A friction pair lubrication state automatic identification method based on triboelectric signal

Abstract

The application discloses a kind of based on triboelectric signal's friction pair lubrication state automatic identification method, including the original triboelectric signal set generated by respectively collecting friction pair under different lubrication states and carrying out friction, EEMD decomposition is carried out to each original triboelectric signal, obtain the IMF component and residual component after decomposition, the cross-correlation coefficient of each IMF component and residual component is calculated based on cross-correlation coefficient method, dominant IMF component is selected, and the reconstructed signal is obtained by superimposing dominant IMF component, and the time domain waveform diagram is drawn according to the reconstructed signal under different lubrication states;Convolutional neural network model is constructed, and the trained convolutional neural network is obtained according to the training set under different lubrication states Training convolutional neural network and testing, according to the trained convolutional neural network, the lubrication state corresponding to the original triboelectric signal to be identified is obtained.The identification precision is higher, does not need a lot of artificial feature extraction, and the automatic identification of lubrication state can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automatic identification of lubrication state of friction pair, and particularly relates to a method for automatically identifying lubrication state of friction pair based on friction electric signal. BACKGROUND

[0002] In recent years, the problem of wear caused by friction makes the consumption of energy and the damage of equipment more and more serious, and each country pays more and more attention to the problem of wear. It is expected that the lubrication state of friction pair can be identified and monitored in real time through a method or means, so that the actual situation of mechanical equipment in different friction pair lubrication states such as lubrication failure and excessive wear can be accurately evaluated, and the property loss and energy consumption caused by wear can be avoided or reduced. At present, the commonly used methods for real-time monitoring of the lubrication state of friction pair include friction heat, friction vibration, oil monitoring and oil film resistance, but these methods have some defects to a certain extent, for example, the precision and sensitivity are relatively poor, and the lubrication state of friction pair cannot be reflected in time. Moreover, the traditional identification method depends on manual feature extraction, requires a large amount of signal processing and professional knowledge, and has very low identification accuracy for non-professionals. SUMMARY

[0003] The present application provides a method for automatically identifying the lubrication state of friction pair based on friction electric signal to overcome the above technical problems.

[0004] A method for automatically identifying the lubrication state of friction pair based on friction electric signal, comprising,

[0005] Step one, a set of original friction electric signals generated by friction of the friction pair under different lubrication states are collected respectively, the different lubrication states include liquid lubrication state, mixed lubrication state and boundary lubrication state,

[0006] Step two, the number of EEMD decomposition layers is set to n, and each original friction electric signal is subjected to EEMD decomposition to obtain n IMF components and a residual component after decomposition,

[0007] Step three, the cross-correlation coefficient of each IMF component and the residual component is calculated based on the cross-correlation coefficient method, a first threshold is set, the IMF component whose cross-correlation coefficient exceeds the first threshold is regarded as a dominant IMF component, and the dominant IMF component is superimposed to obtain a reconstructed signal,

[0008] Step four, a time domain waveform diagram is drawn according to the reconstructed signal under different lubrication states.

[0009] Step five, constructing a convolutional neural network model, respectively dividing the time domain waveform diagrams under different lubrication states into a training set and a test set, training the convolutional neural network according to the training set under different lubrication states, testing the trained convolutional neural network through the test set under different lubrication states, calculating the accuracy of the trained convolutional neural network, setting a second threshold, when the accuracy exceeds the second threshold, obtaining the trained convolutional neural network, identifying the time domain waveform diagram of the original friction electric signal to be identified according to the trained convolutional neural network, and obtaining the lubrication state corresponding to the original friction electric signal.

[0010] Preferably, the EEMD decomposition of each original friction electric signal comprises,

[0011] Step one, adding a noise-containing signal to the original friction electric signal at the i th decomposition according to formula (1),

[0012] x i (t)=x(t)+α·a i (t) i=1,2,...,n (1)

[0013] Wherein, α is the noise amplitude coefficient, a i (t) is the noise-containing signal, x(t) is the original friction electric signal, n is the decomposition layer number, x i (t) is the noise-containing signal,

[0014] Step two, the i th EMD processing is performed on the noise-containing signal x i (t) according to formula (2), to obtain n C i (t) and an r i (t),

[0015]

[0016] Wherein, C i (t) is the IMF component, r i (t) is the residual component,

[0017] Step three, repeat steps one and two n times, obtain the i th IMF component average x i (t) and the residual component average r n (t) according to formulas (3) and (4),

[0018]

[0019]

[0020] Step four, obtaining the EEMD decomposed frictional electricity signal according to formula (5), the EEMD decomposed frictional electricity signal including n IMF component averages and one residual component average,

[0021]

[0022] Wherein, X(t) is the EEMD decomposed frictional electricity signal.

[0023] Preferably, the method of calculating the cross-correlation coefficient of each IMF component and the residual component respectively includes calculating the cross-correlation coefficient of each IMF component and the residual component according to formula (6),

[0024]

[0025] Wherein, x is the original frictional electricity signal, IMF i is the i-th IMF component, x is the average value of the original frictional electricity signal, is the average value of the i-th IMF component, n is the total number of IMF components and residual components, and p(x, IMFi) is the cross-correlation coefficient of the i-th IMF component.

[0026] Preferably, the superposition of the dominant IMF components to obtain the reconstructed signal includes superimposing the dominant IMF components according to formula (7),

[0027]

[0028] Wherein, x i (t) is the i-th dominant IMF component, m is the number of dominant IMF components, and Y(t) is the reconstructed signal.

[0029] A friction pair lubrication state automatic identification system based on frictional electricity signal, comprising a data acquisition module, an EEMD decomposition module, a dominant IMF component reconstruction module, a waveform diagram drawing module and a lubrication state identification module,

[0030] The data acquisition module is used for acquiring the original frictional electricity signal, including acquiring a set of original frictional electricity signals generated by the friction pair under different lubrication states, respectively, the different lubrication states including liquid lubrication state, mixed lubrication state, boundary lubrication state,

[0031] The EEMD decomposition module is used for EEMD decomposition of the original frictional electricity signal, including setting the number of EEMD decomposition layers as n, EEMD decomposition of each original frictional electricity signal, obtaining n IMF components and one residual component after decomposition,

[0032] The advantage IMF component reconstruction module is used for selecting an advantage component and obtaining a reconstructed signal according to the advantage component, and comprises the following steps: calculating a cross-correlation coefficient of each IMF component and a residual component based on a cross-correlation coefficient method, setting a first threshold, taking the IMF component whose cross-correlation coefficient exceeds the first threshold as an advantage IMF component, and superimposing the advantage IMF component to obtain the reconstructed signal,

[0033] The waveform drawing module is used for drawing a time-domain waveform graph according to the reconstructed signal under different lubricating states,

[0034] The lubricating state recognition module is used for constructing and training a convolutional neural network, obtaining a lubricating state corresponding to an original frictional electric signal according to the trained convolutional neural network, and comprises the following steps: constructing a convolutional neural network model, dividing the time-domain waveform graphs under different lubricating states into a training set and a test set, training the convolutional neural network according to the training set under different lubricating states, testing the trained convolutional neural network through the test set under different lubricating states, calculating an accuracy rate of the trained convolutional neural network, setting a second threshold, obtaining the trained convolutional neural network when the accuracy rate exceeds the second threshold, and recognizing the time-domain waveform graph of the original frictional electric signal to be recognized according to the trained convolutional neural network, and obtaining the lubricating state corresponding to the original frictional electric signal.

[0035] The application provides a friction pair lubricating state automatic recognition method based on a frictional electric signal, which first recognizes and monitors the lubricating state of a friction pair by using a frictional electric signal generated in a friction and wear process of a metal friction pair, has the advantages of easy real-time acquisition of signals, low cost of signal acquisition equipment and high signal sensitivity, can accurately recognize and has high precision by performing EEMD decomposition and noise reduction on an original frictional electric signal and recognizing the lubricating state of the friction pair based on a convolutional neural network, and can realize automatic recognition of the lubricating state without a large amount of manual feature extraction and professional knowledge. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0037] Figure 1 is a method flowchart of the present application;

[0038] Figure 2 is a schematic diagram of a friction form of a friction pair of the present application;

[0039] Figure 3 is a time-domain waveform graph of an original frictional electric signal of the present application;

[0040] Figure 4 is the frictional electricity signal time domain waveform diagram before and after noise reduction of the application;

[0041] Figure 5 is the first 9 IMF components and residual components obtained after EEMD decomposition of the application;

[0042] Figure 6a is the frictional electricity signal time domain waveform diagram under liquid lubrication of the application;

[0043] Figure 6b is the frictional electricity signal time domain waveform diagram under mixed lubrication of the application;

[0044] Figure 6c is the frictional electricity signal time domain waveform diagram under boundary lubrication of the application;

[0045] Figure 7 is the convolutional neural network model structure diagram of the application;

[0046] Figure 8 is the schematic diagram of the friction and wear testing machine and the frictional electricity signal collection system of the application;

[0047] Explanation of reference numerals:

[0048] 1, drive motor; 2, friction torque sensor; 3, loading device; 4, high resistance electrometer; 5, NI data acquisition card; 6, insulating pad; 7, ball sample; 8, disc sample; 9, eccentric mechanism. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are some embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0050] Figure 1 is the method flowchart of the application, as shown in the method of the present embodiment can include: Figure 1

[0051] A friction pair lubrication state automatic identification method based on frictional electricity signal, comprising,

[0052] Step one, respectively collect the original frictional electricity signal set generated by the friction of the friction pair under different lubrication states, Figure 2 is the schematic diagram of the friction form of the friction pair, wherein A is a disc sample, B is a ball sample, and C is a sliding direction, Figure 3 ​The original frictional electricity signal time domain waveform diagram is the original frictional electricity signal time domain waveform diagram of the application, the different lubrication states include liquid lubrication state, mixed lubrication state, boundary lubrication state,

[0053] Step two, setting the number of EEMD decomposition layers as n, performing EEMD decomposition on each original frictional electricity signal, and obtaining n IMF components and a residual component after decomposition,

[0054] The EEMD decomposition on each original frictional electricity signal comprises,

[0055] Step one, according to formula (1), adding a noise-containing signal to the original frictional electricity signal at the i-th decomposition,

[0056] x i (t)=x(t)+α·a i (t) i=1,2,...,n (1)

[0057] Wherein, α is a noise amplitude coefficient, a i (t) is a noise-containing signal, x(t) is an original frictional electricity signal, n is a decomposition layer number, x i (t) is a noise-containing signal,

[0058] Step two, according to formula (2), performing i-th EMD processing on the noise-containing signal x i (t), obtaining n C i (t) and a r i (t),

[0059]

[0060] Wherein, C i (t) is an IMF component, r i (t) is a residual component,

[0061] Step three, repeating step one and step two n times, obtaining i-th IMF component average x i (t) and residual component average r n (t) according to formula (3) and (4),

[0062]

[0063]

[0064] Step four, obtaining the EEMD decomposed frictional electricity signal according to formula (5), the EEMD decomposed frictional electricity signal comprises n IMF component averages and a residual component average,

[0065]

[0066] Where X(t) is the triboelectric signal after EEMD decomposition;

[0067] Step 3: Calculate the cross-correlation coefficient between each IMF component and the residual component using the cross-correlation coefficient method. This calculation includes calculating the cross-correlation coefficient between each IMF component and the residual component according to formula (6).

[0068]

[0069] Where x is the original triboelectric signal, and IMFi is the i-th IMF component. This represents the average value of the original triboelectric signal. Let be the average value of the i-th IMF component, n be the total number of IMF components and residual components, and ρ(x, IMFi) be the cross-correlation coefficient of the i-th IMF component.

[0070] In this embodiment, the cross-correlation coefficients between different IMF components and the original signal are shown in Table 1.

[0071] Table 1. Cross-correlation coefficients between different IMF components and the original signal

[0072]

[0073] A first threshold is set, and IMF components with cross-correlation coefficients exceeding the first threshold are taken as dominant IMF components. The dominant IMF components are superimposed to obtain a reconstructed signal. The superposition of dominant IMF components to obtain a reconstructed signal includes superimposing the dominant IMF components according to formula (7).

[0074]

[0075] Where, x i Y(t) represents the i-th dominant IMF component, m represents the number of dominant IMF components, and Y(t) represents the reconstructed signal.

[0076] Step 4: Plot the time-domain waveform based on the reconstructed signals under different lubrication conditions. Figure 4 The time-domain waveforms of the triboelectric signal before and after noise reduction are shown.

[0077] Triboelectric signals, generated during the friction process of a friction pair, contain a series of information about the operating state of the friction pair. They are easy to acquire and have high signal accuracy, providing a new approach for identifying the lubrication state of friction pairs. The specific implementation process is as follows: Triboelectric signals of the metal friction pair are acquired using a KEITHLEY 6517B high-resistivity electrometer and an NI data acquisition card under different lubrication states (liquid lubrication, mixed lubrication, and boundary lubrication). The sampling frequency is 512Hz, and continuous acquisition lasts for 30 minutes. After the triboelectric signals stabilize, a 10-second signal is selected as the signal to be processed. The original triboelectric signals are then subjected to EEMD decomposition, with the decomposition layer number manually set to 9. This yields 9 IMF components and 1 residual component, as shown below. Figure 5 As shown, the dominant component is then selected using the autocorrelation coefficient method, with an autocorrelation coefficient threshold of 0.5. All dominant components are then synthesized to reconstruct the signal, thus achieving noise reduction. Figure 6a , Figure 6b , Figure 6c The time-domain waveforms of triboelectric signals under liquid lubrication, mixed lubrication, and boundary lubrication conditions are shown.

[0078] Step 5: Construct a convolutional neural network model. This model comprises five convolutional layers: the first layer is a 7*7*64 convolutional layer; the second layer consists of two 3*3*64 convolutional layers; the third layer consists of two 3*3*128 convolutional layers; the fourth layer consists of two 3*3*256 convolutional layers; and the fifth layer consists of two 3*3*512 convolutional layers. The structure of the convolutional neural network model is as follows: Figure 7 As shown.

[0079] The convolutional neural network model uses ResNet-18, and the expressions for the ReLU activation function and Softmax function of the ResNet-18 network model structure are as follows;

[0080]

[0081]

[0082] In the formula, x is the input triboelectric signal.

[0083] The time-domain waveforms under different lubrication conditions are divided into training and testing sets. A convolutional neural network is trained using the training set under different lubrication conditions, and the trained convolutional neural network is tested using the testing set under different lubrication conditions. The accuracy of the trained convolutional neural network is calculated, and a second threshold is set. When the accuracy exceeds the second threshold, the trained convolutional neural network is obtained. The time-domain waveform of the original triboelectric signal to be identified is identified using the trained convolutional neural network to obtain the lubrication state corresponding to the original triboelectric signal.

[0084] The process of training a convolutional neural network based on training sets under different lubrication conditions can be divided into:

[0085] S1. Obtain 300 time-domain waveform images of triboelectric signals under three different lubrication conditions. The images are divided into training and test sets in a 2:1 ratio. That is, for each condition, 200 images are used for model training and parameter updating, and 100 images are used for model verification.

[0086] S2. In model training and parameter optimization, the Adam optimization method is used to gradually update the model parameters until the model converges. The learning rate is selected as 0.0002. At the same time, batch processing is used for training, and the batch size is selected as 64. All training samples complete one model update process, which is called a training round. 100 rounds of training are performed. The model parameters when the model classification accuracy is the highest are recorded and saved. The model at this time is the optimal model.

[0087] S3. Input the test images into the optimal model to verify the accuracy, thereby validating the model's effectiveness in classifying and recognizing the three lubrication states. The accuracy is defined as:

[0088]

[0089] S4. Once the test set verification is successful, the time-domain waveform image under any lubrication state can be input into the model to obtain the corresponding classification, thus achieving the purpose of state recognition.

[0090] An automatic identification system for the lubrication state of friction pairs based on triboelectric signals includes a data acquisition module, an EEMD decomposition module, a dominant IMF component reconstruction module, a waveform plotting module, and a lubrication state identification module.

[0091] The data acquisition module is used to collect raw triboelectric signals, including sets of raw triboelectric signals generated by friction between the friction pair under different lubrication conditions, such as liquid lubrication, mixed lubrication, and boundary lubrication.

[0092] The EEMD decomposition module is used to perform EEMD decomposition on the original triboelectric signal. This includes setting the number of EEMD decomposition layers to n, performing EEMD decomposition on each original triboelectric signal, and obtaining n IMF components and one residual component after decomposition.

[0093] The dominant IMF component reconstruction module is used to select dominant components and obtain a reconstructed signal based on them. This includes calculating the cross-correlation coefficient between each IMF component and the residual components using the cross-correlation coefficient method, setting a first threshold, identifying IMF components with cross-correlation coefficients exceeding the first threshold as dominant IMF components, and superimposing the dominant IMF components to obtain the reconstructed signal.

[0094] The waveform plotting module is used to plot time-domain waveforms based on reconstructed signals under different lubrication conditions.

[0095] The lubrication state recognition module is used to construct and train a convolutional neural network (CNN). Based on the trained CNN, it obtains the lubrication state corresponding to the original triboelectric signal. This includes constructing a CNN model, dividing the time-domain waveforms under different lubrication states into training and testing sets, training the CNN using the training set under different lubrication states, testing the trained CNN using the testing set under different lubrication states, calculating the accuracy of the trained CNN, setting a second threshold, and when the accuracy exceeds the second threshold, acquiring the trained CNN. The module then identifies the time-domain waveform of the original triboelectric signal to be recognized based on the trained CNN, thus obtaining the lubrication state corresponding to the original triboelectric signal.

[0096] This embodiment is based on a friction and wear testing machine. Figure 8 As shown, the testing machine consists of a drive motor, a loading device, and friction pair components. The drive motor 1 is connected to the disc sample 8 via an eccentric wheel mechanism 9, converting the rotational motion of the drive motor into the reciprocating motion of the disc sample. The ball sample 7 is placed above the disc sample 8, and each is fixed and installed on the testing machine by a fixing clamp. The loading device 3 is placed above the ball sample 7, and a preset load is applied to the disc sample 8 via a standard spring through the ball sample 7. The insulating pads 6 are placed below the ball sample 7 and the disc sample 8, respectively. The high-resistance electrometer 4 is connected in series with the ball sample 7 and the disc sample 8 via copper wires.

[0097] More preferably, in this embodiment, the load range of the friction and wear testing machine is 10–200 N; a ball-disc friction pair is used, with a ball sample diameter of 6 mm and a disk sample diameter of 25 mm, and an initial surface roughness Ra = 0.336; the ball sample material is Gcr15 steel, and the disk sample material is 45 steel, using CD40 lubricating oil; the drive motor speed range is 200–2200 r / min; the reciprocating sliding frequency of the friction pair is 3–45 Hz, and the sliding length is 0.5–25 mm; the triboelectric signal collected by the KEITHLEY 6517B high-resistance electrometer 4 is acquired and stored through the NI signal acquisition card 5 and the LabVIEW acquisition software program, and the current measurement range of the KEITHLEY 6517B high-resistance electrometer is 1 fA–20 mA. In this experiment, a range of 2 mA is selected, the sampling frequency of the LabVIEW-based signal acquisition system is set to 512 Hz, continuous acquisition is performed for 30 min, and after the signal stabilizes, a 10 s signal is selected as the signal to be processed.

[0098] The acquisition of raw triboelectric signals can be divided into:

[0099] Step 1: Fix the ball sample 7 and the disc sample 8 in the fixture and install them on the wear tester and connect them in series with the high resistance electrometer 4 through copper wires.

[0100] Step 2: Set the drive motor speed to 400 r / min, the loading device load to 20 N, and the reciprocating friction distance to 5 mm. Conduct friction and wear tests under liquid lubrication, mixed lubrication, and boundary lubrication conditions respectively. The lubricating oil is CD40 lubricating oil.

[0101] Step 3: After starting the friction and wear test, the loading device gradually applies a load of 20N. Then, the LabVIEW-based triboelectric signal acquisition software and NI data acquisition card are used to collect the triboelectric signals generated during the friction and wear process of the metal friction pair. The sampling frequency is 512Hz, and the data is continuously collected for 30 minutes. The collected triboelectric signals are then input into the computer for storage.

[0102] Overall beneficial effects:

[0103] This invention provides an automatic identification method for the lubrication state of friction pairs based on triboelectric signals. First, the triboelectric signals generated during the friction and wear process of metal friction pairs are used to identify and monitor the lubrication state of the friction pairs. This method has the advantages of easy real-time signal acquisition, low cost of signal acquisition equipment, and high sensitivity of signal measurement. By performing EEMD decomposition and noise reduction on the original triboelectric signals and identifying the lubrication state of the friction pairs based on convolutional neural networks, it can not only accurately identify the lubrication state with high precision, but also achieve automatic identification of lubrication state without requiring a large amount of manual feature extraction and professional knowledge.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically identifying the lubrication state of a friction pair based on triboelectric signals, characterized in that, include, Step 1: Collect the raw triboelectric signals generated by the friction pair under different lubrication conditions. These different lubrication conditions include liquid lubrication, mixed lubrication, and boundary lubrication. Step 2: Set the number of EEMD decomposition layers to n, perform EEMD decomposition on each original triboelectric signal, and obtain n IMF components and one residual component after decomposition. Step 3: Calculate the cross-correlation coefficient between each IMF component and the residual component using the cross-correlation coefficient method. Set a first threshold, and identify IMF components with cross-correlation coefficients exceeding the first threshold as dominant IMF components. Superimpose the dominant IMF components to obtain the reconstructed signal. Step 4: Plot the time-domain waveform based on the reconstructed signals under different lubrication conditions; Step 5: Construct a convolutional neural network model. Divide the time-domain waveforms under different lubrication states into training and testing sets. Train the convolutional neural network using the training set under different lubrication states, and test the trained convolutional neural network using the testing set under different lubrication states. Calculate the accuracy of the trained convolutional neural network and set a second threshold. When the accuracy exceeds the second threshold, acquire the trained convolutional neural network. Use the trained convolutional neural network to identify the time-domain waveform of the original triboelectric signal to be identified, and obtain the lubrication state corresponding to the original triboelectric signal.

2. The method for automatic identification of the lubrication state of a friction pair based on triboelectric signals according to claim 1, characterized in that, The EEMD decomposition of each original triboelectric signal includes, Step 1: According to formula (1), add a noisy signal to the original triboelectric signal during the i-th decomposition. x i (t)=x(t)+α·a i (t)i=1,2,...,n (1) Where α is the noise amplitude coefficient, a i (t) represents the noisy signal, x(t) represents the original triboelectric signal, n represents the number of decomposition layers, and x i (t) represents a noisy signal. Step 2: Apply formula (2) to the noisy signal x i (t) Perform the i-th EMD process to obtain n Cs i (t) and an r i (t), Among them, C i (t) represents the IMF component, r i (t) represents the residual component. Step 3: Repeat Step 1 and Step 2 n times, and obtain the average value x of the i-th IMF component according to formulas (3) and (4). i (t) and the average value of the residual components r n (t), Step 4: Obtain the triboelectric signal after EEMD decomposition according to formula (5). The triboelectric signal after EEMD decomposition includes the average value of n IMF components and the average value of a residual component. Where X(t) is the triboelectric signal after EEMD decomposition.

3. The method for automatic identification of the lubrication state of a friction pair based on triboelectric signals according to claim 1, characterized in that, The calculation of the cross-correlation coefficient between each IMF component and the residual component based on the cross-correlation coefficient method includes calculating the cross-correlation coefficient between each IMF component and the residual component according to formula (6). Where x is the original triboelectric signal, IMF i For the i-th IMF component, This represents the average value of the original triboelectric signal. Let be the average value of the i-th IMF component, n be the total number of IMF components and residual components, and ρ(x,IMFi) be the cross-correlation coefficient of the i-th IMF component.

4. The method for automatic identification of the lubrication state of a friction pair based on triboelectric signals according to claim 1, characterized in that, The process of superimposing the dominant IMF components to obtain the reconstructed signal includes superimposing the dominant IMF components according to formula (7). Where, x i Y(t) represents the i-th dominant IMF component, m represents the number of dominant IMF components, and Y(t) represents the reconstructed signal.

5. An automatic identification system for the lubrication state of a friction pair based on triboelectric signals, characterized in that, It includes a data acquisition module, an EEMD decomposition module, a dominant IMF component reconstruction module, a waveform plotting module, and a lubrication condition identification module. The data acquisition module is used to collect raw triboelectric signals, including sets of raw triboelectric signals generated by friction between the friction pair under different lubrication conditions, such as liquid lubrication, mixed lubrication, and boundary lubrication. The EEMD decomposition module is used to perform EEMD decomposition on the original triboelectric signal. This includes setting the number of EEMD decomposition layers to n, performing EEMD decomposition on each original triboelectric signal, and obtaining n IMF components and one residual component after decomposition. The dominant IMF component reconstruction module is used to select dominant components and obtain a reconstructed signal based on them. This includes calculating the cross-correlation coefficient between each IMF component and the residual components using the cross-correlation coefficient method, setting a first threshold, identifying IMF components with cross-correlation coefficients exceeding the first threshold as dominant IMF components, and superimposing the dominant IMF components to obtain the reconstructed signal. The waveform plotting module is used to plot time-domain waveforms based on reconstructed signals under different lubrication conditions. The lubrication state recognition module is used to construct and train a convolutional neural network (CNN). Based on the trained CNN, it obtains the lubrication state corresponding to the original triboelectric signal. This includes constructing a CNN model, dividing the time-domain waveforms under different lubrication states into training and testing sets, training the CNN using the training set under different lubrication states, testing the trained CNN using the testing set under different lubrication states, calculating the accuracy of the trained CNN, setting a second threshold, and when the accuracy exceeds the second threshold, acquiring the trained CNN. The module then identifies the time-domain waveform of the original triboelectric signal to be recognized based on the trained CNN, thus obtaining the lubrication state corresponding to the original triboelectric signal.

Images