Gear fault diagnosis method based on linear discriminant analysis and random forest
By combining linear discriminant analysis and random forest methods, LDA-RFA gear fault diagnosis model is constructed and the model parameters are optimized, which solves the problem of relying on a large amount of label data and poor generalization ability in the existing technology, and realizes the noise environment High accuracy fault diagnosis.
Patent Information
- Application Number
- CN202310995364.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-08-09
- Publication Date
- 2025-02-18
AI Technical Summary
The prior art relies on a large amount of labeled fault data in gear fault diagnosis, and the model generalization ability is reduced, making it difficult to achieve accurate fault diagnosis in a noisy environment.
The gear fault diagnosis method based on linear discriminant analysis and random forests is adopted to construct the fault diagnosis model of LDA-RFA through pre-processing and feature extraction of vibration acceleration signals, and the model parameters are optimized using the mutant particle swarm optimization algorithm.
It realizes automatic extraction of fault characteristics and end-to-end intelligent fault diagnosis in a noisy environment, improving the accuracy of gear fault diagnosis and generalization capabilities of model.
Smart Images
Figure CN119469761A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of gear fault diagnosis, and in particular relates to a gear fault diagnosis method based on linear discriminant analysis and random forest. Background Art
[0002] Planetary gear transmission system has unique advantages such as compact structure, small size, light weight, strong load-bearing capacity, large transmission ratio, high transmission efficiency and stable operation. Therefore, it is widely used in various fields such as machine tools, petrochemical industry, metallurgy and ore. However, in the actual working process, due to the harsh working environment and the long-term service of the equipment, the gears are very prone to wear and surface damage, which in turn affects the healthy operation of the system and even causes catastrophic accidents. Therefore, the research on intelligent diagnosis of gears is very important, which can not only ensure the normal operation of the equipment to a certain extent, but also realize the early detection and early resolution of faults on this basis, providing important protection for the safety of equipment and personnel.
[0003] Early research on gear fault diagnosis was completed by processing and analyzing the vibration signals collected by sensors. However, with the continuous advancement of data acquisition methods, the fault data set obtained from experiments has continued to expand, and a single signal analysis technology can no longer accurately judge gear faults. In recent years, the field of machine learning has developed rapidly, setting off another wave of artificial intelligence development, and this technology has also been applied to fault diagnosis. Therefore, intelligent diagnosis technology that organically combines machine learning models such as neural networks, random forests, and support vector machines with the characteristic parameters of gears for comprehensive analysis has developed rapidly.
[0004] Although machine learning has achieved end-to-end gear fault diagnosis, its characteristic of adaptively updating network model parameters according to fault data leads to various problems such as model parameter and hyperparameter combination. A large amount of labeled data is required to achieve the ideal convergence state. Therefore, this method relies on a sufficient amount of labeled fault data. However, in the actual working process of the equipment, it is almost in a healthy service state, which makes it extremely difficult to obtain fault data, resulting in insufficient labeled fault data for some parts, making it difficult to conduct accurate model training. In addition, the distribution between the training set and the test set will change due to different operating conditions, which reduces the generalization ability of the fault diagnosis method based on machine learning. Summary of the invention
[0005] Based on the above deficiencies in the prior art, the technical problem solved by the present invention is to provide a gear fault diagnosis method based on linear discriminant analysis and random forest, which utilizes vibration acceleration signals to construct a fault diagnosis model, and uses a combination of linear discriminant analysis and random forest to accurately extract fault features.
[0006] In order to solve the above technical problems, the present invention is implemented by the following technical solutions: The present invention provides a gear fault diagnosis method based on linear discriminant analysis and random forest, comprising:
[0007] S1: Under normal working and fault working conditions of the gear, the vibration acceleration signal is collected by using the acceleration sensor, and the fault sample data set is constructed by random sampling method;
[0008] S2: According to the fault signal obtained in step S1, the obtained vibration signal is subjected to improved wavelet threshold denoising combined with CEEMDAN to perform signal preprocessing;
[0009] S3: Based on the signal processed in step S2, the signal containing the fault is input into the random forest classifier, and the gear fault diagnosis model of LDA-RFA is established by combining the linear discriminant analysis method;
[0010] S4: Based on the fault diagnosis model constructed in step S3, a mutation particle swarm optimization algorithm is used to optimize the parameters of the constructed gear fault diagnosis model, obtain the setting of the optimal parameters of the model, and train the model;
[0011] S5: Use the experimental data to input the established model and conduct model testing to complete the gear fault diagnosis work.
[0012] Furthermore, the step S1 constructs a fault sample set for each sensor, specifically including:
[0013] S11: Under the five conditions of normal gear and different faults, vibration acceleration signals returned by sensors at different positions are collected respectively;
[0014] S12: Continuously sample each vibration acceleration signal returned by each sensor in various states, once every T moments, and collect M samples in total;
[0015] S13: Connect samples collected by the same sensor in various states to establish a fault sample set under the same sensor.
[0016] Further, the step S2 performs improved wavelet threshold denoising combined with CEEMDAN on the obtained vibration signal to perform signal preprocessing, specifically including:
[0017] S21: Add Gaussian white noise to the original signal to obtain a noisy signal, perform EMD decomposition on the noisy signal to obtain a first-order intrinsic mode component, perform overall averaging on the obtained n first-order intrinsic mode components to obtain the first intrinsic mode component IMF1 of CEEMDAN decomposition, and calculate the residual after removing the first mode component;
[0018] S22: Add positive and negative paired Gaussian white noise to the obtained residual signal to obtain a new signal, perform EMD decomposition on the new signal, take the first-order intrinsic mode component after decomposition, and then calculate the second intrinsic mode IMF2 decomposed by CEEMDAN, calculate the residual after removing the second mode component, repeat the above steps until the obtained residual signal is a monotonic function and cannot be further decomposed, at this time, K intrinsic mode components and residuals of the original signal are obtained;
[0019] S23: Smoothing the correlated components after CEEMDAN decomposition, performing wavelet decomposition on the uncorrelated components, performing wavelet threshold denoising on the obtained high-frequency signals, and reconstructing the low-frequency signals together with the smoothed components.
[0020] Further, the step S3 is based on the random forest classifier and combined with the linear discriminant analysis method to establish a gear fault diagnosis model of LDA-RFA, which specifically includes:
[0021] S31: Using a linear discriminant analysis method to perform dimensionality reduction processing on the collected signal to form a fault feature set A;
[0022] S32: Uniformly sample the fault feature set A using a sampling method with replacement, and randomly generate n fault feature sets , according to the obtained training set, the corresponding decision tree is obtained ;
[0023] S33: Input the test set sample X into each decision tree for discrimination, and obtain the discrimination results of n decision trees Among the n decision tree output results, the number with the largest number of output results is taken as the final result and as the category to which the test set sample X belongs.
[0024] Further, the step S4 optimizes the parameters of the constructed gear fault diagnosis model to obtain the optimal parameters of the model, which specifically includes:
[0025] S41: Initialize the parameters of the variant particle swarm algorithm and the parameters to be optimized, and randomly set the initial position and initial velocity of each particle;
[0026] S42: using LDA and RFA to process the gear fault data set, and using the data classification accuracy as the fitness, to obtain the optimal position of the initial particle and the optimal position of all particles in the population;
[0027] S43: updating the speed and position of all particles and performing mutation operations on all particles, processing the gear fault data set using LDA and RFA again, calculating the particle fitness and updating the optimal position of the initial particle and the optimal position of all particles in the population;
[0028] S44: Determine whether the model termination condition is met. If so, the optimization ends; if not, continue the optimization and finally obtain the optimal parameters.
[0029] Furthermore, the step S5 specifically includes:
[0030] S51: Under the five states of normal gear and different faults, respectively collecting the vibration acceleration signals returned by the acceleration sensor;
[0031] S52: The vibration acceleration signal samples are calculated according to the fault diagnosis model evaluated in step S4, and the frequency domain characteristics of the vibration acceleration signal samples are calculated for fault diagnosis;
[0032] S53: The gear is subdivided into five states, namely, normal state and different fault states, to determine the accuracy of fault diagnosis in each state.
[0033] From the above, the gear fault diagnosis method based on linear discriminant analysis and random forest of the present invention has at least the following beneficial effects:
[0034] 1. The present invention realizes the automatic extraction of fault diagnosis and end-to-end intelligent fault diagnosis in a noisy environment. First, the vibration signal of the gear is collected and extracted by the sensor, and then the obtained vibration signal is subjected to the improved wavelet threshold denoising combined with CEEMDAN to pre-process the signal; then, based on the random forest classifier and the linear discriminant analysis method, the gear fault diagnosis model of LDA-RFA is established, and the parameters of the constructed gear fault diagnosis model are optimized to obtain the optimal parameters of the model. The test data is input into the diagnosis for diagnosis to complete the gear fault detection; this method uses a fault diagnosis method based on the combination of linear discriminant analysis and random forest for feature processing, accurately extracts fault features, and improves the accuracy of bearing fault diagnosis.
[0035] 2. The present invention performs improved wavelet threshold denoising combined with CEEMDAN on the obtained vibration signal and pre-processes the original signal. Secondly, the present invention realizes the identification and diagnosis of the fault state based on a fault diagnosis method combining linear discriminant analysis and random forest.
[0036] 3. The method proposed in the present invention has higher fault diagnosis accuracy compared with other methods, and can obtain accurate fault diagnosis results even in the presence of noise interference and small sample data. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0038] Figure 1 is a flow chart of the gear fault diagnosis method of the present invention;
[0039] Figure 2 Performance analysis of classifier models for fault diagnosis;
[0040] Figure 3 is the accuracy of the fault diagnosis model. DETAILED DESCRIPTION
[0041] The specific implementation of the gear fault diagnosis method based on linear discriminant analysis and random forest of the present invention is described in detail below with reference to the accompanying drawings.
[0042] like Figures 1 to 3 As shown, the present invention provides a gear fault diagnosis method based on linear discriminant analysis and random forest, comprising the following steps:
[0043] S1: Under normal working and fault working conditions of the gear, the vibration acceleration signal is collected by using the acceleration sensor, and the fault sample data set is constructed by random sampling method;
[0044] S2: According to the fault signal obtained in step S1, the obtained vibration signal is subjected to improved wavelet threshold denoising combined with CEEMDAN to perform signal preprocessing;
[0045] S3: Based on the signal processed in step S2, the signal containing the fault is input into the random forest classifier, and the gear fault diagnosis model of LDA-RFA is established by combining the linear discriminant analysis method;
[0046] S4: Based on the fault diagnosis model constructed in step S3, a mutation particle swarm optimization algorithm is used to optimize the parameters of the constructed gear fault diagnosis model, obtain the setting of the optimal parameters of the model, and train the model;
[0047] S5: Use the experimental data to input the established model and conduct model testing to complete the gear fault diagnosis work.
[0048] The step S1 constructs a fault sample set for each sensor, specifically including:
[0049] S11: Under the five conditions of normal gear and different faults, vibration acceleration signals returned by sensors at different positions are collected respectively;
[0050] S12: Continuously sample each vibration acceleration signal returned by each sensor in various states, once every T moments, and collect M samples in total;
[0051] S13: Connect samples collected by the same sensor in various states to establish a fault sample set under the same sensor.
[0052] The step S2 performs improved wavelet threshold denoising combined with CEEMDAN on the obtained vibration signal to perform signal preprocessing, specifically including:
[0053] S21: Add Gaussian white noise to the original signal to obtain a noisy signal, perform EMD decomposition on the noisy signal to obtain a first-order intrinsic mode component, perform overall averaging on the obtained n first-order intrinsic mode components to obtain the first intrinsic mode component IMF1 of CEEMDAN decomposition, and calculate the residual after removing the first mode component;
[0054] S22: Add positive and negative paired Gaussian white noise to the obtained residual signal to obtain a new signal, perform EMD decomposition on the new signal, take the first-order intrinsic mode component after decomposition, and then calculate the second intrinsic mode IMF2 decomposed by CEEMDAN, calculate the residual after removing the second mode component, repeat the above steps until the obtained residual signal is a monotonic function and cannot be further decomposed, at this time, K intrinsic mode components and residuals of the original signal are obtained;
[0055] S23: Smooth the relevant components after CEEMDAN decomposition, perform wavelet decomposition on the unrelated components, perform wavelet threshold denoising on the obtained high-frequency signals, and compare the low-frequency signals with the smoothed ones.
[0056] The step S3 is based on the random forest classifier and combined with the linear discriminant analysis method to establish a gear fault diagnosis model of LDA-RFA, which specifically includes:
[0057] S31: Using a linear discriminant analysis method to perform dimensionality reduction processing on the collected signal to form a fault feature set A;
[0058] S32: Uniformly sample the fault feature set A using a sampling method with replacement, and randomly generate n fault feature sets , according to the obtained training set, the corresponding decision tree is obtained ;
[0059] S33: Input the test set sample X into each decision tree for discrimination, and obtain the discrimination results of n decision trees Among the n decision tree output results, the number with the largest number of output results is taken as the final result and as the category to which the test set sample X belongs.
[0060] The step S4 optimizes the parameters of the constructed gear fault diagnosis model to obtain the optimal parameters of the model, specifically including:
[0061] S41: Initialize the parameters of the variant particle swarm algorithm and the parameters to be optimized, and randomly set the initial position and initial velocity of each particle;
[0062] S42: using LDA and RFA to process the gear fault data set, and using the data classification accuracy as the fitness, to obtain the optimal position of the initial particle and the optimal position of all particles in the population;
[0063] S43: updating the speed and position of all particles and performing mutation operations on all particles, processing the gear fault data set using LDA and RFA again, calculating the particle fitness and updating the optimal position of the initial particle and the optimal position of all particles in the population;
[0064] S44: Determine whether the model termination condition is met. If so, the optimization ends; if not, continue the optimization and finally obtain the optimal parameters.
[0065] Furthermore, the step S5 specifically includes:
[0066] S51: Under the five states of normal gear and different faults, respectively collecting the vibration acceleration signals returned by the acceleration sensor;
[0067] S52: the vibration acceleration signal samples are preprocessed and calculated according to the fault diagnosis model evaluated in step S4, and fault diagnosis is performed according to the frequency domain characteristics of the vibration acceleration signal samples;
[0068] S53: The gear is subdivided into five states: normal, pitting, wear, crack, and broken tooth faults, to determine the accuracy of fault diagnosis in each state.
[0069] The experimental data set sampling signal of the present invention is the gear normal, crack, pitting, wear and broken tooth data with a speed set to 500r / min and a sampling frequency of 50KHz. The training set is 3600 and the test set is 200. After preprocessing, the data is transmitted to the network model for experiments.
[0070] Assume the dimension of the optimization space is , the total number of particles is N , among which The position and velocity of a particle are and , the position and velocity of each generation of particles are iteratively updated according to the following formula.
[0071]
[0072] is the number of iterations;
[0073]
[0074] is the inertia weight factor, For the The optimal position searched by a particle; is the optimal position of all particles in the population; and is the acceleration step length; and is a random number in the interval (0,1);
[0075] Initialize the parameters of the variant particle swarm algorithm and the parameters to be optimized, and randomly set the initial position and initial velocity of each particle;
[0076] Calculate the parameter value corresponding to each particle according to the initial position of the particle;
[0077] LDA-RFA is used to process the gear fault data set, and the data classification accuracy is used as the fitness to obtain the initial and ;
[0078] Update the speed and position of all particles according to the iterative update formula, and perform mutation operations on all particles;
[0079] Use LDA-RFA to process the gear fault dataset again, calculate the particle fitness, and update and ;
[0080] Determine whether the model termination condition is met (the maximum number of iterations is reached or the difference in fitness between two consecutive iterations is less than the set threshold). If so, the optimization ends; if not, the optimization continues.
[0081] The established fault diagnosis model is trained according to the obtained optimal parameters to complete the gear fault diagnosis work.
[0082] Finally, it should be noted that the above is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and changes can be made without departing from the principle of the present invention, and these improvements and changes are also regarded as the protection scope of the present invention.
Claims
1. A gear fault diagnosis method based on linear discriminant analysis and random forest, characterized in that: include: S1: Under normal working and fault working conditions of the gear, the vibration acceleration signal is collected by using the acceleration sensor, and the fault sample data set is constructed by random sampling method; S2: According to the fault signal obtained in step S1, the obtained vibration signal is subjected to improved wavelet threshold denoising combined with CEEMDAN to perform signal preprocessing; S3: Based on the signal processed in step S2, the signal containing the fault is input into the random forest classifier, and the gear fault diagnosis model of LDA-RFA is established by combining the linear discriminant analysis method; S4: Based on the fault diagnosis model constructed in step S3, a mutation particle swarm optimization algorithm is used to optimize the parameters of the constructed gear fault diagnosis model, obtain the setting of the optimal parameters of the model, and train the model; S5: Use the experimental data to input the established model and conduct model testing to complete the gear fault diagnosis work.
2. The gear fault diagnosis method based on linear discriminant analysis and random forest according to claim 1 is characterized in that: The step S1 constructs a gear fault sample data set, specifically comprising: S11: Under the five conditions of normal gear and different faults, vibration acceleration signals returned by sensors at different positions are collected respectively; S12: Continuously sample each vibration acceleration signal returned by each sensor in various states, once every T moments, and collect M samples in total; S13: Connect samples collected by the same sensor in various states to establish a fault sample set under the same sensor.
3. The gear fault diagnosis method based on linear discriminant analysis and random forest according to claim 1 is characterized in that: The step S2 performs improved wavelet threshold denoising combined with CEEMDAN on the obtained vibration signal to perform signal preprocessing, specifically including: S21: Add Gaussian white noise to the original signal to obtain a noisy signal, perform EMD decomposition on the noisy signal to obtain a first-order intrinsic mode component; perform overall averaging on the obtained n first-order intrinsic mode components to obtain the first intrinsic mode component IMF1 decomposed by CEEMDAN and calculate the residual after removing the first mode component; S22: Add positive and negative paired Gaussian white noise to the obtained residual signal to obtain a new signal, perform EMD decomposition on the new signal, take the first-order intrinsic mode component after decomposition, and then calculate the second intrinsic mode IMF2 decomposed by CEEMDAN, and calculate the residual after removing the second mode component; repeat the above steps until the obtained residual signal is a monotonic function and cannot be further decomposed, at this time, K intrinsic mode components and residuals of the original signal are obtained; S23: Smoothing the correlated components after CEEMDAN decomposition, performing wavelet decomposition on the uncorrelated components, performing wavelet threshold denoising on the obtained high-frequency signals, and reconstructing the low-frequency signals together with the smoothed components.
4. The gear fault diagnosis method based on linear discriminant analysis and random forest according to claim 1 is characterized in that: The step S3 is based on the random forest classifier and combined with the linear discriminant analysis method to establish a gear fault diagnosis model of LDA-RFA, which specifically includes: S31: Using a linear discriminant analysis method to perform dimensionality reduction processing on the collected signal to form a fault feature set A; S32: Uniformly sample the fault feature set A using a sampling method with replacement, and randomly generate n fault feature sets ， Get the corresponding decision tree based on the obtained training set ； S33: Input the test set sample X into each decision tree for discrimination, and obtain the discrimination results of n decision trees Among the n decision tree output results, the number with the largest number of output results is taken as the final result and as the category to which the test set sample X belongs.
5. The gear fault diagnosis method based on linear discriminant analysis and random forest according to claim 1, characterized in that: The step S4 optimizes the parameters of the constructed gear fault diagnosis model to obtain the optimal parameters of the model, specifically including: S41: Initialize the parameters of the variant particle swarm algorithm and the parameters to be optimized, and randomly set the initial position and initial velocity of each particle; S42: using LDA and RFA to process the gear fault data set, and using the data classification accuracy as the fitness, to obtain the optimal position of the initial particle and the optimal position of all particles in the population; S43: updating the speed and position of all particles and performing mutation operations on all particles, processing the gear fault data set using LDA-RFA again, calculating the particle fitness and updating the optimal position of the initial particle and the optimal position of all particles in the population; S44: Determine whether the model termination condition is met. If so, the optimization ends; if not, continue the optimization and finally obtain the optimal parameters.
6. The gear fault diagnosis method based on linear discriminant analysis and random forest according to claim 1, characterized in that: The step S5 specifically includes: S51: Under the five states of normal gear and different faults, respectively collecting the returned vibration acceleration signals collected by the acceleration sensor; S52: The vibration acceleration signal samples are calculated according to the fault diagnosis model evaluated in step S4, and the frequency domain characteristics of the vibration acceleration signal samples are calculated for fault diagnosis; S53: The gear is subdivided into five states, namely, normal state and different fault states, to determine the accuracy of fault diagnosis in each state.