A frame die casting process parameter optimization system

Abstract

This invention discloses a system for optimizing process parameters in vehicle frame die casting, relating to the field of vehicle manufacturing technology. The system divides high-speed images of the vehicle frame die casting process into multiple image segments. Edge features are extracted from each image segment to generate an edge bonding chain, and each edge bonding chain is assigned a sequence number. A lightweight model is used to monitor for anomalies in the image segments in real time. For anomalies, the edge bonding chains are used for separation and image segment duplication. After receiving all uploaded image segments, the system verifies and stitches the high-speed images using the sequence number and edge bonding chains. A defect identification mechanism is used to mark defect areas in the high-speed images, thereby determining the optimal die casting process parameters. This system solves the problems of difficulty in capturing local anomalies, untimely anomaly handling, and inaccurate defect identification caused by overall image analysis in traditional vehicle frame die casting process monitoring.

Description

Technical Field

[0001] This invention relates to the field of vehicle manufacturing technology, specifically to a system for optimizing process parameters in vehicle frame die casting. Background Technology

[0002] In the die-casting production of vehicle frames, monitoring and optimization of process parameters are necessary to ensure product quality. Traditional methods mainly rely on monitoring overall images or real-time acquisition of a limited number of key parameters, which is insufficient to comprehensively and meticulously reflect the complexities of the die-casting process. With the development of image processing and intelligent monitoring technologies, it has become possible to conduct more in-depth analysis of the die-casting process using high-speed images.

[0003] Traditional processes rely on preset injection speed, pressure, and mold temperature profiles. In actual production, these parameters are affected by equipment fluctuations and environmental disturbances, leading to molding defects such as porosity, shrinkage, warping, and cold shuts. Existing technologies mostly collect pressure and temperature signals through sensors and adjust process parameters using offline modeling or simple open-loop models. However, they lack intuitive capture of the real-time die-casting process and are difficult to adjust efficiently under complex operating conditions.

[0004] Therefore, in order to address the above problems, there is an urgent need for a vehicle frame die-casting process parameter optimization system. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a vehicle frame die casting process parameter optimization system, which solves the problems of difficulty in capturing local anomalies, untimely anomaly handling, inaccurate defect identification, and chaotic production data management caused by overall image analysis in traditional vehicle frame die casting process monitoring.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a vehicle frame die-casting process parameter optimization system, comprising: an image acquisition and segmentation module, used to receive high-speed images of the vehicle frame die-casting process, and to segment the high-speed images into multiple image segments according to preset rules, each image segment containing die-casting process parameters; an adhesive chain generation module, used to extract edge features from each image segment, generate edge adhesive chains for that image segment, assign a sequence number to each edge adhesive chain, and associate and store the sequence number, the die-casting process parameters of the corresponding image segment, and the edge adhesive chains; and transmission monitoring and anomaly detection. The processing module deploys a lightweight model at the edge bonding chain for each uploaded image segment. This lightweight model monitors the image segment for anomalies in real time, and uses the edge bonding chain to separate and copy the image segments for any detected anomalies. The image stitching and defect recognition module, after receiving all uploaded image segments, verifies and stitches the high-speed image using sequence symbols and the edge bonding chain. It uses the edge bonding chain to form a defect recognition mechanism, marks defective areas in the high-speed image, and determines the optimal die-casting process parameters through parameter simulation of the defective areas.

[0007] Furthermore, the preset rules specifically include: dividing the high-speed image into multiple adjacent image segments according to a spatial grid, and leaving an overlapping area for adjacent image segments, so that the edge of any image segment contains die-casting process parameters, and automatically associating the acquisition time and chassis position of each image segment after division.

[0008] Furthermore, the adhesive chain generation module generates a unique sequence symbol for the edge adhesive chain of each image segment. The sequence symbol includes the image segment number, the high-speed image sequence number, and the position index. The image segment number indicates the sequential position of the image segment in the high-speed image, the high-speed image sequence number indicates the temporal order of the high-speed image in the entire frame die-casting process, and the position index is the physical coordinate of the image segment in the frame model, which is used to indicate the geometric position of the image segment in the mold.

[0009] Furthermore, the edge bonding chain is composed of an edge feature point sequence and connection relationships, and sequentially stores the edge coordinates, curvature information, corresponding die-casting process parameters, and check codes of the image segment. The edge bonding chain structure is implemented based on a bidirectional join table.

[0010] Furthermore, the lightweight model is a lightweight neural network based on distance metric, used to calculate the transmission delay and retrieval check code status between the current image slice and adjacent image slices in real time. If the transmission delay exceeds the threshold and the check code does not change, it is determined to be an image slice delay. If the check code changes, it is determined to be an image slice intrusion.

[0011] Further, the specific analysis of the separation and image segment copying processing using the edge bonding chain is as follows: For image segment delays, the position where the image segment determined by the lightweight model to be delayed should be stitched is located based on the sequence number. The edge bonding chain of the current image segment is instructed to separate from the image segment and proceed to the located position to merge with its adjacent image segments in advance. Using the edge coordinates, curvature information, and corresponding die-casting process parameters of the image segments stored in the edge bonding chain, the image completion model built into the edge bonding chain is called to generate a new copied segment at the merging point. During the generation of the copied segment, texture synthesis is performed by combining the edge shapes of adjacent image segments, and missing die-casting process parameters are filled in. For image segment intrusions, a new copied segment is generated according to the image segment delay processing method. While the edge bonding chain of the current image segment is separated from the image segment, the internal resources of the separated image segment are used for self-destruction. When a new copied segment is generated, the weights of the lightweight model are reinitialized, and the sequence numbers of all edge bonding chains are updated synchronously.

[0012] Furthermore, the specific analysis of the defect recognition mechanism formed by utilizing edge bonding chains is as follows: After all image segments are stitched together by verifying the sequence number and edge bonding chains, the computing resources occupied by each segment during the generation and monitoring of edge bonding chains are integrated into a defect recognition mechanism according to a resource reorganization strategy. The integration method of the defect recognition mechanism is as follows: the convolution kernel of the lightweight model and the edge features stored in the edge bonding chains are used as pre-trained features, and the defect recognition mechanism is constructed by combining the centrally stored die-casting process image database. The defect recognition mechanism has multi-scale convolutional layers and attention modules, which are used to extract features from the macro framework to the micro texture level for the stitched high-speed image step by step, and identify the defect category caused by process deviation.

[0013] Furthermore, the specific analysis of the parameter simulation of the defect area is as follows: After the defect area is marked using the defect identification mechanism, based on the defect type and its spatial distribution in the high-speed image, the process parameter simulation mechanism is invoked to explore multi-dimensional parameters. The process parameter simulation mechanism is used to establish a digital twin model of the chassis die-casting process, mapping the spatiotemporal position of the image slice where the defect area is located to physical coordinates in the chassis die-casting process, and determining the die-casting process parameters when the defect occurs by combining the sequence number. By monitoring the effect of simulating different die-casting process parameters, the optimal die-casting process parameters are selected. The optimal die-casting process parameters are compared with the current die-casting process parameters, and adjustment and optimization suggestions are generated and fed back to the die-casting equipment control unit.

[0014] The present invention has the following beneficial effects:

[0015] This vehicle frame die-casting process parameter optimization system splits high-speed images into multiple image segments and extracts edge features from each segment to generate an edge bonding chain. This allows for meticulous monitoring of local details during the die-casting process, facilitating more accurate monitoring of process parameters and the detection of potential problems. Compared to overall image monitoring, it improves the ability to capture subtle anomalies. A lightweight model is deployed at the edge bonding chain for each uploaded image segment, enabling real-time monitoring of anomalies. Once an anomaly is detected, the edge bonding chain can be used for rapid separation and image segment copying, addressing the anomaly promptly and preventing further escalation and impact on the die-casting process. This improves casting quality and enhances the stability and reliability of the production process. After stitching together high-speed images, an edge-bonding chain is used to form a defect identification mechanism, accurately marking defect areas. By simulating the parameters of these defect areas, optimal die-casting process parameters can be determined more effectively, helping to improve the quality and performance of chassis die-casting, reduce scrap rates, and lower production costs. Linking and storing the sequence number, corresponding image segments, and die-casting process parameters with the edge-bonding chain ensures clear logical relationships and traceability throughout the die-casting process, facilitating subsequent analysis, querying, and management of production data, and providing strong data support for process improvement and production optimization.

[0016] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0017] Figure 1 This is a structural diagram of a vehicle frame die-casting process parameter optimization system according to the present invention.

[0018] Figure 2 This is a flowchart of a method for optimizing process parameters of a vehicle frame die casting system according to the present invention.

[0019] Figure 3 This is a schematic diagram of the logic flow of a vehicle frame die casting process parameter optimization system according to the present invention. Detailed Implementation

[0020] This application embodiment utilizes a chassis die-casting process parameter optimization system to achieve precise image segmentation monitoring, real-time anomaly separation processing, accurate defect identification based on edge bonding chains, and parameter simulation optimization, effectively improving chassis die-casting quality and production stability.

[0021] The overall concept of this application's embodiments is as follows:

[0022] High-speed industrial cameras are used to capture video of the die-casting process. An edge computing gateway divides each frame into multiple image segments according to a preset grid, and features are extracted from the edges of these segments to generate an adhesive chain. This adhesive chain carries process parameters and sequence symbols, used to detect missing or tampered segments during upload. In case of an anomaly, the saved edge information is used to generate a duplicate segment and pre-merge it with adjacent segments to maintain transmission integrity. After all segments are uploaded, the images are stitched together based on the sequence symbols and the edge adhesive chains. The resources used by the adhesive chains are then reorganized into a defect detection mechanism to perform defect detection on the stitched image. Optimal die-casting parameters are determined through process parameter simulation and fed back to the die-casting machine actuator, achieving closed-loop control.

[0023] Please see Figure 1 , Figure 2 , Figure 3 This invention provides a technical solution: a vehicle frame die-casting process parameter optimization system, comprising: an image acquisition and segmentation module, used to receive high-speed images of the vehicle frame die-casting process, and to segment the high-speed images into multiple image segments according to preset rules, each image segment containing die-casting process parameters; an adhesive chain generation module, used to extract edge features from each image segment, generate an edge adhesive chain for that image segment, assign a sequence number to each edge adhesive chain, and associate and store the sequence number, the die-casting process parameters of the corresponding image segment, and the edge adhesive chain; a transmission monitoring and anomaly handling module, used to deploy a lightweight model at the edge adhesive chain for each uploaded image segment, use the lightweight model to monitor in real time whether there are any anomalies in the image segment, and use the edge adhesive chain to separate and copy the image segments for the detected anomalies; and an image stitching and defect recognition module, used to verify and stitch the high-speed images by using the sequence number and the edge adhesive chain after receiving all uploaded image segments, use the edge adhesive chain to form a defect recognition mechanism, use the defect recognition mechanism to mark the defect areas in the high-speed images, and determine the optimal die-casting process parameters by simulating the parameters of the defect areas.

[0024] Specifically, die-casting process parameters refer to the key process variables that directly affect the filling, solidification, and casting quality of molten metal during the die-casting production of vehicle frames. These parameters are recorded in real time and embedded into high-speed image slices, serving as the core basis for defect analysis and parameter optimization. The embedding methods include, but are not limited to: adding structured data fields containing process parameters to the file header or metadata area of ​​the image slice; or encoding the process parameters as invisible watermarks and embedding them in the frequency domain of the sliced ​​image. Die-casting process parameters specifically include the following parameters: injection-related parameters: injection speed, injection pressure; temperature-related parameters: mold temperature, pouring temperature; vacuum and venting parameters: vacuum degree, venting time; time-related parameters: holding time, cooling time, mold opening time; auxiliary process parameters: release agent parameters, pressure chamber filling degree.

[0025] Die-casting process parameters are collected in real time by sensors and associated with corresponding areas of high-speed images. For example, a certain injection speed corresponds to the image of molten metal filling a certain part of the frame. The image is then embedded into segments by the image acquisition and segmentation module, and then bound to the edge features of the segments by an adhesive chain. Finally, during defect identification, the mapping between the defect area and the corresponding process parameters is realized, providing specific adjustment objects for parameter simulation and optimization.

[0026] Specifically, the association is established as follows: for each image acquisition moment or frame, the timestamp is synchronously bound to all die-casting process parameter data acquired by the sensor at that moment. During image segmentation, based on the timestamp and a preset coordinate mapping table, the process parameter data for the same moment is embedded into the image slice corresponding to the physical coordinate region. For example, the injection velocity value at a certain moment will be associated and stored with the slice in the image corresponding to the molten metal front position at that moment.

[0027] The image acquisition and segmentation module is deployed on a high-speed industrial camera and an edge computing gateway. The high-speed industrial camera is fixed in the die-casting machine's working area and connected to the gateway via a high-speed interface to acquire high-speed images of the die-casting process in real time. Due to the rapid changes in the die-casting process, this module employs high frame rate exposure and high-speed transmission technology to ensure the capture of the dynamics of the cast liquid filling and mold cavity filling.

[0028] In this implementation scheme, the acquired high-speed images are segmented on the edge computing gateway according to preset rules. These preset rules divide the high-speed images into multiple adjacent image slices based on a spatial grid. The grid size is adaptively adjusted based on the die-casting equipment's field of view, the spatial scale of process parameter changes, and camera resolution. The minimum acceptable slice size is determined by calculating the movement speed and temperature gradient of the process area. The adaptive grid size adjustment algorithm is based on the following rules: First, an upper limit for the number of image slices is determined based on the die-casting equipment's field of view. Second, a lower limit for the physical size of each slice is determined based on the camera resolution and the spatial scale of process parameter changes to ensure that key process changes can be captured. Finally, the slice size is dynamically adjusted based on the calculated average movement speed of the process area; the faster the movement speed, the larger the slice size, but it must not be less than the aforementioned lower limit of physical size. To ensure smooth splicing of adjacent slices, an overlapping area is set at the grid boundary, and the overlap width is determined based on the camera resolution and process parameter accuracy requirements.

[0029] Each image segment is automatically associated with its acquisition timestamp, physical coordinates in the chassis mold coordinate system, and die-casting process parameters after splitting. These process parameters are derived from injection speed, pressure curves, mold temperature, and vacuum level acquired by the PLC. By embedding these parameters into the segments, it ensures that even if any segment is lost, the complete process state can still be inferred from the remaining segments.

[0030] The image acquisition and segmentation module enables precise segmentation of the high-speed die-casting process, binding each segment to process parameters. This not only provides sufficient segmentation redundancy to ensure transmission reliability, but also lays the foundation for subsequent edge bonding chain generation and parameter reconstruction.

[0031] Specifically, the adhesive chain generation module is deployed on the acceleration chip of the edge computing gateway. It is used to extract edge features from each image segment and generate edge adhesive chains for the segment. Employing gradient- and texture-based edge detection algorithms, combined with the flow characteristics of molten metal during die casting, the pixel gradients at the segment edges are analyzed to extract edge feature point sequences.

[0032] In this implementation scheme, during edge feature extraction, edge curvature information and corresponding die-casting process parameters are recorded to construct an edge feature vector. Sequence symbols are then assigned to each edge bonding chain according to a preset sequence rule. The sequence symbol is a unique identifier composed of an image segment number, a high-speed image sequence number, and physical coordinates within the chassis model. Specifically, the image segment number indicates the relative sequential position of the segment in the high-speed image; the high-speed image sequence number indicates the temporal order of the high-speed images in the video; and the physical coordinates indicate the geometric position of the segment within the mold.

[0033] The edge bonding chain structure is implemented using a bidirectional join table, with nodes storing edge feature point sequences and die-casting process parameters. Both ends of the edge bonding chain are connected to adjacent segments via pointers, forming a ring topology. Each chain also generates a checksum, calculated from the edge features and process parameters using a hash function, used to detect whether the segment has been tampered with or damaged during transmission.

[0034] The edge bonding chain not only carries edge information for subsequent stitching, but also stores the process parameters, texture information, and texture weights of the corresponding image segments. When an image segment is lost, the information in the edge bonding chain is used to generate a new duplicate segment in the overlapping area of ​​adjacent image segments, ensuring that each image segment contains complete process parameters.

[0035] By using an edge bonding chain, each image segment is tightly bound to edge features and process parameters. The design of the sequence symbol not only marks the spatial and temporal position of the segment, but also ensures that the system can quickly locate and reconstruct delayed segments, improving data integrity and security, and providing a reliable basis for subsequent splicing and defect identification.

[0036] Specifically, the transmission monitoring and anomaly handling module is deployed in the real-time operating system of the edge computing gateway. A lightweight model is deployed at the edge glue chain of each uploaded image segment. This lightweight model runs in an independent thread to monitor the transmission status of each segment. The lightweight model uses deep network compression and pruning techniques to reduce the number of parameters while maintaining recognition performance, enabling it to run efficiently on resource-constrained gateway devices.

[0037] The lightweight model's inputs include the transmission delay between the current fragment and its adjacent fragments, checksum changes, and sequencer differences. By calculating the transmission distance between fragments (i.e., the transmission time difference and network hop count) and comparing the checksums, it determines in real-time whether fragments exhibit abnormal conditions. When the transmission delay exceeds a set threshold and the checksum remains unchanged, the fragment is considered to have exceeded the expected delay and can be considered lost. When the checksum changes, the fragment is considered to have been compromised or a transmission error has occurred. For accurate differentiation, the judgment is based on the transmission delay: if the transmission delay is greater than the threshold and the checksum changes, the fragment is considered highly likely to have been compromised; if the transmission delay does not exceed the threshold but the checksum changes, it is primarily considered a transmission error or data corruption.

[0038] Once an anomaly is detected, the adhesive chain is used to perform piece separation and replication. For a missing piece, the lightweight model instructs the chain to separate from that piece and locates the position where it should be spliced ​​using a sequence number. The chain then proceeds to the located adjacent piece and merges it in advance. During the merging process, the lightweight model calls the image completion model inside the chain, using the edge shapes of the adjacent pieces as constraints, and utilizes the saved edge features, curvature information, and process parameters to generate a new replicated piece.

[0039] For fragments detected as compromised, the processing method is similar, except that when the chain is separated from the fragment, some resources are reserved for self-destruction and obfuscation to prevent the leakage of sensitive information. After separation and replication are completed, a new glue chain and lightweight model are generated and synchronized with the sequence number of other fragments.

[0040] The lightweight model architecture includes an input layer, several quantized convolutional layers, and an attention mechanism layer to extract transmission state features. Pruning and quantization reduce the model size. Its output, after passing through an activation function, provides anomaly classification results. The lightweight model interacts with the edge gateway's transmission protocol stack, obtaining transmission statistics through a socket interface and sending anomaly events to the upper layers to trigger chain processing logic.

[0041] Image completion models are generative models that reconstruct missing regions based on surrounding content. Essentially, they utilize the texture, color, and structural information of existing regions to infer pixels in occluded or damaged areas using deep neural networks. Specifically, image completion models can be viewed as a conditional generation task: the input is an image with missing or occluded parts, and the output is a complete image with the missing parts filled in. During the training phase, image completion models learn the statistical properties and structural patterns of natural images, enabling them to generate details consistent with the original image during the inference phase.

[0042] Image completion models are constructed using an encoder-decoder structure or a generative adversarial network (GAN). Specifically, convolutional layers encode the image into multi-scale features, capturing contextual information around the missing regions. The decoder then decodes these features into pixel values ​​to reconstruct the missing regions. To better handle large-area missing parts, the model often integrates self-attention mechanisms or residual blocks, allowing the network to focus on key textures and edge cues globally. During training, several regions of the complete image are randomly occluded to construct training samples. The model is then optimized using pixel reconstruction loss, perceptual loss, and adversarial loss, enabling it to learn to generate natural and coherent content in the missing regions.

[0043] An image completion model is embedded within the bonding chain to generate replacement fragments for lost or compromised image pieces. Its main functions include: performing texture synthesis and edge extension on missing areas based on the edge features and curvature information of adjacent fragments, ensuring a visually seamless stitched image. Utilizing die-casting process parameters stored in the chain, these parameters are filled into the newly generated fragments, ensuring that the generated fragments not only visually match the original... Figure 1 Furthermore, it contains complete process data. In the event of an anomaly during transmission, missing segments are quickly copied to maintain the integrity of the data chain, improve system robustness, and prevent the loss of a single segment from affecting the restoration of the entire process parameters.

[0044] By deploying a lightweight model on edge devices, real-time transmission monitoring and anomaly detection are achieved. This not only solves the problem of deploying traditional deep learning models in embedded environments but also ensures the consistency and security of image fragments during transmission. Fragments can be immediately copied and recovered after loss or intrusion, improving the system's robustness.

[0045] Specifically, once all image segments are uploaded to the central server via the edge computing gateway, the image stitching and defect identification module begins operation. This module is deployed on a combined cloud server and edge gateway platform. It verifies the integrity of the uploaded segments based on the sequence number and the edge bonding chain. Utilizing edge features and curvature information within the chain, it seamlessly stitches the segments into a complete high-speed image sequence according to the spatial and temporal order determined by the sequence number. During the stitching process, overlapping areas are fused, and gaps and visual discontinuities are eliminated through weighted averaging and texture matching.

[0046] In this implementation scheme, after image stitching is completed, the computational resources occupied by each segment during the generation and monitoring of the bonding chain are integrated to form a defect recognition mechanism. The resource integration strategy includes using the convolutional kernels and attention weights in the lightweight model as pre-trained features, and training a deep defect recognition network together with the edge features preserved by the edge bonding chain. This network has multi-scale convolutional layers, spatial attention, and channel attention modules, and can extract features at multiple levels from macroscopic contours to microscopic textures, identifying various defect categories such as incomplete filling and porosity.

[0047] The defect identification mechanism maintains a communication link with a centrally stored database of die-casting process images. It utilizes a large number of annotated reference process images from the database to construct a comparative learning task, thereby improving identification accuracy. By comparing the stitched image with the reference images, it identifies the spatial location of the defect area, the defect type, and the time of defect formation.

[0048] After identifying the defective area, the sequence symbols of the defective segments are mapped to process parameters, and the process parameter simulation mechanism is invoked for parameter optimization. If necessary, the edge computing gateway can be called back to adjust the compensation strategy in real time, and the optimal parameters are fed back to the die-casting machine PLC control logic to form a closed-loop control.

[0049] By utilizing the computational resources recombined from the adhesive chain to construct a defect identification mechanism, additional hardware consumption is avoided; multi-scale convolution and attention mechanisms improve the accuracy and robustness of defect detection; at the same time, linkage with the process database enables high-precision comparison, forming a closed loop between defect identification and parameter optimization, thereby improving production quality.

[0050] Specifically, the process parameter simulation mechanism is deployed on a collaborative computing platform between a cloud server and the die-casting machine controller. It is used to simulate and analyze die-casting process parameters based on defect identification results. The process parameter simulation mechanism constructs a digital twin model of the chassis die-casting process, mapping the spatial location and timestamp of the defect area in the image to specific parts and process stages in the physical casting.

[0051] Based on the mapping results, key process variables such as injection speed, pressure gradient, mold temperature, vacuum degree, and alloy composition are selected. A parameter-quality response relationship is established in the physical model. By calling the simulation module, different parameter combinations are traversed to predict their impact on defect indicators. Sensitivity analysis is used to screen out the process parameters that have the greatest impact on defects. Defect indicators include porosity, surface roughness, and dimensional deviation.

[0052] Based on the comprehensive simulation results, an optimal set of die-casting process parameters is selected from the multi-dimensional parameter space. This parameter combination minimizes defect indicators and meets production cycle time. The optimal parameters are compared with the current production parameters to generate adjustment suggestions, which are then sent to the die-casting machine's PLC control unit via an edge gateway. The PLC controls actuators such as servo valves, hydraulic systems, heaters, and vacuum pumps to regulate injection speed, pressure, and temperature, thereby optimizing the die-casting process in real time. Pressure and temperature data collected by sensors are fed back to the simulation model, forming a closed-loop update.

[0053] By utilizing digital twin models and sensitivity analysis, key process parameters leading to defects can be quickly located and optimized solutions can be generated. The closed loop formed with PLC and sensors enables adjustments to be implemented in real time, achieving precise control, improving product quality, and reducing scrap rates.

[0054] Specifically, in the transmission monitoring module, abnormal situations are divided into two categories: fragment loss and fragment intrusion.

[0055] Fragment loss detection: The lightweight model continuously calculates the transmission time difference and network hop count of adjacent fragments. If the transmission delay of a fragment exceeds a threshold and its corresponding checksum remains unchanged in the network, the fragment is considered to have been unexpectedly lost during transmission. Fragment loss may be caused by network congestion, buffer overflow, or link failure.

[0056] Fragment intrusion detection: If the transmission delay is greater than the threshold and the fragment checksum changes, i.e., it is inconsistent with the checksum calculated at the time of generation, then the fragment is considered to have been tampered with or intruded upon during transmission.

[0057] For missing segments, the replication process is initiated immediately: the edge bonding chain detaches from the segment and locates the target splicing position using a sequence number. The edge bonding chain moves to the located position and merges with the adjacent chain there in advance; using the edge coordinates, curvature information, and process parameters stored in the edge bonding chain, a new segment is reconstructed using a built-in image completion model. The completion model is based on a convolutional neural network, combining the texture features of adjacent segments to generate the missing parts and supplementing the missing process parameters. After replication, a new bonding chain and a lightweight model are generated to replace the missing segment.

[0058] For cases where a segment is compromised, the missing segment is recovered through a copying process. After the chain is separated from the compromised segment, some resources are used to perform blurring, which obscures or scrambles the pixels of the compromised segment to protect process secrets, and then the segment is destroyed. The new copied segment is uploaded along with other segments to ensure image integrity.

[0059] After the fragmentation anomaly is handled, all glue chains are reordered according to the new sequencer and synchronized to other edge nodes via control messages to ensure consistency in subsequent splicing.

[0060] By accurately distinguishing between different anomalies such as data loss and intrusion and adopting targeted processing strategies, both data integrity and transmission continuity are ensured, while security is strengthened. The sharded replication model improves fault tolerance, enabling the complex die-casting process to operate stably even under network fluctuations.

[0061] Specifically, the defect identification mechanism is an intelligent analysis module established after image stitching is completed by integrating the computational resources occupied by the edge bonding chain. The convolutional kernels and attention weights of the lightweight model deployed during edge bonding chain generation and transmission monitoring, the edge features stored in the chain, and the intermediate features of the temporarily used image completion model are unified into a multi-task training library.

[0062] When establishing the defect recognition mechanism, these pre-trained features are normalized and weighted to form initial model parameters. Reference samples from the die-casting process image database are then used for transfer learning of the recognition network. The recognition network employs a pyramid-shaped multi-scale convolutional structure, gradually increasing the receptive field from top to bottom. A spatial attention module focuses on the defect region, while a channel attention module filters out feature channels related to the defect, improving the model's recognition sensitivity.

[0063] The defect identification mechanism extracts local features from the stitched image using a sliding window, performs similarity matching with reference image features, and marks the specific location and category of the defect region. Simultaneously, it calculates the size, shape, and duration of the defect region over time, providing necessary information for subsequent parameter simulation. The identified defect regions are linked to process parameters using a sequence mapping, thus forming a three-dimensional correlation index of time, space, and parameters.

[0064] The completed defect identification mechanism serves as a cloud service, interacting with edge gateways, PLCs, and databases through standardized interfaces. The mechanism achieves efficient construction through resource integration and transfer learning, leveraging an attention mechanism to enhance its ability to focus on defect features. Training strategies improve the system's identification performance for new defect types, significantly enhancing the quality control level of the die-casting process.

[0065] Specifically, the parameter simulation method for defect regions is based on the aforementioned digital twin model and sensitivity analysis. For each marked defect region, the process parameters at the time of its generation are retrieved using sequence symbols to construct a parameter vector. Subsequently, partial differential equations are established in the physical simulation model to describe the molten metal flow, heat transfer, and solidification processes. Through finite element or finite volume discretization, the influence of different parameter combinations on defect indices is obtained.

[0066] A global optimization algorithm is employed to search for the optimal solution in a multi-dimensional parameter space, comprehensively considering the influence of variables such as injection velocity curve, pressure gradient, alloy temperature, and mold surface temperature on defect formation. Simultaneously, data collected in real-time by sensors on the die-casting machine is incorporated to perform online corrections to the simulation model, improving prediction accuracy.

[0067] To avoid the time-consuming large-scale computations, a simulation model is combined with a machine learning proxy model. The proxy model is used to quickly evaluate the merits of parameter combinations, and the system dynamically switches between numerical simulation and the proxy model. The final set of optimal process parameters is sent to the die-casting machine's PLC via a control protocol. The PLC then controls the servo valves to adjust the injection speed, the heaters to adjust the mold temperature, and the vacuum pumps to adjust the vacuum level inside the mold cavity, thereby achieving targeted optimization of defective areas.

[0068] By employing a parameter simulation method that combines digital twin simulation with surrogate models and global optimization algorithms, a near-optimal combination of process parameters can be found in a short time. By combining real-time sensor data to improve model accuracy, the defect rate can be effectively reduced, thereby improving the consistency and reliability of die-cast products.

[0069] In summary, this application has at least the following effects:

[0070] Image segmentation ensures each piece contains process parameters, avoiding parameter fragmentation, reducing resampling costs, and adapting to continuous production. The bonding chain combines edge features and sequence symbols to improve splicing accuracy and enable rapid traceability of defects and parameters. Lightweight models deployed at the edges monitor transmission anomalies in real time, generating duplicate fragments when data is lost and isolating and destroying them in case of intrusion, ensuring continuous data security. Defect identification reuses bonding chain resources, accurately associating abnormal parameters and outputting the optimal solution through simulation, eliminating the need for trial and error. The entire system forms a closed loop, adapting to the high safety requirements of vehicle frames, improving yield rates, and reducing costs and the risk of production interruptions.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] This invention is described with reference to a flowchart of a method according to embodiments of the invention. It should be understood that the combination of each step in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A device for a function specified in one or more processes.

[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 The function specified in one or more processes.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0075] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0076] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A system for optimizing process parameters in vehicle frame die casting, characterized in that, include: The image acquisition and segmentation module is used to receive high-speed images of the chassis die-casting process and split the high-speed images into multiple image segments according to preset rules. Each image segment contains die-casting process parameters. The adhesive chain generation module is used to extract edge features for each image segment, generate an edge adhesive chain for that image segment, assign a sequence number to each edge adhesive chain, and associate and store the sequence number, the die-casting process parameters of the corresponding image segment with the edge adhesive chain. The edge adhesive chain is composed of an edge feature point sequence and connection relationship, and stores the edge coordinates, curvature information, corresponding die-casting process parameters and check code of the image segment in sequence. The edge adhesive chain structure is implemented based on a bidirectional join table. The transmission monitoring and anomaly handling module is used to deploy a lightweight model at the edge bonding chain for each uploaded image segment. The lightweight model is used to monitor whether there are any anomalies in the image segment in real time. For the detected anomalies, the edge bonding chain is used to separate and copy the image segments. The image stitching and defect recognition module is used to verify and stitch high-speed images by using sequence symbols and edge bonding chains after receiving all uploaded image segments. It uses edge bonding chains to form a defect recognition mechanism, marks defect areas in high-speed images, and determines the optimal die-casting process parameters by simulating the parameters of the defect areas. The specific analysis of the defect recognition mechanism formed by the edge bonding chain is as follows: After all image segments are stitched together by verifying the sequence number and the edge bonding chain, the computing resources occupied by each segment in the edge bonding chain generation and monitoring process are integrated into a defect recognition mechanism according to the resource reorganization strategy. The integration method of the defect recognition mechanism is as follows: the convolution kernel of the lightweight model and the edge features stored in the edge bonding chain are used as pre-trained features, and the defect recognition mechanism is constructed by combining the centrally stored die casting process image database. The defect recognition mechanism has multi-scale convolutional layers and attention modules, which are used to extract features from the macro framework to the micro texture level for the stitched high-speed image step by step, and identify the defect category caused by process deviation. The specific analysis of parameter simulation for the defect area is as follows: After the defect area is marked using the defect identification mechanism, the process parameter simulation mechanism is invoked to explore multi-dimensional parameters based on the defect type and its spatial distribution in the high-speed image. The process parameter simulation mechanism is used to establish a digital twin model of the chassis die-casting process, mapping the spatiotemporal position of the image slice where the defect area is located to physical coordinates in the chassis die-casting process, and determining the die-casting process parameters when the defect occurs by combining the sequence number. The optimal die-casting process parameters are selected by monitoring the effect of simulating different die-casting process parameters, and the optimal die-casting process parameters are compared with the current die-casting process parameters to generate adjustment and optimization suggestions and feed them back to the die-casting equipment control unit.

2. The vehicle frame die-casting process parameter optimization system according to claim 1, characterized in that, The preset rules specifically include: dividing the high-speed image into multiple adjacent image segments according to the spatial grid, and leaving an overlapping area for adjacent image segments, so that the edge of any image segment contains the die-casting process parameters, and automatically associating the acquisition time and chassis position of each image segment after division.

3. The vehicle frame die-casting process parameter optimization system according to claim 1, characterized in that, The adhesive chain generation module generates a unique sequence symbol for the edge adhesive chain of each image segment. The sequence symbol includes the image segment number, the high-speed image sequence number, and the position index. The image segment number indicates the sequential position of the image segment in the high-speed image, the high-speed image sequence number indicates the temporal order of the high-speed image in the entire frame die casting process, and the position index is the physical coordinate of the image segment in the frame model, which is used to indicate the geometric position of the image segment in the mold.

4. The vehicle frame die-casting process parameter optimization system according to claim 1, characterized in that, The lightweight model is a lightweight neural network based on distance metric, used to calculate the transmission delay and retrieval checksum status between the current image slice and its neighboring image slices in real time. If the transmission delay exceeds the threshold and the checksum does not change, it is determined to be an image slice delay. If the checksum changes, it is determined that the image slice has been compromised.

5. The vehicle frame die-casting process parameter optimization system according to claim 4, characterized in that, The specific analysis of the separation and image segmentation copying process using edge adhesive chains is as follows: To address image segmentation delay, the lightweight model locates the position where image segments identified as having delays should be stitched together based on the sequence number. It then instructs the edge bonding chain of the current image segment to separate from the image segment and move to the located position to merge with its adjacent image segments in advance. Using the edge coordinates, curvature information, and corresponding die-casting process parameters of the image segments stored in the edge bonding chain, the built-in image completion model of the edge bonding chain is called to generate a new copy segment at the merging point. During the generation of the copy segment, texture synthesis is performed by combining the edge shapes of adjacent image segments, and missing die-casting process parameters are filled in. In response to the intrusion of image segments, a new copy segment is generated according to the image segment delay processing method. While the edge glue chain of the current image segment is separated from the image segment, the internal resources of the separated image segment are used to perform self-destruction processing. When a new copy fragment is generated, the weights of the lightweight model are reinitialized, and the sequence signs of all edge glue chains are updated synchronously.

Images