Aircraft engine gear damage analysis method and system based on image recognition

Abstract

The present application discloses an aircraft engine gear damage analysis method and system based on image recognition, which relates to the field of image recognition technology. The method includes: obtaining gear surface images of the aircraft engine gear to be inspected at multiple angles; preprocessing the gear surface images; performing gear region segmentation, damage region location, and damage feature quantification processing on the preprocessed gear surface images in sequence to obtain gear damage features; performing damage analysis on each damaged region based on the gear damage features to obtain damage analysis results, which at least include damage type determination results, damage severity classification results, and remaining life estimation results. The present application can solve the technical problems of low efficiency and high missed detection rate of aircraft engine gear damage detection, difficulty in detecting tooth surface damage, inability to predict damage evolution and assess remaining life, and failure to meet aircraft engine gear maintenance requirements.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to an aircraft engine gear damage analysis method and system based on image recognition. Background Art

[0002] As a critical component for transmitting power, aircraft engine gears' operating conditions directly impact flight safety. In practice, gears are subjected to extreme operating conditions such as high speed, heavy load, and high temperature for extended periods, making them susceptible to various forms of damage, including cracks, pitting, spalling, wear, and deformation. Failure to promptly detect and assess these damages can lead to serious mechanical failures and even endanger flight safety. Therefore, extremely high standards are placed on damage detection and assessment technologies for aircraft engine gears.

[0003] Currently, aircraft engine gear damage detection primarily relies on manual visual inspection, penetrant testing, magnetic particle testing, or eddy current testing. Manual visual inspection relies on direct observation of the gear surface to determine damage; penetrant testing uses a penetrant to penetrate defects and then reveals them with a developer; magnetic particle testing uses the principle of magnetic flux leakage to detect surface and near-surface defects in ferromagnetic materials; and eddy current testing uses electromagnetic induction to detect defects.

[0004] However, traditional manual visual inspection is not only inefficient and unable to meet large-scale inspection needs, but also suffers from a high rate of missed detections, and can easily lead to inaccurate test results due to human factors. Furthermore, current inspection technologies primarily focus on detecting damage to gear end faces. Due to limitations in detection principles and methods, effective detection of damage such as contact fatigue and frictional wear on the tooth surfaces is difficult. Furthermore, existing technologies lack the ability to predict damage evolution and are unable to dynamically assess the remaining useful life of gears based on the current damage state, making it difficult to meet the reliability and safety requirements of aircraft engine maintenance. Summary of the Invention

[0005] In view of this, the present application provides an aircraft engine gear damage analysis method and system based on image recognition, which can solve the technical problems of low efficiency and high missed detection rate of aircraft engine gear damage detection, difficulty in detecting tooth surface damage, inability to predict damage evolution and evaluate remaining life, and failure to meet the maintenance needs of aircraft engine gears.

[0006] According to a first aspect of the present application, a method for analyzing damage of an aircraft engine gear based on image recognition is provided, comprising:

[0007] Acquire gear surface images of the aircraft engine gear to be inspected at multiple angles;

[0008] Preprocessing the gear surface image, wherein the preprocessing includes at least image noise reduction, image contrast enhancement, and viewpoint normalization;

[0009] Performing gear region segmentation and damage region location processing on the preprocessed gear surface image in sequence, and quantifying damage features of each located damage region to obtain gear damage features, wherein the gear damage features include at least geometric features, texture features, and spatial features;

[0010] Based on the gear damage characteristics, damage analysis is performed on each of the damaged areas to obtain damage analysis results, which at least include damage type determination results, damage severity classification results and remaining life estimation results.

[0011] According to a second aspect of the present application, there is provided an aircraft engine gear damage analysis system based on image recognition, comprising: a data acquisition module, an image analysis module, and a gear damage assessment module;

[0012] The data acquisition module is used to acquire gear surface images of the aircraft engine gear to be inspected at multiple angles;

[0013] The image analysis module is used to preprocess the gear surface image, perform gear region segmentation and damage region location processing on the preprocessed gear surface image in sequence, and quantify damage features of each located damage region to obtain gear damage features, wherein the preprocessing includes at least image noise reduction, image contrast enhancement, and viewpoint normalization, and the gear damage features include at least geometric features, texture features, and spatial features;

[0014] The gear damage assessment module is used to perform damage analysis on each of the damaged areas based on the gear damage characteristics to obtain damage analysis results, which at least include damage type determination results, damage severity classification results and remaining life estimation results.

[0015] According to a third aspect of the present application, a storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the above-mentioned aircraft engine gear damage analysis method based on image recognition is implemented.

[0016] According to the fourth aspect of the present application, an electronic device is provided, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein when the processor executes the program, the above-mentioned image recognition-based aircraft engine gear damage analysis method is implemented.

[0017] By leveraging the aforementioned technical solution, the present application provides an image recognition-based aircraft engine gear damage analysis method and system. After acquiring gear surface images of the aircraft engine gear to be inspected from multiple angles, the system preprocesses the gear surface images to improve image quality for accurate segmentation and location of damaged areas. Furthermore, the system quantifies the damage characteristics of each located damaged area to obtain gear damage signatures, which are then used for damage analysis. This method effectively detects damage types such as tooth surface contact fatigue and friction wear, avoiding the inefficiency and missed detection issues of manual visual inspection. Furthermore, by determining damage type, grading severity, and estimating remaining life, it is possible to predict damage evolution, meeting the reliability and safety requirements of aircraft engine gear maintenance.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a process flow of an aircraft engine gear damage analysis method based on image recognition provided by an embodiment of the present application is shown;

[0020] Figure 2 A schematic flow chart of an aircraft engine gear damage analysis method based on image recognition according to another embodiment of the present application is shown;

[0021] Figure 3 A schematic diagram of the system structure of an aircraft engine gear damage analysis system based on image recognition provided by an embodiment of the present application is shown;

[0022] In the picture;

[0023] 310-data acquisition module, 320-image analysis module, 330-gear damage assessment module, 340-system platform module, 3401-human-computer interaction submodule, 3402-data management submodule, 3403-parameter mapping submodule. DETAILED DESCRIPTION

[0024] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0025] Currently, aircraft engine gear damage detection primarily relies on manual visual inspection, penetrant testing, magnetic particle testing, or eddy current testing. Manual visual inspection relies on direct observation of the gear surface to determine damage; penetrant testing uses a penetrant to penetrate defects and then reveals them with a developer; magnetic particle testing uses the principle of magnetic flux leakage to detect surface and near-surface defects in ferromagnetic materials; and eddy current testing uses electromagnetic induction to detect defects.

[0026] However, traditional manual visual inspection is not only inefficient and unable to meet large-scale inspection needs, but also suffers from a high rate of missed detections, and can easily lead to inaccurate test results due to human factors. Furthermore, current inspection technologies primarily focus on detecting damage to gear end faces. Due to limitations in detection principles and methods, effective detection of damage such as contact fatigue and frictional wear on the tooth surfaces is difficult. Furthermore, existing technologies lack the ability to predict damage evolution and are unable to dynamically assess the remaining useful life of gears based on the current damage state, making it difficult to meet the reliability and safety requirements of aircraft engine maintenance.

[0027] In order to solve the above problems, an embodiment of the present invention provides an aircraft engine gear damage analysis method based on image recognition, such as Figure 1 As shown, the method includes:

[0028] Step 110: Acquire gear surface images of the aircraft engine gear to be inspected at multiple angles.

[0029] Among them, the aircraft engine gear to be inspected refers to the gear component whose damage condition needs to be inspected during the operation of the aircraft engine; the gear surface image at multiple angles refers to the image obtained by driving the gear to rotate with a precision rotation and translation fixture and using a high-resolution industrial camera to capture the gear surface from multiple different angles.

[0030] In specific application scenarios, the aircraft engine gear to be inspected can be fixedly mounted on the loading platform of a six-degree-of-freedom rotation and translation fixture. The fixture's robotic arm can drive the gear to perform submicron-level translation (X, Y, Z axes) and rotation (pitch, yaw, and roll) in three-dimensional space. Specifically, the gear's center of rotation is strictly coaxial with the fixture's main shaft rotation center, ensuring that when rotating at fixed angle intervals (such as every 5° step), each area of ​​the gear tooth surface can sequentially face the high-resolution industrial camera lens. At the same time, the fixture can translate in a direction perpendicular to the gear axis to adjust the imaging position of the gear in the image. In conjunction with the camera's focal length adjustment, different parts such as the tooth root, tooth surface, and tooth top are all within the optimal imaging depth of field. This positional relationship design ensures image clarity and position consistency in each area of ​​the gear surface during multi-angle acquisition, laying the foundation for subsequent viewpoint normalization processing.

[0031] In the disclosed embodiments, industrial inspection equipment can be used to capture comprehensive images of aircraft engine gears. Specifically, a high-resolution camera can be used in conjunction with the precision rotary fixture to rotate the gear at fixed angle intervals (e.g., 5° steps), simultaneously capturing multiple sets of surface images to form a multi-angle image set covering the tooth root, tooth flank, and tooth top. This process can also utilize a combined light source system (coaxial white light and side-polarized blue light). By switching light source modes (coaxial light highlights surface undulations, side light enhances texture details), damage features such as cracks and pitting can be clearly visualized in the images, providing multi-dimensional visual data for subsequent damage analysis.

[0032] Traditional methods can only inspect gear end faces or surfaces at a single angle. However, multi-angle acquisition, through 360° rotational imaging, can improve the coverage of tooth surface damage detection, especially addressing blind spots in curved areas such as tooth root fillets and pitch lines. Furthermore, automated acquisition replaces manual inspection, shortening single-gear inspection time and reducing missed detection rates. Pre-processing multi-angle images eliminates light and shadow distortion, providing a high-quality data foundation for deep learning algorithms to accurately locate damage and ensuring the accuracy of subsequent feature quantification and life assessment.

[0033] Step 120: Preprocess the gear surface image.

[0034] Preprocessing gear surface images refers to the process of optimizing image quality through a series of algorithms, including but not limited to image noise reduction, image contrast enhancement, and viewpoint normalization. In the following disclosed embodiments, the preprocessing steps, including image noise reduction, image contrast enhancement, and viewpoint normalization, are used as an example to illustrate the technical solutions in this application, but do not constitute a specific limitation. Among them, image denoising refers to the technology of eliminating noise in gear surface images through filtering algorithms. Specifically, Gaussian filtering can be used to perform weighted averaging of pixel values ​​using Gaussian functions, smooth the image surface, suppress random noise interference, make the gear surface texture features purer, and provide clear input for subsequent damage identification; image contrast enhancement refers to the technology of improving the brightness and darkness differences in different areas by adjusting the grayscale distribution of the image. Specifically, histogram equalization can be used to map the image grayscale values ​​to a wider range, so that the black outline of the crack and the gray background of the tooth surface are more distinct, especially enhancing the visual recognition of shallow damage (such as wear lines); viewpoint normalization processing refers to the technology of geometric correction of gear surface images based on camera calibration parameters and gear posture data. Through perspective transformation, gear surface images taken at different angles are unified to a standard perspective, solving the shape distortion caused by camera tilt or gear rotation, so that the size and shape of the same damage in different gear surface images are consistent, which is convenient for subsequent quantitative analysis.

[0035] For the embodiments of the present disclosure, Gaussian filtering can be used to reduce the noise of the gear surface image first, and the random noise generated by sensor noise or environmental interference during the shooting process can be eliminated through weighted averaging of the Gaussian function; then histogram equalization is used to redistribute the image grayscale values, stretch the grayscale distribution range, and enhance the contrast between the damaged area and the normal tooth surface; finally, based on the camera calibration parameters and gear posture information, the gear surface image is subjected to geometric distortion correction and perspective transformation, and the gear surface images collected at different angles are unified to a standard perspective, eliminating the shape distortion caused by differences in shooting angles, so that the gear surface features are truly restored in the gear surface image.

[0036] Step 130: perform gear region segmentation and damage region location processing on the pre-processed gear surface image, and quantify the damage features of each located damage region to obtain gear damage features. The gear damage features include at least geometric features, texture features, and spatial features.

[0037] Among them, gear area segmentation refers to the process of extracting gear contours through edge detection algorithms (such as Canny) and separating the gears in the image from the background to avoid background noise interfering with damage detection; damage area positioning refers to automatically marking the specific locations of damage such as cracks and pitting within the gear area; the damage area refers to various defective areas on the surface of aircraft engine gears caused by high speed, heavy load, high temperature and other working conditions, mainly including cracks, pitting, spalling, wear and plastic deformation; damage feature quantification refers to the process of extracting geometric, texture, spatial and other multi-dimensional features of the located gear damage area through algorithms and converting them into numerical parameters; gear damage features are multi-dimensional parameters for quantitative extraction of damage areas. A set of visual features is converted into computable values ​​through algorithms, including geometric features, texture features, and spatial features. Geometric features are used to describe parameters of the physical morphology of damage, which may include crack length, maximum crack width, average crack width, crack area, crack contour perimeter, crack orientation angle, single pitting / spalling area, total cumulative pitting / spalling area, number of pitting / spalling points per unit area, depth-to-diameter ratio per unit area of ​​pitting / spalling, area of ​​the spalling zone, shape factor, etc., which are used to measure the "size and shape" of damage. Texture features are used to reflect parameters of the texture characteristics of the damaged surface, such as grayscale statistical characteristics (variance, entropy, etc.) within the damaged area, local binary pattern (LBP) descriptors, and surface roughness-related metrics, which are used to distinguish different types of damage (such as linear texture of cracks and rough texture of wear). Spatial features are used to represent the location information of damage on the gear surface (spatial coordinates and their mutual relationship), such as whether it is located at the tooth root, pitch line, or tooth top, as well as the spacing and distribution pattern between damages, which are used to analyze the distribution law of damage and potential risks.

[0038] For the embodiments of the present disclosure, the gear surface image that has been pre-processed by denoising, contrast enhancement, and perspective unification is firstly processed by using image processing techniques such as edge detection to accurately extract the gear part from the background; then, within the separated gear image area, with the help of an improved deep learning algorithm, the specific locations of various types of damage such as cracks and pitting are automatically identified and marked; finally, for each marked damage area, its geometric shape (such as crack length, pitting area), surface texture (such as grain roughness, grayscale change) and spatial position (specific coordinates at the gear tooth root and tooth surface, distance between damages) and other information are further extracted and converted into specific numerical parameters to form gear damage characteristic data that comprehensively describes the gear damage condition.

[0039] This series of operations automatically separates the gear area from the background, facilitating the precise location of damage such as cracks and pitting. It also calculates the damage size, surface roughness, and distribution on the gear. This allows for quick and accurate identification of gear damage and its severity without the need for human visual inspection. This provides detailed data for subsequent determination of damage type and lifespan, resolving the inefficiency and inaccuracy of traditional methods.

[0040] Step 140: Perform damage analysis on each damaged area based on the gear damage characteristics to obtain damage analysis results. The damage analysis results at least include damage type determination results, damage severity classification results, and remaining life estimation results.

[0041] Among them, damage analysis is a process of comprehensively evaluating damage based on gear damage characteristic data, using a combination of rules, models and algorithms, aiming to clarify the nature, severity and development trend of the damage; the damage type determination result is the specific damage category obtained by comparing the gear damage characteristics with preset rules or classification models, such as fatigue cracks, contact fatigue pitting, friction wear, etc., providing a basis for subsequent targeted treatment; the damage severity classification result is a classification of the degree of damage hazard based on the damage type and characteristic data, with reference to established standards, usually divided into mild, moderate, severe, etc., to facilitate maintenance personnel to quickly judge the urgent need for damage treatment; the remaining life estimation result is the time or number of cycles that the gear can still operate safely in the current damage state predicted by the damage evolution model based on factors such as damage type, severity, gear material properties and actual operating conditions, providing a decision-making reference for preventive maintenance.

[0042] In the disclosed embodiment, after obtaining characteristic data such as geometry, texture, and space of gear damage, an in-depth analysis is performed on each damaged area. First, based on these characteristic data, they are compared with pre-set rules or models to determine whether the damage type is cracks, pitting, wear, etc.; then, based on the damage type and characteristic data, and in accordance with established standards, the severity of the damage is classified into different levels such as mild, moderate, and severe; finally, combining information such as damage type, severity, gear material, and operating conditions, and using corresponding prediction models, the safe use life of the gear in the damaged area is estimated, ultimately obtaining a complete analysis result including damage type, severity classification, and remaining life estimation.

[0043] This systematic damage analysis can bid farewell to the limitations of traditional manual inspections that rely solely on subjective judgment based on experience. Based on precise quantitative data, the specific type of gear damage can be quickly and accurately determined, avoiding misjudgments due to human factors. At the same time, a standardized classification of damage severity allows maintenance personnel to intuitively understand the degree of damage danger. The remaining life estimation provides a scientific planning basis for the repair and replacement of aircraft engine gears, allowing maintenance plans to be arranged in advance to avoid safety hazards caused by sudden failures, greatly improving the reliability and safety of aircraft engine operations and reducing maintenance costs and risks.

[0044] In summary, the image recognition-based aircraft engine gear damage analysis method provided by the present invention, after acquiring gear surface images of the aircraft engine gear to be inspected at multiple angles, can improve image quality by preprocessing the gear surface images to accurately segment and locate damaged areas. Furthermore, the damage characteristics of each located damaged area can be quantified to obtain gear damage characteristics, which can be used for damage analysis. This method can effectively detect damage types such as tooth surface contact fatigue and friction wear, avoiding the inefficiency and missed detection problems of manual visual inspection. Furthermore, by determining damage type, grading severity, and estimating remaining life, it can predict damage evolution, meeting the reliability and safety requirements of aircraft engine gear maintenance.

[0045] Furthermore, as a refinement and expansion of the specific implementation of the above embodiment, in order to fully illustrate the implementation of this embodiment, this embodiment also provides another aircraft engine gear damage analysis method based on image recognition, such as Figure 2 As shown, the method includes:

[0046] Step 210: Acquire gear surface images of the aircraft engine gear to be inspected at multiple angles.

[0047] For the embodiment of the present disclosure, the specific implementation process can be found in the relevant description of step 110 of the embodiment, which will not be repeated here.

[0048] Step 220: pre-process the gear surface image, where the pre-processing includes at least image noise reduction, image contrast enhancement, and viewpoint normalization.

[0049] Accordingly, step 220 of the embodiment may specifically include the following steps:

[0050] Step 220 - 1 : Use Gaussian filtering to perform noise reduction on the gear surface image.

[0051] Among them, Gaussian filtering is a linear smoothing filtering algorithm that calculates pixel weights through Gaussian functions and performs convolution operations on images to eliminate Gaussian noise.

[0052] In the disclosed embodiments, a Gaussian function can be used to construct a filtering template, performing a weighted average of each pixel and its neighborhood in the gear surface image. The weights follow a Gaussian distribution with the distance from the pixel to the center (for example, when σ = 1.2, the weights decrease with increasing distance from the center). This process effectively smooths random noise generated by camera sensor noise, ambient light interference, and other factors, resulting in a purer gear surface texture and providing a clear image foundation for subsequent damage identification.

[0053] Step 220 - 2 : Use histogram equalization to perform contrast enhancement on the gear surface image after noise reduction.

[0054] Among them, histogram equalization refers to a technology that enhances image contrast by redistributing the grayscale values ​​of image pixels and expanding the histogram distribution to the entire grayscale range. It can effectively improve the recognizability of low-contrast damage.

[0055] In the disclosed embodiments, histogram equalization can be used to map the grayscale distribution of the denoised gear surface image to a wider range, stretching the originally concentrated grayscale range. For example, the denoised image is processed in blocks, with the grayscale histogram independently adjusted within each block. This increases the grayscale difference between the damaged area (such as the black outline of a crack) and the normal tooth surface (gray background), highlighting the visual characteristics of shallow damage as small as 0.1 mm. Contrast enhancement can increase the grayscale difference between the damaged area and the background, improving the visibility of shallow pitting and enhancing the recognition rate of wear lines.

[0056] Step 220-3: Based on the camera calibration parameters and the gear posture information, perform geometric distortion correction and perspective transformation on the gear surface image after contrast enhancement to achieve viewpoint normalization so that the gear surface images collected at different angles have a unified viewing angle reference.

[0057] Among them, camera calibration parameters are a set of key data used to describe the geometric relationship of camera imaging. They are obtained through calibration experiments and are mainly used to correct lens distortion and establish a mapping relationship between image pixels and actual physical space. They may include focal length, distortion coefficient, and camera intrinsic parameter matrix, etc.; gear pose information refers to the spatial position and attitude parameters of the gear relative to the camera during the detection process, which is recorded in real time by a precision rotation and translation fixture and used for spatial alignment of multi-view images. They may include rotation angle, translation coordinates, and attitude parameters (such as pitch angle, yaw angle, roll angle), etc.; geometric distortion correction refers to a technology that corrects the barrel or pillow-shaped deformation caused by lens distortion in the image based on camera calibration parameters, so that straight line features remain straight in the image; perspective transformation is the process of mapping an image from one viewing plane to another, which can eliminate the perspective distortion caused by the tilt of the shooting angle, so that images from different perspectives have a unified spatial reference.

[0058] In the disclosed embodiment, image distortion can be corrected through camera calibration and pose data. First, camera calibration parameters (e.g., focal length f = 50mm, distortion coefficient k1 = -0.15) are used to eliminate image distortion caused by lens distortion. Then, based on the gear's pose information in the fixture (e.g., rotation angle and translation coordinates), perspective transformation is used to map images captured at different angles to a unified viewpoint. For example, 72 sets of images captured at 5° intervals are uniformly transformed to a standard viewpoint perpendicular to the tooth surface, ensuring a faithful reproduction of the gear profile in the image. Viewpoint normalization reduces measurement errors in the gear edge profile and addresses tooth surface curvature distortion caused by the shooting angle.

[0059] Step 230: The pre-processed gear surface image is sequentially subjected to gear region segmentation and damage region location processing, and damage features of each located damage region are quantified to obtain gear damage features, which include at least geometric features, texture features, and spatial features.

[0060] In the embodiment of the present disclosure, when the gear region segmentation and damage region location processing are sequentially performed on the pre-processed gear surface image, step 230 may include the following steps:

[0061] Step 230 - 1 : Use edge detection image processing technology to perform gear region segmentation on the pre-processed gear surface image to obtain a gear region separated from the background region.

[0062] In the disclosed embodiment, the Canny algorithm can be first used to calculate the horizontal and vertical gradients of each pixel in the preprocessed gear surface image using the Sobel operator, synthesizing the gradient amplitude and direction angle. Next, neighboring pixels are compared along the gradient direction, retaining only the pixels with the largest gradient amplitude to refine the edge to a single pixel width. High and low thresholds are then set to filter edges, with gradients above the high threshold marked as strong edges, those between the high and low thresholds as weak edges, and those below the low threshold considered background. Weak edges are retained only when connected to strong edges, ensuring edge continuity and suppressing isolated noise. Ultimately, a complete gear contour is obtained, achieving precise separation of the gear region from the background. This operation effectively removes background noise interference, improves segmentation accuracy, and reduces gear contour extraction errors, allowing the gear region to be presented independently in the image, providing a clean region of interest (ROI) for subsequent damage detection.

[0063] Step 230 - 2: Apply the improved deep learning YOLO target detection algorithm to locate the damaged area within the segmented gear area to obtain each damaged area. The deformable convolution module is introduced into the improved deep learning YOLO target detection algorithm.

[0064] After completing gear region segmentation, an improved YOLO target detection algorithm (GearYOLO-Net) can be used to locate gear surface damage: the algorithm introduces a deformable convolution module (DCM), which uses a trainable weight matrix to predict spatial offsets from feature maps extracted from the backbone network and dynamically adjust the convolution sampling point set (for example, adaptively generating irregular sampling points for bending cracks). This module can overcome the limitations of traditional convolution fixed sampling points, dynamically capture the boundary features of irregular damage such as cracks and pitting, and combine the multi-scale feature pyramid (FPN) to fuse features from different levels, ultimately accurately marking the damage location within the gear region (for example, marking a tooth root crack with a rectangular box). The improved deep learning YOLO target detection algorithm can improve damage localization accuracy and increase the efficiency of locating damaged areas, meeting the rapid detection needs of aviation maintenance.

[0065] The improved deep learning YOLO target detection algorithm (GearYOLO-Net) adopts the network structure of "backbone network + deformable convolution module + multi-scale feature pyramid". When applying the improved deep learning YOLO target detection algorithm to locate the damaged area: the segmented gear area image can be first input into the SkeletonNet backbone network, and the feature map containing texture, contour and other information can be extracted through multi-layer convolution operations; the deformable convolution module (DCM) is introduced in the key layer of the backbone network, and the trainable weight matrix Woffset is used to learn and predict the spatial offset from the feature map. Δpn is used to adjust the sampling point position of the convolution kernel so that it adaptively fits the edges of irregular damage such as cracks and pitting. The feature map after DCM processing is input into the multi-scale feature pyramid (FPN), which fuses features at different levels (bottom-level detail features and high-level semantic features) to enhance the detection ability of damage of different sizes, such as microcracks (such as 0.1mm level) and large-area spalling. Finally, the fused features are classified and regressed through multi-task detection heads (such as crack detection head and pitting detection head), and the category (crack / pitting, etc.), pixel-level coordinates and bounding box of each damaged area are output to achieve accurate labeling of the damage location.

[0066] Among them, the deformable convolution module is used to: receive the feature map from the backbone network as input, predict the spatial offset from the feature map through the trainable convolution kernel weight matrix to generate an adaptively adjusted sampling point set; use the sampling point set to perform convolution operations on the feature map to dynamically capture the boundary features of irregular damage; and output an adaptive damage feature map based on the boundary features for subsequent multi-scale feature fusion and damage area positioning.

[0067] After gear region segmentation and damage region location processing, at least one damaged region can be obtained. The damage characteristics of each damaged region can then be quantified to obtain the gear damage signature. When quantifying the damage characteristics of each damaged region, geometric features such as crack length and pitting area can be obtained through contour extraction and geometric calculations. Texture features such as variance and entropy are then extracted using grayscale statistics and the LBP operator. Spatial features such as the location of the damage on the tooth surface and the spacing between adjacent teeth are determined based on the gear's three-dimensional model and pose information. Ultimately, these geometric, texture, and spatial parameters are digitized to form the gear damage signature.

[0068] Step 240: Determine the damage type based on the gear damage characteristics to obtain a damage type determination result for the corresponding damage area.

[0069] Damage types may include friction damage, fatigue damage, non-fatigue fracture, and plastic deformation. Fatigue damage may further include tooth root fatigue damage and tooth surface contact fatigue damage. It should be noted that if the damage type determination result for the damaged area is either friction damage or fatigue damage, steps 250 to 260 of the embodiment may be continued to perform damage severity classification and remaining life estimation for the damaged area. Damage severity classification and life estimation allow for timely intervention (e.g., repair) before the damage reaches a critical state, keeping safety risks within an acceptable range and avoiding serious accidents such as gear fracture caused by damage progression. If the damage type determination result for the damaged area is non-fatigue fracture or plastic deformation, the aircraft engine gear under inspection should be directly eliminated or replaced, given the significant impact of such damage on aircraft engine operational stability and safety. Therefore, to conserve computing resources, the system may automatically terminate the damage analysis process after outputting a corresponding prompt message, and will not execute the subsequent evaluation operations in steps 250 to 260.

[0070] For the embodiments of the present disclosure, the steps of the embodiments may specifically include: inputting the gear damage features into a preset damage classification model for damage classification processing to obtain a damage type determination result for the corresponding damage area; wherein, the preset damage classification model performs hierarchical matching on the gear damage features based on a preset rule set when performing damage classification processing; if in the hierarchical matching results, different feature types all point to the same damage type, then it is used as the damage type determination result for the corresponding damage area; if in the hierarchical matching results, different feature types point to different damage types, then according to the weight priority corresponding to each feature type, the damage type with the highest corresponding weight priority is screened from at least two matching damage types as the damage type determination result for the corresponding damage area, and the preset rule set is constructed based on the gear material properties, failure mechanism and historical fault data.

[0071] The preset damage classification model is an algorithmic model built based on gear failure knowledge and historical data. It includes a set of rules and weighted priority logic, used to map damage features to specific damage types (e.g., fatigue cracks, friction and wear). The preset rule set is a set of matching rules derived from gear material properties (e.g., the fatigue limit of 18CrNiMo7-6 steel), failure mechanisms (e.g., the conditions for contact fatigue pitting), and historical failure data (e.g., 80% of the root cracks on a certain type of gear are greater than 1.5 mm in length). Hierarchical matching involves progressively verifying the damage type matching logic based on a specific order of feature types, such as "geometric features → texture features → spatial features." The first layer of features narrows the candidate types, while the next layer further confirms them, improving the accuracy of the judgment. The specific order can be customized based on the actual application scenario. In the following embodiments of this disclosure, the specific order of "geometric features → texture features → spatial features" is used as an example to illustrate the technical solutions of this application, but this does not constitute a specific limitation. Weight priority refers to the importance coefficient set for different feature types (such as 40% weight for geometric features, 35% weight for texture features, and 25% weight for spatial features). It is used to resolve type decisions when feature matching conflicts occur and ensure that key features dominate the judgment results.

[0072] Specifically, after inputting gear damage features (such as geometry, texture, and spatial parameters) into a pre-set damage classification model, the model performs a hierarchical matching of the gear damage features based on a pre-set rule set constructed from gear material properties, failure mechanisms, and historical failure data. For example, the model first preliminarily determines the damage type based on geometric features (such as crack length and pitting shape), and then further verifies it using texture features (LBP pattern, grayscale entropy) and spatial features (damage location and distribution). If the matching results for each feature type all point to the same type (e.g., geometric features showing a slender crack, texture features showing a linear LBP pattern, and spatial features located in the high-stress area of ​​the tooth root all indicate fatigue cracks), the damage type is directly determined as that type. If there are conflicting feature matching results (e.g., geometric features suggest wear damage, but texture features are consistent with fatigue damage), the damage type corresponding to the feature type with the highest weight is selected according to a pre-set weight priority (e.g., geometric features weight 40% > texture features 35% > spatial features 25%), and this is used as the final damage type determination result.

[0073] This hierarchical matching mechanism improves the accuracy of damage type identification and the efficiency of handling combined damage. For example, in the case of 18CrNiMo7-6 steel gears, traditional methods have a high misjudgment rate when the wear area is accompanied by microcracks. However, by prioritizing weights (geometric features first), it can accurately distinguish combined damage types (primarily friction wear and fatigue crack initiation), effectively reducing the misjudgment rate.

[0074] Step 250: Based on the gear damage characteristics and damage type determination results, the gear damage model is used to perform damage severity classification to obtain a damage severity classification result for the corresponding damage area.

[0075] Among them, gear damage models include fatigue crack theory model, tooth surface contact fatigue theory model and wear theory model.

[0076] For the embodiment of the present disclosure, step 250 may specifically include the following steps:

[0077] Step 250-1. If the damage type is determined to be tooth root fatigue damage, fatigue damage assessment parameters are calculated using a fatigue crack theory model based on the crack depth, geometric shape parameters, material constants, and tooth root nominal stress in the gear damage characteristics. The fatigue damage assessment parameters are compared with a first preset classification threshold to obtain a damage severity classification result for the corresponding damaged area. Crack propagation-related parameters include at least a stress intensity factor range and a crack propagation rate.

[0078] Among them, tooth root fatigue damage is fatigue crack initiation and expansion damage caused by cyclic bending load in the high stress area of ​​the tooth root, and is characterized by cracks mostly originating from the fillet of the tooth root and extending along the tooth thickness direction; the first preset classification threshold is a multi-segment quantitative standard set based on the fatigue characteristics, failure mechanism and engineering experience of the gear material, which divides the damage severity level by parameter intervals such as ΔK and da / dN to achieve refined classification of the evaluation results; the fatigue crack theoretical model is an empirical model with the Paris law model as the core, which describes the expansion law of cracks under cyclic loads through mathematical relationships, and is the physical basis for connecting damage characteristics and remaining life; the stress intensity factor range (ΔK) is the fluctuation amplitude of the stress field intensity at the crack tip under alternating loads, which is related to the crack depth, nominal stress and geometric shape, and the unit is , which directly affects the crack growth rate; the crack growth rate (da / dN) is the increment of crack depth per cycle (unit: mm / c, c represents cycle), which is calculated by the Paris formula. It is the core parameter reflecting the deterioration rate of fatigue damage and the key comparison indicator of multi-segment thresholds.

[0079] For the embodiment of the present disclosure, when the damage type is determined to be tooth root fatigue damage, the crack depth (a), geometric shape parameters (such as geometric correction factor Y), material constants (C, m) and tooth root nominal stress (σ) in the gear damage characteristics need to be substituted into the fatigue crack theoretical model to calculate the fatigue damage assessment parameters. The specific calculation process is: First, through the formula Calculate the stress intensity factor range ΔK and then use the formula Crack growth rate yes / dN , and finally ΔK, yes / dN The parameters are compared with the first preset grading threshold to determine the injury severity grading result.

[0080] For example, taking 18CrNiMo7-6 steel gear as an example, it is known that: crack depth a = 0.5 mm, geometric correction factor Y = 1.12 (semi-elliptical surface crack), tooth root nominal stress σ = 850 MPa, material constant C = 1.2×10 -12 , m=3.0, the first preset classification threshold is:

[0081] Level 1 minor damage: da / dN<1×10 -8 mm / c and ΔK<15 ;

[0082] Level 2 Moderate Damage: or 15 ≤ΔK<25 ;

[0083] Level 3 Severe Injury: or 25 ≤ΔK<35 ;

[0084] Level 4 Dangerous Injury: or ΔK ≥ 35 .

[0085] Calculate the stress intensity factor range ΔK:

[0086]

[0087] Calculate the crack growth rate da / dN:

[0088]

[0089] Compare the fatigue damage assessment parameters with the first preset classification threshold to obtain the damage severity classification result of the corresponding damage area:

[0090] ΔK=28.5 Belongs to Level 3 (25≤ΔK<35);

[0091] da / dN=2.78×10 -6 mm / c is greater than 1×10 -6 mm / c, judged as Level 4 (dangerous damage).

[0092] When the two conflict, according to the principle of "taking the highest level", the final damage severity classification result of the corresponding damaged area of ​​18CrNiMo7-6 steel gear is determined to be Level 4 (dangerous damage).

[0093] Step 250-2: If the damage type is determined to be tooth surface contact fatigue damage, the tooth surface contact fatigue theoretical model is used to calculate contact fatigue damage assessment parameters based on the contact pressure, friction coefficient, and gear material hardness in the gear damage characteristics. The contact fatigue damage assessment parameters are compared with the second preset classification threshold to obtain a damage severity classification result for the corresponding damage area. The contact fatigue damage assessment parameters include at least the tooth surface shear stress amplitude, the high stress area volume, and the crack initiation life.

[0094] Among them, tooth surface contact fatigue damage is material fatigue spalling or pitting damage caused by cyclic contact stress on the tooth surface, characterized by pit-shaped defects on the tooth surface, mostly distributed in the pitch line contact area; the tooth surface contact fatigue theoretical model is a physical model based on the Lundberg-Palmgren theory, which predicts the crack initiation life through parameters such as contact stress and material properties. The core is to establish a mathematical relationship between shear stress amplitude and fatigue life; tooth surface shear stress amplitude ( t 0) is the maximum orthogonal shear stress of the secondary surface of the tooth contact area, which is related to the contact pressure and friction coefficient, and is expressed in MPa. It is the key driving parameter for contact fatigue damage. The volume of the high stress area (V) is the volume of the area in the contact area where the shear stress exceeds the allowable value of the material, and is expressed in , reflects the spatial range of stress concentration, and is related to the contact area geometry and material hardness; crack initiation life (N) is the number of cycles from no damage to the initiation of fatigue cracks on the tooth surface, which is a direct quantitative indicator of the severity of contact fatigue damage and the core comparison parameter of the multi-segment threshold; the second preset classification threshold is a multi-segment quantitative standard set based on the gear contact fatigue failure mechanism, material properties and operating conditions, through t The damage levels are divided into parameter intervals such as 0, V, and N, supporting the refined evaluation of aircraft engine gears.

[0095] For the embodiment of the present disclosure, when the damage type is determined to be tooth surface contact fatigue damage, it is necessary to determine the damage type based on the contact pressure ( p 0), friction coefficient ( m ) and gear material hardness (HV), and substitute into the tooth surface contact fatigue theoretical model to calculate the contact fatigue damage assessment parameters. The specific calculation process is: First, through the formula Calculate the shear stress amplitude on the tooth surface ( t 0), combined with the contact area ellipse parameters (major semi-axis a, minor semi-axis b) and the material allowable shear stress ( t crit =0.25×HV) to calculate the volume of the high stress area (V), the calculation formula is: , and finally use (c is the material constant, usually 7-9, which needs to be calibrated by fatigue test) to obtain the crack initiation life (N), that is, the crack initiation life N is related to t The product of 0 to the power of c and V is inversely proportional. t Parameters such as 0, V, and N are compared with the second preset grading thresholds of multiple segments, and the injury severity grading result is determined through interval matching. Among them, β is the attenuation coefficient, which is generally set as β=2.5 / b according to the empirical formula, and γ is the shape factor, which is usually set as γ=1.5.

[0096] For example, taking 18CrNiMo7-6 carburized steel gear as an example, it is known that: contact pressure p 0 = 1800 MPa, friction coefficient μ = 0.15, material hardness HV = 650 (τcrit = 0.25 × 650 = 162.5 MPa), contact area ellipse parameters: major semi-axis a = 0.3 mm, minor semi-axis b = 0.15 mm, material constant c = 8, the second preset classification threshold is:

[0097] Level 1 Minor Injury: t 0<100MPa and V<0.05 And N>1×10 6 cycle;

[0098] Level 2 moderate damage: 100MPa≤ t 0<150MPa or 0.05 ≤V<0.1 or ;

[0099] Level 3 Severe Damage: 150 ≤ t 0<180MPa or 0.1 ≤V<0.2 or ;

[0100] Level 4 Dangerous Injury: t 0≥180MPa or V≥0.2 or .

[0101] Calculation of tooth surface shear stress amplitude t 0:

[0102] t 0=0.25×1800× ≈450×1.044=470MPa

[0103] Calculate the volume V of the high stress area:

[0104] Attenuation coefficient β=2.5 / b=2.5 / 0.15≈16.67, shape factor γ=1.5,

[0105] =1.5×(3.14×0.3×0.15)×e^(-16.67×470 / 162.5)≈1.5×0.1414×e^(-48.5)≈2.54× (because t 0 is much larger than τcrit, V approaches 0)

[0106] Calculate the crack initiation life N:

[0107] because t 0=470MPa is much larger than τcrit=162.5MPa. It can be seen that the theoretical value of N is extremely small (far less than 1× cycle), thus triggering the Level 4 classification conditions.

[0108] The contact fatigue damage assessment parameters are compared with the second preset classification threshold to obtain the damage severity classification result of the corresponding damage area:

[0109] t 0=470MPa≥180MPa, judged as Level 4 (dangerous damage),

[0110] V=2.54× Although small, t 0 triggers the highest level. When the two conflict, the principle of "taking the highest level" is followed. Therefore, the final injury severity classification result of the corresponding damaged area is determined to be Level 4 (dangerous injury).

[0111] Step 250-3: If the damage type is determined to be friction damage, the friction and wear assessment parameters are calculated using a wear theory model based on the gear meshing normal load, total sliding distance, gear material hardness, and wear coefficient in the gear damage characteristics. The friction and wear assessment parameters are compared with a third preset classification threshold to obtain a damage severity classification result for the corresponding damaged area. The friction and wear assessment parameters include at least the gear wear volume and wear amount.

[0112] Among them, friction damage is the gradual removal of material due to the relative sliding of the gear meshing surfaces, characterized by a decrease in tooth surface smoothness, scratches or uniform wear bands, mostly distributed in the transition zone between the tooth top and the tooth root; the wear theory model is an empirical model based on the Archard formula, which is used to describe the proportional relationship between wear volume and normal load and sliding distance, and the inverse relationship with material hardness. It is the core tool for quantifying friction damage. Gear wear volume (V) is the amount of material removed due to friction damage, and its unit is , which directly reflects the cumulative effect of the degree of wear; the wear amount, usually expressed as linear wear amount (Δh), in mm, refers to the reduction in tooth surface thickness, which is related to the wear volume and contact area, and is a key parameter for evaluating gear meshing accuracy; the third preset classification threshold is a multi-segment quantitative standard set based on the wear resistance of gear material, operating load and accuracy requirements, which divides the wear severity level into grades through parameter intervals such as V and Δh, supporting preventive maintenance decisions for aviation gears.

[0113] For the embodiment of the present disclosure, when the damage type is determined to be friction damage, it is necessary to determine the gear damage characteristics based on the meshing normal load ( F N )、Total sliding distance( S total ), material hardness (H) and wear coefficient ( K ), substitute into the wear theory model to calculate the friction and wear evaluation parameters. First, use the formula Calculate the gear wear volume (V), and convert V into linear wear (e.g., tooth surface thickness reduction Δh = V / S, where S is the wear contact area). Finally, compare parameters such as V and Δh with the third preset classification threshold of multiple segments to determine the damage severity classification result. or is the lubrication correction factor, β is the roughness correction factor.

[0114] For example, suppose a gear is damaged during operation and is judged to be friction damage. The parameters related to the gear damage characteristics are as follows: meshing normal load F N = 5000N, total sliding distance S total =1000m, material hardness H=200HV, wear coefficient K=1× , lubrication correction factor η=0.8, roughness correction factor β=0.9, wear contact area S=0.01 The third preset classification threshold is:

[0115] Level 1 Minor Damage: Wear Volume And the linear wear amount Δh<0.5mm;

[0116] Level 2 Moderate Damage: or 0.5mm≤Δh<1mm;

[0117] Level 3 Severe Injury: And Δh≥1mm.

[0118] Calculate the gear wear volume (V):

[0119]

[0120] Calculate the linear wear Δh:

[0121] Δh=V / S=3.6× ÷0.01=3.6× m=0.36mm

[0122] Since V = 3.6× <5× And Δh=0.36mm<0.5mm, which meets the mild damage condition. Therefore, the friction damage severity level of the gear is classified as Level 1 mild damage.

[0123] Step 260: Based on the gear damage characteristics, damage type determination results, and damage severity classification results, the damage evolution model is used to estimate the remaining life to obtain the remaining life estimation result of the corresponding damage area.

[0124] Among them, the damage evolution model includes fatigue crack evolution model, contact fatigue evolution model and wear evolution model.

[0125] For the embodiment of the present disclosure, step 260 may specifically include the following steps:

[0126] Step 260-1: If the damage type is determined to be tooth root fatigue damage, based on the crack depth, geometric parameters, and material constants in the gear damage characteristics, combined with the crack growth rate in the damage severity classification result, the fatigue crack evolution model is used to predict the first remaining number of cycles for the crack to grow from the current state to the critical state, and the first remaining number of cycles is converted into a remaining life estimation result for the corresponding damaged area.

[0127] Among them, the fatigue crack evolution model is a physical model based on the Paris law, which describes the relationship between the crack growth rate and the stress intensity factor, and is used to quantify the expansion process of the fatigue crack from the current state to the critical state; the first remaining number of cycles (Np) is the number of cyclic loads required for the crack to grow from the current depth to the critical depth, which is calculated by the Paris law integral and is the core parameter for remaining life estimation.

[0128] For the embodiment of the present disclosure, when the damage type is determined to be tooth root fatigue damage, the crack depth (a), geometric shape parameters (such as geometric correction factor Y) and material constants (C, m) in the gear damage characteristics are combined with the crack growth rate (da / dN) obtained by the damage severity classification, and the fatigue crack evolution model is used to predict the crack growth rate from the current crack depth a0 to the critical crack depth a1. c The first remaining number of cycles (N p Specifically, we can first use the formula Calculate the stress intensity factor range (ΔK), where σ is the nominal stress at the tooth root, and substitute it into the Paris law formula The relationship between crack growth rate and crack depth is obtained by Integral calculation from the current crack depth a0 to the critical crack depth a c The integral value is the first remaining number of cycles N p Finally, N is set according to the gear running frequency. p Convert to remaining life (e.g. hours).

[0129] Step 260-2: If the damage type is determined to be tooth surface contact fatigue damage, based on the contact pressure, friction coefficient, and gear material hardness in the gear damage characteristics, combined with the volume of the high stress area and the crack initiation life in the damage severity classification results, the contact fatigue evolution model is used to predict the second remaining cycles to reach critical damage based on the working load spectrum, and the second remaining cycles are converted into a remaining life estimation result for the corresponding damaged area.

[0130] The contact fatigue evolution model is a physical model based on the Lundberg-Palmgren theory. It predicts the crack initiation and propagation life through parameters such as contact stress and material properties. The core is to establish an inverse relationship between shear stress amplitude and fatigue life. f ) refers to the number of cycles required for the gear to develop from the current contact fatigue damage state to the critical damage state, which is calculated using the Lundberg-Palmgren formula combined with load spectrum correction.

[0131] For the embodiment of the present disclosure, when the damage type is determined to be tooth surface contact fatigue damage, it is necessary to determine the damage type based on the contact pressure ( p 0), friction coefficient ( m ) and material hardness (HV), combined with the high stress area volume (V) and crack initiation life (N) obtained by damage severity classification, using contact fatigue evolution model (such as Lundberg-Palmgren theory), according to the working load spectrum to predict the second remaining cycle number (N) from the current damage state to critical damage (such as pitting depth of 0.5mm). fSpecifically, first use the formula Calculate the shear stress amplitude on the tooth surface ( t 0), combined with the allowable shear stress of the material ( t crit =0.25×HV) to determine the stress excess area, using the Lundberg-Palmgren formula

[0132] (c is the material constant, usually 7-9, which needs to be calibrated by fatigue test) reverse the number of remaining cycles, and finally get the corrected value based on the load spectrum And converted into remaining life (such as hours).

[0133] Step 260-3: If the damage type is determined to be friction damage, based on the wear amount in the gear damage characteristics, the gear material hardness, the gear meshing normal load and the wear coefficient, combined with the wear amount in the damage severity classification result, the wear evolution model is used to reversely infer the third remaining cycles of the gear meshing through the maximum allowable wear amount, and the third remaining cycles are converted into the remaining life estimation result of the corresponding damage area.

[0134] The wear evolution model is a physical model based on the Archard formula, which describes the proportional relationship between wear volume and normal load and sliding distance, and the inverse relationship with material hardness. It is used to quantify the evolution of wear with the number of cycles. The third residual cycle number ( ) is the number of meshing cycles required for the gear to develop from its current wear state to the maximum allowable wear amount, which is obtained by reverse deduction of the Archard formula and is the core parameter for remaining life estimation.

[0135] For the embodiment of the present disclosure, when the damage type is determined to be friction damage, it is necessary to determine the damage type based on the wear amount (Δh), material hardness (HV), meshing normal load ( ) and wear coefficient (K), combined with the wear data obtained from the damage severity classification, using the wear evolution model (such as Archard formula), through the maximum allowable wear ( ) The third remaining cycle number of reverse gear meshing ( Specifically, first use the Archard formula (V is the wear volume, L is the sliding distance) The relationship between the linear wear volume and the number of cycles is derived, and then the remaining number of cycles is inversely solved by the maximum allowable wear volume. , and finally converted into remaining life (such as operating hours).

[0136] In summary, the technical solution in this application can effectively solve many problems of traditional detection technology in aircraft engine gear damage detection through automated image acquisition and analysis processes. Specifically, by collecting multi-angle gear surface images and pre-processing them such as noise reduction, contrast enhancement and viewpoint normalization, the image quality can be improved for subsequent processing. By segmenting the gear area through edge detection, locating the damaged area with the help of an improved deep learning algorithm and quantifying its geometric, texture and spatial features, accurate identification and description of the damage can be achieved, which can greatly improve the detection efficiency and reduce the missed detection rate. Furthermore, based on the preset damage classification model constructed based on the gear material properties, failure mechanism, etc., the damage type is determined, and the corresponding theoretical model is combined to evaluate the severity of the damage, which can solve the problem of traditional technology's difficulty in detecting damage such as tooth surface contact fatigue and friction wear. Finally, the use of the damage evolution model to predict the remaining life can provide a reliable basis for the maintenance of aircraft engine gears and meet their requirements for reliability and safety.

[0137] Furthermore, in order to fully illustrate the implementation of this embodiment, this embodiment also provides an aircraft engine gear damage analysis system based on image recognition, such as Figure 3 As shown, the system includes: a data acquisition module 310, an image analysis module 320, and a gear damage assessment module 330. The data acquisition module 310 is used to collect gear surface images of the aircraft engine gear to be inspected at multiple angles; the image analysis module 320 is used to preprocess the gear surface images, sequentially perform gear region segmentation and damage region location processing on the preprocessed gear surface images, and quantify the damage features of each located damage region to obtain gear damage features. The preprocessing includes at least image noise reduction, image contrast enhancement, and viewpoint normalization processing. The gear damage features include at least geometric features, texture features, and spatial features. The gear damage assessment module 330 is used to perform damage analysis on each damaged region based on the gear damage features to obtain damage analysis results. The damage analysis results include at least damage type determination results, damage severity classification results, and remaining life estimation results.

[0138] For the relevant technical implementation of the data acquisition module 310, the image analysis module 320 and the gear damage assessment module 330, please refer to the relevant embodiment description of the above-mentioned aircraft engine gear damage analysis method based on image recognition, which will not be repeated here.

[0139] In specific application scenarios, such as Figure 3As shown, the system also includes a system platform module 340, which consists of a human-computer interaction submodule 3401, a data management submodule 3402, and a parameter mapping submodule 3403. The human-computer interaction submodule 3401 provides a user interface for interacting with the system, supporting the input of damage detection parameters (such as crack depth and contact pressure), the visualization of damage assessment results (such as damage grading maps and remaining life curves), and interactive decision-making for maintenance strategies. For example, users can use this module to upload gear inspection data, view 3D damage reconstruction models, and receive maintenance alerts. The data management submodule 3402 is responsible for the storage, retrieval, and management of gear damage inspection data, including historical inspection data (such as wear volume and crack growth rate from each inspection), a material parameter library (such as the hardness and fatigue constant of 18CrNiMo7-6 steel), and a damage case library. Data encryption and permission management are supported to ensure the security and traceability of aircraft engine gear inspection data. The parameter mapping submodule 3403 establishes a mapping relationship between gear damage characteristic parameters (such as stress intensity factor and wear volume) and assessment model parameters (such as Paris law constant and Archard wear coefficient). This module optimizes parameter matching accuracy through machine learning algorithms. For example, the real-time detected tooth root crack depth is automatically mapped to the corresponding geometric correction factor and material constant, improving the accuracy of damage assessment.

[0140] Accordingly, the system platform module 340 includes: a human-computer interaction submodule 3401, configured to provide a user interaction interface for inputting gear damage characteristic parameters, displaying damage severity grading results and remaining life estimation results; a data management submodule 3402, configured to store gear material parameters, historical inspection data and damage cases, and support data retrieval and permission management; a parameter mapping submodule 3403, configured to establish a mapping relationship between damage characteristic parameters and evaluation model parameters, and optimize the calculation parameters of the fatigue crack theory model and the wear theory model based on the mapping relationship.

[0141] Accordingly, the human-computer interaction submodule 3401 is further configured to: receive damage characteristic parameters such as meshing normal load, crack depth, contact pressure, etc. input by the user; display the damage severity classification results (such as Level 1-4) and remaining life estimation results (such as remaining number of cycles, operating time) in the form of a chart; generate maintenance strategy recommendations (such as immediate repair, regular monitoring) and support user confirmation operations.

[0142] The data management submodule 3402 is further configured to: establish a material parameter library containing gear material hardness, fatigue constant, and wear coefficient; store damage evolution data such as wear volume and crack growth rate of each gear test; and encrypt and store the test data based on blockchain technology to ensure that the data cannot be tampered with.

[0143] The parameter mapping submodule 3403 is further configured as follows: based on the deep learning model, the real-time detected crack depth is mapped to the corresponding geometric correction factor (Y) and material constant (C, m); according to the hardness and lubrication conditions of the gear material, the wear coefficient (K) and correction factor (η, β) are automatically matched; the parameter mapping relationship library is dynamically updated, and the mapping accuracy is optimized by comparing historical detection data with actual damage results.

[0144] In summary, the technical solution in this application can effectively solve many problems of traditional detection technology in aircraft engine gear damage detection through automated image acquisition and analysis processes. Specifically, by collecting multi-angle gear surface images and pre-processing them such as noise reduction, contrast enhancement and viewpoint normalization, the image quality can be improved for subsequent processing. By segmenting the gear area through edge detection, locating the damaged area with the help of an improved deep learning algorithm and quantifying its geometric, texture and spatial features, accurate identification and description of the damage can be achieved, which can greatly improve the detection efficiency and reduce the missed detection rate. Furthermore, based on the preset damage classification model constructed based on the gear material properties, failure mechanism, etc., the damage type is determined, and the corresponding theoretical model is combined to evaluate the severity of the damage, which can solve the problem of traditional technology's difficulty in detecting damage such as tooth surface contact fatigue and friction wear. Finally, the use of the damage evolution model to predict the remaining life can provide a reliable basis for the maintenance of aircraft engine gears and meet their requirements for reliability and safety.

[0145] Based on the above Figure 1 The method shown in FIG. 1 is a method for performing the above-mentioned operation. Accordingly, this embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned Figure 1 The image recognition-based damage analysis method for aircraft engine gears is shown.

[0146] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), and includes a number of instructions for enabling an electronic device (which can be a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the present application.

[0147] Based on the above Figure 1 、 2 The method shown, and Figure 3 In order to achieve the above-mentioned purpose, the embodiment of the present application further provides an electronic device, which can be a personal computer, a tablet computer, a server, or other network equipment, etc. The device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1 、 2The image recognition-based damage analysis method for aircraft engine gears is shown.

[0148] Optionally, the physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, and the like. The user interface may include a display screen and an input unit such as a keyboard. Optional user interfaces may also include a USB interface and a card reader interface. Optionally, the network interface may include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0149] Those skilled in the art will understand that the above-mentioned physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0150] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device, supporting the execution of information processing programs and other software and / or programs. The network communication module is used to enable communication between components within the storage medium, as well as with other hardware and software within the physical information processing device.

[0151] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform, or by hardware.

[0152] Those skilled in the art will understand that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required to implement the present application. Those skilled in the art will understand that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0153] The serial numbers of the above application are for descriptive purposes only and do not represent the advantages or disadvantages of the implementation scenarios. The above disclosure only discloses several specific implementation scenarios of the present application, but the present application is not limited thereto. Any changes that can be conceived by those skilled in the art should fall within the scope of protection of the present application.

Claims

1. A method for analyzing damage of aircraft engine gears based on image recognition, characterized in that: include: Acquire gear surface images of the aircraft engine gear to be inspected at multiple angles; Preprocessing the gear surface image, wherein the preprocessing includes at least image noise reduction, image contrast enhancement, and viewpoint normalization; Performing gear region segmentation and damage region location processing on the preprocessed gear surface image in sequence, and quantifying damage features of each located damage region to obtain gear damage features, wherein the gear damage features include at least geometric features, texture features, and spatial features; Among them, the gear surface image after the preprocessing is sequentially subjected to gear region segmentation and damage region positioning processing, including: using edge detection image processing technology to perform gear region segmentation on the gear surface image after the preprocessing to obtain a gear region separated from the background region; within the segmented gear region, applying the improved deep learning YOLO target detection algorithm to locate the damage region to obtain each damage region, and the improved deep learning YOLO target detection algorithm introduces a deformable convolution module, wherein the deformable convolution module is used to: receive a feature map from the backbone network as input, and predict the spatial offset from the feature map through a trainable convolution kernel weight matrix to generate an adaptively adjusted sampling point set; use the sampling point set to perform a convolution operation on the feature map to dynamically capture the boundary features of irregular damage; output an adaptive damage feature map based on the boundary features for subsequent multi-scale feature fusion and damage region positioning; Based on the gear damage characteristics, damage analysis is performed on each of the damaged areas to obtain damage analysis results, which at least include damage type determination results, damage severity classification results and remaining life estimation results.

2. The method according to claim 1, characterized in that The gear surface image is preprocessed, wherein the preprocessing includes at least image noise reduction, image contrast enhancement, and viewpoint normalization, including: Using Gaussian filtering to perform noise reduction on the gear surface image; The contrast of the gear surface image after noise reduction is enhanced by using histogram equalization; Based on the camera calibration parameters and gear pose information, geometric distortion correction and perspective transformation are performed on the gear surface image after contrast enhancement to achieve viewpoint normalization, so that the gear surface images collected at different angles have a unified viewing angle reference.

3. The method according to claim 1, characterized in that The damage analysis is performed on each of the damaged areas based on the gear damage characteristics to obtain a damage analysis result, wherein the damage analysis result at least includes a damage type determination result, a damage severity classification result, and a remaining life estimation result, including: Determine the damage type based on the gear damage characteristics to obtain a damage type determination result for the corresponding damage area; Based on the gear damage characteristics and the damage type determination result, a gear damage model is used to perform damage severity classification to obtain a damage severity classification result for the corresponding damage area; Based on the gear damage characteristics, the damage type determination result and the damage severity classification result, the remaining life is estimated using a damage evolution model to obtain a remaining life estimation result for the corresponding damage area.

4. The method according to claim 3, characterized in that The determining of the damage type based on the gear damage characteristics to obtain a damage type determination result corresponding to the damage area includes: Inputting the gear damage characteristics into a preset damage classification model for damage classification processing to obtain a damage type determination result for the corresponding damage area; Among them, when the preset damage classification model performs damage classification processing, the gear damage features are hierarchically matched based on a preset rule set; if in the hierarchical matching results, different feature types all point to the same damage type, it is used as the damage type determination result of the corresponding damage area; if in the hierarchical matching results, different feature types point to different damage types, then according to the weight priority corresponding to each feature type, the damage type corresponding to the highest weight priority is screened from at least two matching damage types as the damage type determination result of the corresponding damage area, and the preset rule set is constructed based on gear material properties, failure mechanism and historical fault data.

5. The method according to claim 4, characterized in that The gear damage model includes a fatigue crack theory model, a tooth surface contact fatigue theory model and a wear theory model; Based on the gear damage characteristics and the damage type determination result, the gear damage model is used to perform damage severity classification, and the damage severity classification result of the corresponding damage area is obtained, including: If the damage type is determined to be tooth root fatigue damage, then based on the crack depth, geometric shape parameters, material constants, and tooth root nominal stress in the gear damage characteristics, a fatigue damage assessment parameter is calculated using the fatigue crack theory model, and the fatigue damage assessment parameter is compared with a first preset classification threshold to obtain a damage severity classification result for the corresponding damaged area, wherein the crack growth-related parameters include at least a stress intensity factor range and a crack growth rate; If the damage type is determined to be tooth surface contact fatigue damage, then based on the contact pressure, friction coefficient, and gear material hardness in the gear damage characteristics, the tooth surface contact fatigue theoretical model is used to calculate contact fatigue damage assessment parameters, and the contact fatigue damage assessment parameters are compared with a second preset classification threshold to obtain a damage severity classification result for the corresponding damage area, wherein the contact fatigue damage assessment parameters include at least the tooth surface shear stress amplitude, the high stress area volume, and the crack initiation life; If the damage type is determined to be friction damage, the friction and wear evaluation parameters are calculated using the wear theory model based on the gear meshing normal load, total sliding distance, gear material hardness and wear coefficient in the gear damage characteristics. The friction and wear evaluation parameters are compared with the third preset classification threshold to obtain the damage severity classification result of the corresponding damaged area. The friction and wear evaluation parameters include at least the gear wear volume and wear amount.

6. The method according to claim 5, characterized in that The damage evolution model includes a fatigue crack evolution model, a contact fatigue evolution model and a wear evolution model; Based on the gear damage characteristics, the damage type determination result, and the damage severity classification result, the remaining life is estimated using a damage evolution model to obtain a remaining life estimation result for the corresponding damage area, including: If the damage type determination result is tooth root fatigue damage, then based on the crack depth, geometric shape parameters, and material constants in the gear damage characteristics, combined with the crack growth rate in the damage severity classification result, the fatigue crack evolution model is used to predict a first remaining number of cycles for the crack to grow from the current state to the critical state, and the first remaining number of cycles is converted into a remaining life estimation result for the corresponding damaged area; If the damage type is determined to be tooth surface contact fatigue damage, then based on the contact pressure, friction coefficient, and gear material hardness in the gear damage characteristics, combined with the volume of the high stress area and the crack initiation life in the damage severity classification result, the contact fatigue evolution model is used to predict the second remaining number of cycles to reach critical damage based on the working load spectrum, and the second remaining number of cycles is converted into a remaining life estimation result for the corresponding damaged area; If the damage type is determined to be friction damage, based on the wear amount, gear material hardness, gear meshing normal load and wear coefficient in the gear damage characteristics, combined with the wear amount in the damage severity classification result, the wear evolution model is used to reversely infer the third remaining cycles of gear meshing through the maximum allowable wear amount, and the third remaining cycles are converted into a remaining life estimation result for the corresponding damage area.

7. An aircraft engine gear damage analysis system based on image recognition, characterized in that: include: Data acquisition module, image analysis module and gear damage assessment module; The data acquisition module is used to acquire gear surface images of the aircraft engine gear to be inspected at multiple angles; The image analysis module is used to preprocess the gear surface image, perform gear region segmentation and damage region location processing on the preprocessed gear surface image in sequence, and quantify damage features of each located damage region to obtain gear damage features, wherein the preprocessing includes at least image noise reduction, image contrast enhancement, and viewpoint normalization, and the gear damage features include at least geometric features, texture features, and spatial features; Among them, the gear surface image after the preprocessing is sequentially subjected to gear region segmentation and damage region positioning processing, including: using edge detection image processing technology to perform gear region segmentation on the gear surface image after the preprocessing to obtain a gear region separated from the background region; within the segmented gear region, applying the improved deep learning YOLO target detection algorithm to locate the damage region to obtain each damage region, and the improved deep learning YOLO target detection algorithm introduces a deformable convolution module, wherein the deformable convolution module is used to: receive a feature map from the backbone network as input, and predict the spatial offset from the feature map through a trainable convolution kernel weight matrix to generate an adaptively adjusted sampling point set; use the sampling point set to perform a convolution operation on the feature map to dynamically capture the boundary features of irregular damage; output an adaptive damage feature map based on the boundary features for subsequent multi-scale feature fusion and damage region positioning; The gear damage assessment module is used to perform damage analysis on each of the damaged areas based on the gear damage characteristics to obtain damage analysis results, which at least include damage type determination results, damage severity classification results and remaining life estimation results.

8. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

Images