A four-ball wear scar intelligent evaluation method based on deep learning

Abstract

The application relates to the technical field of tribology and wear characteristic analysis, and discloses a four-ball wear scar wear intelligent evaluation method based on deep learning, first, an image acquisition device is used to carry out wear scar image acquisition and pretreatment; then, multi-wear type recognition and segmentation are carried out, a deep learning image segmentation model is used to recognize and segment different wear areas in the wear scar image, and a segmentation mask corresponding to different wear types is output; then, after the segmentation masks of different wear types are obtained, pixel-level statistics are carried out on each wear area, wear characteristic evaluation indexes are constructed, and corresponding comprehensive damage indexes are obtained; then, after the comprehensive damage indexes corresponding to different wear types are obtained, the dominant wear mechanism of the four-ball wear scar is judged; finally, after the comprehensive damage indexes of different wear types are obtained, the wear degree is divided into wear grades. The application can improve the objectivity, stability and machine interpretation ability of four-ball wear scar analysis.

Description

Technical Field

[0001] This invention relates to the field of tribology and wear characteristic analysis technology, and in particular to a deep learning-based intelligent assessment method for four-ball wear scars. Background Technology

[0002] The four-ball test is an important method for evaluating the anti-wear performance of lubricants and the wear behavior of friction pairs. After the test, the morphology of the wear scars formed on the surface of the steel balls reflects the anti-wear performance of the lubricant under load, speed, temperature, and frictional contact conditions. Traditional four-ball wear scar analysis mainly relies on manual observation and measurement of the wear scar diameter, and usually calculates the area in an approximate circular or elliptical manner to analyze and evaluate the wear scar. Although this method is simple to operate, it mainly reflects the overall size change of the wear scar and is difficult to fully analyze the complex local damage inside the wear scar.

[0003] In recent years, with the development of image processing and intelligent analysis technologies, some wear mark analysis methods have begun to employ techniques such as edge detection, region segmentation, morphological analysis, and image recognition to improve wear mark detection efficiency and the degree of automation in analysis. These methods have shown some improvement in areas such as wear mark contour extraction, morphological feature analysis, and abnormal state recognition.

[0004] However, existing methods for analyzing wear scar images still have the following shortcomings:

[0005] 1. Existing methods are mostly focused on identifying the wear scar contour, wear scar direction, shape distortion or abnormal state. The main evaluation object is still biased towards the overall wear scar geometry, which is difficult to reflect the wear differences of different local damage areas inside the wear scar.

[0006] 2. In four-ball wear scars, burn areas, scratch areas and spalling areas often coexist and are characterized by blurred boundaries, irregular shapes, large size differences and discrete local distribution. It is difficult to accurately determine the dominant wear mechanism based solely on the wear scar diameter and total area, and it is also difficult to identify the complex wear state formed by the combined action of multiple wear mechanisms.

[0007] 3. Existing methods lack a continuous analytical process from wear area identification, damage feature quantification, dominant wear mechanism judgment to wear level classification, making it difficult to provide stable, objective, and reproducible multidimensional evaluation results for lubricant anti-wear performance evaluation and tribological mechanism analysis. Summary of the Invention

[0008] To achieve the above objectives, this invention addresses the aforementioned problems by providing a deep learning-based intelligent assessment method for four-ball wear scars. This method solves the problems of low efficiency in manual measurement, strong subjectivity of results, difficulty in distinguishing different wear areas such as burns, scratches, and spalling, and difficulty in objectively evaluating the dominant wear mechanism and complex wear state in traditional four-ball wear scar analysis. This method aims to improve the objectivity, stability, and mechanistic explanation capabilities of four-ball wear scar analysis.

[0009] This invention adopts the following technical solution: a deep learning-based intelligent assessment method for four-ball wear scars, comprising the following steps:

[0010] Step 1: Use image acquisition equipment to acquire and preprocess images of wear marks;

[0011] Step 2: Perform multi-wear type identification and segmentation. Input the preprocessed wear mark image into the deep learning image segmentation model. Use the deep learning image segmentation model to identify and segment different wear regions in the wear mark image, and output the segmentation mask corresponding to different wear types.

[0012] Step 3: Construct wear feature evaluation index. After obtaining the segmentation mask for different wear types, perform pixel-level statistics on each wear area to construct wear feature evaluation index and obtain the comprehensive damage index corresponding to different wear types.

[0013] Step 4: Determine the dominant wear mechanism. After obtaining the comprehensive damage index corresponding to different wear types, determine the dominant wear mechanism of the four-ball wear scar.

[0014] Step 5: Wear level classification. After obtaining the comprehensive damage index of different wear types, wear levels are classified by using a robust statistical threshold based on quartiles.

[0015] Furthermore, in step 1, to obtain high-quality wear scar data, the wear scars of the test steel balls under the four-ball test are first photographed using image acquisition equipment such as an optical microscope or a high-resolution camera, ensuring that the wear scar images cover typical wear states under different working conditions and lubrication conditions. Image acquisition should ensure uniform illumination and sufficiently high resolution to capture fine scratches and localized burn areas. The preprocessing of the wear scar images in step 1 includes image cropping, size normalization, brightness adjustment and contrast enhancement, noise suppression, and removal of invalid background areas.

[0016] Furthermore, in step 2, a deep learning image segmentation model is used to analyze the wear scar image to achieve automatic segmentation of different wear types. The wear areas identified in step 2 include at least the wear scar contour area, burn area, scratch area, and spalling area. The wear scar contour area reflects the overall wear area range and is used for overall wear assessment. The burn area is a thermally damaged area caused by localized high-temperature friction, typically with blurred boundaries and low contrast. The scratch area presents a slender, directional scratch area, possibly formed by hard particles or strong shearing action. Spalling wear refers to irregular areas formed by the shedding of wear debris.

[0017] Therefore, the deep learning image segmentation model can output segmentation masks corresponding to different wear types, transforming various wear regions from indistinguishable overall wear scar states into structured wear information that can be independently statistically analyzed, providing a data foundation for the subsequent construction of multidimensional wear assessment indicators.

[0018] Furthermore, the spatial distribution morphology of different wear mechanisms in four-ball wear scars shows significant differences. Extensive sample studies have revealed that:

[0019] Burn-type abrasion typically exhibits characteristics of large-area continuous distribution, blurred boundaries, and localized aggregation. The degree of damage is mainly reflected in the overall expansion of the abrasion area.

[0020] Scratch-type wear typically exhibits a thin, elongated, multi-directional discrete distribution. The degree of damage depends not only on the area variation but also on the number of local damage instances and the degree of spatial dispersion.

[0021] Exfoliation wear typically manifests as multiple irregular discrete regions. Its formation process is related to the propagation of local fatigue cracks and material detachment. It exhibits both area expansion characteristics and obvious discrete distribution characteristics of multiple instances.

[0022] Based on the spatial distribution patterns corresponding to the different wear mechanisms mentioned above, this invention not only statistically analyzes the area of ​​the wear region, but also introduces relative area proportion, relative area growth rate, and normalized instance density, and constructs a comprehensive damage index for wear mechanism analysis, thereby realizing the transformation from traditional geometric dimension measurement to intelligent analysis of wear mechanisms.

[0023] First, based on the segmentation results, pixel-level area calculations are performed for each wear type to avoid errors caused by the use of circular or elliptical approximations in traditional methods. The total pixel area of ​​the wear scar contour obtained from the segmentation mask is... , of which The area of ​​the wear-like region is Then the relative area percentage of this type of wear is:

[0024]

[0025] in, Indicates the first The percentage of the area of ​​wear-like regions in the overall wear scar area.

[0026] Compared to absolute area, relative area ratio can eliminate the influence of differences in wear scar size, making it more suitable for comparing the degree of wear between different samples.

[0027] Furthermore, to eliminate the influence of natural area differences among different wear types, a relative area growth rate is introduced:

[0028]

[0029] in:

[0030] The median area percentage of this wear type across all samples is given by e, where e is a natural constant.

[0031] This indicates that the area of ​​this type of wear in the image is higher than the typical level;

[0032] This indicates a level below typical;

[0033] When the first wear scar is identified When the wear area is 0, it indicates that this wear type does not exist in the sample. In this case, the relative area growth rate of this wear type is not calculated; only the area of ​​the wear type is considered. The wear type is calculated.

[0034] Furthermore, the relative area growth rate cannot reflect the dispersion of the wear region; therefore, a connected component analysis is performed on the wear region to statistically analyze the first... Number of instances of wear type in wear scar images The normalized instance density is defined as:

[0035]

[0036] in:

[0037] This represents the median number of instances of this wear type across all samples.

[0038] No. When the number of wear type instances is 0, it means that this wear type does not exist in the sample.

[0039] Since different wear mechanisms manifest not only as changes in wear area but also as different spatial distribution patterns, simply using area indicators is insufficient to fully reflect the actual wear state. Therefore, this invention constructs a comprehensive damage index by integrating the relative area growth rate and normalized instance density to simultaneously characterize the expansion intensity and spatial dispersion characteristics of the wear region, thereby improving the ability to distinguish between different wear mechanisms. The comprehensive damage index is defined as follows:

[0040]

[0041] in:

[0042] For the first The overall damage index of wear-like areas;

[0043] This represents the relative area growth rate.

[0044] To normalize instance density;

[0045] and These are weighting coefficients, with different weighting combinations used for different wear types.

[0046] The comprehensive damage index is used to comprehensively characterize the severity and spatial distribution characteristics of different wear types.

[0047] Through the aforementioned wear characteristic evaluation indicators, this invention can achieve independent quantitative analysis of different wear types and establish a multi-dimensional intelligent wear evaluation system that includes wear area identification, feature statistics, and comprehensive damage assessment, providing a basis for subsequent judgment of dominant wear mechanisms and wear level classification.

[0048] Furthermore, the comprehensive damage index corresponding to the burn area, scratch area, and peeling area was calculated separately. The study compared the overall damage levels among different wear types. Since different wear mechanisms correspond to different spatial damage patterns, the overall damage index of different wear types can reflect the activity level of the corresponding wear mechanism.

[0049] A high burn damage index indicates that there is significant large-area thermal damage propagation behavior in the wear scar.

[0050] A high scratch damage index indicates that there is significant directional plowing damage in the wear scar.

[0051] A high overall exfoliation damage index indicates significant material detachment within the wear scar.

[0052] Therefore, by comparing the comprehensive damage indices corresponding to different wear types, the dominant wear mechanism can be intelligently determined. If all wear indices are similar, it indicates a complex wear state, meaning that multiple wear mechanisms act on the friction surface together and simultaneously affect the wear scar morphology.

[0053] To improve the stability of the determination of the dominant wear mechanism, a threshold for determining the wear mechanism can be set. When the difference between the comprehensive damage indices of different wear types is less than the aforementioned threshold, it is considered that the corresponding wear mechanisms jointly play a dominant role, and different thresholds can be used for different working conditions.

[0054] By using the above-mentioned method for determining the dominant wear mechanism, this invention can transform the single-dimensional measurement results in traditional four-ball wear scar analysis into intelligent analysis results oriented towards wear mechanism, thereby achieving automatic identification and mechanism analysis of different wear states.

[0055] Furthermore, after obtaining the comprehensive damage index for different wear types, a robust statistical threshold based on quartiles was used to classify the degree of discrete damage-type wear. This was applied to all test samples. Calculate the first quartile respectively and the third and fourth quartiles :

[0056] Then category The degree of wear can be classified as follows:

[0057]

[0058] By employing the robust statistical threshold based on quartiles for wear level classification, the influence of extreme or abnormal wear samples on the evaluation threshold can be reduced, making the wear level determination more stable and objective. Through this classification method, the present invention can output the wear level corresponding to various wear regions based on the comprehensive damage index of different wear types, providing a quantitative basis for evaluating the anti-wear performance of lubricants and analyzing the results of four-ball tests.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] 1. This invention provides a deep learning-based intelligent assessment method for four-ball wear scars. Through a deep learning image segmentation model, it automatically identifies and separates different wear regions in four-ball wear scar images, enabling independent analysis of wear scar contours, burn areas, scratch areas, and spalling areas. Compared to traditional methods relying on manual measurement of wear scar diameter or area, this invention reduces the subjectivity and error caused by human experience, improving the automation, efficiency, and consistency of wear scar analysis in complex wear scenarios.

[0061] 2. This invention provides a deep learning-based intelligent wear assessment method for four-ball wear scars. It constructs a multi-dimensional wear assessment index system including wear area proportion, relative area growth rate, normalized instance density, and comprehensive damage index. This system extends the traditional single wear scar size evaluation to a comprehensive quantitative analysis of the severity, spatial distribution characteristics, and discrete damage degree of different wear types. By performing pixel-level statistical analysis and multi-index fusion analysis on different wear areas, it can more comprehensively reflect the true wear state of four-ball wear scars under complex working conditions, improving the accuracy and objectivity of wear assessment results.

[0062] 3. This invention provides a deep learning-based intelligent assessment method for four-ball wear scars, establishing a correlation between the morphological characteristics of the wear area and the wear mechanism. Compared to traditional evaluation methods based solely on wear scar diameter or total area, this method can more accurately distinguish between thermal wear, ploughing wear, and fatigue spalling wear, improving the accuracy of identifying the dominant wear mechanism under complex composite wear conditions, thereby enhancing the stability and reliability of wear level determination. By establishing a complete intelligent assessment process from wear area identification, feature quantification analysis, dominant wear mechanism judgment to wear level classification, this method provides a more objective, stable, and quantifiable analytical basis for evaluating the anti-wear performance of lubricants, tribological research, and the analysis of complex wear states. Attached Figure Description

[0063] Figure 1 This is a flowchart illustrating the overall process of a deep learning-based intelligent assessment method for four-ball wear scars according to the present invention.

[0064] Figure 2 This is a schematic diagram of the original images of the steel ball wear scars under different working conditions in the four-ball test of this invention, wherein:

[0065] Figure 2 (a) in the image is the original image of the wear scar under the first set of working conditions;

[0066] Figure 2 (b) in the image is the original image of the wear scar under the second set of working conditions;

[0067] Figure 2 (c) in the image is the original image of the wear scar under the third working condition;

[0068] Figure 3 For the corresponding Figure 2 A schematic diagram of the deep learning multi-type wear region segmentation results for the original images of each wear scar, wherein:

[0069] Figure 3 (a) in the middle is Figure 2 (a) shows the segmentation results of the wear scar contour, burn area, scratch area and peeling area.

[0070] Figure 3(b) in the middle is Figure 2 The segmentation results of wear scar contour, burn area, scratch area and peeling area corresponding to (b) in the figure;

[0071] Figure 3 (c) in the middle is Figure 2 The segmentation results of wear scar contour, burn area, scratch area and peeling area corresponding to (c) in the figure;

[0072] Figure 4 A bar chart comparing the comprehensive damage index of three wear scar samples for different wear types;

[0073] Figure 5 The distribution of comprehensive damage index levels for different wear types in three wear scar samples. Detailed Implementation

[0074] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0075] Example 1

[0076] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0077] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings.

[0078] Conversely, this invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the invention as defined in the claims. Furthermore, to provide a better understanding of the invention, certain specific details are described in detail below. However, those skilled in the art will fully understand the invention even without these detailed descriptions.

[0079] This embodiment provides a deep learning-based intelligent assessment method for four-ball wear scars, primarily used for automatic identification, wear region separation, wear feature quantification, dominant wear mechanism determination, and wear level classification of wear scar images on the surface of steel balls after a four-ball test. Unlike traditional methods that rely solely on manual measurement of wear scar diameter or area, this invention does not simply obtain a single wear scar size result. Instead, it separates different wear types within the wear scar image and calculates their respective area proportions, relative area growth rates, instance distribution characteristics, and comprehensive damage indices, thereby constructing a multi-dimensional intelligent wear assessment system for four-ball wear scars.

[0080] This embodiment provides a deep learning-based intelligent assessment method for four-ball wear scars, such as... Figure 1 As shown, it includes the following steps:

[0081] Step 1: Acquisition and Preprocessing of Wear Scratch Images. First, acquire wear scratch images of the test steel balls after the four-ball friction and wear test. Image acquisition equipment can include an optical microscope, industrial camera, high-definition microscopic imaging system, or other imaging devices capable of obtaining wear scratch images. During acquisition, the wear scratch area should be located as centrally as possible in the image, ensuring that features such as the wear scratch outline, internal scratches, localized burns, and flaking areas are clearly recorded.

[0082] In a preferred embodiment, the acquired raw wear-spot images can be preprocessed, including but not limited to image cropping, size normalization, brightness adjustment, contrast enhancement, noise suppression, and removal of invalid background regions. Preprocessing reduces the impact of uneven illumination, metallic reflections, background textures, and image scale differences on subsequent wear area recognition results, improving the stability of deep learning image segmentation models for recognizing different wear types. Specifically, image cropping preserves the effective image range containing the wear-spot regions; size normalization adjusts images acquired from different acquisition devices or at different magnifications to a uniform input size; brightness adjustment and contrast enhancement improve the visual differences between burn areas, scratch areas, and the background; and noise suppression reduces the interference of local noise on the segmentation mask. These preprocessing steps can be selectively performed based on the actual image quality; it is not required that all images undergo the exact same preprocessing procedure.

[0083] Step 2: Multi-wear type identification and segmentation. The preprocessed wear image is input into a deep learning image segmentation model, which identifies and segments different wear regions within the image. The deep learning image segmentation model can be a semantic segmentation model, an instance segmentation model, or a multi-label segmentation model; its specific network structure is not limited, as long as it can output segmentation masks corresponding to different wear types. The deep learning image segmentation model can be Mask2Former, U-Net, DeepLab, SegFormer, or other image segmentation models capable of outputting segmentation masks for different wear types. In this embodiment, Mask2Former is used, and the training data consists of 500 manually labeled four-ball wear images, with label categories including wear outline, burns, scratches, and peeling.

[0084] In a preferred embodiment, the wear area includes at least a wear scar contour area, a burn area, a scratch area, and a spalling area. The wear scar contour area characterizes the overall wear range; the burn area characterizes the degree of localized thermal damage; the scratch area characterizes the directional damage characteristics formed by ploughing wear or scratches from hard particles; and the spalling area characterizes the irregular damage characteristics formed by fatigue spalling or the shedding of wear debris.

[0085] The output of a deep learning image segmentation model can be a binary mask or a multi-class mask corresponding to each type of wear region. For a certain type of wear region, pixels belonging to that type of wear region in the mask are recorded as valid pixels, and pixels not belonging to that type of wear region are recorded as invalid pixels. Through the above masks, the different wear types that originally existed mixed in the overall wear mark can be transformed into structured wear information that can be independently statistically analyzed.

[0086] In this invention, the deep learning image segmentation model primarily serves as a tool for identifying and separating wear regions, aiming to obtain the location and pixel range of regions with different wear types. In other words, the core of this invention is not to limit itself to a specific neural network structure, but rather to further construct a multidimensional wear assessment system suitable for four-ball wear scar analysis based on the segmentation results. Therefore, in practical implementation, different deep learning segmentation models can be selected according to data scale, image quality, computing resources, and recognition accuracy requirements.

[0087] Step 3: Construction of Wear Feature Evaluation Indicators. After obtaining the segmentation masks for different wear types, pixel-level statistics are performed on each wear region to construct wear feature evaluation indicators. These indicators include area percentage, relative area growth rate, normalized instance density, and comprehensive damage index.

[0088] First, the total area of ​​the wear scar is calculated based on the segmentation mask of the wear scar contour region. Let the total pixel area of ​​the wear scar contour obtained by the segmentation mask be... , No. The pixel area of ​​the wear-like region is , of which The wear area can be a burned area, a scratched area, or a peeling area. Relative area percentage of wear-like regions Defined as:

[0089]

[0090] in, Indicates the first The percentage of the wear-like area within the overall wear scar area. Compared to using the absolute area directly, the percentage of the area can mitigate the impact of differences in wear scar size, making the wear conditions between different samples more comparable.

[0091] Furthermore, to reduce the impact of differences in the natural area distribution of different wear types on the evaluation results, a relative area growth rate is introduced. Let the... The median area percentage of wear-like regions in the sample set is Then the first The relative area growth rate of wear-like regions Defined as:

[0092]

[0093] in, Used to characterize the first in the current wear scar image The extent to which the wear-like region expands relative to the typical level of that type of wear. When When, it indicates that the area of ​​this type of wear region is higher than the typical level in the sample set; when When this occurs, it indicates that the area of ​​this type of wear region is lower than the typical level in the sample set. If the value is zero, it means that there is no such wear area in the current wear scar image. In this case, the evaluation result corresponding to this type of wear can be marked as non-existent, or this type of wear index can be excluded from the comprehensive evaluation.

[0094] Relative area percentage and relative area growth rate are mainly used to characterize the extent of area expansion of the wear region, but they cannot fully reflect the discrete distribution of the wear region. For example, under the same area percentage, a concentrated, contiguous burn area may reflect a different wear state than multiple dispersed burn areas. Therefore, this invention further introduces instance number statistics and normalized instance density.

[0095] Specifically, for the first Connectivity analysis is performed on the segmentation mask of the wear-type region to count the number of instances of this type of wear-type region in the current wear scar image. Let the first... The median number of instances of wear-type regions in the sample set is Then the normalized instance density Defined as:

[0096]

[0097] in, Used to characterize the The spatial dispersion of wear-like regions and the characteristics of the number of local damage areas. When When the size is large, it indicates that this type of wear area exhibits a more obvious dispersed distribution or multi-point damage characteristics in the wear scar; when When the value is smaller, it indicates that the number of such wear areas is small or the degree of concentration is high.

[0098] After obtaining the relative area growth rate and normalized instance density, a comprehensive damage index is constructed. Used to comprehensively reflect the first The extent of area expansion and the intensity of discrete distribution of wear-like regions:

[0099]

[0100] in:

[0101] For the first The overall damage index of wear-like areas;

[0102] For the first The relative area growth rate of wear-like regions;

[0103] For the first Normalized instance density of wear-like regions;

[0104] and These are weighting coefficients, with different weighting combinations used for different wear types.

[0105] In a preferred embodiment, the area feature of the burn region has a higher weight, the instance density feature of the scratch region has a higher weight, and the peeling region uses a balanced weight.

[0106] in:

[0107] and These represent the area feature weighting coefficient and the instance density weighting coefficient of the burn region, respectively.

[0108] and These represent the area feature weighting coefficient and the instance density weighting coefficient of the scratch region, respectively.

[0109] and These represent the area feature weighting coefficient and the instance density weighting coefficient of the peeled-off region, respectively.

[0110] In one implementation:

[0111] , ;

[0112] , ;

[0113] , .

[0114] In a preferred embodiment, adjustments can be made according to different test conditions or evaluation requirements. and For example, when the research focus is on large-area burn injuries, the weight of area features can be increased accordingly. When the research focus is on localized multi-point peeling or scattered scratches, the instance density weight can be increased accordingly. Therefore, the weighting coefficients can be set according to the actual application scenario.

[0115] By using the above indicators, this invention transforms different wear areas in the wear scar image into statistically comparable wear features, realizing the transformation from evaluation based on a single wear scar size to a comprehensive evaluation based on multiple wear types and multiple indicators.

[0116] Step 4: Determining the Dominant Wear Mechanism. After obtaining the comprehensive damage index corresponding to different wear types, the dominant wear mechanism of the four-ball wear scar is determined. Specifically, the comprehensive damage index corresponding to the burn area, scratch area, and spalling area is calculated separately, and the comprehensive damage degree between different wear types is compared.

[0117] To avoid misjudgment when the overall damage index of different wear types is similar, a wear mechanism judgment threshold can be set. .

[0118] When the difference between the comprehensive damage index corresponding to a certain wear type and the comprehensive damage index corresponding to other wear types is greater than the wear mechanism determination threshold... If the wear pattern is as described above, then the wear mechanism corresponding to that wear type is determined to be the dominant wear mechanism. Specifically, when the comprehensive damage index of the burn area is the highest, the wear scar is determined to be dominated by thermal wear or localized thermal damage; when the comprehensive damage index of the scratch area is the highest, the wear scar is determined to be dominated by ploughing wear or abrasive scratches; when the comprehensive damage index of the spalling area is the highest, the wear scar is determined to be dominated by fatigue spalling wear.

[0119] If the difference between the combined damage indices of two or more wear types is less than the wear mechanism determination threshold... If the corresponding wear mechanisms play a dominant role, then the wear scar is classified as a composite wear state.

[0120] For example, let the comprehensive damage index of the burn area, scratch area, and peeling area be respectively... , and .like The difference between the comprehensive damage index corresponding to the other two types of wear areas is greater than the threshold. If so, it is determined to be dominated by thermal wear; if and The difference between them is less than the threshold. If both values ​​are higher than the comprehensive damage index corresponding to the spalling area, then it can be determined as a composite wear state caused by the combined effects of thermal wear and plowing wear.

[0121] Based on the above-mentioned dominant wear mechanism, this invention can further transform the single-dimensional measurement results in traditional four-ball wear scar analysis into analysis results oriented towards wear mechanism, which helps to determine the changes in wear behavior under different lubricants or different operating conditions.

[0122] Step 5: Wear Level Classification. After obtaining the comprehensive damage index for different wear types, a robust statistical threshold based on quartiles is used to classify the wear degree. This method does not rely on fixed empirical thresholds, but determines the evaluation threshold based on the distribution of the comprehensive damage index in the sample set, thus reducing the impact of extreme or abnormal wear samples on the evaluation results.

[0123] For the sample set, the th Comprehensive damage index of wear-like areas Calculate its first quartile. and the third quartile, where, For the first The first quartile of the comprehensive damage index for wear-like areas. For the first The third quartile of the comprehensive damage index for wear-like areas.

[0124] In a preferred embodiment, based on the first quartile and the third quartile, the first... The wear degree of wear-like areas is divided into slight wear, moderate wear, and severe wear:

[0125]

[0126] Through the above-mentioned grading method, the present invention can output the corresponding wear level according to the comprehensive damage index of different wear types. The grading result can be used to judge the wear severity of a single sample, or to compare the wear state under different lubricants, different loads, different speeds, or different test times.

[0127] Example 2

[0128] Based on the deep learning-based intelligent assessment method for four-ball wear scars in Example 1, this example selects the lower test steel ball after the four-ball friction and wear test as the test object, and uses an optical microscope to acquire images of the wear scars on the surface of the steel ball to obtain the original wear scar images including the wear scar contour, burn area, scratch area and peeling area.

[0129] like Figure 2 As shown, in this embodiment, three sets of test steel balls after four-ball friction and wear tests under different working conditions were selected as test objects. An optical microscope was used to acquire images of the wear marks on the steel ball surface, obtaining original wear mark images under different wear states. Figure 2(a) Figure 2 (b) and Figure 2 (c) in the image corresponds to the typical original wear scar morphology formed under three different lubrication conditions.

[0130] First, the acquired raw wear mark images are preprocessed. This preprocessing includes image cropping, size normalization, brightness adjustment, contrast enhancement, and noise suppression to reduce the impact of background texture, metallic reflection, and uneven lighting on subsequent recognition results, thereby improving the stability of wear area recognition under different working conditions.

[0131] Subsequently, the preprocessed wear mark image is input into a deep learning image segmentation model to automatically identify and segment the wear mark contour region, burn region, scratch region, and peeling region.

[0132] like Figure 3 As shown, the deep learning image segmentation model can effectively distinguish between different wear types. Wherein: Figure 3 In (a) of the image, the burned area and the area of ​​the flaking area in the wear scar are both larger than those in conventional wear. Figure 3 In (b) of the image, the burn area in the wear scar is relatively large and there is no peeling. Figure 3 The wear patterns in the wear scars corresponding to (c) are not significantly different.

[0133] based on Figure 3 The segmentation results shown are used to perform pixel-level statistics on different wear regions, and to calculate indicators such as area percentage, relative area growth rate, and normalized instance density. Based on this, the comprehensive damage index (SDI) corresponding to different wear types is further calculated.

[0134] like Figure 4 As shown, a statistical analysis of the comprehensive damage index for different wear types in three wear scar samples was performed. The results indicate that there are significant differences in the comprehensive damage index corresponding to various wear regions under different working conditions. Specifically, Figure 4 When the combined damage indices of the three wear types in (a) are close, it indicates that the wear scar is in a state of complex wear. Figure 4 The comprehensive damage index of the burn area in (b) is higher than that of the scratch area, indicating that the wear scar is dominated by thermal wear mechanism under this working condition. Figure 4 The comprehensive damage index of each wear area in (c) is low, indicating that the wear scar is in a low-degree complex wear state;

[0135] like Figure 5 As shown, the wear levels are classified according to the statistical distribution of the comprehensive damage index for different wear types. Figure 5In (a) of the study, the comprehensive damage indices corresponding to burns, scratches and peeling are 1.25, 1.12 and 1.16 respectively, all of which are at a high level and are judged to be in a state of severe wear or close to severe wear. Figure 5 In (b) of the study, the comprehensive damage index of the burn area was 1.17, the comprehensive damage index of the scratch area was 0.71, and the comprehensive damage index of the peeling area was 0.00, indicating that the wear scar was mainly caused by burn wear. Figure 5 In (c), the comprehensive damage indices for burns, scratches, and spalling are 0.84, 0.88, and 0.97, respectively, all falling within the moderate wear range. Therefore, this invention can achieve quantitative grading and judgment of different wear types based on the comprehensive damage index.

[0136] As can be seen from the above embodiments, the present invention can realize the automatic identification of different wear types of four-ball wear scars, analysis of spatial distribution characteristics, judgment of dominant wear mechanism, and classification of wear level, thereby improving the objectivity, stability and automation of four-ball wear scar analysis results.

[0137] The embodiments and implementation process of the present invention have been described in detail above with reference to the accompanying drawings and tables, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments, including components, without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A deep learning-based intelligent assessment method for four-ball wear scars, characterized in that, Includes the following steps, Step 1: Use image acquisition equipment to acquire and preprocess images of wear marks; Step 2: Perform multi-wear type identification and segmentation. Input the preprocessed wear mark image into the deep learning image segmentation model. Use the deep learning image segmentation model to identify and segment different wear regions in the wear mark image and output the segmentation mask corresponding to different wear types. Step 3: Construct wear feature evaluation index. After obtaining the segmentation mask for different wear types, perform pixel-level statistics on each wear area to construct wear feature evaluation index and obtain the comprehensive damage index corresponding to different wear types. Step 4: Determine the dominant wear mechanism. After obtaining the comprehensive damage index corresponding to different wear types, determine the dominant wear mechanism of the four-ball wear scar. Step 5: Wear level classification. After obtaining the comprehensive damage index of different wear types, wear levels are classified by using robust statistical thresholds based on quartiles.

2. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 1, characterized in that: The preprocessing of the abrasion mark image in step 1 includes image cropping, size normalization, brightness adjustment and contrast enhancement, noise suppression, and removal of invalid background areas.

3. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 2, characterized in that: The wear area in step 2 includes at least the wear scar outline area, the burn area, the scratch area, and the peeling area.

4. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 3, characterized in that: The wear characteristic assessment indicators in step 3 include relative area percentage, relative area growth rate, normalized instance density, and comprehensive damage index.

5. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 4, characterized in that: The total area of ​​the wear mark is calculated based on the segmentation mask of the wear mark contour region; let the total pixel area of ​​the wear mark contour obtained by the segmentation mask be... , No. The pixel area of ​​the wear-like region is , of which The wear area can be a burned area, a scratched area, or a peeling area. Relative area percentage of wear-like regions Defined as: ； in, Indicates the first The relative area percentage of wear-like regions within the overall wear scar area.

6. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 5, characterized in that: To reduce the impact of differences in natural area distribution among different wear types on the evaluation results, a relative area growth rate is introduced; let the first... The median area percentage of wear-like regions in the sample set is Then the first The relative area growth rate of wear-like regions Defined as: ； in, Used to characterize the first in the current wear scar image The extent to which the wear-type region expands relative to the typical level of that type of wear; when When this occurs, it indicates that the area of ​​this type of wear region is higher than the typical level in the sample set; when When this occurs, it indicates that the area of ​​this type of wear region is lower than the typical level in the sample set; when When the value is zero, it means that the relative area growth rate and normalized instance density of this type of wear are not calculated, and the comprehensive damage index is considered to be 0.

7. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 6, characterized in that: For the first Connectivity analysis is performed on the segmentation mask of the wear-type region to count the number of instances of this type of wear-type region in the current wear scar image. Let the first The median number of instances of wear-type regions in the sample set is Then the normalized instance density Defined as: ； in, Used to characterize the The spatial dispersion of wear-like areas and the characteristics of the number of local damages.

8. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 7, characterized in that: Construct a comprehensive damage index Used to comprehensively reflect the first The extent of area expansion and the intensity of discrete distribution of wear-like regions: ； in: For the first The overall damage index of wear-like areas; and These are weighting coefficients; different weighting combinations are used for different wear types. For the first The relative area growth rate of wear-like regions; For the first Normalized instance density of wear-like regions.

9. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 8, characterized in that: In step 4, a wear mechanism determination threshold is set. When the difference between the comprehensive damage index corresponding to a certain wear type and the comprehensive damage index corresponding to each of the other wear types is greater than the wear mechanism determination threshold, If so, the wear mechanism corresponding to this wear type is determined to be the dominant wear mechanism; If the difference between the combined damage indices of two or more wear types is less than the wear mechanism determination threshold... If the corresponding wear mechanisms play a dominant role, then the wear scar is classified as a composite wear state.

10. The intelligent assessment method for four-ball wear scars based on deep learning according to claim 9, characterized in that: Step 5 specifically involves, for the first sample in the sample set... Comprehensive damage index of wear-like areas Calculate its first quartile. and the third and fourth quartiles ,in, For the first The first quartile of the comprehensive damage index for wear-like areas. For the first The third quartile of the comprehensive damage index for wear-like areas; Based on the first quartile and the third quartile, the 1st quartile... The wear degree of wear-like areas is divided into slight wear, moderate wear, and severe wear: 。

Images