RV reducer gear weak fault diagnosis method based on adaptive MCKD

Abstract

The invention relates to the technical field of mechanical state monitoring, solves the technical problem that the effect of extracting weak fault feature information is poor due to the fact that MCKD input parameters depend on artificial experience selection, and particularly relates to an adaptive MCKD-based RV reducer gear weak fault diagnosis method. The method comprises the following steps: acquiring an original signal x of an RV reducer gear in an operation state; and adaptively determining an optimal diagnosis parameter of the maximum correlation kurtosis deconvolution MCKD under the optimal fitness function value by adopting an SCSSA search algorithm. The method is used for RV reducer gear fault diagnosis under different working conditions, the advantages of MCKD in the aspect of fault feature extraction are fully utilized, the effectiveness of the proposed method is proved through case research, the classification and recognition accuracy of the method reaches 97.5% or above, and efficient RV reducer fault diagnosis is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of mechanical state monitoring, and in particular to a method for diagnosing weak faults of RV reducer gears based on adaptive MCKD. Background Art

[0002] In the actual operation of the rotating vector (RV) reducer, the impact characteristics generated when a local fault occurs in the RV reducer gear are very weak. Due to the influence of the working environment and external factors, it is difficult to accurately extract the fault characteristics. Considering that the collected signal is affected by strong noise and complex transmission paths, the deconvolution method is considered to be an effective tool for extracting periodic fault pulses.

[0003] At present, the Maximum Correlated Kurtosis Deconvolution (MCKD) algorithm can enhance the impulse characteristics of the signal under low signal-to-noise ratio conditions, thereby improving the kurtosis value of the original signal and deconvolving the periodic pulse components from the fault signal, effectively extracting weak fault feature information. However, the effect of the MCKD algorithm is mainly determined by three input parameters: filter length L, deconvolution period T, and shift number M. The rationality of parameter setting has a crucial impact on the deconvolution effect, but currently it mainly relies on manual experience selection, which leads to low superiority in weak fault feature extraction, and it is more difficult to extract fault features under the interference of transmission path and strong noise. Therefore, a method is needed to adaptively determine the optimal combination of L, T and M in the MCKD algorithm. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method for diagnosing weak faults of RV reducer gears based on adaptive MCKD, which solves the technical problem that the MCKD input parameters rely on artificial experience selection, resulting in poor effect in extracting weak fault feature information.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for diagnosing weak faults of RV reducer gears based on adaptive MCKD, the method comprising the following steps:

[0006] S1, obtain the original signal x of the RV reducer gear in the running state;

[0007] S2, using the SCSSA search algorithm to adaptively determine the optimal diagnostic parameters of the maximum correlation kurtosis deconvolution MCKD under the optimal fitness function value;

[0008] S3, using the optimal diagnostic parameters to perform MCKD deconvolution on the original signal x, and obtain the enhanced periodic pulse component signal y representing the fault in the original signal x;

[0009] S4, decomposing the periodic pulse component signal y by ensemble empirical mode decomposition (EEMD) to obtain n modal components in different frequency bands;

[0010] S5, calculating the multi-scale fuzzy entropy of n-1 modal components and constructing a multi-scale fuzzy entropy feature set vector;

[0011] S6. Input the multi-scale fuzzy entropy feature set vector into the classifier for classification and identification to obtain weak fault diagnosis results.

[0012] Furthermore, in step S2, the optimal fitness function value is to use the maximum value of the envelope spectrum peak factor Ec as the objective function, and the optimal diagnostic parameters include the filter length L, the deconvolution period T and the filter shift number M.

[0013] Furthermore, in step S2, the specific process includes the following steps:

[0014] S21, define the initial parameters of the SCSSA search algorithm and the optimization interval of the MCKD influencing parameters;

[0015] S22, randomly generate Q parameter combinations [L, T, M] in the optimization interval as the initial population Xij, and generate the refraction reverse population X^ij on this basis;

[0016] S23, merge the initial population Xij and the refracted reverse population X^ij, calculate the fitness value of the current position of each parameter combination [L, T, M] in the merged population and sort them;

[0017] S24, select the top Q with high fitness value in the merged population i The parameter combination [L, T, M] is used as the final initial population to obtain the current optimal fitness individual and its position X best ;

[0018] S25, setting the discoverer ratio PD, the vigilant ratio SD, the safety value ST and the warning value R2, and continuously iteratively updating the positions of the discoverers and followers in the merged population;

[0019] S26. Compare the values ​​of the envelope spectrum peak factor Ec in the iterative update process, and continuously update the optimal fitness individual and the corresponding optimal position in the merged population.

[0020] If the maximum number of iterations is met, the iteration is stopped and the optimal diagnostic parameters [L, T, M] at the optimal fitness value are output;

[0021] If not, go to step S25.

[0022] Further, in step S21, defining the initial parameters of the SCSSA search algorithm includes: setting the sparrow population size N, the maximum number of iterations Z, the discoverer ratio PD, the alerter ratio SD, the safety value ST and the alert value R2;

[0023] Defining the optimization interval of the MCKD influencing parameters includes: setting the value range of the filter length L, the deconvolution period T and the filter shift number M to form the optimization interval.

[0024] Furthermore, the expression of envelope spectrum peak factor Ec is:

[0025]

[0026] Where X(N) is the envelope spectrum in [N*f r ,f s ]; f r is the frequency conversion; f s is the system sampling frequency; the value of N should be no less than 2.

[0027] Furthermore, the weak fault diagnosis results include tooth surface defects, broken teeth, tooth root cracks, and tooth surface wear.

[0028] Furthermore, in step S5, it specifically includes:

[0029] Define the fuzzy entropy of n modes and get the original sequence X of length N i ={x1,x2,…,x N};

[0030] For the original sequence X of length N i ={x1,x2,…,x N Given the embedding dimension m and similarity tolerance r in advance, a new coarse-grained vector y is established based on the original sequence j (τ), that is:

[0031]

[0032] In the formula, τ=1,2,… are scale factors. In fact, when τ=1, y j (1) is the original sequence. For non-zero τ, the original sequence X i is divided into τ coarse-grained sequences y of length N / τ each j (τ);

[0033] Calculate the fuzzy entropy of each coarse-grained sequence and plot it as a function of the scale factor, namely:

[0034] sampEn(i,j)=FuzzyEn(z,m,r,n,tau)

[0035] Where z is the one-dimensional vector to be analyzed; m is the embedding dimension, m=2; r is the similarity tolerance, r=0.15Std; n is the fuzzy power, n=2; tau is the time delay, usually 1.

[0036] Furthermore, we define the fuzzy entropy of n modes and obtain the original sequence X with a length of N. i ={x1,x2,…,x N}, specifically including the following steps:

[0037] S51. For the original time series {u(i):1≤i≤N} of length N corresponding to n modes, an m-dimensional vector is constructed in sequence, that is:

[0038]

[0039] In the formula, represents the value of the continuous dimension m original time series u starting from the i-th point; u0(i) is the mean value of the continuous dimension m original time series u starting from the i-th point;

[0040] Remove the mean u0(i), where

[0041]

[0042] S52. Definition and The distance between is the maximum value of the difference between the corresponding elements of the two, that is:

[0043]

[0044] i,j=1,2,…,Nm,i≠j

[0045] In the formula, represents the value of the original time series u with continuous dimension m starting from the jth point; u0(j) is the mean value of the original time series u with continuous dimension m starting from the jth point;

[0046] S53, through the fuzzy function Calculate vector and Similarity Right now:

[0047]

[0048] Where n is the number of modal components, n = 2; r is the similarity tolerance, r = 0.15Std;

[0049] S54. Based on similarity Define the function φ m(n,r), that is:

[0050]

[0051] S55, the function φ m The dimension of (n,r) is m+1, and the above steps S51-S54 are repeated to obtain:

[0052]

[0053] S56. Define fuzzy entropy and obtain the original sequence X of length N i ={x1,x2,…,x N}, the fuzzy entropy is defined as:

[0054]

[0055] When N is a finite number, the above formula is expressed as:

[0056] FuzzyEn(m,n,r,N)=lnφ m (n,r)-lnφ m+1 (n,r).

[0057] By means of the above technical solution, the present invention provides a method for diagnosing weak faults of RV reducer gears based on adaptive MCKD, which has at least the following beneficial effects:

[0058] 1. The present invention uses the maximum value of the envelope spectrum peak factor as the fitness function, and then adaptively determines the optimal diagnostic parameters through the SCSSA search algorithm. It not only inherits the advantages of MCKD in fault feature extraction, but also has the function of adaptive parameter adjustment, avoiding the problem of poor feature extraction performance caused by manual selection of parameters.

[0059] 2. The present invention uses the RV reducer gear weak fault diagnosis method combining MCKD with SCSSA for RV reducer gear fault diagnosis under different working conditions, fully utilizing the advantages of MCKD in fault feature extraction. Case studies have proven the effectiveness of the proposed method, and the classification and recognition accuracy of the method reaches more than 97.5%, achieving efficient RV reducer fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0061] Figure 1 It is a flow chart of the RV reducer gear weak fault diagnosis method of the present invention;

[0062] Figure 2 Iterative curve diagram of optimizing input parameters of the SCSSA search algorithm of the present invention;

[0063] Figure 3 It is a time domain diagram of the initial signal of the tooth root crack fault of the present invention;

[0064] Figure 4 It is a time domain diagram of the initial signal of the tooth root crack fault after deconvolution in the present invention;

[0065] Figure 5 is the classification recognition rate of the present invention under different operating conditions of the RV reducer gear;

[0066] Figure 6 It is the classification recognition rate of the gear fault under the working condition of 1800rpm speed and 7.32Nm load of the present invention;

[0067] Figure 7 This is a distribution diagram of the fault feature set of the RV reducer gear under different operating conditions of the present invention;

[0068] Figure 8 This is a distribution diagram of the fault feature set under the working condition of a speed of 1800 rpm and a load of 7.32 Nm according to the present invention. DETAILED DESCRIPTION

[0069] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods, so that the implementation process of how the present application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0070] Rotary vector (RV) reducer is one of the core components of industrial robots and is widely used in key parts such as robot joint drive. In RV reducer, gear plays the role of energy transmission between the motor side and the load side. It is also one of the components with the highest failure frequency in the reducer. Once failure occurs, it will seriously affect the operating accuracy and reliability of rotating mechanical equipment. If the reducer gear can be clearly diagnosed at the early stage of failure, and the faulty parts can be repaired or replaced in time by taking corresponding measures, the stable operation of the mechanical equipment can be well ensured. However, in the actual operation process, the impact characteristics generated when the RV reducer gear has a local failure are very weak, and it is difficult to accurately extract the fault characteristics due to the influence of the working environment and external factors.

[0071] Considering that the collected signal is affected by strong noise and complex transmission path, the deconvolution method is considered to be an effective tool for extracting periodic fault pulses. At present, the noise reduction method based on minimum entropy deconvolution (MED) has been promoted in practical applications. However, when the noise completely submerges the impulse signal, MED will fail. Therefore, on this basis, the maximum correlated kurtosis deconvolution (MCKD) is proposed. This algorithm can enhance the impulse characteristics of the signal under low signal-to-noise ratio conditions, thereby improving the kurtosis value of the original signal and deconvolving the periodic pulse component from the fault signal, effectively extracting weak fault feature information, and making up for the shortcomings of MED. However, the effect of the MCKD algorithm is mainly determined by three input parameters: filter length L, deconvolution period T, and shift number M. The rationality of parameter setting has a crucial impact on the deconvolution effect. Therefore, a method is needed to adaptively determine the optimal combination of L, T, and M in the MCKD algorithm.

[0072] In order to solve the technical problem that the MCKD input parameters are dependent on manual experience selection, resulting in poor extraction of weak fault feature information, please refer to Figure 1-Figure 8 This embodiment proposes a method for RV reducer gear weak fault diagnosis based on adaptive MCKD, thereby solving the problem that the MCKD input parameters depend on artificial experience selection, giving full play to the superiority of this method in weak fault feature extraction, and solving the problem that the non-stationary nonlinear characteristics of the RV reducer fault feature signal are easily affected by the transmission path and strong noise interference, making it difficult to extract the fault feature, and realizing efficient RV reducer fault diagnosis. The method includes the following steps:

[0073] S1. Obtain the original signal x of the RV reducer gear in the running state; the original signal x is usually obtained by using an acceleration sensor to collect the vibration signal of the RV reducer gear. The acceleration sensor can convert the vibration of the RV reducer gear into an electrical signal of the original signal x, which is convenient for subsequent data processing and analysis, so as to realize the weak fault diagnosis of the RV reducer gear.

[0074] S2. The SCSSA search algorithm is used to adaptively determine the optimal diagnostic parameters of the maximum correlation kurtosis deconvolution MCKD under the optimal fitness function value. The optimal fitness function value is to use the maximum value of the envelope spectrum peak factor Ec as the objective function. The envelope spectrum peak factor represents the impact characteristics of the original signal, and its value represents the strength of the periodic impact. The richer the fault information and the more regular the waveform pulse, the larger the value of the envelope spectrum peak factor.

[0075] The optimal diagnostic parameters include the filter length L, the deconvolution period T and the filter shift number M. Among them, the sparrow search algorithm combining sine-cosine and Cauchy mutation (SCSSA) introduces a refraction reverse learning strategy for initializing individuals, thereby reducing the risk of early convergence to the local optimal solution. In addition, in the sparrow foraging scene, the two types of sparrows, the finder and the follower, usually work closely together to complete the predation process. The sine-cosine strategy and the Cauchy mutation strategy are also introduced to replace the update formula of the finder and follower positions of the original algorithm to increase the diversity of individual positions within the population, increase the search range, solve the problem of insufficient ability of the traditional sparrow search algorithm (SSA) in taking into account both global and local optimization, and improve the global optimization accuracy and speed of SSA. It has the advantages of simple structure, few adjustment parameters, fast convergence speed, strong global search ability and high accuracy. Therefore, in step S2, the specific process includes the following steps:

[0076] S21. Define the initial parameters of the SCSSA search algorithm and the optimization interval of the MCKD influencing parameters. The initial parameters of the SCSSA search algorithm include: setting the sparrow population size N, the maximum number of iterations Z, the proportion of discoverers PD, the proportion of alerters SD, the safety value ST and the alert value R2; the optimization interval of the MCKD influencing parameters includes: setting the value ranges of the filter length L, the deconvolution period T and the filter shift number M to form the optimization interval. In this embodiment, the effect of the MCKD algorithm is mainly determined by the three input parameters of the filter length L, the deconvolution period T and the filter shift number M. The correct selection of the input parameters plays a vital role in improving the filtering effect of the MCKD. In the SCSSA search algorithm that integrates the sine, cosine and Cauchy variations proposed in the present invention, it is necessary to set the value ranges of the three input parameters to meet the setting of the optimization interval, as follows:

[0077] Filter length L:

[0078] The filter length L determines the number of samples calculated in each iteration, and its size directly affects the effect of pulse sequence extraction. In order to ensure that the extracted impulse sequence can cover the entire frequency band of the fault, the filter length L should satisfy Shannon's theorem:

[0079] L>2f s / f i

[0080] In the formula, f s is the sampling frequency; f iis the natural frequency of the system. Under the premise of satisfying the sampling theorem, the larger L is, the better the filtering effect is. However, increasing L will take more computing time. Therefore, the selection of filter length should comprehensively consider the influence of filtering effect and computing time.

[0081] Deconvolution period T:

[0082] Deconvolution period T = f s / f i , considering the influence of actual related parameters, when the fault characteristic frequency is unknown, the original signal can be analyzed by envelope spectrum to find the prominent frequency band, and f can be reasonably selected within this frequency band. i The interval value of is used to determine the optimal range of T;

[0083] The number of filter shifts M:

[0084] The larger the filter shift number M is, the more fault pulses there are in the deconvolution signal. However, the corresponding requirement for the accuracy of T is higher. When the value of M is greater than 7, the data will be distorted. Therefore, the value range of M is generally 1-7.

[0085] The sparrow search algorithm (SCSSA) that combines sine, cosine and Cauchy mutations is used to optimize the MCKD input parameters. The adaptive MCKD model selects the dimensionless index envelope spectrum crest factor (Ec) as the fitness function. The value of Ec represents the strength of the periodic shock. The richer the fault information and the more regular the waveform pulse, the greater the value of the envelope spectrum crest factor. The expression of the envelope spectrum crest factor Ec is as follows:

[0086]

[0087] Where X(N) is the envelope spectrum in [N*f r ,f s ]; f r is the frequency conversion; f s is the system sampling frequency; the value of N should be no less than 2 to eliminate the influence of frequency conversion on the envelope spectrum peak factor Ec value.

[0088] S22. Randomly generate Q parameter combinations [L, T, M] in the optimization interval as the initial population Xij, and generate the refraction reverse population X^ij on this basis, wherein the number of the refraction reverse population X^ij is also Q, and the formula for generating the refraction reverse population X^ij is:

[0089]

[0090] In the formula, x i,jis the position of the i-th parameter combination [L, T, M] in the population in the j-dimensional space, where i = 1, 2, ..., D; j = 1, 2, ..., N, D is the number of populations, and N is the dimension; l j 、u j are the minimum and maximum values ​​of the j-th dimension of the search space, respectively.

[0091] S23, merge the initial population Xij and the refraction reverse population X^ij, calculate the fitness value of the current position of each parameter combination [L, T, M] in the merged population and sort them, wherein, in the merged population, a total of Q initial populations Xij and Q refraction reverse populations X^ij are included, a total of 2Q, then calculate the corresponding fitness values ​​and sort them, and finally select Q population positions with high fitness values ​​as the final population. The calculation of the fitness value can be calculated using the calculation formula used in the existing sparrow search algorithm, which is used as the prior art here, so it is not repeated in detail.

[0092] S24, select the top Q with high fitness value in the merged population i The parameter combination [L, T, M] is used as the final initial population to obtain the current optimal fitness individual and its position X best , where i∈[2,10];

[0093] S25, set the discoverer ratio PD, the vigilant ratio SD, the safety value ST and the warning value R2, and continuously iterate and update the positions of the discoverers and followers in the merged population. In this embodiment, the following formula can be used for implementation, specifically:

[0094] Update the position of the discoverer in the merged population to:

[0095]

[0096] Where ω is the nonlinear weight factor; Z is the maximum number of iterations; r′ is the step-size nonlinear decreasing search factor; a=1; η≥1 is the adjustment coefficient; represents the position of the i-th parameter combination [L, T, M] in the j-dimension in the t-th iteration; the random number x∈[0,2π] represents the moving distance of the parameter combination [L, T, M]; the random number y∈[0,2π] represents the influence of the optimal position individual parameter combination [L, T, M] on the next position; the warning value R2∈(0,1], the safety value ST∈(0.5,1], when R2<ST, it means that the surrounding environment is safe and the discoverer can search for food in the surrounding area; when R2≥ST, it means that the surrounding environment is dangerous and the population needs to go to other safe areas to forage.

[0097] Update the position of the followers in the merged population to:

[0098]

[0099] Where cauchy(0,1) is the standard Cauchy distribution function; It means multiplication.

[0100] The expression for setting the vigilant is:

[0101]

[0102] In the formula, represents the best position found by the sentinel after t iterations; β is the step correction coefficient, which obeys the standard normal distribution; K∈(0,1) is a random number; X worst is the global worst position under the current number of iterations; ε is a minimum constant; f i is the fitness of the parameter combination [L, T, M] after t iterations; f w and f g Respectively represent the minimum fitness value and optimal fitness of individuals in the current population; when f i >f g When f , it means that the parameter combination [L, T, M] is at the outermost edge of the population, which is easy to become prey and needs to move to the center of the population; when f i =f g When , it means that the parameter combination [L, T, M] at the optimal position among the groups senses danger and should go to a safe place immediately.

[0103] The setting of the safety value ST and the warning value R2 can be set according to different experimental objects and actual conditions, and no limitation is made here. The above setting and updating position implementation can be directly implemented by existing technical means, and will not be listed in detail in this embodiment.

[0104] S26. Compare the values ​​of the envelope spectrum peak factor Ec in the iterative update process, and continuously update the optimal fitness individual and the corresponding optimal position in the merged population.

[0105] If the maximum number of iterations is met, the iteration is stopped and the optimal diagnostic parameters [L, T, M] at the optimal fitness value are output;

[0106] If not, go to step S25.

[0107] This embodiment uses the maximum value of the envelope spectrum peak factor as the fitness function, and then adaptively determines the optimal diagnostic parameters through the SCSSA search algorithm. It not only inherits the advantages of MCKD in fault feature extraction, but also has the function of adaptive parameter adjustment, avoiding the problem of poor feature extraction performance caused by manual selection of parameters.

[0108] S3, using the optimal diagnostic parameters to perform MCKD deconvolution on the original signal x, and obtain the enhanced periodic pulse component signal y representing the fault in the original signal x;

[0109] S4, decomposing the periodic pulse component signal y by ensemble empirical mode decomposition (EEMD) to obtain n modal components in different frequency bands;

[0110] S5. Calculate the multiscale fuzzy entropy of n-1 modal components and construct a multiscale fuzzy entropy feature set vector. Since the nth mode may have a poor decomposition effect when performing EEMD decomposition, the first n-1 modal components are selected to calculate the multiscale fuzzy entropy. In the process of calculating the fuzzy entropy, set the parameters m = 2, r = 0.15Std, n = 2, tau = 1, and obtain the multiscale fuzzy entropy feature set sampEn (i, j) of each mode to construct a feature set vector. Therefore, the multiscale fuzzy entropy of the first n-1 modal components is calculated as follows:

[0111] First, define fuzzy entropy as follows:

[0112] S51. For the original time series {u(i):1≤i≤N} of length N corresponding to n modes, an m-dimensional vector is constructed in sequence, that is:

[0113]

[0114] i=1,2,…,N-m+1}

[0115] In the formula, Represents the value of the continuous dimension m original time series u starting from the i-th point; u0(i) is the mean value of the continuous dimension m original time series u starting from the i-th point.

[0116] Remove the mean u0(i), where

[0117]

[0118] S52. Definition and The distance between is the maximum value of the difference between the corresponding elements of the two, that is:

[0119]

[0120] i,j=1,2,…,Nm,i≠j

[0121] In the formula, Represents the value of the original time series u with continuous dimension m starting from the jth point; u0(j) is the mean value of the original time series u with continuous dimension m starting from the jth point.

[0122] S53, through the fuzzy function Calculate vector and Similarity Right now:

[0123]

[0124] In the formula, the fuzzy function is an exponential function, n and r are the gradient and width of its boundary respectively.

[0125] S54. Based on similarity Define the function φ m (n,r), that is:

[0126]

[0127] S55, the function φ m The dimension of (n,r) is m+1, and the above steps S51-S54 are repeated to obtain:

[0128]

[0129] S56. Define fuzzy entropy and obtain the original sequence X of length N i ={x1,x2,…,x N}, the fuzzy entropy is defined as:

[0130]

[0131] When N is a finite number, the above formula is expressed as:

[0132] FuzzyEn(m,n,r,N)=lnφ m (n,r)-lnφ m+1 (n,r)

[0133] Then the original time series is coarse-grained. For the original sequence X of length N i ={x1,x2,…,x N}, given the embedding dimension m and similarity tolerance r in advance, a new coarse grained vector is established based on the original sequence, that is:

[0134]

[0135] In the formula, τ=1,2,… are scale factors. In fact, when τ=1, y j (1) is the original sequence. For non-zero τ, the original sequence X i is divided into τ coarse-grained sequences y of length N / τ each j (τ).

[0136] The fuzzy entropy of each coarse-grained sequence is calculated and plotted as a function of the scale factor. Here, when calculating the fuzzy entropy of each coarse-grained sequence, the similarity tolerance r remains unchanged, which is 0.1-0.25Std, where Std is the standard deviation of the original data.

[0137] Then the multi-scale fuzzy entropy of the modal component is:

[0138] sampEn(i,j)=FuzzyEn(z,m,r,n,tau)

[0139] Where z is the one-dimensional vector to be analyzed; m is the embedding dimension, m=2; r is the similarity tolerance, r=0.15Std; n is the fuzzy power, n=2; tau is the time delay, usually 1.

[0140] S6. Input the multi-scale fuzzy entropy feature set vector into the classifier for classification and identification to obtain weak fault diagnosis results. The weak fault diagnosis results include tooth surface defects, broken teeth, tooth root cracks, and tooth surface wear.

[0141] In order to verify the effectiveness of the RV reducer gear weak fault diagnosis method based on adaptive MCKD proposed in this application, the gearbox data set of Southeast University was first used for verification. The vibration data of the gears in different states under two working conditions, 1200rpm without load and 1800rpm with load of 7.32Nm, were simulated and collected on the experimental platform, including healthy, tooth surface defect, broken tooth, tooth root crack and tooth surface wear species, with a sampling frequency of 5120Hz.

[0142] The method proposed in this application is first used to analyze the root crack fault signal under the no-load condition with a speed of 1200rpm. The population number and the maximum number of iterations in the SCSSA search algorithm are set to 30. According to the characteristics of the fault signal, the value range of the optimization parameters is set to L∈[100,650], T∈[50,350], M∈[1,7]. Then the SCSSA search algorithm is used to optimize these three important parameters of MCKD. The optimization iteration curve is as follows: Figure 2 As shown in the results, it can be seen that during the 30 optimization iterations, the fitness value of the objective function continued to increase. At the 15th generation, the speed of the fitness function curve slowed down and became stable, and the individual fitness basically stopped changing. It can be considered that the optimal result has been found and the optimization has ended. The optimal diagnostic parameter combination for the maximum correlation kurtosis deconvolution MCKD is [100, 84, 1].

[0143] Set the optimal diagnostic parameters for MCKD deconvolution. Figure 3 and Figure 4is the initial signal of the tooth root crack fault and the time domain diagram after deconvolution. Figure 3 From the time domain diagram of the initial signal, it can be seen that the original vibration signal of the gear is seriously affected by noise, and the impact component of the fault signal is almost submerged by the noise. Figure 4 From the time domain waveform after SCSSA-MCKD enhancement, it can be seen that the impulse component of the signal has been significantly increased.

[0144] Similarly, for the other four different types of vibration signals, the optimal diagnostic parameters are obtained after optimization. After the vibration signal is enhanced using the optimal diagnostic parameter combination, 100 groups of data samples are extracted from each of the five different types of vibration signals, for a total of 500 groups of data. The length of each data sample is 1000 points. Each group of samples is decomposed by EEMD, and then the multi-scale fuzzy entropy of the modal component is calculated after decomposition. In the process of calculating the multi-scale fuzzy entropy, the parameters m = 2, r = 0.15Std, n = 2, tau = 1 are set to obtain the multi-scale fuzzy entropy feature set of each mode, and the multi-scale fuzzy entropy feature set vector is constructed.

[0145] Use 60% of the data samples as training samples and 40% as test samples to construct different sample labels. Then input the constructed multi-scale fuzzy entropy feature set vector into the support vector machine for classification and recognition, such as Figure 5 The results shown show that the classification recognition rate under different operating conditions is 97.50%.

[0146] Similarly, the method proposed in this paper is used for gear fault analysis under the working condition of 1800rpm speed and 7.32Nm load. After deconvolution of the vibration signal using the optimal diagnostic parameter combination, the multi-scale fuzzy entropy feature set vector is constructed according to the same method as above for classification and identification, as shown in the following figure. Figure 6 The results shown show that the classification recognition rate under different operating conditions is 98.50%.

[0147] In order to further analyze the multi-scale fuzzy entropy feature set vector constructed by the method in this paper, the T-SNE method is introduced to visualize the multi-scale fuzzy entropy feature set vector and observe the distribution of the five signals in the low-dimensional space. Figure 7 and Figure 8 The distribution of fault feature sets under the above two working conditions is shown respectively.

[0148] The present invention uses the RV reducer gear weak fault diagnosis method combining MCKD with SCSSA for RV reducer gear fault diagnosis under different working conditions, making full use of the advantages of MCKD in fault feature extraction. Case studies have proved the effectiveness of the proposed method, and the classification and recognition accuracy of the method reaches more than 97.5%.

[0149] Those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, so the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0150] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the above embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0151] The above implementation methods have been described in detail. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for RV reducer gear weak fault diagnosis based on adaptive MCKD, characterized in that: The method comprises the following steps: S1, obtain the original signal x of the RV reducer gear in the running state; S2, using the SCSSA search algorithm to adaptively determine the optimal diagnostic parameters of the maximum correlation kurtosis deconvolution MCKD under the optimal fitness function value; S3, using the optimal diagnostic parameters to perform MCKD deconvolution on the original signal x, and obtain the enhanced periodic pulse component signal y representing the fault in the original signal x; S4, decomposing the periodic pulse component signal y by ensemble empirical mode decomposition (EEMD) to obtain n modal components in different frequency bands; S5, calculating the multi-scale fuzzy entropy of n-1 modal components and constructing a multi-scale fuzzy entropy feature set vector; S6. Input the multi-scale fuzzy entropy feature set vector into the classifier for classification and identification to obtain weak fault diagnosis results.

2. The RV reducer gear weak fault diagnosis method according to claim 1 is characterized in that: In step S2, the optimal fitness function value is to use the maximum value of the envelope spectrum peak factor Ec as the objective function, and the optimal diagnostic parameters include the filter length L, the deconvolution period T and the filter shift number M.

3. The RV reducer gear weak fault diagnosis method according to claim 1, characterized in that: In step S2, the specific process includes the following steps: S21, define the initial parameters of the SCSSA search algorithm and the optimization interval of the MCKD influencing parameters; S22, randomly generate Q parameter combinations [L, T, M] in the optimization interval as the initial population Xij, and generate the refraction reverse population X^ij on this basis; S23, merge the initial population Xij and the refracted reverse population X^ij, calculate the fitness value of the current position of each parameter combination [L, T, M] in the merged population and sort them; S24, select the top Q with high fitness value in the merged population i The parameter combination [L, T, M] is used as the final initial population to obtain the current optimal fitness individual and its position X best ; S25, setting the discoverer ratio PD, the vigilant ratio SD, the safety value ST and the warning value R2, and continuously iteratively updating the positions of the discoverers and followers in the merged population; S26. Compare the values ​​of the envelope spectrum peak factor Ec in the iterative update process, and continuously update the optimal fitness individual and the corresponding optimal position in the merged population. If the maximum number of iterations is met, the iteration is stopped and the optimal diagnostic parameters [L, T, M] at the optimal fitness value are output; If not, go to step S25.

4. The RV reducer gear weak fault diagnosis method according to claim 3, characterized in that: In step S21, defining the initial parameters of the SCSSA search algorithm includes: setting the sparrow population size N, the maximum number of iterations Z, the discoverer ratio PD, the vigilant ratio SD, the safety value ST and the warning value R2; Defining the optimization interval of the MCKD influencing parameters includes: setting the value range of the filter length L, the deconvolution period T and the filter shift number M to form the optimization interval.

5. The RV reducer gear weak fault diagnosis method according to claim 3, characterized in that: The expression of envelope spectrum peak factor Ec is: Where X(N) is the envelope spectrum in [N*f r ,f s ]; f r is the frequency conversion; f s is the system sampling frequency; the value of N should be no less than 2.

6. The RV reducer gear weak fault diagnosis method according to claim 1, characterized in that: The weak fault diagnosis results include tooth surface defects, broken teeth, tooth root cracks, and tooth surface wear.

7. The RV reducer gear weak fault diagnosis method according to claim 1, characterized in that: In step S5, it specifically includes: Define the fuzzy entropy of n modes and get the original sequence X of length N i ={x1,x2,…,x N }; For the original sequence X of length N i ={x1,x2,…,x N Given the embedding dimension m and similarity tolerance r in advance, a new coarse-grained vector y is established based on the original sequence j (τ), that is: In the formula, τ=1,2,… are scale factors. In fact, when τ=1, y j (1) is the original sequence. For non-zero τ, the original sequence X i is divided into τ coarse-grained sequences y of length N / τ each j (τ); Calculate the fuzzy entropy of each coarse-grained sequence and plot it as a function of the scale factor, namely: sampEn(i,j)=FuzzyEn(z,m,r,n,tau) Where z is the one-dimensional vector to be analyzed; m is the embedding dimension, m=2; r is the similarity tolerance, r=0.15Std; n is the fuzzy power, n=2; tau is the time delay, which is 1.

8. The RV reducer gear weak fault diagnosis method according to claim 7, characterized in that: Define the fuzzy entropy of n modes and get the original sequence X of length N i ={x1,x2,…,x N }, specifically including the following steps: S51. For the original time series {u(i):1≤i≤N} of length N corresponding to n modes, an m-dimensional vector is constructed in sequence, that is: In the formula, represents the value of the continuous dimension m original time series u starting from the i-th point; u0(i) is the mean value of the continuous dimension m original time series u starting from the i-th point; Remove the mean u0(i), where S52. Definition and The distance between is the maximum value of the difference between the corresponding elements of the two, that is: In the formula, represents the value of the original time series u with continuous dimension m starting from the jth point; u0(j) is the mean value of the original time series u with continuous dimension m starting from the jth point; S53, through the fuzzy function Calculate vector and Similarity Right now: In the formula, the fuzzy function is an exponential function, n and r are the gradient and width of its boundary respectively; S54. Based on similarity Define the function φ m (n,r), that is: S55, the function φ m The dimension of (n,r) is m+1, and the above steps S51-S54 are repeated to obtain: S56. Define fuzzy entropy and obtain the original sequence X of length N i ={x1,x2,…,x N }, the fuzzy entropy is defined as: When N is a finite number, the above formula is expressed as: FuzzyEn(m,n,r,N)=lnφ m (n,r)-lnφ m+1 (n,r)。

Images