Automobile armrest plate polymer cladding piece interface peeling early warning method and system

Abstract

The application discloses a kind of automobile handrail plate high polymer covering piece interface peeling early warning method and system, belong to fault early warning technical field, comprising the following steps: S1, the unique identification information of workpiece is acquired, workpiece is carried out multi-source nondestructive testing, and detection data is bound with unique identification information;S2, based on the physical detection area division rule established in advance, the workpiece is mapped to at least two independent physical detection areas, and the workpiece level feature vector containing area attribute is constructed;S3, call the lightweight regression mapping model matched with the current workpiece material, infer the workpiece level feature vector, and output the interface equivalent adhesive strength prediction value of each physical detection area;S4, compare interface equivalent adhesive strength prediction value with preset safety threshold, generate grading early warning instruction according to comparison result and feedback to production line control system, improve the accuracy of automobile handrail plate high polymer covering piece interface peeling early warning.

Description

Technical Field

[0001] This invention relates to the field of fault warning technology, and in particular to a method and system for early warning of interface peeling of polymer-coated parts for automotive armrests. Background Technology

[0002] Automotive interior parts are widely used in areas such as doors, dashboards, and armrests. Armrest panels, in particular, typically employ a composite structure consisting of a base material and a skin-like outer layer. This type of structure generally uses plastic or composite materials as the base, with a polymer skin layer (such as polyvinyl chloride, polyurethane, or thermoplastic polyolefin) formed on its surface through molding, hot pressing, or adhesive bonding processes. This enhances the product's appearance, feel, and wear resistance. During the coating process, the quality of the interfacial bonding is affected by various process parameters, including temperature control, pressure distribution, adhesive layer thickness, mold fitting accuracy, and material shrinkage characteristics. Poor interfacial bonding can easily lead to peeling and failure of the coating layer.

[0003] In existing production processes and quality inspections, the evaluation of interface quality for polymer-coated parts mainly focuses on the finished product stage. Common methods include manual visual inspection, machine vision surface defect identification, and sampling offline destructive peel testing, such as tensile peel tests or 180-degree peel strength tests. Some production lines use online vision systems or basic ultrasonic testing to identify and screen for geometric defects such as surface bubbles, wrinkles, and obvious edge warping, and combine this with process parameter records for process traceability analysis.

[0004] However, existing technologies for interface quality inspection of polymer-coated components such as automotive armrests cannot achieve accurate early warning of non-destructive interface delamination. Specifically, current online inspection methods mainly rely on the principles of light reflection or sound wave reflection, which can only identify geometric defects such as bubbles, wrinkles, or macroscopic debonding. In reality, however, the failure of polymer coatings often begins with the decay of interfacial bonding energy, such as a decrease in adhesive strength due to adhesive aging or temperature fluctuations. At this point, the interface may still be geometrically intact, and the non-destructive testing signal may not show significant changes, but the actual mechanical strength can no longer meet safety requirements. Existing technologies struggle to establish an effective quantitative correspondence between the acquired multi-dimensional non-destructive testing signals and the actual mechanical decay process of the interface, resulting in the inability to obtain continuous real-time interface strength indicators and thus failing to issue effective early warnings before macroscopic delamination of the coating occurs.

[0005] Furthermore, existing technologies rely on real mechanical data obtained through offline destructive testing as labels, but this faces several insurmountable bottlenecks in practical engineering applications. First, the inability to achieve full inspection results in extremely small sample sizes. Fast production line cycles make it impossible to perform destructive testing on every finished product to obtain true strength. The very limited amount of true data obtained from offline sampling is insufficient to meet the training needs of complex deep neural networks or large models, easily leading to model overfitting or inaccurate predictions. Second, the representativeness of the test pieces is problematic. To avoid damaging the finished product, small test pieces are usually formed simultaneously at the mold edge. However, the heating history, stress conditions, and glue flow of these test pieces are not entirely consistent with the actual state of the complex curved surface area of ​​the handrail. Directly applying the strength of the test pieces or attempting to extrapolate the strength of the finished product using complex three-dimensional spatial mapping algorithms results in excessive errors, rendering the technology meaningless. Third, data alignment is extremely cumbersome. Even if destructive testing is performed on the finished product, the overall destructive force is measured. However, non-destructive testing probes only scan a few local areas. Accurately aligning the local signals with the overall destructive force requires the construction of a complex three-dimensional digital model, which is extremely difficult to implement. A large amount of computing power and time is consumed in the alignment process, making it difficult to meet the real-time and reliability requirements of quality control in the automated production of automotive interiors. This results in low accuracy of the interface peeling warning for polymer-coated automotive armrest panels. Summary of the Invention

[0006] To address the low accuracy of existing technologies for warning of interface peeling in polymer-coated automotive armrest components, this invention provides a method and system for warning of interface peeling in polymer-coated automotive armrest components. The technical solution is as follows:

[0007] On the one hand, a method for early warning of interface peeling of polymer-coated automotive armrest parts is provided. The method includes: S1, obtaining the unique identification information of the workpiece, performing multi-source non-destructive testing on the workpiece, and binding the test data with the unique identification information; S2, based on the pre-established physical inspection area division rules, mapping the workpiece into at least two independent physical inspection areas, and constructing a workpiece-level feature vector containing area attributes; S3, calling a lightweight regression mapping model that matches the current workpiece material, inferring the workpiece-level feature vector, and outputting the predicted value of the interface equivalent bonding strength of each physical inspection area; S4, comparing the predicted value of the interface equivalent bonding strength with a preset safety threshold, generating a graded early warning instruction based on the comparison result, and feeding it back to the production line control system.

[0008] On the other hand, a system for early warning of interface peeling of polymer-coated automotive armrest panels is provided, comprising: a data acquisition and binding module, a region mapping and feature construction module, an interface strength prediction module, and a graded early warning and feedback module. The data acquisition and binding module is used to acquire the unique identification information of the workpiece, perform multi-source non-destructive testing on the workpiece, and bind the test data with the unique identification information. The region mapping and feature construction module is used to map the workpiece into at least two independent physical testing regions based on pre-established physical testing region division rules, and construct a workpiece-level feature vector containing region attributes. The interface strength prediction module is used to call a lightweight regression mapping model matching the current workpiece material, infer the workpiece-level feature vector, and output the predicted value of the interface equivalent bonding strength of each physical testing region. The graded early warning and feedback module is used to compare the predicted value of the interface equivalent bonding strength with a preset safety threshold, generate graded early warning instructions based on the comparison results, and feed them back to the production line control system.

[0009] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0010] 1. This invention provides a method for early warning of interface peeling in polymer-coated automotive armrest panels. By acquiring the unique identification information of the workpiece and performing multi-source non-destructive testing, the test data is bound to the unique identification information, achieving accurate traceability and full lifecycle tracking of the test data to specific workpieces. Compared with the problems of isolated test data and difficulty in corresponding with specific products in existing technologies, which leads to difficulties in quality traceability, this invention maps the workpiece into at least two independent physical test areas based on pre-established physical test area division rules, constructs a workpiece-level feature vector containing regional attributes, and achieves structured and modular characterization of complex workpieces, accurately extracting feature information of key areas, and improving the accuracy and targeting of defect identification. Compared with the problems of using a general overall assessment in existing technologies, which easily masks local minor defects, this invention finally uses a lightweight regression mapping model that matches the current workpiece material to infer the feature vector, outputs the predicted value of the interface equivalent bonding strength of each physical test area, and compares the predicted value with a preset safety threshold to generate a graded early warning instruction fed back to the production line control system. This achieves online, real-time, and intelligent prediction and graded closed-loop management of interface peeling risk, improving the accuracy of interface peeling warning for polymer-coated automotive armrest panels.

[0011] 2. This invention introduces an incremental calibration step based on small sample data after model initialization. Specifically, it uses the loading of a new batch of parameterized formula templates or the accumulation of prediction deviations as trigger conditions to obtain the offline destructive test true values ​​and online predicted values ​​of a small number of small sample specimens to calculate calibration parameters. Based on this, the output bias term or slope of the lightweight regression mapping model is fine-tuned online. This enables the model to quickly and adaptively adjust to new material batches or process fluctuations. Compared with the existing technology where the model is fixed once trained and cannot cope with the decline in prediction accuracy caused by differences in raw material batches, this invention significantly improves the model's generalization ability and prediction accuracy in actual production environments.

[0012] 3. By constructing a health baseline model based on qualified workpiece samples, the distribution characteristics of the predicted values ​​of the equivalent bonding strength of the interface of qualified workpieces are monitored in real time within a preset statistical period. When an overall shift in the distribution characteristics is detected, environmental or equipment drift is determined to have occurred, and the drift compensation amount is automatically calculated to dynamically adjust the model output bias parameters or preset safety thresholds. This achieves real-time compensation for systematic errors caused by production environment noise, such as temperature and humidity changes and equipment condition degradation. Compared with the existing technology, which lacks a dynamic correction mechanism for environmental interference and is prone to false alarms or missed alarms due to environmental fluctuations, this effectively improves the robustness and long-term stability of the early warning system. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating an interface peeling warning method for a polymer-coated automotive armrest panel, provided in this application embodiment;

[0015] Figure 2 A flowchart for generating graded early warning instructions provided in an embodiment of this application;

[0016] Figure 3 A flowchart for incremental model calibration based on small sample data provided in this application embodiment;

[0017] Figure 4 A flowchart of residual feedback self-correction based on furnace-fed test specimens provided for embodiments of this application. Detailed Implementation

[0018] The following provides explanations for some of the terms used in this application. It should be noted that these explanations are for the convenience of those skilled in the art and do not constitute a limitation on the scope of protection claimed in this application.

[0019] The embodiments of this application involve at least one, including one or more; where "multiple" means two or more. Furthermore, it should be understood that in the description of this specification, terms such as "first," "second," and "third" are used only for descriptive purposes and should not be construed as indicating relative importance or order. For example, "first device" and "second device" do not represent the degree of importance of the two or their order, but are merely for descriptive distinction. In the embodiments of this application, "and / or" merely describes an association relationship, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0020] The directional terms mentioned in the embodiments of this application, such as "up", "down", "left", "right", "inner", and "outer", are only for reference to the directions in the accompanying drawings. Therefore, the directional terms used are for better and clearer explanation and understanding of the embodiments of this application, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0021] References to "one embodiment," "in some examples," or "some embodiments" as described in the embodiments of this application mean that one or more embodiments of this specification include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in some examples," "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] like Figure 1 The diagram shown is a flowchart of a method for early warning of interface peeling of a polymer-coated automotive armrest panel according to an embodiment of this application. The method includes the following steps:

[0024] S1: Obtain the unique identification information of the workpiece, perform multi-source non-destructive testing on the workpiece, and bind the test data with the unique identification information; a combination of QR code and radio frequency identification (RFID) technology is used to achieve efficient and accurate identification management; specifically, QR codes are used to record static information of the workpiece, such as production batch, material formula, and process parameters, while RFID tags are used to track the flow status of the workpiece in the production line in real time, such as the current processing step, dwell time, and process execution status; production batch information is used to trace the specific production time and production line information of the workpiece, the material formula records the composition and proportion of raw materials in detail, and the process parameters cover key variables such as processing temperature, pressure, and time. This information provides an important basis for subsequent quality analysis and process optimization; the rules for generating identification information follow the international standard ISO / IEC. The system employs a unique coding system (15693) and integrates it with the enterprise's custom coding system to ensure the global uniqueness of information. For example, the beginning of the identification information uses a fixed-length code to represent the production year and month of the workpiece, the middle part generates a unique identifier through a specific algorithm to distinguish different workpieces within the same batch, and the end part is reserved for enterprises to extend the coding according to their own needs. In addition, during the information storage process, the system first performs a hash operation on the identification information to generate a fixed-length hash value, which is then stored in a distributed database along with the original information, thereby improving data security and tamper resistance. By binding the encrypted identification information with the unique identifier of the workpiece, the system further improves the standardization and traceability of data management, providing reliable data support for subsequent multi-source non-destructive testing and interface stripping early warning.

[0025] Furthermore, multi-source non-destructive testing is performed on the workpiece, and the test data is bound to unique identification information, including:

[0026] The system acquires visual images, mechanical response signals, acoustic response signals, and process parameters of the workpiece. Visual image acquisition utilizes a high-resolution industrial camera with a resolution of 4096×2160 pixels and a frame rate of 30fps to meet the image capture requirements of rapid production. Mechanical response signals are acquired using piezoelectric force sensors with a sampling frequency of 10kHz to accurately record changes in force distribution on the workpiece surface. Acoustic response signals are acquired by acoustic emission sensors with a frequency range covering 50kHz to 500kHz to detect weak acoustic signals generated during interface peeling. Process parameters are recorded in real-time through an embedded sensor network. Specifically, temperature sensors employ thermocouple technology to measure temperature changes on the workpiece surface with an accuracy of ±0.5℃; pressure sensors, based on strain gauge principles, accurately capture changes in pressure distribution. Time parameters are recorded using a built-in clock module to ensure data synchronization. All sensors are connected to the central data processing unit via wireless communication protocols, forming a highly efficient distributed acquisition system. All acquisition devices undergo rigorous calibration to ensure data accuracy and consistency. The acquired data is stored in a structured format for easy subsequent processing and analysis.

[0027] Utilizing the non-processing intervals during workpiece flow on the production line, an asynchronous parallel processing architecture is employed to preprocess and fuse visual images, mechanical response signals, acoustic response signals, and process parameters in the background, generating a workpiece-level multi-source dataset. Specifically, a timestamp synchronization mechanism is used to associate different types of data streams with workpiece identification information, ensuring temporal consistency and spatial correspondence of the data. Data cleaning, format conversion, and feature extraction operations are performed for each type of data, tailored to their characteristics. For example, visual image data is filtered by Gaussian to remove noise, and then a convolutional neural network is used to extract key visual features; mechanical response signals are converted to frequency domain features using a fast Fourier transform to highlight periodic components. All preprocessed data is integrated into a workpiece-level multi-source dataset, providing high-quality data support for subsequent region segmentation and feature vector construction.

[0028] S2, based on the pre-established physical detection area division rules, maps the workpiece into at least two independent physical detection areas and constructs a workpiece-level feature vector containing area attributes.

[0029] Furthermore, based on pre-established physical inspection area division rules, the workpiece is mapped into at least two independent physical inspection areas, including:

[0030] The workpiece is pre-divided into at least two fixed physical regions containing critical bonding risk points. Critical bonding risk points refer to areas on the workpiece surface that are prone to interfacial delamination due to factors such as geometry, material distribution characteristics, or bonding process parameters. To accurately identify critical bonding risk points, the workpiece is pre-divided into at least two fixed physical regions, each containing potential weak bonding areas. The division of regions is mainly based on the structural design drawings of the workpiece and the analysis results of historical bonding failure data. For example, based on the geometry and material distribution characteristics of the workpiece, areas prone to stress concentration are identified as key detection targets, such as surfaces with large curvature changes or locations of abrupt changes in material thickness. At the same time, the region boundaries are further refined by combining bonding process parameters, such as the adhesive application path and curing temperature distribution. The criteria for dividing fixed physical regions follow the principle of minimizing overlap between regions and ensuring complete coverage of all high-risk areas. Region boundaries are determined using a graph-based segmentation algorithm. This algorithm generates continuous and closed region boundaries by calculating the curvature and normal vector changes of the workpiece surface. Specifically, the workpiece surface is first discretized using finite element analysis software, extracting the curvature and normal vector information of each node. Then, a weighted undirected graph is constructed based on these geometric properties, where nodes represent discrete points on the workpiece surface, and edge weights reflect the geometric similarity between adjacent points. Finally, the minimum cut algorithm is used to segment the graph into multiple sub-regions, each corresponding to a fixed physical detection region. This approach not only highlights the importance of high-risk areas but also lays the foundation for accurate classification of subsequent detection data.

[0031] A positional mapping relationship is established between the probe position of multi-source non-destructive testing and a fixed physical area. A high-precision laser scanner is used to scan the workpiece surface line by line along a preset path, recording the spatial coordinates of each scan point. The data obtained from multiple scans are stitched together into a complete workpiece surface model using a point cloud registration algorithm. The acquired point cloud data not only contains rich geometric details but also reflects the microscopic morphological features of the workpiece surface, providing an important foundation for the subsequent accurate registration of the probe position. The iterative nearest point algorithm is used to register the probe installation position with the workpiece surface model, thereby determining the precise coordinates of the probe in the workpiece coordinate system. Specifically, the coarse position of the probe is initialized, and then the nearest point distance between the key points on the probe and the point cloud on the workpiece surface is calculated. The rigid body transformation parameters, including the rotation matrix and translation vector, are solved using the least squares method. The probe position is updated according to the solved transformation parameters, and the above steps are repeated until convergence.

[0032] To further optimize the positional mapping between the workpiece surface and the multi-source non-destructive testing probe, a Voronoi diagram segmentation algorithm is used to divide the workpiece surface into several sub-regions and assign the nearest probe position to each sub-region. The Voronoi diagram is a segmentation method based on spatial proximity. Its basic principle is to divide the plane into multiple cells, where the distance from any point in each cell to the corresponding generated point in that cell is less than the distance to other generated points. The generated point is the installation position of the probe, and the cell is the detection sub-region of the workpiece surface. The threshold for the nearest probe distance is theoretically calculated based on the density and noise level of the workpiece surface point cloud data to ensure the accuracy of the mapping relationship, thereby achieving efficient classification of the detection data.

[0033] The detection data from the workpiece-level multi-source dataset are directly categorized into corresponding fixed physical regions, generating a unified feature vector with region labels. The feature vector consists of visual features, mechanical features, acoustic features, and process parameter features. Visual features primarily describe the appearance of the workpiece surface, such as the morphology and distribution of defects like cracks and bubbles, directly reflecting the integrity of the bonding interface. Mechanical features focus on the dynamic response of the workpiece under stress conditions, such as vibration frequency and energy distribution, effectively revealing the internal stress state of the bonding area. Acoustic features analyze the propagation characteristics of ultrasonic waves in materials to assess the bonding strength and defect size of the bonding interface. Process parameter features record key variables in the manufacturing process, such as adhesive application temperature and curing time, providing important clues for analyzing the causes of interface delamination. For example, regarding visual features… A pre-trained ResNet-50 model is used to extract deep semantic information from regional images. The high-resolution image of the workpiece surface is segmented into several sub-images corresponding to fixed physical regions. These sub-images are then input into the ResNet-50 model for feature extraction. The convolutional neural network automatically learns abstract features in the image, such as texture and edges, thereby generating feature vectors with strong representation capabilities. For mechanical features, wavelet packet transform is used to decompose the mechanical response signal and obtain its energy distribution in different frequency bands. Wavelet packet transform can decompose complex time-domain signals into multiple sub-frequency bands, thereby revealing hidden local features in the signal. For acoustic features and process parameter features, support vector machine classifiers and principal component analysis methods are combined respectively to further optimize the feature extraction and selection process, thereby improving the overall quality and applicability of the feature vectors.

[0034] Specifically, establishing a mapping set is fundamental to achieving precise correspondence between the workpiece and the multi-source non-destructive testing probe. The method for setting this mapping set mainly includes the principles for establishing the mapping relationship, data sources, and accuracy assurance measures. First, the mapping relationship follows the nearest neighbor principle, meaning that the detection data within each fixed physical area is only associated with the nearest probe position to avoid data redundancy or confusion. Second, the data sources cover both workpiece surface point cloud data and probe installation coordinates. The point cloud data is obtained through 3D laser scanning technology, while the probe coordinates are determined using an iterative nearest point algorithm registered with the workpiece surface model. To ensure the accuracy and completeness of the mapping set, a cross-validation mechanism is introduced. Different probe layout schemes are simulated to evaluate their impact on data classification accuracy, and the optimal scheme is selected as the final mapping set. This mapping set setting method effectively improves the positioning accuracy of the detection data, laying a solid foundation for subsequent feature extraction and early warning analysis.

[0035] S3 invokes a lightweight regression mapping model matching the current workpiece material to infer the workpiece-level feature vector and outputs the predicted value of the interface equivalent bond strength of each physical detection area. The workpiece-level feature vector mainly consists of physical, chemical, and process parameters related to the delamination of the polymer coating interface. These parameters include key variables such as material surface energy, elastic modulus, coefficient of thermal expansion, and temperature and pressure during the production process. The specific inference process of the lightweight regression mapping model is to take the workpiece-level feature vector as input and combine it with the benchmark feature parameters loaded from the parameterized formula template library and the pre-trained regression coefficients. In this process, the model adopts a combination of linear regression and nonlinear activation functions to perform multi-layer mapping and transformation on the feature vector to capture complex nonlinear relationships. For example, for the material formula of rigid matrix and elastic coating, the model will dynamically adjust the weight coefficients according to the difference in surface energy and elastic modulus to achieve more accurate prediction. The predicted value of the interface equivalent bond strength of each physical detection area output by the model is presented in a structured data format, which usually includes the predicted value, confidence interval, and corresponding detection area identifier.

[0036] Furthermore, a lightweight regression mapping model matching the current workpiece material is invoked, including:

[0037] A parametric formulation template library was pre-established for different material formulations. The library stores the baseline characteristic parameters and pre-trained regression coefficients for each material. The construction process of the parametric formulation template library includes three main steps: material formulation classification, determination of baseline characteristic parameters, and acquisition of pre-trained regression coefficients. Specifically, based on the chemical composition and physical properties of polymer materials, formulations are categorized into rigid matrices and elastic coatings, flexible matrices and rigid coatings, etc., and a hierarchical classification system is established. Then, for each type of material formulation, key parameters that significantly affect its adhesive performance are selected as baseline characteristics, such as surface energy, elastic modulus, and coefficient of thermal expansion. These parameters are determined through a combination of experimental measurements and literature review, and are stored in the template library as key-value pairs. The data is stored in the form of [database name missing], and finally, a lightweight regression mapping model is trained using historical production data to obtain pre-trained regression coefficients corresponding to each material formulation. The parameterized formulation template library is managed using a NoSQL database, supporting efficient query and update operations, thus providing a reliable guarantee for the rapid initialization of the model. Surface energy reflects the wettability and adhesion of the material surface, and its value can be calculated by combining contact angle measurement experiments with the Owens-Wendt-Rabel-Kaelble equation. The elastic modulus characterizes the material's resistance to deformation under stress and can be obtained through dynamic mechanical analysis or static tensile tests. The coefficient of thermal expansion affects the dimensional stability of the material under temperature change conditions, and its determination usually relies on a thermomechanical analyzer.

[0038] When a production line changeover signal is detected, the corresponding baseline parameter template is directly loaded from the parametric formula template library based on the material information of the current workpiece, and the lightweight regression mapping model is initialized. The specific process is as follows: First, the material identification module decodes the workpiece identification information and extracts the material formula code contained therein. For example, RG-45 represents a combination of a rigid matrix and an elastic coating. Based on a predefined hierarchical classification system, this code is mapped to a specific classification node in the template library. Second, during the template library construction phase, each material formula is assigned a unique hash value, which is generated using the SHA-256 algorithm to ensure uniqueness. To address the issues of uniformity and irreversibility, during model initialization, the corresponding hash value is calculated based on the material formula code output by the material recognition module. This hash value is then used to quickly locate the target template record. After loading, the baseline feature parameters and pre-trained regression coefficients are directly assigned to the lightweight regression mapping model as initial inputs. Finally, the model is fine-tuned based on the specific process parameters of the current workpiece, such as temperature and pressure, to adapt to actual production conditions. To ensure accurate matching between stored materials and the model, a hash table-based indexing mechanism is introduced. The target template is quickly located by calculating the hash value of the material formula, thereby significantly improving the efficiency of model initialization.

[0039] Specifically, obtaining the pre-trained regression coefficients corresponding to each material formulation is a crucial step in constructing a lightweight regression mapping model. Training the model using historical production data involves three key steps: data preprocessing, model selection, and training methods. In the data preprocessing stage, the raw data is first cleaned and normalized to eliminate the impact of outliers and dimensional differences on model performance. Then, principal component analysis (PCA) is used to reduce the dimensionality of high-dimensional features, retaining the principal components that contribute most to the prediction of adhesive performance, thereby improving the model's computational efficiency and generalization ability. For model selection, a lightweight regression model based on gradient boosting trees is chosen. This model is widely used in industrial forecasting tasks due to its strong fitting ability to nonlinear relationships and robustness to noisy data. During training, five-fold cross-validation is used to optimize the model's hyperparameters, and an early stopping mechanism is employed to avoid overfitting and improve the model's convergence speed. Finally, pre-trained regression coefficients are obtained for different material formulations and stored in a parametric formulation template library, providing a reliable guarantee for subsequent model initialization.

[0040] The lightweight regression mapping model is constructed using any one of the following algorithms: gradient boosting decision tree algorithm, lightweight gradient boosting machine algorithm, random forest algorithm, partial least squares algorithm, or support vector regression algorithm. The cleaned and normalized dataset is divided into training, validation, and test sets in a 7:2:1 ratio to ensure the model's performance is representative across different datasets. The training set is used to learn the model parameters, the validation set is used to tune hyperparameters and evaluate the model's generalization ability, and the test set is used for final performance validation. In the training of the lightweight gradient boosting machine model, the core hyperparameters are first initialized, with the learning rate set to 0.1, the maximum tree depth set to 5, and the number of iterations set to 100. During training, an early stopping mechanism is used to avoid overfitting, i.e., stopping when the root mean square error on the validation set is continuously below 1. Training was stopped when the performance did not decrease after zero iterations. Five-fold cross-validation was used to split the training set, randomly dividing it into five subsets. Each time, one subset was selected as the validation set, and the remaining four subsets were used as the training set. This process was repeated five times, and the average performance metric was calculated. Cross-validation not only comprehensively evaluated the model's performance on different data subsets but also effectively reduced bias caused by improper data partitioning. Based on the cross-validation results, the model was further optimized to improve its performance. To address potential overfitting, the regularization parameter was adjusted to limit model complexity, thereby reducing excessive reliance on training data. The learning rate and number of iterations were dynamically adjusted, employing a gradually decreasing learning rate strategy to allow for more refined parameter optimization in the later stages of training.

[0041] S4 compares the predicted value of the equivalent bonding strength of the interface with the preset safety threshold, generates a graded early warning instruction based on the comparison result, and feeds it back to the production line control system.

[0042] like Figure 2 The flowchart for generating graded early warning instructions shown further includes comparing the predicted value of the equivalent bonding strength of the interface with a preset safety threshold, and generating graded early warning instructions based on the comparison result, including:

[0043] A multi-level judgment standard is set, including warning thresholds and rejection thresholds, with the warning threshold being higher than the rejection threshold; the preset safety thresholds include warning thresholds and rejection thresholds.

[0044] When the predicted value of the equivalent bond strength of the interface is lower than the warning threshold but greater than or equal to the rejection threshold, a first-level warning instruction is generated, the current workpiece's inspection data is recorded, and the manual re-inspection process is triggered. This means that the workpiece's inspection data is uploaded to the quality management system, and the on-site operators are notified to conduct visual inspection and supplementary testing on the target workpiece. At the same time, relevant information is recorded in the traceability database.

[0045] When the predicted value of the equivalent bonding strength of the interface is lower than the rejection threshold, a secondary warning instruction is generated, triggering the workpiece rejection and equipment shutdown inspection process. That is, the production line control system automatically locks the relevant workstation to prevent unqualified products from flowing into the next process, and at the same time sends an alarm message to the maintenance personnel to prompt them to check the equipment failure.

[0046] The preset safety threshold is a fixed mechanical strength value set based on national standards or OEM enterprise standards. In order to cope with the uncertainty of the production environment, a preset safety margin is introduced, that is, the warning threshold is increased by 5% and the rejection threshold is decreased by 5%, thereby improving the fault tolerance of the interface peeling warning. The warning threshold is set to 90% of the average equivalent bonding strength of the interface of qualified workpieces to identify potential risks, while the rejection threshold is set to 80% of the average value to determine obviously unqualified products.

[0047] It should be noted that raising the warning threshold increases the threshold for entering the warning zone, ensuring that most workpieces with potential risks are included in the monitoring scope and avoiding missing early failure signals. Lowering the rejection threshold is intended to offset the inherent measurement uncertainty of online non-destructive testing and prevent the false rejection of qualified products due to signal noise. It does not lower the quality and safety standards, because workpieces on the boundary between the rejection threshold and the warning threshold have been classified as first-level warning areas and must undergo a more precise manual re-inspection process for final decision. This minimizes the false scrap rate of the production line while ensuring zero defects in product quality.

[0048] This application provides an interface peeling early warning system for polymer-coated automotive armrest panels, comprising: a data acquisition and binding module, a region mapping and feature construction module, an interface strength prediction module, and a graded early warning and feedback module.

[0049] The data acquisition and binding module is used to obtain the unique identification information of the workpiece, perform multi-source non-destructive testing on the workpiece, and bind the test data with the unique identification information to ensure the traceability and integrity of the data. The output of this module serves as the input of the subsequent region mapping and feature construction module, providing basic data support for the operation of the entire early warning system.

[0050] The region mapping and feature construction module is used to map the workpiece into at least two independent physical detection regions based on pre-established physical detection region division rules, and construct workpiece-level feature vectors containing region attributes. Through region-level data slicing, the original feature distribution of key bonding risk points is accurately preserved, and the sensitivity of feature data to local interface attenuation is significantly improved.

[0051] The interface strength prediction module is used to call a lightweight regression mapping model that matches the current workpiece material, infer the workpiece-level feature vector, and output the predicted value of the interface equivalent bonding strength of each physical detection area; thus achieving efficient adaptation to different types of workpieces.

[0052] The graded early warning and feedback module compares the predicted value of the equivalent bonding strength of the interface with the preset safety threshold, generates graded early warning instructions based on the comparison results, and feeds them back to the production line control system. This enables accurate identification of risks of different degrees and real-time control of the production process.

[0053] Example 2: As production progresses, model predictions may deviate due to factors such as material batch fluctuations. To address this issue, based on Example 1, after initializing the lightweight regression mapping model, an incremental model calibration step based on small sample data is also included:

[0054] like Figure 3The flowchart shown illustrates incremental model calibration based on small sample data. When a new parametric formula template is loaded or model prediction bias accumulation is detected, the true values ​​of offline destructive tests on a set number of small sample specimens are obtained. When model prediction bias accumulation is detected, the mean square error between the predicted value and the true value of the offline destructive test is calculated in real time, and a bias accumulation threshold is set to determine whether the model needs calibration. The bias accumulation threshold is set based primarily on historical data distribution characteristics, process tolerance range, and initial model performance. The true value of the offline destructive test is the actual equivalent interfacial bond strength value obtained through offline destructive testing of several randomly selected small sample specimens from the current production batch. The offline destructive test methods include tensile shear tests and peel strength tests. The test results are averaged after three repeated measurements to ensure data reliability. In the tensile shear test, the specimen is fixed on a special fixture and a uniform load is applied axially until the specimen fails. The maximum load value at failure is recorded, and the interfacial bond strength is calculated using a formula. In the peel strength test, the coating material on the surface of the specimen is peeled off at a uniform speed, and the force change during the peeling process is measured to calculate the average peel strength.

[0055] By using the true values ​​of offline destructive testing of small sample specimens and the corresponding online predicted values, the residuals between the two are calculated. The calibration parameters are then fitted using the least squares method. The optimal calibration parameters are found by minimizing the sum of squared errors between the predicted and actual values.

[0056] Based on calibration parameters, the output bias term or slope of the lightweight regression mapping model is fine-tuned online to complete the incremental calibration of the model.

[0057] Specifically, for fine-tuning the bias term, if the model's overall predicted values ​​are detected to be too high or too low, for example due to differences in the curing rates of different batches of glue, the bias parameters in the model configuration file are updated. For tree models, such as GBDT, this is achieved by adjusting the average output of all leaf nodes or by directly adding the calibration parameters of the bias term to the final output value. For linear models, such as PLS, the intercept term of the model is directly modified to correct the batch baseline bias.

[0058] For slope fine-tuning, if the model is found to have insufficient ability to distinguish between different intensity ranges, such as the predicted value distribution range being too narrow, the slope calibration parameter will be updated to the model output module. During actual inference, the original score output by the model will be automatically multiplied by the slope calibration parameter to scale it, so as to adapt it to the mechanical response range of different material formulations.

[0059] Example 3: To mitigate the impact of environmental and equipment condition fluctuations on model prediction accuracy, based on Example 1, after completing the incremental calibration of the model, a dynamic correction step for health baseline drift is also included:

[0060] A health baseline model based on qualified workpiece samples is constructed. The construction of the health baseline model first requires selecting representative qualified workpiece samples from the production process to ensure that the baseline can accurately reflect the performance of the system under normal operating conditions. Specifically, the selection of workpiece samples that are judged to be qualified within a preset statistical period must follow strict standards and methods. The criteria for judging qualified workpieces mainly include whether the predicted value of its interface equivalent bonding strength falls within the preset safety threshold range, and whether there are other obvious quality defects.

[0061] Based on the selection of qualified workpiece samples, the distribution characteristics of the predicted interface equivalent bond strength of workpieces judged as qualified within a preset statistical period are statistically analyzed in real time. The distribution characteristics include the mean and standard deviation of the predicted interface equivalent bond strength. The healthy baseline model performs statistical analysis on the predicted interface equivalent bond strength of workpieces judged as qualified within a preset statistical period, calculates its mean and standard deviation, and uses them as reference indicators for the baseline. For example, if the statistical period is one week, the predicted values ​​of all qualified workpieces in that week are added together and divided by the sample size to obtain the mean. An unbiased estimation method is used, that is, the standard deviation is obtained by taking the square root of the sample variance.

[0062] When an overall shift in distribution characteristics is detected, environmental or equipment drift is determined, and drift compensation is automatically calculated. Specifically, when the mean deviates from the historical baseline mean by more than two standard deviations, a significant drift is considered to have occurred. In this case, the difference between the mean in the current sliding window and the historical baseline mean is recorded as the drift compensation for the mean. When the rate of change of the standard deviation exceeds 15% of the historical baseline standard deviation, random interference in the production process is determined, such as uneven temperature distribution in the curing oven. In this case, the difference between the rate of change of the standard deviation in the current sliding window and the historical baseline standard deviation is recorded as the drift compensation for the standard deviation.

[0063] The output bias parameters or preset safety thresholds of the lightweight regression mapping model are dynamically adjusted using drift compensation to counteract the effects of environmental noise. Specifically, the adjusted output bias term can be represented as the difference between the original output bias term and the drift compensation amount of the mean. Similarly, the preset safety threshold is adjusted, and the adjusted preset safety threshold can be represented as the difference between the original preset safety threshold and the drift compensation amount of the standard deviation. This maintains the stability and accuracy of the prediction results, effectively reduces the impact of environmental noise on model performance, and improves the robustness of interface stripping warnings. Furthermore, to prevent extreme situations, such as sensor malfunctions causing numerical abrupt changes, an upper limit for the drift compensation amount is usually set. For example, the single drift compensation amount is limited to no more than 10% of the historical baseline mean. If the calculated drift compensation amount is too large, it should be judged as a sensor malfunction rather than drift, thus triggering a shutdown alarm instead of automatic compensation.

[0064] Example 4: To address the issue of representativeness bias in routine sampling tests on the production line, based on Example 1, a residual feedback self-correction step based on the furnace-fed test pieces is also included:

[0065] like Figure 4 The flowchart shown is a residual feedback self-correction process based on in-furnace test pieces. In the production process of the workpiece, in-furnace test pieces are generated synchronously, and the association between the in-furnace test pieces and the physical inspection areas of the corresponding workpieces is established. The selection of in-furnace test pieces follows the principle of random sampling. 5% of the workpieces in each batch of production are selected as test piece samples, and a one-to-one mapping relationship is established with their corresponding physical inspection areas. In the production process of the workpiece, each physical inspection area is assigned a unique identifier, and the in-furnace test pieces are associated with their corresponding workpieces by sharing the same identifier.

[0066] The true peel strength obtained from the offline destructive test of the furnace-loaded specimen is obtained. The residual between the true peel strength obtained from the offline destructive test and the corresponding predicted value of the equivalent bond strength of the interface is obtained. Tensile shear test and peel strength test are used to evaluate the bonding performance of the specimen under different stress conditions. In the tensile shear test, the specimen is fixed on a special fixture and a uniform tensile force is applied along the bonding interface until the interface peels off. In the peel strength test, the required peel force is measured by gradually peeling off the adhesive layer. Each test is repeated three times and the average value is taken as the final test result.

[0067] When the residual exceeds the preset tolerance range, the model self-correction mechanism is triggered, and the output bias parameters or slope of the model are updated through the gradient descent algorithm, thereby gradually reducing the prediction error. The setting of the tolerance range needs to take into account both the model prediction accuracy requirements and the actual production conditions. Usually, 1.5 times the standard deviation of the model prediction value is taken as the threshold.

[0068] It should be noted that the above-mentioned incremental calibration of the model based on small sample data, dynamic correction of health baseline drift, and residual feedback self-correction based on in-furnace test pieces can be implemented as independent technical means, or two or three of them can be combined according to the actual computing power and quality control level of the production line, all of which fall within the protection scope of this invention.

[0069] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)). Where there is no conflict, the solutions in the above embodiments can be combined.

[0070] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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.

[0071] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, 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 apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] 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 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] 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 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope and intent of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A method for early warning of interface peeling of polymer-coated parts for automotive armrests, characterized in that, Includes the following steps: S1, obtain the unique identification information of the workpiece, perform multi-source non-destructive testing on the workpiece, and bind the test data with the unique identification information; S2, based on the pre-established physical detection area division rules, the workpiece is mapped into at least two independent physical detection areas, and a workpiece-level feature vector containing area attributes is constructed. S3 calls a lightweight regression mapping model that matches the current workpiece material to infer the workpiece-level feature vector and outputs the predicted value of the interface equivalent bonding strength of each physical detection area. S4 compares the predicted value of the equivalent bonding strength of the interface with the preset safety threshold, generates a graded early warning instruction based on the comparison result, and feeds it back to the production line control system.

2. The method for early warning of interface peeling of polymer-coated automotive armrest panels as described in claim 1, characterized in that: The process of performing multi-source non-destructive testing on the workpiece and binding the test data with unique identification information includes: Acquire visual images, mechanical response signals, acoustic response signals, and process parameters of the workpiece; By utilizing the non-processing gap time of the workpiece during its flow on the production line, an asynchronous parallel processing architecture is adopted to preprocess and fuse visual images, mechanical response signals, acoustic response signals and process parameters in the background to generate a workpiece-level multi-source dataset.

3. The method for early warning of interface peeling of polymer-coated automotive armrest panels as described in claim 2, characterized in that: The method of mapping the workpiece into at least two independent physical detection areas based on pre-established physical detection area division rules includes: The workpiece is pre-divided into at least two fixed physical areas containing critical bonding risk points; Establish a position mapping relationship between the probe position of multi-source non-destructive testing and a fixed physical area; The detection data in the workpiece-level multi-source dataset is directly classified into the corresponding fixed physical region, and a region-level unified feature vector with region label is generated.

4. The method for early warning of interface peeling of polymer-coated automotive armrest panels as described in claim 1, characterized in that: The invocation of the lightweight regression mapping model that matches the current workpiece material includes: A parameterized formula template library is pre-established for different material formulations. The template library stores the benchmark feature parameters and pre-trained regression coefficients corresponding to each material. When a production line change signal is detected, the corresponding baseline parameter template is directly loaded from the parametric recipe template library based on the material information of the current workpiece, and the lightweight regression mapping model is initialized.

5. The method for early warning of interface peeling of polymer-coated automotive armrest panels as described in claim 4, characterized in that: After initializing the lightweight regression mapping model, the process also includes an incremental model calibration step based on small sample data: When a new parametric recipe template is loaded or when model prediction bias is detected to accumulate, obtain the offline destructive test truth values ​​of a set number of small sample specimens. Calibration parameters are calculated using the true values ​​of offline destructive testing on a small sample of specimens and the corresponding online predicted values. Based on calibration parameters, the output bias term or slope of the lightweight regression mapping model is fine-tuned online to complete the incremental calibration of the model.

6. The method for early warning of interface peeling of polymer-coated automotive armrest panels as described in claim 5, characterized in that: After the incremental calibration of the completed model, a dynamic correction step for the health baseline drift is also included: A health baseline model based on qualified workpiece samples is constructed, and the distribution characteristics of the predicted interface equivalent bond strength of workpieces judged to be qualified within a preset statistical period are statistically analyzed in real time. When an overall shift in the distribution characteristics is detected, it is determined that environmental or equipment drift has occurred, and the drift compensation amount is automatically calculated. The output bias parameters or preset safety thresholds of the lightweight regression mapping model can be dynamically adjusted using drift compensation.

7. The method for early warning of interface peeling of polymer-coated automotive armrest panels as described in claim 6, characterized in that: It also includes a residual feedback self-correction step based on the furnace-fed test specimen: During the production process of the workpiece, furnace-fed test pieces are generated simultaneously, and the correlation between the furnace-fed test pieces and the physical testing areas of the corresponding workpiece is established. Obtain the true peel strength from the offline destructive test of the furnace-fed specimen, and calculate the residual between it and the predicted value of the equivalent interfacial bond strength; When the residual exceeds the preset tolerance range, the model self-correction mechanism is triggered, and the output bias parameters or slope of the lightweight regression mapping model are updated using the residual.

8. The method for early warning of interface peeling of polymer-coated automotive armrest panels as described in claim 1, characterized in that: The step of comparing the predicted value of the equivalent bonding strength of the interface with a preset safety threshold and generating a graded early warning instruction based on the comparison result includes: A multi-level judgment standard is set, including a warning threshold and a rejection threshold, where the warning threshold is higher than the rejection threshold; When the predicted value of the equivalent bonding strength of the interface is lower than the warning threshold but greater than or equal to the rejection threshold, a first-level warning instruction is generated, the current workpiece's inspection data is recorded, and the manual re-inspection process is triggered. When the predicted value of the equivalent bonding strength of the interface is lower than the rejection threshold, a secondary warning instruction is generated, triggering the workpiece rejection and equipment shutdown inspection process.

9. The method for early warning of interface peeling of polymer-coated automotive armrest panels as described in claim 1, characterized in that: The lightweight regression mapping model is constructed using any one of the following algorithms: gradient boosting decision tree algorithm, lightweight gradient boosting machine algorithm, random forest algorithm, partial least squares method, or support vector regression algorithm. The preset safety threshold is a fixed mechanical strength value set based on national standards or OEM enterprise standards, and includes a preset safety margin.

10. A pre-warning system for interface peeling of polymer-coated automotive armrest panels, characterized in that, include: The module includes data acquisition and binding, region mapping and feature construction, interface intensity prediction, and hierarchical early warning and feedback. The data acquisition and binding module is used to obtain the unique identification information of the workpiece, perform multi-source non-destructive testing on the workpiece, and bind the test data with the unique identification information. The region mapping and feature construction module is used to map the workpiece into at least two independent physical detection regions based on pre-established physical detection region division rules, and construct a workpiece-level feature vector containing region attributes. The interface strength prediction module is used to call a lightweight regression mapping model that matches the current workpiece material, infer the workpiece-level feature vector, and output the predicted value of the interface equivalent bonding strength of each physical detection area. The graded early warning and feedback module is used to compare the predicted value of the equivalent bonding strength of the interface with the preset safety threshold, generate graded early warning instructions based on the comparison results, and feed them back to the production line control system.

Images