Method and device for detecting a toothed wheel defect in a compressor stator inner cavity

Abstract

The present application relates to the field of industrial vision technology, and discloses a compressor stator inner cavity toothing defect detection method and device, which improves the efficiency and accuracy of the compressor stator inner cavity toothing defect detection. The method comprises the following steps: S1. An imaging device is used to take pictures at a preset angle to collect stator inner cavity images, and the stator is controlled to rotate at a preset step length to ensure that the collected images cover all areas of the stator inner cavity; S2. The collected stator inner cavity images are preprocessed; S3. The preprocessed stator inner cavity images are subjected to threshold segmentation to screen out target detection areas; S4. For the screened target detection areas, a target detection model based on deep learning is used to identify toothing defects; and S5. For the area where the toothing defects identified in step S4 are located, a traditional image processing algorithm is used to perform secondary verification on the identified toothing defects based on the brightness and texture features of the area to obtain the final detection result.

Description

Technical Field

[0001] This invention relates to the field of industrial vision technology, specifically to a method and device for detecting tooth rolling defects in the inner cavity of a compressor stator. Background Technology

[0002] As a core power component in many industrial and civil equipment such as refrigeration and air conditioning systems, the integrity of the compressor stator's internal structure directly determines the compressor's operating efficiency, service life, and safety stability. Internal tooth rolling defects are one of the most common and serious quality problems encountered during stator production. These defects are often caused by insufficient stamping precision, mechanical friction during assembly, and uneven material stress release. They result in abnormal clearance between the stator and rotor, increased operating resistance, and consequently, increased equipment vibration and energy consumption. In severe cases, they can even lead to stator steel sheet breakage, causing overall compressor failure. Therefore, detecting stator internal tooth rolling defects is a crucial aspect of the compressor production quality control system.

[0003] Currently, in industrial production, the detection of stator internal cavity tooth rolling defects mainly relies on manual visual assessment. Inspectors observe the surface of the stator internal cavity using magnifying glasses or simple imaging equipment, judging the presence and severity of defects based on personal experience. However, this traditional inspection method has several drawbacks: First, it is inefficient. The stator internal cavity structure is complex, and the field of view is limited. Manual inspection of each piece is time-consuming and cannot meet the high-speed mass production requirements of modern production lines. Second, inspection standards are inconsistent. Different inspectors have varying experience and visual sensitivity, leading to inconsistent defect judgment thresholds and significant fluctuations in product quality, making it difficult to establish a standardized quality control system. Third, it is susceptible to subjective factors. Prolonged inspection can cause visual fatigue among inspectors, resulting in increased rates of missed and false detections, and compromising the stability and reliability of inspection results. Fourth, labor costs are high. Large-scale production requires a large number of inspectors, and the long training period significantly increases the company's production costs.

[0004] To address the aforementioned issues with manual inspection, some companies have attempted to automate the process using image processing algorithms, extracting defect features through edge detection and threshold segmentation. However, due to the weak reflectivity and low image contrast of the black steel sheet area within the stator cavity, and its susceptibility to interference from impurities such as plastic scraps and oil stains in the production environment, traditional algorithms exhibit poor adaptability to complex scenarios and insufficient precision in defect feature extraction, resulting in low robustness and difficulty in controlling the false negative rate. Other solutions employ direct deep learning algorithms for defect detection. While this improves feature extraction capabilities, the lack of targeted optimization for specific scenarios within the stator cavity leads to insufficient ability to distinguish between defects and impurities such as plastic scraps, resulting in a high false positive rate. Furthermore, their performance is easily affected in low-contrast image scenarios, making it difficult to meet the stringent precision requirements of industrial production. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and device for detecting tooth rolling defects in the inner cavity of a compressor stator, thereby improving the efficiency and accuracy of detecting tooth rolling defects in the inner cavity of a compressor stator.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] On one hand, the present invention provides a method for detecting tooth rolling defects in the inner cavity of a compressor stator, comprising the following steps:

[0008] S1. Use an imaging device to capture images of the stator cavity from a preset angle, while controlling the stator to rotate at a preset step size to ensure that the captured images cover the entire area of ​​the stator cavity;

[0009] S2. Preprocess the acquired images of the stator cavity;

[0010] S3. Perform threshold segmentation on the preprocessed stator cavity image to filter out the target detection area;

[0011] S4. For the selected target detection areas, a deep learning-based target detection model is used to identify tooth rolling defects;

[0012] S5. For the area where the tooth-rubbing defect is located, the traditional image processing algorithm is used to perform secondary verification of the identified tooth-rubbing defect based on the brightness and texture features of the area, and the final detection result is obtained.

[0013] Furthermore, in step S1, the imaging angle of the imaging device is 45°, and the rotation step of the stator is 60°.

[0014] Furthermore, in step S2, the preprocessing includes brightness and contrast enhancement processing.

[0015] Furthermore, in step S3, the threshold segmentation of the preprocessed stator cavity image to filter out the target detection region specifically includes:

[0016] First, Gaussian filtering is applied to the preprocessed image. Then, half of the average gray level of a single image is used as a threshold to segment the image. The segmented image is then processed in reverse to obtain a binary image. After dilation and erosion processing is applied to the binary image, the black steel sheet region is determined by the area of ​​the connected components and used as the target detection region.

[0017] Furthermore, in step S4, the target detection model adopts a trained YOLO algorithm. The training method includes: collecting images of the inner cavity of the compressor stator and using the Labelimg annotation tool to annotate the tooth rolling defects in the black steel sheet area of ​​the stator inner cavity image to construct a training dataset, and using the training dataset to train the YOLO algorithm.

[0018] Furthermore, in step S4, the output results of the target detection model for identifying tooth rolling defects include: the location coordinates of the tooth rolling defects and the confidence level of the detection results.

[0019] Furthermore, in step S5, the use of traditional image processing algorithms to perform secondary verification of the identified jagged edges based on the brightness and texture features of the region specifically includes:

[0020] For the region where the tooth-rubbing defect is located, identified in step S4, the brightness value and texture feature complexity of the region are calculated. The confidence level, brightness value, and texture complexity of the detection results are compared with set thresholds to determine whether the tooth-rubbing defect identified in step S4 is a true tooth-rubbing defect.

[0021] ;

[0022] in, Represented as plastic scraps, This indicates a true tooth rolling defect. Indicates the confidence level of the test results. This represents the average brightness value of the image. This represents the texture complexity value of the image.

[0023] Furthermore, the texture complexity value of the image is obtained by calculating the gray-level co-occurrence matrix:

[0024] The pixel spacing and orientation in the image are determined. By statistically analyzing the occurrence frequency of two-pixel grayscale value combinations under specified spacing and orientation, a matrix is ​​formed and normalized to a probability matrix. Then, based on the probability matrix, the sum of squares of all elements is calculated to obtain the energy value of the grayscale co-occurrence matrix, which is used as the texture complexity value of the image.

[0025] Alternatively, the texture complexity value of the image can be evaluated using image entropy:

[0026] ;

[0027] in, Represents image entropy, Represents pixels The probability density of grayscale values ​​among all pixels in this region. For the first image region line, number The number of pixels in a column.

[0028] On the other hand, the present invention also provides a device for detecting defects in the tooth rolling of a compressor stator cavity, comprising:

[0029] The image acquisition module is used to capture images of the stator cavity from a preset angle.

[0030] The image processing module is used to preprocess the acquired stator cavity image and perform threshold segmentation on the preprocessed stator cavity image to filter out the target detection area.

[0031] The defect detection module is used to identify tooth rolling defects in the selected target detection areas using a deep learning-based target detection model.

[0032] The result verification module is used to perform secondary verification of the identified tooth-rubbing defects in the area using traditional image processing algorithms based on the brightness and texture features of the area, and obtain the final detection result.

[0033] The beneficial effects of this invention are:

[0034] (1) Improve detection accuracy and stability:

[0035] This invention utilizes a fusion technology of deep learning and traditional image processing. First, it uses a deep learning target detection algorithm to accurately identify tooth-rubbing defects. Then, it uses a traditional algorithm to calculate brightness and texture features to perform secondary filtering on the detection results. This effectively distinguishes tooth-rubbing defects from plastic debris, significantly improves detection accuracy, reduces false detection rate, ensures stable and reliable detection results, and avoids quality fluctuations caused by subjective human factors and the limitations of a single algorithm.

[0036] (2) Improve detection efficiency and automation level:

[0037] This invention employs a high-resolution industrial camera to capture images from a preset angle while simultaneously controlling the stator to rotate in steps. This image acquisition method enables 360-degree full coverage of the stator's inner cavity, achieving efficient image acquisition without manual intervention. Furthermore, the entire testing process can be integrated into automated equipment, directly adapting to the production line to achieve a fully automated closed loop, significantly improving testing efficiency and saving labor costs. Attached Figure Description

[0038] Figure 1 This is a flowchart of the compressor stator inner cavity tooth rolling defect detection method in an embodiment of the present invention.

[0039] Figure 2 This is a structural diagram of the compressor stator inner cavity tooth rolling defect detection device in an embodiment of the present invention. Detailed Implementation

[0040] This invention aims to provide a method and apparatus for detecting tooth rolling defects in the internal cavity of a compressor stator, improving the efficiency and accuracy of such detection. The core idea is to utilize a high-resolution industrial camera to capture images from a preset angle, while simultaneously controlling the stator to rotate in step increments. This approach solves the problem of limited field of view within the cavity, achieving 360-degree full-area image coverage without blind spots. High-resolution imaging also ensures detailed rendering of the low-contrast black steel sheet area, providing high-quality data support for subsequent defect detection and avoiding detection deviations caused by incomplete information or poor image quality. For the acquired images, image enhancement is first used to improve the low contrast of the black steel sheet. Then, threshold segmentation, dilation erosion, and connected component filtering are used to accurately extract the tooth rolling defect detection area. Subsequently, the strong feature extraction capability of deep learning object detection algorithms is used to initially identify the defect. Finally, traditional image processing algorithms are used to calculate brightness values ​​and texture features for secondary verification of the detection results, accurately distinguishing tooth rolling defects from impurities such as plastic debris, thus reducing the false detection rate while ensuring detection accuracy.

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0042] Example:

[0043] This embodiment first provides a method for detecting tooth rolling defects in the internal cavity of a compressor stator. See [link to relevant documentation]. Figure 1 The implementation process includes the following steps:

[0044] S1. Acquire images of the stator cavity:

[0045] In this step, an imaging device is used to capture images of the stator cavity from a preset angle, while the stator is rotated in preset steps to ensure that the captured images cover the entire area of ​​the stator cavity.

[0046] In one exemplary implementation, a high-resolution industrial camera is mounted above the compressor stator and images of the stator cavity are captured by shooting from a 45° angle. The distance needs to be adjusted according to the camera's field of view and the size of the stator. In this example, a 40-megapixel industrial camera is used to cover one-third of the cavity area at a time, and the stator is rotated in 60° increments each time, so that six images can be captured continuously to ensure full coverage of the stator cavity.

[0047] S2. Preprocess the acquired images of the stator cavity;

[0048] In this step, the brightness and contrast of the image of the compressor stator cavity are enhanced to obtain an enhanced image. In an exemplary implementation, based on the characteristic that the black steel sheet in the stator cavity image has poor reflectivity, the brightness and contrast enhancement algorithm is implemented using a direct multiplication method. First, the image is converted into floating-point data, and then multiplication is performed. Data greater than 255 is assigned a value of 255. The specific formula is as follows:

[0049] ;

[0050] ;

[0051] Where I is the original image, and k is the multiplier. This is the enhanced image.

[0052] S3. Filter target detection areas:

[0053] In this step, threshold segmentation is performed on the preprocessed stator cavity image to filter out the target detection area.

[0054] In one exemplary implementation, the preprocessed image is first subjected to Gaussian filtering to remove high-frequency noise. Then, threshold segmentation is performed, since the black steel sheet region is a dark area. Next, the segmented image is reverse-processed to obtain a binary image where the dark steel sheet region is 1 and other regions are 0. Then, dilation and erosion are performed on the binary image to remove interference points. Finally, the location of the black steel sheet region, i.e., the tooth-knotting defect detection area image, is obtained by thresholding the connected component area, and this image serves as the target detection area image.

[0055] S4. Deep learning to identify tooth-rolling defects:

[0056] In this step, a deep learning-based target detection model is used to identify tooth-rolling defects within the selected target detection regions. In an exemplary implementation, the target detection model can employ the YOLOv8 target detection algorithm, and its training process is as follows:

[0057] Dataset construction: Collect multiple images of black steel sheets containing different degrees of tooth rolling defects (covering three scenarios: slight tooth rolling, severe tooth rolling, and no defects). Use the Labelimg annotation tool to annotate the defect areas, including the defect category tooth rolling and the coordinates of the defect location.

[0058] Dataset partitioning: The dataset processed above is divided into training set and validation set according to a certain ratio.

[0059] Model training: Set training parameters, including batch size, learning rate, number of iterations and optimizer, train the model on the training set, and verify the model's performance using the validation set. When the model performance meets the target or the parameters converge, the trained model is obtained.

[0060] Using this trained model, the target detection area can be detected and identified, and the location coordinates of the tooth rolling defect and the confidence level of the detection result can be output.

[0061] S5. Secondary verification using traditional image processing algorithms:

[0062] In this step, for the area where the tooth-rubbing defect is located, the traditional image processing algorithm is used to perform secondary verification of the identified tooth-rubbing defect based on the brightness and texture features of the area, and the final detection result is obtained.

[0063] In one exemplary implementation, to further reduce false detections caused by plastic debris, a traditional image processing algorithm is used to perform secondary verification on the deep learning target detection results. Specifically, this is implemented as follows:

[0064] First, the area for detecting tooth rubbing is segmented to form a sub-image. Brightness and texture calculations are then performed on this sub-image. Since the plastic debris exhibits high brightness after k-fold image processing in step S2, it is used as one of the filtering factors. Simultaneously, the plastic debris is blocky with simple texture features; therefore, a gray-level co-occurrence matrix (GLCM) is used to describe the texture features as one of the classification conditions. Specifically, detection results with a confidence level (conf) greater than 0.1 are considered candidate targets for tooth rubbing defects. The average brightness of the region where each candidate target is located is calculated, and the GLCM is also calculated. The energy in the GLCM is taken as the texture description feature value. The verification condition judgment logic is as follows:

[0065] ;

[0066] in, Represented as plastic scraps, This indicates a true tooth rolling defect. Indicates the confidence level of the test results. This represents the average brightness value of the image. This represents the texture complexity value of the image.

[0067] In addition, for the sake of simplifying calculations, image entropy can also be used to evaluate the texture complexity of an image:

[0068] ;

[0069] in, Represents image entropy, Represents pixels The probability density of grayscale values ​​among all pixels in this region. For the first image region line, number The number of pixels in the column. The calculated image entropy value is used as the texture complexity value.

[0070] This embodiment also provides a device for detecting tooth rolling defects in the inner cavity of a compressor stator. See [link to relevant documentation]. Figure 2 It includes an image acquisition module, an image processing module, a defect detection module, and a result verification module; the functions of each module are as follows:

[0071] The image acquisition module is used to capture images of the stator cavity from a preset angle.

[0072] The image processing module is used to preprocess the acquired stator cavity image and perform threshold segmentation on the preprocessed stator cavity image to filter out the target detection area.

[0073] The defect detection module is used to identify tooth rolling defects in the selected target detection areas using a deep learning-based target detection model.

[0074] The result verification module is used to perform secondary verification of the identified tooth-rubbing defects in the area using traditional image processing algorithms based on the brightness and texture features of the area, and obtain the final detection result.

[0075] Since the functions of each module in the detection device are consistent with the steps in the aforementioned detection method, and the specific implementation of each step in the detection method has been described in detail, the specific implementation of each module in the detection device will not be repeated here.

[0076] Although embodiments of the present invention have been described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present invention, and all such changes and alterations shall not depart from the protection scope of the present invention.

Claims

1. A method for detecting tooth rolling defects in the inner cavity of a compressor stator, characterized in that, Includes the following steps: S1. Use an imaging device to capture images of the stator cavity from a preset angle, while controlling the stator to rotate at a preset step size to ensure that the captured images cover the entire area of ​​the stator cavity; S2. Preprocess the acquired images of the stator cavity; S3. Perform threshold segmentation on the preprocessed stator cavity image to filter out the target detection area; S4. For the selected target detection areas, a deep learning-based target detection model is used to identify tooth rolling defects; S5. For the area where the tooth-rubbing defect is located, the traditional image processing algorithm is used to perform secondary verification of the identified tooth-rubbing defect based on the brightness and texture features of the area to obtain the final detection result; The method employs traditional image processing algorithms to perform secondary verification of the identified tooth-rubbing defects based on the brightness and texture features of the region. Specifically, this includes: calculating the brightness value and texture feature complexity of the region where the tooth-rubbing defect is located, and comparing the confidence level, brightness value, and texture complexity of the detection results with set thresholds to determine whether the tooth-rubbing defect identified in step S4 is a genuine tooth-rubbing defect. ； in, Represented as plastic scraps, This indicates a true tooth rolling defect. Indicates the confidence level of the test results. This represents the average brightness value of the image. This represents the texture complexity value of the image.

2. The method for detecting tooth rolling defects in the inner cavity of a compressor stator as described in claim 1, characterized in that, In step S1, the imaging device has a shooting angle of 45° and a stator rotation step of 60°.

3. The method for detecting tooth rolling defects in the inner cavity of a compressor stator as described in claim 1, characterized in that, In step S2, the preprocessing includes brightness and contrast enhancement processing.

4. The method for detecting tooth rolling defects in the inner cavity of a compressor stator as described in claim 1, characterized in that, In step S3, the threshold segmentation of the preprocessed stator cavity image to filter out the target detection region specifically includes: First, Gaussian filtering is applied to the preprocessed image. Then, half of the average gray level of a single image is used as a threshold to segment the image. The segmented image is then processed in reverse to obtain a binary image. After dilation and erosion processing is applied to the binary image, the black steel sheet region is determined by the area of ​​the connected components and used as the target detection region.

5. The method for detecting tooth rolling defects in the inner cavity of a compressor stator as described in claim 1, characterized in that, In step S4, the target detection model uses a trained YOLO algorithm. The training method includes: collecting images of the inner cavity of the compressor stator and using the Labelimg annotation tool to annotate the tooth rolling defects in the black steel sheet area of ​​the stator inner cavity image to construct a training dataset, and using the training dataset to train the YOLO algorithm.

6. The method for detecting tooth rolling defects in the inner cavity of a compressor stator as described in claim 1, characterized in that, In step S4, the output of the target detection model for identifying tooth-rolling defects includes: the location coordinates of the tooth-rolling defects and the confidence level of the detection results.

7. The method for detecting tooth rolling defects in the inner cavity of a compressor stator as described in claim 1, characterized in that, The texture complexity value of the image is obtained by calculating the gray-level co-occurrence matrix: The pixel spacing and orientation in the image are determined. By statistically analyzing the occurrence frequency of two-pixel grayscale value combinations under specified spacing and orientation, a matrix is ​​formed and normalized to a probability matrix. Then, based on the probability matrix, the sum of squares of all elements is calculated to obtain the energy value of the grayscale co-occurrence matrix, which is used as the texture complexity value of the image.

8. The method for detecting tooth rolling defects in the inner cavity of a compressor stator as described in claim 1, characterized in that, The texture complexity of the image is evaluated using image entropy: ； in, Represents image entropy, Represents pixels The probability density of grayscale values ​​among all pixels in this region. For the first image region line, number The number of pixels in a column.

9. A device for detecting defects in the tooth rolling inside the compressor stator cavity, characterized in that, include: The image acquisition module is used to capture images of the stator cavity from a preset angle. During image acquisition, the stator is rotated in preset steps to ensure that the acquired image covers the entire area of ​​the stator cavity. The image processing module is used to preprocess the acquired stator cavity image and perform threshold segmentation on the preprocessed stator cavity image to filter out the target detection area. The defect detection module is used to identify tooth rolling defects in the selected target detection areas using a deep learning-based target detection model. The result verification module is used to perform secondary verification of the identified tooth-rubbing defects in the area where they are located using traditional image processing algorithms based on the brightness and texture features of the area, and to obtain the final detection result. The method employs traditional image processing algorithms to perform secondary verification of the identified jagged edges based on the brightness and texture features of the region. Specifically, this includes: For the region where the tooth-rubbing defect is located, identified in step S4, the brightness value and texture feature complexity of the region are calculated. The confidence level, brightness value, and texture complexity of the detection results are compared with set thresholds to determine whether the tooth-rubbing defect identified in step S4 is a true tooth-rubbing defect. ； in, Represented as plastic scraps, This indicates a true tooth rolling defect. Indicates the confidence level of the test results. This represents the average brightness value of the image. This represents the texture complexity value of the image.

Images