Method and system for identifying hot-pressing defects of plastic cellular board

Abstract

The invention provides a plastic cellular board hot-pressing defect identification method and system, and relates to the technical field of defect identification. The method comprises the following steps: monitoring the position of a plate and generating position information; applying instantaneous local thermal excitation to the surface of the plate to generate a thermal excitation area; optical image data and infrared thermal image data are collected; obtaining temperature difference data according to the infrared thermal image data; distortion correction and noise filtering are carried out on the optical image data to obtain corrected optical image data; registering the temperature difference data with the corrected optical image data to obtain registered multi-modal data; and the defect problem is judged by analyzing the registered multi-modal data. The method provided by the invention overcomes the problem that the defect optical and infrared thermal imaging characteristics are difficult to recognize along with time attenuation when the plate is in a rapid moving and continuous cooling environment, and realizes robustness recognition and accurate positioning of hot pressing defects below millimeter level.

Description

Technical Field

[0001] The present invention relates to the technical field of defect identification, and in particular to a method and system for identifying hot pressing defects of plastic honeycomb panels. Background Art

[0002] As a lightweight, high-strength structural material, plastic honeycomb panels are widely used in aerospace, rail transportation, construction, automotive, and other fields. The hot pressing process is a critical step in its manufacturing process, determining the ultimate performance and quality of the panels. However, the hot pressing process can introduce various internal or surface defects, such as delamination, voids, cracks, and poor welds. These defects can seriously affect the structural integrity and service life of the panels. Therefore, efficient and accurate quality inspection of hot-pressed plastic honeycomb panels is crucial.

[0003] Traditional inspection methods, such as visual inspection, ultrasonic testing, and radiographic testing, have limitations. Visual inspection relies on manual experience, is inefficient, and struggles to detect internal defects. While ultrasonic and radiographic testing can detect internal defects, they often require couplant or radiation protection for operators. Furthermore, their speed is slow, making them incapable of meeting the demands of modern, high-speed production lines. Furthermore, some traditional methods can cause minor damage to the sheet metal. Therefore, the development of a non-destructive, efficient, and automated online inspection technology is a pressing need in the industry.

[0004] On an actual production line, after the plastic honeycomb panel is removed from the hot pressing mold, it is usually continuously transported to subsequent processes, including quality inspection, via a conveyor belt. At this time, the panel still has significant residual heat and continues to dissipate heat to the environment during movement. This high temperature and dynamic cooling characteristic poses unique challenges to non-destructive testing. Existing automated inspection systems often integrate optical imaging equipment (such as visible light cameras) and infrared thermal imaging equipment. Optical imaging is used to capture visible light images of the panel surface to detect surface defects; infrared thermal imaging is used to obtain surface temperature distribution images of the panel and infer internal defects by analyzing temperature anomalies.

[0005] However, single-modality inspection methods have limitations in environments where the sheet metal is moving rapidly and cooling continuously. First, the inspection process needs to be performed while the production line is running at high speed, requiring extremely fast image acquisition speeds to avoid motion blur and ensure coverage. Second, the surface temperature distribution of the sheet metal changes continuously over time. Internal defects (such as delamination or voids) may appear as subtle localized temperature anomalies when the sheet metal is at high temperatures (for example, delamination areas with slow heat dissipation may appear as hot spots, while void areas may appear as hot or cold spots due to air insulation). However, as the sheet metal cools, these subtle temperature differences quickly decrease or even disappear, making the defects difficult to detect using infrared thermal imaging. Similarly, certain surface defects (such as fine cracks) may be more obvious at high temperatures due to the thermal expansion and contraction of the material, but become less noticeable after cooling, making them more difficult to identify using optical imaging. This results in significant differences in the appearance of defect characteristics at different temperatures and time points, even attenuating to the point of being unrecognizable.

[0006] To improve detection robustness, attempts are underway to fuse optical and infrared thermal images. This requires addressing data registration challenges arising from different imaging principles, resolutions, fields of view, and high-speed plate movement. Even if successful, extracting robust features from the fused data that are robust to defects of various types and sizes (especially millimeter-scale or smaller) while maintaining effective identification during dynamic changes in plate temperature remains a key technical challenge. For example, a tiny internal cavity may appear as a faint hotspot when the plate is freshly molded, disappearing as it cools. Simultaneously, this cavity may induce subtle surface deformation in the optical image. Effectively integrating this dynamically changing, time-correlated, yet distinctly different multimodal data, and accurately determining defect type and location, remains a significant challenge facing existing technologies. Furthermore, factors such as ambient light fluctuations, airflow, and internal heating of the inspection equipment at the production site can also interfere with the quality of optical and infrared images, further complicating detection. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and system for identifying hot-pressing defects in plastic honeycomb panels, which overcomes the problem of difficulty in identification caused by the decay of optical and infrared thermal imaging features of defects over time in a fast-moving and continuously cooling environment, and realizes robust identification and precise positioning of hot-pressing defects below the millimeter level.

[0008] In a first aspect, the present invention provides a method for identifying hot pressing defects of a plastic honeycomb panel, comprising the following steps:

[0009] After the plate conveying device is controlled to convey the plastic honeycomb plate to the detection area, the position of the plate is monitored in real time through a position sensor or encoder to generate position information;

[0010] According to the position information, after determining that the plate has reached the designated position in the detection area, the local weak thermal excitation device is controlled to apply instantaneous local thermal excitation to the surface of the plate to generate a thermal excitation area;

[0011] Controlling the optical imaging unit and the infrared thermal imaging unit to synchronously collect the optical image and the infrared thermal image of the plate in the thermal excitation area to obtain the optical image data and the infrared thermal image data;

[0012] Based on the infrared thermal image data, the temperature difference data is obtained by calculating the temperature difference map of the thermal excitation area before and after thermal excitation, or the temperature response characteristic data is obtained by analyzing the temperature response curve of the thermal excitation area during the thermal excitation process;

[0013] Performing distortion correction and noise filtering on the optical image data to obtain corrected optical image data;

[0014] registering the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain registered multimodal data;

[0015] By analyzing the registered multimodal data, we can determine whether there are defects, the defect type, defect location and defect size.

[0016] The method for identifying hot-pressed defects of plastic honeycomb panels provided by the present invention effectively enhances the difference in thermal response between defective areas and normal areas by introducing local weak thermal excitation during the online detection process of the plastic honeycomb panels after hot-pressing, in a dynamic environment of high-speed movement and continuous cooling on the production line. This significantly overcomes the problem of difficulty in identification caused by the decay of defect optical and infrared thermal imaging features over time in the fast-moving and continuous cooling environment of the panels. Combining the precise spatiotemporal synchronous acquisition of high-frame-rate optical imaging and infrared thermal imaging with multimodal data fusion processing, this solution improves the sensitivity and accuracy of identifying various types of (internal and surface) micro-defects below the millimeter level in complex and dynamic environments, achieving non-destructive, efficient, and robust detection of hot-pressed defects of plastic honeycomb panels, significantly improving the level of product quality control.

[0017] In a second aspect, the present invention provides a plastic honeycomb panel hot pressing defect identification system, comprising:

[0018] The first control module is used to control the plate conveying device to convey the plastic honeycomb plate to the detection area, monitor the position of the plate in real time through a position sensor or encoder, and generate position information;

[0019] The thermal excitation module is used to determine that the plate has reached the designated position in the detection area based on the position information, and then control the local weak thermal excitation device to apply instantaneous local thermal excitation to the surface of the plate to generate a thermal excitation area;

[0020] The second control module is used to control the optical imaging unit and the infrared thermal imaging unit to synchronously capture the optical image and the infrared thermal image of the plate in the thermal excitation area to obtain optical image data and infrared thermal image data;

[0021] An acquisition module is used to obtain temperature difference data by calculating a temperature difference map of a thermally actuated area before and after thermal actuation based on infrared thermal image data, or to obtain temperature response characteristic data by analyzing a temperature response curve of the thermally actuated area during thermal actuation;

[0022] A correction module, used to perform distortion correction and noise filtering on the optical image data to obtain corrected optical image data;

[0023] A registration module, used to register the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain registered multimodal data;

[0024] The analysis module is used to determine whether there are defects, the defect type, defect location and defect size by analyzing the registered multimodal data.

[0025] As can be seen from the above, the method for identifying hot pressing defects of plastic honeycomb panels provided by the present invention applies instantaneous, localized, weak thermal excitation to the surface of the panel when the panel moves along the production line to a specific detection area. Due to the difference in thermal conductivity between the defective area and the normal area, an enhanced temperature response will be generated under weak thermal excitation. After the excitation occurs or during the excitation process, the system synchronously collects optical images and infrared thermal images of the panel. By analyzing the temperature changes before and after the excitation or the temperature response differences during the excitation process, combined with the surface features of the optical image, the defects can be identified and located. The overall concept is to amplify the thermal characteristics of the defects through external auxiliary means, overcome the challenges of small temperature differences and unclear features under natural cooling conditions, and use optical imaging for supplementary verification and precise position determination.

[0026] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the embodiments of the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flow chart of a method for identifying hot pressing defects of plastic honeycomb panels provided by an embodiment of the present invention.

[0028] Figure 2 A schematic structural diagram of a plastic honeycomb panel hot pressing defect identification system provided by an embodiment of the present invention.

[0029] Description of labels:

[0030] 100, first control module; 200, thermal actuation module; 300, second control module; 400, acquisition module; 500, correction module; 600, registration module; 700, analysis module. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0032] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are used only to distinguish the description and should not be understood as indicating or implying relative importance.

[0033] Reference Attachment Figure 1 The present invention provides a method for identifying hot pressing defects of a plastic honeycomb panel, comprising the following steps:

[0034] After the plate conveying device is controlled to convey the plastic honeycomb plate to the detection area, the position of the plate is monitored in real time through a position sensor or encoder to generate position information;

[0035] According to the position information, after determining that the plate has reached the designated position in the detection area, the local weak thermal excitation device is controlled to apply instantaneous local thermal excitation to the surface of the plate to generate a thermal excitation area;

[0036] Controlling the optical imaging unit and the infrared thermal imaging unit to synchronously collect the optical image and the infrared thermal image of the plate in the thermal excitation area to obtain the optical image data and the infrared thermal image data;

[0037] Based on the infrared thermal image data, the temperature difference data is obtained by calculating the temperature difference map of the thermal excitation area before and after thermal excitation, or the temperature response characteristic data is obtained by analyzing the temperature response curve of the thermal excitation area during the thermal excitation process;

[0038] Performing distortion correction and noise filtering on the optical image data to obtain corrected optical image data;

[0039] registering the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain registered multimodal data;

[0040] By analyzing the registered multimodal data, the thermal response abnormality area is identified. Then, it is checked whether there are corresponding surface features in the thermal response abnormality area on the corrected optical image data. A comprehensive analysis is performed on the thermal response abnormality type, intensity, spatial distribution and corresponding optical features. Finally, based on the comprehensive analysis results, it is determined whether there is a defect, the defect type, the defect location and the defect size.

[0041] This method aims to address the technical challenge of accurately identifying hot-pressed defects in plastic honeycomb panels during panel movement and dynamic temperature fluctuations. By introducing localized weak thermal excitation and multimodal data fusion, the robustness of defect detection is enhanced. The method first controls panel movement and monitors its position to ensure that subsequent inspection operations are accurately performed when the panel reaches the designated area. Position information is essential for automated and synchronized inspection. After the panel reaches the designated location, a transient localized thermal excitation is applied. This step injects a small amount of heat into the panel, creating a detectable temperature difference or different temperature decay rates between the defective area and the normal area. This controlled thermal excitation overcomes the potential rapid decay of defect thermal signatures during natural cooling, enhancing the defect signal. Subsequently, optical and infrared thermal images of the panel are simultaneously acquired in the thermally stimulated area. The optical image provides visual information of the panel surface for surface defect detection and serves as a registration reference. The infrared thermal image captures the temperature distribution on the panel surface, detecting thermal anomalies caused by internal defects. Synchronous acquisition ensures that the two modal data correspond to the same time and physical area, laying the foundation for subsequent fusion. The infrared thermal image data is processed to calculate temperature difference maps or analyze temperature response curves. The temperature difference map visually displays the difference in temperature before and after thermal stimulation, while the temperature response curve reflects the heat diffusion and conduction characteristics within the plate. These processing steps extract defect-related thermal signature data from the raw thermal image. Distortion correction and noise filtering are performed on the optical image data to improve image quality. Distortion correction eliminates inherent geometric distortion in the optical system, and noise filtering reduces interference introduced by the environment or the sensor, allowing the optical image to more accurately reflect the true surface conditions of the plate, facilitating subsequent registration and surface feature analysis. The processed temperature difference data or temperature response signature data is registered with the corrected optical image data. Because optical and infrared imaging principles, fields of view, and resolutions may differ, and because the plate is in motion, registration is a key step in achieving multimodal data fusion. It spatially aligns the two modal data, ensuring that thermal and optical signatures from the same physical location are correlated. Finally, the registered multimodal data is analyzed to identify defects. This step combines information about thermal response anomalies with corresponding surface features in the optical image. By comprehensively analyzing the characteristics of thermal anomalies and their appearance in optical images, different types of defects can be more accurately distinguished, false positives eliminated, and the presence, type, location, and size of defects determined. This multimodal comprehensive analysis leverages the complementary strengths of the two imaging technologies to improve the accuracy and robustness of defect identification.

[0042] Specifically, the sheet material is continuously moved by a conveyor, its position monitored in real time by a position sensor or encoder. When a specific area of ​​the sheet material reaches a preset detection position, a localized weak thermal excitation device is triggered based on the position information. This device applies a brief heat pulse to the sheet material's surface, for example, using a flash lamp or a focused infrared light source. As a result, the heat conduction or heat dissipation characteristics of defective areas within or on the sheet material's surface differ from those of normal areas, resulting in a transient temperature difference or a different temperature decay process in the thermally stimulated area. Simultaneously with or immediately following the application of the thermal excitation, an optical imaging unit (such as a visible light camera) and an infrared thermal imaging unit simultaneously capture images of the thermally stimulated area. The optical image records the visible features of the sheet material's surface, while the infrared thermal image records the temperature distribution on the sheet material's surface. From the infrared thermal image data, the temperature difference before and after the thermal excitation can be calculated to form a temperature difference map, or the temperature curve of the area after the thermal excitation can be analyzed over time to extract the temperature response characteristics. This temperature difference data or temperature response characteristics characterize the thermophysical properties of the sheet material in that area. Simultaneously, the captured optical image data is processed, including removing geometric distortion caused by the camera lens and filtering out interference introduced by ambient light variations or sensor noise, to produce high-quality corrected optical image data. Subsequently, the extracted thermal signature data (temperature difference data or temperature response signature data) is spatially registered with the corrected optical image data. This is typically achieved by establishing a coordinate transformation relationship between the two imaging units, ensuring that the location of the thermal anomaly is accurately mapped to the corresponding location on the optical image. After registration, registered multimodal data is obtained, containing spatially aligned thermal and optical information. Finally, the registered multimodal data is analyzed. First, areas of thermal response anomaly are identified, which may indicate internal defects. These areas of thermal response anomaly are then examined for corresponding surface features in the corrected optical image, such as surface texture changes, subtle deformations, or color anomalies. By comprehensively analyzing the type of thermal anomaly (e.g., hot or cold spots), intensity, spatial distribution, and corresponding appearance in the optical image, it is possible to determine whether the thermal anomaly is caused by an internal defect or a surface feature (e.g., stain). Based on this comprehensive assessment, the presence of defects, their specific type (e.g., delamination, voids), precise location, and size are ultimately determined. This approach combines the sensitivity of thermal imaging to internal defects with the surface feature recognition capabilities of optical imaging, and enhances defect signals through controlled thermal excitation in a dynamic environment, improving detection accuracy and reliability.

[0043] The working principle of the present invention is to exploit the difference in thermal conductivity between defective areas (such as delamination and voids) and normal areas in plastic honeycomb panels. Under natural cooling conditions, the surface temperature difference caused by this difference may be very weak and decay rapidly over time. This solution artificially introduces a thermal disturbance within the panel by applying a transient, localized weak thermal excitation (such as transient cooling or heating) to the surface of the panel. Because the thermal resistance or heat capacity of the defective area is different from that of the normal area, the speed and amplitude of their response to this external thermal disturbance will also be different. For example, due to the insulation of air within an internal void, the area above it may cool more slowly than the surrounding solid area when locally cooled, thus forming a relative hot spot after the excitation; when locally heated, the temperature may rise faster than the surrounding solid area, or cool more slowly than the surrounding area, also forming a thermal anomaly. Delaminated areas may have heat conduction obstructed due to interlayer air or poor bonding, and their response to external thermal excitation will also be different from that of normal areas. The temperature response difference caused by this instantaneous excitation is short-lived, but its temperature difference is usually more significant than the steady-state temperature difference under natural cooling conditions, and is easier to be captured by high-sensitivity infrared thermal imagers. By precisely controlling the timing and position of the excitation, and synchronously acquiring high-frame-rate optical and thermal imaging data immediately after the excitation, the system can capture the thermal characteristics of the defect area after it is magnified. Subsequently, these thermal anomalies are further highlighted through data processing (such as temperature difference calculation), and are fused and aligned with the surface information provided by the optical image. Ultimately, by analyzing the fused multimodal features, it is determined whether there is a defect, its type, and its location. The optical image provides surface details, which helps to identify surface defects and accurately determine the defect boundaries; the thermal imaging data provides internal heat conduction information, which helps to detect internal defects. The combination of the two improves the comprehensiveness and robustness of detection.

[0044] In some embodiments, a plastic honeycomb panel moves on a conveyor belt at a speed of 0.5 m / s. An encoder mounted on the conveyor rollers monitors the longitudinal position of the panel in real time. When the leading edge of the panel reaches the entrance to the inspection zone, the encoder reading triggers the system to enter inspection mode. When the predetermined inspection area on the panel reaches the center of the inspection station (determined by the encoder reading), the control system activates a localized weak thermal excitation device. This device, consisting of an array of infrared lamps with adjustable power, applies a transient heat pulse of 500 watts with a duration of 100 milliseconds to the panel surface, generating a thermal excitation area approximately 100 mm x 100 mm. Subsequently, a visible light camera with a resolution of 1280 x 1024 pixels and an infrared thermal imager with a resolution of 640 x 512 pixels simultaneously capture images of this area. The infrared thermal imager continuously captures image sequences at a rate of 50 frames / s. From the infrared image sequence, the pixel-level temperature difference between the frame before thermal excitation and the tenth frame after thermal excitation is calculated to generate a temperature difference map. Images captured by the visible light camera are dedistorted using pre-calibrated camera parameters to remove radial and tangential distortion. A median filter is also applied to remove image noise. The temperature difference map is registered with the corrected visible light image by identifying pre-defined calibration points (e.g., plate edges or special markings) in both images and calculating an affine transformation matrix. After registration, hotspots are identified in the temperature difference map, where the temperature is 2°C higher than the surrounding area. Each hotspot is compared to its corresponding position in the registered visible light image. If the hotspot lacks obvious surface features (e.g., a flat surface) in the visible light image, it is considered an internal defect, such as delamination or a void. If the hotspot corresponds to a surface stain or scratch in the visible light image, it is considered a surface feature rather than an internal defect. Based on the size and shape of the hotspot and its appearance in the visible light image, the defect type is further classified (e.g., a round hotspot may be a void, an irregular hotspot may be a delamination) and the defect size is estimated. The defect location is determined using the registered coordinates.

[0045] In some embodiments, after determining, based on the position information, that the plate has reached a designated location in the detection area, the step of applying instantaneous local thermal excitation to the surface of the plate by controlling the local weak thermal excitation device to generate the thermal excitation area includes:

[0046] The temperature distribution of the plate surface is monitored in real time through the temperature sensor array to generate a temperature distribution map;

[0047] Calculating the temperature gradient of each area of ​​the plate according to the temperature distribution diagram, and defining the area where the temperature gradient exceeds a first preset threshold as a temperature non-uniform area;

[0048] For areas with non-uniform temperature, the excitation intensity and duration of the local weak thermal excitation device are adaptively adjusted according to the size of the temperature gradient; the larger the temperature gradient, the smaller the excitation intensity and the shorter the excitation duration;

[0049] According to the adjusted excitation intensity and duration, the local weak thermal excitation device is controlled to apply instantaneous local thermal excitation to the temperature non-uniform area on the surface of the plate to generate a thermal excitation area.

[0050] The temperature distribution of the plate surface is monitored in real time using a temperature sensor array. A non-contact infrared temperature sensor array can be placed above the plate to continuously collect surface temperature data and integrate this data into a two-dimensional temperature distribution map. The temperature gradient of each region of the plate can be calculated by performing spatial differentiation on the temperature distribution map, for example, using the Sobel or Prewitt operator to calculate the horizontal and vertical temperature gradients and then calculating the gradient modulus. Regions where the temperature gradient exceeds a first preset threshold are identified as temperature non-uniform regions. This threshold can be set based on actual production experience or experimental data. For temperature non-uniform regions, the excitation intensity and duration are adaptively adjusted based on the magnitude of the temperature gradient. A functional relationship or lookup table can be established between the temperature gradient value and the excitation parameters (intensity and duration). The temperature gradient value is input into this relationship or table, and the corresponding excitation intensity and duration values ​​are output. A localized weak thermal excitation device is controlled to apply instantaneous local thermal excitation to the temperature non-uniform region on the plate surface. A focused infrared lamp or laser can be used as the excitation source. The excitation intensity and duration are adjusted by controlling its power and irradiation time. The scanning or positioning system is then controlled to apply the excitation to the temperature non-uniform region.

[0051] Specifically, this technical solution aims to address the impact of initial plate temperature non-uniformity on thermal excitation. After the plate is moved to a designated location within the inspection area, a temperature sensor array is first used to acquire real-time surface temperature distribution information. Based on this temperature distribution information, the temperature gradients of each region of the plate are calculated to quantify the degree of surface temperature non-uniformity. Areas with large temperature gradients indicate a dramatic initial temperature change. Directly applying a fixed-intensity thermal excitation may result in an over- or under-temperature response in these areas, affecting subsequent thermal imaging detection of defects. Therefore, this solution identifies areas where the temperature gradient exceeds a preset threshold as areas of temperature non-uniformity requiring special treatment. For these non-uniform areas, the excitation intensity and duration of the localized weak thermal excitation device are dynamically adjusted based on the magnitude of the temperature gradient. A larger temperature gradient indicates a more rapid initial temperature change in that area. To avoid overheating or masking defect signals, the applied excitation intensity and duration are reduced. Conversely, areas with smaller temperature gradients can receive relatively stronger excitation. This adaptive adjustment ensures that, even in the presence of initial temperature non-uniformity, the applied thermal excitation more effectively stimulates the thermal response difference between the defect area and the surrounding normal areas, thereby improving the sensitivity and accuracy of defect detection using infrared thermal imaging. Finally, according to the adjusted excitation parameters, the local weak thermal excitation device applies instantaneous local thermal excitation to the temperature non-uniform area on the surface of the plate, generating a thermal excitation area for subsequent infrared thermal imaging analysis.

[0052] In some specific embodiments, an array of 32x32 infrared temperature sensors can be used to collect temperature data from the plate surface at a rate of 100 frames per second. After the temperature distribution map is generated, the temperature gradient between each pixel and its adjacent pixels is calculated using the central difference method. A first preset threshold is set at 1 degree Celsius per millimeter. Areas where the calculated temperature gradient exceeds this threshold are marked as temperature non-uniform regions. For these non-uniform regions, a mapping relationship between temperature gradient values ​​and excitation parameters is established: when the temperature gradient is 1-2 degrees Celsius per millimeter, the excitation intensity is set to 100 watts and the duration is 100 milliseconds; when the temperature gradient is 2-3 degrees Celsius per millimeter, the excitation intensity is set to 80 watts and the duration is 80 milliseconds; when the temperature gradient is greater than 3 degrees Celsius per millimeter, the excitation intensity is set to 60 watts and the duration is 60 milliseconds. The localized weak thermal excitation device uses a variable-power focused infrared lamp. A two-dimensional scanning galvanometer focuses the infrared beam and rapidly scans it onto the identified temperature non-uniform areas. The infrared lamp power and irradiation time are controlled based on the calculated excitation intensity and duration. Therefore, when the initial temperature distribution of the plate is uneven, adaptive thermal excitation can be applied to enhance the thermal response signal of the defect and improve the reliability of defect detection.

[0053] In some embodiments, for a temperature non-uniform area, the step of adaptively adjusting the excitation intensity and duration of a local weak thermal excitation device according to the temperature gradient includes:

[0054] Establishing a mapping relationship model between temperature gradient and excitation parameters. The mapping relationship model is a nonlinear model and is used to characterize the dynamic influence of temperature gradient on excitation parameters. The excitation parameters include excitation intensity and excitation duration.

[0055] According to the temperature gradient value of the temperature non-uniform area, combined with the mapping relationship model, the initial excitation intensity value and the initial excitation duration value corresponding to each temperature non-uniform area are calculated;

[0056] Considering the differences in thermal properties of materials in different areas of the plastic honeycomb panel, by introducing a thermal conductivity correction factor, and based on the thermal conductivity of the materials in each temperature non-uniform area, the initial excitation intensity value and the initial excitation duration value are corrected to obtain the corrected excitation intensity value and the corrected excitation duration value;

[0057] The excitation intensity adjustment range and the excitation duration adjustment range are set, and the corrected excitation intensity value and the corrected excitation duration value are limited to the corresponding adjustment range to obtain the final excitation intensity value and the final excitation duration value for the subsequent thermal excitation process.

[0058] A mapping relationship model between temperature gradient and excitation parameters is established. This mapping relationship model is a nonlinear model used to characterize the dynamic influence of temperature gradient on excitation parameters. As a result, the thermal excitation parameters required under different initial temperature gradients can be determined more accurately, overcoming the limitations of simple linear relationships. According to the temperature gradient values ​​in the temperature non-uniform area, combined with the nonlinear model, the preliminary excitation parameter values ​​are calculated. Furthermore, a thermal conductivity correction factor is introduced, and the preliminary excitation parameters are corrected according to the thermal conductivity of the material in each area. This compensates for the influence of differences in the thermal properties of the material on thermal conduction and thermal response, and improves the adaptability of the excitation. The adjustment range of the excitation intensity and duration is set, and the corrected parameters are limited to these ranges. This ensures that the applied thermal excitation is within the effective range and avoids excessive or weak excitation.

[0059] Specifically, after the plastic honeycomb panel is transported to the detection area and confirmed to have reached the designated location, a temperature distribution map of the panel surface is first acquired via a temperature sensor array. Based on this temperature distribution map, the temperature gradient of each region of the panel is calculated, and temperature-inhomogeneous regions where the temperature gradient exceeds a first preset threshold are identified. For these temperature-inhomogeneous regions, an initial excitation intensity and initial excitation duration values ​​are calculated based on the temperature gradient values ​​of each region using a pre-established nonlinear mapping model between temperature gradients and excitation parameters. This nonlinear model reflects the complex impact of temperature gradient variations on the required excitation parameters. To account for potential differences in material thermal conductivity across different regions of the plastic honeycomb panel, a thermal conductivity correction factor is introduced. For example, this correction factor can be determined based on the ratio of the local material thermal conductivity to a reference thermal conductivity. The initial excitation parameter value is multiplied by the corresponding thermal conductivity correction factor to obtain a corrected excitation intensity and duration value. Finally, the corrected excitation parameter value is compared with a preset excitation intensity and duration adjustment range, and the parameter value is constrained within the corresponding range to obtain the final excitation intensity and duration for the local weak thermal excitation device. Based on these final parameters, the localized weak thermal excitation device applies instantaneous local thermal excitation to the temperature-inhomogeneous region, generating a thermally excited region. This adaptive adjustment process ensures that even when the initial temperature state and material properties of the plate are non-uniform, the applied thermal excitation effectively enhances the thermal response signal of the defect region, improving the accuracy of subsequent defect identification.

[0060] In some embodiments, a nonlinear mapping model between temperature gradient and excitation parameters can be established by fitting experimental data, for example, using a polynomial function or a lookup table. For example, the mapping relationship between excitation intensity E and temperature gradient G can be expressed as E = aG² + bG + c, and the mapping relationship between excitation duration D and temperature gradient G can be expressed as D = dG² + eG + f, where a, b, c, d, e, and f are coefficients determined experimentally. The thermal conductivity correction factor can be defined as k_ref / k_local, where k_ref is the thermal conductivity of a reference material and k_local is the thermal conductivity of the material in the current region. k_local can be obtained from a pre-scan or material type information. The corrected excitation intensity E_corr = E*(k_ref / k_local), and the corrected excitation duration D_corr = D*(k_ref / k_local). The excitation intensity adjustment range can be set to [E_min, E_max], and the excitation duration adjustment range can be set to [D_min, D_max]. The final excitation intensity E_final = max(E_min, min(E_max, E_corr)), and the final excitation duration D_final = max(D_min, min(D_max, D_corr)). For example, if the calculated corrected excitation intensity is 120W and the adjustment range is [50W, 100W], the final excitation intensity is limited to 100W. If the corrected excitation duration is 0.8s and the adjustment range is [0.1s, 0.5s], the final duration is limited to 0.5s. In this way, the applied thermal excitation can be finely adjusted according to the actual state of the plate, improving the effectiveness of thermal excitation in enhancing defect signals.

[0061] In some embodiments, the step of performing distortion correction and noise filtering on the optical image data to obtain corrected optical image data includes:

[0062] After using a multispectral camera as an optical imaging unit to obtain multispectral image data of the plate surface as optical image data, a spectral feature vector is constructed based on the multispectral image data by extracting the spectral reflectance value of each pixel in different bands; the multispectral image data contains spectral information of multiple bands, and each band corresponds to a different spectral reflectance;

[0063] Establish a spectral feature database of pollutants and defects, which contains the spectral reflectance range of known pollutants and defects in various bands;

[0064] Match the spectral feature vector with the spectral feature database. If the spectral feature vector matches the pollutant spectral feature, the pixel is determined to be a pollutant, otherwise it is determined to be a non-pollutant.

[0065] Distortion correction and noise filtering are performed on the optical image data corresponding to the pixel points determined to be non-pollutants to obtain corrected optical image data.

[0066] Based on the multispectral image data, for each pixel, the spectral reflectance values ​​in multiple discrete spectral bands are extracted, thereby constructing a spectral feature vector. This vector characterizes the reflectance characteristics of the pixel at different wavelengths. Furthermore, a spectral feature database is pre-established, storing the reflectance ranges or typical spectral curves of known contaminants and defects in various spectral bands. The spectral feature vector of each pixel is matched with the spectral features in the spectral feature database. This matching process can utilize spectral distance calculations, such as spectral angle or Euclidean distance, and then compare them with a preset threshold. If the similarity between the pixel's spectral feature and the contaminant's spectral feature meets the judgment criteria, the pixel is classified as a contaminant. This generates a pixel classification result, distinguishing between contaminant and non-contaminant areas in the image. Finally, distortion correction and noise filtering are performed only on the optical image data corresponding to pixels classified as non-contaminants. This can be achieved by creating a mask that marks the non-contaminant areas, and then applying the correction and filtering algorithms to the areas covered by the mask. This selective processing prevents contaminants from interfering with the image correction and filtering processes, improving processing accuracy.

[0067] Specifically, before performing distortion correction and noise filtering on optical image data, it is necessary to address the potential interference from surface contaminants in the optical image. These contaminants have specific appearances in optical images. Direct processing can result in contaminant signatures being retained or misidentified, impacting the accuracy of subsequent defect identification. This solution addresses this issue by introducing multispectral imaging and spectral signature analysis. First, a multispectral camera is used to acquire multispectral image data of the plate surface. This data contains reflectance information from the plate surface in multiple spectral bands. Based on the acquired multispectral image data, the spectral reflectance values ​​in different bands are extracted for each pixel in the image to construct a spectral signature vector. This vector represents the spectral characteristics of the material or surface covering at that pixel. Simultaneously, a spectral signature database containing the spectral signatures of known contaminants and defects is established. This database serves as a reference for identifying contaminants and defects. The spectral signature vector of each pixel is matched against the spectral signatures in the spectral signature database. By comparing the similarity between the pixel's spectral signature and the contaminant signatures in the database, a contaminant is determined for that pixel. Based on the determination results, a classification mask is generated to distinguish between contaminant and non-contaminant pixels in the image. Finally, distortion correction and noise filtering are performed only on the optical image data corresponding to pixels determined to be non-contaminants. This excludes data from the contaminated areas from the correction and filtering process, ensuring that the processing primarily affects the sheet material itself. This improves the accuracy of the corrected optical image data and provides more reliable foundational data for subsequent registration and defect analysis.

[0068] In some specific embodiments, a multispectral camera with 12 spectral bands can be used, covering the visible to near-infrared band, for example, from 450 nm to 950 nm, with band spacing of approximately 40 nm. In a laboratory environment, multispectral images of known contaminants, such as oil droplets, dust particles, and defect-free plate surfaces, are collected, and their spectral reflectance curves are extracted to establish a database of contaminant spectral signatures. The database stores the average reflectance values ​​and standard deviations for each contaminant type in the 12 bands. During online inspection, the multispectral camera captures images of the plate. For each pixel in the image, its reflectance values ​​in the 12 bands are extracted to construct a 12-dimensional spectral signature vector. This spectral signature vector is then compared with the contaminant spectral signatures in the database. For example, the spectral angle between the pixel spectral vector and the average spectral vector for each contaminant in the database is calculated. If the calculated spectral angle is less than a preset threshold, for example, 0.1 radians, the pixel is determined to be a contaminant of the corresponding type. This generates a binary mask image, in which a pixel value of 1 indicates a non-contaminant and a pixel value of 0 indicates a contaminant. Next, a distortion correction algorithm, for example, based on pre-calibrated camera parameters, and a noise filtering algorithm, such as a median filter or a Gaussian filter, are applied to the raw optical image data. These algorithms process only those regions within the mask image where the pixel value is 1. For example, when calculating the filter window, only pixels within the window where the mask value is 1 are considered. This leaves the pixel values ​​in the contaminated areas unaffected by the correction and filtering, resulting in corrected optical image data free of contaminant interference.

[0069] In some embodiments, the step of registering the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain registered multimodal data includes:

[0070] Establish the correspondence between the optical image coordinate system and the infrared thermal image coordinate system, calculate the coordinate transformation matrix by calibrating the coordinates of the characteristic points of the plate in the optical image and the infrared thermal image, and use this coordinate transformation matrix to transform the infrared thermal image coordinates into the optical image coordinate system;

[0071] Extract the center coordinates of the abnormal thermal response area based on the temperature difference data or the temperature response characteristic data, and transform the center coordinates of the abnormal thermal response area into the optical image coordinate system using the coordinate transformation matrix to obtain the center coordinates of the abnormal thermal response area after registration;

[0072] For the corrected optical image data, the key point detection algorithm of SIFT features is used to extract the image key points;

[0073] Taking the center coordinate of the thermal response anomaly area after registration as the center, search for the N key points closest to the center coordinate among all the key points extracted, and calculate the distance and angle between each key point and the center coordinate;

[0074] According to the calculated distances and angles, N key point spatial relationship feature vectors are constructed, and the spatial relationship feature vectors are fused with the temperature difference data or temperature response feature data to obtain the registered multimodal data.

[0075] Establish the correspondence between the optical image coordinate system and the infrared thermal image coordinate system, and calculate the coordinate transformation matrix through the coordinates of the feature points of the calibration plate in the optical image and the infrared thermal image. The coordinate transformation matrix is ​​used to transform the infrared thermal image coordinates into the optical image coordinate system. In specific implementation, a calibration plate containing known geometric features (for example, a checkerboard pattern) can be used. The calibration plate is placed in the detection area, and its image is synchronously acquired using the optical imaging unit and the infrared thermal imaging unit. In the acquired optical image and infrared thermal image, the pixel coordinates of the corresponding feature points (for example, checkerboard corners) on the calibration plate are detected and extracted. Using these corresponding point pairs, a coordinate transformation matrix, such as an affine transformation matrix or a perspective transformation matrix, can be calculated. This matrix can map the pixel coordinates in the infrared thermal image to the corresponding positions in the optical image coordinate system.

[0076] Specifically, this method first establishes a correspondence between the optical image coordinate system and the infrared thermal image coordinate system. Using the coordinates of the characteristic points of the calibration plate in both the optical and infrared thermal images, a coordinate transformation matrix is ​​calculated. This coordinate transformation matrix is ​​used to transform the infrared thermal image coordinates into the optical image coordinate system. This resolves the coordinate system inconsistency between the two imaging modalities. Next, the center coordinates of the thermal response anomaly region are extracted based on the temperature difference data or temperature response feature data. Using the previously calculated coordinate transformation matrix, the center coordinates of the thermal response anomaly region are transformed into the optical image coordinate system, resulting in the center coordinates of the registered thermal response anomaly region. This step determines the approximate location of the thermal anomaly region in the optical image. The SIFT feature-based keypoint detection algorithm is used to extract image keypoints from the corrected optical image data. The SIFT algorithm detects local feature points in an image that are scale- and rotation-invariant. Using the center coordinates of the registered thermal response anomaly region as the center, the N closest keypoints are searched for among all extracted keypoints. The distance and angle between each keypoint and the center coordinate are calculated. This step focuses on the adjacent areas of the thermal anomaly on the optical image, screens out optical feature points that may be related to the thermal anomaly, and describes their spatial distribution. Based on the calculated distances and angles, N key point spatial relationship feature vectors are constructed. Finally, the spatial relationship feature vector is fused with the temperature difference data or temperature response feature data to obtain the registered multimodal data. This fusion method combines the intensity or feature information of the thermal anomaly with the spatial structure of the local area around it in the optical image, forming a multimodal data representation that contains thermal information and optical spatial structure information, providing richer information for subsequent identification.

[0077] In some specific embodiments, a calibration plate containing a 10x10 checkerboard pattern can be used for coordinate system calibration. The optical imaging unit and the infrared thermal imaging unit synchronously capture the calibration plate image. The 9x9 inner corner points of the checkerboard are detected in the optical image and the infrared thermal image respectively to obtain two sets of corresponding pixel coordinates. Using these corresponding point pairs, a 3x3 perspective transformation matrix is ​​calculated as the coordinate transformation matrix. When an abnormal thermal response area is detected, for example, the pixel coordinates of its center in the infrared thermal image coordinate system are (x_t, y_t), then the calculated perspective transformation matrix is ​​used to convert it to the optical image coordinate system to obtain the aligned center coordinates (x_o, y_o). Run the SIFT algorithm on the corrected optical image to extract all key points. With (x_o, y_o) as the center, calculate the Euclidean distance of all SIFT key points to the center. Select the closest N=20 key points. For each of these 20 keypoints, calculate its distance from (x_o, y_o) and the angle between the vector pointing from (x_o, y_o) to that keypoint and the horizontal. These 20 (distance, angle) pairs are sorted by distance to form a 40-dimensional spatial relationship feature vector. This 40-dimensional vector is concatenated or combined with the temperature difference value (e.g., 5 degrees Celsius) or temperature response feature value (e.g., peak temperature rise rate) of the thermal response anomaly area to form the final registered multimodal data.

[0078] In some embodiments, the steps of searching for N key points closest to the center coordinates of the registered thermal response abnormal region among all extracted key points, and calculating the distance and angle between each key point and the center coordinates include:

[0079] Calculate the descriptor of each key point and compare its similarity with the descriptor of the center coordinates of the thermal response abnormal area after registration;

[0080] Based on the comparison results, key points with similarity less than a second preset threshold are removed to improve the accuracy of key point screening;

[0081] Taking the center coordinate of the thermal response anomaly area after registration as the center, search for the N key points closest to the center coordinate among all the retained key points, and calculate the distance and angle between each key point and the center coordinate.

[0082] The descriptor of each key point is calculated to characterize the local image features of the area around the key point. The descriptor is a vector representation that captures the texture, gradient and other information of the neighborhood of the key point. The descriptor of each key point is compared with the descriptor corresponding to the center coordinate of the thermal response anomaly area after registration for similarity, and the degree of correlation between the key point and the center area in local features is quantified. According to the comparison results, the key points with a similarity less than the second preset threshold are eliminated, and key point filtering based on feature similarity is achieved. In this way, key points that may be close in space but have irrelevant image features are removed. In the filtered key point set, with the center coordinate of the thermal response anomaly area after registration as the center, the N key points closest to the center coordinate are searched, and the distance and angle between these key points and the center coordinate are calculated to construct the spatial relationship feature vector.

[0083] Specifically, this technical solution improves the keypoint selection process for constructing spatial relationship feature vectors, aiming to address the problem that periodic texture or random noise on the surface of plastic honeycomb panels can cause the keypoint detection algorithm to extract a large number of mismatched points, affecting the accuracy of the spatial relationship feature vector. First, a descriptor is calculated for each keypoint extracted from the optical image. Simultaneously, the descriptor of the optical image region corresponding to the center coordinates of the thermal response anomaly region after registration is obtained. Next, the similarity between the descriptor of each keypoint and the descriptor of the center region is calculated. For example, the Euclidean distance or cosine similarity between the descriptor vectors can be used as a similarity metric. A second preset threshold is then set. Keypoints with a similarity below this threshold are removed from the keypoint set. These removed keypoints are considered uncorrelated in image features with the center of the thermal response anomaly region, even if they may be spatially close. For the remaining set of keypoints that share a certain degree of feature similarity with the center region, the spatial distance and angle between each retained keypoint and the reference point are calculated, using the center coordinates of the thermal response anomaly region after registration as the reference point. Finally, the N keypoints closest to the reference point are selected from these retained keypoints. By introducing a filtering step based on descriptor similarity, the final N key points selected are not only spatially close to the center of the thermally abnormal region but also correlated with local image features. The resulting spatial relationship feature vector more accurately reflects the local structure and texture information of the thermally abnormal region in the optical image, improving the quality of the registered multimodal data and thus enhancing the accuracy and robustness of subsequent defect recognition.

[0084] In some specific embodiments, the SIFT algorithm is used to calculate the descriptor of each key point. The SIFT descriptor of the optical image area corresponding to the center coordinate of the abnormal thermal response area after registration is also calculated. The Euclidean distance between SIFT descriptors is used as a similarity measure. The second preset threshold is set as the maximum allowable value of the Euclidean distance of the descriptor, for example, it is set to a certain percentage of the maximum possible Euclidean distance of the descriptor. The Euclidean distance between each key point descriptor and the central area descriptor is calculated. If the distance is greater than the second preset threshold, the key point is eliminated. Among the remaining key points, the Euclidean distance of each key point to the center coordinate and the polar angle relative to the center coordinate are calculated. Select the N key points with the closest distance, for example, N is set to 10. These selected key points and their distances and angles from the center coordinate are used to construct a spatial relationship feature vector.

[0085] In some embodiments, the step of analyzing the registered multimodal data to determine whether a defect exists, the defect type, the defect location, and the defect size includes:

[0086] Analyze the registered multimodal data, extract the temperature gradient of the thermal response abnormal area and the texture features of the optical image, calculate the mutual correlation coefficient between the temperature gradient and the texture features, and obtain the mutual correlation coefficient map;

[0087] According to the cross-correlation coefficient map, an adaptive threshold segmentation algorithm is used to segment the areas with high cross-correlation coefficients as candidate defect areas;

[0088] For the candidate defect area, the shape features, size features and position features are extracted to construct a defect feature vector, which is then input into a pre-trained defect classifier to obtain the defect type, defect location and defect size prediction results.

[0089] When analyzing the registered multimodal data, a temperature gradient is extracted from the area with abnormal thermal response. This gradient quantifies the spatial rate of change of the thermal anomaly. Texture features are extracted from the optical image. These features describe the grayscale changes or spatial structure of the local area of ​​the image. The cross-correlation coefficient between the temperature gradient and the texture features is calculated to generate a cross-correlation coefficient map, which reflects the degree of spatial correlation between the thermal anomaly and the surface texture. Regions with high cross-correlation coefficients indicate the possibility of the simultaneous occurrence of thermal anomalies and specific surface textures. An adaptive threshold segmentation algorithm is used to process the cross-correlation coefficient map. The segmentation threshold is dynamically determined based on the statistical characteristics of the map, and regions with cross-correlation coefficients above the threshold are identified as candidate defect regions. For the candidate defect regions, their geometric attributes, including shape, size, and spatial position, are extracted and constructed into a vector. This vector is input into a pre-trained classifier, which analyzes the vector based on the learned pattern and outputs the defect category, spatial coordinates, and physical size.

[0090] Specifically, during the inspection process of plastic honeycomb panels after hot pressing, internal or surface defects may occur. These defects manifest as dynamically changing thermal and optical signatures during the cooling process. To accurately identify these defects, the collected, registered multimodal data is first analyzed. Temperature gradient information is extracted from regions with thermally anomalous responses, capturing the intensity and diffusion characteristics of the thermal anomalies. Simultaneously, texture features are extracted from the registered optical images to describe the local morphology or structure of the panel surface. A cross-correlation coefficient map is generated by calculating the cross-correlation coefficient between the temperature gradient and texture features. This map combines information from the thermal and optical data, highlighting areas where thermal anomalies and surface texture coexist—potential defect locations. Regions with high cross-correlation coefficients indicate a correlation between thermal anomalies and specific surface textures, improving the reliability of defect identification. Next, an adaptive threshold segmentation algorithm is applied to the cross-correlation coefficient map. Based on the distribution characteristics of the cross-correlation coefficient, a segmentation threshold is automatically determined. Regions with cross-correlation coefficients exceeding the threshold are segmented to identify candidate defect regions. This adaptive segmentation method adapts to variations in cross-correlation coefficients caused by different defect types and environmental factors, improving segmentation accuracy. Finally, for the segmented candidate defect regions, geometric features such as shape, size, and location are extracted and constructed into a feature vector. This feature vector is then fed into a pre-trained defect classifier. Based on the input feature vector, the classifier outputs the predicted defect type, defect location on the sheet, and physical dimensions. This multimodal data fusion, feature extraction, region segmentation, and machine learning classification method achieves automated and accurate identification and classification of hot-pressing defects in plastic honeycomb panels, overcoming the limitations of single-modality detection and improving detection robustness.

[0091] In some specific embodiments, when analyzing the multimodal data after registration, the temperature gradient along a specific direction in the thermal response abnormality area, such as the direction of the maximum temperature gradient, can be calculated. The texture features of the optical image can be extracted using a local binary pattern (LBP) or a Gabor filter bank. The pixel-level mutual correlation coefficient between the temperature gradient map and the texture feature map is calculated to generate a mutual correlation coefficient map. For example, each pixel value in the mutual correlation coefficient map represents the correlation between the temperature gradient value and the texture feature value at the corresponding position. The mutual correlation coefficient map is segmented using the Otsu method or an adaptive threshold segmentation algorithm based on local window statistics, and the minimum connected area threshold is set, and areas smaller than the threshold are eliminated. For the retained candidate defect areas, their centroid coordinates are extracted as position features, and their area and perimeter are calculated as size and shape features. These features are combined into a vector and input into a support vector machine (SVM) or convolutional neural network (CNN) classifier that has been pre-trained using known defect samples to obtain the classification results and position and size information of the defects. For example, a candidate region has a high correlation coefficient, an area of ​​10 square millimeters, a shape close to a circle, and a centroid coordinate of (x, y). After these features are input into the classifier, the classifier outputs the prediction result as "internal void", the position is (x, y), and the size is about 3.5 mm in diameter.

[0092] Furthermore, the threshold of the adaptive threshold segmentation algorithm is dynamically adjusted according to the mean and standard deviation of the cross-correlation coefficient map, and the candidate defect areas are segmented to meet the minimum area threshold.

[0093] This technical solution improves the accuracy of identifying candidate defect areas by improving the adaptive threshold segmentation algorithm. First, the segmentation threshold is no longer a fixed value or based only on local information, but is dynamically adjusted according to the global statistical characteristics (mean and standard deviation) of the entire cross-correlation coefficient map. This adjustment enables the threshold to adapt to the differences in the distribution of cross-correlation coefficients generated by different images or different batches of plates, thereby improving the robustness of the segmentation. Secondly, after the preliminary segmentation is completed, the area of ​​the connected areas obtained by the segmentation is checked, and only areas with an area greater than or equal to the preset minimum area threshold are retained as candidate defect areas. This step effectively filters out areas with too small an area, which are usually caused by noise or non-defect factors. By setting the minimum area requirement, false alarms are reduced and the reliability of the candidate areas is improved. The dynamic adjustment of the threshold and the setting of the minimum area threshold work together to make the candidate defect areas extracted from the cross-correlation coefficient map more accurately correspond to the actual potential defect locations.

[0094] Specifically, after analyzing the registered multimodal data, the temperature gradients and texture features of the optical image in the thermally abnormal region are extracted. The cross-correlation coefficients between the temperature gradients and texture features are calculated to obtain a cross-correlation coefficient map. This cross-correlation coefficient map is then processed using an adaptive threshold segmentation algorithm. The threshold of this adaptive threshold segmentation algorithm is calculated based on the overall mean and standard deviation of the current cross-correlation coefficient map. For example, the threshold can be set to the mean plus a multiple of the standard deviation. This allows the threshold to be dynamically adjusted based on the overall distribution of the image content, avoiding the over-segmentation or under-segmentation issues that can result from using a fixed threshold. After initial segmentation using this dynamically adjusted threshold, a series of connected regions are obtained. The areas of these connected regions are then calculated and compared to a preset minimum area threshold. Regions with areas smaller than the minimum area threshold are identified as noise or non-defect regions and are eliminated. Only regions with areas greater than or equal to the minimum area threshold are retained as final candidate defect regions. By combining dynamic threshold adjustment with minimum area filtering, this method can more accurately identify regions associated with potential defects from the cross-correlation coefficient map, reducing false positives caused by noise or minor local variations and improving the accuracy of subsequent defect assessment.

[0095] Furthermore, the defect classifier is constructed using a support vector machine or a convolutional neural network, and the output result of the defect classifier is subjected to confidence evaluation, and the prediction result is valid if the confidence is higher than a third preset threshold.

[0096] The classifier is constructed using a support vector machine (SVM) or a convolutional neural network (CNN). SVMs achieve classification by constructing hyperplanes. Convolutional neural networks extract and learn features through a multi-layered network structure. The classifier receives a defect feature vector as input and outputs a prediction result, including defect type, location, and size. The classifier output undergoes a confidence assessment. This assessment determines the model's degree of certainty in the prediction result. The confidence score is then compared with a third preset threshold. If the confidence score exceeds the third preset threshold, the prediction result is deemed valid.

[0097] Specifically, to address potential concerns about classifier output, a defect classifier is constructed using a support vector machine or convolutional neural network. This classifier receives a defect feature vector and outputs a predicted defect type, location, and size. To ensure the credibility of the prediction, the classifier output undergoes a confidence assessment. The confidence assessment quantifies the model's degree of certainty in the prediction. This confidence is then compared to a pre-set third threshold. Only when the confidence exceeds the third threshold is the prediction accepted and used to determine the presence, type, location, and size of a defect. This reduces the risk of misjudgment and improves the accuracy of the identification process.

[0098] In some specific embodiments, a convolutional neural network is used as a defect classifier. After training, the network can predict defect types such as delamination, voids, and cracks based on the input defect feature vector and output the probability value of the corresponding category. For example, for an input feature vector, the network outputs the probability of each defect type as follows: delamination 0.92, void 0.05, and crack 0.03. In this case, the prediction result is delamination, and the confidence level is 0.92. If the third preset threshold is set to 0.90, the predicted delamination result is judged to be valid. If the network outputs the probability of delamination as 0.85, which is lower than the threshold of 0.90, the prediction result is judged to be invalid and is not used for the final defect judgment.

[0099] Reference Attachment Figure 2 The present invention provides a plastic honeycomb panel hot pressing defect identification system, comprising:

[0100] The first control module 100 is used to control the plate conveying device to convey the plastic honeycomb plate to the detection area, monitor the plate position in real time through a position sensor or encoder, and generate position information;

[0101] The thermal excitation module 200 is used to determine, based on the position information, that the plate has reached the designated position in the detection area, and then control the local weak thermal excitation device to apply instantaneous local thermal excitation to the surface of the plate to generate a thermal excitation area;

[0102] The second control module 300 is used to control the optical imaging unit and the infrared thermal imaging unit to synchronously capture the optical image and the infrared thermal image of the plate in the thermal excitation area to obtain optical image data and infrared thermal image data;

[0103] The acquisition module 400 is used to obtain temperature difference data by calculating the temperature difference map of the thermal actuation area before and after thermal actuation based on the infrared thermal image data, or to obtain temperature response characteristic data by analyzing the temperature response curve of the thermal actuation area during the thermal actuation process;

[0104] The correction module 500 is used to perform distortion correction and noise filtering on the optical image data to obtain corrected optical image data;

[0105] A registration module 600 is used to register the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain registered multimodal data;

[0106] The analysis module 700 is used to determine whether there is a defect, the defect type, the defect location and the defect size by analyzing the registered multimodal data.

[0107] In some embodiments, after the thermal actuation module 200 determines that the plate has reached a designated location in the detection area based on the position information, it controls the local weak thermal actuation device to apply instantaneous local thermal actuation to the surface of the plate to generate the thermal actuation area, thereby executing:

[0108] The temperature distribution of the plate surface is monitored in real time through the temperature sensor array to generate a temperature distribution map;

[0109] Calculating the temperature gradient of each area of ​​the plate according to the temperature distribution diagram, and defining the area where the temperature gradient exceeds a first preset threshold as a temperature non-uniform area;

[0110] For areas with non-uniform temperature, the excitation intensity and duration of the local weak thermal excitation device are adaptively adjusted according to the size of the temperature gradient; the larger the temperature gradient, the smaller the excitation intensity and the shorter the excitation duration;

[0111] According to the adjusted excitation intensity and duration, the local weak thermal excitation device is controlled to apply instantaneous local thermal excitation to the temperature non-uniform area on the surface of the plate to generate a thermal excitation area.

[0112] In some embodiments, the thermal actuation module 200 performs the following when used to adaptively adjust the actuation intensity and duration of a local weak thermal actuation device according to the temperature gradient in a temperature-nonuniform region:

[0113] Establishing a mapping relationship model between temperature gradient and excitation parameters. The mapping relationship model is a nonlinear model and is used to characterize the dynamic influence of temperature gradient on excitation parameters. The excitation parameters include excitation intensity and excitation duration.

[0114] According to the temperature gradient value of the temperature non-uniform area, combined with the mapping relationship model, the initial excitation intensity value and the initial excitation duration value corresponding to each temperature non-uniform area are calculated;

[0115] By introducing a thermal conductivity correction factor and correcting the initial excitation intensity value and the initial excitation duration value according to the material thermal conductivity of each temperature non-uniform area, a corrected excitation intensity value and a corrected excitation duration value are obtained;

[0116] The excitation intensity adjustment range and the excitation duration adjustment range are set, and the corrected excitation intensity value and the corrected excitation duration value are limited within the corresponding adjustment range to obtain the final excitation intensity value and the final excitation duration value.

[0117] In some embodiments, when the correction module 500 is used to perform distortion correction and noise filtering on the optical image data to obtain the corrected optical image data, it executes:

[0118] After using a multispectral camera as an optical imaging unit to obtain multispectral image data of the plate surface as optical image data, a spectral feature vector is constructed based on the multispectral image data by extracting the spectral reflectance value of each pixel in different bands; the multispectral image data contains spectral information of multiple bands, and each band corresponds to a different spectral reflectance;

[0119] Establish a spectral feature database of pollutants and defects, which contains the spectral reflectance range of known pollutants and defects in various bands;

[0120] Match the spectral feature vector with the spectral feature database. If the spectral feature vector matches the pollutant spectral feature, the pixel is determined to be a pollutant, otherwise it is determined to be a non-pollutant.

[0121] Distortion correction and noise filtering are performed on the optical image data corresponding to the pixel points determined to be non-pollutants to obtain corrected optical image data.

[0122] In some embodiments, the registration module 600 performs the following when registering the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain the registered multimodal data:

[0123] Establish the correspondence between the optical image coordinate system and the infrared thermal image coordinate system, calculate the coordinate transformation matrix by calibrating the coordinates of the characteristic points of the plate in the optical image and the infrared thermal image, and use this coordinate transformation matrix to transform the infrared thermal image coordinates into the optical image coordinate system;

[0124] Extract the center coordinates of the abnormal thermal response area based on the temperature difference data or the temperature response characteristic data, and transform the center coordinates of the abnormal thermal response area into the optical image coordinate system using the coordinate transformation matrix to obtain the center coordinates of the abnormal thermal response area after registration;

[0125] For the corrected optical image data, the key point detection algorithm of SIFT features is used to extract the image key points;

[0126] Taking the center coordinate of the thermal response anomaly area after registration as the center, search for the N key points closest to the center coordinate among all the key points extracted, and calculate the distance and angle between each key point and the center coordinate;

[0127] According to the calculated distances and angles, N key point spatial relationship feature vectors are constructed, and the spatial relationship feature vectors are fused with the temperature difference data or temperature response feature data to obtain the registered multimodal data.

[0128] In some embodiments, the registration module 600 is used to search for the N key points closest to the center coordinate of the registered thermal response abnormal region among all extracted key points, and calculate the distance and angle between each key point and the center coordinate when performing the following operations:

[0129] Calculate the descriptor of each key point and compare its similarity with the descriptor of the center coordinates of the thermal response abnormal area after registration;

[0130] According to the comparison results, key points with a similarity less than a second preset threshold are removed;

[0131] Taking the center coordinate of the thermal response anomaly area after registration as the center, search for the N key points closest to the center coordinate among all the retained key points, and calculate the distance and angle between each key point and the center coordinate.

[0132] In some embodiments, the analysis module 700 is configured to determine the presence of a defect, the defect type, the defect location, and the defect size by analyzing the registered multimodal data:

[0133] Analyze the registered multimodal data, extract the temperature gradient of the thermal response abnormal area and the texture features of the optical image, calculate the mutual correlation coefficient between the temperature gradient and the texture features, and obtain the mutual correlation coefficient map;

[0134] According to the cross-correlation coefficient map, an adaptive threshold segmentation algorithm is used to segment the areas with high cross-correlation coefficients as candidate defect areas;

[0135] For the candidate defect area, the shape features, size features and position features are extracted to construct a defect feature vector, which is then input into a pre-trained defect classifier to obtain the defect type, defect location and defect size prediction results.

[0136] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any actual relationship or order between these entities or operations.

[0137] The foregoing description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for identifying hot pressing defects of plastic honeycomb panels, characterized in that: The following steps are involved: After the plate conveying device is controlled to convey the plastic honeycomb plate to the detection area, the position of the plate is monitored in real time through a position sensor or encoder to generate position information; According to the position information, after determining that the plate has reached the designated position in the detection area, the local weak thermal excitation device is controlled to apply instantaneous local thermal excitation to the surface of the plate to generate a thermal excitation area; Controlling the optical imaging unit and the infrared thermal imaging unit to synchronously collect the optical image and the infrared thermal image of the plate in the thermal excitation area to obtain the optical image data and the infrared thermal image data; Based on the infrared thermal image data, the temperature difference data is obtained by calculating the temperature difference map of the thermal excitation area before and after thermal excitation, or the temperature response characteristic data is obtained by analyzing the temperature response curve of the thermal excitation area during the thermal excitation process; Performing distortion correction and noise filtering on the optical image data to obtain corrected optical image data; registering the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain registered multimodal data; By analyzing the registered multimodal data, we can determine whether there are defects, the defect type, defect location and defect size.

2. The method for identifying hot pressing defects of plastic honeycomb panels according to claim 1, characterized in that: After determining that the plate has reached the designated position in the detection area based on the position information, the steps of applying instantaneous local thermal excitation to the surface of the plate by controlling the local weak thermal excitation device to generate the thermal excitation area include: The temperature distribution of the plate surface is monitored in real time through the temperature sensor array to generate a temperature distribution map; Calculating the temperature gradient of each area of ​​the plate according to the temperature distribution diagram, and defining the area where the temperature gradient exceeds a first preset threshold as a temperature non-uniform area; For areas with non-uniform temperature, the excitation intensity and duration of the local weak thermal excitation device are adaptively adjusted according to the size of the temperature gradient; the larger the temperature gradient, the smaller the excitation intensity and the shorter the excitation duration; According to the adjusted excitation intensity and duration, the local weak thermal excitation device is controlled to apply instantaneous local thermal excitation to the temperature non-uniform area on the surface of the plate to generate a thermal excitation area.

3. The method for identifying hot pressing defects of plastic honeycomb panels according to claim 2, characterized in that: For the temperature non-uniform area, the steps of adaptively adjusting the excitation intensity and duration of the local weak thermal excitation device according to the temperature gradient include: Establishing a mapping relationship model between temperature gradient and excitation parameters. The mapping relationship model is a nonlinear model and is used to characterize the dynamic influence of temperature gradient on excitation parameters. The excitation parameters include excitation intensity and excitation duration. According to the temperature gradient value of the temperature non-uniform area, combined with the mapping relationship model, the initial excitation intensity value and the initial excitation duration value corresponding to each temperature non-uniform area are calculated; By introducing a thermal conductivity correction factor and correcting the initial excitation intensity value and the initial excitation duration value according to the material thermal conductivity of each temperature non-uniform area, a corrected excitation intensity value and a corrected excitation duration value are obtained; The excitation intensity adjustment range and the excitation duration adjustment range are set, and the corrected excitation intensity value and the corrected excitation duration value are limited within the corresponding adjustment range to obtain the final excitation intensity value and the final excitation duration value.

4. The method for identifying hot pressing defects of plastic honeycomb panels according to claim 1, characterized in that: The steps of performing distortion correction and noise filtering on the optical image data to obtain corrected optical image data include: After using a multispectral camera as an optical imaging unit to obtain multispectral image data of the plate surface as optical image data, a spectral feature vector is constructed based on the multispectral image data by extracting the spectral reflectance value of each pixel in different bands; the multispectral image data contains spectral information of multiple bands, and each band corresponds to a different spectral reflectance; Establish a spectral feature database of pollutants and defects, which contains the spectral reflectance range of known pollutants and defects in various bands; Match the spectral feature vector with the spectral feature database. If the spectral feature vector matches the pollutant spectral feature, the pixel is determined to be a pollutant, otherwise it is determined to be a non-pollutant. Distortion correction and noise filtering are performed on the optical image data corresponding to the pixel points determined to be non-pollutants to obtain corrected optical image data.

5. The method for identifying hot pressing defects of plastic honeycomb panels according to claim 1, characterized in that: The step of registering the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain registered multimodal data includes: Establish the correspondence between the optical image coordinate system and the infrared thermal image coordinate system, calculate the coordinate transformation matrix by calibrating the coordinates of the characteristic points of the plate in the optical image and the infrared thermal image, and use this coordinate transformation matrix to transform the infrared thermal image coordinates into the optical image coordinate system; Extract the center coordinates of the abnormal thermal response area based on the temperature difference data or the temperature response characteristic data, and transform the center coordinates of the abnormal thermal response area into the optical image coordinate system using the coordinate transformation matrix to obtain the center coordinates of the abnormal thermal response area after registration; For the corrected optical image data, the key point detection algorithm of SIFT features is used to extract the image key points; Taking the center coordinate of the thermal response anomaly area after registration as the center, search for the N key points closest to the center coordinate among all the key points extracted, and calculate the distance and angle between each key point and the center coordinate; According to the calculated distances and angles, N key point spatial relationship feature vectors are constructed, and the spatial relationship feature vectors are fused with the temperature difference data or temperature response feature data to obtain the registered multimodal data.

6. The method for identifying hot pressing defects of plastic honeycomb panels according to claim 5, characterized in that: Taking the center coordinate of the thermal response anomaly area after registration as the center, searching for the N key points closest to the center coordinate among all the extracted key points, and calculating the distance and angle between each key point and the center coordinate include: Calculate the descriptor of each key point and compare its similarity with the descriptor of the center coordinates of the thermal response abnormal area after registration; According to the comparison results, key points with a similarity less than a second preset threshold are removed; Taking the center coordinate of the thermal response anomaly area after registration as the center, search for the N key points closest to the center coordinate among all the retained key points, and calculate the distance and angle between each key point and the center coordinate.

7. The method for identifying hot pressing defects of plastic honeycomb panels according to claim 1, characterized in that: The steps of analyzing the registered multimodal data to determine whether there is a defect, the defect type, the defect location, and the defect size include: Analyze the registered multimodal data, extract the temperature gradient of the thermal response abnormal area and the texture features of the optical image, calculate the mutual correlation coefficient between the temperature gradient and the texture features, and obtain the mutual correlation coefficient map; According to the cross-correlation coefficient map, an adaptive threshold segmentation algorithm is used to segment the areas with high cross-correlation coefficients as candidate defect areas; For the candidate defect area, the shape features, size features and position features are extracted to construct a defect feature vector, which is then input into a pre-trained defect classifier to obtain the defect type, defect location and defect size prediction results.

8. The method for identifying hot pressing defects of plastic honeycomb panels according to claim 7, characterized in that: The threshold of the adaptive threshold segmentation algorithm is dynamically adjusted according to the mean and standard deviation of the cross-correlation coefficient map, and the candidate defect area is segmented to meet the minimum area threshold.

9. The method for identifying hot pressing defects of plastic honeycomb panels according to claim 7, characterized in that: The defect classifier is constructed using support vector machines or convolutional neural networks.

10. A plastic honeycomb panel hot pressing defect recognition system, characterized in that: include: The first control module is used to control the plate conveying device to convey the plastic honeycomb plate to the detection area, monitor the position of the plate in real time through a position sensor or encoder, and generate position information; The thermal excitation module is used to determine that the plate has reached the designated position in the detection area based on the position information, and then control the local weak thermal excitation device to apply instantaneous local thermal excitation to the surface of the plate to generate a thermal excitation area; The second control module is used to control the optical imaging unit and the infrared thermal imaging unit to synchronously capture the optical image and the infrared thermal image of the plate in the thermal excitation area to obtain optical image data and infrared thermal image data; An acquisition module is used to obtain temperature difference data by calculating a temperature difference map of a thermally actuated area before and after thermal actuation based on infrared thermal image data, or to obtain temperature response characteristic data by analyzing a temperature response curve of the thermally actuated area during thermal actuation; A correction module, used to perform distortion correction and noise filtering on the optical image data to obtain corrected optical image data; A registration module, used to register the temperature difference data or the temperature response characteristic data with the corrected optical image data to obtain registered multimodal data; The analysis module is used to determine whether there are defects, the defect type, defect location and defect size by analyzing the registered multimodal data.

Images