Ball screw pair fault identification method and system based on energy entropy and ANFIS

Abstract

The invention discloses a ball screw pair fault identification method and system based on energy entropy and ANFIS, and relates to the technical field of detection, and the method comprises the steps: obtaining fault sample data, carrying out the preprocessing of the fault sample data, carrying out the decomposition through employing an ICEEMDAN algorithm, and obtaining a plurality of IMF modal components; extracting fault features in the IMF modal component, and constructing a fault feature matrix vector of the vibration signal of the ball screw pair based on the fault features; and inputting the fault feature matrix vector into a preset ANFIS model to realize fault type identification of the ball screw pair. According to the invention, different fault types of the ball screw pair can be identified.

Description

Technical Field

[0001] The present invention relates to the field of detection technology, and more specifically to a ball screw pair fault identification method and system based on energy entropy and ANFIS. Background Art

[0002] As a mechanical transmission component, ball screw pairs are widely used in various industrial devices and precision instruments. They efficiently convert rotational motion into linear motion, and have the advantages of high transmission accuracy, high efficiency, and long life. They occupy an important position in the transmission field.

[0003] In recent years, there has been more and more research on the performance and dynamic analysis of ball screw pairs, and the reliability of ball screw pairs has been continuously improved to improve the processing accuracy of CNC machine tools. However, in the long-term operation of ball screws, continuous acceleration and deceleration, emergency stops, and even insufficient lubrication will continue to increase the wear of the ball and raceway surface, resulting in an increase in the gap between the ball and the raceway, which directly affects the positioning accuracy of the feed system, thereby affecting the product quality of the processed parts, accelerating the wear and damage of the tool, and causing failures, resulting in huge economic losses to the company. These failures will not only hinder the normal operation of the equipment, but may even cause the entire production line to stop working, causing huge economic losses to the company.

[0004] Therefore, fault diagnosis of ball screw pairs to ensure their stable and reliable operation has become a hot topic in the current mechanical field. Summary of the invention

[0005] In view of this, the present invention provides a ball screw pair fault identification method and system based on energy entropy and ANFIS, which can realize real-time identification of ball screw pair faults.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] The ball screw pair fault identification method based on energy entropy and ANFIS includes:

[0008] Acquire fault sample data, preprocess the fault sample data, decompose the data using the ICEEMDAN algorithm, and acquire a number of IMF modal components;

[0009] Extracting fault features from the IMF modal components, and constructing a fault feature matrix vector of the ball screw pair vibration signal based on the fault features;

[0010] The fault feature matrix vector is input into a preset ANFIS model to realize the identification of the ball screw pair fault type.

[0011] Preferably, the obtaining of a plurality of IMF modal components specifically includes: obtaining a first residual: Then calculate the first IMF modal component based on the first residual Among them, M(.) is the local mean calculated from the sequence x(t), and N is the number of means.

[0012] Preferably, the extraction of fault features in the IMF modal components specifically includes: calculating the energy entropy of each modal component as a frequency domain feature, simultaneously calculating the time domain index of the vibration signal, and obtaining the time domain feature by reducing the dimension through principal component analysis, combining the time domain feature and the frequency domain feature, and constructing a fault feature matrix vector of the ball screw pair vibration signal.

[0013] Preferably, the time domain indicators include: mean value, standard deviation, absolute error, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, crest factor, waveform factor and impulse factor.

[0014] Preferably, the calculating the energy entropy of each modal component specifically includes:

[0015]

[0016] Among them, p i =E i / E represents the i-th IMF modal component E i The energy in the total energy The proportion of i=1,...,n,H EM is the energy entropy value.

[0017] The ball screw fault identification system based on energy entropy and ANFIS includes:

[0018] A modal component acquisition module acquires fault sample data, pre-processes the fault sample data, decomposes the data using the ICEEMDAN algorithm, and acquires a number of IMF modal components;

[0019] A matrix vector construction module extracts fault features in the IMF modal components and constructs a fault feature matrix vector of the ball screw pair vibration signal based on the fault features;

[0020] The identification module inputs the fault feature matrix vector into a preset ANFIS model to realize the identification of the ball screw pair fault type.

[0021] It can be known from the above technical solutions that, compared with the prior art, the present invention discloses a ball screw pair fault identification method and system based on energy entropy and ANFIS, which uses energy entropy to calculate the characteristics of the IMF component. Energy entropy can quantify the significant characteristics of the signal in the time-frequency domain, and the difference is obvious under different fault conditions. The fault feature vector is constructed in combination with time domain indicators. This multi-dimensional feature extraction method comprehensively and accurately captures the fault information of the ball screw pair, effectively solves the complex problem of fault feature extraction, and can better reflect the essential characteristics of the fault state compared to a single feature extraction method. The ANFIS model itself has self-learning and adaptive capabilities, combining the advantages of fuzzy logic and neural networks. In the fault identification process, the fuzzy rules and membership functions can be automatically adjusted through training, and different fault types of the ball screw pair have a high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0023] Figure 1 ANFIS structure diagram provided by the present invention;

[0024] FIG. 2( a ) is a normal state diagram provided by the present invention;

[0025] FIG2( b ) is a ball failure diagram provided by the present invention;

[0026] FIG2( c ) is a screw failure diagram provided by the present invention;

[0027] FIG2( d ) is a nut failure diagram provided by the present invention;

[0028] Figure 3 A fault identification flow chart provided by the present invention;

[0029] Figure 4 The principal component contribution rate diagram provided by the present invention;

[0030] Figure 5 ANFIS test result diagram provided by the present invention;

[0031] Figure 6 The SVM test result diagram provided by the present invention;

[0032] Figure 7 This is a KNN test result diagram provided by the present invention. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] The embodiment of the present invention discloses a ball screw pair fault identification method based on energy entropy and ANFIS, and the specific steps are as follows: Figure 3 As shown, including:

[0035] Obtain fault sample data, preprocess the fault sample data, decompose it using the ICEEMDAN algorithm, and obtain several IMF modal components;

[0036] Extract the fault features in the IMF modal components and construct the fault feature matrix vector of the ball screw pair vibration signal based on the fault features;

[0037] The fault feature matrix vector is input into the preset ANFIS model to realize the identification of the ball screw pair fault type.

[0038] Among them, 12 indicators refer to time domain and frequency domain characteristic indicators, including mean value, standard deviation, absolute error, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, crest factor, waveform factor, impulse factor, and energy entropy.

[0039] In a specific embodiment, obtaining a plurality of IMF modal components specifically includes: obtaining a first residual: Then calculate the first IMF modal component based on the first residual Among them, M(.) is the local mean calculated from the sequence x(t), and N is the number of means.

[0040] ICEEMDAN is an improved empirical mode decomposition (EMD) method that aims to solve the endpoint effect and mode aliasing problems existing in the traditional EMD method. This method can more effectively decompose complex signals into a series of intrinsic mode function (IMF) components by cleverly introducing adaptive noise and integration strategies. Each IMF component reflects the characteristics of the signal at different time scales. The specific steps of decomposition are as follows:

[0041] First, we need to obtain the first residual, which can be calculated by formula (1):

[0042]

[0043] M(.) is the local mean calculated from the sequence x(t).

[0044] From this, we can get As shown in formula (2):

[0045]

[0046] Repeat the calculation for the kth residual and The calculation of is shown in equations (3) and (4).

[0047]

[0048] by Take for example, as shown in formula (5).

[0049]

[0050] Where ε represents a coefficient related to the signal-to-noise ratio (SNR).

[0051] When a fault occurs in different parts of the ball screw pair, the frequency distribution of the collected vibration signal will also change, and the energy distribution of the fault vibration signal will also change accordingly. Therefore, the ICEEMDAN decomposition method can be used to decompose the vibration signal of the ball screw pair and calculate the energy entropy of each IMF component. This energy entropy can be used as a feature vector to determine the type of fault in the ball screw pair.

[0052] In a specific embodiment, the preset ANFIS model adopts the ANFIS algorithm. The adaptive fuzzy neural network (ANFIS) is an intelligent algorithm that combines the advantages of fuzzy logic and neural network, and has self-learning and self-adaptive capabilities. Compared with traditional neural networks, the outstanding advantage is that it is more convenient and more efficient. The general ANFIS structure is as follows: Figure 1 As shown:

[0053] The fuzzy neural network model uses the feature space relationship provided during the training process to generate rules, which are the reasoning and judgment rules of the fuzzy neural network. Generally, the adaptive fuzzy neural network has a total of five layers:

[0054] The first layer is the fuzzy layer. The eigenvalues ​​extracted from the processed vibration signal are directly input into the fuzzy layer for calculation. The output is:

[0055]

[0056] In formula (6), A i is the i-th characteristic component of the vibration signal i=(1,2,…n), representing a series of fuzzy sets, each of which is associated with the degree of a linguistic variable at the node, i represents the node, and x represents the input variable of node i; Indicates the membership value corresponding to the i-th node in the first layer; u AiRepresents the membership function of x. This degree can be such as "high", "medium", or "low", etc. It is the membership degree belonging to the set, describing the degree of belonging. Common membership functions include Gaussian membership function (Gaussfm), bell-shaped membership function (Gbellmf), and S-shaped membership function (Sigmf), etc. The bell-shaped membership function (Gbellmf) selected in the present invention has the formula shown in Equation (7):

[0057]

[0058] Where:

[0059] 1) μ(x) represents the membership degree of element x belonging to a certain fuzzy set, and its value range is [0, 1].

[0060] 2) x is the input variable, representing the element whose membership degree is to be evaluated.

[0061] 3) c is the center (mean value) of the Gaussian function, representing the center point or the most typical member of the fuzzy set.

[0062] 4) σ is the standard deviation of the Gaussian function, which determines the width and smoothness of the membership function curve.

[0063] The second layer is the rule layer, and its main function is to calculate the activation degree of each fuzzy rule according to the output signal of the membership function layer. These activation degrees reflect the degree to which the input data satisfies each fuzzy rule.

[0064]

[0065] Where, w i represents the triggering intensity; u represents the membership function; A ij represents multiplying the output signal A i , A j to calculate the triggering intensity of each rule;

[0066] The third layer is the weight assignment layer, also called the normalization layer. The main function of this layer is to normalize the triggering intensity of each rule output by the second layer, so as to obtain the relative importance or triggering proportion of each rule in the final decision.

[0067] Its formula is:

[0068]

[0069] Where, represents normalizing the output data of the previous layer to obtain the triggering proportion of each feature on this rule.

[0070] The fourth layer is the clarification layer. The main task of this layer is to convert the results of fuzzy reasoning into a clear, single output value. Since the output of the fuzzy rule layer is the membership value or trigger strength of the fuzzy set, and a specific value is often required as the output in practical applications, the clarification layer is responsible for completing this conversion process. There are many ways to implement the clarification layer, the most common of which is the weighted average method, and its formula is:

[0071]

[0072] The fifth layer reduces the dimension of the results by summing them up to obtain the sum of a single feature in each rule.

[0073]

[0074] In a specific embodiment, extracting fault features from IMF modal components specifically includes: calculating the energy entropy of each modal component as a frequency domain feature, calculating the time domain index of the vibration signal at the same time, and obtaining the time domain feature through dimensionality reduction by principal component analysis, combining the time domain feature and the frequency domain feature, and constructing a fault feature matrix vector of the ball screw pair vibration signal.

[0075] In a specific embodiment, the time domain indicators include: mean value, standard deviation, absolute error, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, crest factor, crest factor and impulse factor.

[0076] In a specific embodiment, calculating the energy entropy of each modal component specifically includes:

[0077]

[0078] Among them, p i =E i / E represents the i-th IMF modal component E i The energy in the total energy The proportion of i=1,...,n,H EM is the energy entropy value.

[0079] Four types of ball screw pair fault vibration acceleration signals are selected for analysis. By performing ICEEMDAN decomposition on the original vibration acceleration signal of the ball screw pair, several IMF components can be obtained, such as Figure 2(a)-Figure 2(d) As shown in Table 1, the entropy values ​​corresponding to different fault positions of the ball screw pair are shown.

[0080] Table 1 Entropy value table

[0081]

[0082] As shown in Table 1, under normal operating conditions, the ICEEMDAN energy entropy of the ball screw pair is higher than that of the other three abnormal states. This is because, under normal conditions, the vibration signal of the ball screw pair tends to be stable, and its energy distribution is relatively uniform and uncertain. On the contrary, when a ball, screw or nut fails, a resonant frequency will be generated within a specific frequency range, causing the energy to concentrate in these frequency bands, thereby reducing the uncertainty of energy distribution and causing the entropy value to decrease accordingly. Generally speaking, when a fault occurs, the faulty part will interfere with the vibration signal and resonate and demodulate with the vibration signal of the screw. This effect will have different degrees of impact on the stability of the vibration acceleration signal under the three fault conditions. As the distance between the screw and the fault point increases, this impact on stability will gradually weaken, which is reflected in the entropy value as the screw fault has the greatest impact, followed by the ball fault, and finally the nut fault.

[0083] Combining the advantages of ICEEMDAN in the field of signal decomposition and the significant differences in energy entropy under different fault conditions, a new method for ball screw pair fault identification combining ICEEMDAN energy entropy with ANFIS model is proposed. The preprocessed ball screw pair fault vibration signal is first decomposed by ICEEMDAN, and multiple modal components are obtained after decomposition. The modal components with a large energy share are selected from them to obtain the frequency domain characteristic energy entropy of the signal; then 11 time domain indicators such as the average value of the fault signal are calculated, and the main time domain features of the fault signal are obtained through principal component analysis (PCA) dimensionality reduction. All the features after dimensionality reduction constitute the fault feature matrix as the input of the ANFIS model, thereby realizing the classification and identification of faults.

[0084] The ball screw fault identification system based on energy entropy and ANFIS includes:

[0085] The modal component acquisition module acquires fault sample data, pre-processes the fault sample data, decomposes it using the ICEEMDAN algorithm, and acquires several IMF modal components;

[0086] The matrix vector construction module extracts the fault features in the IMF modal components and constructs the fault feature matrix vector of the ball screw pair vibration signal based on the fault features;

[0087] The identification module inputs the fault feature matrix vector into the preset ANFIS model to realize the identification of the ball screw pair fault type.

[0088] In order to conduct the fault diagnosis experiment of the ball screw pair, four states are selected: normal state, ball fault, screw fault and nut fault. In order to facilitate the collection of vibration signals, a test platform for collecting vibration signals of ball screw pairs is constructed. The experimental platform includes three core components: the platform body, the drive control system and the signal acquisition system. During the experiment, the YMC121A100 acceleration sensor, YMC9216 signal collector and YMC9800 software are used for signal analysis. The acceleration performance parameter indicators and the specifications of the collector are shown in the table:

[0089] Table 2 Signal collector parameter

[0090] Indicator parameters Specifications Number of channels 8 (Synchronous sampling) Resolution 24b Measurement Type Voltage, full bridge Sampling rate 250ks / s Analog-to-digital conversion Analog + anti-aliasing filtering Operating temperature -10℃ to 50℃ Voltage range ±10V to ±1V, ±100mV

[0091] Table 3 Acceleration sensor specifications

[0092] Indicator name Technical specifications Sensitivity 100mv / g Range <![CDATA[450 / s 2 ]]> quality 9.5g Operating temperature -40 to 121°C Frequency Response 1-8kHz Current range 2-10mA Voltage range 18-30v

[0093] For each fault type, the present invention selects 20 groups of data to form a training sample set, totaling 80×4 groups of data. For these samples, 11 time domain feature indicators are calculated, and the frequency domain feature energy entropy is further extracted with the help of ICEEMDAN decomposition. Therefore, under each fault state, 20 groups of data are obtained, each with 12 eigenvalues. Fault data sets. Taking the vibration data of ball bearing fault as an example, some of its eigenvalues ​​are shown in Table 4. It can be observed from the data in Table 4 that even for the 10 groups of data for the ball bearing fault state, the amount of feature information is quite rich. However, this high-dimensional feature vector may not only bring about the problem of information overlap in the time domain indicators, but also increase the complexity of the calculation process, thereby prolonging the time required for fault diagnosis. In order to improve operating efficiency and reduce the problem of information overlap, principal component analysis (PCA) is used to perform nonlinear dimensionality reduction on the time domain indicators. This effectively reduces the data dimension and retains the features that contribute the most in the data set. The results are as follows. Figure 4 shown.

[0094] Table 4 Characteristic values ​​of ball failure

[0095]

[0096] according to Figure 4As shown in the figure, the first three principal components have the highest contribution rate, reaching 99.6%. Therefore, the original features corresponding to the three principal components with the highest contribution rate are selected as the final time domain features, and combined with the frequency domain feature of energy entropy, the database required for vibration fault diagnosis is jointly constructed. By recombining these four eigenvectors, a matrix containing 4×80 eigenvalues ​​can be constructed. In this matrix, each sample consists of four eigenvalues, which together describe the eigenvector of a specific state. There are a total of 80 such eigenvectors in four different states. Subsequently, these 80 eigenvectors are input into the ANFIS model as a training data set to further realize the identification and diagnosis of the working state of the ball screw pair. At the output end, clear labels are set, as shown in Table 5:

[0097] Table 5 Output terminal label

[0098]

[0099] When the test sample is input into the optimized ANFIS model, by adjusting the parameters and membership functions, the output results of the model will be judged according to the following criteria: if the output value is in the interval [0.5, 1.5], it is judged to be in a normal state; if the output value is in the interval [1.5, 2.5], it is considered to be a ball fault; if the output value is in the interval [2.5, 3.5], it is judged to be a screw fault; if the output value is in the interval [3.5, 4.5], it is considered to be a nut fault

[16] .

[0100] Figure 5 The test results of the trained ANFIS model in four states are shown, where "○" represents the actual target value and "*" represents the model predicted value. As can be seen from the figure, within the allowable error range, the recognition accuracy of the model in each state reaches 100%, indicating that the trained ANFIS model has good prediction accuracy.

[0101] In order to better reflect the good fault identification effect of ANFIS model, the feature vectors are input into SVM (support vector machine) model and KNN (K-nearest nearest neighbor algorithm) model for comparative analysis. Similarly, "○" represents the actual target output and "*" represents the model predicted output. The results are as follows: Figure 6 , Figure 7 .

[0102] Depend on Figure 6 It can be seen that the SVM model has good overall recognition, and there is a group of recognition errors in the normal state, the screw fault state, and the nut fault state. Figure 7 It can be seen that the overall recognition of the KNN model is relatively good. There are two groups of recognition errors in the normal state and two groups of recognition errors in the ball fault state. The recognition results of the three faults are plotted in Table 6 for comparison.

[0103] Comparison of Test Results in Table 6

[0104]

[0105] As can be seen from Table 3, when 80 groups of fault data are used for identification, within a certain error range, the identification rate of the ANFIS model can reach 100%; the SVM model has three identification errors, and the overall identification rate can reach 96.25%; the KNN model has six identification errors, and the overall identification rate can reach 94.94%. All three models have achieved good identification effects, but the identification effect of the ANFIS model is more significant.

[0106] A fault identification method for ball screw pairs based on ICEEMDAN energy entropy and ANFIS analyzes the fault data of ball screw pairs, and uses a comparative experiment to verify the effectiveness and accuracy of the ANFIS model, obtaining the following conclusions:

[0107] 1) Under different fault states, the frequency distribution and energy distribution of the vibration signals of ball screw pairs are different, and their corresponding energy entropies have obvious differences. In this paper, the energy entropy of each IMF component is calculated, and the energy entropy of the main IMF components is extracted and combined with time-domain indicators to form a fault feature vector, which can effectively identify the fault states of ball screw pairs.

[0108] 2) Using the ANFIS model for fault identification, this model effectively combines fuzzy logic and neural networks. Through the forward propagation and backpropagation learning algorithms, it can automatically adjust the fuzzy rules and membership functions. Compared with the SVM model and the KNN model, the ANFIS model has a better fault identification effect.

[0109] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0110] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A ball screw pair fault identification method based on energy entropy and ANFIS, characterized in that: include: Acquire fault sample data, preprocess the fault sample data, decompose the data using the ICEEMDAN algorithm, and acquire a number of IMF modal components; Extracting fault features from the IMF modal components, and constructing a fault feature matrix vector of the ball screw pair vibration signal based on the fault features; The fault feature matrix vector is input into a preset ANFIS model to realize the identification of the ball screw pair fault type.

2. The ball screw pair fault identification method based on energy entropy and ANFIS according to claim 1 is characterized in that: The obtaining of a plurality of IMF modal components specifically includes: obtaining a first residual: Then calculate the first IMF modal component based on the first residual Among them, M(.) is the local mean calculated from the sequence x(t), and N is the number of means.

3. The ball screw pair fault identification method based on energy entropy and ANFIS according to claim 1 is characterized in that: The method of extracting fault features from the IMF modal components specifically includes: calculating the energy entropy of each modal component as a frequency domain feature, calculating the time domain index of the vibration signal at the same time, and obtaining the time domain feature by reducing the dimension through principal component analysis, combining the time domain feature and the frequency domain feature, and constructing a fault feature matrix vector of the ball screw pair vibration signal.

4. The ball screw pair fault identification method based on energy entropy and ANFIS according to claim 3 is characterized in that: The time domain indicators include: mean value, standard deviation, absolute error, kurtosis, maximum value, minimum value, peak-to-peak value, root mean square, crest factor, waveform factor and impulse factor.

5. The ball screw pair fault identification method based on ICEEMDAN energy entropy and ANFIS according to claim 3 is characterized in that: The calculation of the energy entropy of each modal component specifically includes: Among them, p i =E i / E represents the i-th IMF modal component E i The energy in the total energy The proportion of i=1,...,n,H EM is the energy entropy value.

6. A ball screw pair fault identification system based on energy entropy and ANFIS, applying the ball screw pair fault identification method based on energy entropy and ANFIS as described in any one of claims 1 to 5, characterized in that: include: A modal component acquisition module acquires fault sample data, pre-processes the fault sample data, decomposes the data using the ICEEMDAN algorithm, and acquires a number of IMF modal components; A matrix vector construction module extracts fault features in the IMF modal components and constructs a fault feature matrix vector of the ball screw pair vibration signal based on the fault features; The identification module inputs the fault feature matrix vector into a preset ANFIS model to realize the identification of the ball screw pair fault type.

Images