Automated testing line for multidimensional parameters of rolling bearings and method for evaluating the grade of rolling bearings

Abstract

This invention proposes an automated testing line for multi-dimensional parameters of rolling bearings and a method for evaluating the grade of rolling bearings. The testing line includes a control system, a conveyor belt device connected to the control system, and a robotic arm device located outside the conveyor belt device. Along the length of the conveyor belt device are modules for testing rolling bearing stiffness, radial clearance, and vibration characteristics. By incorporating the conveyor belt device and robotic arm device, this invention can automatically detect parameters such as radial and axial support stiffness, radial clearance, and vibration characteristics of rolling bearings. In actual production, this invention can replace some single-function bearing performance testing devices, reducing or eliminating the need for manufacturers to construct additional rolling bearing testing and screening lines. This invention can promptly identify and screen out unqualified bearings, helping companies improve product quality and achieve the effects of reducing costs and increasing production efficiency.

Description

Technical Field

[0001] This invention belongs to the field of bearing testing technology, specifically relating to an automated testing line for multi-dimensional parameters of rolling bearings and a method for evaluating the grade of rolling bearings. Background Technology

[0002] Rolling bearings are critical components in rotating machinery, and their quality and performance have a significant impact on the reliability, safety, and service performance of the machinery. To ensure the stable and reliable operation of mechanical products, designers and manufacturers must test key parameters such as radial clearance, support stiffness, and vibration characteristics of rolling bearings before installation, and select bearings that meet the relevant performance indicators.

[0003] Currently, most testing equipment can only detect a single parameter. For example, patents CN202210906441.2 and CN201310304120.6 can only detect the radial clearance of rolling bearings, patent CN202110389138.5 can only detect the stiffness of rolling bearings, and patents CN202111575216.7 and CN201410519880.3 can only detect the vibration characteristics of rolling bearings. However, in actual engineering scenarios, rolling bearings in mechanical products often require the detection and evaluation of multiple parameters.

[0004] To meet the requirements of large-scale production, relevant enterprises must purchase a large number of testing equipment. However, different types of testing equipment generally come from different manufacturers, resulting in low levels of collaboration between these testing devices. Production units need to build additional rolling bearing testing production lines to enable these devices to work together, which undoubtedly consumes a lot of manpower, material resources, and financial resources, leading to increased production costs and relatively low production efficiency. Moreover, since the testing parameters of each testing device are different, manual multi-dimensional comprehensive evaluation of whether the rolling bearing is qualified is required, which is a large workload. Summary of the Invention

[0005] The present invention aims to solve the technical problems existing in the prior art. The first objective of the present invention is to provide an automated detection line for multi-dimensional parameters of rolling bearings. The second objective of the present invention is to provide a method for evaluating the grade of rolling bearings using the aforementioned detection line.

[0006] To achieve the first objective mentioned above, the present invention adopts the following technical solution: an automated multi-dimensional parameter testing line for rolling bearings, comprising a control system, a conveyor belt device for transporting rolling bearings connected to the control system, and a robotic arm device located outside the conveyor belt device. Multiple testing modules, each connected to the control system, are also arranged along the length of the conveyor belt device. These testing modules include a rolling bearing stiffness testing module, a rolling bearing radial clearance testing module, and a rolling bearing vibration characteristic testing module. A bearing support is mounted on the conveyor belt device, allowing the rolling bearing under test to be placed on the bearing support and to move along with the bearing support on the conveyor belt device. The robotic arm device is used to grip the rolling bearing under test located on the bearing support and send it into each test module for testing. After the test is completed, the rolling bearing under test is removed from each test module and placed back on the bearing support of the conveyor belt device. The control system controls the rolling bearing stiffness test module, rolling bearing radial clearance test module and rolling bearing vibration characteristic test module to perform stiffness testing, radial clearance testing and vibration characteristic testing on the rolling bearing under test. The control system acquires the radial stiffness parameters, axial stiffness parameters, radial clearance parameters, vibration velocity parameters and vibration acceleration parameters of the rolling bearing under test.

[0007] The above-described technical solution, by incorporating a conveyor belt device and a robotic arm device, can automatically detect key parameters of rolling bearings, such as radial and axial support stiffness, radial clearance, and vibration characteristics. In actual production operations, this invention can replace some single-function bearing performance testing devices, enabling manufacturing enterprises to reduce or eliminate the need for additional rolling bearing testing lines.

[0008] In a preferred embodiment of the present invention, the radial clearance testing module detects the radial clearance of each tested rolling bearing at three different angular positions of 0°, 120° and 240°, and takes the average value as the radial clearance of the rolling bearing.

[0009] In a preferred embodiment of the present invention, the inspection line further includes a laser marking device connected to the control system at the beginning of the conveyor belt device. The control system has the rolling bearing number entered into it, and the laser marking device marks the outer ring surface of the rolling bearings of the batch to be inspected with the corresponding mark and number according to the rolling bearing number entered into the control system.

[0010] The above technical solution uses a laser marking device to mark corresponding marks and numbers on the outer ring surface of the rolling bearings in the batch to be inspected, thereby ensuring that the detection data acquired in the control system corresponds to the rolling bearings under inspection.

[0011] In a preferred embodiment of the present invention, the control system includes a hardware device control module, a detection data storage module, and a detection data analysis module. The hardware device control module controls the operating state of the detection line and expands the functionality of the detection line by modifying the program in the hardware device control module. The detection data storage module automatically saves the stiffness test results, radial clearance test results, and vibration characteristic test results of the tested rolling bearings, and can display and view the stiffness test results, radial clearance test results, and vibration characteristic test results of the rolling bearings. The detection data analysis module performs stiffness analysis, vibration analysis, and statistical analysis on the rolling bearings. The stiffness analysis can display the values ​​of the radial and axial loads applied to the rolling bearings during the test by the rolling bearing stiffness test module, as well as the corresponding radial and axial displacement curves of the bearings, and calculates the radial and axial stiffness of the rolling bearings based on the relationship between load and displacement. The vibration analysis calculates the vibration velocity characteristics and vibration acceleration characteristics of the tested rolling bearings based on the data collected by the rolling bearing vibration characteristic test module. The statistical analysis function organizes the measurement data of the rolling bearings with specified numbers and can evaluate the accuracy of the test data.

[0012] In a preferred embodiment of the present invention, the control system includes a rolling bearing evaluation module, which is capable of reading information from the module, evaluating the rolling bearing under test based on the information, and screening and grouping the rolling bearings based on the evaluation results.

[0013] The above technical solution, by setting up a rolling bearing evaluation module, facilitates the screening and grouping of multiple rolling bearings, enabling manufacturers to reduce or eliminate the need to build additional rolling bearing screening production lines.

[0014] To achieve the second objective mentioned above, the present invention adopts the following technical solution: a rolling bearing grade evaluation method, which utilizes the detection line provided by the present invention. The rolling bearing evaluation module includes a first machine learning model and a second machine learning model. The grade evaluation method includes the following steps.

[0015] 1) Obtain raw sample data as training data, prepare rolling bearings of different grades in advance, and detect their axial stiffness, radial stiffness, radial clearance, vibration velocity and vibration acceleration through the detection line;

[0016] 2) Data preprocessing: Calculate the root mean square value (v) of the vibration velocity in the low-frequency band based on the vibration velocity collected in the previous step. LRMS The root mean square value of vibration velocity v in the intermediate frequency band MRMS and the root mean square value of vibration velocity v in the high frequency band HRMS Combine it with axial stiffness K a Radial stiffness K r Radial clearance G rConcatenate into a single feature vector

[0017] v features =[K a ,K r G r ,v LRMS ,v MRMS ,v HRMS The vibration acceleration data was then subjected to maximum and minimum normalization, and a Fourier transform was performed to obtain its frequency domain representation. The amplitude corresponding to the frequency range of 0-5000Hz was denoted as v. a And perform max-min normalization on it;

[0018] 3) Transfer the feature vector v features Input the first machine learning model, and the first machine learning model will process the feature vector v. features Process it;

[0019] 4) The vibration acceleration amplitude v a The signal is input into a second machine learning model, which extracts signal features from the frequency domain signal of the vibration acceleration.

[0020] 5) Let the output of the first machine learning model be o1, the output of the second machine learning model be o2, and the true result corresponding to the training data be y.

[0021] Calculate the difference L1 between the output of the first machine learning model and the actual result, and the difference L2 between the output of the second machine learning model and the actual result;

[0022]

[0023]

[0024] Where m is the label of the level, n is the total number of levels, and o1(m) represents the probability value of the output result o1 at level m. o2(m) represents the probability value of the output result o2 at level m, and

[0025] 6) When the network converges or reaches the set maximum number of training rounds, training is terminated and the model is saved. The saved model is then encapsulated into the rolling bearing evaluation module. After the multi-dimensional parameter detection of the rolling bearing is completed, this module automatically reads the corresponding data from the detection data storage module and inputs it into the machine learning model to obtain the grade evaluation result of the inspected rolling bearing.

[0026] In the above technical solution, the input to the first machine learning model is the axial stiffness K of the rolling bearing under test. a Radial stiffness K rRadial clearance G r The root mean square value of vibration velocity in the low-frequency band, v LRMS The root mean square value of vibration velocity v in the mid-frequency band MRMS The root mean square value of vibration velocity v in the high frequency band HRMS The eigenvector v features The first machine learning model is used to process the feature vector v features The input to the second machine learning model is the frequency domain amplitude v of the vibration acceleration of the tested rolling bearing. a The second machine learning model is used to extract signal features from the frequency domain signal of vibration acceleration. The rolling bearing evaluation module of this invention employs two machine learning models that process two different data streams respectively, effectively utilizing the various detection information of the rolling bearings detected by the detection line of this invention. This invention can promptly identify and screen out unqualified inspected bearings, helping enterprises improve product quality levels, reduce costs, and increase production efficiency. Furthermore, the rolling bearing evaluation module of this invention is based on a deep learning model. Before evaluating the grade of the inspected rolling bearings, the deep learning model needs to be trained according to steps 1) to 5) to learn the task pattern related to rolling bearing evaluation.

[0027] In another preferred embodiment of the present invention, in step 5), the rolling bearing evaluation module adopts the following weighting rule:

[0028] o=w1o1+w2o2

[0029] L = |L1 - L2|

[0030]

[0031] Where o represents the final weighted sum of the outputs of the two machine learning models, w1 and w2 represent the weights assigned to o1 and o2 respectively, L represents the distance between L1 and L2, min(L1,L2) represents the smaller of L1 and L2, and max(L1,L2) represents the larger of L1 and L2. This indicates that o is assigned the corresponding smaller difference value. i The weight,

[0032] This indicates that o is assigned a larger difference. i The weight.

[0033] The above technical solution, by designing weighting rules, assigns greater weight to models whose output differs more from the actual results, and vice versa, so that the model's output is closer to the actual results.

[0034] In another preferred embodiment of the invention, in step 5), if L is greater than twice the smaller of L1 and L2, then the output of the i-th model with the smallest difference is assigned a weight of 1, and it is used as the final output, while the output of the model with the larger difference is ignored; if L1 and L2 are equal, then the outputs of the two models contribute equally to the final result, w1 = w2 = 1 / 2; in other cases, using and Calculate the weights assigned to each model output, and constrain the output weights so that w1 + w2 = 1;

[0035] Considering the distance L between L1 and L2, such that m i The value can be dynamically adjusted according to the changes in L1 and L2, so that a smaller difference corresponds to a larger weight and a larger difference corresponds to a smaller weight.

[0036] The above technical solution analyzes multiple scenarios to achieve the effect of smaller differences corresponding to larger weights and larger differences corresponding to smaller weights.

[0037] In another preferred embodiment of the present invention, in step 3), the first machine learning model consists of two fully connected layers followed by a ReLU activation layer, and one fully connected layer followed by a Softmax activation layer.

[0038] In another preferred embodiment of the present invention, the second machine learning model contains four convolutional blocks, denoted as ①, ②, ③, and ④, respectively; wherein convolutional blocks ①, ③, and ④ are composed of a stack of one-dimensional convolutional layers, batch normalization layers, ReLU activation layers, and max pooling layers, and convolutional block ② is composed of a stack of one-dimensional convolutional layers, batch normalization layers, ReLU activation layers, and average pooling layers, and v a The output after inputting convolution block ① is The output after inputting convolution block ② is The output after inputting convolution block ③ is Will As input to convolutional block ④, convolutional block ④ is connected to a classifier network, which consists of one flattened layer, one randomly deactivated layer, two fully connected layers followed by a ReLU activation layer, and one fully connected layer followed by a Softmax activation layer.

[0039] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0040] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0041] Figure 1 This is a schematic diagram of the structure of an automated detection line for multidimensional parameters of rolling bearings according to Embodiment 1.

[0042] Figure 2 This is a schematic diagram of the model structure of the rolling bearing evaluation module in Embodiment 2.

[0043] The reference numerals in the accompanying drawings include: control system 1, conveyor belt device 2, bearing support seat 21, robotic arm device 3, rolling bearing stiffness testing module 4, rolling bearing radial clearance testing module 5, rolling bearing vibration characteristic testing module 6, and laser marking device 7. Detailed Implementation

[0044] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0045] In the description of this invention, it should be understood that the terms "longitudinal", "lateral", "vertical", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0046] In the description of this invention, unless otherwise specified and limited, it should be noted that the terms "installation", "connection" and "linking" should be interpreted broadly. For example, they can refer to mechanical or electrical connections, or internal connections between two components. They can be direct connections or indirect connections through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0047] Example 1

[0048] This embodiment provides an automated detection line for multi-dimensional parameters of rolling bearings, such as... Figure 1As shown, in a preferred embodiment, the testing line includes a control system 1, a conveyor belt device 2 for conveying rolling bearings connected to the control system 1, and a robotic arm device 3 located outside the conveyor belt device 2. Multiple test modules connected to the control system 1 are also provided along the length direction of the conveyor belt device 2. The test modules include a rolling bearing stiffness test module 4, a rolling bearing radial clearance test module 5, and a rolling bearing vibration characteristic test module 6 arranged sequentially along the length direction of the conveyor belt device 2.

[0049] The conveyor belt device 2 is equipped with a bearing support seat 21, on which the rolling bearing under test can be placed and moved along with the bearing support seat 21 on the conveyor belt device 2. The robotic arm device 3 is used to grip the rolling bearing under test located on the bearing support seat 21, send it to each test module for testing, and after the testing is completed, remove the rolling bearing under test from each test module and put it back on the bearing support seat 21 of the conveyor belt device 2.

[0050] Control system 1 controls the operation of rolling bearing stiffness testing module 4, rolling bearing radial clearance testing module 5 and rolling bearing vibration characteristic testing module 6 to perform stiffness testing, radial clearance testing and vibration characteristic testing on the rolling bearing under test. Control system 1 acquires the radial stiffness parameters, axial stiffness parameters, radial clearance parameters, vibration velocity parameters and vibration acceleration parameters of the rolling bearing under test.

[0051] In this invention, the function of the rolling bearing stiffness testing module 4 is to detect the radial stiffness and axial stiffness of the rolling bearing. Both the radial stiffness test and the axial stiffness test are performed by applying a certain load to the rolling bearing through a loading device, and at the same time, the elastic deformation of the bearing under this load is obtained through a displacement sensor. Then, the stiffness value of the bearing is obtained according to the relationship between the applied force and the elastic deformation.

[0052] In this invention, the function of the rolling bearing radial clearance test module 5 is to detect the radial clearance of the rolling bearing. In order to detect the radial clearance of the bearing more accurately, the radial clearance test module of this invention detects the radial clearance of each bearing at three different angular positions of 0°, 120° and 240°, and takes the average value as the radial clearance of the bearing.

[0053] In this invention, the function of the rolling bearing vibration characteristic test module 6 is to measure the vibration velocity and vibration acceleration of the rolling bearing.

[0054] It should be noted that the rolling bearing stiffness test module 4, the rolling bearing radial clearance test module 5, and the rolling bearing vibration characteristic test module 6 can all adopt existing technologies and are not the innovations of this invention. Their structures and working principles will not be described in detail here.

[0055] In another preferred embodiment, the inspection line also includes a laser marking device 7 connected to the control system 1 at the beginning of the conveyor belt device 2. The control system 1 has the rolling bearing number entered into it. The laser marking device 7 marks the outer ring surface of the rolling bearings in the batch to be inspected with the corresponding mark and number according to the rolling bearing number entered into the control system 1.

[0056] In this invention, the control system 1 includes a hardware device control module, a detection data storage module, a detection data analysis module, and a rolling bearing evaluation module.

[0057] Specifically, the hardware device control module is the control center of the entire testing line. Operators can use this module to control the operating status of the testing line and expand the functionality of the testing line by modifying the program in the hardware device control module.

[0058] The test data storage module automatically saves the stiffness test results, radial clearance test results, vibration characteristic test results, and bearing grade evaluation results of the tested rolling bearings. It can also display and view these results. The vibration characteristic test results include the vibration velocity characteristics of the tested rolling bearings (including waveforms, spectrum diagrams, peak values, peak factor, and RMS values; RMS values, mean values, peak values, peak factor, pulse counts, and kurtosis for each frequency band) and vibration acceleration characteristics (including waveforms, RMS values, peak values, and peak factor). Operators can add, delete, export, or reset information in this test data storage module. Simultaneously, the rolling bearing evaluation module can read the information from this module and perform a grade evaluation of the tested rolling bearings based on this information.

[0059] The test data analysis module is used to perform stiffness analysis, vibration analysis, and statistical analysis on rolling bearings. Specifically, the stiffness analysis can display the values ​​of radial and axial loads applied to the rolling bearing during the test by the rolling bearing stiffness test module 4, as well as the corresponding radial and axial displacement curves of the bearing, and calculate the radial and axial stiffness of the rolling bearing based on the relationship between load and displacement; the vibration analysis calculates the vibration velocity characteristics and vibration acceleration characteristics of the tested rolling bearing based on the data collected by the rolling bearing vibration characteristic test module 6; the statistical analysis function organizes the measurement data of rolling bearings with specified numbers and can evaluate the accuracy of the test data.

[0060] The workflow of the automated multi-dimensional parameter detection line for rolling bearings in this embodiment is as follows:

[0061] 1) Place the rolling bearing to be inspected on the bearing support seat 21 at the beginning of the conveyor belt device 2.

[0062] 2) The conveyor belt device 2 operates, transporting the inspected rolling bearing to the work station of the laser marking device 7. The laser marking device 7 performs laser marking on the inspected rolling bearing according to the bearing number entered in the control system 1, thereby ensuring that the test data and the inspected rolling bearing can correspond to each other.

[0063] (3) After laser marking is completed, the conveyor belt device 2 transports the tested rolling bearing to the workstation of the rolling bearing stiffness testing module 4. The robotic arm device 3 places the rolling bearing in the rolling bearing stiffness testing module 4 for radial and axial stiffness testing. Preferably, by gradually increasing the radial load, the bearing radial stiffness under six different radial loads is finally obtained, and the final radial stiffness of the bearing is obtained by fitting these six radial stiffnesses; by gradually increasing the axial load, the bearing axial stiffness under six different axial loads is finally obtained, and the final axial stiffness of the bearing is obtained by fitting these six axial stiffnesses.

[0064] (4) After completing the rolling bearing stiffness test, the robotic arm device 3 places the rolling bearing on the bearing support 21 and transports it to the workstation of the rolling bearing radial clearance test module 5 via the conveyor belt device 2. The robotic arm device 3 then places the rolling bearing in the rolling bearing radial clearance test module 5 for radial clearance testing. Preferably, after completing one radial clearance measurement, the mandrel of the rolling bearing radial clearance test module 5 is rotated 120° clockwise to continue testing, ultimately obtaining the radial clearance of the rolling bearing at three angular positions: 0°, 120°, and 240°. The average value is taken as the final radial clearance of the rolling bearing.

[0065] (5) After completing the radial clearance test of the rolling bearing, the robotic arm device 3 places the rolling bearing on the bearing support 21 and transports it to the workstation of the rolling bearing vibration characteristic test module 6 via the conveyor belt device 2. Then, the robotic arm device 3 places the rolling bearing in the rolling bearing vibration characteristic test module 6 for rolling bearing vibration testing. Preferably, the vibration velocity and vibration acceleration characteristics of the rolling bearing are measured at low frequency (50-300Hz), medium frequency (300-1800Hz), and high frequency (1800-10000Hz).

[0066] (6) During the above testing process, after each test is completed, the results of the rolling bearing are automatically entered into the detection data storage module of the control system 1. The rolling bearing evaluation module performs an automated level evaluation of the tested rolling bearing based on the test results of each rolling bearing. Then, the robotic arm device 3 places the rolling bearing in the post-inspection sample placement area according to the evaluation results, selects the rolling bearings that meet the specified requirements, and removes the unqualified rolling bearings. Thus, a complete automatic detection, evaluation, and screening process for rolling bearings is completed.

[0067] Example 2

[0068] This embodiment provides a rolling bearing grade evaluation method using the automated multi-dimensional parameter detection line of rolling bearings in Embodiment 1. In this embodiment, the rolling bearing evaluation module is a module based on a deep learning model. The rolling bearing evaluation module includes two machine learning models: a first machine learning model and a second machine learning model, which can effectively utilize the various detection information of the rolling bearings detected by the detection line of this invention.

[0069] like Figure 2 As shown, ①, ②, ③, and ④ represent four convolutional blocks; "1D-CNN" represents a one-dimensional convolutional neural network layer; "BN" represents a batch normalization layer; "ReLU" represents a ReLU activation layer; "MaxPool" represents a max pooling layer; "AveragePool" represents an average pooling layer; and "Softmax" represents a softmax activation layer. The input to the first machine learning model is the axial stiffness K of the tested rolling bearing. a Radial stiffness K r Radial clearance G r The root mean square value of vibration velocity v in the low frequency (50-300Hz) band LRMS The root mean square value of vibration velocity v in the mid-frequency (300-1800Hz) band MRMS The root mean square value of vibration velocity v in the high frequency band (1800-10000Hz) HRMS The feature vector is composed of the frequency domain amplitude of the vibration acceleration of the tested rolling bearing (e.g., the amplitude corresponding to a frequency of 0-5000Hz).

[0070] The specific steps of the rolling bearing grade evaluation method in this embodiment are as follows:

[0071] 1) Obtain the original sample data as training data. Prepare rolling bearings of different grades in advance. For example, prepare several rolling bearings of grades A, B, C, and D. Detect their axial stiffness, radial stiffness, radial clearance, vibration velocity, and vibration acceleration through the detection line of this invention. Obtain the known axial stiffness, radial stiffness, radial clearance, vibration velocity, and vibration acceleration of the rolling bearings of grades A, B, C, and D.

[0072] 2) Data preprocessing: Calculate the root mean square value (v) of the vibration velocity in the low-frequency (50-300Hz) band based on the vibration velocity collected in the previous step. LRMS The root mean square value of vibration velocity v in the mid-frequency (300-1800Hz) band MRMS The root mean square value of vibration velocity v in the high frequency band (1800-10000Hz) HRMS Combine it with axial stiffness K a Radial stiffness K rRadial clearance G r Concatenate them into a single feature vector v features =

[0073] [K a ,K r G r ,v LRMS ,v MRMS ,v HRMS The data is then subjected to maximum and minimum normalization, the specific maximum and minimum normalization being a conventional technique in this field and not detailed here. A Fourier transform is performed on the vibration acceleration data to obtain its frequency domain representation, and the amplitude corresponding to frequencies from 0 to 5000 Hz is denoted as v. a And perform max-min normalization on it.

[0074] 3) Transfer the feature vector v features Input the first machine learning model, and the first machine learning model will process the feature vector v. features The first machine learning model consists of two fully connected layers followed by ReLU activation layers, and one fully connected layer followed by a Softmax activation layer.

[0075] 4) The vibration acceleration amplitude v a The input is fed into a second machine learning model, which extracts signal features from the frequency domain signal of vibration acceleration. The second machine learning model contains four convolutional blocks, denoted as ①, ②, ③, and ④. Convolutional blocks ①, ③, and ④ are composed of stacked one-dimensional convolutional layers, batch normalization layers, ReLU activation layers, and max pooling layers. Convolutional block ② is composed of stacked one-dimensional convolutional layers, batch normalization layers, ReLU activation layers, and average pooling layers. The input is then processed by the second machine learning model. a The output after inputting convolution block ① is The output after inputting convolution block ② is The output after inputting convolution block ③ is Will As input to convolutional block ④, convolutional block ④ is connected to a classifier network, which consists of one flattened layer, one randomly deactivated layer, two fully connected layers followed by a ReLU activation layer, and one fully connected layer followed by a Softmax activation layer.

[0076] 5) To obtain the best bearing evaluation results, it is necessary to comprehensively consider the output results of the two machine learning models. Let the output result of the first machine learning model be o1, the output result of the second machine learning model be o2, and the true result corresponding to the training data be y. Calculate the difference L1 between the output result of the first machine learning model and the true result, and the difference L2 between the output result of the second machine learning model and the true result.

[0077]

[0078]

[0079] Where m is the level label, n is the total number of levels, and in this embodiment, four levels A, B, C, and D are set as an example, then m = 1, 2, 3, and 4 represent the four levels A, B, C, and D respectively; o1(m) represents the probability value of the output result o1 at level m, and 1, o2(m) represents the probability value of the output result o2 at level m, and

[0080] The smaller the difference between the output of two machine learning models and the true result, the closer the model's output is to the true result, and therefore, a larger weight should be assigned to that model's output; conversely, a larger difference should be assigned a smaller weight. Preferably, the output of the two machine learning models should be considered using the following weighting rule:

[0081] o=w1o1+w2o2

[0082] L = |L1 - L2|

[0083]

[0084] Where o represents the final weighted sum of the outputs of the two machine learning models, w1 and w2 represent the weights assigned to o1 and o2 respectively, L represents the distance between L1 and L2, min(L1,L2) represents the smaller of L1 and L2, and max(L1,L2) represents the larger of L1 and L2. This indicates that o is assigned the corresponding smaller difference value. i The weight,

[0085] This indicates that o is assigned a larger difference. i The weight.

[0086] If L is greater than twice the smaller of L1 and L2, then the output of the i-th model with the smallest difference is weighted with 1, and this is used as the final output, while the output of the model with the larger difference is ignored. If L1 and L2 are equal, then the outputs of the two models contribute equally to the final result, w1 = w2 = 1 / 2. In other cases, the formula is used. and The weights assigned to each model output are calculated using a formula that takes into account the distance L between L1 and L2, such that m i The value can be dynamically adjusted according to changes in L1 and L2, achieving the effect of smaller differences corresponding to larger weights and larger differences corresponding to smaller weights. (Formula) Then the weights of the output are constrained so that w1 + w2 = 1.

[0087] 6) When the network converges or reaches the set maximum number of training rounds, training is terminated and the model is saved. The saved model is then encapsulated into the rolling bearing evaluation module. After the multi-dimensional parameter detection of the rolling bearing is completed, this module automatically reads the corresponding data from the detection data storage module and inputs it into the machine learning model to obtain the grade evaluation result of the inspected rolling bearing.

[0088] In the description of this specification, references to terms such as "preferred embodiment," "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0089] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A multi-dimensional parameter automatic detection line for rolling bearings, characterized in that, The control system, the conveyor belt device with the control system connected, the mechanical arm device arranged outside the conveyor belt device, a plurality of test modules respectively connected with the control system are arranged along the length direction of the conveyor belt device, the test modules include rolling bearing stiffness test module, rolling bearing radial clearance test module and rolling bearing vibration characteristic test module; The bearing support seat is arranged on the conveyor belt device, the rolling bearing under test can be arranged on the bearing support seat and can move with the bearing support seat on the conveyor belt device; The mechanical arm device is used for clamping the rolling bearing under test on the bearing support seat and sending it into each test module for detection, and after the detection is completed, the rolling bearing under test is taken out from each test module and placed on the bearing support seat of the conveyor belt device; The control system controls the rolling bearing stiffness test module, the rolling bearing radial clearance test module and the rolling bearing vibration characteristic test module to work to detect the stiffness, the radial clearance and the vibration characteristic of the rolling bearing under test, and obtains the radial stiffness parameter, the axial stiffness parameter, the radial clearance parameter, the vibration velocity parameter and the vibration acceleration parameter of the rolling bearing under test; The control system includes a hardware device control module, a detection data storage module and a detection data analysis module; The hardware device control module is used for controlling the running state of the detection line, and the function expansion of the detection line is realized by modifying the program in the hardware device control module; The detection data storage module is used for automatically saving the stiffness detection result, the radial clearance detection result and the vibration characteristic detection result of the rolling bearing under test, and can display and view the stiffness detection result, the radial clearance detection result and the vibration characteristic detection result of the rolling bearing under test; The detection data analysis module is used for stiffness analysis, vibration analysis and statistical analysis of the rolling bearing; the stiffness analysis can display the values of the radial load and the axial load applied on the rolling bearing in the rolling bearing stiffness test module test process, and the corresponding bearing radial displacement and axial displacement curves, and calculate the radial stiffness and the axial stiffness of the rolling bearing according to the relationship between the load and the displacement; the vibration analysis calculates the vibration velocity characteristic and the vibration acceleration characteristic of the rolling bearing under test according to the data collected by the rolling bearing vibration characteristic test module; The statistical analysis function realizes the arrangement of the measurement data of the rolling bearing with the specified number, and can evaluate the accuracy of the test data.

2. The multi-dimensional parameter automatic detection line of the rolling bearing according to claim 1, characterized in that, The radial clearance test module detects the radial clearance of each rolling bearing under test at three different angle positions of 0°, 120° and 240°, and takes the average value as the radial clearance of the rolling bearing.

3. The rolling bearing multi-dimensional parameter automatic detection line according to claim 1, characterized in that, The laser marking device connected with the control system is arranged at the starting end of the conveyor belt device, the rolling bearing number is recorded in the control system, and the laser marking device marks the corresponding mark and number on the outer ring surface of the rolling bearing under test according to the rolling bearing number recorded in the control system.

4. The multi-dimensional parameter automatic detection line of the rolling bearing according to any one of claims 1-3, characterized in that, The control system comprises a rolling bearing evaluation module capable of reading information in the module, evaluating the rolling bearing under test according to the information, and screening and grouping the rolling bearing according to the evaluation result.

5. The rolling bearing grade evaluation method using the rolling bearing multi-dimensional parameter automated inspection line according to claim 4, characterized by The rolling bearing evaluation module comprises a first machine learning model and a second machine learning model, and the grade evaluation method comprises the following steps: 1) Obtain original sample data as training data, prepare rolling bearings of different grades in advance, and detect the axial stiffness, radial stiffness, radial clearance, vibration velocity and vibration acceleration of the rolling bearings through the detection line; 2) Data preprocessing, calculate the root mean square value of the vibration speed in the low frequency band by the vibration speed collected in the previous step the root mean square value of the vibration speed in the medium frequency band and the root mean square value of the vibration speed in the high frequency band , and splice it into a feature vector with the axial stiffness , the radial stiffness , the radial clearance , and perform maximum and minimum normalization processing, and perform Fourier transform on the vibration acceleration data to obtain its frequency domain representation, take the amplitude value corresponding to the frequency of 0-5000Hz, and record it as , and perform maximum and minimum normalization processing;​ 3) inputting the feature vector into a first machine learning model processing the feature vector by the first machine learning model ; 4) inputting the vibration acceleration amplitude to a second machine learning model, from which the second machine learning model extracts signal features from the frequency domain signal of the vibration acceleration; 5) the output of the first machine learning model is , the output of the second machine learning model is , the true result corresponding to the training data is , the difference between the output of the first machine learning model and the true result is calculated , and the difference between the output of the second machine learning model and the true result is calculated ; ， ， wherein, the label of the class, n is the total number of classes, denotes the output result the probability value on the class , and , denotes the output result the probability value on the class , and ; 6) Terminate the training and save the model when the network converges or reaches the set maximum training round, encapsulate the saved model into the rolling bearing evaluation module, and automatically read the corresponding data from the detection data storage module and input the machine learning model to obtain the grade evaluation result of the rolling bearing under test after the multi-dimensional parameter detection of the rolling bearing is completed.

6. The rolling bearing grade evaluation method according to claim 5, characterized in that, In step 5), the rolling bearing evaluation module adopts the following weighting rules: ， ， ， wherein, is a final result after weighting the output results of the two machine learning models, and respectively represent weights given to and represents a distance between and represents one of and with a smaller median value, represents one of and with a larger median value, represents a weight given to corresponding to a smaller difference value, represents a weight given to corresponding to a larger difference value.​​ 7. The rolling bearing grade evaluation method according to claim 6, characterized in that, In step 5), if More than twice and The one with the smaller median value is then the one with the smallest difference. The output of each model is assigned a weight of 1, and that is used as the final output, while the output of the model with the larger difference is ignored; If and are equal, the contribution of the output of the 2 models to the final result is equally important, ; In other cases, the use of and weights assigned to each model output result, the weights of the outputs are constrained so that ; The distance between and is considered , so that the value can be dynamically adjusted according to and changes, realizing the effect of smaller difference corresponding to larger weight and larger difference corresponding to smaller weight.

8. The rolling bearing grade evaluation method according to claim 5, characterized in that, In step 3), the first machine learning model is composed of two fully connected layers and the subsequent ReLU activation layer, one fully connected layer and the subsequent Softmax activation layer.

9. The rolling bearing grade evaluation method according to claim 5, characterized by, In step 3), the second machine learning model contains four convolutional blocks, which are represented by ①, ②, ③ and ④ respectively. Among them, convolutional blocks ①, ③, and ④ are composed of stacked one-dimensional convolutional layers, batch normalization layers, ReLU activation layers, and max pooling layers, while convolutional block ② is composed of stacked one-dimensional convolutional layers, batch normalization layers, ReLU activation layers, and average pooling layers. The output after inputting convolution block ① is , The output after inputting convolution block ② is , The output after inputting convolution block ③ is ,Will As input to convolutional block ④, convolutional block ④ is connected to a classifier network, which consists of one flattened layer, one randomly deactivated layer, two fully connected layers followed by a ReLU activation layer, and one fully connected layer followed by a Softmax activation layer.

Images