Variable-speed bearing fault diagnosis method based on improved crown porcupine optimized random forest

Abstract

The invention relates to the technical field of bearing fault diagnosis, in particular to a variable-speed bearing fault diagnosis method based on an improved crown porcupine optimized random forest, and the method comprises the steps: collecting a bearing vibration signal; converting the vibration signal from a time domain signal to an angle domain signal by using COT; searching a modal component and a penalty factor of the VMD by using OCSSA, obtaining an optimal IMF component by using a criterion of minimum envelope entropy, and extracting a fault feature vector; cPO is improved by using Logistic chaotic mapping and a cyclic population reduction technology; the number of decision trees and the depth of the trees of the RF are searched using the improved CPO. The problems that fault features of the variable-speed rolling bearing are not easy to extract, faults are difficult to recognize, and fault diagnosis and classification accuracy of the variable-speed rolling bearing is not high are solved.

Description

Technical Field

[0001] The invention relates to the technical field of bearing fault diagnosis, and in particular to a variable speed bearing fault diagnosis method based on an improved crested porcupine optimized random forest. Background Art

[0002] Rolling bearings are key components in rotating machinery and play an indispensable role in the operation of equipment.

[0003] Existing vibration analysis methods for bearing fault diagnosis are aimed at constant speed conditions, such as fast Fourier transform, power spectrum analysis, envelope spectrum analysis, wavelet analysis, empirical mode decomposition, etc.; under variable speed conditions, traditional methods inevitably lead to spectral tailing of signal analysis.

[0004] Liu Qianjin et al. proposed a method of decomposing and fusing COT and VMD and combining it with deep learning, which solved the problem of difficulty in extracting fault features under variable speed conditions. However, their method has a low fault identification accuracy when the data set is limited.

[0005] Traditional CPO has strong robustness and search capabilities, but there are still some problems: the initial position of the porcupines is not adjusted, and the behavior choices of the porcupines converge with the increase in the number of iterations, which may cause the algorithm to fall into a local optimal solution.

[0006] In addition, the random forest algorithm requires many hyperparameters to be adjusted. Manual adjustment is a very tedious process and cannot guarantee the classification effect of the model. Summary of the invention

[0007] In view of the shortcomings of the existing methods, the present invention aims at the difficulty in extracting fault features of variable speed rolling bearings and the difficulty in fault identification. The computational order tracking algorithm (COT) is used to perform equal-angle resampling on the original vibration signal to convert the time domain non-stationary signal into an angular domain stationary signal; VMD is used to perform variational mode decomposition on the angular domain signal to find the optimal IMF component and extract its fault feature as a feature vector; the improved crown porcupine algorithm is used to optimize the number of decision trees and the depth of the random forest (RF), and the CPO-RF is trained using a training set to obtain a bearing fault classification model, thereby effectively solving the problem of difficulty in classifying variable speed rolling bearing faults and effectively improving the accuracy of variable speed rolling fault diagnosis and classification.

[0008] The technical solution adopted by the present invention is: a variable speed bearing fault diagnosis method based on improved crown porcupine optimized random forest includes the following steps:

[0009] Step 1: Collect bearing vibration signals;

[0010] As a preferred embodiment of the present invention, the states of the bearing include: normal, inner ring fault and outer ring fault.

[0011] Step 2: Use COT to convert the vibration signal from a time domain signal to an angle domain signal;

[0012] As a preferred embodiment of the present invention, step 2 specifically includes:

[0013] The vibration signal is converted from time sampling to angle sampling using phase information to determine the resampling time point;

[0014] According to the resampling time point, the vibration signal is interpolated, the amplitude of the vibration signal in the angular domain corresponding to the sampling time point is calculated, and the resampling signal in the angular domain is obtained.

[0015] Step 3: Use OCSSA to search for the modal components and penalty factors of VMD, and use the criterion of minimum envelope entropy to obtain the optimal IMF components and extract the fault feature vector;

[0016] As a preferred embodiment of the present invention, step three specifically includes:

[0017] Step 31, initializing the parameters of the sparrow algorithm and the optimization range of VMD parameters;

[0018] Step 32, initialize the sparrow population vector position, and VMD decompose the angle domain signal;

[0019] Step 33, using the minimum envelope entropy as the fitness value to update the optimal fitness value and the optimal position of the sparrow population;

[0020] Step 34, determine the tuning parameters in VMD;

[0021] Step 35, continuously updating the fitness values ​​and optimal positions of the sparrows, and re-ranking them;

[0022] Step 36: Determine whether the termination condition is met;

[0023] Step 37: Select the modal component and penalty factor with the minimum envelope entropy value.

[0024] As a preferred implementation manner of the present invention, the fault feature vector includes: mean, variance, peak, kurtosis, effective value, peak factor, pulse factor, waveform factor and margin factor.

[0025] Step 4: Improve CPO using Logistic chaos mapping and cyclic population reduction techniques;

[0026] As a preferred embodiment of the present invention, step 4 specifically includes:

[0027] Use Logistic chaos mapping to increase the diversity of the initialization population;

[0028] Cyclic population reduction was used to select non-threatened porcupines, and global search was performed by dynamic shrinkage.

[0029] Step 5: Use the improved CPO to search the number of decision trees and the depth of the tree in the random forest algorithm training model to find the optimal value;

[0030] As a preferred implementation of the present invention, step five specifically includes:

[0031] Step 51, initializing the crested porcupine algorithm parameters and the RF parameter optimization range;

[0032] Step 52: taking the out-of-bag error as the fitness value;

[0033] Step 53, sorting according to the fitness value of each porcupine to obtain the current optimal crown porcupine position;

[0034] Step 54: Determine RF tuning parameters;

[0035] Step 55, using a cyclic population reduction strategy to update the population size, continuously updating the optimal fitness value and the best position of the porcupines, and re-ranking;

[0036] Step 56: Determine whether the termination condition is met;

[0037] Step 57: Output the optimal RF parameters.

[0038] As a preferred embodiment of the present invention, a variable speed bearing fault diagnosis system based on an improved crown porcupine optimized random forest includes: a memory for storing instructions executable by a processor; a processor for executing instructions to implement a variable speed bearing fault diagnosis method based on an improved crown porcupine optimized random forest.

[0039] As a preferred embodiment of the present invention, a computer readable medium stores a computer program code, and when the computer program code is executed by a processor, a variable speed bearing fault diagnosis method based on an improved crested porcupine optimized random forest is implemented.

[0040] Beneficial effects of the present invention:

[0041] 1. Combine COT with VMD decomposition to extract the best IMF component fault characteristics of each fault type as the feature vector, and effectively extract the variable speed bearing fault characteristics;

[0042] 2. Use Logistic chaos mapping and cyclic population reduction technology to improve CPO and make it converge faster;

[0043] 3. Use improved CPO to optimize the number of decision trees and the depth of the tree of RF, effectively improving the classification accuracy of variable speed rolling fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of the variable speed bearing fault diagnosis method based on the improved crested porcupine optimized random forest of the present invention;

[0045] Figure 2 Schematic diagram of the improved CPO optimization RF step of the present invention;

[0046] Figure 3 (ab) is a schematic diagram of the result of using COT to convert the vibration signal from a time domain signal to an angle domain signal;

[0047] Figure 4 It is a schematic diagram of iterative comparison between the improved CPO of the present invention and the traditional CPO optimized random forest under multi-operating fault conditions;

[0048] Figure 5 (ab) is a schematic diagram comparing the fault diagnosis results of CPO-RF and RF single working conditions of the present invention;

[0049] Figure 6 (ab) are schematic diagrams comparing the CPO-RF and RF multi-operating condition fault diagnosis results of the present invention. DETAILED DESCRIPTION

[0050] The present invention is further described below in conjunction with the accompanying drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, and therefore it only shows the components related to the present invention.

[0051] like Figure 1 As shown, the variable speed bearing fault diagnosis method based on improved crown porcupine optimized random forest includes the following steps:

[0052] Step 1: Collect bearing vibration signals;

[0053] Take the variable speed bearing data set of the University of Ottawa in Canada as an example; the sampling frequency (fs) of this data set is 200kHz, the sampling time is 10 seconds, the bearing states include: normal, inner ring fault and outer ring fault, and the speed changes are speed increase, speed decrease, speed increase first and then decrease, and speed decrease first and then increase.

[0054] In order to test the diagnostic performance of the method of the present invention under different fault states, the vibration data with increasing speed is selected to generate a single-condition data set; the vibration data of 5 to 8 seconds with 4 speed changes are selected to form a multi-condition data set; each data is subjected to sliding window processing with a window of 1fs and a step size of 0.05fs. A total of 3*180 sample data are obtained for the single condition, and 3*240 sample data are obtained for the multi-condition. 70% of them are selected as training sets to train the model, and 30% are selected as test sets to verify the model.

[0055] Step 2: Use COT (Computed Order Tracking) to convert the vibration signal from a time domain signal to an angle domain signal;

[0056] like Figure 3 , use COT to convert the vibration signal from Figure 3 The time domain signal of (a) is converted into Figure 3 (b) Schematic diagram of the angle domain signal results,

[0057] COT specifically includes:

[0058] Firstly, the vibration signal is converted from time sampling to angle sampling using phase information to determine the resampling time point;

[0059] Secondly, according to the resampling time, the vibration signal is interpolated to find the amplitude of the vibration signal in the angular domain corresponding to the sampling time point, thereby obtaining a stable resampling signal in the angular domain;

[0060] The formula for the resampling time point is:

[0061] θ=β 0 +β 1 t+β 2 t 2 ; (1)

[0062]

[0063] Among them, t is the resampling time point; θ is the angle of the rotor rotation; β 0 , β 1 , β 2 are the initial position, angular velocity and angular acceleration respectively; t 1 ,t 2 ,t 3 are the arrival times of three consecutive pulses; δ is the angular interval between every two pulses.

[0064] The Osprey optimization algorithm, Cauchy mutation strategy and Sparrow algorithm are integrated to improve the Sparrow algorithm (OCSSA), including:

[0065] First, Logistic chaotic mapping is used to increase the diversity of the initial sparrow population;

[0066] The specific formula for generating the initial population of sparrows using Logistic chaotic mapping is:

[0067]

[0068] Among them, a is the control parameter, satisfying a∈[0,4]; i is the current iteration number, is the position of the nth sparrow in the ith iteration, is the position of the nth sparrow in the i+1th iteration.

[0069] Secondly, the original sparrow algorithm is replaced by the osprey optimization algorithm to update the discoverer's position;

[0070] The specific formula for updating the discoverer position in the first phase of the Osprey optimization algorithm is:

[0071]

[0072] in, is the new position of the nth osprey in the mth dimension at the i-th iteration; is the new position of the nth osprey in the mth dimension at the i+1th iteration; SF n,m is a food location detected randomly; r is a random number in the interval [0,1]; I is a random number in the set {1,2};

[0073] Finally, the Cauchy mutation strategy is used to replace the original sparrow algorithm to update the follower position;

[0074] The specific formula for updating the follower position using the Cauchy mutation strategy is:

[0075]

[0076] in, is the new position of the nth osprey in the mth dimension at the i+1th iteration, is the best sparrow position in the i-th iteration, cauchy(0,1) is the standard Cauchy distribution function, Represents the meaning of multiplication.

[0077] Step 3: Use OCSSA to search for the optimal VMD parameter combination [K, α], and use the minimum envelope entropy criterion to obtain the optimal IMF component and extract the fault feature vector;

[0078] Searching for optimal VMD parameters includes:

[0079] Step 31, initializing the improved sparrow algorithm parameters and VMD parameter optimization range, and initializing the sparrow population position vector;

[0080] Step 32, initialize the sparrow population vector position, and VMD decompose the angle domain signal;

[0081] Step 33, using the minimum envelope entropy as the fitness value to update the optimal fitness value and the optimal position of the sparrow population;

[0082] The minimum envelope entropy updates the optimal fitness value of the sparrow population by calculating the fitness value corresponding to each sparrow through formula (7), which is:

[0083]

[0084] Where N is the number of signal sampling points; b ij is the envelope signal obtained after modulation of each modal component; E bi is the envelope entropy; K is the number of modal components.

[0085] Step 34, determine the parameters [K, α] that need to be tuned in VMD, where a is the penalty factor;

[0086] Step 35, continuously update and calculate the fitness values ​​and optimal positions of the sparrows, and re-sort them;

[0087] Step 36, determine whether the termination condition is met; when the number of iterations meets the termination condition, terminate the iteration, and output the parameters [K, α] corresponding to the best sparrow position, and output the optimal solution, otherwise return to step 35.

[0088] Step 37, the optimization is finished, and K and α with the minimum envelope entropy value are selected.

[0089] The extracted feature vectors include: mean, variance, peak, kurtosis, effective value, peak factor, pulse factor, waveform factor and margin factor.

[0090] like Figure 2 As shown, step 4, using Logistic chaos mapping and cyclic population reduction technology to improve CPO;

[0091] The four existing defense strategies of CPO porcupines include:

[0092] The first defense strategy is that when the CP detects the predator, the predator has two choices: move towards it or move away from it. The specific formula is as follows:

[0093]

[0094] in, is the optimal solution of the evaluation function t; is the position of the i-th porcupine at the t-th iteration; is the position of the i-th porcupine at the t+1th iteration; is the vector generated between the current porcupine and a randomly selected porcupine from the population, indicating the position of the predator at the tth iteration; τ 1 is a normally distributed random number; τ 2 is a random value between the interval [0,1];

[0095] The second defense strategy, CP, threatens predators by making noise. When the predator approaches, the porcupine will become louder. The specific formula is as follows:

[0096]

[0097] Among them, τ 3 is a random value generated between 0 and 1, r1 and r2 are two random integers between [1, N]; To simulate the distance between the predator and the porcupine, there are three types of sound; Indicates the location of the predator;

[0098] The third defense strategy is that CP will secrete a foul odor in the surrounding area to prevent predators from approaching. The specific formula is as follows:

[0099]

[0100] Where r2 is a random number between [1, N], γ t , They are the parameters controlling the search direction, the defense factor, and the odor diffusion factor. The specific calculation formula is as follows:

[0101]

[0102] in, is a vector of randomly generated values ​​between [0,1]; rand is a variable randomly generated between [0,1]; N is the population size; t max is the maximum number of iterations; is the objective function value of the i-th porcupine at the t-th iteration, is the objective function value of the i-th porcupine at the t-th iteration; ε is a small value to avoid division by zero;

[0103] Used to simulate three situations that occur in this strategy:

[0104] when When it is 0, the predator stops approaching and both sides maintain a constant distance; if is 1, which means that the predator is close to the CP and the odor release is enhanced; A parameter combination between 0 and 1 means that the predator and the CP remain within an appropriate safe distance, so the CP does not need to emit excessive odor to maintain this relatively stable state.

[0105] The fourth defense strategy is that when the predator is very close to it and attacks, the CP takes physical attack. In this process, the two objects merge strongly, indicating a one-dimensional inelastic collision. The specific formula is as follows:

[0106]

[0107] in, is the optimal solution obtained, which represents the position of the i-th porcupine at the t-th iteration of CP; α is the convergence speed factor; τ 4 and τ 5 is a random value in the interval [0,1]; is the average force of the porcupine affecting the i-th predator at the t-th iteration.

[0108] Improvements to CPO include:

[0109] Use Logistic chaos mapping to increase the diversity of the initialization population;

[0110] The specific formula for generating the initial population of porcupines using the Logistic chaotic map is:

[0111]

[0112] Among them, a is the control parameter, satisfying a∈[0,4]; i is the current iteration number, is the position of the nth porcupine in the ith iteration, is the position of the nth porcupine in the i+1th iteration.

[0113] The cyclic population reduction technique is used to select some porcupines that are not threatened, so as to accelerate the convergence of the global search through dynamic shrinkage, and reintegrate these solutions into the population in the subsequent stage, thereby promoting the diversification of the population structure and effectively preventing falling into the local optimal solution. The specific formula is:

[0114]

[0115] Where N is the number of individuals in the newly generated population; N min is the minimum number of individuals in the newly generated population, so the population size cannot be less than N min ; N′ is the derivative of N; t is the current function evaluation; % represents the remainder or modulus operator; T max is the maximum number of function evaluations; T is a variable that determines the number of loops.

[0116] Step 5: Figure 2 As shown, the improved CPO is used to search for the optimal parameters [n,m] of the random forest algorithm (RF) to build an improved CPO-RF model;

[0117] Step 51, initializing the optimized range of improved CPO parameters and RF parameters and randomly generating a porcupine population;

[0118] Step 52: taking the out-of-bag error as the fitness value;

[0119] Step 53, sorting according to the fitness value of each porcupine and obtaining the current optimal crown porcupine position;

[0120] Step 54: Determine the parameters [n, m] to be tuned in the RF algorithm; where n is the number of decision trees n_estimators, and m is the depth max_depth of the tree.

[0121] Step 55, using a cyclic population reduction strategy to update the population size, continuously updating the optimal fitness value and the best position of the porcupines, and re-ranking;

[0122] Step 56, determine whether the termination condition is met. During the iterative optimization process, iterate based on the fitness value; when the number of iterations meets the termination condition, terminate the iteration, and output the parameters [n, m] corresponding to the best position, and output the optimal solution, otherwise return to step 55;

[0123] Step 57, the optimization is completed; the parameters [n, m] output in step 56 are used as the modeling parameters of the random forest to obtain the best classification model after training.

[0124] Use the improved CPO-RF to build a fault diagnosis classification model, use the training set to train the classification model; use the test set to verify the model and determine the performance of the classification model in fault classification;

[0125] like Figure 4 As shown in the figure, using a multi-condition fault data set, the quality performance of traditional CPO and improved CPO in finding the RF optimal solution is compared. It can be seen that the improved CPO has a faster iteration speed.

[0126] like Figure 5 (a) and (b) show the accuracy comparison between the improved CPO-RF and RF when using a single-condition fault data set; Figure 6 As shown in (a) and (b), the accuracy of the improved CPO-RF and RF is compared when using a multi-condition fault data set. It can be seen that the classification accuracy of the improved CPO-RF is significantly improved.

[0127] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A variable speed bearing fault diagnosis method based on improved crown porcupine optimized random forest, characterized in that: The following steps are involved: Step 1: Collect bearing vibration signals; Step 2: Use COT to convert the vibration signal from a time domain signal to an angle domain signal; Step 3: Use OCSSA to search for the modal components and penalty factors of VMD, and use the criterion of minimum envelope entropy to obtain the optimal IMF components and extract the fault feature vector; Step 4: Improve CPO using Logistic chaos mapping and cyclic population reduction techniques; Step 5: Use the improved CPO to search for the optimal values ​​of the number of decision trees and the depth of the tree in the random forest algorithm training model.

2. The variable speed bearing fault diagnosis method based on improved crown porcupine optimized random forest according to claim 1 is characterized in that: Step 2 specifically includes: The vibration signal is converted from time sampling to angle sampling using phase information to determine the resampling time point; According to the resampling time point, the vibration signal is interpolated, the amplitude of the vibration signal in the angular domain corresponding to the sampling time point is calculated, and the resampling signal in the angular domain is obtained.

3. The variable speed bearing fault diagnosis method based on improved crested porcupine optimized random forest according to claim 1 is characterized in that: Step three specifically includes: Step 31, initializing the parameters of the sparrow algorithm and the optimization range of VMD parameters; Step 32, initialize the sparrow population vector position, and VMD decompose the angle domain signal; Step 33, using the minimum envelope entropy as the fitness value to update the optimal fitness value and the optimal position of the sparrow population; Step 34, determine the tuning parameters in VMD; Step 35, continuously updating the fitness values ​​and optimal positions of the sparrows, and re-ranking them; Step 36: Determine whether the termination condition is met; Step 37: Obtain the optimal parameter modal components and penalty factors.

4. The variable speed bearing fault diagnosis method based on improved crested porcupine optimized random forest according to claim 1 is characterized in that: Step 4 specifically includes: Use Logistic chaos mapping to increase the diversity of the initialization population; Cyclic population reduction was used to select non-threatened porcupines, and global search was performed by dynamic shrinkage.

5. The variable speed bearing fault diagnosis method based on improved crested porcupine optimized random forest according to claim 1 is characterized in that: Step 5 specifically includes: Step 51, initializing the crested porcupine algorithm parameters and the RF parameter optimization range; Step 52: taking the out-of-bag error as the fitness value; Step 53, sorting according to the fitness value of each porcupine to obtain the current optimal crown porcupine position; Step 54: Determine RF tuning parameters; Step 55, using a cyclic population reduction strategy to update the population size, continuously updating the optimal fitness value and the best position of the porcupines, and re-ranking; Step 56: Determine whether the termination condition is met; Step 57: Output the optimal RF parameters.

6. The variable speed bearing fault diagnosis method based on improved crested porcupine optimized random forest according to claim 1 is characterized in that: The bearing status includes: normal, inner ring fault and outer ring fault.

7. The variable speed bearing fault diagnosis method based on improved crested porcupine optimized random forest according to claim 1 is characterized in that: The fault feature vector includes: mean, variance, peak, kurtosis, effective value, peak factor, pulse factor, waveform factor and margin factor.

8. A variable speed bearing fault diagnosis system based on improved crown porcupine optimized random forest, characterized in that: include: a memory for storing instructions executable by a processor; A processor, used to execute instructions to implement the variable speed bearing fault diagnosis method based on improved crested porcupine optimized random forest as described in any one of claims 1-7.

9. A computer readable medium storing a computer program code, characterized in that: When the computer program code is executed by a processor, the variable speed bearing fault diagnosis method based on the improved crested porcupine optimized random forest as described in any one of claims 1 to 7 is implemented.

Images