Intelligent production method, device and medium for heat exchanger

Abstract

This application discloses a smart production method, equipment, and medium for heat exchangers, relating to the field of computer technology. The method includes: generating a digital identifier for each heat exchanger and storing it in association with the heat exchanger's process parameter set; collecting surface temperature field data and visible light image data of the heat exchanger under test during the cooling stage after brazing is completed, and performing multimodal fusion processing to generate brazing quality inspection results, and storing the brazing quality inspection results in association with the digital identifier of the heat exchanger under test; if the brazing quality inspection results are lower than a preset quality threshold, retrieving the process parameter set associated with the digital identifier of the heat exchanger under test, and identifying the preceding process node and corresponding abnormal parameters that caused the current brazing quality abnormality based on the process parameter set; generating process correction data corresponding to the preceding process node based on the abnormal parameters and sending it to the production equipment of the preceding process node to drive the preceding process node to adjust its operating parameters.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a smart production method, equipment and medium for heat exchangers. Background Technology

[0002] Heat exchangers, as core components of thermal management systems, are widely used in automotive transmission oil coolers, hybrid vehicle cooling systems, and power battery thermal management. Plate-fin and finned heat exchangers have become the mainstream technology due to their compact structure and high heat exchange efficiency. The manufacturing of these heat exchangers typically involves multiple processes, including core assembly, fin installation, pipe riveting, and vacuum brazing. Vacuum brazing is a critical step determining the final product quality; factors such as temperature uniformity, holding time, and cooling rate directly affect each weld joint. Defects in brazing quality can lead to leaks or even explosions during operation, severely impacting the safety and reliability of the entire vehicle.

[0003] Currently, the traditional methods for inspecting the brazing quality of heat exchangers mainly include sampling burst testing and offline metallographic examination. Burst testing involves pressurizing a sample of each batch of products until it breaks down to determine if the brazing process of that batch is up to standard. This method is destructive and a post-inspection process, making it difficult to cover all products. Even if a sampled product passes inspection, it does not guarantee that other products in the same batch are free of welding defects, resulting in a high risk of missed defects. Metallographic examination, on the other hand, requires cutting, sample preparation, and microscopic observation of the product. This process is time-consuming and also destructive, making online full inspection impossible. Summary of the Invention

[0004] This application provides a method, equipment, and medium for intelligent production of heat exchangers to solve the above-mentioned technical problems.

[0005] On one hand, embodiments of this application provide a smart manufacturing method for heat exchangers, including: A unique digital identifier is generated for each heat exchanger produced online, and the digital identifier is associated with and stored in relation to the process parameter set of the corresponding heat exchanger in multiple production processes; During the cooling stage after brazing, surface temperature field data and visible light image data of the heat exchanger under test are collected simultaneously. Multimodal fusion processing is performed on the surface temperature field data and visible light image data to generate brazing quality inspection results. The brazing quality inspection results are then associated with and stored with the digital identifier of the heat exchanger under test. If the brazing quality test result is lower than the preset quality threshold, the process parameter set associated with the digital identifier of the heat exchanger under test is retrieved, and based on the process parameter set, the preceding process node that caused the current brazing quality abnormality and the abnormal parameters corresponding to the preceding process node are identified. Based on the abnormal parameters, process correction data corresponding to the preceding process node is generated, and the process correction data is sent to the production equipment of the preceding process node to drive the preceding process node to adjust its operating parameters.

[0006] In one implementation of this application, during the cooling stage after brazing is completed, surface temperature field data and visible light image data of the heat exchanger under test are simultaneously acquired, specifically including: An infrared thermal imaging acquisition device and a visible light imaging device are installed on the outer side of the brazing furnace outlet end, so that the optical axes of the infrared thermal imaging acquisition device and the visible light imaging device are parallel to each other. According to the preset sampling frequency, the infrared thermal imaging acquisition device continuously acquires multiple frames of infrared thermal images of the heat exchanger under test as it passes through multiple acquisition positions in sequence on the cooling path, and the visible light imaging device continuously acquires multiple frames of visible light images of the heat exchanger under test as it passes through the same acquisition position in sequence on the same cooling path. For each frame of infrared thermal image, preprocessing is performed to generate surface temperature field data of the heat exchanger under test; the preprocessing includes non-uniformity correction processing, bad pixel compensation processing, and temperature inversion calculation; Median filtering is performed on each frame of the visible light image to remove random impulse noise from the visible light image. Histogram equalization is performed to adjust the pixel grayscale distribution of the visible light image, and Gaussian noise in the visible light image is suppressed while preserving the welding contour edges to generate surface visible light image data.

[0007] In one implementation of this application, multimodal fusion processing is performed on the surface temperature field data and the surface visible light image data to generate brazing quality inspection results, specifically including: Based on the three-dimensional digital model data of the heat exchanger under test, the welding structures in the heat exchanger under test are classified, and regions of interest of multiple key welding nodes are extracted from the surface temperature field data and surface visible light image data; the key welding nodes include at least one of the following: the top plate cup-mouth welding area, the fin welding area, and the flat plate protrusion welding area. The surface temperature field data and surface visible light image data within the region of interest of each key welding node are subjected to multimodal fusion processing to generate a fused feature map corresponding to the key welding node. Each fused feature map is input into a pre-trained quality assessment network to output a welding quality score for each key welding node. The welding quality scores of all key welding areas are then weighted and merged and compared with a preset welding quality pass threshold to generate brazing quality inspection results.

[0008] In one implementation of this application, based on the three-dimensional digital model data of the heat exchanger under test, the welded structures in the heat exchanger under test are classified, and regions of interest of multiple key weld nodes are extracted from the surface temperature field data and surface visible light image data, specifically including: The three-dimensional coordinates and welding profile geometric parameters of each welding area of ​​the heat exchanger under test are retrieved from the pre-established database. Based on the spatial pose parameters of the infrared imaging acquisition device and the visible light imaging device relative to the heat exchanger under test, a projection transformation matrix is ​​established from the world coordinate system to the surface temperature field data image coordinate system and the surface visible light image data image coordinate system. Based on the projection transformation matrix, the three-dimensional coordinates of multiple preset key welding nodes are mapped to the pixel coordinate positions corresponding to the surface temperature field data and the surface visible light image data. Using the pixel coordinates as the center point, the region of interest is extended outward along the welding contour geometry parameters to the geometric region corresponding to the key welding node to generate the region of interest mask for the key welding node. The region of interest mask is then superimposed on the surface temperature field data and the surface visible light image data.

[0009] In one implementation of this application, the surface temperature field data and visible light image data of the surface within the region of interest of each key welding node are subjected to multimodal fusion processing to generate a fused feature map corresponding to the key welding node, specifically including: The surface temperature field data and surface visible light image data of the region of interest of the key welding node are respectively cropped into sub-images of the same size; The sub-images of the surface temperature field data are normalized to map the pixel values ​​of the sub-images of the surface temperature field data to the same numerical range as the sub-images of the surface visible light image data. The normalized surface temperature field sub-image and the surface visible light image sub-image are concatenated along the channel dimension to form a multi-channel input tensor. The multi-channel input tensor is then input into an image fusion network with a pre-constructed encoder-decoder structure to output a fused feature map. The fused feature map includes temperature gradient distribution information and welding contour geometric parameters.

[0010] In one implementation of this application, when the brazing quality test result is lower than a preset quality threshold, the process parameter set associated with the digital identifier of the heat exchanger under test is retrieved, and based on the process parameter set, the preceding process node that caused the current brazing quality abnormality, and the abnormal parameters corresponding to the preceding process node, are identified, specifically including: If the brazing quality test result is lower than the preset quality threshold, based on the digital identifier of the heat exchanger under test, all process parameter sets recorded in the preceding process nodes of the heat exchanger under test are traversed and extracted from the database. The process parameter set includes at least the assembly pressure value in the core assembly process, the fin pitch deviation value in the automatic fin forming process, the riveting force curve characteristic parameters in the pipe riveting process, and the segmented temperature values ​​of the furnace temperature curve and the conveyor belt speed value in the brazing process. The process parameter set is input into a pre-trained inverse mapping model to output the abnormal contribution score of each preceding process node, and to identify the preceding process node whose abnormal contribution score exceeds a preset abnormal threshold, as well as the corresponding process parameter item as the abnormal source process and abnormal parameter.

[0011] In one implementation of this application, after sending the process correction data to the production equipment of the preceding process node to drive the preceding process node to adjust its operating parameters, the method further includes: After the process correction data is issued, the trend of the welding quality inspection results of the heat exchangers produced in the subsequent process nodes is continuously monitored. When the welding quality test results of the predetermined number of heat exchangers produced in subsequent continuous production all exceed the preset quality threshold, the current combination of process parameters is solidified into the process parameter knowledge base of the preceding process node, and used as the optimized process benchmark parameters of the preceding process node for the current heat exchanger model. If, in the subsequent continuous production of a predetermined number of heat exchangers, there are still abnormal heat exchangers whose welding quality test results are lower than the preset quality threshold, a multi-level alarm is triggered to package the abnormal heat exchanger's digital identifier, abnormal process information, and the input and output data of the current fault propagation inverse mapping model into an abnormal report. The anomaly report is sent to the administrator, and the production task of the production line where the preceding process node is located is suspended until a production resumption signal is received to resume production.

[0012] In one implementation of this application, a unique digital identifier is generated for each heat exchanger produced online, and the digital identifier is associated with and stored in relation to the process parameter set of the corresponding heat exchanger in multiple production steps. Specifically, this includes: Before the heat exchanger enters the first production station, a unique digital identifier is generated for the top plate surface of the heat exchanger by a writing device; As the heat exchanger passes through each production station in sequence, the process parameters of the heat exchanger at each production station are acquired in real time through data acquisition terminals installed at each production station. The production stations include core assembly station, fin installation station, top plate assembly station, bottom plate assembly station, press-fitting station, and vacuum brazing station. The process parameters include core assembly pressure value, fin positioning deviation value, top plate riveting force curve, and pressing displacement and pressing force value of the press-fitting fixture. Upon completion of each production process, the process parameters of the current process are merged and stored with the existing records in the associated database, using the digital identifier of the heat exchanger as an index, to form a historical archive of the entire process parameters of the heat exchanger from core assembly to brazing completion.

[0013] On the other hand, this application also provides an intelligent production device for heat exchangers, the device comprising: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a heat exchanger intelligent production method as described above.

[0014] On the other hand, this application embodiment also provides a non-volatile computer storage medium storing computer-executable instructions, which, when executed, implement the above-described intelligent production method for heat exchangers.

[0015] This application provides a smart production method, equipment, and medium for heat exchangers, which has at least the following beneficial effects: By assigning a unique digital identifier to each heat exchanger in online production and linking these identifiers with process parameter sets across multiple production steps, a traceable data archive for the entire process was established. During the cooling phase after brazing, surface temperature field data and visible light image data of the heat exchanger were simultaneously collected. Multimodal fusion processing of these two types of data was then performed to generate brazing quality inspection results. This comprehensively utilized the sensitivity of temperature field distribution to internal weld defects and the ability of visible light images to distinguish the geometric features of the weld surface contour, enabling the inspection system to simultaneously capture geometric defects and improve the accuracy of brazing quality inspection. By retrieving all process parameter sets linked to the digital identifier of the abnormal heat exchanger and identifying the preceding process nodes and their corresponding abnormal parameters that caused the current brazing quality anomaly, reverse precision localization from terminal quality defects to source process deviations was achieved, shortening the time for root cause localization. Based on the identified abnormal parameters, process correction data for the preceding process nodes was automatically generated and distributed to the corresponding production equipment, achieving full automation from quality inspection to anomaly tracing and process correction. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A schematic flowchart of a smart production method for heat exchangers provided in an embodiment of this application; Figure 2 This is a schematic diagram of the internal structure of an intelligent production equipment for heat exchangers provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0019] Figure 1 This is a schematic flowchart of a smart production method for heat exchangers provided in an embodiment of this application.

[0020] The analysis method involved in the embodiments of this application can be implemented by a terminal device or a server, and this application does not impose any special limitations on it. For ease of understanding and description, the following embodiments are all described in detail using a server as an example.

[0021] It should be noted that the server can be a single device or a system composed of multiple devices, i.e., a distributed server. This application does not make any specific limitations on this.

[0022] like Figure 1 As shown in the embodiment of this application, a smart manufacturing method for heat exchangers includes: Step 101: Generate a unique digital identifier for each heat exchanger produced online, and associate and store the digital identifier with the process parameter set of the corresponding heat exchanger in multiple production processes.

[0023] In this embodiment, a unique digital identifier is assigned to each online production heat exchanger as an identity code. Its core function is to provide a unified index key for the massive amounts of process data generated in subsequent processes, ensuring that the entire lifecycle data of each heat exchanger, from core assembly to finished product completion, can be accurately collected and retrieved in a structured manner. For example, the digital identifier can be implemented using various physical carriers. For instance, a QR code or one-dimensional barcode pattern can be formed by laser marking on a predetermined area of ​​the heat exchanger's top plate surface. Alternatively, an RFID tag writing device can be used to write the digital code into an RFID tag pasted on the non-welding area of ​​the heat exchanger. Or, a plaintext digital string can be directly stamped or etched onto the heat exchanger body. It should be noted that regardless of the form used, the unique digital identifier must be reliably readable by the corresponding reading device throughout the entire production process and able to withstand the high-temperature environment of subsequent vacuum brazing without damage or detachment. Specifically, before the heat exchanger enters the first production station, a unique digital identifier is generated and assigned to the heat exchanger through the aforementioned writing device. At the same time, an independent data storage partition is created for the digital identifier in the database of the production management system for subsequent association and storage of the heat exchanger's process parameters in each process.

[0024] As the heat exchanger undergoes multiple production processes, each production station is equipped with a corresponding data acquisition terminal. For example, these production stations include a core assembly station, a fin installation station, a top plate assembly station, a bottom plate assembly station, a press-fitting station, and a vacuum brazing station. The data acquisition terminal can be a sensor and programmable logic controller (PLC) data acquisition module integrated within the production equipment, or it can be a separately installed acquisition system composed of displacement sensors, pressure sensors, force sensors, and other detection elements combined with a data acquisition card. Each data acquisition terminal acquires the process parameters of the heat exchanger at its current station in real time according to a preset sampling frequency.

[0025] Specifically, the parameters collected at the core assembly station include at least the assembly pressure value. This value reflects the pressure exerted on each layer of plates during the pressing process. Deviations in the assembly pressure value directly affect the tightness of the fit between the plates and the flow and filling effect of the brazing filler metal during subsequent brazing. The parameters collected at the fin assembly station include at least the fin positioning deviation value. This value reflects the offset between the placement position of the inner fin on the plate and its theoretical design position. Excessive offset leads to insufficient contact area between the fin and the plate, thus affecting the welding bonding rate. The parameters collected at the top plate assembly station and the bottom plate assembly station include at least the riveting force curve. This curve records the trajectory of the riveting force over time during the riveting process. The peak value and waveform characteristics of the trajectory can reflect whether the riveting is in place and whether there are cases of excessively tight or loose riveting.

[0026] The parameters collected at the press-fitting station include at least the pressing displacement and pressing force of the press-fitting fixture. These two parameters together determine the initial pre-tightening state of the core before it enters the brazing furnace. The magnitude of the pre-tightening force directly affects the relative sliding and thermal deformation coordination between the layers of plates during the brazing heating process. The parameters collected at the vacuum brazing station include at least the segmented temperature values ​​of the furnace temperature curve and the conveyor belt speed. The furnace temperature curve reflects the temperature change over time during the brazing process. The heating rate, holding temperature, holding time, and cooling rate all have a decisive influence on the metallurgical bonding quality of the brazed joint. The conveyor belt speed determines the residence time distribution of the heat exchanger core in each temperature zone of the furnace.

[0027] Understandably, at the completion of each production process, the process parameters of the current process collected for the corresponding process are merged and stored with the existing records in the database associated with the current digital identifier, using the digital identifier of the heat exchanger as an index. This process is continuously updated until a complete historical archive of process parameters for the entire heat exchanger process, from core assembly to brazing, is formed. This complete historical archive of process parameters provides an accurate and complete data foundation for tracing the source of quality anomalies in subsequent processes.

[0028] Step 102: During the cooling stage after brazing is completed, the surface temperature field data and visible light image data of the heat exchanger under test are collected simultaneously. The surface temperature field data and visible light image data are then subjected to multimodal fusion processing to generate brazing quality inspection results. The brazing quality inspection results are then associated with and stored with the digital identifier of the heat exchanger under test.

[0029] In this embodiment, the cooling stage after brazing is a critical window for inspecting brazing quality. Specifically, an inspection station is arranged outside the outlet of the vacuum brazing furnace, along the path of the heat exchanger core after it exits the furnace and cools naturally or by forced cooling. Each inspection station is equipped with at least one set of infrared thermal imaging and visible light imaging devices, spatially adjacent and with parallel optical axes. After pre-calibration of the field of view, their imaging fields of view are substantially overlapped at predetermined inspection positions along the cooling path. The infrared thermal imaging device is used to capture the temperature field distribution information on the heat exchanger surface caused by differences in welding quality. Areas with high brazing pass rates have relatively low surface temperatures due to good weld bonding and continuous metal heat conduction paths, resulting in faster heat dissipation during cooling. Conversely, areas with low brazing pass rates have relatively high surface temperatures due to defects such as incomplete welds, lack of fusion, or porosity, which obstruct heat conduction at the weld interface, leading to greater heat accumulation. The visible light imaging device is used to capture the geometric information of the weld contour on the heat exchanger surface, including the continuity of the weld, the boundary position of the weld area, and the presence of visible defects such as surface cracks.

[0030] When the heat exchanger under test moves to the predetermined detection position with the conveyor belt, the infrared thermal imaging acquisition device is triggered to continuously acquire multiple frames of infrared thermal images at a preset sampling frequency. Simultaneously, the visible light imaging device is triggered to synchronously acquire multiple frames of visible light images at the same sampling frequency, ensuring that the two types of data are strictly aligned in the time dimension. For each frame of the acquired infrared thermal image, non-uniformity correction processing is first performed to eliminate fixed pattern noise caused by inconsistent response rates between pixels of the infrared detector. Then, bad pixel compensation processing is performed to replace abnormal pixel values ​​in the infrared thermal image caused by detector pixel failure. Finally, based on the blackbody calibration parameters, the grayscale value of each pixel is calculated by temperature inversion and converted into the corresponding absolute temperature value to generate the surface temperature field data of the heat exchanger at the corresponding position on the cooling path.

[0031] For each frame of synchronously acquired visible light image, median filtering is first performed to remove random impulse noise from the image while preserving the edge information of the welding area. Then, histogram equalization is performed to adjust the pixel grayscale distribution of the image and stretch the contrast, making the welding area, which was originally blurred due to uneven lighting, clear. Finally, edge-preserving filtering is used to further suppress Gaussian noise and enhance the sharp edges of the welding contour, generating visible light image data of the surface.

[0032] Subsequently, the collected surface temperature field data and visible light image data are subjected to multimodal fusion processing to generate brazing quality inspection results. Specifically, firstly, based on the three-dimensional digital model data of the heat exchanger under test, the welded structures in the heat exchanger are classified, and regions of interest (ROIs) of multiple key weld nodes are extracted from the surface temperature field data and visible light image data. It should be noted that key weld nodes include at least one or more of the following types: the top plate cup-mouth weld area, the fin weld area, and the flat plate bump weld area.

[0033] When extracting the region of interest (ROI), the three-dimensional coordinates and welding contour geometric parameters of each welded area of ​​the heat exchanger under test are retrieved from a pre-established heat exchanger product structure database. Based on the spatial pose parameters of the infrared thermal imaging acquisition device and the visible light imaging device relative to the heat exchanger under test, a projection transformation matrix is ​​established from the world coordinate system to the surface temperature field data image coordinate system and the surface visible light image data image coordinate system. Then, based on the projection transformation matrix, the three-dimensional coordinates of multiple preset key weld nodes are mapped to the corresponding pixel coordinate positions in the surface temperature field data and surface visible light image data. Next, using the pixel coordinate positions as the center point, the process extends outward along the welding contour geometric parameters to the geometric region corresponding to the key weld nodes, generating a ROI mask for the key weld nodes. This ROI mask is then superimposed on the surface temperature field data and surface visible light image data, completing the automatic annotation of the key weld nodes.

[0034] For each critical welding node, the surface temperature field data and the visible light image data within the region of interest are cropped into sub-images of the same size. The surface temperature field data sub-image is then normalized, mapping pixel values ​​to the same numerical range as the visible light image data sub-image, thereby eliminating dimensional differences between the two modal data. Then, the normalized surface temperature field sub-image and the visible light image sub-image are concatenated along the channel dimension to form a multi-channel input tensor. This multi-channel input tensor is then fed into a pre-constructed encoder-decoder structured image fusion network.

[0035] It should be noted that the encoder part of the image fusion network consists of multiple stacked convolutional layers. Each convolutional layer is followed by a batch normalization layer and a modified linear unit activation layer, used to extract shallow geometric features and deep temperature distribution features layer by layer from the input multi-channel tensor. The decoder part consists of multiple stacked upsampling layers. Each upsampling layer is followed by a convolutional layer and a skip connection structure. The skip connection structure concatenates the feature map of the same level in the encoder part with the upsampled feature map in the channel dimension to compensate for the geometric details lost during the upsampling process. The last convolutional layer of the decoder part outputs a single-channel fused feature map with the same size as the input image. The fused feature map retains both temperature gradient distribution information and welding contour geometric parameters.

[0036] Understandably, image fusion networks can be configured with differentiated feature processing modules for different types of critical welding nodes. For example, when the input critical welding node is the welding area of ​​the cup rim on the top plate, a channel-spatial serial attention module is added at the end of the encoder. This module first calculates the global average pooling vector and global max pooling vector of the input feature map along the channel dimension. After nonlinear transformation using a shared multilayer perceptron, the two vectors are added together, and then a normalized exponential function is used to generate channel attention weights. These weights are multiplied channel-by-channel with the input feature map to obtain a channel recalibration feature map, thereby strengthening the response intensity of channels sensitive to temperature field distribution characteristics. Then, the global maximum and global average values ​​of the channel recalibration feature map are calculated along the spatial dimension. These values ​​are then used by a convolutional layer to generate spatial attention weights. These weights are multiplied position-by-position with the channel recalibration feature map and output to the decoder, thereby focusing on local areas with abnormal heat diffusion in the temperature field and achieving accurate detection of weld defects in the cup rim welding area.

[0037] When the key welding node input is the welding area where the fins and the plate meet, a skip-layer edge enhancement module is inserted between the upsampling layer and the convolutional layer in the decoder. The skip-layer edge enhancement module takes the feature map output by the current decoding layer as input, calculates the gradient magnitude of each pixel in the feature map along the horizontal and vertical directions respectively, and adds and fuses the gradient magnitude map with the original feature map pixel by pixel according to the preset weights to obtain the feature map after edge enhancement. This highlights the geometric contour of the welding contact boundary and enables accurate detection of geometric defects such as insufficient fin welding length and insufficient contact area.

[0038] The fused feature maps of each key welding node are input into a pre-trained quality assessment network. It should be noted that the quality assessment network employs a multi-branch parallel convolutional structure, with each branch receiving the fused feature map of the corresponding type of key welding node. After alternating stacking of successive convolutional layers, batch normalization layers, and modified linear unit activation layers, the welding quality representation vector of the key welding node is extracted. Then, the representation vectors output from all branches are concatenated along the feature dimension to form a comprehensive quality feature vector. This vector is then subjected to layer-by-layer nonlinear transformation and dimensionality compression by a regression network composed of stacked fully connected layers, ultimately outputting the welding quality score for each key welding node. The welding quality scores of all key welding areas are weighted and merged, then compared with a preset welding quality pass threshold to generate the brazing quality inspection result of the heat exchanger under test. The brazing quality inspection result is then associated with and stored with the unique digital identifier of the heat exchanger under test, serving as an important component of the heat exchanger's quality archive.

[0039] Step 103: If the brazing quality test result is lower than the preset quality threshold, retrieve the process parameter set associated with the digital identifier of the heat exchanger under test, and based on the process parameter set, identify the preceding process node that caused the current brazing quality abnormality, as well as the abnormal parameters corresponding to the preceding process node.

[0040] In this embodiment, the preset quality threshold is a welding quality acceptance limit value pre-set according to the design requirements of the heat exchanger product and industry standards. When the brazing quality test result of a heat exchanger is lower than the preset quality threshold, it indicates that there is a welding quality defect in the heat exchanger during the brazing process, and the abnormality tracing process needs to be initiated immediately. At this time, the system uses the digital identifier of the heat exchanger under test as an index to traverse and extract all process parameter sets associated with the digital identifier from the database of the production management system. It should be noted that the process parameter set covers all process parameters recorded in all preceding processes of the heat exchanger from the core assembly process to the vacuum brazing process, including at least the assembly pressure value in the core assembly process, the fin pitch deviation value in the automatic fin forming process, the riveting force curve characteristic parameters in the pipe riveting process, and the segmented temperature values ​​of the furnace temperature curve and the conveyor belt speed values ​​in the brazing process.

[0041] Subsequently, the extracted set of all process parameters is input into a pre-trained inverse mapping model. The inverse mapping model is a machine learning model pre-trained based on multiple sets of historical training samples. Each set of training samples contains the complete process parameter set of a historical heat exchanger and its corresponding final brazing quality detection result. During training, the process parameters of each step in the complete process parameter set are used as input features, and the corresponding brazing quality detection results are used as supervision labels. Training is performed based on a random forest algorithm or structural equation model, enabling the model to learn and establish a complex mapping relationship between the process parameters of each preceding step and the final brazing quality result, and quantitatively output the influence weight coefficients of each preceding step process parameter on the final brazing quality result. After inputting the complete set of process parameters of the current abnormal heat exchanger into the pre-trained inverse mapping model, the model outputs the abnormal contribution score of each preceding step. It should be noted that the abnormal contribution score quantitatively reflects the degree of contribution of the deviation of each preceding step process parameter to the final brazing quality abnormality. Preceding step nodes whose abnormal contribution scores exceed a preset abnormality threshold and their corresponding process parameter items are identified as the abnormal source processes and abnormal parameters causing the current brazing quality abnormality.

[0042] Step 104: Based on the abnormal parameters, generate the process correction data corresponding to the preceding process node, and send the process correction data to the production equipment of the preceding process node to drive the preceding process node to adjust its operating parameters.

[0043] In this embodiment, after identifying the abnormal source process and its corresponding abnormal parameters, the system generates process correction data for the preceding process nodes based on the abnormal parameters. Specifically, the system retrieves the standard process parameter values ​​of the abnormal source process under the current heat exchanger model from the process parameter knowledge base, calculates the deviation between the current actual process parameter value and the standard parameter value, and then generates process correction data according to the process control logic and parameter adjustment rules of the corresponding process. It should be noted that the process correction data includes the parameter adjustment direction and adjustment step size. For example, if the abnormal source process is a vacuum brazing process and the abnormal parameter is that the actual temperature of a certain temperature zone deviates from the standard temperature curve, then the process correction data includes instructions to increase or decrease the set temperature of the temperature zone by a specific amount; if the abnormal source process is a core assembly process and the abnormal parameter is that the assembly pressure value is too low, then the process correction data includes instructions to increase the assembly pressure to the target value. Process correction data is sent to the controller of the production equipment in the corresponding preceding process node via industrial Ethernet or fieldbus network in a control command format that the equipment can recognize, and drives the production equipment to automatically adjust the corresponding operating parameters so that the process parameters of the heat exchangers produced in subsequent processes return to the standard range.

[0044] After the process correction data is issued, the system continuously monitors the trend of welding quality inspection results of heat exchangers produced in subsequent processes during the brazing quality inspection stage. When the welding quality inspection results of the predetermined number of heat exchangers produced in subsequent consecutive processes all recover to above the preset quality threshold, it indicates that the process correction is effective. The system then solidifies the currently effective combination of process parameters into the process parameter knowledge base of the previous process node, serving as the optimized process benchmark parameters for the current heat exchanger model for subsequent production reference.

[0045] If, among the predetermined number of heat exchangers produced in subsequent continuous production, there are still abnormal heat exchangers whose welding quality inspection results are lower than the preset quality threshold, it indicates that the abnormality may involve more complex multi-process coupling factors or a fault in the equipment itself. At this time, the system triggers multi-level alarms and packages the digital identifier of the abnormal heat exchanger, the abnormal process information, and the input and output data of the current reverse mapping model into an abnormal report, which is sent to the administrator's terminal device. At the same time, the continuous production task of the production line where the preceding process node is located is suspended until the production can be resumed after receiving the production resumption signal from the administrator. This can prevent the continuous production of defective products from causing greater quality and cost losses.

[0046] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide an intelligent production equipment for heat exchangers, the structure of which is as follows: Figure 2 As shown.

[0047] Figure 2This is a schematic diagram of the internal structure of an intelligent production equipment for heat exchangers provided in an embodiment of this application. Figure 2 As shown, the device includes: At least one processor; And, a memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, and the instructions, when executed by at least one processor, enable at least one processor to: A unique digital identifier is generated for each heat exchanger produced online, and the digital identifier is associated with and stored in relation to the process parameter set of the corresponding heat exchanger in multiple production processes; During the cooling stage after brazing, surface temperature field data and visible light image data of the heat exchanger under test are collected simultaneously. Multimodal fusion processing is performed on the surface temperature field data and visible light image data to generate brazing quality inspection results. The brazing quality inspection results are then associated with and stored with the digital identifier of the heat exchanger under test. If the brazing quality test result is lower than the preset quality threshold, the process parameter set associated with the digital identifier of the heat exchanger under test is retrieved, and based on the process parameter set, the preceding process node that caused the current brazing quality abnormality and the abnormal parameters corresponding to the preceding process node are identified. Based on abnormal parameters, process correction data corresponding to the preceding process node is generated, and the process correction data is sent to the production equipment of the preceding process node to drive the preceding process node to adjust its operating parameters.

[0048] This application also provides a non-volatile computer storage medium storing computer-executable instructions, which, when executed, can: A unique digital identifier is generated for each heat exchanger produced online, and the digital identifier is associated with and stored in relation to the process parameter set of the corresponding heat exchanger in multiple production processes; During the cooling stage after brazing, surface temperature field data and visible light image data of the heat exchanger under test are collected simultaneously. Multimodal fusion processing is performed on the surface temperature field data and visible light image data to generate brazing quality inspection results. The brazing quality inspection results are then associated with and stored with the digital identifier of the heat exchanger under test. If the brazing quality test result is lower than the preset quality threshold, the process parameter set associated with the digital identifier of the heat exchanger under test is retrieved, and based on the process parameter set, the preceding process node that caused the current brazing quality abnormality and the abnormal parameters corresponding to the preceding process node are identified. Based on abnormal parameters, process correction data corresponding to the preceding process node is generated, and the process correction data is sent to the production equipment of the preceding process node to drive the preceding process node to adjust its operating parameters.

[0049] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0050] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for intelligent production of heat exchangers, characterized in that, The method includes: A unique digital identifier is generated for each heat exchanger produced online, and the digital identifier is associated with and stored in relation to the process parameter set of the corresponding heat exchanger in multiple production processes; During the cooling stage after brazing, surface temperature field data and visible light image data of the heat exchanger under test are collected simultaneously. Multimodal fusion processing is performed on the surface temperature field data and visible light image data to generate brazing quality inspection results. The brazing quality inspection results are then associated with and stored with the digital identifier of the heat exchanger under test. If the brazing quality test result is lower than the preset quality threshold, the process parameter set associated with the digital identifier of the heat exchanger under test is retrieved, and based on the process parameter set, the preceding process node that caused the current brazing quality abnormality and the abnormal parameters corresponding to the preceding process node are identified. Based on the abnormal parameters, process correction data corresponding to the preceding process node is generated, and the process correction data is sent to the production equipment of the preceding process node to drive the preceding process node to adjust its operating parameters.

2. The intelligent production method for heat exchangers according to claim 1, characterized in that, During the cooling phase after brazing, surface temperature field data and visible light image data of the heat exchanger under test are simultaneously acquired, specifically including: An infrared thermal imaging acquisition device and a visible light imaging device are installed on the outer side of the brazing furnace outlet end, so that the optical axes of the infrared thermal imaging acquisition device and the visible light imaging device are parallel to each other. According to the preset sampling frequency, the infrared thermal imaging acquisition device continuously acquires multiple frames of infrared thermal images of the heat exchanger under test as it passes through multiple acquisition positions in sequence on the cooling path, and the visible light imaging device continuously acquires multiple frames of visible light images of the heat exchanger under test as it passes through the same acquisition position in sequence on the same cooling path. For each frame of infrared thermal image, preprocessing is performed to generate surface temperature field data of the heat exchanger under test; the preprocessing includes non-uniformity correction processing, bad pixel compensation processing, and temperature inversion calculation; Median filtering is performed on each frame of the visible light image to remove random impulse noise from the visible light image. Histogram equalization is performed to adjust the pixel grayscale distribution of the visible light image, and Gaussian noise in the visible light image is suppressed while preserving the welding contour edges to generate surface visible light image data.

3. The intelligent production method for heat exchangers according to claim 1, characterized in that, The surface temperature field data and surface visible light image data are subjected to multimodal fusion processing to generate brazing quality inspection results, specifically including: Based on the three-dimensional digital model data of the heat exchanger under test, the welding structures in the heat exchanger under test are classified, and regions of interest of multiple key welding nodes are extracted from the surface temperature field data and surface visible light image data; the key welding nodes include at least one of the following: the top plate cup-mouth welding area, the fin welding area, and the flat plate protrusion welding area. The surface temperature field data and surface visible light image data within the region of interest of each key welding node are subjected to multimodal fusion processing to generate a fused feature map corresponding to the key welding node. Each fused feature map is input into a pre-trained quality assessment network to output a welding quality score for each key welding node. The welding quality scores of all key welding areas are then weighted and merged and compared with a preset welding quality pass threshold to generate brazing quality inspection results.

4. The intelligent production method for heat exchangers according to claim 3, characterized in that, Based on the three-dimensional digital model data of the heat exchanger under test, the welded structures in the heat exchanger under test are classified, and regions of interest for multiple key welded nodes are extracted from the surface temperature field data and surface visible light image data, specifically including: The three-dimensional coordinates and welding profile geometric parameters of each welding area of ​​the heat exchanger under test are retrieved from the pre-established database. Based on the spatial pose parameters of the infrared imaging acquisition device and the visible light imaging device relative to the heat exchanger under test, a projection transformation matrix is ​​established from the world coordinate system to the surface temperature field data image coordinate system and the surface visible light image data image coordinate system. Based on the projection transformation matrix, the three-dimensional coordinates of multiple preset key welding nodes are mapped to the pixel coordinate positions corresponding to the surface temperature field data and the surface visible light image data. Using the pixel coordinates as the center point, the region of interest is extended outward along the welding contour geometry parameters to the geometric region corresponding to the key welding node to generate the region of interest mask for the key welding node. The region of interest mask is then superimposed on the surface temperature field data and the surface visible light image data.

5. The intelligent production method for heat exchangers according to claim 3, characterized in that, The surface temperature field data and visible light image data within the region of interest of each key welding node are subjected to multimodal fusion processing to generate a fused feature map corresponding to the key welding node, specifically including: The surface temperature field data and visible light image data of the region of interest of the key welding node are cropped into sub-images of the same size. The sub-images of the surface temperature field data are normalized to map the pixel values ​​of the sub-images of the surface temperature field data to the same numerical range as the sub-images of the surface visible light image data. The normalized surface temperature field sub-image and the surface visible light image sub-image are concatenated along the channel dimension to form a multi-channel input tensor. The multi-channel input tensor is then input into an image fusion network with a pre-constructed encoder-decoder structure to output a fused feature map. The fused feature map includes temperature gradient distribution information and welding contour geometric parameters.

6. The intelligent production method for heat exchangers according to claim 1, characterized in that, If the brazing quality test result is lower than a preset quality threshold, the process parameter set associated with the digital identifier of the heat exchanger under test is retrieved. Based on the process parameter set, the preceding process node that caused the current brazing quality abnormality, and the abnormal parameters corresponding to the preceding process node, are identified, specifically including: If the brazing quality test result is lower than the preset quality threshold, based on the digital identifier of the heat exchanger under test, all process parameter sets recorded in the preceding process nodes of the heat exchanger under test are traversed and extracted from the database. The process parameter set includes at least the assembly pressure value in the core assembly process, the fin pitch deviation value in the automatic fin forming process, the riveting force curve characteristic parameters in the pipe riveting process, and the segmented temperature values ​​of the furnace temperature curve and the conveyor belt speed value in the brazing process. The process parameter set is input into a pre-trained inverse mapping model to output the abnormal contribution score of each preceding process node, and to identify the preceding process node whose abnormal contribution score exceeds a preset abnormal threshold, as well as the corresponding process parameter item as the abnormal source process and abnormal parameter.

7. The intelligent production method for heat exchangers according to claim 1, characterized in that, After sending the process correction data to the production equipment of the preceding process node to drive the preceding process node to adjust its operating parameters, the method further includes: After the process correction data is issued, the trend of the welding quality inspection results of the heat exchangers produced in the subsequent process nodes is continuously monitored. When the welding quality test results of the predetermined number of heat exchangers produced in subsequent continuous production all exceed the preset quality threshold, the current combination of process parameters is solidified into the process parameter knowledge base of the preceding process node, and used as the optimized process benchmark parameters of the preceding process node for the current heat exchanger model. If, in the subsequent continuous production of a predetermined number of heat exchangers, there are still abnormal heat exchangers whose welding quality test results are lower than the preset quality threshold, a multi-level alarm is triggered to package the abnormal heat exchanger's digital identifier, abnormal process information, and the input and output data of the current fault propagation inverse mapping model into an abnormal report. The anomaly report is sent to the administrator, and the production task of the production line where the preceding process node is located is suspended until a production resumption signal is received to resume production.

8. The intelligent production method for heat exchangers according to claim 1, characterized in that, A unique digital identifier is generated for each heat exchanger produced online, and this digital identifier is associated with and stored in relation to the process parameter set of the corresponding heat exchanger across multiple production stages. Specifically, this includes: Before the heat exchanger enters the first production station, a unique digital identifier is generated for the top plate surface of the heat exchanger by a writing device; As the heat exchanger passes through each production station in sequence, the process parameters of the heat exchanger at each production station are acquired in real time through data acquisition terminals installed at each production station. The production stations include core assembly station, fin installation station, top plate assembly station, bottom plate assembly station, press-fitting station, and vacuum brazing station. The process parameters include core assembly pressure value, fin positioning deviation value, top plate riveting force curve, and pressing displacement and pressing force value of the press-fitting fixture. Upon completion of each production process, the process parameters of the current process are merged and stored with the existing records in the associated database, using the digital identifier of the heat exchanger as an index, to form a historical archive of the entire process parameters of the heat exchanger from core assembly to brazing completion.

9. A smart production equipment for heat exchangers, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a heat exchanger intelligent production method as described in any one of claims 1-8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, When the computer-executable instructions are executed, a smart production method for heat exchangers as described in any one of claims 1-8 is implemented.

Images