Nondestructive testing method and system for surfaces of main shaft parts

Abstract

The invention relates to the technical field of image processing, in particular to a nondestructive testing method and system for the surface of a spindle part, and the method comprises the steps: obtaining a historical grey-scale map of the surface of a preprocessed part, and marking the historical grey-scale map to obtain a plurality of defect regions and corresponding historical defect types; dividing and accumulating the gradient of each defect area based on the optimal division parameter to obtain a gradient histogram of each defect in the historical grey-scale map; and obtaining a real-time grey-scale map of the surface of the preprocessed part, extracting a defect area of the real-time grey-scale map, obtaining a gradient histogram of each real-time defect in the real-time grey-scale map by using the optimal division parameter, comparing the gradient histograms of the real-time defects with the gradient histograms of the historical defects, and outputting a result which is the defect type of the part. According to the method, the gradient histogram difference between similar defects is minimized, the difference between different types of defects is maximized, and division parameters are optimized so as to improve the accuracy of real-time defect detection.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more specifically, to a method and system for nondestructive testing of the surface of spindle parts. Background Art

[0002] Spindle parts are key rotating parts in machinery, mainly used to transmit torque, bear loads, and ensure the precise operation of the machine. As a vital component of the mechanical system, the surface quality of spindle parts directly affects the normal operation and service life of mechanical equipment. For example, defects on the shaft surface, such as cracks and silver streaks, can cause sudden failure of the equipment during operation and even cause serious safety accidents. Through effective surface defect detection, these potential risks can be discovered and dealt with in a timely manner, thereby ensuring the safe operation of the equipment.

[0003] With the advancement of machine vision and image processing technology, automatic detection has become an important means to improve detection efficiency and accuracy. At present, the existing technology, such as the patent application document with publication number CN113838054A, discloses a method for detecting surface damage of mechanical parts based on artificial intelligence, including: first obtaining an image of the surface of the part; constructing a mixed Gaussian model of the initial damage area and the normal area according to the image data; using the constructed model to calculate the probability that each pixel in the image belongs to the damage area or the normal area; calculating the weights between the pixels in the image; combining the weights, damage probability and normal probability of the pixel points to construct a loss function model; and determining the final target damage area through the loss function model. This method can accurately identify the damaged area on the surface of the part by constructing a mixed Gaussian model to process the image.

[0004] However, there are interferences from irrelevant details in the image, and the damage types are diverse. The use of methods such as mixed Gaussian models cannot well identify the characteristics of the corresponding defects, which reduces the accuracy of defect detection. Summary of the invention

[0005] In order to solve the above-mentioned technical problem of reduced accuracy of defect detection, the present invention provides solutions in the following aspects.

[0006] In a first aspect, a method for nondestructive testing of the surface of a spindle-type part comprises: Obtain a historical grayscale image of the preprocessed part surface, mark the historical grayscale image to obtain multiple defect areas and corresponding historical defect types; divide and accumulate the gradient of each defect area based on the optimal division parameter to obtain a gradient histogram of each defect in the historical grayscale image; Obtain the real-time grayscale image of the preprocessed part surface, extract the defect area of ​​the real-time grayscale image, use the optimal segmentation parameters to obtain the gradient histogram of each real-time defect in the real-time grayscale image, compare the gradient histogram of the real-time defect with the gradient histogram of the historical defect, and output the defect type of the part; The optimal partitioning parameter is positively correlated with a preset objective function, and the objective function is positively correlated with the difference in gradient histograms between historical defects of the same type, and negatively correlated with the difference in gradient histograms between historical defects of different types.

[0007] The present invention constructs an objective function by utilizing historical grayscale images and corresponding defect type information, so that the gradient histogram differences between defects of the same type are minimized and the gradient histogram differences between defects of different types are maximized, and then the partitioning parameters are adjusted to achieve the optimum. Under the application of the optimal partitioning parameters, gradient partitioning and accumulation are performed on each extracted real-time defect area, and the gradient histogram of each real-time defect is calculated, which further ensures that the gradient histogram of the real-time defect can accurately reflect the characteristics of the defect, so that the defect type can be judged more accurately in the subsequent comparison with the historical defects, and the accuracy of defect detection is significantly improved.

[0008] Preferably, the process of obtaining the objective function includes: Presetting an initial gradient direction interval size, and obtaining a gradient histogram of all historical defects under the division of the initial gradient direction interval size; Calculate the sum of the gradient differences between each gradient histogram of historical defects of any defect type and the gradient histogram of historical defects of all other defect types, and record it as the first parameter; Calculate the sum of the gradient differences between each gradient histogram of historical defects of any defect type and other gradient histograms of historical defects of the same defect type, and record it as the second parameter; The sum of the ratios of the first parameter to the second parameter of all historical defects of all defect types is calculated to obtain the value of the objective function.

[0009] By optimizing the objective function, we can not only measure the discrimination of defect features, but also optimize the defect detection model to improve the accuracy and efficiency of defect detection.

[0010] Preferably, when the value of the objective function is maximum, the corresponding initial gradient direction interval size is the optimal division parameter.

[0011] In machine learning or deep learning models, the maximum value of the objective function often represents the best fit of the model to the current data set. By adjusting the size of the initial gradient direction interval and finding the parameter that maximizes the objective function value, the model's recognition accuracy for defects, images, or other data features can be significantly improved.

[0012] Preferably, it also includes: Under the partitioning of the optimal partitioning parameters, the absolute difference between the mean values ​​of the gradient values ​​of any two types of historical defects in the same gradient direction interval is calculated, and the absolute difference between the mean values ​​of the gradient values ​​is normalized to obtain the difference weight of the corresponding gradient direction interval.

[0013] By calculating the difference weights, we can identify which gradient direction intervals are most important for defect detection, so that these intervals can be given priority in feature selection, reducing unnecessary calculations and improving detection efficiency.

[0014] Preferably, the gradient histogram of the real-time defect is compared with the gradient histogram of the historical defect, and the output result is the defect type of the part including: In any gradient direction interval, the absolute value of the difference between the gradient value of the real-time defect and the gradient mean value of the historical defect is calculated and recorded as the third parameter; The product of the third parameter and the difference weight of the corresponding gradient direction interval is used as the difference measure between the real-time defect and the historical defect in the gradient direction interval; If the difference metric is less than a preset threshold, the real-time defect and the historical defect are the same type of defect.

[0015] By combining gradient histogram comparison and difference weights, fast and accurate detection and classification of real-time defects are achieved.

[0016] Preferably, comparing the gradient histogram of the real-time defect with the gradient histogram of the historical defect comprises: The gradient histogram of each real-time defect is matched with the gradient histogram of each historical defect. If the Euclidean distance between the gradient histogram of a historical defect and the gradient histogram of the real-time defect is the smallest, the real-time defect is considered to belong to the defect type corresponding to the historical defect.

[0017] Preferably, comparing the gradient histogram of the real-time defect with the gradient histogram of the historical defect comprises: The peak positions of the gradient histogram of the real-time defect and the gradient histogram of the historical defect are compared. If the distance between the peaks of the two is less than a preset position threshold, the real-time defect is considered to belong to the defect type corresponding to the historical defect.

[0018] In a second aspect, a system for nondestructive testing of the surface of spindle parts includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned method for nondestructive testing of the surface of spindle parts is implemented.

[0019] The beneficial effects of the present invention are: The present invention obtains an optimal division parameter by analyzing the characteristics of historical defects, extracts the defect area by acquiring the historical grayscale image and the real-time grayscale image of the preprocessed part surface, and performs gradient division and accumulation under the application of the optimal division parameter to obtain the gradient histogram of each defect, then compares the gradient histogram of the real-time defect with the gradient histogram of the historical defect, and outputs the defect type of the part, thereby improving the efficiency and accuracy of surface defect detection of spindle parts.

[0020] By presetting the objective function and determining the optimal partitioning parameters, the method can adaptively adjust the detection parameters to adapt to defects of different types and degrees. This makes the method very applicable and can be applied to various types of spindle parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein: Figure 1 It is a method flow chart of steps S1 to S3 in a method for non-destructive testing of the surface of spindle parts in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] 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 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0023] The embodiment of the present invention discloses a non-destructive detection method for the surface of a spindle part, referring to Figure 1 , including steps S1 to S3, which are specifically as follows: S1: Obtain a historical grayscale image of the preprocessed part surface, and mark the historical grayscale image to obtain multiple defect areas and corresponding historical defect types.

[0024] Use a high-resolution camera or image acquisition device to photograph or scan the surface of the part to obtain a clear original image, perform denoising on the original image to eliminate noise points in the image, and then convert the image into a grayscale image to simplify subsequent processing steps.

[0025] The grayscale image is further preprocessed, such as contrast enhancement and smoothing filtering, to highlight the defect features.

[0026] The grayscale image is processed using an edge detection algorithm (such as the Canny edge detector). Based on the edge detection results, the grayscale image is segmented into multiple regions using an image segmentation algorithm (such as threshold segmentation, region growing, etc.), which include normal regions and defect regions. Defect regions usually have different grayscale values ​​or texture features from normal regions.

[0027] Traverse the segmented image to identify defect areas with specific features. Use marking tools (such as the brush tool in the image processing software) to mark the defect areas.

[0028] Finally, feature extraction is performed on the marked defective areas, and the extracted features are classified and identified using machine learning algorithms (such as support vector machines, neural networks, etc.). Based on the classification results, the defect type corresponding to each defective area is determined.

[0029] It should be noted that what is obtained is a historical grayscale image of the part surface, so the historical defects and the corresponding historical defect types are also marked on the historical grayscale image.

[0030] S2: Divide and accumulate the gradients of each defect area based on the optimal division parameters to obtain the gradient histogram of each defect in the historical grayscale image.

[0031] Gradient is the rate of change of image pixel values ​​in the spatial domain, which can reflect the edge and texture information in the image. In defect detection, edge and texture features are often the key to distinguish defective areas from normal areas. Gradient histogram is an effective feature descriptor that can count the distribution of gradient direction and amplitude in an image. By calculating the gradient histogram of each defective area, information reflecting the defect characteristics can be extracted to provide a basis for subsequent classification and identification.

[0032] When calculating the gradient histogram, traditional methods usually divide the gradient direction into fixed intervals (such as 0°~180°, with every 20° as an interval), then accumulate all the gradient values ​​in a certain gradient direction interval, and obtain the gradient histogram after traversing all gradient direction intervals. However, different features may have different gradient directions and amplitude distributions. Uniformly processing the gradients of all features may ignore these differences, resulting in the inability to accurately reflect the uniqueness of different features in some cases. Therefore, in the gradient histogram feature extraction, by selecting appropriate partitioning parameters, the edge and texture information in the image can be captured more accurately, thereby providing strong support for subsequent image analysis and recognition tasks.

[0033] Specifically, when calculating the gradient histogram features, an objective function is constructed so that the difference in gradient histograms between defects of the same type is smaller and the difference in gradient histograms between defects of different types is larger, thereby obtaining an optimal segmentation parameter.

[0034] First, preset an initial gradient direction interval size , within the preset gradient direction interval, all historical defects are analyzed and the corresponding gradient histogram is generated; For any type of historical defects, the sum of the gradient differences between each gradient histogram in this type and the gradient histograms of historical defects of all other defect types is calculated and recorded as the first parameter; this first parameter reflects the overall difference in gradient characteristics between a certain type of defect and other types of defects.

[0035] Similarly, for any type of historical defects, the sum of the gradient differences between each gradient histogram in the class and other gradient histograms in the same class is calculated and recorded as the second parameter; this second parameter reflects the degree of change in the gradient characteristics within the same type of defects.

[0036] Finally, the sum of the ratios of the first parameter to the second parameter of all historical defects of all defect types is calculated to obtain the value of the objective function. This objective function is used to evaluate the discrimination between different types of defects and the consistency within the same type of defects. Ideally, the gradient difference between different types of defects should be larger (i.e., the first parameter is larger), while the gradient difference between the same type of defects should be smaller (i.e., the second parameter is smaller), so that the value of the objective function is larger, indicating a better classification effect.

[0037] The process of calculating the gradient difference between gradient histograms is as follows: First, determine two gradient histograms such as and The number of gradient direction intervals and the gradient value corresponding to each interval; Then, for each gradient direction interval such as , calculate the gradient histogram and The difference in gradient values ​​over this interval; Finally, the gradient histogram is calculated and The gradient difference satisfies the following relationship:

[0038] In the formula, is the gradient histogram and The gradient difference, In the gradient direction interval Upper gradient histogram The corresponding gradient value is In the gradient direction interval Upper gradient histogram The corresponding gradient value.

[0039] This gradient difference is used to measure the similarity or difference between two gradient histograms. The smaller the gradient difference value, the more similar the two histograms are; the larger the gradient difference value, the greater the difference between the two histograms.

[0040] In another embodiment, a calculation method with a relatively simple calculation process is provided, that is, the following relationship is satisfied:

[0041] In the formula, is the gradient histogram and The gradient difference, In the gradient direction interval Upper gradient histogram The corresponding gradient value is In the gradient direction interval Upper gradient histogram The corresponding gradient value.

[0042] Among them, the constructed objective function is expressed as follows:

[0043] In the formula, is the size of the initial gradient direction interval, is the first parameter, is the second parameter, is the number of historical defects in any type of historical defects, is the total number of historical defect types.

[0044] The first parameter satisfies the relationship:

[0045] In the formula, For the The first of the historical defects The gradient histogram of the historical defects, For the The first of the historical defects Gradient histogram of historical defects.

[0046] The second parameter satisfies the relationship:

[0047] In the formula, For the The first of the historical defects The gradient histogram of the historical defects, For the The first of the historical defects Gradient histogram of historical defects.

[0048] After constructing the objective function, the optimal gradient direction interval size can be found by maximizing the objective function. When the objective function reaches its maximum value, it means that the histogram difference between defects of different categories is the largest, while the histogram difference between defects of the same category is the smallest, which is conducive to the classification of defects.

[0049] Furthermore, when the objective function reaches a maximum value, the input initial gradient direction interval size is used as the optimal partition parameter.

[0050] The optimal division parameters calculated above are used to divide and accumulate the gradients of each defect area to obtain the gradient histogram of each defect in the historical grayscale image.

[0051] S3: Obtain a real-time grayscale image of the preprocessed part surface, extract the defect area of ​​the real-time grayscale image, use the optimal segmentation parameters to obtain the gradient histogram of each real-time defect in the real-time grayscale image, compare the gradient histogram of the real-time defect with the gradient histogram of the historical defect, and output the defect type of the part.

[0052] Under the optimal partitioning parameters, the calculation is performed in any interval such as Below, the differences between different kinds of historical defects.

[0053] For example, in In the interval, the difference between any two types of defects satisfies the following relationship:

[0054] In the formula, For the The gradient difference of the interval, For the All gradient histograms in the defect class are in The mean of the gradient values ​​in the gradient direction interval, For the All gradient histograms in the defect class are in The mean of the gradient values ​​in the gradient direction interval.

[0055] Furthermore, according to the calculation The gradient difference of each interval is used to obtain the difference weight of the interval, which satisfies the following relationship:

[0056] In the formula, For the The difference weight of the interval, For the The gradient difference of the interval.

[0057] By calculating the gradient difference in each gradient direction interval, important features about the defect can be extracted. These features can reflect the differences in gradient distribution of different defects, thus helping to distinguish them. The calculation of difference weights further emphasizes those gradient direction intervals with the largest differences, which may contain the most useful information for classification or recognition.

[0058] In an embodiment of the present invention, a high-resolution camera or image acquisition device is first used to photograph or scan the surface of a part in real time, and the acquired image is preprocessed, such as denoising, contrast enhancement, color space conversion, etc., to obtain a real-time grayscale image of the part surface.

[0059] In the preprocessed real-time grayscale image, image processing techniques (such as threshold segmentation, edge detection, morphological operations, etc.) are used to extract the defect area.

[0060] The above optimal partitioning parameters are used to calculate the gradient histogram of the extracted real-time defects, and the gradient histogram of the real-time defects is compared with the gradient histogram of the historical defects. The general steps of the comparison are as follows: In any gradient direction interval, the absolute value of the difference between the gradient value of the real-time defect and the gradient mean value of the historical defect is calculated and recorded as the third parameter; The product of the third parameter and the difference weight of the corresponding gradient direction interval is used as the difference measure between the real-time defect and the historical defect in the gradient direction interval; If the difference metric is less than the preset threshold, the real-time defect and the historical defect are the same type of defect. The setting of the threshold needs to be determined based on a large amount of experimental data and prior knowledge to ensure its rationality and effectiveness. On the contrary, if the difference metric is greater than or equal to the preset threshold, the real-time defect is a new defect, and the operator can perform the next step of analysis for the new defect.

[0061] Exemplarily, in order to reduce the amount of calculation, the historical grayscale image is set to include three defects: cracks, burrs and pits. The difference metrics between the real-time defect and the three defects in the gradient direction interval are calculated respectively, and three difference metrics are obtained, which are compared with the preset thresholds respectively. If the difference metric corresponding to the real-time defect and the crack defect is less than the preset threshold, the real-time defect is a crack defect.

[0062] In another embodiment, for each gradient histogram of the real-time defect, it is matched with the gradient histograms of each historical defect. If the Euclidean distance between the gradient histogram of a historical defect and the gradient histogram of the real-time defect is the smallest, the real-time defect is considered to belong to the defect type corresponding to the historical defect.

[0063] It should be noted that a distance threshold may be set, and if the minimum distance exceeds the threshold, the real-time defect is considered not to belong to any known historical defect category and may need to be further analyzed or marked as a new type.

[0064] In another embodiment, the peak positions of the gradient histogram of the real-time defect and the gradient histogram of the historical defect may be compared. If the distance between the peaks of the two is less than a preset position threshold, the real-time defect is considered to belong to the defect type corresponding to the historical defect.

[0065] In general, the main difference between the above three comparison methods lies in the different ways they process and compare data. The first method is to quantify the difference and set a threshold to make a judgment; the second method is to minimize the difference by calculating the Euclidean distance for matching; the third method focuses on comparing the proximity of the peak positions. These different methods can be selected and used according to specific application scenarios and needs to improve the accuracy and efficiency of non-destructive testing.

[0066] In addition, in the real-time grayscale image after preprocessing, if the defective area is not extracted using image processing technology (such as threshold segmentation, edge detection, morphological operation, etc.), the part is output as qualified.

[0067] An embodiment of the present invention further discloses a system for nondestructive testing of the surface of spindle parts, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a method for nondestructive testing of the surface of spindle parts according to the present invention is implemented.

[0068] The system also includes other components familiar to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0069] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module or both. Any such computer storage medium may be part of a device or accessible or connectable to a device.

[0070] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0071] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A non-destructive testing method for the surface of spindle parts, characterized in that: include: Obtain a historical grayscale image of the preprocessed part surface, and mark the historical grayscale image to obtain multiple defect areas and corresponding historical defect types; The gradient of each defect area is divided and accumulated based on the optimal division parameter to obtain the gradient histogram of each defect in the historical grayscale image; Obtain the real-time grayscale image of the preprocessed part surface, extract the defect area of ​​the real-time grayscale image, use the optimal segmentation parameters to obtain the gradient histogram of each real-time defect in the real-time grayscale image, compare the gradient histogram of the real-time defect with the gradient histogram of the defect in the historical grayscale image, and output the defect type of the part; The optimal partitioning parameter is positively correlated with a preset objective function, and the objective function is positively correlated with the difference in gradient histograms between historical defects of the same type, and negatively correlated with the difference in gradient histograms between historical defects of different types.

2. A non-destructive testing method for the surface of spindle parts according to claim 1, characterized in that: The process of obtaining the objective function includes: Presetting an initial gradient direction interval size, and obtaining a gradient histogram of all historical defects under the division of the initial gradient direction interval size; Calculate the sum of the gradient differences between each gradient histogram of historical defects of any defect type and the gradient histogram of historical defects of all other defect types, and record it as the first parameter; Calculate the sum of the gradient differences between each gradient histogram of historical defects of any defect type and other gradient histograms of historical defects of the same defect type, and record it as the second parameter; The sum of the ratios of the first parameter to the second parameter of all historical defects of all defect types is calculated to obtain the value of the objective function.

3. A non-destructive testing method for the surface of spindle parts according to claim 2, characterized in that: When the value of the objective function is the largest, the corresponding initial gradient direction interval size is the optimal partition parameter.

4. A non-destructive testing method for the surface of spindle parts according to claim 3, characterized in that: Also includes: Under the partitioning with the optimal partitioning parameters, the absolute difference between the mean values ​​of the gradient values ​​of any two types of historical defects in the same gradient direction interval is calculated, and the absolute difference between the mean values ​​of the gradient values ​​is normalized to obtain the difference weight of the corresponding gradient direction interval.

5. A non-destructive testing method for the surface of spindle parts according to claim 4, characterized in that: The gradient histogram of the real-time defect is compared with the gradient histogram of the historical defect, and the output result is that the defect types of the part include: In any gradient direction interval, the absolute value of the difference between the gradient value of the real-time defect and the gradient mean value of the historical defect is calculated and recorded as the third parameter; The product of the third parameter and the difference weight of the corresponding gradient direction interval is used as the difference measure between the real-time defect and the historical defect in the gradient direction interval; If the difference metric is less than a preset threshold, the real-time defect and the historical defect are the same type of defect.

6. A non-destructive testing method for the surface of spindle parts according to claim 1, characterized in that: The comparing the gradient histogram of the real-time defect with the gradient histogram of the historical defect includes: The gradient histogram of each real-time defect is matched with the gradient histogram of each historical defect. If the Euclidean distance between the gradient histogram of a historical defect and the gradient histogram of the real-time defect is the smallest, the real-time defect is considered to belong to the defect type corresponding to the historical defect.

7. A non-destructive testing method for the surface of spindle parts according to claim 1, characterized in that: The comparing the gradient histogram of the real-time defect with the gradient histogram of the historical defect includes: The peak positions of the gradient histogram of the real-time defect and the gradient histogram of the historical defect are compared. If the distance between the peaks of the two is less than a preset position threshold, the real-time defect is considered to belong to the defect type corresponding to the historical defect.

8. A non-destructive testing system for the surface of spindle parts, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the method for non-destructive surface testing of spindle parts according to any one of claims 1 to 7 is implemented.

Images