the welded end face image, and a visual feature vector representing the distribution state of the release agent residue is output. Based on the multimodal fusion model, time-series modeling is performed on the real-time sensing data to determine the time-series change feature vector of the target mechanical part during the welding process. Based on the multimodal fusion model, the release agent process information and the release agent residue treatment parameters are encoded, and a statistical feature vector representing the pre-welding process state of the target mechanical part is output. The visual feature vector, the temporal variation feature vector, and the statistical feature vector are fused to obtain a multimodal fusion feature. The multimodal fusion feature is then input into a classification network to perform welding quality anomaly analysis on the target mechanical part. The root cause of the current welding quality anomaly is output, which includes at least one of the following: abnormal release agent spraying, abnormal mold temperature, insufficient shot blasting, excessive cleaning residue, and poor machining of the weld front end face.

6. The method for quality inspection of friction welding of mechanical parts based on multi-mode information according to claim 5, characterized in that, The training process of the multimodal fusion model includes: Obtain a training dataset, which includes multiple sample mechanical parts, each corresponding to a sample of mold release agent process information, mold release agent residue treatment parameters, welding end face image samples, welding process sensor data samples, and pre-labeled abnormal root cause type labels. An initial multimodal fusion model is constructed, which includes a visual feature extraction network, a temporal feature extraction network, a statistical feature encoding network, a feature fusion layer, and a classification network; The sample of the release agent process information and the sample of the release agent residue treatment parameters are input into the statistical feature encoding network to output a statistical feature vector sample. The sample of the weld end face image is input into the visual feature extraction network to output a visual feature vector sample. The sample of the welding process sensor data is input into the temporal feature extraction network to output a temporal feature vector sample. The statistical feature vector samples, the visual feature vector samples, and the temporal feature vector samples are input into the feature fusion layer for fusion to obtain multimodal fusion feature samples. The multimodal fusion feature samples are then input into the classification network to output the predicted anomaly root cause category. Based on the difference between the predicted root cause category and the corresponding root cause type label, a loss function value is calculated, and the network parameters of the multimodal fusion model are iteratively updated according to the loss function value until the loss function value converges, thus obtaining the trained multimodal fusion model.

7. The method for quality inspection of friction welding of mechanical parts based on multi-mode information according to claim 1, characterized in that, The step of comparing the real-time sensing data with standard sensing data under a preset normal welding mode, and determining that the target mechanical part has a welding quality abnormality when the real-time sensing data deviates from the standard sensing data by a set state, includes: The torque signal, thermal imaging signal, acoustic emission signal, and vibration signal in the real-time sensing data are compared with their respective preset standard sensing data. When any signal comparison result exceeds the corresponding preset threshold range, it is determined that the real-time sensing data is in a set deviation state relative to the standard sensing data, and the target mechanical part is judged to have welding quality abnormality.

8. A quality inspection system for friction welding of mechanical parts based on multi-modal information, characterized in that, include: The information acquisition module is used to acquire the mold release agent process information of the target mechanical part in the die casting process, the mold release agent residue treatment parameters in the die casting post-processing process, and the welding end face image in the welding process. The mold release agent process information includes mold release agent spraying parameters and mold temperature distribution information. The mold release agent residue treatment parameters include at least one of shot blasting parameters, cleaning parameters, and welding end face machining parameters. The comparison module is used to collect real-time sensing data of the target mechanical part during the friction welding process, compare the real-time sensing data with the standard sensing data of the preset normal welding mode, and determine that the target mechanical part has a welding quality abnormality when the real-time sensing data is in a set deviation state relative to the standard sensing data. The real-time sensing data includes at least one of the torque signal, thermal imaging signal, acoustic emission signal and vibration signal of the target mechanical part during the welding process. The quality inspection module is used to perform welding quality anomaly analysis on the target mechanical part based on the release agent process information, the release agent residue treatment parameters, and the welding end face image, determine the root cause of the current welding quality anomaly, and output the friction welding quality inspection result of the target mechanical part according to the root cause.

9. An electronic device, characterized in that, include: Memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-mode information-based friction welding quality inspection method for mechanical parts as described in any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the friction welding quality inspection method for mechanical parts based on multi-mode information as described in any one of claims 1-7.

Images