Gearbox gear defect detection method, system and equipment

Abstract

The invention relates to the technical field of surface defect detection, and discloses a gearbox gear defect detection method. The method comprises the steps of collecting a full-surface image of a gear, performing multi-layer feature extraction through image preprocessing in combination with a convolutional neural network, performing gridding segmentation and defect probability calculation through a Bayesian discrimination model, performing defect detection and classification on a defect-prone area and a defect-non-prone area by adopting a single-category model and a multi-category model respectively, and obtaining a defect-prone area and a defect-non-prone area. Generating a defect report; compared with the problems that in the prior art, manual detection is relied on, the speed is low, precision is insufficient, and particularly under the condition of a high-speed production line or uneven illumination, a traditional detection technology is difficult to effectively recognize small defects, real-time and high-precision gear surface defect detection is achieved by combining automatic detection based on deep learning, and the detection efficiency is improved. Therefore, the problems of missing detection and false detection in manual detection are avoided, the detection efficiency and accuracy are remarkably improved, and the method is suitable for a modern automatic production line.

Description

Technical Field

[0001] The present invention belongs to the technical field of gearbox gear defect detection, and in particular relates to a gearbox gear defect detection method, system and equipment. Background Art

[0002] At present, in the production process of gearbox gears, surface defect detection is usually carried out by manual inspection or static image acquisition. However, these traditional inspection methods have obvious limitations and cannot meet the high efficiency and high precision requirements of modern automated production lines.

[0003] Manual inspection relies on the experience and visual judgment of quality inspectors. Although it can find relatively obvious surface defects in some cases, it is highly subjective, has high rates of missed and false detections, and is inefficient. Due to differences in the judgment criteria of different inspectors, the inspection results are quite inconsistent, which affects the quality control of the product. In addition, manual inspection requires checking each gear one by one, which reduces production efficiency. Especially in large-scale production, manual inspection is costly and time-consuming.

[0004] Although static image detection technology has improved the level of automation of detection to a certain extent, the technology still has significant defects. Static image detection cannot achieve real-time online detection. Most of the time, it is carried out after the gear is produced or when production is interrupted. The feedback is not timely, which affects the continuity of production. Due to the limitations of external factors such as lighting conditions and camera angles, the static image detection system is prone to collect distorted or blurred images under conditions of uneven lighting, high-speed rotation of gears or vibration of the production line, resulting in insufficient detection accuracy. In particular, traditional static detection systems mainly rely on fixed algorithms or simple image processing methods, which make it difficult to accurately identify complex surface defects.

[0005] Existing technologies cannot fully meet the needs of modern automated production lines for real-time, efficient, and accurate detection, especially in scenarios such as high-speed production lines, complex lighting conditions, and production line vibration interference. Existing detection methods are difficult to provide reliable detection results. Therefore, there is an urgent need for a surface defect detection technology that can still achieve real-time online, automated, and high-precision detection in high-speed production lines and complex production environments. Summary of the invention

[0006] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to propose a gearbox gear defect detection method, aiming to solve the technical problem that the detection method in the prior art is difficult to accurately identify subtle defects on complex surfaces, especially in a noisy data environment, and cannot achieve accurate detection of micron-level defects.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions: The present invention provides a gearbox gear defect detection method,

[0008] The gearbox gear defect detection method comprises:

[0009] Step S10: A 120-frame / second CMOS industrial camera is used to acquire a full-surface image of the gear in real time. The industrial camera is fixed by a mechanical arm above the production line, maintaining a 90-degree vertical angle with the gear, and an LED adjustable light source system is used to ensure uniform illumination;

[0010] Step S20: preprocessing the collected gear surface image, including:

[0011] Median filter: Use a 3x3 median filter to remove noise interference in the image;

[0012] Histogram equalization: Perform histogram equalization on images with uneven grayscale value distribution;

[0013] Edge enhancement processing: Use the Canny edge detection algorithm to enhance the edge contour of the gear;

[0014] The whole process of image preprocessing is completed within 1ms to ensure that the speed of real-time detection is not affected;

[0015] Step S30: Perform multi-layer feature extraction on the preprocessed gear surface image based on the convolutional neural network CNN. The convolutional neural network CNN consists of 3 convolutional layers and 2 fully connected layers. The size of the convolution kernel is set to 3x3 and the step size is set to 1. After each convolution layer, the maximum pooling operation is used, and the pooling window size is 2x2. The extracted features include:

[0016] Texture features: used to analyze the texture irregularities on the gear surface, especially to detect abnormal textures caused by defects in the manufacturing process;

[0017] Grayscale gradient feature: used to capture subtle depressions or protrusions on the gear surface based on grayscale gradient changes to identify tiny defects;

[0018] Geometric features: used to extract the tooth edge and contact surface structure of the gear, focusing on identifying wear and crack geometric anomalies;

[0019] Step S40: mesh the gear surface according to a predetermined mesh size, dynamically adjust the size of each mesh based on the extracted geometric features and surface roughness, and calculate the defect probability of each mesh using a probability model based on Bayesian discrimination. When the defect probability of a mesh exceeds 0.75, the mesh area is marked as a defect-prone area, and the remaining mesh areas are treated as non-defect-prone areas.

[0020] Step S50: Use a single-category convolutional neural network model to identify defect-prone areas. The training samples of the model are mainly based on historical defect data, including cracks, pitting and edge damage on the gear surface. The recognition process is binary classification. The model outputs whether there are surface defects and gives the specific defect type when there are defects. For non-defect-prone areas, a multi-category Softmax classifier is used to detect different surface defect categories. The model is optimized through a weighted loss function to ensure the recognition accuracy of low-frequency defect categories.

[0021] Step S60: Generate a test report based on the defect identification results, the report includes the following contents:

[0022] Defect type and location: For each grid unit, the specific type of defect (such as crack, wear, pitting) and its location are marked, with an accuracy range of ±0.1mm;

[0023] Defect area and depth estimation: For defects such as cracks and wear, the area and depth of the defect are calculated by combining grayscale gradient and geometric features;

[0024] Maintenance suggestions: Automatically generate maintenance suggestions based on the severity of the defect. If the crack depth exceeds 0.5mm or the wear area exceeds 1mm 2 , it is recommended to replace the gear immediately; for minor defects, it is recommended to perform surface repair during the next maintenance cycle.

[0025] Preferably, in step S40, the step of dynamically adjusting the size of each grid based on the extracted geometric features and surface roughness adopts the formula:

[0026]

[0027] Among them, G size Indicates the size of the grid, D max is the maximum diameter of the gear, σ r is the standard deviation of surface roughness, and α is the adjustment coefficient.

[0028] Preferably, in step S40, the step of calculating the defect probability of each grid using a probability model based on Bayesian discrimination adopts the formula:

[0029]

[0030] Among them, P(defect|features) is the posterior probability, which indicates the probability of a defect occurring in the region after the preset features are observed; P(features|defect)·P(defect) is the likelihood probability, which indicates the probability of the preset features appearing under the premise that a defect is known to exist in a certain region; P(defect) is the prior probability, which indicates the probability of a defect occurring in a certain region when the preset features are not observed; P(features) is the marginal probability of the feature, which indicates the probability of the preset features appearing in all cases.

[0031] Preferably, in step S60, the step of calculating the area and depth of the defect adopts the formula:

[0032]

[0033] Where I is the gray value distribution of the defect area, and x and y are the coordinate axes in the image.

[0034] Preferably, in step S10, the image acquisition process is performed at a rotation speed of the gear of 500 to 1000 revolutions per minute to ensure that the captured image is clear.

[0035] Preferably, in step S10, the adjustable light source system automatically adjusts the light intensity according to the reflection condition of the gear surface.

[0036] Preferably, in step S40, the predetermined grid size is that the side length of each grid unit is 0.5 mm.

[0037] The present invention also provides a gearbox gear defect detection system comprising:

[0038] Image acquisition module, used to acquire the full surface image of the gear in real time using a 120-frame / second CMOS industrial camera during the production process of automobile gearbox gears. The industrial camera is fixed by a mechanical arm above the production line, maintaining a 90-degree vertical angle with the gear, and an LED adjustable light source system is used to ensure uniform illumination;

[0039] The image preprocessing module is used to preprocess the collected gear surface images, including:

[0040] Median filter: Use a 3x3 median filter to remove noise interference in the image;

[0041] Histogram equalization: Perform histogram equalization on images with uneven grayscale value distribution;

[0042] Edge enhancement processing: Use the Canny edge detection algorithm to enhance the edge contour of the gear;

[0043] The whole process of image preprocessing is completed within 1ms to ensure that the speed of real-time detection is not affected;

[0044] The feature extraction module is used to perform multi-layer feature extraction on the preprocessed gear surface image based on the convolutional neural network CNN. The convolutional neural network CNN consists of 3 convolutional layers and 2 fully connected layers. The size of the convolution kernel is set to 3x3 and the step size is set to 1. After each convolution layer, the maximum pooling operation is used, and the pooling window size is 2x2. The extracted features include:

[0045] Texture features: used to analyze the texture irregularities on the gear surface, especially to detect abnormal textures caused by defects in the manufacturing process;

[0046] Grayscale gradient feature: used to capture subtle depressions or protrusions on the gear surface based on grayscale gradient changes to identify tiny defects;

[0047] Geometric features: used to extract the tooth edge and contact surface structure of the gear, focusing on identifying wear and crack geometric anomalies;

[0048] The meshing module is used to mesh the gear surface according to the predetermined mesh size, dynamically adjust the size of each mesh based on the extracted geometric features and surface roughness, and calculate the defect probability of each mesh using a probability model based on Bayesian discrimination. If the defect probability of a mesh exceeds 0.75, the mesh area is marked as a defect-prone area, and the remaining mesh areas are treated as non-defect-prone areas;

[0049] The defect recognition module is used to identify defect-prone areas using a single-category convolutional neural network model. The training samples of this model are mainly based on historical defect data, including cracks, pitting and edge damage on the gear surface. The recognition process is binary classification. The model outputs whether there are surface defects and gives the specific defect type when there are defects. For non-defect-prone areas, a multi-category Softmax classifier is used to detect different surface defect categories. The model is optimized through a weighted loss function to ensure the recognition accuracy of low-frequency defect categories.

[0050] The report generation module is used to generate a test report based on the defect identification results. The report includes the following contents:

[0051] Defect type and location: For each grid unit, the specific type of defect (such as crack, wear, pitting) and its location are marked, with an accuracy range of ±0.1mm;

[0052] Defect area and depth estimation: For defects such as cracks and wear, the area and depth of the defect are calculated by combining grayscale gradient and geometric features;

[0053] Maintenance suggestions: Automatically generate maintenance suggestions based on the severity of the defect. If the crack depth exceeds 0.5mm or the wear area exceeds 1mm 2 , it is recommended to replace the gear immediately; for minor defects, it is recommended to perform surface repair during the next maintenance cycle.

[0054] The present invention also provides a gearbox gear defect detection device, comprising a memory, a processor, and a gearbox gear defect detection program stored in the memory and executable on the processor, wherein the gearbox gear defect detection program implements the gearbox gear defect detection method when executed by the processor.

[0055] The present invention also provides a computer program product, including a gearbox gear defect detection program, which implements the gearbox gear defect detection method when executed by a processor.

[0056] The beneficial effects of the present invention are as follows: compared with the prior art which relies on manual inspection, which is slow and lacks precision, especially under high-speed production lines or under conditions of uneven lighting, traditional inspection technology has difficulty in effectively identifying small defects. This application combines automated inspection based on deep learning to achieve real-time and high-precision gear surface defect detection, thereby avoiding the problems of missed detection and false detection in manual inspection, significantly improving inspection efficiency and accuracy, and being suitable for modern automated production lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0058] Figure 1 It is a schematic flow chart of a first embodiment of a gearbox gear defect detection method of the present invention.

[0059] Figure 2 It is a schematic diagram of equipment for a method for detecting gearbox gear defects of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Embodiment 1: Figure 1 , which is a flow chart of the first embodiment of the gearbox gear defect detection method of the present invention, and provides the first embodiment of the gearbox gear defect detection method of the present invention.

[0062] In a first embodiment, the gearbox gear defect detection method comprises:

[0063] Step S10: A 120-frame / second CMOS industrial camera is used to acquire a full-surface image of the gear in real time. The industrial camera is fixed by a mechanical arm above the production line, maintaining a 90-degree vertical angle with the gear, and an LED adjustable light source system is used to ensure uniform illumination;

[0064] It should be noted that the frame rate of the CMOS industrial camera is 120 frames per second, which is designed to ensure that clear images can be captured even when the gears rotate at high speed. This high frame rate can significantly reduce the motion blur problem caused by rotation, ensure the clarity of each frame of the image, and thus improve the accuracy of subsequent image processing and defect detection. The fixed position and angle design of the robotic arm ensures that the camera always remains perpendicular to the gear surface, which helps to reduce image distortion caused by changes in viewing angle.

[0065] It is understandable that the use of LED adjustable light source system can dynamically adjust the light intensity to adapt to different ambient light conditions. This design is particularly suitable for unstable lighting environment on the production line. By adjusting the light brightness, it ensures that every part of the gear surface is evenly illuminated, avoiding image contrast problems caused by insufficient or excessive light.

[0066] For example, on a gearbox gear production line, the gears rotate at a speed of 1,000 revolutions per minute. Under such high-speed rotation, if a camera with a frame rate lower than 120 frames per second is used for image acquisition, motion blur may occur, resulting in the inability of the subsequent detection system to accurately identify tiny defects on the gear surface. However, by using the CMOS camera and adjustable light source system of the present invention, clear, distortion-free images can be obtained even under such high-speed rotation conditions, thereby ensuring the accuracy of defect detection.

[0067] Step S20: preprocessing the collected gear surface image, including:

[0068] Median filter: Use a 3x3 median filter to remove noise interference in the image;

[0069] Histogram equalization: Perform histogram equalization on images with uneven grayscale value distribution;

[0070] Edge enhancement processing: Use the Canny edge detection algorithm to enhance the edge contour of the gear;

[0071] The whole process of image preprocessing is completed within 1ms to ensure that the speed of real-time detection is not affected;

[0072] It should be noted that each step of the preprocessing has been optimized to ensure that it can be completed in a very short time to meet the real-time requirements of the industrial production line. The 3x3 median filter can quickly filter out noise while maintaining image details, and the histogram equalization ensures the uniformity of the overall brightness distribution of the image, making subsequent detection more accurate. Edge enhancement processing is an important part of image preprocessing. The Canny algorithm can effectively identify the edge structure of the gear surface and avoid the problem of edge loss caused by blur or uneven lighting. The multi-level detection mechanism of the Canny algorithm not only ensures the accuracy of edge detection, but also reduces the appearance of pseudo edges.

[0073] It should be understood that the 1ms pre-processing time is to ensure that the system can detect in real time at the speed of the gear production line. Through hardware acceleration or parallel processing technology, each step of image processing operation can be completed in a very short time, so as not to delay the overall production process.

[0074] Step S30: Perform multi-layer feature extraction on the preprocessed gear surface image based on the convolutional neural network CNN. The convolutional neural network CNN consists of 3 convolutional layers and 2 fully connected layers. The size of the convolution kernel is set to 3x3 and the step size is set to 1. After each convolution layer, the maximum pooling operation is used, and the pooling window size is 2x2. The extracted features include:

[0075] Texture features: used to analyze the texture irregularities on the gear surface, especially to detect abnormal textures caused by defects in the manufacturing process;

[0076] Grayscale gradient feature: used to capture subtle depressions or protrusions on the gear surface based on grayscale gradient changes to identify tiny defects;

[0077] Geometric features: used to extract the tooth edge and contact surface structure of the gear, focusing on identifying wear and crack geometric anomalies;

[0078] It should be noted that the convolution layer in the convolutional neural network is mainly used to extract local features of the image. The 3x3 convolution kernel can capture small-scale local changes in the image, such as the subtle texture and unevenness of the gear surface. The step size is set to 1 to ensure that the convolution operation does not skip important image details. The maximum pooling operation is used after each convolution layer to reduce the size of the feature map through a 2x2 pooling window while retaining significant features.

[0079] It is understandable that the convolution layer reduces the number of parameters through local connections, and the maximum pooling operation can effectively reduce the computational complexity and avoid the waste of computing resources caused by the feature map being too large. The extracted texture features, grayscale gradient features, and geometric features cover different dimensions of the image, making this method not only suitable for detecting conventional surface defects, but also able to identify more complex defect types.

[0080] It should be understood that the multi-layer structure of CNN can extract higher and higher level features through convolution. For example, the lower convolution layers focus more on the edges and simple textures of the image, while the higher convolution layers can capture the overall geometry and complex patterns of the gears. The fully connected layers aggregate the features extracted by the previous layers to generate the final classification result or prediction.

[0081] For example, in a batch of gears produced, some gears have slight wear or cracks on their surfaces, which are easily overlooked in early manual inspections. However, using the convolutional neural network in the present invention, by analyzing the texture features and grayscale gradient features of the surface, even tiny bumps or cracks can be clearly identified. Especially in the production process, wear often produces irregular texture changes, which can be effectively detected through CNN's multi-layer feature extraction, thereby achieving high-precision automated defect detection.

[0082] Step S40: mesh the gear surface according to a predetermined mesh size, dynamically adjust the size of each mesh based on the extracted geometric features and surface roughness, and calculate the defect probability of each mesh using a probability model based on Bayesian discrimination. When the defect probability of a mesh exceeds 0.75, the mesh area is marked as a defect-prone area, and the remaining mesh areas are treated as non-defect-prone areas.

[0083] It should be noted that the size of the grid segmentation is not a fixed value, but is dynamically adjusted according to the specific geometric features and roughness of the gear surface. Larger grid units may be used in areas with relatively flat gear surfaces, while smaller grid units may be used in areas with complex edge details and high surface roughness to ensure the accuracy of detection. Through this dynamic adjustment, more detailed detection can be guaranteed in important high-risk areas, while the consumption of computing resources can be reduced in relatively flat areas. The probability model of Bayesian discrimination can estimate whether there are defects in the grid area based on empirical data by calculating the distribution of features within each grid. The use of Bayesian models helps to accurately predict the probability of defect occurrence in each area based on previous defect data in complex production environments. This probability-based method can effectively deal with uncertainty, especially when some surface features are vague or unclear, and can still provide relatively reliable defect detection results.

[0084] It should be understood that the defect probability threshold of 0.75 is a reasonable value based on a large amount of experimental data, which is used to balance the accuracy and error rate of detection. If the defect probability of a grid exceeds this threshold, it is marked as a defect-prone area and requires more sophisticated follow-up inspection. If the probability is lower than this value, the area is considered to be a non-defect-prone area, which can reduce resource investment and thus improve the efficiency of the entire detection system.

[0085] Step S50: Use a single-category convolutional neural network model to identify defect-prone areas. The training samples of the model are mainly based on historical defect data, including cracks, pitting and edge damage on the gear surface. The recognition process is binary classification. The model outputs whether there are surface defects and gives the specific defect type when there are defects. For non-defect-prone areas, a multi-category Softmax classifier is used to detect different surface defect categories. The model is optimized through a weighted loss function to ensure the recognition accuracy of low-frequency defect categories.

[0086] It should be noted that the single-class convolutional neural network (CNN) model focuses on detecting high-risk areas where defects are most likely to occur, such as the top and root of the gear. These areas have abundant historical defect data, and the model is mainly trained using sample data of these high-frequency defects, so that the model can accurately determine whether there are defects such as cracks and pitting in specific areas. Since the model only needs to determine whether these defects exist, the use of the binary classification method simplifies the calculation and greatly improves the recognition speed.

[0087] It is understandable that in non-defect prone areas, due to the presence of more types of defects, such as slight wear and surface scratches, the use of a multi-class Softmax classifier can finely distinguish these different types of defects. Through the multi-class output of the Softmax function, the system can calculate each possible defect category and its corresponding probability, ensuring that even in areas with low defect probabilities, multiple different types of surface defects can be accurately identified.

[0088] It should be understood that in order to ensure the accuracy of the recognition of low-frequency defects (such as minor scratches, tiny cracks, etc.), the model uses a weighted loss function for optimization. The weighted loss function gives higher weights to those defect categories that are relatively rare in the training data, so that the model will not ignore these low-frequency defect categories during classification. This optimization method can significantly improve the recognition effect of rare defects and avoid the risk of the model being overly biased towards common defects and ignoring other important defects.

[0089] Step S60: Generate a test report based on the defect identification results, the report includes the following contents:

[0090] Defect type and location: For each grid unit, the specific type of defect (such as crack, wear, pitting) and its location are marked, with an accuracy range of ±0.1mm;

[0091] Defect area and depth estimation: For defects such as cracks and wear, the area and depth of the defect are calculated by combining grayscale gradient and geometric features;

[0092] Maintenance recommendations: Maintenance recommendations are automatically generated based on the severity of the defect. If the crack depth exceeds 0.5mm or the wear area exceeds 1mm2, it is recommended to replace the gear immediately; for minor defects, it is recommended to perform surface repairs during the next maintenance cycle.

[0093] It should be understood that the generation process of the inspection report is automated, and the system generates maintenance recommendations based on the identified defect type and location, combined with historical data and preset maintenance rules. This process reduces the time cost of manual analysis and ensures that each decision is supported by data, ensuring the continuity of production and the reliability of the equipment. For example, during a certain inspection process, the system identified a 0.6mm deep crack at the root of the gear tooth. The inspection report immediately indicated the specific location and depth of the crack and recommended that the gear be replaced immediately. At the same time, the system also identified slight wear on the gear surface with an area of ​​0.8mm2. The inspection report recommended surface repair during the next maintenance cycle. With the automatically generated report, operators can immediately take action based on the system's recommendations, thereby avoiding potential equipment failures and unnecessary downtime.

[0094] In addition, the present invention also provides a gearbox gear defect detection system, the gearbox gear defect detection system comprising:

[0095] Image acquisition module, used to acquire the full surface image of the gear in real time using a 120-frame / second CMOS industrial camera during the production process of automobile gearbox gears. The industrial camera is fixed by a mechanical arm above the production line, maintaining a 90-degree vertical angle with the gear, and an LED adjustable light source system is used to ensure uniform illumination;

[0096] The image preprocessing module is used to preprocess the collected gear surface images, including:

[0097] Median filter: Use a 3x3 median filter to remove noise interference in the image;

[0098] Histogram equalization: Perform histogram equalization on images with uneven grayscale value distribution;

[0099] Edge enhancement processing: Use the Canny edge detection algorithm to enhance the edge contour of the gear;

[0100] The whole process of image preprocessing is completed within 1ms to ensure that the speed of real-time detection is not affected;

[0101] The feature extraction module is used to perform multi-layer feature extraction on the preprocessed gear surface image based on the convolutional neural network CNN. The convolutional neural network CNN consists of 3 convolutional layers and 2 fully connected layers. The size of the convolution kernel is set to 3x3 and the step size is set to 1. After each convolution layer, the maximum pooling operation is used, and the pooling window size is 2x2. The extracted features include:

[0102] Texture features: used to analyze the texture irregularities on the gear surface, especially to detect abnormal textures caused by defects in the manufacturing process;

[0103] Grayscale gradient feature: used to capture subtle depressions or protrusions on the gear surface based on grayscale gradient changes to identify tiny defects;

[0104] Geometric features: used to extract the tooth edge and contact surface structure of the gear, focusing on identifying wear and crack geometric anomalies;

[0105] The meshing module is used to mesh the gear surface according to the predetermined mesh size. The size of each mesh is dynamically adjusted based on the extracted geometric features and surface roughness. The defect probability of each mesh is calculated using a probability model based on Bayesian discrimination. When the defect probability of a mesh exceeds 0.75, the mesh area is marked as a defect-prone area, and the remaining mesh areas are treated as non-defect-prone areas.

[0106] The defect recognition module is used to identify defect-prone areas using a single-category convolutional neural network model. The training samples of this model are mainly based on historical defect data, including cracks, pitting and edge damage on the gear surface. The recognition process is binary classification. The model outputs whether there are surface defects and gives the specific defect type when there are defects. For non-defect-prone areas, a multi-category Softmax classifier is used to detect different surface defect categories. The model is optimized through a weighted loss function to ensure the recognition accuracy of low-frequency defect categories.

[0107] The report generation module is used to generate a test report based on the defect identification results. The report includes the following contents:

[0108] Defect type and location: For each grid unit, the specific type of defect (such as crack, wear, pitting) and its location are marked, with an accuracy range of ±0.1mm;

[0109] Defect area and depth estimation: For defects such as cracks and wear, the area and depth of the defect are calculated by combining grayscale gradient and geometric features;

[0110] Maintenance suggestions: Automatically generate maintenance suggestions based on the severity of the defect. If the crack depth exceeds 0.5mm or the wear area exceeds 1mm 2 , it is recommended to replace the gear immediately; for minor defects, it is recommended to perform surface repair during the next maintenance cycle.

[0111] The present invention provides a gearbox gear defect detection system, which adopts a gearbox gear defect detection method in the above embodiment, and can solve a technical problem of gearbox gear defect detection. Compared with the prior art, the beneficial effects of the gearbox gear defect detection system provided by the present invention are the same as the beneficial effects of the gearbox gear defect detection method provided by the above embodiment, and other technical features of the gearbox gear defect detection system are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0112] The present invention provides a gearbox gear defect detection device, please refer to Figure 2, a gearbox gear defect detection device includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a gearbox gear defect detection method in the above-mentioned embodiment 1. A gearbox gear defect detection device in the embodiment of the present invention may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. A gearbox gear defect detection device is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention. A gearbox gear defect detection device may include a processing device 1001 (such as a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM) 1004. In RAM1004, various programs and data required for the operation of a gearbox gear defect detection device are also stored. The processing device 1001, ROM1002 and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow a gearbox gear defect detection device to communicate wirelessly or wired with other devices to exchange data. Although a gearbox gear defect detection device with various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0113] The present invention also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the steps of the above-mentioned method for detecting a gearbox gear defect are implemented. The computer program product provided by the present invention can solve a technical problem of detecting a gearbox gear defect. Compared with the prior art, the beneficial effects of the computer program product provided by the present invention are the same as the beneficial effects of the method for detecting a gearbox gear defect provided by the above-mentioned embodiment, and will not be described in detail here.

[0114] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present invention are executed.

[0115] It should be understood that the various parts disclosed in the present invention can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0116] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A gearbox gear defect detection method, characterized in that: The gearbox gear defect detection method comprises: Step S10: A 120-frame / second CMOS industrial camera is used to acquire a full-surface image of the gear in real time. The industrial camera is fixed by a mechanical arm above the production line, maintaining a 90-degree vertical angle with the gear, and an LED adjustable light source system is used to ensure uniform illumination; Step S20: preprocessing the collected gear surface image, including: Median filter: Use a 3x3 median filter to remove noise interference in the image; Histogram equalization: Perform histogram equalization on images with uneven grayscale value distribution; Edge enhancement processing: Use the Canny edge detection algorithm to enhance the edge contour of the gear; The whole process of image preprocessing is completed within 1ms; Step S30: Perform multi-layer feature extraction on the preprocessed gear surface image based on the convolutional neural network CNN. The convolutional neural network CNN consists of 3 convolutional layers and 2 fully connected layers. The size of the convolution kernel is set to 3x3 and the step size is set to 1. After each convolution layer, the maximum pooling operation is used, and the pooling window size is 2x2. The extracted features include: Texture features: used to analyze the texture irregularities on the gear surface; Grayscale gradient feature: used to capture subtle depressions or protrusions on the gear surface based on grayscale gradient changes to identify tiny defects; Geometric features: used to extract the tooth edge and contact surface structure of the gear and identify wear and crack geometric anomalies; Step S40: mesh the gear surface according to a predetermined mesh size, dynamically adjust the size of each mesh based on the extracted geometric features and surface roughness, and calculate the defect probability of each mesh using a probability model based on Bayesian discrimination. When the defect probability of a mesh exceeds 0.75, the mesh area is marked as a defect-prone area, and the remaining mesh areas are treated as non-defect-prone areas. Step S50: Use a single-category convolutional neural network model to identify defect-prone areas. The training samples of the model are mainly based on historical defect data, including cracks, pitting and edge damage on the gear surface. The recognition process is binary classification. The model determines and outputs whether there are surface defects and gives the specific defect type when there are defects. For non-defect-prone areas, a multi-category Softmax classifier is used to detect different surface defect categories. The model is optimized through a weighted loss function to ensure the recognition accuracy of low-frequency defect categories. Step S60: Generate a test report based on the defect identification results, the report includes the following contents: Defect type and location: For each grid unit, the specific type of defect including crack, wear, pitting and its location are marked, with an accuracy range of ±0.1mm; Defect area and depth estimation: For defects such as cracks and wear, the area and depth of the defect are calculated by combining grayscale gradient and geometric features; Maintenance suggestions: Automatically generate maintenance suggestions based on the severity of the defect. If the crack depth exceeds 0.5mm or the wear area exceeds 1mm 2 , it is recommended to replace the gear immediately; for minor defects, it is recommended to perform surface repair during the next maintenance cycle.

2. A gearbox gear defect detection method according to claim 1, characterized in that: In step S40, the step of dynamically adjusting the size of each grid based on the extracted geometric features and surface roughness adopts the formula: Among them, G size Indicates the size of the grid, D max is the maximum diameter of the gear, σ e is the standard deviation of surface roughness, and α is the adjustment coefficient.

3. A gearbox gear defect detection method as claimed in claim 1, characterized in that: In step S40, the step of calculating the defect probability of each grid using the probability model based on Bayesian discrimination adopts the formula: Among them, P(defect|features) is the posterior probability, which indicates the probability of defects occurring in the area after the preset features are observed; P(features|defect) indicates the possibility of feature occurrence under the premise that defects are known to exist, which is the frequency of occurrence of certain features under defective conditions; P(defect) is the prior probability, which indicates the probability of defects occurring in a certain area when the preset features are not observed; P(features) is the marginal probability of the feature, which indicates the probability of the preset features occurring in all cases.

4. A gearbox gear defect detection method as claimed in claim 1, characterized in that: In step S60, the step of calculating the area and depth of the defect adopts the formula: Wherein, I is the gray value distribution of the defect area, and x and y are the horizontal and vertical axes in the image.

5. A gearbox gear defect detection method as claimed in claim 1, characterized in that: In step S10, the image acquisition process is performed at a rotation speed of the gear of 500 to 1000 revolutions per minute.

6. A gearbox gear defect detection method according to claim 1, characterized in that: In step S10, the adjustable light source system automatically adjusts the light intensity according to the reflection condition of the gear surface.

7. A gearbox gear defect detection method as claimed in claim 1, characterized in that: In step S40, the predetermined grid size is that the side length of each grid unit is 0.5 mm.

8. A gearbox gear defect detection system, characterized in that: The gearbox gear defect detection system comprises: Image acquisition module, used to acquire the full surface image of the gear in real time using a 120-frame / second CMOS industrial camera during the production process of automobile gearbox gears. The industrial camera is fixed by a mechanical arm above the production line, maintaining a 90-degree vertical angle with the gear, and an LED adjustable light source system is used to ensure uniform illumination; The image preprocessing module is used to preprocess the collected gear surface images, including: Median filter: Use a 3x3 median filter to remove noise interference in the image; Histogram equalization: Perform histogram equalization on images with uneven grayscale value distribution; Edge enhancement processing: Use the Canny edge detection algorithm to enhance the edge contour of the gear; The whole process of image preprocessing is completed within 1ms to ensure that the speed of real-time detection is not affected; The feature extraction module is used to perform multi-layer feature extraction on the preprocessed gear surface image based on the convolutional neural network CNN. The convolutional neural network CNN consists of 3 convolutional layers and 2 fully connected layers. The size of the convolution kernel is set to 3x3 and the step size is set to 1. After each convolution layer, the maximum pooling operation is used, and the pooling window size is 2x2. The extracted features include: Texture features: used to analyze the texture irregularities on the gear surface, especially to detect abnormal textures caused by defects in the manufacturing process; Grayscale gradient feature: used to capture subtle depressions or protrusions on the gear surface based on grayscale gradient changes to identify tiny defects; Geometric features: used to extract the tooth edge and contact surface structure of the gear, focusing on identifying wear and crack geometric anomalies; The meshing module is used to mesh the gear surface according to the predetermined mesh size. The size of each mesh is dynamically adjusted based on the extracted geometric features and surface roughness. The defect probability of each mesh is calculated using a probability model based on Bayesian discrimination. When the defect probability of a mesh exceeds 0.75, the mesh area is marked as a defect-prone area, and the remaining mesh areas are treated as non-defect-prone areas. The defect recognition module is used to identify defect-prone areas using a single-category convolutional neural network model. The training samples of this model are mainly based on historical defect data, including cracks, pitting and edge damage on the gear surface. The recognition process is binary classification. The model outputs whether there are surface defects and gives the specific defect type when there are defects. For non-defect-prone areas, a multi-category Softmax classifier is used to detect different surface defect categories. The model is optimized through a weighted loss function to ensure the recognition accuracy of low-frequency defect categories. The report generation module is used to generate a test report based on the defect identification results. The report includes the following contents: Defect type and location: For each grid unit, the specific type of defect (such as crack, wear, pitting) and its location are marked, with an accuracy range of ±0.1mm; Defect area and depth estimation: For defects such as cracks and wear, the area and depth of the defect are calculated by combining grayscale gradient and geometric features; Maintenance recommendations: Maintenance recommendations are automatically generated based on the severity of the defect. If the crack depth exceeds 0.5mm or the wear area exceeds 1mm2, it is recommended to replace the gear immediately; for minor defects, it is recommended to perform surface repairs during the next maintenance cycle.

9. A gearbox gear defect detection device, characterized in that: The gearbox gear defect detection device comprises: A memory, a processor, and a gearbox gear defect detection program stored in the memory and executable on the processor, wherein the gearbox gear defect detection program, when executed by the processor, implements the gearbox gear defect detection method described in any one of claims 1 to 7.

10. A computer program product, characterized in that The computer program product includes a gearbox gear defect detection program, which, when executed by a processor, implements the gearbox gear defect detection method described in any one of claims 1 to 7.

Images