Surface defect detection method for aircraft composite materials

Abstract

This invention relates to the field of composite material testing technology, specifically to a method for detecting surface defects in composite material structures. This invention employs hyperspectral imaging technology to achieve a rapid method for detecting surface defects in composite material structures, resulting in fast data acquisition and processing speeds and high identification accuracy. This invention can not only quickly and accurately detect the location and type of surface defects in composite materials, but also classify and determine them further, thereby greatly reducing the influence of subjective human factors on defect determination. Simultaneously, it can directly obtain defect location data without the need for secondary measurement work, improving work efficiency. This invention simultaneously utilizes the image and spectral dimensions of the hyperspectral imaging system to detect fabric defects, making the results more intuitive and accurate.

Description

Technical Field

[0001] This invention relates to the field of composite material testing technology, and more specifically to a method for detecting surface defects in composite material structures. Background Technology

[0002] Because composite materials are made by mixing different types of materials in a certain proportion, their internal stability is not as good as that of single-type materials. Therefore, they are prone to damage during processing, which manifests as cracks, delamination, and other phenomena. Current inspection methods rely on manual inspection, where operators use tools such as flashlights and reflectors to inspect with the naked eye. However, since aircraft extensively use composite materials, the areas and number of composite structures to be inspected are large, and manual methods may miss some. Furthermore, for surface defects such as scratches, the depth and width of the scratches cannot be directly determined by the naked eye, making it impossible to determine whether the material can continue to be used. It is still necessary to request a metrology center to measure the surface defects of the composite material to determine the result. The above results prove that the traditional method has disadvantages such as easy omissions, low efficiency, reliance on metrology, and long working time. Therefore, there is an urgent need for an efficient method for detecting surface defects in composite material structures. Summary of the Invention

[0003] The purpose of this invention is to solve the problems in the prior art and propose a method for detecting surface defects in composite material structures. This method uses hyperspectral remote sensing imaging technology, a parallel classification algorithm for the spectral angle of remote sensing images, and a neural network structure model to achieve accurate identification of defect images, enabling rapid online detection of surface defects in aircraft composite material structures.

[0004] To achieve the above objectives, this invention proposes a method for detecting surface defects in composite material structures, comprising the following steps:

[0005] Step 1: Under the same working environment, use a hyperspectral remote sensing imaging system to collect hyperspectral remote sensing data and images of several composite materials with various surface defects, forming a database. The composite material structures to be inspected include aircraft skin, hatches, control surfaces, and doors. The surface defects of the above composite structures include stains, scratches, dents, bulges, etc.

[0006] The difference between hyperspectral remote sensing images of composite materials with defects and those without defects is reflected in the composite defect quantification value. Data and image acquisition equipment obtain the average brightness of the defective area of ​​the composite material structure and the average brightness of the background area of ​​the defective area. The image contrast of the defective area is calculated using both values. Furthermore, the defect severity of the composite material defective area is calculated using its area and depth, thereby determining the composite defect quantification value. Specifically, the formula for calculating the composite defect quantification value is as follows:

[0007]

[0008] Among them, Q fc K represents the quantification value of the composite defect in the defect region of the composite structure. s L is the correction factor for the defect area of ​​composite materials (which varies depending on the type of composite material). q L represents the average brightness of the defective areas in the composite material. b S represents the average background brightness of the composite material. q K represents the area of ​​the defective part in the composite material. h H is the depth correction factor for defects in composite materials (varying depending on the type of composite material). q This represents the depth value of the defective part in the composite material.

[0009] The equipment for acquiring hyperspectral remote sensing data and images of composite materials with various surface defects mainly adopts camera equipment using visual guidance technology.

[0010] Step 2: Because the environment during actual operation differs from the environment when the database was established, it is necessary to process and convert various hyperspectral remote sensing data and images during actual use. Methods such as remote sensing radiometric correction, exposure reduction, and image enhancement are used to convert them into information that can be recognized and processed by the computer or analysis module based on the data collected.

[0011] During the processing and conversion, remote sensing radiometric correction is particularly important. By adjusting and correcting the radiometric values ​​in remote sensing images, the influence of factors such as the atmosphere, surface reflectivity, and sensors that may have affected the data during acquisition can be eliminated. This avoids inaccurate or inconsistent radiometric values ​​in the images caused by these factors, which could affect the accuracy of the detection results. To this end, a correction formula is provided:

[0012]

[0013] Among them, P m For the hyperspectral remote sensing image after radiometric correction, P l A is the hyperspectral remote sensing image before radiometric correction, A is the theoretical image of the composite structure under test, and k is the theoretical image of the composite structure under test. y This is the remote sensing radiometric correction factor.

[0014] Step 3: Utilize a parallel classification algorithm for spectral angles of remote sensing images to form fitted image information from the processed and transformed hyperspectral remote sensing data and images, thereby revealing surface defects in the composite material structure. The principle of this parallel classification algorithm is to project the spectrum as a vector onto an N-dimensional space, where N is the number of bands being calculated. In this N-dimensional space, the spectral curve is a vector with direction and length. The angle formed between different vectors is the spectral angle; the smaller the spectral angle between different spectral curves, the more similar they are.

[0015]

[0016] Where α is the spectral angle, k is the correction coefficient, n is the total number of bands, Xa is the measured spectral curve, and Xb is the theoretical spectral curve for different defect forms.

[0017] The depth and length of the damaged area were obtained using remote sensing image processing software.

[0018] Step 4: Establish a linear neural network structure model

[0019] A linear neural network model is established, similar to a neuron forming a multi-input single-output information processing unit. Based on the characteristics and functions of neurons, the neuron can be abstracted into a simple mathematical model algorithm, which includes an input layer, a decision layer, and an output layer. Hyperspectral remote sensing data and images enter the decision layer through the input layer (as described in step one), and then reach the output layer through the decision layer (step three), thus obtaining the surface defect type and damage degree of the composite structure.

[0020] Step 5: Utilize the established linear neural network structure model to detect the defect types of the sample under test and formulate a treatment plan.

[0021] The composite material sample to be tested is operated according to steps one to three. Then, the hyperspectral remote sensing image obtained from step two and the depth and length of the damage site obtained from step three are fed into the established linear neural network structure model to obtain the type and degree of damage on the surface of the composite material structure.

[0022] The beneficial effects of this invention are:

[0023] This invention employs hyperspectral imaging technology to obtain a rapid method for detecting surface defects in composite material structures, enabling fast data acquisition and processing with high identification accuracy. This invention not only quickly and accurately detects the location and type of surface defects in composite materials but also classifies and determines them, significantly reducing the influence of subjective human factors on defect identification. Furthermore, it directly obtains defect location data without the need for secondary measurement, improving work efficiency. This invention simultaneously utilizes both the image and spectral dimensions of the hyperspectral imaging system to detect fabric defects, resulting in more intuitive and accurate results. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the process of surface defect detection methods for composite material structures.

[0025] Figure 2 This is a flowchart of the algorithm of the linear neural network structure model of the present invention based on the surface defect detection method of composite material structure;

[0026] Figure 3This is a schematic diagram of the training of the linear neural network structure model of the present invention based on the surface defect detection method of composite material structure. Detailed Implementation

[0027] Reference Figure 1 This invention is based on a method for detecting surface defects in composite material structures, comprising the following steps:

[0028] S1. Use a hyperspectral imaging system to collect hyperspectral data and images of several types of fabrics with defects:

[0029] First, under the same working environment, a hyperspectral remote sensing imaging system was used to collect hyperspectral remote sensing data and images of several composite materials with various surface defects to form a database. The composite material structures to be inspected included aircraft skin, hatches, control surfaces, and doors. The surface defects of the above composite structures included stains, scratches, dents, bulges, etc. In this way, a set of composite structure defects was formed for subsequent missing parts detection and comparison.

[0030] The difference between hyperspectral remote sensing images of composite materials with defects and those without defects is reflected in the composite defect quantification value. Data and image acquisition equipment obtain the average brightness of the defective area of ​​the composite material structure and the average brightness of the background area of ​​the defective area. The image contrast of the defective area is calculated using both values. Furthermore, the defect severity of the composite material defective area is calculated using the area and depth of the defective area, thereby determining the composite defect quantification value. The formula for calculating the defect quantification value is shown below:

[0031]

[0032] One example is selected: images and data of scratches measuring 10mm in length, 2mm in width, and 1mm in depth on the surface of the composite material, as well as images and data of dents measuring 8mm in length, 4mm in width, and 6mm in depth, are collected to form a database.

[0033] S2. Processing and converting measured hyperspectral remote sensing data and images:

[0034] The main methods employed include remote sensing radiometric correction, exposure reduction, and image enhancement, converting the data into information that can be recognized and processed by computers or analysis modules based on the acquired data. Remote sensing radiometric correction is particularly important during the conversion process. By adjusting and correcting the radiometric values ​​in remote sensing images, it eliminates the influence of factors such as atmospheric conditions, surface reflectivity, and sensor performance that may have affected the data during acquisition. This prevents inaccurate or inconsistent radiometric values ​​in the images from affecting the accuracy of the detection results. Therefore, the correction formula is adopted:

[0035]

[0036] Step 3: Utilize a parallel classification algorithm for spectral angles of remote sensing images to form fitted image information from the processed and transformed hyperspectral remote sensing data and images, thereby revealing surface defects in the composite material structure. The principle of this parallel classification algorithm is to project the spectrum as a vector onto an N-dimensional space, where N is the number of bands being calculated. In this N-dimensional space, the spectral curve is a vector with direction and length, and the angle formed between different vectors is the spectral angle. The smaller the spectral angle between different spectral curves, the more similar they are.

[0037]

[0038] One example is selected: the depth and length of the damaged area are obtained using remote sensing image processing software: the measured surface defect information of the composite material is a scratch image with a length of 8mm, a width of 1.5mm and a depth of 0.5mm.

[0039] Step 4: Establish a linear neural network structure model

[0040] A linear neural network model is established, similar to a neuron forming a multi-input single-output information processing unit. Based on the characteristics and functions of neurons, the neuron can be abstracted into a simple mathematical model algorithm, which includes an input layer, a decision layer, and an output layer. Hyperspectral remote sensing data and images enter the decision layer through the input layer (as described in step one), and then reach the output layer through the decision layer (step three), thus obtaining the surface defect type and damage degree of the composite structure.

[0041] One example is selected: by comparing the measured surface defect information of the composite material with the defect database information, the defect type is determined to be "scratches" and the defect severity is "minor".

[0042] Step 5: Utilize the established linear neural network structure model to detect the defect types of the sample under test and formulate a treatment plan.

[0043] The composite material sample to be tested is operated according to steps one to three. Then, the hyperspectral remote sensing image obtained from step two and the depth and length of the damage site obtained from step three are fed into the established linear neural network structure model to obtain the type and degree of damage on the surface of the composite material structure.

Claims

1. A method for detecting surface defects in aircraft composite material structures, characterized in that, Includes the following steps: Step 1: Under the same working environment, use a hyperspectral remote sensing imaging system to collect hyperspectral remote sensing data and images of several composite materials with various surface defects, and form a database. The difference in data between hyperspectral remote sensing defect images and defect-free images of composite materials is reflected in the composite defect quantification value, which is calculated using the following formula: Among them, Q fc K represents the quantification value of the composite defect in the defect region of the composite structure. s L is the correction factor for the defect area of ​​composite materials. q L represents the average brightness of the defective areas in the composite material. b S represents the average background brightness of the composite material. q K represents the area of ​​the defective part in the composite material. h H is the depth correction factor for defects in composite materials. q This represents the depth of the defect in the composite material. Step 2: Process and convert the various hyperspectral remote sensing data and images by using remote sensing radiometric correction, exposure reduction, and image enhancement methods to convert them into information that can be recognized and processed by a computer or analysis module based on the data collected. Step 3: The parallel classification algorithm of spectral angle of remote sensing images is used to form fitted image information from the processed and transformed hyperspectral remote sensing data and images to display surface defects of composite material structures. The principle of the parallel classification algorithm of spectral angle of remote sensing images is to project the spectrum as a vector onto an N-dimensional space, where the dimension N is the number of bands calculated. In the N-dimensional space, the spectral curve is a vector with direction and length. The angle formed between different vectors is the spectral angle. The smaller the spectral angle between different spectral curves, the more similar they are. Where α is the spectral angle, k is the correction coefficient, n is the total number of bands, Xa is the measured spectral curve, and Xb is the theoretical spectral curve for different defect forms; Use remote sensing image processing software to obtain the depth and length of the damaged area; Step 4: Establish a linear neural network structure model A linear neural network model is established, which includes an input layer, a decision layer, and an output layer. Hyperspectral remote sensing data and images enter the decision layer through the input layer, and then reach the output layer after passing through the decision layer, so as to obtain the surface defect type and damage degree of the composite structure. Step 5: Use the established linear neural network structure model to detect the defect type of the sample under test and formulate a treatment plan.

2. The method for detecting surface defects in aircraft composite material structures as described in claim 1, characterized in that, The composite material structures to be tested include aircraft skin, hatches, control surfaces, and cabin doors. Surface defects of the composite material structures include stains, scratches, dents, and bulges.

3. The method for detecting surface defects in aircraft composite material structures as described in claim 1 or 2, characterized in that, In step two, during the processing and conversion, a correction formula is provided by adjusting and correcting the radiometric values ​​in the remote sensing image: Among them, P m For the hyperspectral remote sensing image after radiometric correction, P l A is the hyperspectral remote sensing image before radiometric correction, A is the theoretical image of the composite structure under test, and k is the theoretical image of the composite structure under test. y This is the remote sensing radiometric correction factor.

4. The method for detecting surface defects in aircraft composite material structures as described in claim 1 or 2, characterized in that, The equipment for acquiring hyperspectral remote sensing data and images of the composite materials uses a camera device with visual guidance technology.

5. The method for detecting surface defects in aircraft composite material structures as described in claim 3, characterized in that, The equipment for acquiring hyperspectral remote sensing data and images of the composite materials uses a camera device with visual guidance technology.

6. The method for detecting surface defects in aircraft composite material structures as described in claim 1, 2, or 5, characterized in that, The composite material sample to be tested is operated according to steps one to three. Then, the hyperspectral remote sensing image obtained from step two and the depth and length of the damage site obtained from step three are fed into the established linear neural network structure model to obtain the type and degree of damage on the surface of the composite material structure.

7. The method for detecting surface defects in aircraft composite material structures as described in claim 3, characterized in that, The composite material sample to be tested is operated according to steps one to three. Then, the hyperspectral remote sensing image obtained from step two and the depth and length of the damage site obtained from step three are fed into the established linear neural network structure model to obtain the type and degree of damage on the surface of the composite material structure.

8. The method for detecting surface defects in aircraft composite material structures as described in claim 4, characterized in that, The composite material sample to be tested is operated according to steps one to three. Then, the hyperspectral remote sensing image obtained from step two and the depth and length of the damage site obtained from step three are fed into the established linear neural network structure model to obtain the type and degree of damage on the surface of the composite material structure.

Images