Aircraft health reliability analysis method and system

The noise in the hole probe image was removed by wavelet decomposition and the model parameters were determined using improved genetic algorithms, which solved the problems of poor noise removal effect and low training efficiency in aircraft engine blade damage detection, and achieved efficient and accurate damage recognition.

CN118396953BActive Publication Date: 2024-11-19AVIC LITTON (BEIJING) AVIATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410498180.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-24
Publication Date
2024-11-19
Estimated Expiration
2044-04-24

AI Technical Summary

Technical Problem

The prior art has problems such as poor denoising effect and low model training efficiency in aircraft engine blade damage detection, resulting in low recognition accuracy and efficiency.

Method used

Wavelet decomposition is used to identify and remove noise in the hole probe image, and combined with the improved genetic algorithm to determine the optimal parameters of the model to improve the efficiency and accuracy of damage recognition.

Benefits of technology

Through wavelet decomposition and improved genetic algorithms, the denoising effect of mixed noise and model training efficiency are significantly improved, and the accuracy and efficiency of aircraft blade damage detection are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118396953B_ABST
    Figure CN118396953B_ABST
Patent Text Reader

Abstract

The present invention discloses a health reliability analysis method and system for an aircraft. The present invention decomposes an original image, performs noise identification on the decomposed signal, then denoises the noise signal, and then reconstructs the image using the denoised signal and the non-noise signal, thereby obtaining a denoised original image; then, the denoised original image is input into an aircraft blade damage recognition model to obtain a damage recognition result, and finally, based on the damage recognition result, a health reliability analysis result can be obtained; in this way, through the above-mentioned design, noises of different frequency bands and frequencies can be removed to the greatest extent, thereby reducing the influence of mixed noise on damage recognition, thereby improving the accuracy of model recognition; at the same time, when performing damage recognition, the optimal parameters of the model are determined by an improved genetic algorithm, and multiple sets of model parameters can be evaluated in one iteration process, thereby reducing the model training time and improving the recognition efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of aircraft reliability analysis, and in particular relates to a health reliability analysis method and system for an aircraft. Background Art

[0002] As the heart of an aircraft, the core components of an aircraft engine usually operate in a harsh environment of high temperature, high pressure and high speed. As the key components of the engine, aircraft engine blades are susceptible to factors such as damage from foreign objects, combustion, corrosion, and rotation-static friction when working in the aforementioned harsh environment, which can cause various types of damage (such as curling, cracks, etc.); therefore, damage detection of aircraft engine blades is an important part of conducting aircraft health and reliability analysis and ensuring the healthy operation of aircraft.

[0003] At present, the blade damage detection of aircraft engines mostly adopts boresight technology, which plays a vital role in aviation maintenance work. The boresight technology displays the internal condition of the engine to the maintenance personnel in the form of an image, so that the maintenance personnel can complete the damage detection of the engine blades based on the image; at the same time, with the development of computer vision and the improvement of image processing level, engine blade damage detection has gradually been transformed from manual to computer execution. Compared with manual detection, the blade damage detection method based on machine vision has improved the detection efficiency and accuracy and has been widely used.

[0004] However, the existing machine vision-based engine blade damage detection methods have the following shortcomings: (1) Since the borescope image will be affected by various factors such as the acquisition environment during the acquisition process, it will contain a lot of mixed noise. The existing technology usually uses a simple filtering method for denoising, which has a weak denoising effect on the mixed noise in the borescope image. This will have a great impact on the model recognition result, thereby reducing the accuracy of model recognition; (2) When training the model, traditional technology usually uses a Bayesian optimization search method to determine the optimal parameters of the model. However, the aforementioned method can only evaluate one set of parameters in one iteration process, which will cause the problem of long model training time. Based on this, how to provide an efficient and accurate aircraft health reliability analysis method has become a problem to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to provide an aircraft health reliability analysis method and system to solve the problem of low model recognition accuracy caused by poor denoising effect and low model training efficiency in the prior art.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, a health reliability analysis method for an aircraft is provided, comprising:

[0008] Acquire borescope images of engine blades of target aircraft;

[0009] Performing wavelet decomposition processing on the borescope image to obtain wavelet coefficients at different decomposition scales and a wavelet decomposition image signal corresponding to each wavelet coefficient;

[0010] Performing signal recognition processing on the wavelet decomposition image signal corresponding to each wavelet coefficient to recognize a noise image signal from a plurality of wavelet decomposition image signals;

[0011] Based on the wavelet coefficients of the noise image signal, denoising the noise image signal to obtain a denoised image signal;

[0012] Using the target image signal and the denoised image signal, wavelet reconstruction processing is performed to obtain a denoised boresight image after the wavelet reconstruction processing, wherein the target image signal is all wavelet decomposed image signals among the plurality of wavelet decomposed image signals except the noise image signal;

[0013] Acquire an aircraft blade damage recognition model, wherein the aircraft blade damage recognition model is obtained by determining optimal model parameters through an improved genetic algorithm, and based on the optimal model parameters, taking sample denoised boresight images of engine blades of several sample aircraft as inputs and outputting damage recognition results of engine blades of each sample aircraft for training;

[0014] Inputting the denoised boresight image into the aircraft blade damage identification model for damage identification processing to obtain a damage identification result of the engine blade of the target aircraft;

[0015] According to the damage identification results of the engine blades of the target aircraft, the health reliability analysis results of the target aircraft are obtained.

[0016] Based on the above disclosed content, after acquiring the borescope image of the engine blade of the target aircraft, the present invention first performs wavelet decomposition on it to obtain wavelet decomposition image signals at different decomposition scales; then, signal recognition is performed on each wavelet decomposition image signal, thereby identifying the noise image signal in the wavelet decomposition image signals; then, based on the wavelet coefficient of the noise image signal, the noise image signal is denoised to obtain a denoised image signal; finally, after completing the denoising of the noise image signal, the image can be reconstructed in combination with the aforementioned non-noise wavelet decomposition image signal to obtain a denoised borescope image; in this way, the aforementioned steps are equivalent to decomposing the original image into signals of different frequencies and frequency bands, and then noise recognition and noise removal are performed on the signals of each frequency and frequency band respectively. Based on this, compared with traditional filtering denoising, the present invention can remove noise of different frequency bands and frequencies to the greatest extent, thereby improving the denoising effect of mixed noise.

[0017] At the same time, the present invention adopts an improved genetic algorithm to determine the optimal parameters of the model during model training. Therefore, the population in the genetic algorithm can be used to evaluate multiple sets of model parameters in one iteration. In this way, compared with traditional technologies, the model training time can be reduced, thereby improving the damage identification efficiency. Based on this, after the aircraft blade damage identification model is obtained by training based on the aforementioned method, the aforementioned denoised boresight image is input into the aircraft blade damage identification model to obtain the damage identification result of the blade. Finally, based on the damage identification result of the blade, the health reliability analysis result of the target aircraft can be obtained.

[0018] Through the above design, the present invention uses wavelet decomposition to decompose the original image before damage detection, so as to obtain image signals of different frequency bands and frequencies; then, the image signals of different frequency bands and frequencies are identified to obtain image signals belonging to noise, and then the image signals belonging to noise are denoised to obtain denoised image signals; finally, the denoised image signals and non-noise image signals are used to reconstruct the image, so as to obtain denoised boresight images; thus, compared with traditional filtering denoising, the present invention can remove noise of different frequency bands and frequencies to the greatest extent, thereby improving the denoising effect of mixed noise; at the same time, when training the blade damage model, the present invention uses an improved genetic algorithm to determine the optimal parameters of the model. Compared with traditional technologies, the genetic population can be used to evaluate multiple groups of model parameters in one iteration process, so the model training time can be reduced, thereby improving the damage recognition efficiency; based on this, the present invention can reduce the model training time while ensuring the recognition accuracy, so it is very suitable for large-scale application and promotion in the field of engine blade damage detection.

[0019] In a possible design, a signal recognition process is performed on the wavelet decomposition image signal corresponding to each wavelet coefficient to recognize a noise image signal from a plurality of wavelet decomposition image signals, including:

[0020] For any wavelet decomposition image signal, obtaining all pixel points in the wavelet decomposition image signal;

[0021] Based on all the pixels in any one of the wavelet decomposition image signals, and using a binary classification algorithm, determine a binary classification value of any one of the wavelet decomposition image signals;

[0022] According to the binary classification value of any wavelet decomposition image signal, the signal classification result of any wavelet decomposition image signal is determined, wherein if the binary classification value of any wavelet decomposition image signal is less than 0, the signal classification result of any wavelet decomposition image signal is obtained as a noise image signal, and if the binary classification value of any wavelet decomposition image signal is greater than or equal to 0, the signal classification result of any wavelet decomposition image signal is obtained as a non-noise image signal.

[0023] In a possible design, based on all pixels in any wavelet decomposition image signal, and using a binary classification algorithm, determining a binary classification value of any wavelet decomposition image signal includes:

[0024] According to the coordinates of each pixel point in any wavelet decomposition image signal, and according to the following formula (1), a binary classification bias value is determined;

[0025]

[0026] In the above formula (1), represents the binary classification bias value, δ represents the Lagrange multiplier, x j ,y j are the horizontal coordinate and vertical coordinate of the j-th pixel point in any wavelet decomposition image signal, n represents the total number of pixels in any wavelet decomposition image signal, and λ1 and λ2 both represent binary classification parameters;

[0027] Using the binary classification offset value and the coordinates of each pixel point in any wavelet decomposition image signal, and according to the following formula (2), determine the binary classification value of any wavelet decomposition image signal;

[0028]

[0029] In the above formula (2), Y represents the binary classification value of any wavelet decomposition image signal, K() represents the Gaussian radial basis kernel function, and sign[] represents the sign function, wherein when the value in the square brackets is less than 0, the value returned by sign[] is -1, and when the value in the square brackets is greater than 0, the value returned by sign[] is 1.

[0030] In a possible design, based on the wavelet coefficients of the noisy image signal, denoising the noisy image signal to obtain a denoised image signal includes:

[0031] Acquire an initial wavelet threshold, and determine a first wavelet threshold and a second wavelet threshold based on the initial wavelet threshold, wherein the first wavelet threshold and the second wavelet threshold are both greater than 0, and the first wavelet threshold is less than the initial wavelet threshold, and the initial wavelet threshold is less than the second wavelet threshold;

[0032] Determining a wavelet adjustment parameter based on the wavelet coefficient of the noise image signal, the first wavelet threshold, and the second wavelet threshold;

[0033] The wavelet coefficients of the noisy image signal are adjusted according to the wavelet adjustment parameter, the first wavelet threshold, and the second wavelet threshold, so as to obtain the denoised image signal after the wavelet coefficients are adjusted.

[0034] In a possible design, determining a wavelet adjustment parameter based on the wavelet coefficient of the noise image signal, the first wavelet threshold, and the second wavelet threshold includes:

[0035] According to the following formula (3), the wavelet adjustment parameter is calculated;

[0036]

[0037] In the above formula (3), μ represents the wavelet adjustment parameter, σ1 and σ2 represent the first wavelet threshold and the second wavelet threshold respectively, and a represents the wavelet coefficient of the noise image signal;

[0038] Correspondingly, adjusting the wavelet coefficients of the noise image signal according to the wavelet adjustment parameter, the first wavelet threshold, and the second wavelet threshold includes:

[0039] According to the following formula (4), the wavelet coefficients of the noise image signal are adjusted;

[0040]

[0041] In the above formula (4), a′ represents the adjusted wavelet coefficient, and sign() represents the sign function.

[0042] In a possible design, the optimal model parameters are determined by the improved genetic algorithm, including:

[0043] Obtain multiple groups of model parameters, and use each group of model parameters as a chromosome to form an initial chromosome population;

[0044] Based on the model parameters corresponding to each chromosome in the initial chromosome population, a number of initial aircraft blade damage recognition models are constructed, and each initial aircraft blade damage recognition model is trained using a training set to obtain a number of trained initial aircraft blade damage recognition models, wherein the training set includes sample denoised boresight images of engine blades of a number of sample aircraft;

[0045] Inputting the same test data into each trained initial aircraft blade damage identification model to perform blade damage identification processing, so as to obtain model outputs of each trained initial aircraft blade damage identification model respectively;

[0046] Calculating the fitness of each chromosome according to the model output of each trained initial aircraft blade damage recognition model and the label value of the test data;

[0047] Filter out the global optimal fitness from various fitnesses, and determine whether the global optimal fitness meets the preset conditions;

[0048] If not, then according to the fitness of each chromosome, a number of chromosomes are selected from the initial chromosome population to form a parent chromosome population;

[0049] Selecting a plurality of parent chromosomes from the parent chromosome population, and using an improved crossover algorithm to perform individual crossover operations on the selected plurality of parent chromosomes, so as to obtain a plurality of crossover chromosomes after the individual crossover operations;

[0050] Selecting a number of target chromosomes from the target chromosome population, and using the improved mutation algorithm to perform mutation operations on each of the selected target chromosomes, so as to obtain a number of mutant chromosomes after the mutation operation, wherein the target chromosome population includes a plurality of crossover chromosomes and unselected parent chromosomes in the parent chromosome population;

[0051] Using the mutant chromosomes and the unselected target chromosomes in the target chromosome population, a progeny chromosome population is formed;

[0052] The initial chromosome population is updated to the offspring chromosome population, and a number of initial aircraft blade damage identification models are reconstructed based on the model parameters corresponding to each chromosome in the initial chromosome population until the global optimal fitness meets the preset conditions, and the model parameters of the chromosome corresponding to the global optimal fitness that meets the preset conditions are used as the optimal parameters of the model.

[0053] In a possible design, according to the fitness of each chromosome, several chromosomes are selected from the chromosome population to form a parent chromosome population, including:

[0054] Sort each chromosome in descending order of fitness to obtain a chromosome sorting sequence;

[0055] Classifying the chromosomes in the chromosome sorting sequence into three categories from front to back to obtain a first category of chromosomes, a second category of chromosomes and a third category of chromosomes, respectively, wherein the fitness of the last chromosome in the first category of chromosomes is greater than the fitness of the first chromosome in the second category of chromosomes, and the fitness of the last chromosome in the second category of chromosomes is greater than the fitness of the first chromosome in the third category of chromosomes;

[0056] An individual selection probability is obtained, and a plurality of chromosomes are selected from the third type of chromosomes based on the individual selection probability, and the parent chromosome population is formed using the first type of chromosomes, the second type of chromosomes, and a plurality of chromosomes selected from the third type of chromosomes.

[0057] In a possible design, an improved crossover algorithm is used to perform individual crossover operations on multiple selected parent chromosomes, so as to obtain multiple crossover chromosomes after the individual crossover operations, including:

[0058] For any two parent chromosomes among the selected multiple parent chromosomes, obtaining genes at the crossover position of the any two parent chromosomes;

[0059] Generate a first random number between 0 and 1 for any two parent chromosomes;

[0060] According to the first random number and the two genes at the crossover position, and in accordance with the following formula (5), an individual crossover operation is performed on the arbitrary two parent chromosomes, so as to obtain two crossover chromosomes after the individual crossover operation;

[0061]

[0062] In the above formula (5), G b,d represents the gene at the crossover position of one of the two parent chromosomes, G′b,d represents the gene after the crossover of one of the parent chromosomes at the crossover position, G c,d represents the gene of the other parent chromosome in the crossover position of any two parent chromosomes, G′ c,d represents the gene of the other parent chromosome after crossover at the crossover position, and p represents the first random number.

[0063] In a possible design, the improved mutation algorithm is used to perform mutation operations on each selected target chromosome, so that after the mutation operation, several mutant chromosomes are obtained, including:

[0064] For any target chromosome among the selected target chromosomes, generate a second random number corresponding to the target chromosome, wherein the value interval of the second random number is [0, 1];

[0065] Determine the variation position of any target chromosome, and obtain the gene of any target chromosome at the variation position and the maximum value and minimum value of the gene at the variation position;

[0066] According to the maximum value, the minimum value, the second random number and the gene of any target chromosome at the mutation position, the mutated gene of any target chromosome at the mutation position is determined by using the following formula (6);

[0067]

[0068] In the above formula (6), G′ s represents the gene of any target chromosome after mutation at the mutation position, G s represents the gene at the variant position of any target chromosome, G max represents the maximum value, G min represents the minimum value, θ represents the second random number, τ represents the current iteration number, τ max Indicates the maximum number of iterations;

[0069] The gene at the mutation position of any target chromosome is updated to the mutated gene, so as to obtain the mutated chromosome corresponding to any target chromosome after the update.

[0070] In a second aspect, a health reliability analysis system for an aircraft is provided, comprising:

[0071] An acquisition unit, used for acquiring a borescope image of an engine blade of a target aircraft;

[0072] A wavelet decomposition unit, used for performing wavelet decomposition processing on the borescope image to obtain wavelet coefficients at different decomposition scales and a wavelet decomposition image signal corresponding to each wavelet coefficient;

[0073] A signal recognition unit, used for performing signal recognition processing on the wavelet decomposition image signal corresponding to each wavelet coefficient, so as to recognize a noise image signal from a plurality of wavelet decomposition image signals;

[0074] A denoising unit, configured to perform denoising processing on the noisy image signal based on the wavelet coefficients of the noisy image signal to obtain a denoised image signal;

[0075] The denoising unit is further used to perform wavelet reconstruction processing using the target image signal and the denoised image signal to obtain a denoised boresight image after the wavelet reconstruction processing, wherein the target image signal is all wavelet decomposed image signals among the plurality of wavelet decomposed image signals except the noise image signal;

[0076] A blade damage identification unit is used to obtain an aircraft blade damage identification model, wherein the aircraft blade damage identification model is obtained by training by determining optimal model parameters through an improved genetic algorithm, and taking sample denoised boresight images of engine blades of several sample aircraft as inputs and outputting damage identification results of engine blades of each sample aircraft as outputs based on the optimal model parameters;

[0077] The blade damage identification unit is further used to input the denoised boresight image into the aircraft blade damage identification model for damage identification processing to obtain a damage identification result of the engine blade of the target aircraft;

[0078] The reliability analysis unit is used to obtain a health reliability analysis result of the target aircraft based on the damage identification result of the engine blade of the target aircraft.

[0079] According to a third aspect, another device for analyzing the health and reliability of an aircraft is provided. Taking the device as an electronic device as an example, the device includes a memory, a processor and a transceiver which are communicatively connected in sequence, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute a method for analyzing the health and reliability of the aircraft as in the first aspect or any possible design of the first aspect.

[0080] In a fourth aspect, a storage medium is provided, on which instructions are stored. When the instructions are executed on a computer, the health reliability analysis method of the aircraft as in the first aspect or any possible design of the first aspect is executed.

[0081] In a fifth aspect, a computer program product comprising instructions is provided, which, when executed on a computer, causes the computer to execute the health reliability analysis method for an aircraft as described in the first aspect or any possible design of the first aspect.

[0082] Beneficial effects:

[0083] (1) Before damage detection, the present invention uses wavelet decomposition to decompose the original image, so as to obtain image signals of different frequency bands and frequencies; then, the image signals of different frequency bands and frequencies are identified to obtain image signals belonging to noise, and then the image signals belonging to noise are denoised to obtain denoised image signals; finally, the denoised image signals and non-noise image signals are used to reconstruct the image, so as to obtain denoised borescope images; thus, compared with traditional filtering denoising, the present invention can remove noise of different frequency bands and frequencies to the greatest extent, thereby improving the denoising effect of mixed noise; at the same time, when training the blade damage model, the present invention uses an improved genetic algorithm to determine the optimal parameters of the model. Compared with traditional technologies, the present invention can use genetic populations to evaluate multiple groups of model parameters in one iteration process, thereby reducing the model training time and improving the damage recognition efficiency; based on this, the present invention can reduce the model training time while ensuring the recognition accuracy, and therefore is very suitable for large-scale application and promotion in the field of engine blade damage detection.

[0084] (2) When the present invention utilizes wavelet coefficients for denoising, two wavelet thresholds are re-determined based on the initial wavelet threshold, and based on this, the wavelet adjustment parameters are calculated; then, an adaptive wavelet coefficient adjustment function is constructed according to the wavelet adjustment parameters and the aforementioned two wavelet thresholds; thus, compared with the traditional hard threshold adjustment method, the present invention can maintain the smoothness and detail information of the image after denoising, thereby further improving the denoising effect.

[0085] (3) When determining the optimal parameters of the model, the present invention uses an improved crossover and mutation algorithm to perform individual crossover and mutation operations. In the crossover operation, the traditional gene position exchange is no longer used, but the first random number and gene value of the two crossover parties are introduced to recalculate the gene; similarly, during mutation, the second random number and gene corresponding to the mutated individual are introduced to recalculate; in this way, compared with traditional technologies, the population can be made more diversified, thereby avoiding the problem of traditional genetic algorithms being prone to falling into local optimality. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 A schematic flow chart of the steps of a method for analyzing the health reliability of an aircraft provided by an embodiment of the present invention;

[0087] Figure 2A structural diagram of an aircraft blade damage identification model provided by an embodiment of the present invention;

[0088] Figure 3 A schematic diagram of the structure of a feature extraction module provided in an embodiment of the present invention;

[0089] Figure 4 A schematic diagram of the structure of a health reliability analysis system for an aircraft provided by an embodiment of the present invention;

[0090] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0091] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0092] It should be understood that although the terms first, second, etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another unit. For example, a first unit can be referred to as a second unit, and similarly, a second unit can be referred to as a first unit without departing from the scope of the exemplary embodiments of the present invention.

[0093] It should be understood that the term "and / or" that may appear in this article is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" that may appear in this article describes another type of association object relationship, indicating that two relationships may exist. For example, A / and B can represent two situations: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this article generally indicates that the previous and next associated objects are in an "or" relationship.

[0094] Example:

[0095] See also Figure 1As shown, the aircraft health reliability analysis method provided by this embodiment, when performing denoising, separates the signals of different frequencies and frequency bands in the original image, and performs noise identification respectively, so as to accurately identify the image noise in different frequencies and frequency bands in the original image; then, the identified image noise is denoised using wavelet coefficients to obtain a denoised signal; then, the image is reconstructed with the non-noise signal to obtain a denoised boresight image; finally, the denoised boresight image is input into the aircraft blade damage identification model to obtain a damage identification result; thus, compared with the traditional technology, the present method can remove the signals of different frequency bands and frequencies to the greatest extent. Noise reduces the influence of mixed noise on damage identification, thereby improving the accuracy of model identification; at the same time, when performing damage identification, the optimal parameters of the model are determined by the improved genetic algorithm, and multiple groups of model parameters can be evaluated in one iteration. Therefore, the model training time can be reduced, thereby improving the recognition efficiency; based on this, the method can reduce the model training time while ensuring the recognition accuracy, and is very suitable for large-scale application and promotion in the field of engine blade damage detection; among them, for example, the method can be but not limited to running on the aircraft detection end side, optionally, the aircraft detection end can be but not limited to a personal computer (PC), a tablet computer or a smart phone. It can be understood that the aforementioned execution subject does not constitute a limitation on the embodiments of the present application. Accordingly, the operation steps of the method can be but not limited to the following steps S1 to S8.

[0096] S1. Acquire a boresight image of the engine blade of the target aircraft; in this embodiment, a boresight device may be used to acquire a boresight image of the engine blade, and then the boresight image is transmitted to the aircraft detection end for damage detection of the engine blade; wherein, before damage detection, the boresight image needs to be denoised to reduce the influence of mixed noise in the image on subsequent damage identification; specifically, the image denoising process may be, but is not limited to, as shown in the following steps S2 to S5.

[0097] S2. Perform wavelet decomposition processing on the borescope image to obtain wavelet coefficients at different decomposition scales and wavelet decomposition image signals corresponding to each wavelet coefficient; in specific implementation, before performing wavelet decomposition, the borescope image needs to be grayscaled to reduce the amount of data to be processed; wherein, this embodiment uses an improved grayscale conversion algorithm to perform grayscale processing on the borescope image to obtain a grayscale image; specifically, the aforementioned improved grayscale conversion algorithm can be expressed by the following formula (7).

[0098] f(l)=0.299×R(l)+0.518×G(l)+0.114×B(l) (7)

[0099] In the above formula (7), f(l) represents the grayscale value of the lth pixel in the boresight image (l=1, 2, ..., L, L is the total number of pixels in the boresight image), and R(l), G(l), and B(l) respectively represent the red component value, green component value, and blue component value of the lth pixel in the boresight image; thus, based on the above formula (7), different weighted values ​​for the three primary colors can be used to perform grayscale conversion, thereby improving the grayscale conversion effect.

[0100] At the same time, since the probe of the borescope equipment has its own light source to facilitate photography, there will be local light noise in the obtained borescope image. Therefore, in order to reduce this noise, this embodiment also performs contrast adjustment processing on the grayscale image to obtain an adjusted grayscale image; optionally, the following formula (8) may be used, for example but not limited to, to perform contrast adjustment processing on the grayscale image.

[0101] f′(l)=βlg(1+f(l)) (8)

[0102] In the above formula (8), f′(l) represents the adjusted grayscale value corresponding to the l-th pixel, and β represents the grayscale adjustment coefficient. In this embodiment, the grayscale adjustment coefficient can be, but is not limited to, 20.

[0103] Thus, after the aforementioned processing, the adjusted grayscale image can be subjected to wavelet decomposition processing, thereby obtaining signals at different frequencies and frequency bands in the original image, as well as the wavelet coefficients corresponding to each signal; of course, wavelet decomposition is a commonly used technique for signal processing, and its principle will not be repeated here.

[0104] After obtaining the image signals at different frequencies and frequency bands in the original image, noise recognition can be performed to identify the signals belonging to the noise from the several image signals obtained by decomposition (i.e., the aforementioned wavelet decomposition image signals); wherein the noise recognition process can be but is not limited to the one shown in the following step S3.

[0105] S3. Perform signal recognition processing on the wavelet decomposition image signal corresponding to each wavelet coefficient to identify the noise image signal from a number of wavelet decomposition image signals; in specific applications, this embodiment uses a binary classification algorithm to perform binary classification on each wavelet decomposition image, so as to determine the signal category of each wavelet decomposition image; optionally, the specific process of using the binary classification algorithm for signal recognition can be but is not limited to the following steps S31 to S33.

[0106] S31. For any wavelet decomposition image signal, obtain all pixel points in the wavelet decomposition image signal.

[0107] S32. Based on all the pixels in any one of the wavelet decomposition image signals and using a binary classification algorithm, determine the binary classification value of any one of the wavelet decomposition image signals; in specific implementation, it can be but not limited to first determining the binary classification bias value based on the coordinates of each pixel point in any one of the wavelet decomposition image signals; then, use the binary classification bias value and the coordinates of each pixel point in any one of the wavelet decomposition image signals to determine the binary classification value of any one of the wavelet decomposition image signals.

[0108] Optionally, the binary classification bias value may be calculated by, for example but not limited to, using the following formula (1).

[0109]

[0110] In the above formula (1), represents the binary classification bias value, δ represents the Lagrange multiplier, x j ,y j are respectively the abscissa and ordinate of the j-th pixel point in any wavelet decomposition image signal, n represents the total number of pixels in any wavelet decomposition image signal, and λ1 and λ2 both represent binary classification parameters.

[0111] In this embodiment, the binary classification algorithm is a machine learning algorithm, wherein the aforementioned binary classification category parameters can be obtained by machine learning; specifically, the binary classification algorithm can be used to construct a binary classification model (such as a support vector machine model), and then, the sample pixel points of the image signal are decomposed using different sample wavelets to train the binary classification model, thereby obtaining the corresponding binary classification category parameters; specifically, the binary classification category parameters can be learned as feature vectors belonging to noise signals and feature vectors belonging to non-noise signals.

[0112] Based on this, after obtaining the binary classification parameters, they can be substituted into the aforementioned formula (1) to obtain the binary classification bias value used in this signal recognition; and after obtaining the binary classification bias value, the following formula (2) can be used to calculate the binary classification value in any of the aforementioned wavelet decomposition image signals.

[0113]

[0114] In the above formula (2), Y represents the binary classification value of any wavelet decomposition image signal, K() represents the Gaussian radial basis kernel function, and sign[] represents the sign function, wherein when the value in the square brackets is less than 0, the value returned by sign[] is -1, and when the value in the square brackets is greater than 0, the value returned by sign[] is 1; in this embodiment, when the value in the square brackets is equal to 0, it returns 0; in this way, based on the above formula (2), the binary classification value can be mapped to -1, 0 or 1; then, the signal classification result of any wavelet decomposition image signal can be determined according to the above binary classification value, and the process is shown in the following step S33.

[0115] S33. According to the binary classification value of any wavelet decomposition image signal, determine the signal classification result of any wavelet decomposition image signal, wherein if the binary classification value of any wavelet decomposition image signal is less than 0, then the signal classification result of any wavelet decomposition image signal is obtained as a noise image signal; if the binary classification value of any wavelet decomposition image signal is greater than or equal to 0, then the signal classification result of any wavelet decomposition image signal is obtained as a non-noise image signal;

[0116] Therefore, through the aforementioned steps S31 to S33, the noise image signal can be identified from all wavelet decomposed image signals; then, the wavelet coefficients corresponding to the noise image signal can be used to perform signal denoising; wherein, the signal denoising process can be but is not limited to the following step S4.

[0117] S4. Based on the wavelet coefficients of the noise image signal, the noise image signal is denoised to obtain a denoised image signal; in specific applications, this embodiment is improved on the basis of traditional hard threshold denoising to overcome the problem of discontinuity of the denoised image existing in hard threshold denoising; specifically, the aforementioned denoising process is shown in the following steps S41 to S43.

[0118] S41. Obtain an initial wavelet threshold, and based on the initial wavelet threshold, determine a first wavelet threshold and a second wavelet threshold, wherein the first wavelet threshold and the second wavelet threshold are both greater than 0, and the first wavelet threshold is less than the initial wavelet threshold, and the initial wavelet threshold is less than the second wavelet threshold; in specific applications, the initial wavelet threshold can be estimated using noise variance, which is a commonly used estimation method for wavelet thresholds in wavelet denoising, and its principle will not be repeated here.

[0119] In this way, after randomly generating the first wavelet threshold and the second wavelet threshold that meet the above-mentioned conditions based on the initial wavelet threshold, the wavelet adjustment parameters can be determined based on the above-mentioned first wavelet threshold and the second wavelet threshold, so as to subsequently adjust the wavelet coefficients of the denoised image signal based on the wavelet adjustment parameters; wherein the specific calculation process of the wavelet adjustment parameters can be but is not limited to as shown in the following step S42.

[0120] S42. Determine a wavelet adjustment parameter based on the wavelet coefficient of the noise image signal, the first wavelet threshold and the second wavelet threshold; in specific implementation, the wavelet adjustment parameter can be calculated by, for example but not limited to, using the following formula (3).

[0121]

[0122] In the above formula (3), μ represents the wavelet adjustment parameter, σ1 and σ2 represent the first wavelet threshold and the second wavelet threshold respectively, and a represents the wavelet coefficient of the noise image signal.

[0123] After the wavelet adjustment parameter is calculated based on the above formula (3), the wavelet adjustment parameter, the above first wavelet threshold and the second wavelet threshold can be used to adjust the wavelet coefficients of the noise image signal; wherein the adjustment process of the wavelet coefficients is shown in the following step S43.

[0124] S43. According to the wavelet adjustment parameter, the first wavelet threshold and the second wavelet threshold, the wavelet coefficients of the noise image signal are adjusted to obtain the denoised image signal after the wavelet coefficients are adjusted. In this embodiment, the wavelet coefficients of the noise image signal can be adjusted according to the following formula (4) for example but not limited thereto.

[0125]

[0126] In the above formula (4), a′ represents the adjusted wavelet coefficient, and sign() represents the sign function.

[0127] Based on the above formula (4), it can be known that when the absolute value of the wavelet coefficient of the noise image signal is less than the first wavelet threshold, it is set to 0; at the same time, since the wavelet coefficient is a constant of the signal item corresponding to the noise image signal in the wavelet decomposition, therefore, setting it to 0 is equivalent to changing the component of the noise image signal to 0. At this time, the noise image signal is filtered; based on this, by adjusting the value of the wavelet coefficient of the noise image signal, the noise signal can be filtered out or weakened, thereby obtaining a denoised image signal; at the same time, this embodiment adopts an improved soft threshold denoising method for denoising, which can ensure the smoothness of denoising compared with the traditional technology, thereby further improving the denoising effect.

[0128] Thus, through the aforementioned steps S41 to S43, the denoising process of the noisy image signal can be completed, thereby obtaining a denoised image signal; then, it can be combined with a number of non-noise image signals in the wavelet decomposition image signals, and wavelet reconstruction processing can be performed, and after the wavelet reconstruction processing, a denoised boresight image can be obtained; wherein, the image reconstruction process can be but is not limited to as shown in the following step S5.

[0129] S5. Perform wavelet reconstruction processing using the target image signal and the denoised image signal to obtain a denoised boresight image after the wavelet reconstruction processing, wherein the target image signal is all wavelet decomposition image signals except the noise image signal from a plurality of wavelet decomposition image signals; in this embodiment, wavelet reconstruction is a commonly used technology for signal denoising, and its principle will not be described in detail; and after obtaining the denoised boresight image, a pre-trained aircraft blade damage recognition model can be obtained, so as to perform damage recognition of the engine blade of the target aircraft based on the model.

[0130] Optionally, the damage identification process may be, but is not limited to, as shown in the following steps S6 and S7.

[0131] S6. Obtain an aircraft blade damage recognition model, wherein the aircraft blade damage recognition model is obtained by training by determining optimal model parameters through an improved genetic algorithm, and based on the optimal model parameters, taking sample denoised boresight images of engine blades of several sample aircraft as input and outputting damage recognition results of engine blades of each sample aircraft; in specific applications, the aircraft blade damage recognition model may be, but is not limited to, an RBF neural network (Radial basis function network); of course, it may also be other neural networks, and in this embodiment, an improved YOLOV4 neural network is provided, and its structure is described in detail below.

[0132] At the same time, since the traditional blade damage recognition model uses a Bayesian optimization search method to determine the optimal parameters of the model during training, one iteration can only evaluate one set of model parameters, which leads to the problem of long training time. This embodiment provides a new method for determining the optimal parameters of the model to reduce the model training time and thus improve the recognition efficiency; specifically, the process of determining the optimal parameters of the aforementioned model can be but is not limited to the following steps S61 to S610.

[0133] S61. Obtain multiple groups of model parameters, and take each group of model parameters as a chromosome to form an initial chromosome population; in this embodiment, taking the aforementioned improved YOLOV4 neural network as an example, a group of model parameters may include but is not limited to the weights of each convolution kernel in the network, regularization coefficient, learning rate, batch size, etc.; wherein, the aforementioned multiple groups of model parameters may be randomly generated first, or selected based on experience, and pre-stored in the aircraft detection end, so that they can be called when used; at the same time, for example, each group of model parameters can be encoded to obtain the corresponding chromosome.

[0134] After the initial chromosome population is constructed based on the aforementioned step S61, the model parameters corresponding to each chromosome in the initial chromosome population can be used to construct an initial aircraft blade damage identification model corresponding to each model parameter, so as to subsequently calculate the fitness of each chromosome according to each initial aircraft blade damage identification model; wherein the aforementioned process can be but is not limited to as shown in the following steps S62 to S64.

[0135] S62. Based on the model parameters corresponding to each chromosome in the initial chromosome population, a number of initial aircraft blade damage recognition models are constructed, and each initial aircraft blade damage recognition model is trained using a training set to obtain a number of trained initial aircraft blade damage recognition models, wherein the training set includes sample denoised boresight images of engine blades of a number of sample aircrafts; in this embodiment, it is equivalent to using the same training set to train each initial aircraft blade damage recognition model, thereby obtaining a number of trained models; of course, the aforementioned training set includes different sample denoised boresight images, and labels corresponding to each sample denoised boresight image; further, the sample denoised boresight images may include denoised boresight images of different damage types, and denoised boresight images of normal blades; of course, the corresponding labels are the damage types (such as chipping, cracking, corrosion or no damage, etc., in this embodiment, different damage types are represented by numbers respectively); in this way, after training, the same test data can be used to perform model testing, so as to calculate the fitness of each of the aforementioned chromosomes based on the model output results, and the process is shown in the following steps S63 and S64.

[0136] S63. Input the same test data into each trained initial aircraft blade damage recognition model for blade damage recognition processing, so as to obtain the model output of each trained initial aircraft blade damage recognition model respectively; in this embodiment, the test data is the denoised boresight image of the aircraft to be detected, and the model output is the damage recognition result; at the same time, the damage recognition result can be represented by, but not limited to, category values, that is, different damage categories correspond to different category values, and of course, also correspond to their own label values.

[0137] After obtaining the model output of each trained initial aircraft blade damage recognition model, the fitness of each chromosome can be calculated based on the label value of the test data, so as to determine whether the genetic iteration can be ended based on the fitness; wherein, the fitness calculation process is shown in the following step S64.

[0138] S64. The fitness of each chromosome is calculated based on the model output of each trained initial aircraft blade damage identification model and the label value of the test data. In specific applications, the error between the model output (i.e., predicted value) and the actual value (i.e., label value) can be used as the fitness function, for example but not limited to. In this way, the fitness of each chromosome can be calculated based on the model output of each of the aforementioned trained initial aircraft blade damage identification models.

[0139] After calculating the fitness of each chromosome, the global optimal fitness can be screened out to determine whether it meets the iteration stop condition; wherein the above process is shown in the following step S65.

[0140] S65. Filter out the global optimal fitness from each fitness, and determine whether the global optimal fitness meets the preset conditions; in this embodiment, the global optimal fitness is the maximum fitness in all iterations; wherein, since this is the beginning of the iteration, the maximum fitness among the corresponding fitness of each chromosome is the global optimal fitness; at the same time, the preset condition can be, but is not limited to, the global optimal fitness being less than a set threshold; of course, the set threshold can be set specifically according to the use, and is not specifically limited here.

[0141] In specific applications, if the global optimal fitness does not meet the preset conditions, genetic iteration is required, that is, the first iteration; wherein, a genetic iteration process mainly includes individual selection, crossover and mutation operations, wherein, in order to avoid the optimization falling into the local optimal solution, this embodiment improves the mutation and crossover (described in detail below); optionally, the genetic iteration process can be but is not limited to the following steps S66 to S610.

[0142] S66. If not, then select a number of chromosomes from the initial chromosome population according to the fitness of each chromosome to form the parent chromosome population; in specific implementation, although the traditional optimal individual selection principle can reduce the risk of the optimal individual being eliminated, selecting the optimal individual for inheritance will reduce the diversity of individuals, which will lead to the problem of falling into a local optimal solution; therefore, this embodiment adopts an improved individual selection method to perform individual selection, and the process is: first, sort each chromosome in order from high to low fitness to obtain a chromosome sorting sequence; then, classify the chromosomes in the chromosome sorting sequence into three categories from front to back to obtain the first category chromosomes, the second category chromosomes and the third category chromosomes respectively; wherein, in this embodiment, the first The fitness of the last chromosome in one category of chromosomes is greater than the fitness of the first chromosome in the second category of chromosomes, and the fitness of the last chromosome in the second category of chromosomes is greater than the fitness of the first chromosome in the third category of chromosomes; in this way, the chromosomes are classified in order of fitness from high to low, thereby dividing them into three categories; specifically, the number of each category is the total number of chromosomes divided by 3; and after the division of chromosomes is completed, individual selection can be performed, that is: obtaining the individual selection probability, and selecting multiple chromosomes from the third category of chromosomes based on the individual selection probability, and using the first category of chromosomes, the second category of chromosomes and the multiple chromosomes selected from the third category of chromosomes to form the parent chromosome population.

[0143] Thus, through the aforementioned improved individual selection method, this embodiment can ensure that the best individual will not be eliminated, and can also ensure the diversity of individuals, thereby avoiding the problem of falling into the local optimal solution; thus, the accuracy of model parameter optimization can be improved.

[0144] After completing the individual selection and obtaining the parent chromosome population, an individual crossover operation can be performed; wherein the individual crossover process can be but is not limited to the one shown in the following step S67.

[0145] S67. Select multiple parent chromosomes from the parent chromosome population, and use the improved crossover algorithm to perform individual crossover operations on the selected multiple parent chromosomes, so as to obtain multiple crossover chromosomes after the individual crossover operations; in this embodiment, in order to improve the diversity of crossover, the following steps S671 to S673 can be used, for example but not limited to, to perform individual crossover operations.

[0146] S671. For any two parent chromosomes among the selected multiple parent chromosomes, obtain the genes at the crossover position of the any two parent chromosomes; in this embodiment, the crossover position can be but is not limited to being pre-set, and the crossover position of this embodiment is set to one, that is, a single-point crossover operation is performed; in this way, after obtaining the genes at the crossover position of any two parent chromosomes, a first random number can be introduced to subsequently combine the genes at the two crossover positions to calculate a new gene for crossover; wherein, the calculation process of the new gene can be but is not limited to as shown in the following steps S672 and step S673.

[0147] S672. Generate a first random number between 0 and 1 for any two parent chromosomes.

[0148] S673. Based on the first random number and the two genes at the crossover position, and in accordance with the following formula (5), an individual crossover operation is performed on the arbitrary two parent chromosomes to obtain two crossover chromosomes after the individual crossover operation.

[0149]

[0150] In the above formula (5), G b,d represents the gene at the crossover position of one of the two parent chromosomes, G′ b,d represents the gene after the crossover of one of the parent chromosomes at the crossover position, G c,d represents the gene of the other parent chromosome in the crossover position of any two parent chromosomes, G′ c,d represents the gene of the other parent chromosome after crossover at the crossover position, and p represents the first random number.

[0151] Thus, by using the aforementioned formula (5), new genes can be calculated for the two parent chromosomes that need to undergo a crossover operation, so that the calculated genes can be used to replace the original genes at their respective crossover positions; of course, each gene on the chromosome refers to a code value in the code string of the corresponding model parameter.

[0152] Therefore, based on the aforementioned steps S671 to S673, the individual crossover operation of each selected parent chromosome can be completed; then, the individual mutation operation can be performed; wherein, the individual mutation operation process is shown in the following step S68.

[0153] S68. Select several target chromosomes from the target chromosome population, and use the improved mutation algorithm to perform mutation operations on each selected target chromosome to obtain several mutant chromosomes after the mutation operation, wherein the target chromosome population includes multiple crossover chromosomes and unselected parent chromosomes in the parent chromosome population; in the specific implementation, this embodiment takes any selected target chromosome as an example for explanation, and its mutation process can be but is not limited to as shown in the following steps S681 to S684.

[0154] S681. For any target chromosome among the selected target chromosomes, generate a second random number corresponding to the target chromosome, wherein the value interval of the second random number is [0,1]; in the present embodiment, the second random number is any number between [0,1], which can be randomly selected each time; after obtaining the second random number, the mutation position and the maximum and minimum values ​​of the gene at the mutation position can be determined, so as to subsequently calculate the mutated gene based on the aforementioned data; wherein the aforementioned process is shown in the following step S682.

[0155] S682. Determine the mutation position of any target chromosome, and obtain the gene of any target chromosome at the mutation position and the maximum and minimum values ​​of the gene at the mutation position; in this embodiment, the mutation position is also pre-set in the aircraft detection end, and can be called when used, and the same is true for the maximum and minimum values ​​of the gene at the mutation position, and the acquisition principle will not be repeated.

[0156] After determining the relevant parameters of the gene at the mutation position of any of the aforementioned target chromosomes, the gene mutation operation can be performed, and the process is shown in the following steps S683 and S684.

[0157] S683. Based on the maximum value, the minimum value, the second random number and the gene of any target chromosome at the mutation position, and using the following formula (6), determine the gene of any target chromosome at the mutation position after mutation.

[0158]

[0159] In the above formula (6), G′ s represents the gene of any target chromosome after mutation at the mutation position, G s represents the gene at the variant position of any target chromosome, G max represents the maximum value, G min represents the minimum value, θ represents the second random number, τ represents the current iteration number, τ max Indicates the maximum number of iterations.

[0160] After the mutated gene is calculated based on the above formula (6), the mutated gene can be used to replace the original gene, thereby realizing the mutation operation of any target chromosome; wherein the above process is shown in the following step S694.

[0161] S684. Update the gene at the mutation position of any target chromosome to the mutated gene, so as to obtain a mutated chromosome corresponding to any target chromosome after the update.

[0162] In this way, through the aforementioned steps S681 to S684, the mutation operation of each selected target chromosome can be completed; of course, the selection of the aforementioned chromosomes for individual crossover and mutation is achieved based on the set crossover probability and mutation probability; at the same time, the aforementioned improved individual crossover method is used to perform the crossover operation, and its crossover is no longer the traditional direct exchange of gene positions, but the gene is recalculated; similarly, when mutating, the second random number and gene corresponding to the mutated individual are introduced for recalculation; in this way, compared with traditional technologies, the population can be made more diversified, thereby avoiding the problem of easy falling into local optimality in traditional genetic algorithms; in addition, the aforementioned mutation method can also enhance the local search ability of the algorithm, and the mutation amplitude gradually decreases with time, so that the convergence speed of the algorithm can be improved.

[0163] After completing the individual mutation operation, the mutated chromosomes and the unselected target chromosomes can be used to form the offspring chromosome population generated by this iteration; then, the initial chromosome population is updated to the offspring chromosome population, and the aforementioned steps S62 to S610 are re-executed until the preset conditions are met, and the optimal parameters of the model can be obtained; wherein, the loop process is shown in the following steps S69 and S610.

[0164] S69. Use the mutant chromosomes and the unselected target chromosomes in the target chromosome population to form a progeny chromosome population.

[0165] S610. Update the initial chromosome population to the offspring chromosome population, and re-construct several initial aircraft blade damage identification models based on the model parameters corresponding to each chromosome in the initial chromosome population, until the global optimal fitness meets the preset conditions, and use the model parameters of the chromosome corresponding to the global optimal fitness that meets the preset conditions as the model optimal parameters; in this embodiment, after completing this iteration, when performing the next iteration, the number of iterations will be increased by 1, and the above steps S62 to S610 will be continuously repeated until the global optimal solution obtained by the iteration meets the above preset conditions; at this time, the model parameters corresponding to the chromosome of the global optimal solution that meets the preset conditions can be used as the model optimal parameters; of course, when the number of iterations reaches the maximum number of iterations, the loop can also be ended.

[0166] Therefore, through the aforementioned steps S61 to S610, the improved genetic algorithm can be used to complete the optimization of the model parameters of the aircraft blade damage recognition model, so as to obtain the model parameters with the highest recognition accuracy; then, the model parameters can be used to train the model to obtain the aircraft blade damage recognition model; of course, since the model corresponding to each chromosome has been trained in the aforementioned optimization process, the initial aircraft blade damage recognition model after training of the chromosome corresponding to the global optimal solution that meets the preset conditions can also be directly used as the final model for the subsequent blade damage recognition of the target aircraft.

[0167] In a specific embodiment, the following discloses a specific structure of an aircraft blade damage identification model:

[0168] See also Figure 2 As shown, for example, it may include but is not limited to: a feature extraction network, a target detection network and a damage recognition network, wherein the feature extraction network may include but is not limited to: a first convolutional structure layer and four feature extraction modules connected in sequence (numbered first to fourth in sequence); specifically, the first convolutional structure layer is connected to the first feature extraction module, the output end of the second feature extraction module is respectively connected to the second convolutional structure layer and the third feature extraction module, the output end of the third feature extraction module is respectively connected to the third convolutional structure layer and the fourth feature extraction module, the output end of the fourth feature extraction module is respectively connected to the target detection network and the fourth convolutional structure layer, the fourth convolutional structure layer is connected to the fifth convolutional structure layer through the maximum pooling pyramid, the output end of the fifth convolutional structure layer, the output end of the second convolutional structure layer and the output end of the third convolutional structure layer are all connected to the target detection network, and the output end of the target detection network is connected to the damage recognition network.

[0169] Optionally, the first convolutional structure layer may include but is not limited to three first convolutional layers connected in sequence (the convolution kernel sizes of the three first convolutional layers are 3×3, and the steps are 2, 1 and 1 respectively); the structures of the second convolutional structure layer, the third convolutional structure layer, the third convolutional structure layer and the fourth convolutional structure layer are the same, and all include a second convolutional layer, a batch normalization layer and a nonlinear activation layer; at the same time, the target detection network may include but is not limited to a PAENT network (Path Aggregation Network, a network for target detection), and the damage identification network may include but is not limited to a YOLOv4 network.

[0170] Furthermore, the following discloses one specific structure of the feature extraction module:

[0171] See also Figure 3 As shown, the example feature extraction module may include but is not limited to 5 second convolutional layers (numbered 1-5), a feature fusion layer, a third convolutional layer, a global average pooling layer, and a fully connected layer; wherein the 5 second convolutional layers are connected in sequence, and the input of the feature extraction module and the output of each second convolutional layer are connected to the feature fusion layer, the output of the feature fusion layer is connected to the third convolutional layer, the third convolutional layer is connected to the global average pooling layer, and the global average pooling layer is connected to the fully connected layer; further, the output of the fully connected layer is linearly multiplied with the output of the third convolutional layer, and then added to the output of the feature extraction module to obtain the output of the feature extraction module after the addition.

[0172] Therefore, the working process of the entire model is:

[0173] When any sample denoised boresight image is input into the aircraft damage recognition model, it is first convolutionally processed by the first convolutional structure layer, and then successively processed by the first and second feature extraction modules for feature extraction to obtain the first feature information, and then the first feature information is input into the PAENT network after passing through the second convolutional structure layer. At the same time, the first feature information is re-extracted by the third feature extraction module to obtain the second feature information, and the second feature information is input into the PAENT network after passing through the third convolutional structure layer; similarly, the third feature information is re-extracted by the fourth feature extraction module, and then successively processed by the fourth convolutional structure layer, the maximum pooling pyramid and the fifth convolutional structure layer for feature extraction, and then input into the PAENT network; at the same time, the fourth feature extraction module will also be directly input into the PAENT network; at this time, the feature extraction stage can be ended and the target detection stage can be entered.

[0174] In the target detection stage, the PAENT network extracts the input feature information again and performs feature fusion, and then outputs three target detection feature maps of different scales to the YOLOv4 network. The YOLOv4 network can derive parameters such as damage type (i.e., category value), damage prediction box coordinates, and confidence based on the three target detection feature maps of different scales. In this way, the damage recognition result of the blade of the aircraft corresponding to any of the aforementioned sample denoised boresight images can be obtained from the final output of the model. Of course, as mentioned above, the PAENT network and the YOLOv4 network are commonly used neural networks, and their working processes will not be repeated one by one.

[0175] Of course, the aforementioned aircraft blade damage identification model may also use an existing RBF network, which is not specifically limited here.

[0176] Therefore, through the aforementioned step S6 and its sub-steps, the optimal parameter optimization and model training of the model can be completed, thereby obtaining an aircraft blade damage identification model; finally, the aircraft blade damage identification model can be used to identify damage to the engine blades of the target aircraft, and the process is shown in the following step S7.

[0177] S7. Input the denoised boresight image into the aircraft blade damage identification model for damage identification processing to obtain damage identification results of the engine blades of the target aircraft. In this embodiment, the working process of the model on the denoised boresight image is the same as the processing process of any of the aforementioned sample denoised boresight images, and its principle is not repeated here. After the damage identification results of the engine blades of the target aircraft are obtained based on the aforementioned aircraft blade damage identification model, the health reliability analysis of the target aircraft can be performed, and the process is shown in the following step S8.

[0178] S8. Based on the damage identification result of the engine blades of the target aircraft, the health reliability analysis result of the target aircraft is obtained; in this embodiment, if the damage identification result is that damage exists (i.e., the damage category is detected), then the health reliability analysis result is that there is a safety hazard, otherwise, the status is normal; among them, if the health reliability analysis result is that there is a safety hazard, a prompt can be sent to the aviation ground service end to prompt the ground maintenance personnel to perform timely maintenance, thereby ensuring the safe operation of the aircraft.

[0179] Thus, according to the aircraft health reliability analysis method described in detail in the aforementioned steps S1 to S8, the present invention decomposes the original image, performs noise identification on the decomposed signal, and then denoises the noise signal, and then reconstructs the image using the denoised signal and the non-noise signal, thereby obtaining the denoised original image; then, the denoised original image is input into the aircraft blade damage recognition model, and the damage recognition result can be obtained. Finally, based on the damage recognition result, the aircraft health reliability analysis result can be obtained; in this way, through the aforementioned design, the noise of different frequency bands and frequencies can be removed to the greatest extent, thereby reducing the influence of mixed noise on damage recognition, thereby improving the accuracy of model recognition; at the same time, when performing damage recognition, the optimal model parameters are determined by the improved genetic algorithm, and multiple sets of model parameters can be evaluated in one iteration process, thereby reducing the model training time and improving the recognition efficiency; based on this, the present invention can reduce the model training time while ensuring the recognition accuracy, and is very suitable for large-scale application and promotion in the field of engine blade damage detection.

[0180] like Figure 4 As shown, the second aspect of this embodiment provides a hardware system for implementing the aircraft health reliability analysis method described in the first aspect of the embodiment, including:

[0181] The acquisition unit is used to acquire the borescope image of the engine blade of the target aircraft.

[0182] The wavelet decomposition unit is used to perform wavelet decomposition processing on the borescope image to obtain wavelet coefficients at different decomposition scales and a wavelet decomposition image signal corresponding to each wavelet coefficient.

[0183] The signal recognition unit is used to perform signal recognition processing on the wavelet decomposition image signal corresponding to each wavelet coefficient, so as to recognize the noise image signal from a plurality of wavelet decomposition image signals.

[0184] The denoising unit is used to perform denoising processing on the noisy image signal based on the wavelet coefficient of the noisy image signal to obtain a denoised image signal.

[0185] The denoising unit is further used to perform wavelet reconstruction processing using the target image signal and the denoised image signal to obtain a denoised boresight image after the wavelet reconstruction processing, wherein the target image signal is all wavelet decomposed image signals among the plurality of wavelet decomposed image signals except the noise image signal.

[0186] The blade damage identification unit is used to obtain an aircraft blade damage identification model, wherein the aircraft blade damage identification model is obtained by determining the optimal parameters of the model through an improved genetic algorithm, and based on the optimal parameters of the model, taking sample denoised boresight images of engine blades of several sample aircraft as input and the damage identification results of the engine blades of each sample aircraft as output.

[0187] The blade damage identification unit is also used to input the denoised boresight image into the aircraft blade damage identification model for damage identification processing to obtain a damage identification result of the engine blade of the target aircraft.

[0188] The reliability analysis unit is used to obtain a health reliability analysis result of the target aircraft based on the damage identification result of the engine blade of the target aircraft.

[0189] The working process, working details and technical effects of the device provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0190] like Figure 5 As shown, the third aspect of this embodiment provides a health reliability analysis device for an aircraft. Taking the device as an electronic device as an example, it includes: a memory, a processor and a transceiver that are communicatively connected in sequence, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer program to execute the health reliability analysis method for the aircraft as described in the first aspect of the embodiment.

[0191] For example, the memory may include, but is not limited to, random access memory (RAM), read only memory (ROM), flash memory, first input first output (FIFO) and / or first in last out (FILO), etc. Specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). At the same time, the processor may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state.

[0192] In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. For example, the processor may not be limited to a microprocessor of the STM32F105 series, a reduced instruction set computer (RISC) microprocessor, an X86 or other architecture processor, or a processor with an integrated embedded neural network processing unit (NPU); the transceiver may be, but is not limited to, a wireless fidelity (WIFI) wireless transceiver, a Bluetooth wireless transceiver, a general packet radio service technology (General Packet Radio Service, GPRS) wireless transceiver, a ZigBee protocol (a low-power LAN protocol based on the IEEE802.15.4 standard, ZigBee) wireless transceiver, a 3G transceiver, a 4G transceiver and / or a 5G transceiver, etc. In addition, the device may also include, but is not limited to, a power module, a display screen and other necessary components.

[0193] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0194] The fourth aspect of this embodiment provides a storage medium storing instructions for the aircraft health reliability analysis method described in the first aspect of the embodiment, that is, the storage medium stores instructions, and when the instructions are run on a computer, the aircraft health reliability analysis method described in the first aspect of the embodiment is executed.

[0195] The storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, a CD, a hard disk, a flash memory, a USB flash drive and / or a memory stick, and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0196] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment and will not be described in detail here.

[0197] The fifth aspect of this embodiment provides a computer program product comprising instructions, which, when executed on a computer, causes the computer to execute the aircraft health reliability analysis method as described in the first aspect of the embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0198] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for analyzing the health reliability of an aircraft, characterized in that: include: Acquire borescope images of engine blades of target aircraft; Performing wavelet decomposition processing on the borescope image to obtain wavelet coefficients at different decomposition scales and a wavelet decomposition image signal corresponding to each wavelet coefficient; Performing signal recognition processing on the wavelet decomposition image signal corresponding to each wavelet coefficient to recognize a noise image signal from a plurality of wavelet decomposition image signals; Based on the wavelet coefficients of the noise image signal, denoising the noise image signal to obtain a denoised image signal; Using the target image signal and the denoised image signal, wavelet reconstruction processing is performed to obtain a denoised boresight image after the wavelet reconstruction processing, wherein the target image signal is all wavelet decomposed image signals among the plurality of wavelet decomposed image signals except the noise image signal; Acquire an aircraft blade damage recognition model, wherein the aircraft blade damage recognition model is obtained by determining optimal model parameters through an improved genetic algorithm, and based on the optimal model parameters, taking sample denoised boresight images of engine blades of several sample aircraft as inputs and outputting damage recognition results of engine blades of each sample aircraft for training; Inputting the denoised boresight image into the aircraft blade damage identification model for damage identification processing to obtain a damage identification result of the engine blade of the target aircraft; Based on the damage identification results of the target aircraft's engine blades, the health reliability analysis results of the target aircraft are obtained; The optimal parameters of the model are determined by the improved genetic algorithm, including: Obtain multiple groups of model parameters, and use each group of model parameters as a chromosome to form an initial chromosome population; Based on the model parameters corresponding to each chromosome in the initial chromosome population, a number of initial aircraft blade damage recognition models are constructed, and each initial aircraft blade damage recognition model is trained using a training set to obtain a number of trained initial aircraft blade damage recognition models, wherein the training set includes sample denoised boresight images of engine blades of a number of sample aircraft; Inputting the same test data into each trained initial aircraft blade damage identification model to perform blade damage identification processing, so as to obtain model outputs of each trained initial aircraft blade damage identification model respectively; Calculating the fitness of each chromosome according to the model output of each trained initial aircraft blade damage recognition model and the label value of the test data; Filter out the global optimal fitness from various fitnesses, and determine whether the global optimal fitness meets the preset conditions; If not, then according to the fitness of each chromosome, a number of chromosomes are selected from the initial chromosome population to form a parent chromosome population; Selecting a plurality of parent chromosomes from the parent chromosome population, and using an improved crossover algorithm to perform individual crossover operations on the selected plurality of parent chromosomes, so as to obtain a plurality of crossover chromosomes after the individual crossover operations; Selecting a number of target chromosomes from the target chromosome population, and using the improved mutation algorithm to perform mutation operations on each of the selected target chromosomes, so as to obtain a number of mutant chromosomes after the mutation operation, wherein the target chromosome population includes a plurality of crossover chromosomes and unselected parent chromosomes in the parent chromosome population; Using the mutant chromosomes and the unselected target chromosomes in the target chromosome population, a progeny chromosome population is formed; The initial chromosome population is updated to the offspring chromosome population, and a number of initial aircraft blade damage recognition models are reconstructed based on the model parameters corresponding to each chromosome in the initial chromosome population until the global optimal fitness satisfies the preset conditions, and the model parameters of the chromosome corresponding to the global optimal fitness that satisfies the preset conditions are used as the model optimal parameters; According to the fitness of each chromosome, a number of chromosomes are selected from the chromosome population to form a parent chromosome population, including: Sort each chromosome in descending order of fitness to obtain a chromosome sorting sequence; Classifying the chromosomes in the chromosome sorting sequence into three categories from front to back to obtain a first category of chromosomes, a second category of chromosomes and a third category of chromosomes, respectively, wherein the fitness of the last chromosome in the first category of chromosomes is greater than the fitness of the first chromosome in the second category of chromosomes, and the fitness of the last chromosome in the second category of chromosomes is greater than the fitness of the first chromosome in the third category of chromosomes; Obtaining individual selection probabilities, and selecting multiple chromosomes from the third type of chromosomes based on the individual selection probabilities, and using the first type of chromosomes, the second type of chromosomes, and the multiple chromosomes selected from the third type of chromosomes to form the parent chromosome population; The improved crossover algorithm is used to perform individual crossover operations on the selected multiple parent chromosomes, so that after the individual crossover operations, multiple crossover chromosomes are obtained, including: For any two parent chromosomes among the selected multiple parent chromosomes, obtaining genes at the crossover position of the any two parent chromosomes; Generate a first random number between 0 and 1 for any two parent chromosomes; According to the first random number and the two genes at the crossover position, and in accordance with the following formula (5), an individual crossover operation is performed on the arbitrary two parent chromosomes, so as to obtain two crossover chromosomes after the individual crossover operation; (5) In the above formula (5), represents a gene at the crossover position of one of the two parent chromosomes, represents the gene after the crossover of one of the parent chromosomes at the crossover position, represents the gene at the crossover position of the other parent chromosome of any two parent chromosomes, represents the gene after the crossover of the other parent chromosome at the crossover position, and represents the first random number; The improved mutation algorithm is used to perform mutation operations on each selected target chromosome, so that after the mutation operation, several mutant chromosomes are obtained, including: For any target chromosome among the selected target chromosomes, generate a second random number corresponding to the target chromosome, wherein the value interval of the second random number is [0, 1]; Determine the variation position of any target chromosome, and obtain the gene of any target chromosome at the variation position and the maximum value and minimum value of the gene at the variation position; According to the maximum value, the minimum value, the second random number and the gene of any target chromosome at the mutation position, the mutated gene of any target chromosome at the mutation position is determined by using the following formula (6); (6) In the above formula (6), represents the gene of any target chromosome after mutation at the mutation position, represents the gene at the variant position of any target chromosome, represents the maximum value, represents the minimum value, represents the second random number, Indicates the current iteration number, Indicates the maximum number of iterations; The gene at the mutation position of any target chromosome is updated to the mutated gene, so as to obtain the mutated chromosome corresponding to any target chromosome after the update.

2. The method according to claim 1, characterized in that The signal recognition processing is performed on the wavelet decomposition image signal corresponding to each wavelet coefficient to recognize the noise image signal from the plurality of wavelet decomposition image signals, including: For any wavelet decomposition image signal, obtaining all pixel points in the wavelet decomposition image signal; Based on all the pixels in any one of the wavelet decomposition image signals, and using a binary classification algorithm, determining a binary classification value of any one of the wavelet decomposition image signals; According to the binary classification value of any wavelet decomposition image signal, the signal classification result of any wavelet decomposition image signal is determined, wherein if the binary classification value of any wavelet decomposition image signal is less than 0, the signal classification result of any wavelet decomposition image signal is obtained as a noise image signal, and if the binary classification value of any wavelet decomposition image signal is greater than or equal to 0, the signal classification result of any wavelet decomposition image signal is obtained as a non-noise image signal.

3. The method according to claim 2, characterized in that Based on all the pixels in any one of the wavelet decomposition image signals, and using a binary classification algorithm, determining a binary classification value of any one of the wavelet decomposition image signals, including: According to the coordinates of each pixel point in any wavelet decomposition image signal, and in accordance with the following formula (1), a binary classification bias value is determined; (1) In the above formula (1), represents the binary classification bias value, represents the Lagrange multiplier, are the horizontal coordinate and vertical coordinate of the j-th pixel point in any wavelet decomposition image signal, respectively. represents the total number of pixels in any wavelet decomposition image signal, All represent binary classification parameters; Using the binary classification offset value and the coordinates of each pixel point in any wavelet decomposition image signal, and according to the following formula (2), determine the binary classification value of any wavelet decomposition image signal; (2) In the above formula (2), represents the binary classification value of any wavelet decomposition image signal, represents the Gaussian radial basis kernel function, represents a sign function, where when the value in the square brackets is less than 0, The returned value is -1 when the value in the square brackets is greater than 0. The returned value is 1.

4. The method according to claim 1, characterized in that: Based on the wavelet coefficients of the noise image signal, denoising the noise image signal to obtain a denoised image signal, including: Acquire an initial wavelet threshold, and determine a first wavelet threshold and a second wavelet threshold based on the initial wavelet threshold, wherein the first wavelet threshold and the second wavelet threshold are both greater than 0, and the first wavelet threshold is less than the initial wavelet threshold, and the initial wavelet threshold is less than the second wavelet threshold; Determining a wavelet adjustment parameter based on the wavelet coefficient of the noise image signal, the first wavelet threshold, and the second wavelet threshold; The wavelet coefficients of the noisy image signal are adjusted according to the wavelet adjustment parameter, the first wavelet threshold, and the second wavelet threshold, so as to obtain the denoised image signal after the wavelet coefficients are adjusted.

5. The method according to claim 4, characterized in that Determining a wavelet adjustment parameter based on the wavelet coefficient of the noise image signal, the first wavelet threshold, and the second wavelet threshold includes: According to the following formula (3), the wavelet adjustment parameter is calculated; (3) In the above formula (3), represents the wavelet adjustment parameter, represent the first wavelet threshold and the second wavelet threshold respectively, A wavelet coefficient representing the noise image signal; Correspondingly, adjusting the wavelet coefficients of the noise image signal according to the wavelet adjustment parameter, the first wavelet threshold, and the second wavelet threshold includes: According to the following formula (4), the wavelet coefficients of the noise image signal are adjusted; (4) In the above formula (4), represents the adjusted wavelet coefficients, Represents a symbolic function.

6. An aircraft health reliability analysis system, characterized in that: include: An acquisition unit, used for acquiring a borescope image of an engine blade of a target aircraft; A wavelet decomposition unit, used for performing wavelet decomposition processing on the borescope image to obtain wavelet coefficients at different decomposition scales and a wavelet decomposition image signal corresponding to each wavelet coefficient; A signal recognition unit, used for performing signal recognition processing on the wavelet decomposition image signal corresponding to each wavelet coefficient, so as to recognize a noise image signal from a plurality of wavelet decomposition image signals; A denoising unit, configured to perform denoising processing on the noisy image signal based on the wavelet coefficients of the noisy image signal to obtain a denoised image signal; The denoising unit is further used to perform wavelet reconstruction processing using the target image signal and the denoised image signal to obtain a denoised boresight image after the wavelet reconstruction processing, wherein the target image signal is all wavelet decomposed image signals among the plurality of wavelet decomposed image signals except the noise image signal; A blade damage identification unit is used to obtain an aircraft blade damage identification model, wherein the aircraft blade damage identification model is obtained by training by determining optimal model parameters through an improved genetic algorithm, and taking sample denoised boresight images of engine blades of several sample aircraft as inputs and outputting damage identification results of engine blades of each sample aircraft as outputs based on the optimal model parameters; The blade damage identification unit is further used to input the denoised boresight image into the aircraft blade damage identification model for damage identification processing to obtain a damage identification result of the engine blade of the target aircraft; A reliability analysis unit, used for obtaining a health reliability analysis result of the target aircraft according to the damage identification result of the engine blade of the target aircraft; The optimal parameters of the model are determined by the improved genetic algorithm, including: Obtain multiple groups of model parameters, and use each group of model parameters as a chromosome to form an initial chromosome population; Based on the model parameters corresponding to each chromosome in the initial chromosome population, a number of initial aircraft blade damage recognition models are constructed, and each initial aircraft blade damage recognition model is trained using a training set to obtain a number of trained initial aircraft blade damage recognition models, wherein the training set includes sample denoised boresight images of engine blades of a number of sample aircraft; Inputting the same test data into each trained initial aircraft blade damage identification model to perform blade damage identification processing, so as to obtain model outputs of each trained initial aircraft blade damage identification model respectively; Calculating the fitness of each chromosome according to the model output of each trained initial aircraft blade damage recognition model and the label value of the test data; Filter out the global optimal fitness from various fitnesses, and determine whether the global optimal fitness meets the preset conditions; If not, then according to the fitness of each chromosome, a number of chromosomes are selected from the initial chromosome population to form a parent chromosome population; Selecting a plurality of parent chromosomes from the parent chromosome population, and using an improved crossover algorithm to perform individual crossover operations on the selected plurality of parent chromosomes, so as to obtain a plurality of crossover chromosomes after the individual crossover operations; Selecting a number of target chromosomes from the target chromosome population, and using the improved mutation algorithm to perform mutation operations on each of the selected target chromosomes, so as to obtain a number of mutant chromosomes after the mutation operation, wherein the target chromosome population includes a plurality of crossover chromosomes and unselected parent chromosomes in the parent chromosome population; Using the mutant chromosomes and the unselected target chromosomes in the target chromosome population, a progeny chromosome population is formed; The initial chromosome population is updated to the offspring chromosome population, and a number of initial aircraft blade damage recognition models are reconstructed based on the model parameters corresponding to each chromosome in the initial chromosome population until the global optimal fitness satisfies the preset conditions, and the model parameters of the chromosome corresponding to the global optimal fitness that satisfies the preset conditions are used as the model optimal parameters; According to the fitness of each chromosome, a number of chromosomes are selected from the chromosome population to form a parent chromosome population, including: Sort each chromosome in descending order of fitness to obtain a chromosome sorting sequence; Classifying the chromosomes in the chromosome sorting sequence into three categories from front to back to obtain a first category of chromosomes, a second category of chromosomes and a third category of chromosomes, respectively, wherein the fitness of the last chromosome in the first category of chromosomes is greater than the fitness of the first chromosome in the second category of chromosomes, and the fitness of the last chromosome in the second category of chromosomes is greater than the fitness of the first chromosome in the third category of chromosomes; Obtaining individual selection probabilities, and selecting multiple chromosomes from the third type of chromosomes based on the individual selection probabilities, and using the first type of chromosomes, the second type of chromosomes, and the multiple chromosomes selected from the third type of chromosomes to form the parent chromosome population; The improved crossover algorithm is used to perform individual crossover operations on the selected multiple parent chromosomes, so that after the individual crossover operations, multiple crossover chromosomes are obtained, including: For any two parent chromosomes among the selected multiple parent chromosomes, obtaining genes at the crossover position of the any two parent chromosomes; Generate a first random number between 0 and 1 for any two parent chromosomes; According to the first random number and the two genes at the crossover position, and in accordance with the following formula (5), an individual crossover operation is performed on the arbitrary two parent chromosomes, so as to obtain two crossover chromosomes after the individual crossover operation; (5) In the above formula (5), represents a gene at the crossover position of one of the two parent chromosomes, represents the gene after the crossover of one of the parent chromosomes at the crossover position, represents the gene at the crossover position of the other parent chromosome of any two parent chromosomes, represents the gene after the crossover of the other parent chromosome at the crossover position, and represents the first random number; The improved mutation algorithm is used to perform mutation operations on each selected target chromosome, so that after the mutation operation, several mutant chromosomes are obtained, including: For any target chromosome among the selected target chromosomes, generate a second random number corresponding to the target chromosome, wherein the value interval of the second random number is [0, 1]; Determine the variation position of any target chromosome, and obtain the gene of any target chromosome at the variation position and the maximum value and minimum value of the gene at the variation position; According to the maximum value, the minimum value, the second random number and the gene of any target chromosome at the mutation position, the mutated gene of any target chromosome at the mutation position is determined by using the following formula (6); (6) In the above formula (6), represents the gene of any target chromosome after mutation at the mutation position, represents the gene at the variant position of any target chromosome, represents the maximum value, represents the minimum value, represents the second random number, Indicates the current iteration number, Indicates the maximum number of iterations; The gene at the mutation position of any target chromosome is updated to the mutated gene, so as to obtain the mutated chromosome corresponding to any target chromosome after the update.