Gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted scatter entropy

Abstract

The invention discloses a gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted spread entropy, comprising a signal acquisition system and a fault diagnosis system; the signal acquisition system acquires gearbox vibration data, is developed based on an STM32 development board, and is composed of a Flash storage chip, a CAN transmission module, an acceleration sensor, and an AD conversion chip; the fault diagnosis system diagnoses the gearbox fault type, and decomposes the collected gearbox vibration data into useful signal components using a swarm intelligent decomposition (SWD) algorithm whose parameters are determined by an optimized bat algorithm; after being processed by the method of the invention, useful features in the signal can be more accurately extracted, useful impact components can be retained, and the reconstructed signal has low distortion; the invention can quickly and accurately detect whether a gearbox fails and the fault type, and then timely handle the fault, thereby avoiding greater economic losses or even safety accidents caused by the gearbox failure.

Description

Technical Field

[0001] The invention relates to the field of agricultural automation machinery, and in particular to a gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted dispersion entropy. Background Art

[0002] With the development of artificial intelligence and the Internet of Things, intelligent diagnosis of mechanical failures has become increasingly popular. As an important production machinery, the intelligence of agricultural machinery has received more and more attention. To realize the intelligence of agricultural machinery, on the one hand, it is necessary to rely on a large number of high-precision sensors to accurately sense the state of the machinery, thereby laying the foundation for subsequent processing strategies; on the other hand, it is necessary to process the acquired data to find out the internal correlation and realize differentiated processing of the machinery. The gearbox of the walking system of the corn cob and stalk harvester often runs in a dusty and vibrating environment. The healthy state of the gearbox is crucial to the reliability of the whole vehicle. The faults of the gear transmission system in the gearbox are mostly progressive faults. When the defects reach a certain level, the components lose their functions and the vehicle is shut down for a long time, causing economic losses or safety accidents, which have a great impact on agricultural production.

[0003] Identifying the type of gear transmission system fault and dealing with it in a targeted manner is an effective means to avoid major faults. At present, the more commonly used method is to obtain the vibration data of the transmission system, extract the impact component caused by the faulty component, and then identify the fault; however, in actual work, the acquired fault vibration signal will be affected by the attenuation of the transmission path and the mutual coupling between the paths, and will be accompanied by strong background noise. At the same time, the signal itself also has obvious nonlinear and non-stationary characteristics; under this premise, it is difficult to intuitively identify the fault from the signal;

[0004] Therefore, it is urgent to design a gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted spread entropy to solve the above technical problems. Summary of the invention

[0005] The present invention mainly aims at the gear transmission system of the gear box of the walking system of the corn cob and stalk harvester, and consists of two parts: a signal acquisition system and a fault diagnosis system; the signal acquisition system is mainly used to collect the vibration signal of the gear box, is developed based on an STM32 development board, and is composed of a Flash storage chip, a CAN sending module, an acceleration sensor, and an AD conversion chip; the fault diagnosis system diagnoses whether the gear box is faulty and the fault type, decomposes the collected gear box vibration data into useful signal components by using a swarm intelligence decomposition algorithm whose parameters are determined by an optimized bat algorithm, extracts signal features by using an improved multi-scale weighted spread entropy algorithm, and determines the fault type by using an SVM classifier whose parameters are determined by an optimized bat algorithm, so as to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A gearbox fault diagnosis system based on group decomposition and improved multi-scale weighted dispersive entropy, wherein the signal acquisition system has a sampling frequency of 5000 Hz, an ADC bit number of 16 bits, a sampling time of 2 seconds, a sampling interval of 15 minutes, and a communication mode of a low-speed CAN bus;

[0008] The signal acquisition system is developed based on the STM32 development board, and the development board uses STM32F103C8T6; the vibration sensor YK-YD20 collects the gearbox vibration signal, which is converted into a digital signal by the AD conversion module with the PCF8591 chip as the core, and the converted data is stored in the W25Q16 Flash storage chip through the single-chip microcomputer; at the sampling interval, the TAJ1050CAN sending module is used to send the data to the vehicle terminal, and the vehicle terminal sends it to the cloud storage;

[0009] A gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted scatter entropy, wherein the method for optimizing the bat algorithm is: introducing an adaptive speed weight factor ω into the formula for local search of the standard bat algorithm individuals i , improve the global optimization speed, the calculation formula is:

[0010]

[0011] In formula (1), rand(·) is a random number generated from (0,1); x i is the current position; x mean is the mean of the global position;

[0012] Calculate ω i After that, the speed formula of updating the bat algorithm is:

[0013]

[0014] In formula (2), is the bat speed at time t+1; is the bat speed at time t; is the bat position at time t; x * is the global optimal position; f i is the current pulse transmission frequency.

[0015] As a further solution of the present invention: the method of group intelligent decomposition of signals is: inputting a vibration signal, determining the key parameter center frequency power density spectrum threshold and iterative error, and then performing signal decomposition.

[0016] As a further solution of the present invention: the fitness function of the SWD bat algorithm is: the inverse of the mean value of the envelope spectrum kurtosis of the decomposed OC component is used as the fitness function to evaluate the decomposition effect. As a further solution of the present invention: the improved multi-scale weighted scatter entropy feature extraction method is: based on the multi-scale scatter entropy, for the input time series X, the maximum value is selected during coarse-graining As a coarse-grained value, in order to extract the weak impact component in the time series to better describe the characteristics of the signal, the calculation formula is:

[0017]

[0018] In formula (3): represents the coarse-grained sequence when the maximum scale factor is τ and the current scale factor is s, and X is the original input discrete time series; is the original sequence from X (i-1)s+τ To X is+τ-1 The maximum value of the interval; N is the signal length.

[0019] Probability of compressed dimensions after coarse-graining When calculating, add the evaluation weight parameter ω that takes into account the concentration of the reconstruction vector i , further highlighting the shock component in the time series, the calculation formula is:

[0020]

[0021] In formula (4): is the arrangement pattern corresponding to the i-th phase space reconstruction vector; m is the embedding dimension; d is the delay; is the arrangement pattern corresponding to the j-th phase space reconstruction vector; the count(·) function represents the statistics from arrive All potential permutations In the current arrangement style The same number; ω i is the weight of the i-th phase space reconstruction vector; Z i is the i-th phase space reconstruction vector; is the mean of the reconstruction vector of the i-th phase space; is the jth value of the i-th phase space reconstruction vector;

[0022] Finally, the improved multi-scale weighted scatter entropy E(X,m,d,c) is calculated using the following formula:

[0023]

[0024] In formula (5), X is the original input discrete time series; m is the embedding dimension; d is the delay; c is the number of classifications during sequence reconstruction.

[0025] As a further solution of the present invention: the method of training the SVM classifier to identify different types of faults is: using the libsvm model; inputting a feature vector, determining the support vector machine type, the kernel function type, the error accuracy for stopping training, the weight occupied by each class, the penalty coefficient, the kernel function coefficient, and then performing fault identification.

[0026] As a further solution of the present invention: the fitness function of the SVM bat algorithm is: using the accuracy of 5-fold cross-validation as the fitness function to evaluate the classification accuracy.

[0027] As a further solution of the present invention: the SVM classifier can identify the following types: normal gears, normal bearings, normal bearing rolling elements, gear eccentricity, gear wear, gear cracks, half-broken gear teeth, broken gear teeth, pitting on the outer ring of the bearing, pitting on the inner ring of the bearing, and pitting on the rolling elements of the bearing.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The swarm intelligence decomposition (SWD) algorithm has relatively good adaptability and decomposition effect in processing strong noise, nonlinear and non-stationary signals. Through the SWD algorithm, useful signal components can be quickly and accurately extracted from the original signal containing strong noise, nonlinearity and non-stationary to lay the foundation for subsequent fault identification;

[0030] As a classifier for identifying fault types, the SVM classifier has good recognition and generalization capabilities for nonlinear problems;

[0031] The SWD algorithm has two important parameters: the center frequency power density spectrum threshold and the iteration error, which determine the number of signal decompositions and the accuracy of each component respectively; SVM also has two important parameters: the penalty coefficient and the coefficient of the kernel function. Different parameter combinations affect the accuracy of the SVM classifier. In order to quickly find the optimal combination of SWD and SVM parameters and improve the overall operation efficiency of the algorithm, the optimized bat algorithm is used for parameter search. Compared with the original bat algorithm, the optimized bat algorithm is faster and more accurate in global optimization.

[0032] As a nonlinear dynamic indicator for evaluating the regularity and complexity of time series, entropy theory is very suitable for characterizing non-stationary gearbox vibration acceleration signals; the spread entropy has fast calculation speed, high stability, and greatly improved noise resistance, and a slight change in the vibration signal will not change its mapped category label, but only considers the absoluteness of the amplitude but not the relativity, and is unable to evaluate the volatility of the signal; the improved multi-scale weighted spread entropy proposed in the present invention has fast calculation speed, is less affected by mutation signals, and considers the size relationship between amplitudes, which has an obvious effect on signal feature extraction and has strong robustness.

[0033] The present invention comprehensively utilizes the advantages of group decomposition and improved multi-scale weighted scatter entropy to perform fault diagnosis on the gear box of the walking system of the corn cob and stalk harvester, and can relatively quickly and accurately extract useful impact features in the signal from a strong noise background for fault identification. The method has high efficiency, is easy to implement, and has good robustness.

[0034] At the same time, the gearbox vibration signal with strong background noise can be processed by this method to more accurately extract useful features in the signal, retain useful impact components, and reconstruct the signal with low distortion; the overall operation efficiency of the algorithm is improved by optimizing the bat algorithm; the present invention can quickly and accurately detect whether the gearbox has a fault and the type of fault, and then deal with it in time to avoid greater economic losses or even safety accidents caused by gearbox failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is the overall flow chart of the gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted dispersion entropy.

[0036] Figure 2 Vibration signal diagram collected for the gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted scattered entropy.

[0037] Figure 3 OC1 component diagram of vibration signal in gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted scatter entropy.

[0038] Figure 4 OC2 component diagram of vibration signal in gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted scattered entropy. DETAILED DESCRIPTION

[0039] 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.

[0040] See also Figures 1 to 4 ,In an embodiment of the present invention, a gearbox fault diagnosis system based on group decomposition and improved multi-scale weighted spread entropy includes a signal acquisition system and a fault diagnosis system;

[0041] The signal acquisition system collects gearbox vibration data and is developed based on the STM32 development board. It consists of a Flash storage chip, a CAN transmission module, an acceleration sensor, and an AD conversion chip.

[0042] The fault diagnosis system diagnoses the type of gearbox fault and decomposes the collected gearbox vibration data into useful signal components using a swarm intelligence decomposition (SWD) algorithm whose parameters are determined by an optimized bat algorithm;

[0043] The improved multi-scale weighted scattered entropy algorithm is used to extract the features of the signal; the extracted features are used to determine the fault type by optimizing the SVM classifier with parameters determined by the bat algorithm;

[0044] The signal acquisition system has a sampling frequency of 5000HZ, an ADC bit number of 16 bits, a sampling time of 2s, a sampling interval of 15 minutes, and a communication mode of low-speed CAN bus; it is developed based on the STM32 development board, and the development board uses STM32F103C8T6; the vibration sensor YK-YD20 collects the gearbox vibration signal, which is converted into a digital signal by the AD conversion module with the PCF8591 chip as the core, and the converted data is stored in the W25Q16 Flash storage chip through the single-chip microcomputer; at the sampling interval, the TAJ1050CAN sending module is used to send the data to the vehicle terminal, and the vehicle terminal sends it to the cloud storage;

[0045] The method of optimizing the bat algorithm is: in the formula for local search of the standard bat algorithm individual, introduce the self

[0046] Adaptation speed weight factor ω i , improve the global optimization speed, the calculation formula is;

[0047]

[0048] In formula (1), rand(·) is a random number generated from (0,1); x i is the current position; x meanis the mean of the global position;

[0049] The fitness function of the SWD bat algorithm is: the inverse of the mean value of the envelope spectrum kurtosis of the decomposed OC component is used as the fitness function to evaluate the decomposition effect;

[0050] The improved multi-scale weighted spread entropy feature extraction method is:

[0051] Step 1) Based on the multi-scale scatter entropy, for the input time series X, select the maximum value when coarsening As a coarse-grained value, in order to extract the weak impact component in the time series to better describe the characteristics of the signal, the calculation formula is:

[0052] In formula (3): represents the coarse-grained sequence when the maximum scale factor is τ and the current scale factor is s, and X is the original input discrete time series; is the original sequence from X (i-1)s+τ To X is+τ-1 The maximum value of the interval; N is the signal length.

[0053] Step 2) Probability of compressed dimension after coarse-graining When calculating, add the evaluation weight parameter ω that takes into account the concentration of the reconstruction vector i , further highlighting the shock component in the time series, the calculation formula is:

[0054]

[0055] In formula (4): is the arrangement pattern corresponding to the i-th phase space reconstruction vector; m is the embedding dimension; d is the delay; is the arrangement pattern corresponding to the j-th phase space reconstruction vector; the count(·) function represents the statistics from arrive All potential permutations In the current arrangement style The same number; ω i is the weight of the i-th phase space reconstruction vector; Z i is the i-th phase space reconstruction vector; is the mean of the reconstruction vector of the i-th phase space; is the jth value of the i-th phase space reconstruction vector;

[0056] Step 3) Finally, the improved multi-scale weighted spread entropy E(X,m,d,c) is calculated using the following formula:

[0057]

[0058] In formula (5), X is the original input discrete time series; m is the embedding dimension; d is the delay; c is the number of classifications when the sequence is reconstructed;

[0059] The parameter settings of SWD determined by the optimized bat algorithm are: population size is 60, variable dimension is 2, maximum number of iterations is 150, parameter center frequency power density spectrum threshold search range is (0.01, 1), and parameter iteration error search range is (0.01, 2).

[0060] After optimizing the bat algorithm, the optimal parameters of the center frequency power density spectrum threshold and iteration error of SWD are (0.19, 0.16). The two OC components of the vibration signal decomposed using the optimal parameters are as follows: Figure 3 shown.

[0061] The improved multi-scale weighted spread entropy calculated for the two OC components is shown in the following table:

[0062]

[0063] The proposed eigenvalues ​​are combined to form a eigenvector, and the length of the eigenvector corresponding to each signal sample is 40.

[0064] The parameter settings of SVM determined by the optimized bat algorithm are: population size is 50, variable dimension is 2, maximum number of iterations is 150, parameter c search range is (0.01, 100), parameter g search range is (1, 100); other SVM parameter settings are C-support vector classifier as type; RBF radial basis kernel function is used as kernel function; the error accuracy for stopping training is 0.001; the weight occupied by each class is 1.

[0065] After optimizing the bat algorithm, the optimal parameters of SVM are (9.13, 17.09);

[0066] The feature vector is input into the SVM classifier for classification and the vibration fault type is pitting of the bearing inner ring.

[0067] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted spread entropy, characterized by: The signal acquisition system collects gearbox vibration data and is developed based on the STM32 development board. The signal acquisition system consists of a Flash storage chip, a CAN transmission module, an acceleration sensor, and an AD conversion chip. The fault diagnosis system diagnoses the gearbox fault type, and uses the SWD algorithm whose parameters are determined by the optimized bat algorithm to decompose the collected gearbox vibration data into useful signal components; the fitness function of the SWD algorithm is: the inverse of the mean value of the envelope spectrum kurtosis of the decomposed OC component is used as the fitness function to evaluate the decomposition effect; The improved multi-scale weighted scattered entropy algorithm is used to extract the signal features; Extract features and determine the fault type by optimizing the SVM classifier with parameters determined by the bat algorithm; the method of identifying different types of faults by the SVM classifier is: using the libsvm model; inputting a feature vector, determining the support vector machine type, the kernel function type, the error accuracy for stopping training, the weight occupied by each class, the penalty coefficient, and the kernel function coefficient, and then performing fault identification; the fitness function of the SVM algorithm is: using the accuracy of 5-fold cross validation as the fitness function to evaluate the classification accuracy; The improved multi-scale weighted spread entropy feature extraction method is: Step 1) Based on the multi-scale scatter entropy, for the input time series X, select the maximum value when coarsening As a coarse-grained value, in order to extract the weak impact component in the time series to better describe the characteristics of the signal, the calculation formula is: In formula (3): represents the coarse-grained sequence when the maximum scale factor is τ and the current scale factor is s, and X is the original input discrete time series; is the original sequence from X (i-1)s+τ To X is+τ-1 The maximum value of the interval; N is the signal length; Step 2) Probability of compressed dimension after coarse-graining When calculating, add the evaluation weight parameter ω that takes into account the concentration of the reconstruction vector i , further highlighting the shock component in the time series, the calculation formula is: In formula (4): is the arrangement pattern corresponding to the i-th phase space reconstruction vector; m is the embedding dimension; d is the delay; is the arrangement pattern corresponding to the j-th phase space reconstruction vector; the count(·) function represents the statistics from arrive All potential permutations In the current arrangement style The same number; ω i is the weight of the i-th phase space reconstruction vector; Z i is the i-th phase space reconstruction vector; is the mean of the reconstruction vector of the i-th phase space; is the jth value of the i-th phase space reconstruction vector; Step 3) Finally, the improved multi-scale weighted spread entropy E(X,m,d,c) is calculated using the following formula: In formula (5), X is the original input discrete time series; m is the embedding dimension; d is the delay; c is the number of classifications during sequence reconstruction.

2. The gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted spread entropy according to claim 1 is characterized in that: The signal acquisition system has a sampling frequency of 5000 Hz and an ADC bit number of 16 bits.

3. The gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted spread entropy according to claim 1 is characterized in that: The signal acquisition system sampling method is: Step 1: The sampling time is 2s, the sampling interval is 15 minutes, and the communication mode is low-speed CAN bus; the development is based on the STM32 development board, and the development board uses STM32F103C8T6; the vibration sensor YK-YD20 collects the gearbox vibration signal, which is converted into a digital signal by the AD conversion module with PCF8591 chip as the core; Step 2: The converted data is stored in the W25Q16 Flash storage chip through the microcontroller; Step 3: Use the TAJ1050CAN sending module to send data to the vehicle terminal during the sampling interval, and the vehicle terminal sends the data to the cloud storage.

4. The gearbox fault diagnosis method based on group decomposition and improved multi-scale weighted spread entropy according to claim 1 is characterized in that: The method for optimizing the bat algorithm is: introducing an adaptive speed weight factor ω into the formula for local search of the standard bat algorithm individual i , improve the global optimization speed, the calculation formula is: In formula (1), rand(·) is a random number generated from (0,1); x i is the current position; x mean is the mean of the global positions.

Images