A recognition method for aircraft skin defect detection
The characteristic parameters of the aircraft skin surface are extracted through the grayscale symbiosis matrix, and combined with fuzzy control and BP neural network technology, the rapid and accurate detection of aircraft skin defects is achieved, and the training convergence speed and low detection accuracy in traditional methods are solved. question.
Patent Information
- Application Number
- CN202111086451.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-16
- Publication Date
- 2025-02-11
- Estimated Expiration
- 2041-09-16
AI Technical Summary
Traditional aircraft skin defect detection methods have problems such as slow training convergence speed, low detection accuracy and low efficiency.
The characteristic parameters of the skin surface are extracted using a grayscale symbiosis matrix, and fuzzy processing is performed through a fuzzy control system, combined with the BP neural network classification recognition technology based on fuzzy control, real-time detection of aircraft skin defects is realized.
This greatly improves the training convergence speed, significantly improves the accuracy and accuracy of defect detection, improves detection efficiency, and reduces errors.
Smart Images
Figure CN113870199B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an online detection technology for aircraft skin, and in particular to an identification method for detecting defects in aircraft skin. Background Art
[0002] With the development of science and technology and economy, aircraft play a key role in various fields such as military, transportation, agriculture, etc. However, during the long-term flight of the aircraft, as an important part of the aircraft structure, the skin surface will be eroded by the atmospheric environment and various impact loads, and defects such as skin surface cracks, surface impact and surface corrosion will appear. Surface defects not only affect the appearance of the aircraft skin surface, but also affect the overall service life of the aircraft. More seriously, it will lead to flight accidents, seriously threatening the safety of pilots and passengers. Therefore, manufacturers attach great importance to the detection of skin surface defects in order to control them in time, thereby effectively controlling the quality of aircraft skin.
[0003] Traditional aircraft skin defect detection often uses the naked eye for identification and judgment. Manual detection is prone to errors and the detection efficiency is very low. With the development of science and technology, domestic and foreign researchers use machine vision for non-destructive testing, and its detection efficiency has been improved, but the detection speed is still slow and the level of intelligence needs to be improved. Especially in the pattern recognition module, the traditional BP neural network is a multi-layer feedforward network based on error direction propagation. It is a supervised learning algorithm that can solve the weight adjustment problem of multi-layer feedforward neural network of nonlinear continuous function. The typical BP neural network uses the gradient descent algorithm. With the increasing complexity of the internal structure of the aircraft and the influence of external uncertainties on the detection of aircraft skin surface defects, there is a certain uncertainty between the characteristic parameters and the type of aircraft skin defects, which makes the aircraft skin defect detection results not accurate enough. The design of the fuzzy controller does not require an accurate mathematical model, but one of the disadvantages of fuzzy logic is that fuzzy rules are usually established based on experience, and it is difficult to establish a stable mathematical model. It itself cannot learn autonomously. With the increase of fuzzy rules, the time of fuzzy reasoning also increases, which greatly reduces the reasoning efficiency. To sum up: when the traditional BP neural network classification technology is used to realize defect recognition, the neural network model training convergence speed is slow, and the use of fuzzy reasoning for defect recognition will also lead to inaccurate detection results and low detection efficiency. Therefore, how to achieve the best detection effect is still a question worth pondering. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide an aircraft skin defect detection recognition method which can not only greatly improve the training convergence speed, but also effectively improve the precision and accuracy of defect detection, thereby improving the detection efficiency.
[0005] In order to solve the above technical problems, the identification method of aircraft skin defect detection of the present invention comprises the following steps:
[0006] Step A: image acquisition, using a CCD camera to take pictures of the aircraft skin and preprocessing the captured original aircraft skin defect images;
[0007] Step B: Using gray-level co-occurrence matrix to extract characteristic parameters of the skin surface under different defect types, so as to establish a set of aircraft skin characteristic parameters;
[0008] Step C: fuzzify the characteristic parameters in step B using a fuzzy control system to obtain a characteristic parameter set expressed in fuzzy language;
[0009] Step D: According to the different types of aircraft skin surface defects, real-time detection of aircraft skin defects is achieved through BP neural network classification and recognition based on fuzzy control.
[0010] In the step A, the original aircraft skin defect image is preprocessed by selecting a median filter to preprocess the original aircraft skin defect image to remove noise through experimental comparison.
[0011] In the specific process of preprocessing the original aircraft skin defect image, a sampling window M containing an odd number of pixels is first selected, and then the window M1 on the original image is moved, and the pixel grayscale values in the original window M1 are sorted according to the size of the acquired image. Finally, the grayscale value in the middle part is selected as the grayscale value of all pixels in the window. The corresponding median filtering formula is:
[0012] k(i,j)=med{h(xt,yj),t,j∈M} (1)
[0013] Where: the gray value of the pixel (i, j) in the original aircraft skin defect image is h(xt, yj), the gray value of the corresponding pixel (i, j) after filtering is k(i, j), t and j are image coordinate variables, M is the sampling window size, and med is the median filter algorithm for eliminating the original image noise.
[0014] The aircraft skin defect image after median filtering is subjected to image binarization processing. The algorithm process is as follows:
[0015]
[0016] Where: (i, j) - the coordinates of the pixel point of the skin image after median filtering; h(i, j) - the gray value of the skin image after median filtering; k(i, j) - the gray value of the skin image after binarization.
[0017] The specific process of the BP neural network classification and recognition technology based on fuzzy control is as follows: fuzzy processing is performed on the extracted feature parameters using fuzzy language, the BP neural network toolbox based on fuzzy control is used in MATLAB to perform corresponding training to obtain a training model, and real-time detection of aircraft skin defects is achieved through fuzzy feature parameter reasoning.
[0018] Through the BP neural network classification and recognition technology based on fuzzy control, it is divided into four unit layers: the first layer is the original aircraft skin defect image feature extraction layer; the second layer is the fuzzification layer; the third layer is the neural network layer; and the fourth layer is the output layer.
[0019] The first aircraft skin defect image feature extraction layer is used to complete the preprocessing of the defect image and the extraction of the feature parameters of the defect image after preprocessing; the second fuzzification layer is used to complete the fuzzification and normalization processing of the feature parameters; the third neural network layer uses the function of the BP neural network to train the fuzzified data, adjust the weight parameters of the system, optimize the system structure, perform pattern classification on the input data, and test the classification accuracy; the fourth layer is the output layer, which uses the BP neural network toolbox based on fuzzy control in MATLAB to perform corresponding training to obtain the training model, and output the skin defect type in real time through fuzzy feature parameter reasoning.
[0020] Compared with the existing technology, the inspection can be completed in front of the computer without the need for professionals. It not only greatly reduces the errors caused by manual inspection, but also effectively improves the precision and accuracy of defect detection. In particular, a special image acquisition method is used to establish a set of aircraft skin feature parameters expressed in fuzzy language, and the BP neural network classification and recognition technology based on the fuzzy logic algorithm is used to achieve rapid and accurate recognition and judgment of aircraft skin surface defects, which effectively verifies the advantages of combining fuzzy control theory with BP neural network technology. Compared with traditional BP neural networks, its training speed is faster and easier to converge. It has significantly improved the accuracy and real-time performance of pattern recognition and classification detection, has good scalability and application prospects, is of great significance to reducing air crashes caused by aircraft skin defects, and also provides a new research direction for the field of non-destructive testing. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 The schematic diagram of the identification method for detecting aircraft skin defects of the present invention;
[0022] Figure 2 The present invention is a flow chart of the identification method for detecting aircraft skin defects. DETAILED DESCRIPTION
[0023] The following is a further detailed description of the aircraft skin defect detection identification method of the present invention in conjunction with the accompanying drawings and specific implementation methods.
[0024] As shown in the figure, the identification method for aircraft skin defect detection of the present invention comprises the following steps:
[0025] Step A: Image acquisition: Use a high-precision CCD camera to take pictures of the aircraft skin surface, and transmit the collected real-time and dynamic skin damage information to the ground health monitoring platform through wireless transmission technology, and then use median filtering and image binarization to pre-process the characteristic image (original aircraft skin defect image) of the taken aircraft skin picture;
[0026] Among them, the feature image (original aircraft skin defect image) is preprocessed by using the median filter method to preprocess the original aircraft skin defect image for denoising through experimental comparison, thereby ensuring that the edge features in the skin surface image are clear and complete while removing the noise.
[0027] In the specific process of preprocessing the feature image (original aircraft skin defect image), a sampling window M containing an odd number of pixels is first selected, and then the original window M1 on the original image is moved, and the pixel gray values in the original window M1 are sorted according to the size of the acquired image. Finally, the gray value in the middle part is selected as the gray value of all pixels in the window. The corresponding median filtering formula is:
[0028] k(i,j)=med{h(xt,yj),t,j∈M} (1)
[0029] Where: the gray value of the pixel point (i, j) in the original image of the aircraft skin defect is h(xt, yj), the gray value of the corresponding pixel point (i, j) after filtering is k(i, j), t and j are image coordinate variables, M is the sampling window size, and med is the median filter algorithm for eliminating the noise of the original image.
[0030] The Otsu algorithm, which can automatically determine the threshold, is used to perform image binarization on the aircraft skin defect image after median filtering. The algorithm process is as follows:
[0031]
[0032] Where: (i, j) - the coordinates of the pixel point of the skin image after median filtering; h(i, j) - the gray value of the skin image after median filtering; k(i, j) - the gray value of the skin image after binarization.
[0033] Step B: extracting the feature parameters of the preprocessed defect image using the gray level co-occurrence matrix, thereby establishing a set of aircraft skin feature parameters;
[0034] The specific process of extracting the characteristic parameters of the skin surface using the gray-level co-occurrence matrix is as follows: four characteristic parameter values of contrast, correlation, energy and homogeneity in different directions (0°, 45°, 90°, 135°) are selected, and the step size δ=1 is taken to construct the gray-level co-occurrence matrix in the four directions of 0°, 45°, 90° and 135°, and the matrix is normalized to obtain the relevant parameters of the gray-level co-occurrence matrix. By selecting high-weight parameters on the basis of multi-dimensional parameters to perform parameter dimensionality reduction, the classification detection of aircraft skin defects can be realized.
[0035] Step C: fuzzify the characteristic parameters in step B using a fuzzy control system to obtain a characteristic parameter set expressed in fuzzy language;
[0036] Step D: According to the different types of aircraft skin surface defects, the BP neural network classification and recognition technology based on fuzzy control is used. The system structure is as follows: Figure 1 As shown in the figure, it is mainly divided into 4 layers; the first layer is the original aircraft skin defect image feature extraction layer, which mainly completes the preprocessing of the defect image and the extraction of the feature parameters of the defect image after preprocessing; the second layer is the fuzzification layer, which mainly completes the fuzzification and normalization processing of the feature parameters; the third layer is the neural network layer, which mainly uses the function of the BP neural network to train the fuzzified data, adjust the weight parameters of the system, optimize the system structure, classify the input data, and test the classification accuracy; the fourth layer is the output layer, which uses the BP neural network toolbox based on fuzzy control in MATLAB to perform corresponding training to obtain the training model, and through fuzzy feature parameter reasoning, the skin defect type is output in real time, that is, the fuzzy neural network (FBP) is used to realize fuzzy feature parameter reasoning; that is to say, the extracted feature parameters are fuzzy processed by fuzzy language, and the fuzzy parameter set is trained accordingly in MATLAB using the BP neural network toolbox based on fuzzy control to obtain the training model, and random testing and verification are performed through fuzzy feature parameter reasoning, so as to quickly and effectively realize the real-time detection of aircraft skin defects.
[0037] The above-mentioned fuzzy processing of the extracted feature parameters using fuzzy language and the addition of fuzzy logic algorithm to the neural network training increase the self-learning function of the BP neural network compared to the traditional fuzzy reasoning, which can effectively reduce the uncertainty of the data caused by fuzzy processing.
[0038] In addition, the specific process of BP neural network classification and recognition technology based on fuzzy control is as follows: first, four characteristic parameters are selected as input layer node parameters after fuzzy processing, and three defect types are selected as the number of nodes in the output layer. Taking aircraft skin defect detection as the application background, the four characteristic parameters are input and collected to obtain 200 characteristic parameters. After fuzzy processing, the characteristic parameters are expressed in the form of fuzzy language and then input into the BP neural network for training to form a fuzzy neural network classifier. By adjusting the system parameters online and optimizing the classifier structure, the defect classification detection can be quickly realized.
[0039] Experimental results show that the identification method for aircraft skin defect detection based on the present invention, after the introduction of fuzzy algorithm, can output the classification summary results in real time on the computer, and the classification accuracy is improved to 93.55%, which is 3.35% higher than the 90.2% accuracy of the BP network. The number of classification cycles is reduced from 1086 times to less than 500 times, which effectively verifies the practicability and accuracy of the defect detection device.
Claims
1. A method for detecting aircraft skin defects, comprising the following steps: Step A: image acquisition, using a CCD camera to take pictures of the aircraft skin and preprocessing the captured original aircraft skin defect images; Step B: Using gray-level co-occurrence matrix to extract characteristic parameters of the skin surface under different defect types, so as to establish a set of aircraft skin characteristic parameters; Step C: fuzzify the characteristic parameters in step B using a fuzzy control system to obtain a characteristic parameter set expressed in fuzzy language; Step D: According to the different types of aircraft skin surface defects, the aircraft skin defects are detected in real time through BP neural network classification and recognition based on fuzzy control; The specific process of the BP neural network classification and recognition technology based on fuzzy control is as follows: fuzzy processing is performed on the extracted feature parameters using fuzzy language, and the BP neural network toolbox based on fuzzy control is used in MATLAB to perform corresponding training to obtain a training model, and real-time detection of aircraft skin defects is achieved through fuzzy feature parameter reasoning; Through the BP neural network classification and recognition technology based on fuzzy control, it is divided into four unit layers: the first layer is the original aircraft skin defect image feature extraction layer; the second layer is the fuzzification layer; the third layer is the neural network layer; the fourth layer is the output layer; The first aircraft skin defect image feature extraction layer is used to complete the preprocessing of the defect image and the extraction of the feature parameters of the defect image after preprocessing; the second fuzzification layer is used to complete the fuzzification and normalization processing of the feature parameters; the third neural network layer uses the function of the BP neural network to train the fuzzified data, adjust the weight parameters of the system, optimize the system structure, perform pattern classification on the input data, and test the classification accuracy; The fourth layer is the output layer, which is trained in MATLAB using the BP neural network toolbox based on fuzzy control to obtain a training model, and outputs the skin defect type in real time through fuzzy feature parameter reasoning.
2. The method for detecting and identifying aircraft skin defects according to claim 1, characterized in that: In the step A, the original aircraft skin defect image is preprocessed by selecting a median filter to preprocess the original aircraft skin defect image to remove noise through experimental comparison.
3. The method for detecting and identifying aircraft skin defects according to claim 2, characterized in that: Said In the specific process of preprocessing the original aircraft skin defect image, a sampling window containing an odd number of pixels is first selected. , and then move the window on the original image , complete the original window according to the size of the acquired image Sort the grayscale values of the pixels in the middle part. The gray value of is used as the gray value of all pixels in the window, and the corresponding median filtering formula is: (1) Where: Pixel points in the original aircraft skin defect image The gray value is , the corresponding pixel after filtering The gray value is , and is the image coordinate variable, is the sampling window size, Representation of the median filter algorithm for removing noise from the original image.
4. The method for detecting and identifying aircraft skin defects according to claim 3 is characterized in that: The aircraft skin defect image after median filtering is subjected to image binarization processing. The algorithm process is as follows: (2) Where: ——The coordinates of the pixel points of the skin image after median filtering; ——The gray value of the skin image after median filtering; ——The grayscale value of the skin image after binarization.