Method and system for detecting infrared thermal imaging defects of carbon fiber plate in real time

Abstract

The invention relates to the technical field of nondestructive testing, in particular to a carbon fiber plate infrared thermal imaging defect real-time detection method and system, and the method comprises the following steps: constructing an improved YOLOv9c target detection network model; performing offline training on the improved network model, and deploying the trained network model in a reasoning module; continuously scanning the surface of the carbon fiber plate by utilizing a line laser source, and acquiring a temperature image sequence of a scanning area by adopting an infrared thermal imager; the method comprises the following steps: constructing a dynamic background model through frame averaging based on multi-frame thermal image data in the initial stage of scanning, and performing frame difference processing on real-time thermal image data acquired subsequently to obtain a difference thermal image; the differential thermogram is input into a reasoning module for defect detection, a defect position bounding box and category confidence are output, and real-time defect detection and tracking are achieved based on continuous multi-frame detection results; the problems that an existing carbon fiber detection method cannot carry out industrial rapid detection, the automation degree is low, and the real-time performance is insufficient are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nondestructive testing, in particular to a real-time detection method and system for infrared thermal imaging defects of carbon fiber plates. BACKGROUND

[0002] Carbon fiber reinforced composites are widely used in key fields such as aerospace, rail transportation, wind turbine blades, and automobile structures due to their excellent specific strength, corrosion resistance, and good fatigue performance. However, during processing and service, carbon fiber plates are prone to defects such as delamination, voids, cracks, and debonding. These defects are often difficult to detect with the naked eye in the early stages, but they can significantly reduce the mechanical properties and service life of the material, posing a significant safety hazard.

[0003] Currently, there are two main methods for detecting defects in carbon fiber plates using infrared thermal imaging: one is fixed excitation position detection, which has limited detection area and low efficiency, and cannot meet the demand for large-area rapid detection; the other is scanning detection, which can achieve large-area coverage and high detection speed through a thermal imager, but the analysis process still relies mainly on manual tracking and post-processing, making it difficult to achieve automated real-time detection and positioning of defects. These two methods are mostly limited to laboratory conditions and cannot form an automatic closed loop on actual production lines.

[0004] Therefore, it is necessary to propose a real-time detection method and system for infrared thermal imaging defects of carbon fiber plates to solve the above problems. SUMMARY

[0005] The present application provides a real-time detection method and system for infrared thermal imaging defects of carbon fiber plates to solve the problems of low automation and insufficient real-time performance of existing carbon fiber detection methods.

[0006] The present application is implemented by using the following technical solutions: A real-time detection method for infrared thermal imaging defects of carbon fiber plates, comprising the following steps: S1: Construct an improved YOLOv9c target detection network model; wherein the improved YOLOv9c target detection network model is constructed by integrating a sparse space-channel hybrid attention module at the front end of the backbone network of the YOLOv9c target detection network, and the sparse space-channel hybrid attention module generates QKV features using convolution kernels and introduces a sparsity regularization loss function after the softmax weight output to enhance feature sparsity and noise resistance; S2: Use labeled thermal image data collected under multiple excitation intensities, scanning directions, and environmental temperature conditions to offline train the improved YOLOv9c target detection network model, obtain a trained network model, and deploy the trained network model in an inference module; S3: continuously scanning the surface of the carbon fiber plate with a line laser source, and collecting a temperature image sequence of the scanning area with an infrared thermal imager; S4: based on the multiple frames of thermal image data collected at the initial stage of scanning, constructing a dynamic background model through frame averaging, performing frame difference processing on the real-time thermal image data collected subsequently, suppressing background thermal noise, and obtaining a difference thermal image; S5: inputting the difference thermal image into the inference module, using the inference module to detect defects in the real-time frame difference thermal image data, outputting the bounding box and class confidence of the defect position, and automatically realizing real-time detection and tracking of the defect based on the detection results of multiple consecutive frames.

[0007] Further, in step S3, the line laser source is a semiconductor laser generator, the working wavelength is 915 nm, the maximum output power is 500 W, the line laser spot length is 40-60 mm, and the width is 4-6 mm.

[0008] Further, in step S3, the resolution of the infrared thermal imager is 640x512 pixels, and the frame rate is 30 Hz.

[0009] Further, in step S3, the laser beam emitted by the line laser source passes through a convex lens, a Fourier lens, and a cylindrical lens combination in sequence for shaping, forming a line-shaped spot in the width direction of the carbon fiber plate.

[0010] Further, in step S3, the carbon fiber plate is fixed on a three-dimensional moving platform, and the moving platform moves at a constant speed of 3-7 mm / s in the scanning direction to realize uniform scanning.

[0011] A carbon fiber plate infrared thermal imaging defect real-time detection system, which is used to execute a carbon fiber plate infrared thermal imaging defect real-time detection method as described in the present application, the detection system comprising: a semiconductor laser generator for generating a line laser excitation for scanning the surface of the carbon fiber plate; an infrared thermal imager for collecting a temperature image sequence of the scanning area; a three-dimensional moving platform for fixing and moving the carbon fiber plate; a lens group for shaping the laser beam emitted by the semiconductor laser generator into a line-shaped spot; an industrial computer internally configured with an inference module for loading the improved YOLOv9c target detection network model and performing frame difference processing, neural network inference, and defect tracking functions.

[0012] Further, the industrial computer realizes GPU inference based on the PyTorch framework and CUDA acceleration.

[0013] The present application provides a kind of carbon fiber plate infrared thermal imaging defect real-time detection method and system, compared with prior art, the present application is combined with line laser scanning and frame difference enhancement image contrast, with improved YOLOv9c model and sparse attention mechanism, effectively improve the accuracy and real-time of detection, and system structure is simple, it is convenient for industrial grade field deployment, significantly improve the automation level and production efficiency of carbon fiber plate defect detection. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 It is the detection method flow chart of the present application.

[0015] Figure 2 It is the actual detection effect of the present application Figure One .

[0016] Figure 3 It is the actual detection effect of the present application Figure Two .

[0017] Figure 4 It is the actual detection effect of the present application Figure Three . DETAILED DESCRIPTION

[0018] The related technical solutions will be described clearly and completely below, obviously, the described embodiments are only a part of the embodiments, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application. EMBODIMENT

[0019] A kind of carbon fiber plate infrared thermal imaging defect real-time detection method, as shown in Fig. Figure 1 It includes the following steps: S1: improved YOLOv9c target detection network model is constructed;Wherein, the improved YOLOv9c target detection network model is integrated in the front end of the backbone network of YOLOv9c target detection network sparse space-channel hybrid attention module and is constituted.

[0020] The sparse space-channel hybrid attention module is based on ACmix framework, and is improved by the following way: the original convolution kernel generating QKV feature is replaced by 3*3 convolution kernel to enhance local receptive field, while introducing sparse regularization loss function after Softmax weight output, to suppress invalid noise features, improve the sparsity and significance of feature expression, ensure the accurate focusing and interference suppression of carbon fiber plate thermal anomaly area.

[0021] S2: Using labeled thermal image data collected under various excitation intensities, scanning directions, and ambient temperature conditions, the improved YOLOv9c target detection network model is trained offline to optimize network weights. The training rounds are no less than 100 rounds until the detection performance is stable. In this embodiment, it is 100 rounds. The trained network model is then deployed to the inference module of an industrial computer.

[0022] S3: The surface of the carbon fiber plate is continuously scanned using a semiconductor laser generator. In this embodiment, the laser power is adjusted to 30 W to generate sufficient thermal excitation effect without damaging the material. At the same time, an infrared thermal imager is used to collect temperature image sequences of the scanned area, and the obtained thermal image data is transmitted to an industrial computer in real time.

[0023] S4: To eliminate the thermal tailing effect and background thermal noise that may be generated during laser scanning, multiple frames of thermal image data collected at the beginning of the scan are selected first. A dynamic background model is constructed by frame averaging. Frame differential processing is performed on each subsequent frame of real-time thermal image data to suppress background thermal noise and obtain differential thermal images, thereby effectively enhancing the temperature contrast of the defect area.

[0024] S5: Input the differential thermal image into the inference module, use the inference module to perform defect detection on the real-time frame differential thermal image data, output the bounding box coordinates of the defect location and category confidence information, and automatically realize real-time defect detection and tracking based on the detection results of multiple consecutive frames, and display the detection results on an external display in real time. The detection effect diagram is attached. Figure 2 ~Appendix Figure 4 As shown.

[0025] The detection method described in this embodiment is performed by a real-time infrared thermal imaging defect detection system for carbon fiber plates, which includes: In this embodiment, the semiconductor laser generator has a working wavelength of 915 nm, a maximum output power of 500 W, a line laser spot length of 50 mm, and a width of 5 mm, and is used to generate line laser excitation to scan the surface of the carbon fiber plate. In this embodiment, the infrared thermal imager has a resolution of 640×512 pixels and a frame rate of 30 Hz, and is used to acquire temperature image sequences of the scanned area. In this embodiment, the three-dimensional moving platform is controlled by a host computer software to fix and move the carbon fiber plate. During the scanning process, the three-dimensional moving platform is controlled to move at a constant speed of 5 mm / s along the scanning direction (X-axis) to ensure that the line laser performs a uniform scan on the surface of the plate. The lens group, in this embodiment, includes a convex lens, a Fourier lens and a cylindrical lens, and is used for shaping the laser beam emitted by the semiconductor laser generator through the convex lens, the Fourier lens and the cylindrical lens in sequence to form a linear light spot along the width direction of the carbon fiber plate on the surface of the carbon fiber plate. The industrial computer, in this embodiment, is configured with an Intel Core i5-13490F CPU and an NVIDIA RTX 4060 GPU, and runs a Windows 10 operating system; an inference module configured in the industrial computer is constructed based on a PyTorch framework and a CUDA technology, and is used for realizing GPU-accelerated inference; the inference module is responsible for loading the improved YOLOv9c target detection network model trained and is used for performing frame difference processing, neural network inference and defect tracking functions.

[0026] In summary, the present application provides a real-time detection method and system for defects of a carbon fiber plate in infrared thermal imaging. Firstly, an improved YOLOv9c target detection network model is constructed and trained, which significantly enhances the recognition ability of thermal anomaly features and the anti-interference performance by integrating a sparse space-channel mixed attention module based on an ACmix framework at the front end of the backbone network and combining a sparse regularization loss function. In the real-time detection stage, temperature image sequences are synchronously collected by line laser scanning and an infrared thermal imager, and frame difference processing based on a dynamic background model is used to effectively suppress thermal tailing effects and background noise and improve defect contrast. Finally, the processed images are input into the trained network model deployed in the inference module of the industrial computer to realize real-time and automatic detection and tracking of defects. The system integrates the method as an integrated solution, has the characteristics of automation, intelligence and high efficiency, and can meet the industrial online nondestructive testing requirements of carbon fiber composite material plates.

[0027] In the description of the present application, it should be understood that the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application.

[0028] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for real-time detection of defects in carbon fiber plate by infrared thermal imaging, characterized by: Comprise the following steps: S1: Construct an improved YOLOv9c target detection network model; wherein the improved YOLOv9c target detection network model is constructed by integrating a sparse space-channel hybrid attention module at the front end of the backbone network of the YOLOv9c target detection network, and the sparse space-channel hybrid attention module generates QKV features using a convolution kernel, and introduces a sparsity regularization loss function after the softmax weight output to enhance feature sparsity and noise resistance; S2: Use the labeled thermal image data collected under multiple excitation intensities, scanning directions and environmental temperature conditions to offline train the improved YOLOv9c target detection network model to obtain a trained network model, and deploy the trained network model in an inference module; S3: Use a line laser source to continuously scan the surface of the carbon fiber plate, and use an infrared thermal imager to collect temperature image sequences of the scanned area; S4: Based on the multiple frames of thermal image data collected at the initial stage of scanning, construct a dynamic background model by frame averaging, perform frame difference processing on the real-time thermal image data collected subsequently, suppress background thermal noise, and obtain a difference thermal image; S5: Input the difference thermal image into the inference module, use the inference module to detect defects in the real-time frame difference thermal image data, output the bounding box and class confidence of the defect position, and automatically realize real-time detection and tracking of the defect based on the detection results of multiple consecutive frames.

2. The method according to claim 1, characterized in that: In step S3, the line laser source is a semiconductor laser generator with a working wavelength of 915 nm and a maximum output power of 500 W, and the line laser spot length is 40-60 mm and the width is 4-6 mm.

3. The method according to claim 1, characterized in that: In step S3, the resolution of the infrared thermal imager is 640×512 pixels, and the frame rate is 30 Hz.

4. The method according to claim 1, characterized in that: In step S3, the laser beam emitted by the line laser source is shaped by a combination of a convex lens, a Fourier lens and a cylindrical lens to form a line-shaped spot in the width direction of the carbon fiber plate.

5. The method according to claim 1, characterized in that: In step S3, the carbon fiber plate is fixed on a three-dimensional moving platform, and the moving platform moves at a constant speed of 3-7 mm / s in the scanning direction to achieve uniform scanning.

6. A real-time detection system for infrared thermographic defects in carbon fiber panels, characterized by: The system is used to perform the real-time defect detection method of the carbon fiber plate infrared thermal imaging as claimed in any one of claims 1-5, and the detection system comprises: A semiconductor laser generator for generating a line laser excitation for scanning the surface of the carbon fiber plate; An infrared thermal imager for collecting temperature image sequences of the scanned area; A three-dimensional moving platform for fixing and moving the carbon fiber plate; A lens group for shaping the laser beam emitted by the semiconductor laser generator into a line-shaped spot; An industrial computer with an inference module inside for loading the improved YOLOv9c target detection network model and performing frame difference processing, neural network inference and defect tracking functions.

7. The real-time system for detecting defects in carbon fiber panels by infrared thermography according to claim 6, characterized in that it comprises: The industrial computer realizes GPU inference based on the PyTorch framework and CUDA acceleration.

Images