A method, system and device for detecting the protrusion amount of a bearing by image vision recognition

Abstract

The present invention relates to a method, system and device for detecting the protrusion amount of a bearing in image vision recognition, belonging to the field of image processing. The target component model of the target device model on which the ball screw support bearing is installed is obtained; taking the target device model and the target component model as scene constraints, a same-scene sample group that meets the scene constraints is collected, and the preload concentration value and the first bearing protrusion amount interval are statistically calculated; through a predefined bearing protrusion amount configuration function, the preload concentration value is processed to obtain a bearing protrusion amount confidence interval; according to the bearing protrusion amount confidence interval, an intersection interval is extracted to obtain a second bearing protrusion amount interval; the bearing protrusion amount detection value of the bearing installation captured image is extracted; when the bearing protrusion amount detection value does not belong to the second bearing protrusion amount interval, the bearing is abnormally marked according to the captured image, solving the technical problems of inaccurate bearing protrusion amount detection data and insufficient adaptability to the scene.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and particularly to a method, system and device for detecting the protrusion amount of a bearing by image vision recognition. Background Art

[0002] The ball screw support bearing is a key transmission component widely used in machine tools and precision machinery. Its main function is to efficiently convert rotational motion into linear motion, or convert torque into axial repetitive force. The ball screw support bearing consists of a screw, a nut, a reverse device and balls. Through the rolling of the balls between the screw and the nut, a low-friction and high-precision transmission effect is achieved. Due to its high precision, reversibility and high efficiency, the ball screw support bearing is widely used in various industrial equipment and precision instruments, such as numerical control machine tools, precision measuring equipment, robots, etc. In these applications, the ball screw support bearing not only undertakes the transmission task, but also directly affects the positioning accuracy, transmission speed and stability of the equipment.

[0003] During the installation and use of the ball screw support bearing, the protrusion amount of the bearing, that is, the relative height difference between the end face of the inner ring and the end face of the outer ring of the bearing, is an important parameter. Accurate control of the protrusion amount of the bearing can ensure the adaptability and normal operation of the bearing, thus guaranteeing the accuracy and stability of the entire transmission system. However, at present, the ball screw support bearing is installed according to the recommended protrusion amount of the bearing. This standardized installation method lacks a quantitative standard closely combined with the application scenario, resulting in frequent mismatch of the protrusion amount of the bearing in actual applications, affecting the performance and service life of the transmission system. In order to accurately detect the protrusion amount of the bearing, traditional methods mainly use contact detection equipment. However, this detection method has some problems. On the one hand, the contact detection equipment is prone to wear during use, especially when a large number of detections are carried out, and the wear situation is particularly serious. This will lead to inaccurate detection data, thus affecting the accurate evaluation of the protrusion amount of the bearing. On the other hand, most of the existing contact detection equipment relies on imports, with high costs and complex maintenance, making it difficult to meet the needs of enterprises for a large number of detections. Therefore, developing a high-precision and non-contact method for detecting the protrusion amount of the bearing is of great significance for improving the adaptability and normal operation ability of the ball screw support bearing and guaranteeing the accuracy and stability of the transmission system. Summary of the Invention

[0004] The present invention aims at the technical problems of inaccurate detection data of the protrusion amount of the bearing and insufficient adaptability to the scenario in the prior art, and provides a method, system and device for detecting the protrusion amount of the bearing by image vision recognition to solve.

[0005] The technical solution of the present invention to solve the above technical problems is as follows:

[0006] In a first aspect, the present invention provides a method for detecting the protruding amount of a bearing through image vision recognition, including: obtaining the target component model of the target equipment model on which a ball screw support bearing is installed; taking the target equipment model and the target component model as scene constraints, collecting a group of same-scene samples that meet the scene constraints, and statistically analyzing the concentrated preload value and the first bearing protruding amount interval; through a predefined bearing protruding amount configuration function, processing the concentrated preload value to obtain a bearing protruding amount confidence interval; extracting an intersection interval based on the first bearing protruding amount interval and the bearing protruding amount confidence interval to obtain a second bearing protruding amount interval; extracting the bearing protruding amount detection value of the bearing installation captured image; when the bearing protruding amount detection value does not belong to the second bearing protruding amount interval, performing an abnormal identification of the bearing protruding amount on the bearing according to the captured image.

[0007] In a second aspect, the present invention provides a system for detecting the protruding amount of a bearing through image vision recognition. The system includes: a model identification module for obtaining the target component model of the target equipment model on which a ball screw support bearing is installed; a collection and statistics module for taking the target equipment model and the target component model as scene constraints, collecting a group of same-scene samples that meet the scene constraints, and statistically analyzing the concentrated preload value and the first bearing protruding amount interval; a numerical processing module for processing the concentrated preload value through a predefined bearing protruding amount configuration function to obtain a bearing protruding amount confidence interval; an interval extraction module for extracting an intersection interval based on the first bearing protruding amount interval and the bearing protruding amount confidence interval to obtain a second bearing protruding amount interval; a numerical extraction module for extracting the bearing protruding amount detection value of the bearing installation captured image; an abnormal identification module for performing an abnormal identification of the bearing protruding amount on the bearing according to the captured image when the bearing protruding amount detection value does not belong to the second bearing protruding amount interval.

[0008] In a third aspect, the present invention provides an electronic device, including: a memory for storing a computer software program; a processor for reading and executing the computer software program, thereby implementing a method for detecting the protruding amount of a bearing through image vision recognition as described in the first aspect.

[0009] The beneficial effects of the present invention are as follows: By combining the target equipment model and the target component model for sample group analysis under scene constraints, and using a predefined configuration function to process the preload data to obtain a confidence interval, and finally by comparing the bearing protruding amount detection value obtained through image recognition with the confidence interval, accurate recognition and identification of abnormal bearing protruding amounts are achieved, and the technical effects of improving detection accuracy and efficiency are achieved. Description of the Drawings

[0010] Figure 1 It is a schematic flowchart of a method for detecting the protruding amount of a bearing through image vision recognition provided by the present invention.

[0011] Figure 2 This is a schematic structural diagram of a bearing protrusion amount detection system for image visual recognition provided by the present invention.

[0012] Figure 3 This is a schematic structural diagram of an electronic device provided by the present invention;

[0013] Figure 4 This is a diagram showing the training effects of replacing different loss functions provided by the present invention.

[0014] Explanation of reference numerals: model recognition module 11, acquisition and statistics module 12, numerical processing module 13, interval extraction module 14, numerical extraction module 15, anomaly identification module 16, electronic device 500, memory 510, processor 520, first computer program 511. Detailed implementation manners

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0016] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined.

[0017] In the description of the present invention, the term "for example" is used to mean "serving as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily construed as being more preferred or having more advantages than other embodiments. In order for any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for purposes of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope that conforms to the principles and features disclosed in the present invention. Embodiment

[0018] Such as Figure 1As shown in the figure, an embodiment of the present invention provides a method for detecting the protrusion amount of a bearing based on image vision recognition, including:

[0019] S10: Obtain the target component model of the target device model where the ball screw support bearing is installed.

[0020] S20: Taking the target device model and the target component model as scene constraints, collect a group of same-scene samples that meet the scene constraints, and statistically analyze the concentrated preload value and the first bearing protrusion amount interval.

[0021] S30: Process the concentrated preload value through a predefined bearing protrusion amount configuration function to obtain the bearing protrusion amount confidence interval.

[0022] S40: Extract the intersection interval based on the first bearing protrusion amount interval and the bearing protrusion amount confidence interval to obtain the second bearing protrusion amount interval.

[0023] S50: Extract the bearing protrusion amount detection value of the bearing installation captured image.

[0024] S60: When the bearing protrusion amount detection value does not belong to the second bearing protrusion amount interval, mark the bearing with an abnormal protrusion amount according to the captured image.

[0025] Exemplarily, the bearing protrusion amount refers to the relative height difference between the end face of the inner ring and the end face of the outer ring of the bearing. After applying a preload to a single bearing, the distance that the end face of the inner ring protrudes from the end face of the outer ring at the same end face of the bearing can ensure the adaptability and normal operation of the bearing. When detecting the bearing protrusion amount based on image vision recognition technology, first, clarify the target to be detected, that is, obtain the target device model where the ball screw support bearing is installed and its corresponding target component model. The target device model refers to the specific model of the machine or device where the ball screw support bearing is installed, which determines the working environment and force conditions of the bearing; while the target component model refers to the specific model of the ball screw support bearing that matches the device, which affects key factors such as the size, performance, and installation method of the bearing. By clarifying these two models, accurate constraint conditions can be provided for subsequent analysis of the bearing protrusion amount based on the specific application scenario, ensuring the accuracy and practicality of the detection results.

[0026] Furthermore, when dealing with the detection task of the protrusion amount of the ball screw support bearing for the target device model and the target component model, this solution adopts a more refined and scientific method rather than directly detecting the protrusion amount of the bearing. The reason is that the protrusion amount of the bearing is not determined by a single factor, and one of its underlying elements is the preload. In the actual application scenario, the preload and the axial stiffness of the bearing act together on the protrusion amount of the bearing, but their relationship is not strictly linear and is affected by various complex factors. Therefore, if only the protrusion amount of the bearing is directly detected without considering the quantification of the preload, the detection result may be interfered by unclear factors, thus affecting the accuracy. To more accurately reflect the actual situation of the protrusion amount of the bearing, a large number of same-scenario sample groups are first collected according to the scenario constraints of the target device model and the target component model to ensure the pertinence and accuracy of subsequent analysis. Specifically, the same-scenario sample groups that meet these scenario constraints are collected, and these sample groups contain the historical data of the ball screw support bearings installed under similar or the same application conditions. By analyzing these samples, the concentrated value of the preload is statistically obtained, and this value reflects the magnitude of the preload that the ball screw support bearing usually needs to be applied to achieve the ideal working state in a specific scenario. At the same time, the first bearing protrusion amount interval is statistically obtained, that is, the distribution of the protrusion amount of the bearing in these samples. The determination of this interval helps to understand the reasonable variation range of the protrusion amount of the bearing under the given preload. Next, the concentrated value of the preload is processed using a predefined bearing protrusion amount configuration function to obtain the theoretically confidence interval of the bearing protrusion amount. The predefined bearing protrusion amount configuration function is a mathematical tool constructed based on a large amount of experimental data and theoretical models, and it can reflect the complex relationship between the preload and the protrusion amount of the bearing. Subsequently, the concentrated value of the preload statistically obtained before is sent into this configuration function as an input parameter, and the function uses the built-in algorithms and logics to perform a series of calculations and processes on the concentrated value of the preload, and finally outputs a confidence interval of the protrusion amount of the bearing. This confidence interval represents the reasonable range within which the protrusion amount of the bearing may fall under the given preload condition, and it provides a scientific and reliable reference standard for judging whether the protrusion amount of the bearing is in a healthy state. This process is equivalent to starting from the underlying element (i.e., the preload) and predicting the possible range of the protrusion amount of the bearing through a mathematical model. Then, the intersection operation is performed between the theoretically confidence interval and the statistically obtained first bearing protrusion amount interval, and the resulting result takes into account both the distribution characteristics of the actual data and the accuracy of the theoretical prediction, so it is more representative. In summary, by starting the analysis from the underlying element of the preload and combining statistical data and theoretical prediction, the healthy range of the protrusion amount of the bearing can be more accurately determined, thus ensuring the accuracy and reliability of the detection result.

[0027] Specifically, the confidence intervals are obtained by processing the preload concentration values through a predefined configuration function, and they each represent the reasonable boundaries within which the bearing protrusion amount may vary under different scenarios or conditions. Subsequently, an extraction operation of the intersection interval is performed on these first bearing protrusion amount intervals and the confidence intervals, which involves finding the part commonly covered by all the given intervals, that is, the overlapping area between them. Through this process, a narrower and more accurate second bearing protrusion amount interval is obtained, which comprehensively represents the most reasonable range of the bearing protrusion amount under all considered factors. The determination of the second bearing protrusion amount interval provides a more stringent basis for judging the health status of the bearing protrusion amount, contributing to improving the stability and reliability of the equipment.

[0028] Finally, the actual state of the bearing protrusion amount is determined by combining image vision recognition technology. A clear image of the bearing installation position is taken by a high-precision image acquisition device. Subsequently, an advanced image processing algorithm is used to extract the detected value of the bearing protrusion amount from the captured image, and this process realizes non-contact and high-precision measurement of the bearing protrusion amount. Immediately afterwards, this detected value is compared with the second bearing protrusion amount interval obtained previously. If the detected value falls outside this interval, it means that the bearing protrusion amount does not meet the preset health standard. At this time, the corresponding bearing is marked with an abnormal protrusion amount based on the captured image. This step not only records the abnormal state of the bearing, but also provides valuable visual evidence for subsequent problem tracing and analysis through the retention of image information. Through this series of processes, abnormal situations of the bearing protrusion amount can be identified in a timely and accurate manner, providing strong support for the maintenance and management of the equipment.

[0029] In a preferred embodiment, with the target equipment model and the target component model as scene constraints, a same-scene sample group that meets the scene constraints is collected, and the preload concentration value and the first bearing protrusion amount interval are statistically analyzed, including: retrieving the first bearing historical installation data of healthy installation samples whose ball screw support bearing model meets the scene constraints, where the first bearing historical installation data includes a set of installation preload record values and a set of bearing protrusion amount detection record values; performing a central tendency analysis on the set of installation preload record values to obtain the preload concentration value; and performing an outlier deletion on the set of bearing protrusion amount detection record values to obtain the first bearing protrusion amount interval.

[0030] Optionally, specific target device and component models are used as scenario constraints to ensure the targetedness and relevance of subsequent analysis. Based on this, a sample group meeting these scenario constraints is collected. This sample group contains historical data for ball screw support bearings installed under similar or identical application conditions. Specifically, historical first bearing installation data for healthy ball screw support bearing models meeting the scenario constraints is retrieved. This data provides detailed information on the preload force and bearing protrusion detection during installation. After acquiring this historical data, a central tendency analysis is performed on the set of installation preload force records. This step calculates statistical quantities such as the mean, median, or mode to determine the central value of the preload force, which reflects the typical preload force required for the ball screw support bearing to achieve ideal operating conditions in a given scenario. Furthermore, outlier removal is performed on the set of bearing protrusion detection records to eliminate extreme values caused by measurement error, abnormal operation, or other atypical factors, thereby obtaining a more realistic and reliable first bearing protrusion range. This range provides a reasonable range of variation for bearing protrusion under normal installation conditions.

[0031] In a preferred embodiment, performing outlier removal on the set of bearing protrusion detection record values to obtain the first bearing protrusion interval includes: sorting the bearing protrusion detection record values in ascending order to obtain a bearing protrusion detection record value sorting result; extracting a first bearing protrusion detection record value with a sequence number that is rounded up by one-fourth of the bearing protrusion detection record value sorting result; and extracting a second bearing protrusion detection record value with a sequence number that is rounded down by three-fourths of the bearing protrusion detection record value sorting result. Calculate the bearing protrusion deviation modulus of the first bearing protrusion detection record value and the second bearing protrusion detection record value; use the first bearing protrusion detection record value minus 1.5 times the bearing protrusion deviation modulus to obtain the bearing protrusion lower limit value, and use the second bearing protrusion detection record value plus 1.5 times the bearing protrusion deviation modulus to obtain the bearing protrusion upper limit value; delete the outlier bearing protrusion detection record values that are less than the bearing protrusion lower limit value or greater than the bearing protrusion upper limit value from the bearing protrusion detection record value set to obtain the first bearing protrusion interval of the retained bearing protrusion detection record value distribution.

[0032] Specifically, when processing the protruding amount data of the ball screw support bearing, in order to ensure the accuracy and practicality of the analysis results, a strict outlier deletion process is adopted to determine the first bearing protruding amount interval. This process begins with sorting the detected record values of the bearing protruding amount, that is, arranging all the record values in ascending order to obtain an ordered sorting result of the detected record values of the bearing protruding amount. Subsequently, key data points are extracted from the sorting result. Specifically, the first detected record value of the bearing protruding amount corresponding to the rounded-up serial number at the one-fourth position after sorting is found, and the second detected record value of the bearing protruding amount corresponding to the rounded-down serial number at the three-fourths position after sorting is found. These two data points respectively represent the preliminary estimates of the lower and upper bounds of the bearing protruding amount distribution. Next, the deviation modulus value between these two data points is calculated, that is, their absolute difference, which reflects the range size of the bearing protruding amount distribution. Based on this deviation modulus value, the upper and lower limit values of the bearing protruding amount are further determined. Specifically, the lower limit value is obtained by subtracting 1.5 times the deviation modulus value from the first detected record value of the bearing protruding amount, and the upper limit value is obtained by adding 1.5 times the deviation modulus value to the second detected record value of the bearing protruding amount. These two boundary values constitute the benchmark for judging whether the bearing protruding amount is abnormal. Finally, all the detected record values of the bearing protruding amount are traversed, and those record values that are less than the lower limit value or greater than the upper limit value are regarded as outliers and deleted. After this step, a distribution interval composed of the remaining detected record values of the bearing protruding amount is obtained, that is, the first bearing protruding amount interval. This interval provides an accurate description of the reasonable variation range of the bearing protruding amount under normal circumstances. Taking the protruding amount data of a certain type of ball screw support bearing as an example, assume there is a set of 100 detected record values. According to the above process, these values are first sorted, and then the 25th (rounded up at the one-fourth position) and 75th (rounded down at the three-fourths position) data points are extracted as the preliminary boundaries. After calculating the deviation modulus value between these two points, the lower limit value and the upper limit value are obtained, and 5 outliers are deleted accordingly. Finally, a first bearing protruding amount interval composed of 95 remaining record values is obtained, which provides strong data support for subsequent analysis and decision-making.

[0033] In a preferred embodiment, the pre-tightening force concentration value is processed through a predefined bearing protruding amount configuration function to obtain a bearing protruding amount confidence interval, including: obtaining the second bearing historical installation data of the ball screw support bearing model, where the second bearing historical installation data includes pre-tightening force record data and bearing protruding amount identification interval record data; constructing a bearing protruding amount configuration loss function according to the physical law loss term and the data-driven loss term; based on the bearing protruding amount configuration loss function, using the bearing protruding amount identification interval record data as supervision and the pre-tightening force record data as input, training the bearing protruding amount configuration function through a BP neural network.

[0034] Preferably, in order to accurately predict the bearing protrusion amount and determine its confidence interval, it is carried out by combining physical laws and data-driven methods. This process first depends on obtaining the second bearing historical installation data of a specific ball screw support bearing model, which contains rich preload records and corresponding bearing protrusion amount identification interval records, providing a basis for subsequent model training. Next, a bearing protrusion amount configuration loss function is constructed, which combines a physical law loss term and a data-driven loss term. The physical law loss term is based on an in-depth understanding of the mechanical characteristics of the bearing, which ensures that the model prediction results conform to basic physical laws. The data-driven loss term uses the actual protrusion amount identification interval in the historical installation data to supervise the training of the model, enabling the model to learn the potential laws and patterns in the data. After having the loss function, a BP (backpropagation) neural network is used to train the bearing protrusion amount configuration function. In this process, the preload record data is used as the input of the neural network, and the bearing protrusion amount identification interval record data is used as the supervision information to guide the weight adjustment of the neural network. Through multiple iterative trainings, the neural network gradually learns how to predict a reasonable bearing protrusion amount and its confidence interval according to the input preload value. Taking a certain model of ball screw support bearing as an example, hundreds of historical installation records may be collected, and each record contains the preload and the corresponding bearing protrusion amount identification interval. Using these data, a loss function is constructed and trained multiple times through a BP neural network. Finally, a configuration function that can accurately predict the bearing protrusion amount and its confidence interval is obtained. This function not only considers physical laws but also makes full use of the information in historical data, providing a powerful tool for the installation and performance evaluation of ball screw support bearings.

[0035] In a preferred embodiment, according to the physical law loss term and the data-driven loss term, a bearing protrusion amount configuration loss function is constructed, including: constructing the physical law loss term: ; wherein, represents the physical law loss value of the i-th training, represents the predicted bearing protrusion amount confidence interval of the i-th training, represents the maximum value of the predicted bearing protrusion amount confidence interval of the i-th training, represents the minimum value of the predicted bearing protrusion amount confidence interval of the i-th training, represents the preload record data of the i-th training, and b represent hyperparameters, w represents a weight parameter, b represents a bias parameter, represents the bearing axial stiffness, is a bearing protrusion amount calculation function obtained by empirical fitting; constructing the data-driven loss term: ; wherein, The data-driven loss value characterizing the i-th training The bearing protrusion amount identification interval record data characterizing the i-th training Characterize the intersection-over-union ratio threshold; construct the bearing protrusion amount configuration loss function: ; where The loss value after every N trainings is characterized Characterize the weight of the physical law loss term, and N characterizes the number of training times of the preset statistical loss value

[0036] Furthermore, construct a physical law loss term, which evaluates the confidence interval of the predicted bearing protrusion amount based on physical laws and mechanical characteristics. Specifically, for each training, calculate the maximum and minimum values of the confidence interval of the predicted bearing protrusion amount, and evaluate the degree of compliance with physical laws according to the preload record data and hyperparameters (such as weight parameter w and bias parameter b) in the model. This loss value reflects the consistency between the prediction result and the physical law. The specific formula for the physical law loss term is: , Characterize the median value of the bearing protrusion amount prediction interval Characterize the predicted value of the bearing protrusion amount fitted according to experience. By Although the predicted value of the bearing protrusion amount obtained by fitting has deviations, it is mainly the fluctuation value generated around a certain accurate value, so it also has reference value. Therefore, the physical law loss term characterizes the deviation between the current model prediction value and the predicted value obtained by . If it is too large, it is abnormal. Introducing the physical law loss term aims to limit the output to conform to physical laws, thereby improving the convergence speed

[0037] Such as Figure 4 As shown, for replacing different loss functions in the simulation code, the experimental result data of the convergence speed and accuracy are obtained. Among them, the double loss function refers to the LOSS function mentioned in the embodiments of the present invention: :

[0038] Next, construct a data-driven loss term, which uses the bearing protrusion amount identification interval record data in the historical installation data to supervise the training of the model. Specifically, calculate the intersection-over-union of the prediction result and the true identification interval, and compare it with a preset intersection-over-union ratio threshold to obtain the data-driven loss value. This loss value reflects the degree of coincidence between the prediction result and the actual data. The specific formula for the data-driven loss term is: ; where The data-driven loss value characterizing the i-th training The bearing protrusion amount identification interval record data characterizing the i-th training Characterize the intersection ratio threshold. Finally, combine the physical law loss term and the data-driven loss term to construct a bearing protrusion amount configuration loss function. This function comprehensively considers the constraints of physical laws and data-driven. By adjusting the weight of the physical law loss term and the number of training times of the preset statistical loss value, the weights between the two can be balanced, so as to obtain a loss function that conforms to physical laws and is close to actual data. After every N training times, calculate the loss value and update the model parameters according to it to optimize the prediction performance. The specific formula of the bearing protrusion amount configuration loss function is: ; where characterizes the loss value after every N training times, characterizes the weight of the physical law loss term, and N characterizes the number of training times of the preset statistical loss value. To sum up, the loss function construction method combines the advantages of physical laws and data-driven, providing an accurate and reliable framework for predicting the protrusion amount of ball screw support bearings. By continuous iterative training and optimization, a model that can accurately predict the protrusion amount of the bearing and its confidence interval can be obtained, providing strong support for the installation and performance evaluation of the bearing.

[0039] In a preferred embodiment, based on the bearing protrusion amount configuration loss function, with the data recorded in the bearing protrusion amount identification interval as the supervision and the pre-tightening force recorded data as the input, train the bearing protrusion amount configuration function through a BP neural network, including: initialize the number of neurons in the hidden layer of the BP neural network to 1, based on the bearing protrusion amount configuration loss function, with the data recorded in the bearing protrusion amount identification interval as the supervision and the pre-tightening force recorded data as the input, train the first bearing protrusion amount configuration function through a BP neural network; when the validation loss value of the first bearing protrusion amount configuration function for a continuous preset number of times is greater than or equal to the convergence loss threshold, increase the number of neurons by 1 and execute a loop until the validation loss value of the Mth bearing protrusion amount configuration function for a continuous preset number of times is less than the convergence loss threshold, and set the Mth bearing protrusion amount configuration function as the bearing protrusion amount configuration function.

[0040] Exemplarily, an iterative training method based on a BP (backpropagation) neural network is adopted. This method combines a bearing protrusion amount configuration loss function to optimize the bearing protrusion amount configuration function. The training process starts with initializing the number of hidden layer neurons in the BP neural network to 1, which is a relatively simple starting point and helps to gradually explore the complexity of the model. Next, using the data recorded in the bearing protrusion amount identification interval as supervision information and the data recorded in the preload as input, the first bearing protrusion amount configuration function is trained through the BP neural network. During the training process, the bearing protrusion amount configuration loss function is continuously calculated. This loss function comprehensively considers physical laws and data-driven constraints, providing a clear direction for the optimization of the model. To evaluate the performance of the model, the validation loss value is calculated after each training and compared with a preset convergence loss threshold. If the validation loss value of the first bearing protrusion amount configuration function for a continuous preset number of times is greater than or equal to the convergence loss threshold, it means that the performance of the current model has not reached the expected standard, and the complexity of the neural network needs to be increased to further improve the prediction ability. Therefore, it enters a loop process. In each loop, the number of hidden layer neurons is increased by 1 and the bearing protrusion amount configuration function is retrained. This iterative process continues until the validation loss value of a certain bearing protrusion amount configuration function (denoted as the Mth bearing protrusion amount configuration function) for a continuous preset number of times is less than the convergence loss threshold. At this time, it is considered that the model has reached sufficient complexity and the prediction performance has reached the expected standard. Finally, this Mth bearing protrusion amount configuration function is set as the final bearing protrusion amount configuration function for subsequent bearing protrusion amount prediction tasks. Through gradually increasing the complexity of the neural network and combining the optimization of the bearing protrusion amount configuration loss function throughout the training process, an accurate and reliable bearing protrusion amount prediction model is achieved.

[0041] In a preferred embodiment, extracting the bearing protrusion amount detection value of the bearing installation captured image includes: collecting the bearing installation captured record image, wherein the bearing installation captured record image has the bearing inner ring end face identification position and the bearing outer ring end face identification position; using the bearing inner ring end face identification position as supervision and the bearing installation captured record image as input, constructing a bearing inner ring end face position extraction channel based on a convolutional neural network; using the bearing outer ring end face identification position as supervision and the bearing installation captured record image as input, constructing a bearing outer ring end face position extraction channel based on a convolutional neural network; configuring an end face relative Euclidean distance calculation layer; connecting the output layers of the bearing inner ring end face position extraction channel and the bearing outer ring end face position extraction channel to the input layer of the end face relative Euclidean distance calculation layer to obtain a bearing protrusion amount detection model, and processing the bearing installation captured image to obtain the bearing protrusion amount detection value.

[0042] Specifically, bearing installation captured images containing the identification positions of the inner ring end face and the outer ring end face of the bearing are collected. These images provide rich visual information and serve as the basis for subsequent position extraction and protrusion amount calculation. Next, a position extraction channel for the inner ring end face of the bearing and a position extraction channel for the outer ring end face of the bearing are respectively constructed based on a convolutional neural network. These two channels use the identification positions of the inner ring end face and the outer ring end face of the bearing as supervision information and the bearing installation captured images as input. Through training, these two channels can accurately extract the position information of the inner and outer ring end faces of the bearing from the images. To calculate the bearing protrusion amount, an end face relative Euclidean distance calculation layer is configured. This calculation layer is responsible for receiving the output information from the position extraction channel of the inner ring end face of the bearing and the position extraction channel of the outer ring end face of the bearing, and calculating the Euclidean distance between the two. This distance value is the required bearing protrusion amount detection value. Finally, the output layers of the position extraction channel of the inner ring end face of the bearing and the position extraction channel of the outer ring end face of the bearing are connected to the input layer of the end face relative Euclidean distance calculation layer, thereby constructing a complete bearing protrusion amount detection model. This model can process the input bearing installation captured images and automatically output the bearing protrusion amount detection value. In summary, an automated and high-precision bearing protrusion amount detection process is achieved by combining a convolutional neural network and Euclidean distance calculation. This process not only improves the detection efficiency but also ensures the accuracy and reliability of the detection results.

[0043] The bearing protrusion amount detection method based on image vision recognition provided by the embodiment of the present invention has at least the following technical effects:

[0044] 1. By introducing the target device model and the target component model as scene constraints, the pre-tightening force concentration value and the first bearing protrusion amount interval of the same-scene sample group of the ball screw support bearing model are analyzed. This prediction method based on scene constraints can more accurately reflect the bearing protrusion amount characteristics in the actual installation environment, improving the accuracy and practicality of the prediction.

[0045] 2. By constructing a physical law loss term and a data-driven loss term, and constructing a bearing protrusion amount configuration loss function based on these loss terms. Then, using a BP neural network for adaptive training, the optimal bearing protrusion amount configuration function is found by gradually increasing the number of neurons in the hidden layer. This adaptive training method can fully consider the dual constraints of physical laws and actual data, improving the generalization ability and prediction accuracy of the model.

[0046] 3. A position extraction channel for the end face of the bearing inner ring and the end face of the outer ring is constructed using a convolutional neural network, and the bearing protrusion amount is detected by calculating the Euclidean distance between the two. This detection method based on the convolutional neural network can automatically extract key position information from the captured image and perform accurate distance calculation, realizing the rapid and accurate detection of the bearing protrusion amount. At the same time, this method also has high robustness and adaptability, and can handle image detection tasks under different shooting angles and lighting conditions. Embodiment

[0047] As Figure 2 shown, based on the same inventive concept as the bearing protrusion amount detection method of image vision recognition provided in Embodiment 1, the embodiment of the present invention also provides an image vision recognition bearing protrusion amount detection system, and the system includes:

[0048] A model identification module 11, configured to obtain a target component model of the target device model on which the ball screw support bearing is installed.

[0049] An acquisition and statistics module 12, configured to take the target device model and the target component model as scene constraints, acquire a same-scene sample group that meets the scene constraints, and statistically analyze the preload concentration value and the first bearing protrusion amount interval.

[0050] A numerical processing module 13, configured to process the preload concentration value through a predefined bearing protrusion amount configuration function to obtain a bearing protrusion amount confidence interval.

[0051] An interval extraction module 14, configured to perform intersection interval extraction according to the first bearing protrusion amount interval and the bearing protrusion amount confidence interval to obtain a second bearing protrusion amount interval.

[0052] A numerical extraction module 15, configured to extract the bearing protrusion amount detection value of the bearing installation captured image.

[0053] An abnormal identification module 16, configured to perform abnormal identification of the bearing protrusion amount on the bearing according to the captured image when the bearing protrusion amount detection value does not belong to the second bearing protrusion amount interval.

[0054] Furthermore, the acquisition and statistics module 12 is further configured to perform the following steps:

[0055] Retrieve the first bearing historical installation data of the healthy installation samples whose ball screw support bearing models meet the scene constraints, where the first bearing historical installation data includes a set of installation preload record values and a set of bearing protrusion amount detection record values; perform central tendency analysis on the set of installation preload record values to obtain the preload concentration value; perform outlier deletion on the set of bearing protrusion amount detection record values to obtain the first bearing protrusion amount interval.

[0056] Furthermore, the acquisition and statistics module 12 is further configured to perform the following steps:

[0057] Sort the detected and recorded values of the bearing protrusion amount in ascending order to obtain the sorting result of the detected and recorded values of the bearing protrusion amount; extract the first detected and recorded value of the bearing protrusion amount at the rounded-up serial number of one-fourth of the sorting result of the detected and recorded values of the bearing protrusion amount; extract the second detected and recorded value of the bearing protrusion amount at the rounded-down serial number of three-fourths of the sorting result of the detected and recorded values of the bearing protrusion amount; calculate the modulus of the bearing protrusion amount deviation between the first detected and recorded value of the bearing protrusion amount and the second detected and recorded value of the bearing protrusion amount; use the first detected and recorded value of the bearing protrusion amount minus 1.5 times the modulus of the bearing protrusion amount deviation to obtain the lower limit value of the bearing protrusion amount, and use the second detected and recorded value of the bearing protrusion amount plus 1.5 times the modulus of the bearing protrusion amount deviation to obtain the upper limit value of the bearing protrusion amount; delete the outlier detected and recorded values of the bearing protrusion amount that are less than the lower limit value of the bearing protrusion amount or greater than the upper limit value of the bearing protrusion amount from the set of detected and recorded values of the bearing protrusion amount to obtain the first bearing protrusion amount interval of the remaining detected and recorded values of the bearing protrusion amount distribution.

[0058] Furthermore, the numerical processing module 13 is further configured to perform the following steps:

[0059] Obtain the second bearing historical installation data of the ball screw support bearing model, where the second bearing historical installation data includes preload record data and bearing protrusion amount identification interval record data; construct a bearing protrusion amount configuration loss function according to the physical law loss term and the data-driven loss term; based on the bearing protrusion amount configuration loss function, with the bearing protrusion amount identification interval record data as the supervision and the preload record data as the input, train the bearing protrusion amount configuration function through a BP neural network.

[0060] Furthermore, the numerical processing module 13 is further configured to perform the following steps:

[0061] Construct the physical law loss term: ; where represents the physical law loss value of the i-th training, represents the predicted bearing protrusion amount confidence interval of the i-th training, represents the maximum value of the predicted bearing protrusion amount confidence interval of the i-th training, represents the minimum value of the predicted bearing protrusion amount confidence interval of the i-th training, represents the preload record data of the i-th training, and b represent hyperparameters, w represents a weight parameter, and b represents a bias parameter; construct the data-driven loss term: ; where represents the data-driven loss value of the i-th training represents the recorded data of the bearing protrusion amount identification interval for the i-th training represents the intersection-over-union ratio threshold; construct the bearing protrusion amount configuration loss function: ; where represents the loss value after every N trainings represents the weight of the physical law loss term, and N represents the preset number of training times for the statistical loss value.

[0062] Furthermore, the numerical processing module 13 is further configured to perform the following steps:

[0063] Initialize the number of neurons in the hidden layer of the BP neural network to 1. Based on the bearing protrusion amount configuration loss function, using the recorded data of the bearing protrusion amount identification interval as supervision and the recorded data of the pre-tightening force as input, train the first bearing protrusion amount configuration function through the BP neural network; when the verification loss value of the first bearing protrusion amount configuration function for a continuous preset number of times is greater than or equal to the convergence loss threshold, increase the number of neurons by 1 and execute the loop until the verification loss value of the M-th bearing protrusion amount configuration function for a continuous preset number of times is less than the convergence loss threshold, and set the M-th bearing protrusion amount configuration function as the bearing protrusion amount configuration function.

[0064] Furthermore, the numerical extraction module 15 is further configured to perform the following steps:

[0065] Collect the bearing installation photographed record image, where the bearing installation photographed record image has the bearing inner ring end face identification position and the bearing outer ring end face identification position; using the bearing inner ring end face identification position as supervision and the bearing installation photographed record image as input, construct the bearing inner ring end face position extraction channel based on the convolutional neural network; using the bearing outer ring end face identification position as supervision and the bearing installation photographed record image as input, construct the bearing outer ring end face position extraction channel based on the convolutional neural network; configure the end face relative Euclidean distance calculation layer; connect the output layers of the bearing inner ring end face position extraction channel and the bearing outer ring end face position extraction channel to the input layer of the end face relative Euclidean distance calculation layer to obtain the bearing protrusion amount detection model, and process the bearing installation photographed image to obtain the bearing protrusion amount detection value.

[0066] Through the foregoing detailed description of a method for detecting the bearing protrusion amount by image vision recognition in this specification, those skilled in the art can clearly know a system for detecting the bearing protrusion amount by image vision recognition in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description in the method part. Embodiment

[0067] See also Figure 3 , Figure 3 Schematic diagram of an embodiment of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 520, and a first computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the first computer program 511, a bearing protrusion detection method based on image visual recognition as described in Example 1 is implemented.

[0068] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0069] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware.

[0070] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0071] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0072] Although preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts become known.

[0073] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the present invention and its equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A method for detecting the protrusion amount of a bearing in image visual recognition, characterized in that, Including: Obtaining a target component model of the target device model for installing a ball screw support bearing; Taking the target device model and the target component model as scenario constraints, collecting a same-scenario sample group that meets the scenario constraints, and statistically analyzing the preload concentration value and the first bearing protrusion amount interval; Processing the preload concentration value through a predefined bearing protrusion amount configuration function to obtain a bearing protrusion amount confidence interval; wherein, processing the preload concentration value through a predefined bearing protrusion amount configuration function to obtain a bearing protrusion amount confidence interval includes: Obtaining second bearing historical installation data of the ball screw support bearing model, where the second bearing historical installation data includes preload record data and bearing protrusion amount identification interval record data; Constructing a bearing protrusion amount configuration loss function according to a physical law loss term and a data-driven loss term; Based on the bearing protrusion amount configuration loss function, using the bearing protrusion amount identification interval record data as supervision and the preload record data as input, training the bearing protrusion amount configuration function through a BP neural network; Performing intersection interval extraction according to the first bearing protrusion amount interval and the bearing protrusion amount confidence interval to obtain a second bearing protrusion amount interval; Extracting a bearing protrusion amount detection value of a bearing installation captured image; When the bearing protrusion amount detection value does not belong to the second bearing protrusion amount interval, performing an abnormal identification of the bearing protrusion amount on the bearing according to the captured image.

2. The method according to claim 1, wherein Taking the target device model and the target component model as scenario constraints, collecting a same-scenario sample group that meets the scenario constraints, and statistically analyzing the preload concentration value and the first bearing protrusion amount interval, including: Retrieving first bearing historical installation data of healthy installation samples of the ball screw support bearing model that meet the scenario constraints, where the first bearing historical installation data includes a set of installation preload record values and a set of bearing protrusion amount detection record values; Performing a central tendency analysis on the set of installation preload record values to obtain the preload concentration value; Performing outlier deletion on the set of bearing protrusion amount detection record values to obtain the first bearing protrusion amount interval.

3. The method according to claim 2, wherein Performing outlier deletion on the set of bearing protrusion amount detection record values to obtain the first bearing protrusion amount interval, including: Sorting the bearing protrusion amount detection record values in ascending order to obtain a sorting result of the bearing protrusion amount detection record values; Extracting a first bearing protrusion amount detection record value at the ceiling of one-fourth of the sorting result of the bearing protrusion amount detection record values; Extracting a second bearing protrusion amount detection record value of the bearing protrusion amount at the floor of three-fourths of the sorting result of the bearing protrusion amount detection record values; Calculating a bearing protrusion amount deviation modulus value between the first bearing protrusion amount detection record value and the second bearing protrusion amount detection record value; Subtracting 1.5 times the bearing protrusion amount deviation modulus value from the first bearing protrusion amount detection record value to obtain a bearing protrusion amount lower limit value, and adding 1.5 times the bearing protrusion amount deviation modulus value to the second bearing protrusion amount detection record value to obtain a bearing protrusion amount upper limit value; Delete the outlier bearing protrusion amount detection record values that are less than the lower limit value of the bearing protrusion amount or greater than the upper limit value of the bearing protrusion amount from the set of bearing protrusion amount detection record values, and obtain the first bearing protrusion amount interval in which the remaining bearing protrusion amount detection record values are distributed.

4. The method according to claim 1, characterized in that Construct a bearing protrusion amount configuration loss function based on the physical law loss term and the data-driven loss term, including: Construct the physical law loss term: ; Among them, represents the physical law loss value of the i-th training, represents the confidence interval of the predicted bearing protrusion amount of the i-th training, represents the maximum value of the confidence interval of the predicted bearing protrusion amount of the i-th training, represents the minimum value of the confidence interval of the predicted bearing protrusion amount of the i-th training, represents the preload force recorded data of the i-th training, w and b represent hyperparameters, w represents the weight parameter, b represents the bias parameter, and k represents the bearing axial stiffness, is the bearing protrusion amount calculation function obtained by empirical fitting; Construct the data-driven loss term: ； Among them, represents the data-driven loss value of the i-th training, represents the bearing protrusion amount identification interval record data of the i-th training, represents the intersection-over-union ratio threshold; Construct the bearing protrusion amount configuration loss function: ； Among them, represents the loss value after every N trainings, represents the weight of the physical law loss term, and N represents the number of training times of the preset statistical loss value.

5. The method according to claim 4, characterized in that, Based on the bearing protrusion amount configuration loss function, using the data recorded in the bearing protrusion amount identification interval as supervision and the preload force recorded data as input, train the bearing protrusion amount configuration function through a BP neural network, including: Initialize the number of neurons in the hidden layer of the BP neural network to 1. Based on the bearing protrusion amount configuration loss function, using the data recorded in the bearing protrusion amount identification interval as supervision and the preload force recorded data as input, train the first bearing protrusion amount configuration function through a BP neural network; When the verification loss value of the first bearing protrusion amount configuration function for a continuous preset number of times is greater than or equal to the convergence loss threshold, increase the number of neurons by 1 and execute a loop until the verification loss value of the Mth bearing protrusion amount configuration function for a continuous preset number of times is less than the convergence loss threshold, and set the Mth bearing protrusion amount configuration function as the bearing protrusion amount configuration function.

6. The method according to claim 1, wherein Extract the bearing protrusion amount detection value of the bearing installation captured image, including: Collect the bearing installation captured record image, where the bearing installation captured record image has the identification positions of the inner ring end face of the bearing and the identification position of the outer ring end face of the bearing; Based on the identification position of the inner ring end face of the bearing as supervision and the bearing installation captured record image as input, construct a bearing inner ring end face position extraction channel based on a convolutional neural network; Based on the identification position of the outer ring end face of the bearing as supervision and the bearing installation captured record image as input, construct a bearing outer ring end face position extraction channel based on a convolutional neural network; Configure an end face relative Euclidean distance calculation layer; Connect the output layers of the bearing inner ring end face position extraction channel and the bearing outer ring end face position extraction channel to the input layer of the end face relative Euclidean distance calculation layer to obtain a bearing protrusion amount detection model, process the bearing installation captured image, and obtain the bearing protrusion amount detection value.

7. A bearing protrusion amount detection system for image visual recognition, characterized in that, A system for implementing the method for detecting the bearing protrusion amount by image vision recognition according to any one of claims 1-6, the system includes: A model identification module for obtaining the target component model of the target device model on which the ball screw support bearing is installed; A collection and statistics module for taking the target device model and the target component model as scene constraints, collecting a group of same-scene samples that meet the scene constraints, and statistically calculating the preload force concentration value and the first bearing protrusion amount interval; A numerical processing module for processing the preload force concentration value through a predefined bearing protrusion amount configuration function to obtain a bearing protrusion amount confidence interval; An interval extraction module for extracting an intersection interval according to the first bearing protrusion amount interval and the bearing protrusion amount confidence interval to obtain a second bearing protrusion amount interval; A numerical extraction module extracts the bearing protrusion amount detection value of the bearing installation captured image; An abnormal identification module is used to abnormally identify the bearing protrusion amount according to the captured image when the bearing protrusion amount detection value does not belong to the second bearing protrusion amount interval.

8. An electronic device, characterized in that, It includes: A memory for storing computer software programs; A processor for reading and executing the computer software program, thereby implementing the method for detecting the bearing protrusion amount by image vision recognition according to any one of claims 1-7.

Images