A method for predicting the lubrication state of low-friction lubrication systems

Abstract

This application relates to the field of friction and lubrication technology, specifically to a method for predicting the lubrication state of low-friction lubrication systems. The method includes: collecting lubricating fluid and multiple material samples, and conducting routine low-friction coefficient tests; calculating the surface roughness of the material sample to be determined; performing a low-friction coefficient recovery test on the material sample to be determined; the low-friction coefficient recovery test tests the friction coefficient by changing the previously applied load and speed; if the low-friction coefficient of the material sample to be determined is lower than 0.02, the material sample is considered reversible; if the low-friction coefficient of the material sample to be determined is higher than 0.02, the material sample is considered failed; inputting the surface roughness, material sample parameters, speed, and load into a failure state judgment model for training; collecting a new dataset and inputting it into the failure state judgment model to predict the probability of the lubrication state of the material sample under low friction. This invention achieves accurate determination of the lubrication state under low friction.

Description

Technical Field

[0001] This application relates to the field of friction and lubrication technology, and specifically to a method for predicting the lubrication state of low-friction lubrication systems. Background Technology

[0002] In current technologies, the lubrication status of conventional lubrication systems can be reflected by vibration signals, temperature, and friction coefficient, but the lubrication status of low-friction lubrication systems cannot be determined. Because the friction coefficient of low-friction lubrication systems is less than 0.02, their lubrication status cannot be identified using conventional methods (such as temperature rise, vibration, and noise). Furthermore, low-friction systems exhibit both reversible and complete failure states, making it impossible to determine system failure based on a friction coefficient greater than 0.02. Inability to accurately determine the lubrication failure mode of such systems leads to excessive waste of lubricating oil, and excessive replacement of corresponding lubrication components also results in economic losses. Therefore, there is an urgent need to propose a method for predicting the failure modes of low-friction systems.

[0003] For example, a Chinese patent, publication number CN106483032A, publication date March 8, 2017, patent title "A method for achieving ultra-low friction in water-lubricated pairs", the specific technical solution of which is as follows: The invention discloses a method for achieving ultra-low friction in water-lubricated pairs, the specific steps of which are: (1) According to the actual operating conditions of the parts, set the same load and speed for initial pre-running, run for 300 seconds under water lubrication condition F; then unload and stop the machine. (1) Separate stationary parts to remove abrasive particles generated between friction interfaces during the pre-running stage; (2) Continue to run under actual operating conditions for 300 seconds; then run under overload conditions T at 2-3 times the actual operating load for 600 seconds; finally, continue to increase the load to 4-5 times the actual operating load and continue to run under overload conditions for 600 seconds; (3) Unload and stop the machine, ultrasonically clean the sample surface, clean the chamber, replace with new water, and completely eliminate the influence of wear particles.

[0004] Although the aforementioned patent achieves low friction by adjusting load and speed to break in parts, the above solution still cannot predict the lubrication state when the workpiece's coefficient of friction is less than 0.02. Summary of the Invention To address the problems existing in the prior art, the present invention provides a method for predicting the lubrication state of low-friction lubrication systems, which can predict the lubrication state when the coefficient of friction is less than 0.02, thereby avoiding increased operating costs due to incorrect state judgment.

[0005] To achieve the above-mentioned technical effects, the technical solution of this application is as follows: A method for predicting the lubrication state of a low-friction lubrication system includes the following steps: S1: Collect lubricating fluid and multiple different material samples to form a material sample sample, and conduct routine low friction coefficient tests. Apply load and speed to the material sample to obtain the friction coefficient of the material, and define the valid material sample and the material sample to be determined. S2: Calculate the roughness within the scratched area of ​​the material sample to be determined, and use it as the surface roughness of the material sample; S3: Perform a low friction coefficient recovery test on the material sample to be determined; the low friction coefficient recovery test tests the friction coefficient of the material sample to be determined by changing the previously applied load and speed. S4: After low friction coefficient recovery test, if the low friction coefficient of the material sample to be determined is lower than 0.02, the material sample is defined as reversible; if the low friction coefficient of the material sample to be determined is higher than 0.02, the material sample is defined as failed. S5: The surface roughness of the material sample, the defined material sample data, and the velocity and load corresponding to each material sample data are used as a dataset and input into the failure state judgment model built using deep learning for training. S6: Collect a new dataset and input it into the trained failure state judgment model to finally predict the probability of the lubrication state of the material sample under low friction.

[0006] Furthermore, in step S1, samples with a friction coefficient lower than 0.02 are defined as valid material samples, and samples with a friction coefficient greater than 0.02 are defined as undetermined material samples.

[0007] Furthermore, the specific steps for conducting a low-friction coefficient routine test in step S1, which involves applying load and velocity to the material sample to obtain the friction coefficient of the material, are as follows: Step a1: Prepare the rolling structure and substrate from the materials of the two different contact bodies respectively; Step a2: Fix the base on the moving platform of the cylinder, apply lubricant to the bottom of the rolling structure, apply force F1 to the rolling structure, and apply force F1 to the base through the rolling structure; Step a3: By rotating the motion platform around the center, the frictional force F2 between the rolling structure and the base is obtained, and thus the coefficient of friction Cof is obtained.

[0008] Furthermore, the specific formula for calculating the friction coefficient Cof is as follows: .

[0009] Furthermore, the specific method steps of step S2 are as follows: Step b1: Randomly select at least four locations in the scratched area and measure the roughness of the entire area along the cross-section of the scratch using a measuring instrument; Step b2: Take the measurement range A of the scratch center position as the accurate value of roughness; the measurement range A of the scratch center position is fifty percent of the scratch cross-sectional length; Step b3: Based on the obtained precise values ​​of multiple roughnesses, calculate the average of the multiple precise values ​​of the roughnesses, and use it as the surface roughness of the material sample.

[0010] Furthermore, the specific method for the low friction coefficient recovery test in step S3 is as follows: with the speed of the rolling structure remaining constant, the low friction coefficient recovery test is performed by increasing or decreasing the load, or with the load of the rolling structure remaining constant, the low friction coefficient recovery test is performed by increasing or decreasing the speed.

[0011] Furthermore, the load is increased by 1.2 to 3 times the load in the previous step S1; the load is decreased by one-third to five-sixths of the load in the previous step S1; the speed is increased by 1.5 to 9 times the speed in the previous step S1; and the speed is decreased by one-ninth to two-thirds of the speed in step S1.

[0012] Furthermore, the failure state judgment model uses Keras in deep learning to build a neural network, determine the network loss function and optimizer; preprocesses the input dataset to obtain tensors of speed, load and lubrication state. The input features of the failure state judgment model are speed and load, and the output feature is lubrication state.

[0013] Furthermore, the dataset is divided into a training set, a test set, and a validation set.

[0014] Furthermore, the neural network comprises an activation function and three linear layers; the number of neurons in the three linear layers are 64, 32, and 3, respectively; the activation function includes a ReLU activation function and a Softmax layer; the network loss function is cross-entropy L, and the optimizer is ADAM.

[0015] Furthermore, the specific expression for the cross-entropy L is as follows: ; In the formula, One-hot encoding of the real label. To predict the category of the model The probability of.

[0016] Furthermore, the neural network is initialized using Python's numpy.random function, and then iteratively trained using the training set until the maximum number of iterations is reached. The validation and test set data are then input into the neural network to check the accuracy, which is used to validate and test the failure state judgment model.

[0017] Furthermore, the lubrication state includes failure, reversibility, and effectiveness.

[0018] Furthermore, the one-hot encoding defines a valid lubrication state as 2, a reversible state as 1, and a failure state as 0.

[0019] Furthermore, the rolling structure is a small ball; the measuring instrument is a white light interferometer or an atomic force microscope.

[0020] Based on the above technical solution, the beneficial effects of this application are as follows: 1. The method of the present invention achieves accurate determination of the lubrication state under low friction by adopting conventional low friction coefficient test and low friction coefficient recovery test, avoiding the technical problem of increased actual use cost due to incorrect judgment of lubrication state under low friction state, and protecting related supporting lubrication components.

[0021] 2. The method of the present invention achieves low friction state control within a certain load and speed range by adjusting the load and speed, clarifies the failure state of the low friction system, and improves the applicability of the low friction system.

[0022] 3. The method of the present invention predicts the lubrication state by constructing a failure state determination model, thereby achieving accurate prediction of the failure state of low friction system under different operating conditions and reducing research and development costs. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the failure state determination model; Figure 2 This is a schematic diagram of the measurement area; Figure 3 A rough selection of schematics Figure 1 ; Figure 4 A rough selection of schematics Figure 2 ; Figure 5 Schematic diagram of a low friction coefficient recovery test to adjust the load. Figure 1 ; Figure 6 Schematic diagram of a low friction coefficient recovery test to adjust the load. Figure 2 ; Figure 7A schematic diagram of a low-friction coefficient recovery test to adjust speed. Figure 1 ; Figure 8 A schematic diagram of a low-friction coefficient recovery test to adjust speed. Figure 2 ; Figure 9 A schematic diagram of the structure for routine testing of low friction coefficient or anti-friction coefficient recovery testing; Figure 10 This is a schematic diagram of the neural network structure for the failure state determination model. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0025] Example 1 A method for predicting the lubrication state of a low-friction lubrication system includes the following steps: S1: Collect lubricating fluid and multiple different material samples to form a material sample sample, and conduct routine low friction coefficient tests. Apply load and speed to the material sample to obtain the friction coefficient of the material, and define the valid material sample and the material sample to be determined. S2: Calculate the roughness within the scratched area of ​​the material sample to be determined, and use it as the surface roughness of the material sample; S3: Perform a low friction coefficient recovery test on the material sample to be determined; the low friction coefficient recovery test tests the friction coefficient of the material sample to be determined by changing the previously applied load and speed. S4: After low friction coefficient recovery test, if the low friction coefficient of the material sample to be determined is lower than 0.02, the material sample is defined as reversible; if the low friction coefficient of the material sample to be determined is higher than 0.02, the material sample is defined as failed. S5: The surface roughness of the material sample, the defined material sample data, and the velocity and load corresponding to each material sample data are used as a dataset and input into the failure state judgment model built using deep learning for training. S6: Collect a new dataset and input it into the trained failure state judgment model to finally predict the probability of the lubrication state of the material sample under low friction.

[0026] In this method, each type of lubricant and each pair of different material samples constitute a low-friction lubrication system. Steps S1 to S6 are used to predict the lubrication state of this low-friction lubrication system, thereby determining the effective probability that conventional low-friction coefficient tests can directly achieve the optimal lubrication state (friction coefficient below 0.02), the reversible probability that changing the load and speed can achieve the optimal lubrication state (friction coefficient below 0.02), and the failure probability that changing the load and speed cannot restore the optimal lubrication state (friction coefficient below 0.02).

[0027] Example 2 Based on Example 1, in step S1, samples with a friction coefficient lower than 0.02 are defined as valid material samples, and samples with a friction coefficient greater than 0.02 are defined as undetermined material samples.

[0028] like Figure 9 As shown, the specific steps for conducting a routine low-friction coefficient test in step S1, which involves applying load and velocity to the material sample to obtain the friction coefficient of the material, are as follows: Step a1: Prepare the rolling structure and the substrate from the materials of the two different contact bodies respectively; the rolling structure is a small ball; Step a2: Fix the base on the cylindrical motion platform, apply lubricant to the bottom of the rolling structure (generally 10ml), and apply a force F1 to the rolling structure, which then transfers the force F1 to the base. A force sensor is installed on the top of the ball. Applying a force F1 vertically from the top of the ball allows for direct observation of the direction and magnitude of the applied force F1. The force sensor is a mature existing mechanism in this field and will not be described in detail here. Step a3: The frictional force F2 between the rolling structure and the base is obtained by rotating the motion platform around the center, thereby obtaining the coefficient of friction Cof; the motion platform can be a commercially available instrument, such as the Brook friction and wear tester; the rolling structure is a small ball.

[0029] The specific formula for calculating the friction coefficient Cof is as follows: .

[0030] like Figure 2 , Figure 3 and Figure 4 As shown, the specific steps of step S2 are as follows: Step b1: Randomly select at least four locations in the scratched area and measure the roughness of the entire area along the cross-section of the scratch using a measuring instrument; the measuring instrument is a white light interferometer or an atomic force microscope; Step b2: The measurement range A at the center of the scratch is taken as the precise value of the roughness; the measurement range A at the center of the scratch is fifty percent of the cross-sectional length of the scratch; for example... Figure 4 As shown, the entire scratch length is approximately 240 μm. However, the surface roughness is very low on both sides of the scratch (0-80 μm, 200-240 μm), and only at 80-200 μm can the true surface roughness be reflected. Figure 3 As shown, the entire scratch length is close to 110 μm. However, the surface roughness is very low on both sides of the scratch (0-33 μm, 77-240 μm). Only at 33-77 μm can the roughness of the real surface be reflected. Step b3: Based on the obtained precise values ​​of multiple roughnesses, calculate the average of the multiple precise values ​​of the roughnesses, and use it as the surface roughness of the material sample.

[0031] Surface roughness can be used to determine the failure of material samples. By conducting small-sample tests, the failure threshold can be determined. For larger sample sizes, it is not necessary to conduct recovery tests to determine the failure state; surface roughness can be tested directly.

[0032] Among them, small sample experiments are a mature existing technology in this field. They are used to describe experiments with a small amount of data, such as a speed range of 1-10 m / s. However, the experiment only needs to complete 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 m / s to derive 1.1, 1.2 m / s, etc., without conducting the experiment.

[0033] The average surface roughness is used to determine the failure threshold of the lubrication system. Subsequent test samples are then tested under normal conditions at low friction coefficients. If the friction coefficient is greater than 0.02, the surface roughness of the sample is characterized. If the result is greater than the failure threshold, the system is considered to have failed.

[0034] Example 3 Based on Example 2, the specific method for the low friction coefficient recovery test in step S3 is as follows: with the speed of the rolling structure remaining constant, the low friction coefficient recovery test is performed by increasing or decreasing the load, or with the load of the rolling structure remaining constant, the low friction coefficient recovery test is performed by increasing or decreasing the speed.

[0035] The load can be increased by 1.2 to 3 times the load in the previous step S1; the load can be decreased by one-third to five-sixths of the load in the previous step S1; the speed can be increased by 1.5 to 9 times the speed in the previous step S1; the speed can be decreased by one-ninth to two-thirds of the speed in step S1.

[0036] The failure status assessment model uses Keras in deep learning to build a neural network, determine the network loss function and optimizer; the input dataset is preprocessed to obtain tensors of speed, load and lubrication status. The input features of the failure status assessment model are speed and load, and the output feature is lubrication status. One-hot encoding defines effective lubrication status as 2, reversible as 1, and failure as 0; the dataset is divided into training set, test set and validation set; lubrication status includes failure, reversible and effective. Among them, effective directly determines that the functional component is operating well and does not require shutdown; reversible indicates that the current operating parameters (including the speed and load involved in the use) cannot reach the optimal lubrication status of the component, but does not determine that the component is faulty or failed, and does not require shutdown and replacement, and can also guide further optimization of operating parameters; failure indicates that the current system has failed and needs to be stopped immediately to ensure that it will not cause further damage to the equipment. The neural network consists of an activation function and three linear layers; the number of neurons in the three linear layers are 64, 32, and 3, respectively; the activation function includes the ReLU activation function and a Softmax layer; the network loss function is cross-entropy L, and the optimizer is ADAM; the specific expression of cross-entropy L is as follows: ; In the formula, One-hot encoding of the real label. To predict the category of the model The probability of.

[0037] The neural network is initialized using Python's numpy.random function, and then iteratively trained using the training set until the maximum number of iterations is reached. The validation and test sets are then input into the neural network to check the accuracy, which is used to validate and test the failure state judgment model.

[0038] Example 4 Based on Example 3, this example provides a specific application embodiment, as follows: Step 1: Perform routine tests on the oxalic acid low-friction system to obtain the friction coefficient of the system under different speed and load conditions. Define the friction coefficient below 0.02 as valid and the friction coefficient above 0.02 as pending. Step 2: Characterize all samples to be determined, such as Figure 2 and Figure 3 The surface roughness of the characteristic region within the scratch shown; Step 3: For samples with a friction coefficient greater than 0.02, conduct low friction coefficient recovery tests. Based on the mechanism of achieving low friction, the low friction state (friction coefficient less than 0.02) is restored by changing specific conditions. Step 4: Define samples that cannot be restored to the low friction state even after the low friction coefficient recovery test as irreversible, and define recoverable samples as reversible; Step 5: As Figure 1 As shown, a failure state determination model is constructed. Based on the data characteristics in this paper, the three states of reversibility, failure, and effectiveness can be determined by velocity and load. Step 1 will be described in further detail, and the friction coefficient under different speed and load conditions will be obtained through the following steps: Stp1: such as Figure 9 As shown, for different lubrication systems, the materials of the two contact bodies are respectively prepared into spheres and substrates; other structures can also be prepared here, not limited to sphere structures; Stp2: Install the small ball in a ball tractor or other tooling, and fix the base to the motion platform; Stp3: Apply lubricating oil before the experiment begins; Stp4: The force F1 is controlled by a mechanical sensor and applied to the base through a small ball; Stp5: The motion platform rotates around the center or reciprocates, and the speed can be controlled by the motion platform; Stp6: The frictional force F2 during the relative motion between the ball and the base is obtained through a mechanical sensor; Stp7: The coefficient of friction Cof can be calculated using Formula 1.

[0039]

[0040] Because the conditions for forming a low-lubrication state in water-based samples are harsh, the surface roughness at stress concentration points in characteristic areas within the scratch is an indicator of the lubrication state, rather than the entire scratch area; for example... Figure 2 As shown, the center position of the scratch is set as the zero point, and the measurement range is ±A (A = the total length of the scratch cross section × 50%).

[0041] The method for determining the lubrication condition is as follows: Determined by load: (1) Reversible A1: Routine tests for low coefficient of friction, such as Figure 5 As shown, the friction coefficient is 0.03 under the conditions of a speed of 0.5 m / s and a load of 1 N; A2: Under constant speed, the load was increased to 3 N, and the coefficient of friction was 0.008 (returning to the low friction state). This result indicates that the low friction state of the sample is reversible.

[0042] (2) Failure E1: Low coefficient of friction standard test, such as Figure 6 As shown, the friction coefficient is 0.03 under the conditions of a speed of 0.5 m / s and a load of 9 N; E2: Under constant speed, the load is reduced to 3 N, and the coefficient of friction is 0.028 (the low friction state cannot be recovered). This result indicates that the low friction state of the sample is a failure.

[0043] Determined by speed: (1) Reversible B1: Routine tests for low coefficient of friction, such as Figure 7 As shown, the friction coefficient is 0.02 under the conditions of a speed of 0.03 m / s and a load of 3 N; B2: Under constant load, the speed was increased to 0.25 m / s, and the friction coefficient was 0.008 (returning to the low friction state). This result indicates that the low friction state of the sample is reversible.

[0044] (2) Failure C1: Low coefficient of friction standard test, such as Figure 8 As shown, under the conditions of a speed of 2 m / s and a load of 3 N, the coefficient of friction is 0.035 (greater than 0.02). C2: Under constant load, when the speed is reduced to 0.25 m / s, the friction coefficient is 0.028 (the low friction state cannot be recovered). This result indicates that the low friction state of the sample is a failure.

[0045] Valid determination: During routine testing for low friction coefficients, the sample's friction coefficient was less than 0.02, thus the sample was deemed valid.

[0046] like Figure 1 and Figure 10 As shown, a failure state determination model is constructed: D1: Preprocessing the input dataset in the input layer: Each data point is processed into a tensor of (velocity, load, lubrication state). The input features of the neural network are (velocity, load), and the output is the lubrication state, which includes effective, reversible, and ineffective states. Based on one-hot encoding, the lubrication state is simplified to effective == 2, reversible == 1, and ineffective == 0. To verify the effectiveness of the network, the dataset is divided into a training set (X_train, Y_train) (70%), a test set (X_test, Y_test) (15%), and a validation set (X_val, Y_val) (15%). D2: Using Keras as the framework, build a neural network, compile the model, determine the network loss function and optimization method. The network consists of 3 linear layers with 64, 32, and 3 neurons in each layer, one ReLU activation function, one Softmax layer, and cross-entropy. As the loss function of the network, One-hot encoding of the real label. To predict the category of the model The probability of ADAM is the optimizer, which combines the model with the loss function, optimizer and accuracy. D3: Initialize and train the network. Use the built-in numpy.random function of Python to initialize the neural network, and use the training set to iteratively train the neural network until the maximum number of iterations is met and training stops.

[0047] D4: Validate and test the model. Input the validation and test set data into the network and check the accuracy.

[0048] like Figure 1 As shown, as the image changes from light gray to dark gray, it indicates that the lubrication system has a higher probability of failure under the corresponding operating conditions. This relationship graph was obtained by training the failure state judgment model after importing actual test data. The probability 0-1 represents the probability of failure under the corresponding operating conditions.

[0049] Different loads and speeds correspond to different failure probabilities. Some of these results are obtained from experiments. By importing experimental data into the model for training, a more accurate judgment model for the oxalic acid system can be obtained, which can then predict the probability of results for other operating parameters that have not been tested.

[0050] As shown in Table 1 below, the probability of failure of the material sample under low friction is finally predicted by inputting arbitrary speed and load into the failure state judgment model.

[0051] Table 1. Calculation results of the prediction model for the lubrication state of the oxalic acid lubrication system.

[0052] In this application, different lubrication systems correspond to different contact bodies and lubricating oils. This application establishes a method for predicting the lubrication state of different low-friction lubrication systems, such as bearings. There are many types of lubricating oils, and their contact bodies are also different, including bearing steel, polyimide, etc. Therefore, this embodiment is only a specific application embodiment, and it can also be applied to the precision transmission mechanism of high-end equipment, such as electric spindles, reducers and bearings.

[0053] The above description is a detailed description of the preferred embodiments of the present invention. However, the embodiments are not intended to limit the scope of the patent application of the present invention. All equivalent changes or modifications made under the technical spirit of the present invention should fall within the patent scope covered by the present invention.

Claims

1. A method for predicting the lubrication state of a low-friction lubrication system, characterized in that, The methods and steps include the following: S1: Collect lubricating fluid and multiple different material samples to form a sample of material samples, and conduct routine low friction coefficient tests. Apply load and speed to the material samples to obtain the friction coefficient of the materials, and define the valid material samples and the material samples to be determined. S2: Calculate the roughness within the scratched area of ​​the material sample to be determined, and use it as the surface roughness of the material sample; S3: Conduct a low friction coefficient recovery test on the material sample to be determined; The low friction coefficient recovery test measures the friction coefficient of a material sample by changing the previously applied load and speed. S4: If the low friction coefficient of the material sample to be determined is less than 0.02 after the low friction coefficient recovery test, the material sample is defined as reversible. If the coefficient of friction of the material sample to be determined is higher than 0.02, then the material sample is defined as failed. S5: The surface roughness of the material sample, the defined material sample data, and the velocity and load corresponding to each material sample data are used as a dataset and input into the failure state judgment model built using deep learning for training. S6: Collect a new dataset and input it into the trained failure state judgment model to finally predict the probability of the lubrication state of the material sample under low friction.

2. The method for predicting the lubrication state of a low-friction lubrication system according to claim 1, characterized in that: In step S1, samples with a friction coefficient lower than 0.02 are defined as valid material samples, and samples with a friction coefficient greater than 0.02 are defined as undetermined material samples.

3. The method for predicting the lubrication state of a low-friction lubrication system according to claim 2, characterized in that: The specific steps for conducting a routine low-friction coefficient test in step S1, which involves applying load and velocity to the material sample to obtain the friction coefficient of the material, are as follows: Step a1: Prepare the rolling structure and substrate from the materials of the two different contact bodies respectively; Step a2: Fix the base on the moving platform of the cylinder, apply lubricant to the bottom of the rolling structure, apply force F1 to the rolling structure, and apply force F1 to the base through the rolling structure; Step a3: By rotating the motion platform around the center, the frictional force F2 between the rolling structure and the base is obtained, and thus the coefficient of friction Cof is obtained.

4. The method for predicting the lubrication state of a low-friction lubrication system according to claim 3, characterized in that: The specific formula for calculating the friction coefficient Cof is as follows: 。 5. The method for predicting the lubrication state of a low-friction lubrication system according to claim 4, characterized in that: The specific steps of step S2 are as follows: Step b1: Randomly select at least four locations in the scratched area and measure the roughness of the entire area along the cross-section of the scratch using a measuring instrument; Step b2: Take the measurement range A of the scratch center position as the accurate value of roughness; the measurement range A of the scratch center position is fifty percent of the scratch cross-sectional length; Step b3: Based on the obtained precise values ​​of multiple roughnesses, calculate the average of the multiple precise values ​​of the roughnesses, and use it as the surface roughness of the material sample.

6. The method for predicting the lubrication state of a low-friction lubrication system according to claim 5, characterized in that: The specific method for the low friction coefficient recovery test in step S3 is as follows: with the speed of the rolling structure remaining constant, the low friction coefficient recovery test is performed by increasing or decreasing the load, or with the load of the rolling structure remaining constant, the low friction coefficient recovery test is performed by increasing or decreasing the speed.

7. The method for predicting the lubrication state of a low-friction lubrication system according to claim 6, characterized in that: The load increases from 1.2 to 3 times the load in the previous step S1; the load decreases from one-third to five-sixths of the load in the previous step S1; the speed increases from 1.5 to 9 times the speed in the previous step S1; and the speed decreases from one-ninth to two-thirds of the speed in step S1.

8. The method for predicting the lubrication state of a low-friction lubrication system according to claim 1, characterized in that: The failure state judgment model uses Keras in deep learning to build a neural network, determine the network loss function and optimizer; it preprocesses the input dataset to obtain tensors of speed, load and lubrication state. The input features of the failure state judgment model are speed and load, and the output feature is lubrication state.

9. The method for predicting the lubrication state of a low-friction lubrication system according to claim 8, characterized in that: The dataset is divided into a training set, a test set, and a validation set.

10. The method for predicting the lubrication state of a low-friction lubrication system according to claim 9, characterized in that: The neural network consists of an activation function and three linear layers; the number of neurons in the three linear layers are 64, 32 and 3, respectively; the activation function includes a ReLU activation function and a Softmax layer; the network loss function is cross-entropy L, and the optimizer is ADAM.

11. The method for predicting the lubrication state of a low-friction lubrication system according to claim 10, characterized in that: The specific expression for the cross-entropy L is as follows: ； In the formula, One-hot encoding of the real label. To predict the category of the model The probability of.

12. The method for predicting the lubrication state of a low-friction lubrication system according to claim 11, characterized in that: The neural network is initialized using Python's numpy.random function, and then iteratively trained using the training set until the maximum number of iterations is reached. The validation and test sets are then input into the neural network to check the accuracy, which is used to validate and test the failure state judgment model.

13. The method for predicting the lubrication state of a low-friction lubrication system according to claim 12, characterized in that: The lubrication state includes failure, reversibility, and effectiveness.

14. The method for predicting the lubrication state of a low-friction lubrication system according to claim 13, characterized in that: The one-hot encoding defines a valid lubrication state as 2, a reversible state as 1, and a failure state as 0.

15. The method for predicting the lubrication state of a low-friction lubrication system according to claim 6, characterized in that: The rolling structure is a small ball; the measuring instrument is a white light interferometer or an atomic force microscope.

Images