Wear state identification method of ball screw pair based on raceway surface profile

Abstract

The present invention discloses a method for identifying the wear state of a ball screw pair based on raceway surface profiles. The method comprises: collecting raceway surface profile curves at three ball screw pair locations, removing the shape and performing Gaussian filtering on the profiles; extracting the main features of the raceway surface profiles using comprehensive statistical analysis, recursive analysis, and fractal analysis methods to construct a mixed feature set; evaluating the importance of each feature in the feature set using a random forest method, and selecting features containing 95% of the information to construct a new feature set; establishing a support vector machine model optimized by a genetic algorithm, training the model using the mixed feature set of two raceway locations, and finding the optimal parameters c and g using the genetic algorithm; and using the trained model to identify the wear state of the raceway locations. The method not only overcomes the noise interference inherent in traditional methods for identifying the wear state of ball screw pairs based on vibration signals, but also offers practicality, convenience, high accuracy, and low error.

Description

Technical Field

[0001] The invention belongs to the technical field of ball screw pairs, and in particular to a method for identifying the wear state of a ball screw pair based on a raceway surface profile. Background Art

[0002] Ball screws are widely used in CNC machine tools due to their excellent positioning accuracy and load-bearing performance. As the primary transmission mechanism of machine tool systems, ball screw wear can lead to continuous performance degradation, even causing damage to the machine tool and endangering human life. Therefore, accurately determining the wear status of ball screws is particularly important.

[0003] Currently, the wear status identification of ball screw pairs is achieved by collecting vibration signals during the operation of the ball screw pair and extracting their features. However, this method is easily affected by noise and other interferences, resulting in poor prediction results. Summary of the Invention

[0004] The purpose of the present invention is to address the problems existing in the prior art and provide a method for identifying the wear status of a ball screw pair based on the raceway surface profile, so as to solve the technical problem that the wear status identification of the ball screw pair is not accurate enough.

[0005] The technical solution to achieve the objectives of this invention is that the wear condition of a ball screw pair during operation can be reflected by changes in the raceway surface profile. Therefore, this invention proposes a method for identifying the wear condition of a ball screw pair based on the raceway surface profile. This method not only overcomes the shortcomings of existing methods but also offers advantages such as practicality, convenience, high accuracy, and low error, providing a new approach for identifying the wear condition of ball screw pairs.

[0006] A method for identifying the wear state of a ball screw pair based on a raceway surface profile, the method comprising the following steps:

[0007] Step 1: Collect the surface profile curves of the ball screw pair's ball screw raceway at three raceway positions, and perform shape removal and Gaussian filtering on the profiles;

[0008] Step 2, determine the wear status of the ball screw pair and use it as a label;

[0009] Step 3: Comprehensive statistical analysis, recursive analysis, and fractal analysis methods are used to extract the main features of the raceway surface profile, including roughness, maximum peak-to-valley height, root mean square, recursive law, fractal dimension, and multifractal spectrum width, and construct a labeled hybrid feature set;

[0010] Step 4: Evaluate the importance of each feature in the hybrid feature set, then sort the features in the hybrid feature set in descending order according to their importance, and extract the top m features that contain more than P% of the original information to construct a new hybrid feature set;

[0011] Step 5: Establish a support vector machine model based on genetic algorithm optimization, normalize the new mixed feature set and extract the mixed feature set of the two raceway positions, then import it into the established model for training, select the optimal penalty factor c and kernel function parameter g according to the genetic algorithm, and complete the model training;

[0012] Step 6: Import the mixed feature set of another raceway position into the trained model to identify the wear state of the raceway position, and compare it with the actual state to obtain the accuracy of the model;

[0013] Step 7: For the ball screw pair to be identified, execute steps 1, 3, and 4 to obtain its unlabeled mixed feature set, and then use the trained model to identify the wear status of the ball screw pair.

[0014] Furthermore, step 1 specifically includes:

[0015] Select three raceway positions of the lead screw in the uniform speed running area and mark them;

[0016] The ball screw pair was placed on a wear test bench. During the first 300,000 revolutions of the ball screw pair, the test bench was stopped and the ball screw pair was removed every 30,000 revolutions. Surface profile curves were collected at three raceway positions using a Taylor Hobson profilometer. After the ball screw pair had run for 300,000 revolutions, the test bench was stopped and the ball screw pair was removed every 60,000 revolutions to collect surface profile curves.

[0017] The surface profile curve is smoothed by Gaussian filtering, and then the fifth-order polynomial method is used to remove the shape of the smoothed curve, and finally the required raceway surface profile curve is derived.

[0018] Furthermore, step 2 is specifically as follows:

[0019] Taking the turning point of the preload change trend of the ball screw pair as the dividing point, the wear state of the ball screw pair is divided into running-in wear, stable wear and rapid wear states in sequence;

[0020] The wear status of the ball screw pair is determined based on the change trend of the current preload force.

[0021] Furthermore, step 3 specifically includes:

[0022] Step 3-1, calculate the roughness, maximum peak-to-valley height, and root mean square characteristics of the raceway surface profile curve using statistical analysis methods. The formula used is:

[0023]

[0024] Rz=z max -z min

[0025]

[0026] Where Ra is the roughness, Rz is the maximum peak-to-valley height, Rms is the root mean square, and z i is the height of the i-th contour sampling point, z min and z max are the minimum profile height and the maximum profile height, is the mean height of the contour, n is the sampling rate;

[0027] Step 3-2, use the recursive analysis method to solve the recursive law of the raceway surface profile. The formula used is:

[0028] r ij =|z i -z j |

[0029] ε=0.5σ

[0030] R ij (ε)=θ(ε-r ij )

[0031]

[0032]

[0033] Among them, RR is the recursive law, z i and z j are all contour heights, i,j=1,2,...,n,r ij Represents the distance between any two points i and j, R ij is an element of the matrix, σ is the standard deviation, ε is the threshold, and θ(x) is the Heaviside function.

[0034] Step 3-3, solving the fractal dimension of the raceway surface profile, the process includes:

[0035] The WM function is used to characterize the nonlinear raceway surface profile. The formula is:

[0036]

[0037] Among them, 1<D<2, γ>1, D is the fractal dimension, G is the height scale coefficient, γ n is the spectrum of the rough surface, n is the sampling rate, n1 is the minimum sampling rate, L is the sampling length, usually γ = 1.5, z(x) is the height of the random contour, and x is the position coordinate of the contour;

[0038] The power spectrum function of the WM function is expressed as:

[0039]

[0040] The incremental variance of z(x) is defined as the structure function, as shown below:

[0041]

[0042] Where τ = nΔL, ΔL is the sampling interval;

[0043] Combining the above formulas, we can get:

[0044]

[0045] Where C = Γ(2D-3)sin((D-1.5)π) / (4-2D)lnγ, Γ(*) is the gamma function,

[0046] Taking the logarithm of both sides of the above equation:

[0047] lgS(τ)=(4-2D)lgτ+lgC+2(D-1)lgG

[0048] According to the above formula, the fractal dimension D can be:

[0049]

[0050] Among them, k is the slope of the straight line, and the fractal dimension can reflect the complexity and irregularity of the raceway profile. The larger the fractal dimension, the more complex the raceway profile and the greater the irregularity.

[0051] In step 3-4, since the fractal dimension only describes the raceway profile from a single measure, and the multifractal spectrum can describe the raceway profile from multiple measures, the multifractal spectrum is introduced.

[0052] The box counting method is used to calculate the multifractal spectrum of the raceway surface profile. The minimum value of the raceway surface profile data is extracted as the lower limit zero to ensure that the amplitude is all positive. Many small boxes with a size of ε' are used to cover the raceway profile, 0<ε'<1, and the total profile height S i (ε) represents the sum of all contour amplitudes in the i-th small box when the box size is ε, then the probability measure P i(ε) is defined as:

[0053]

[0054] Among them, ∑S i (ε) is the sum of the amplitudes of all raceway profile data;

[0055] In the scale-free interval, P i (ε) can also be written in exponential form:

[0056] P i (ε)~ε α

[0057] Among them, α is the singular index, which is used to reflect P i The singular strength of (ε);

[0058] Let the number of boxes with the same singular index α be N α (ε), then in the scale-free interval N α (ε) is written in exponential form:

[0059] N α (ε)~ε -f(α)

[0060] Among them, f(α) represents the fractal dimension corresponding to the singular index α, and the smaller f(α), the greater N α (ε) is smaller;

[0061] Define the partition function of the multifractal as χ q (ε), the formula is:

[0062] χ q (ε)=∑P i (ε) q =ε τ(q)

[0063] Where q is the weight factor and τ(q) is the quality index;

[0064] When ε→0, τ(q) is written as:

[0065]

[0066] Combining the three parameters α, f(α) and τ(α) and according to the Legendre transformation relationship between the three, we can get:

[0067]

[0068] f(α)=q·α(q)-τ(q)

[0069] α and f(a) form a multifractal spectrum;

[0070] The multifractal spectral width Δα is defined as:

[0071] Δα=α max -α min

[0072] Among them, α min and α max are the minimum singular index and the maximum singular index respectively. Δα represents the unevenness of the probability measure of the sequence, which corresponds to the fluctuation range of the surface profile height. The larger the fluctuation range, the larger Δα is.

[0073] In steps 3-5, the roughness, maximum peak-to-valley height, root mean square, recurrence law, fractal dimension, and multifractal spectrum width of the raceway surface profile at the three raceway positions are combined with the label of step 2 to form a labeled hybrid feature set.

[0074] Furthermore, in step 4, the random forest algorithm based on out-of-bag data is used to evaluate the importance of each feature in the mixed feature set. The specific process includes:

[0075] Establish a total data set D, which consists of M samples and N = 6 features; independently sample k times from the data set, each sampling method is bagging resampling, and m samples are randomly selected each time to form a training set S, that is, k independent training data sets are formed. The data not sampled during the entire sampling process is called out-of-bag data, or OOB data;

[0076] Generate a corresponding single decision tree for any training data set, and generate a total of k decision trees. Each decision tree is verified by the OOB data set prediction, and the prediction result is Y p , the true value is Y, and the mean square error between the true value and the predicted value is recorded as ε mse ;

[0077] For a feature n among the N features of the out-of-bag data i Add noise interference, generate a new test set, and recalculate the mean square error between the true value and the predicted value, recorded as i=1,2,…,N;

[0078] The characteristic variable n i The importance of the corresponding single decision tree is recorded as mse i , whose value is

[0079] Traverse the random forest formed by k decision trees to obtain the feature variable n i The importance of the entire random forest is recorded as

[0080] Furthermore, step 5 extracts the mixed feature set of the two raceway positions from the new mixed feature set and performs normalization processing, and then imports it into the established model for training. The optimal penalty factor c and kernel function parameter g are selected according to the genetic algorithm to complete the model training, which specifically includes:

[0081] Step 5-1, randomly generate M1 individuals as the initial population, encode the parameters c and g, and set the maximum number of evolutionary iterations to N1;

[0082] Step 5-2, calculate the fitness of each individual in the group;

[0083] Step 5-3: Decode the optimal individual of the current group and determine whether its fitness meets the conditions or reaches the maximum number of evolutionary iterations of the population. If so, go to step 5-5; otherwise, go to step 5-4.

[0084] Step 5-4: Encode the parameters c and g again, perform selection, crossover and mutation operations on the population to obtain a new population, and go to step 5-3;

[0085] Step 5-5, output the optimal parameters c and g.

[0086] Furthermore, the fitness in step 5-2 is the accuracy of identifying the wear state of the ball screw pair, and the formula is:

[0087]

[0088] Where Fitness is the fitness, s is the number of data samples correctly identified, and t is the total number of data samples.

[0089] Compared with the prior art, the present invention has the following significant advantages:

[0090] 1) The present invention can more accurately and concisely identify the wear status of a ball screw pair, overcoming the drawback of inaccurate wear status identification caused by large noise interference in traditional feature extraction from vibration signals.

[0091] 2) The support vector machine model based on genetic algorithm optimization established in the present invention can well realize the identification of the wear state of the ball screw pair with high accuracy.

[0092] The present invention is further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0093] Figure 1 This is a flow chart of the ball screw pair wear state identification method based on the raceway surface profile of the present invention.

[0094] Figure 2A schematic structural diagram of the ball screw pair raceway surface profile acquisition provided in one embodiment.

[0095] Figure 3 The figure is a schematic structural diagram corresponding to a ball screw pair wear test bench provided in one embodiment.

[0096] Figure 4 A schematic diagram of a process for identifying the wear state of a ball screw pair based on the raceway surface profile provided in one embodiment.

[0097] Figure 5 FIG. 1 is a schematic diagram of fatigue spalling of a ball screw pair provided in one embodiment, wherein Figure 5 (a) is the surface morphology of the raceway in the non-working area of ​​the screw. Figure 5 (b) The surface morphology of the raceway where fatigue spalling occurs in the working area of ​​the screw.

[0098] Figure 6 A schematic diagram of the wear state classification of a ball screw pair provided in one embodiment.

[0099] Figure 7 A schematic diagram of feature importance based on random forest evaluation provided in one embodiment.

[0100] Figure 8 A schematic diagram of model training for a support vector machine optimized based on a genetic algorithm provided in one embodiment.

[0101] Figure 9 A schematic diagram of a confusion matrix for identifying the wear state of a ball screw pair based on a genetic algorithm optimized support vector machine provided in one embodiment.

[0102] Figure 10 A schematic diagram of model training for optimizing a support vector machine based on a genetic algorithm before dimensionality reduction provided in one embodiment.

[0103] Figure 11 A schematic diagram of a confusion matrix for identifying the wear state of a ball screw pair based on a genetic algorithm-optimized support vector machine before dimensionality reduction provided in one embodiment. DETAILED DESCRIPTION

[0104] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0105] Combine Figure 1 , the present invention describes the ball screw pair wear state identification method based on the raceway surface profile. Figure 2The ball screw pair raceway surface profile collection test mainly includes Taylor Hobson profilometer, support unit, PC and V-block. Taylor Hobson profilometer is used to extract the surface profile of the screw raceway, and the ball screw pair is placed on two V-blocks for support. The profilometer needs to be leveled before measurement, and its sampling length is set to 1mm. The measurement position is the 50th raceway, which is notched and marked to record the measurement position. The same operation is performed on the 68th raceway and the 98th raceway to obtain surface profile data. The three selected raceway positions are all in the uniform speed operation area. They are recorded here as position one to position three.

[0106] Then the ball screw pair wear test is carried out, refer to Figure 3 The ball screw wear test bench mainly consists of an eddy current brake, a loaded ball screw pair, a linear guide, a workbench, and a stepper motor. The axial load applied to the ball screw pair can be adjusted by setting the eddy current brake current. In this test, the eddy current brake current was set to 2.5A, corresponding to an axial load of 150kN. The ball screw pair to be tested was installed on the wear test bench. During the first 300,000 revolutions of the ball screw pair, the test bench was stopped and the screw was removed every 30,000 revolutions. The surface profile curves of the three raceway positions were collected using a Taylor Hobson profilometer. After the ball screw pair had run for 300,000 revolutions, the test bench was stopped and the screw was removed every 60,000 revolutions for a profile collection test. The curve was then smoothed using a Gaussian filter, and the smoothed curve was subjected to shape removal using a fifth-order polynomial method to ultimately derive the desired raceway surface profile curve.

[0107] Then, the wear state of the ball screw pair is divided according to the change trend of the preload force during the operation of the ball screw pair.

[0108] Furthermore, the characteristic values ​​of the collected raceway surface profile curve are extracted by statistical analysis method, recursive analysis method and fractal analysis method. The specific steps include:

[0109] First, the roughness, maximum peak-to-valley height, and root mean square characteristics of the extracted raceway profile curve are solved by statistical analysis methods. The features extracted by this method can reflect the most original characteristics of the raceway profile and are more sensitive to changes in wear. The larger the three, the rougher the raceway profile. The formula used in this method is:

[0110]

[0111] Rz=z max -z min

[0112]

[0113] Where Ra is the roughness, Rz is the maximum peak-to-valley height, Rms is the root mean square, and z i is the height of the i-th contour sampling point, z min and z max are the minimum and maximum contour heights, respectively, z is the mean height of the contour, and n is the sampling rate.

[0114] The recursive analysis method is used to solve the recursive law of the extracted raceway profile curve. The recursive law can reflect the volatility and stability of the raceway profile. The larger the recursive law, the more stable the corresponding profile and the smaller the volatility. The formula used in this method is:

[0115] r ij =|z i -z j |

[0116] ε=0.5σ

[0117] R ij (ε)=θ(ε-r ij )

[0118]

[0119]

[0120] Among them, RR is the recursive law, z i and z j are all contour heights, i,j=1,2,...,n,r ij Represents the distance between any two points, R ij is an element of the matrix, σ is the standard deviation, ε is the threshold, and θ(x) is the Heaviside function.

[0121] Statistical analysis methods and recursive analysis methods are both scale-dependent analysis methods, which are related to the sampling length of the equipment. Since fractal dimension has self-similarity and scale-free properties, it is a scale-independent feature, so fractal dimension is introduced. There are many methods for solving fractal dimension, and the structure function method is more accurate. This invention solves fractal dimension based on the structure function method. The WM function is suitable for characterizing the nonlinear raceway surface profile, and the formula is:

[0122]

[0123] Among them, 1<D<2, γ>1, D is the fractal dimension, G is the height scale coefficient, γ n is the spectrum of the rough surface, n is the sampling rate, n1 is the minimum sampling rate, L is the sampling length, usually γ = 1.5, z(x) is the height of the random contour, x is the position coordinate of the contour,

[0124] The power spectrum function of the above formula can be expressed as:

[0125]

[0126] The incremental variance of z(x) is defined as the structure function, as shown below:

[0127]

[0128] Wherein, τ=nΔL, ΔL is the sampling interval.

[0129] Combining the above formulas, we can get:

[0130]

[0131] Where C = Γ(2D-3)sin((D-1.5)π) / (4-2D)lnγ, Γ(*) is the gamma function,

[0132] Taking the logarithm of both sides of the equation gives:

[0133] lgS(τ)=(4-2D)lgτ+lgC+2(D-1)lgG

[0134] According to the above formula, the fractal dimension D can be:

[0135]

[0136] Where k is the slope of the line. The fractal dimension can reflect the complexity and irregularity of the raceway profile. The larger the fractal dimension, the more complex the raceway profile and the greater the irregularity.

[0137] Since the fractal dimension only describes the raceway profile from a single measure, the multifractal spectrum can describe the raceway profile from multiple measures, so the multifractal spectrum is introduced. The box counting method is used to calculate the multifractal spectrum of the raceway profile. The minimum value of the collected raceway profile data is extracted as the lower limit zero to ensure that the amplitude is all positive. Many small boxes with a size of ε (0 < ε < 1) are used to cover the raceway profile. The total profile height S i (ε) represents the sum of all contour amplitudes in the i-th small box when the box size is ε, then the probability measure P i (ε) can be defined as:

[0138]

[0139] Among them, ∑S i (ε) is the sum of the amplitudes of all raceway profile data.

[0140] In the scale-free interval, P i(ε) can also be written in exponential form, such as the formula:

[0141] P i (ε)~ε α

[0142] Among them, α is the singular index, which can reflect the P i The singular strength of (ε).

[0143] Assume that the number of boxes with the same singular index α is N α (ε), then in the scale-free interval N α (ε) can be written in exponential form:

[0144] N α (ε)~ε -f(α)

[0145] Among them, f(α) represents the fractal dimension corresponding to the singular index α, and the smaller f(α), the greater N α (ε) is smaller. Since the number of boxes with singular index is difficult to calculate directly, we introduce the partition function to calculate it. The partition function of multifractal is defined as χ q (ε), the formula is:

[0146] χ q (ε)=∑P i (ε) q =ε τ(q)

[0147] Where q is the weight factor and τ(q) is the quality index. The value of q should be in [-∞, +∞]. However, in actual calculation, the value of q cannot be infinite. Therefore, the present invention takes the value of q to be [-20, 30] with an interval of 1.

[0148] When ε→0, τ(q) can be written as:

[0149]

[0150] By connecting the three parameters α, f(α) and τ(α) and based on the Legendre transformation relationship between the three, we can get the formula:

[0151]

[0152] f(α)=q·α(q)-τ(q)

[0153] α and f(a) form a multifractal spectrum. The multifractal spectrum width Δα is defined as:

[0154] Δα=α max -α min

[0155] Among them, αmin and α max are the minimum singular index and the maximum singular index respectively. Δα represents the unevenness of the probability measure of the sequence, which corresponds to the fluctuation range of the surface profile height. The larger the fluctuation range, the larger Δα.

[0156] The roughness, maximum peak-to-valley height, root mean square, recursion law, fractal dimension and multifractal spectrum width of the surface profile of the three raceway positions are combined by the above method to construct a labeled hybrid feature set.

[0157] Then, dimensionality reduction is performed. The dimensionality reduction method is the random forest algorithm based on out-of-bag data (OOB). The specific process is as follows:

[0158] Create a total dataset D consisting of M samples and N features. Independently sample k times from the dataset, using bagging resampling for each sampling. Each time, m samples are randomly selected to form the training set S, thus forming k independent training datasets. The data not sampled during the entire sampling process is called out-of-bag data, or OOB data.

[0159] Generate a corresponding single decision tree for any training data set, and generate a total of k decision trees. Each decision tree is verified by the OOB data set prediction, and the prediction result is Y p , the true value is Y, and the mean square error between the true value and the predicted value is recorded as ε mse .

[0160] ·For a feature n among the N features of the out-of-bag data i (i=1,2,…,N) Add noise interference to generate a new test set, and recalculate the mean square error between the true value and the predicted value, which is recorded as

[0161] Characteristic variable n i The importance of the corresponding single decision tree is recorded as mse i , whose value is

[0162] Traverse the random forest formed by k decision trees to obtain the feature variable n i The importance of the entire random forest is recorded as

[0163] The importance MSE can be positive, negative, or zero. When the input feature has a strong correlation with the output, adding noise to the feature will weaken the correlation and the importance MSE will be positive. A larger positive value indicates a greater importance of the feature. When the input feature has almost no correlation with the output and adding noise does not change the prediction result, the importance MSE is zero. When adding noise increases the correlation between the input feature and the output, the importance MSE is negative.

[0164] Set the number of decision trees in the random forest to 20, import the mixed feature set in step 4 into Python, extract the features containing 95% of the original information to construct a new mixed feature set, and then achieve dimensionality reduction.

[0165] A support vector machine model based on genetic algorithm optimization is established. First, the mixed feature set of the two raceway positions is normalized and imported into the support vector machine model. Then, the optimal penalty factor c and kernel function parameter g are selected through genetic algorithm to complete the model training. The specific process is as follows:

[0166] Randomly generate M individuals as the initial population, encode the parameters c and g, and set the maximum number of evolutionary iterations to N.

[0167] Evaluate the individuals in the population based on the constructed mixed feature set, calculate the fitness of each individual in the population and then assess the quality of each individual.

[0168] Decode the optimal individual of the current group and determine whether its fitness meets the conditions or whether it reaches the maximum number of evolutionary iterations of the population. If so, go to the last step, otherwise go to the next step.

[0169] Parameters c and g are encoded again, and selection, crossover, and mutation operations are performed on the population. Individuals with high fitness in step 3 are selected to a high degree, and the crossover and mutation probabilities are set to obtain a new population. Then, go to step 3.

[0170] Get the best parameters c and g.

[0171] This application sets the accuracy of the state recognition model as fitness, and its formula is:

[0172]

[0173] Where s is the number of data samples correctly classified, and t is the total number of data samples.

[0174] Finally, the mixed feature set of the third raceway position is normalized and imported into the trained support vector machine model to realize the identification of the wear state of the ball screw pair. The overall flow chart is as follows: Figure 4 shown.

[0175] The method of the present invention is described below by way of an embodiment.

[0176] The test screw was selected as the GD4010 series screw produced by Bot Precision Co., Ltd. in Shandong, China. The main parameters are shown in Table 1.

[0177] Table 1 Ball screw pair parameters

[0178]

[0179] The raceway surface profile data collection is continued until fatigue peeling occurs on the screw raceway surface. At this time, the ball screw pair has run a total of 1.42 million revolutions. Figure 5 The raceway surface profile is processed by Gaussian filtering and the shape is removed by a fifth-order polynomial. Then, six features are extracted according to the feature extraction method mentioned above, which show the changes of the six features with the rotation speed of the ball screw pair.

[0180] Then, the wear state of the ball screw pair is divided according to the change of the preload force during the operation of the ball screw pair, such as Figure 6 As shown in the figure, through curve fitting, it can be found that the preload shows an obvious three-stage change trend, representing the running-in wear, stable wear and rapid wear states, which are used as labels.

[0181] Then the labeled mixed feature set is imported into the random forest evaluation algorithm for dimensionality reduction, and the obtained model scoring results are as follows: Figure 7 As shown in the figure, their importance, from high to low, is root mean square (RMS), roughness, recurrence law, maximum peak-to-valley height, fractal dimension, and multifractal spectral width. It can be seen that the first five features contain more than 95% of the original information, so the multifractal spectral width is discarded and a new hybrid feature set is constructed from the five features of roughness, maximum peak-to-valley height, RMS, recurrence law, and fractal dimension.

[0182] After the support vector machine model based on genetic algorithm optimization is established, the mixed feature sets of the three raceway positions are normalized. Then the reduced dimensional mixed feature sets constructed from the raceway surface profiles at positions 1 and 2 are imported into the model training. The results of the genetic algorithm parameter optimization are as follows: Figure 8 As shown, the training accuracy is 100%.

[0183] Finally, the mixed feature set of the three-way position is imported into the trained model to realize the wear state recognition of the ball screw pair. The confusion matrix of the recognition is as follows: Figure 9 As shown, its recognition accuracy is 96.7%.

[0184] In order to verify the effect of random forest algorithm on dimensionality reduction, the mixed feature set without dimensionality reduction was imported into the support vector machine model optimized by genetic algorithm. The confusion matrices of the training process and the recognition process are as follows: Figure 10 and Figure 11 As shown in the figure, the training accuracy of the model before and after dimensionality reduction is consistent, both reaching 100%. However, the recognition accuracy before dimensionality reduction is only 90%, while the accuracy after dimensionality reduction is increased by nearly 7%. This shows that it is very necessary to evaluate dimensionality reduction through the random forest algorithm.

[0185] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only illustrative of the principles of the present invention. Without departing from the spirit and scope of the present invention, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A method for identifying the wear state of a ball screw pair based on the raceway surface profile, characterized in that: The method comprises the following steps: Step 1: Collect the surface profile curves of the ball screw pair's ball screw raceway at three raceway positions, and perform shape removal and Gaussian filtering on the profiles; Step 2, determine the wear status of the ball screw pair and use it as a label; Step 3: Comprehensive statistical analysis, recursive analysis, and fractal analysis methods are used to extract the main features of the raceway surface profile, including roughness, maximum peak-to-valley height, root mean square, recursive law, fractal dimension, and multifractal spectrum width, and construct a labeled hybrid feature set; Step 4: Evaluate the importance of each feature in the hybrid feature set, then sort the features in the hybrid feature set in descending order according to their importance, and extract the top m features that contain more than P% of the original information to construct a new hybrid feature set; Step 5: Establish a support vector machine model based on genetic algorithm optimization, normalize the new mixed feature set and extract the mixed feature set of the two raceway positions, then import it into the established model for training, select the optimal penalty factor c and kernel function parameter g according to the genetic algorithm, and complete the model training; Step 6: Import the mixed feature set of another raceway position into the trained model to identify the wear state of the raceway position, and compare it with the actual state to obtain the accuracy of the model; Step 7: For the ball screw pair to be identified, execute steps 1, 3, and 4 to obtain its unlabeled mixed feature set, and then use the trained model to identify the wear status of the ball screw pair.

2. The method for identifying the wear state of a ball screw pair based on the raceway surface profile according to claim 1, characterized in that: Step 1 specifically includes: Select three raceway positions of the lead screw in the uniform speed running area and mark them; The ball screw pair was placed on a wear test bench. During the first 300,000 revolutions of the ball screw pair, the test bench was stopped and the ball screw pair was removed every 30,000 revolutions. Surface profile curves were collected at three raceway positions using a Taylor Hobson profilometer. After the ball screw pair had run for 300,000 revolutions, the test bench was stopped and the ball screw pair was removed every 60,000 revolutions to collect surface profile curves. The surface profile curve is smoothed by Gaussian filtering, and then the fifth-order polynomial method is used to remove the shape of the smoothed curve, and finally the required raceway surface profile curve is derived.

3. The method for identifying the wear state of a ball screw pair based on the raceway surface profile according to claim 1, characterized in that: Step 2 is as follows: Taking the turning point of the preload change trend of the ball screw pair as the dividing point, the wear state of the ball screw pair is divided into running-in wear, stable wear and rapid wear states in sequence; The wear status of the ball screw pair is determined based on the change trend of the current preload force.

4. The method for identifying the wear state of a ball screw pair based on the raceway surface profile according to claim 1, characterized in that: Step 3 specifically includes: Step 3-1, calculate the roughness, maximum peak-to-valley height, and root mean square characteristics of the raceway surface profile curve using statistical analysis methods. The formula used is: Rz=z max -With min Where Ra is the roughness, Rz is the maximum peak-to-valley height, Rms is the root mean square, and z i is the height of the i-th contour sampling point, z min and z max are the minimum profile height and the maximum profile height, is the mean height of the contour, n is the sampling rate; Step 3-2, use the recursive analysis method to solve the recursive law of the raceway surface profile. The formula used is: r ij =|z i -z j | ε=0.5σ R ij (ε)=θ(ε-r ij ) Among them, RR is the recursive law, z i and z j are all contour heights, i,j=1,2,...,n,r ij Represents the distance between any two points i and j, R ij is an element of the matrix, σ is the standard deviation, ε is the threshold, and θ(x) is the Heaviside function; Step 3-3, solving the fractal dimension of the raceway surface profile, the process includes: The WM function is used to characterize the nonlinear raceway surface profile. The formula is: Among them, 1<D<2, γ>1, D is the fractal dimension, G is the height scale coefficient, γ n is the spectrum of the rough surface, n is the sampling rate, n1 is the minimum sampling rate, L is the sampling length, usually γ = 1.5, z(x) is the height of the random contour, and x is the position coordinate of the contour; The power spectrum function of the WM function is expressed as: The incremental variance of z(x) is defined as the structure function, as shown below: Where τ = nΔL, ΔL is the sampling interval; Combining the above formulas, we can get: Where C = Γ(2D-3)sin((D-1.5)π) / (4-2D)lnγ, Γ(*) is the gamma function, Taking the logarithm of both sides of the above equation: lgS(τ)=(4-2D)lgτ+lgC+2(D-1)lgG According to the above formula, the fractal dimension D can be: Where k is the slope of the straight line; Step 3-4, the box counting method is used to calculate the multifractal spectrum of the raceway surface profile. The minimum value of the raceway surface profile data is extracted as the lower limit zero to ensure that the amplitude is all positive. Many small boxes with a size of ε' are used to cover the raceway profile, 0<ε′<1, and the total profile height S i (ε) represents the sum of all contour amplitudes in the i-th small box when the box size is ε, then the probability measure P i (ε) is defined as: Among them, ∑S i (ε) is the sum of the amplitudes of all raceway profile data; In the scale-free interval, P i (ε) can also be written in exponential form: P i (e)~e α Among them, α is the singular index, which is used to reflect P i The singular strength of (ε); Let the number of boxes with the same singular index α be N α (ε), then in the scale-free interval N α (ε) is written in exponential form: N α (e)~e -f(α) Among them, f(α) represents the fractal dimension corresponding to the singular index α, and the smaller f(α), the greater N α (ε) is smaller; Define the partition function of the multifractal as χ q (ε), the formula is: x q (e)=∑P i (e) q =e τ(q) Where q is the weight factor and τ(q) is the quality index; When ε→0, τ(q) is written as: Combining the three parameters α, f(α) and τ(α) and according to the Legendre transformation relationship between the three, we can get: f(α)=q·α(q)-τ(q) α and f(a) form a multifractal spectrum; The multifractal spectral width Δα is defined as: Dα=α max -a min Among them, α min and α max are the minimum singular index and the maximum singular index respectively. Δα represents the unevenness of the probability measure of the sequence, which corresponds to the fluctuation range of the surface profile height. The larger the fluctuation range, the larger Δα is. In steps 3-5, the roughness, maximum peak-to-valley height, root mean square, recurrence law, fractal dimension, and multifractal spectrum width of the raceway surface profile at the three raceway positions are combined with the label of step 2 to form a labeled hybrid feature set.

5. The method for identifying the wear state of a ball screw pair based on the raceway surface profile according to claim 1, characterized in that: In step 4, the random forest algorithm based on out-of-bag data is used to evaluate the importance of each feature in the mixed feature set. The specific process includes: Establish a total data set D, which consists of M samples and N = 6 features; independently sample k times from the data set, each sampling method is bagging resampling, and m samples are randomly selected each time to form a training set S, that is, k independent training data sets are formed. The data not sampled during the entire sampling process is called out-of-bag data, or OOB data; Generate a corresponding single decision tree for any training data set, and generate a total of k decision trees. Each decision tree is verified by the OOB data set prediction, and the prediction result is Y p , the true value is Y, and the mean square error between the true value and the predicted value is recorded as ε mse ; For a feature n among the N features of the out-of-bag data i Add noise interference, generate a new test set, and recalculate the mean square error between the true value and the predicted value, recorded as The characteristic variable n i The importance of the corresponding single decision tree is recorded as mse i , whose value is Traverse the random forest formed by k decision trees to obtain the feature variable n i The importance of the entire random forest is recorded as 6. The method for identifying the wear state of a ball screw pair based on the raceway surface profile according to claim 1, characterized in that: Step 5 extracts the mixed feature set of the two raceway positions from the new mixed feature set and performs normalization processing, and then imports it into the established model for training. The optimal penalty factor c and kernel function parameter g are selected according to the genetic algorithm to complete the model training, which specifically includes: Step 5-1, randomly generate M1 individuals as the initial population, encode the parameters c and g, and set the maximum number of evolutionary iterations to N1; Step 5-2, calculate the fitness of each individual in the group; Step 5-3: Decode the optimal individual of the current group and determine whether its fitness meets the conditions or reaches the maximum number of evolutionary iterations of the population. If so, go to step 5-5; otherwise, go to step 5-4. Step 5-4: Encode the parameters c and g again, perform selection, crossover and mutation operations on the population to obtain a new population, and go to step 5-3; Step 5-5, output the optimal parameters c and g.

7. The method for identifying the wear state of a ball screw pair based on the raceway surface profile according to claim 6, characterized in that: The fitness in step 5-2 is the accuracy of identifying the wear state of the ball screw pair, and the formula is: Where Fitness is the fitness, s is the number of data samples correctly identified, and t is the total number of data samples.

Images