Gear fault diagnosis method and system based on adaptive extreme learning machine

Abstract

The application discloses a gear fault diagnosis method based on an adaptive extreme learning machine, and comprises the following steps: obtaining a vibration signal of a gear in a fault type state, and identifying a gear fault state based on the vibration signal; extracting a fault feature of the vibration signal to form a feature sample, and dividing the feature sample into a training set and a test set; training an adaptive extreme learning machine network by using the training set of the extracted feature sample; first, a search mode of dividing an interval of hidden layer node numbers is adopted to obtain an optimal node number search area; then, a simulated annealing algorithm is used to quickly iterate the obtained optimal node number search area to obtain a final adaptive hidden layer optimal node number; and finally, the test set is taken as input, the obtained adaptive hidden layer optimal node number is used to construct an extreme learning machine network, and gear fault diagnosis is performed. The application improves the efficiency of determining the hidden layer node number of the extreme learning machine during gear fault diagnosis, and increases the accuracy and stability of the gear fault diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gear fault diagnosis, and particularly relates to a gear fault diagnosis method and system based on an adaptive extreme learning machine. BACKGROUND

[0002] Gears are the key part of power transmission in mechanical equipment, and are widely used in various mechanical equipment. Gears are also one of the components with a high failure frequency. According to statistics, gear failure accounts for about 80% of mechanical equipment transmission failures. If a gear failure occurs in industrial production and is not timely eliminated, it may lead to a decline in product quality, thereby causing serious economic losses. Therefore, it is of great practical significance to realize gear operation state detection and diagnosis. Gear fault diagnosis methods usually use vibration signal information of gears during operation to evaluate the state. The selection of a classification recognition method plays a very important role in gear fault diagnosis, and the performance of the classification method greatly affects the accuracy of fault diagnosis.

[0003] The extreme learning machine is widely used in the fields of fault diagnosis, image classification, wind speed prediction, and character recognition, due to its simple network structure, fast learning speed, and strong generalization ability. The number of nodes in the hidden layer of the extreme learning machine depends on the size of the training sample, the feature dimension, and the complexity of classification. For a long time, there is still a lack of strict theoretical guidance for determining the number of nodes in the hidden layer of the extreme learning machine. For a classification problem, if the number of nodes in the hidden layer is too small, the recognition performance and fitting performance of the network are poor, and it is difficult to handle complex problems. If the number of nodes is too large, the training time is too long, and over-training is likely to occur, which may reduce the recognition performance. Therefore, selecting an appropriate number of nodes in the hidden layer is one of the key factors to improve the performance of the extreme learning machine. Domestic and foreign scholars have conducted a large number of studies on this issue. Their methods can be summarized into four categories, namely, trial-and-error method, empirical formula method, growth / pruning method, intelligent algorithm, and some other comprehensive optimization methods. The trial-and-error method selects the number of nodes by constantly trying. The empirical formula method calculates the number of nodes in the hidden layer of the network by relying on an empirical formula. The growth method starts with the least number of nodes, and then gradually increases new nodes or sub-networks to optimize the network structure. The pruning method is the opposite. The intelligent algorithm generally uses the growth or pruning method. However, these methods have certain limitations. The trial-and-error method is too blind and has too much computational overhead. The pruning and growth methods are the most studied and most in-depth methods, but they have the problem of when to stop, and the computational overhead is also large, which makes them difficult to apply to large-scale classification. The empirical formula method lacks corresponding theoretical support, is effective for specific samples, and lacks a universal formula. The intelligent algorithm also faces the problems of large computational load and slow search speed, and is not suitable for handling complex classification problems. SUMMARY

[0004] The purpose of this invention is to provide a gear fault diagnosis method and system based on adaptive extreme learning machine, which improves the efficiency of determining the number of hidden layer nodes of extreme learning machine in gear fault diagnosis and increases the accuracy and stability of gear fault diagnosis.

[0005] To solve the above technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a gear fault diagnosis method based on an adaptive limit learning machine is provided, including:

[0007] Step S100: Obtain the vibration signal of the gear in the fault type state, and identify the gear fault state based on the vibration signal;

[0008] Step S200: Extract fault features from vibration signals to form feature samples, and divide the feature samples into training set and test set;

[0009] Step S300: Train an adaptive extreme learning machine network based on the training set. First, use the method of segmenting the interval of the number of hidden layer nodes to search for the optimal number of nodes to obtain the optimal number of nodes search region. Then, use the simulated annealing algorithm to iterate the obtained optimal number of nodes in the adaptive hidden layer to obtain the optimal number of nodes.

[0010] Step S400: Using the test sample set as input, construct an extreme learning machine network with the obtained optimal number of nodes in the adaptive hidden layer and perform gear fault diagnosis.

[0011] Furthermore, in step S100, the vibration signal of the gear running state is measured by an accelerometer installed on the outer surface of the input shaft gearbox housing; the gear fault state types include four types: normal, minor damage, moderate damage, and broken teeth.

[0012] Furthermore, in step S200, the method for extracting the fault features of the vibration signal is as follows:

[0013] The vibration signal is decomposed into wavelet packets, and the features are extracted as gear state features. The features are the wavelet packet entropy corresponding to the vibration signal.

[0014] Furthermore, in step S200, before extracting the fault characteristics of the vibration signal, the following steps are also included:

[0015] Wavelet decomposition and reconstruction method is used to perform noise reduction preprocessing on vibration signals in order to remove noise from the vibration signals.

[0016] Further, step S300 includes:

[0017] Step S310: Determine the initial search interval range using empirical formulas and set the termination threshold condition for interval segmentation;

[0018] The number of nodes is calculated using an empirical formula. Using this number as a baseline, the upper and lower limits of the initial search interval are determined proportionally. A search termination threshold is then set proportionally based on the length of the search interval, serving as the termination condition for the slope-based interval segmentation method iteration. The empirical formula and parameters are as follows:

[0019]

[0020] Where n is the number of nodes, N is the number of samples, C is the classification category, θ is the golden ratio, and P is the feature dimension;

[0021] Step S320: Select the segmentation position based on the search interval range;

[0022] The two dividing points are located at the two golden ratio points near the upper and lower boundaries of the search interval. The selection rules for the dividing point positions are as follows:

[0023] x c =[b-θ(ba)]

[0024] x d =[a+θ(ba)]

[0025] Where a and b are the upper and lower boundary endpoints of the search interval, x c Let c be the dividing point near point a, and x be the dividing point. d Let d be the dividing point near b, and θ be the golden ratio.

[0026] Step S330: Calculate the slope of the accuracy change of each interval of the five-segment segmentation;

[0027] Using the gear fault diagnosis and identification accuracy change curve as the target object, a parabolic model is established. The independent variable is the number of hidden layer nodes, and the dependent variable is the identification accuracy corresponding to the number of nodes. The slope is used to analyze the changing trend of the parabolic model to obtain the next search interval. The slope parameter is calculated as follows:

[0028]

[0029] Where k is the interval slope, h1 and h2 are the number of hidden layer nodes of the extreme learning machine, and ε1 and ε2 are the identification accuracy of gear fault diagnosis corresponding to h1 and h2.

[0030] Step S340: The interval segmentation is actually dividing the interval [a, b] into three segments, and then studying the slope changes of the five intervals contained therein to predict the new search interval; analyze the slope change trend of the interval to obtain the interval with gentle changes and containing extreme values, and then perform the next interval search on the obtained node number area; perform iterative search on the interval, and the iterative search ends according to the set termination threshold, and then obtain the optimal node number search area.

[0031] Step S350: After obtaining the optimal number of nodes search area, use the simulated annealing algorithm to iterate and find the optimal number of hidden layer nodes for gear fault diagnosis.

[0032] Furthermore, in step S350, the simulated annealing algorithm method is as follows:

[0033] Take a state S(i-1) as the original solution, and generate the current solution S(i) according to a certain rule. Use an evaluation function to evaluate the merits of the current solution and the original solution. If the current solution is better, then accept the current solution. If the original solution is better, then choose whether to accept the current solution with a certain probability.

[0034] Secondly, a gear fault diagnosis system based on an adaptive extreme learning machine is provided, comprising a memory and a processor; wherein the memory stores a computer program, which, when executed by the processor, can implement the aforementioned gear fault diagnosis method based on an adaptive extreme learning machine.

[0035] Thirdly, a computer storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the aforementioned gear fault diagnosis method based on an adaptive extreme learning machine.

[0036] The present invention has the following beneficial effects:

[0037] This invention constructs an adaptive extreme learning machine (ELM) network structure and uses a segmentation of the hidden layer node number range and simulated annealing algorithm to determine the optimal number of hidden layer nodes. This effectively achieves the optimal node configuration of the network's hidden layers, overcoming the overfitting or underfitting phenomena that occur during network model training due to too many or too few nodes in existing technologies. This invention improves network performance and thus enhances the accuracy of gear fault diagnosis through optimal hidden layer node configuration. Specifically, this invention improves the efficiency and accuracy of gear fault diagnosis by adaptively determining the number of hidden layer nodes in the extreme learning machine. Attached Figure Description

[0038] Figure 1 This is a flowchart of the gear fault diagnosis method based on adaptive limit learning machine of the present invention;

[0039] Figure 2 This is a schematic diagram of the selection of the segmentation position and the segmentation segment when dividing the interval in an embodiment of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] Please refer to Figure 1This invention relates to a gear fault diagnosis method based on an adaptive limit learning machine, comprising:

[0042] Step S100: Obtain the vibration signal of the gear in the fault type state, and identify the gear fault state based on the vibration signal;

[0043] Step S200: Extract the fault features of the vibration signal to form feature samples, and divide the feature samples into training set and test set; Step S300: Train the adaptive extreme learning machine network based on the training set. First, the optimal node number search region is obtained by segmenting the hidden layer node number interval. Then, the optimal node number search region is obtained by using the simulated annealing algorithm to iteratively obtain the optimal number of nodes in the adaptive hidden layer.

[0044] Step S400: Using the test sample set as input, construct an extreme learning machine network with the obtained optimal number of nodes in the adaptive hidden layer and perform gear fault diagnosis.

[0045] The following sections respectively address... Figure 1 The specific steps in the process will be explained in detail.

[0046] In step S100, the vibration signal of the gear running state in this embodiment is measured by an accelerometer installed on the outer surface of the input shaft gearbox housing; the gear fault state types include four types: normal (N), minor damage (S), moderate damage (M), and broken tooth (B).

[0047] In step S200, the vibration signal is preprocessed to remove noise; the fault features of the denoised vibration signal are extracted, and different gear fault states are characterized by the fault features to form feature samples; the feature samples are divided into training set and test set.

[0048] In this embodiment, preprocessing employs wavelet decomposition and reconstruction to remove noise from the vibration signal. Wavelet packet decomposition is performed on the noise-removed vibration signal, and features are extracted as gear state features, where the feature is the wavelet packet entropy corresponding to the vibration signal. In this embodiment, during gear fault diagnosis, the data feature samples are randomly divided into a training sample set and a test sample set at a ratio of 7:3.

[0049] In step S300, step S300 includes:

[0050] Step S310: Determine the initial search interval range using empirical formulas and set the termination threshold condition for interval segmentation;

[0051] The number of nodes, n, is calculated using an empirical formula. Using this number as a baseline, the upper and lower limits of the initial search interval are determined by a ±30% ratio. A search termination threshold is set based on the length of the search interval, serving as the termination condition for the slope-based interval segmentation method iteration. In this embodiment, the termination threshold is set to 10% of the initial search interval length. The termination threshold can be adjusted according to experimental requirements. The slope-based interval segmentation iteration terminates when the interval length is less than or equal to the termination threshold. The empirical formula and parameters are as follows:

[0052]

[0053] Where n is the number of nodes, N is the number of samples, C is the classification category, θ is the golden ratio, and P is the feature dimension;

[0054] Step S320: Select the segmentation position based on the search interval range;

[0055] The formula selects two node positions to divide the interval. The two dividing points are located at the two golden section points near the upper and lower boundaries of the search interval. Dividing the interval according to the golden ratio is relatively fast. Figure 2 As shown, n is the baseline calculated by the empirical formula, and c and d are the dividing points. The interval is divided into three segments based on the position of these nodes, resulting in a total of five intervals L. ac L ad L cd L cb L db This facilitates analysis from both macro and micro perspectives; the rules for selecting the division point location are as follows:

[0056] x c =[b-θ(ba)] (2)

[0057] x d =[a+θ(ba)] (3)

[0058] Where a and b are the upper and lower boundary endpoints of the search interval, x c Let c be the dividing point near point a, and x be the dividing point. d Let d be the dividing point near b, and θ be the golden ratio.

[0059] Step S330: Calculate the slope of the accuracy change of each interval of the five-segment segmentation;

[0060] Using the gear fault diagnosis and identification accuracy curve (i.e., the Extreme Learning Machine classification task identification accuracy curve) as the target object, a parabolic model is established. The independent variable is the number of hidden layer nodes, and the dependent variable is the identification accuracy corresponding to the number of nodes. The slope is used to analyze the changing trend of the parabolic model. The slope can usually reflect the changing trend of the curve well. By dividing the curve into segments for judgment, the overall and local changing trends of the curve can be quickly obtained, and the next search interval can be predicted. The Extreme Learning Machine obtains the gear fault diagnosis identification accuracy for the number of nodes at four points a, b, c, and d. After completing the above steps, the number of nodes at the interval endpoints and the two dividing points is known, totaling four nodes. At this time, the search interval contains five sub-intervals. The slope of the identification accuracy change between each sub-interval is obtained by the number of nodes at the interval endpoints and the identification accuracy. The identification accuracy is the result of the Extreme Learning Machine performing the gear fault operation state classification task. The slope of the accuracy change of the five segments is calculated by the number of nodes at each point and its corresponding identification accuracy. The slope parameter calculation formula is as follows:

[0061]

[0062] Where k is the interval slope, h1 and h2 are the number of hidden layer nodes of the extreme learning machine, and ε1 and ε2 are the identification accuracy of gear fault diagnosis corresponding to h1 and h2.

[0063] Step S340: The interval segmentation essentially divides the interval [a, b] into three segments, then studies the slope changes of the five segments it contains to predict new search intervals. Based on the slope analysis of the interval's changing trend, the interval with gentle changes and containing extreme values ​​is determined as the next search interval. Gentle changes mean the interval with the lowest slope value for the change in identification accuracy. Using the slope change in accuracy and the identification accuracy of gear fault diagnosis as two criteria, intervals with gentle changes and high accuracy are selected. The iteration stops when the interval length reaches the set termination threshold, obtaining the optimal hidden layer node number search area.

[0064] Step S350: After obtaining the optimal number of nodes search area, use the simulated annealing algorithm to iterate and find the optimal number of hidden layer nodes for gear fault diagnosis;

[0065] Generally, the nodes in the optimal node number search region basically meet the requirements. To obtain the optimal node number and thus improve classification accuracy, after obtaining the optimal node number search region, the simulated annealing algorithm can be used for rapid iteration to find the optimal hidden layer node number that satisfies the requirements of the Extreme Learning Machine for diagnosing gear faults. The simulated annealing algorithm starts from a relatively high initial temperature and, as the temperature parameter decreases, randomly searches for the global optimal solution of the objective function in the solution space by combining the probability jump characteristic. The simulated annealing algorithm includes the following steps:

[0066] Let S0 be the initial solution, with an initial temperature of T0, a final temperature of e, a cooling rate of alpha, an iteration count of M, and an initial state time of i. After each iteration, the temperature decreases, and any state S(i-1) is taken as the original solution. A current solution S(i) is generated according to a predetermined method. An evaluation function is used to compare the current solution with the original solution. If the current solution is better, it is accepted; if the original solution is better, a decision is made with a certain probability as to whether to accept the current solution.

[0067] The formula for temperature decrease is as follows:

[0068] T(i) = alpha * T(i-1) (s)

[0069] The current temperature is the product of the cooling rate and the temperature of the previous stage. When i is 1, the temperature of the next stage can be obtained from T0.

[0070] The evaluation function formula is as follows:

[0071] ΔC=C(S(i))-C(S(i-1)) (6)

[0072] Where S(i) is the number of nodes in the current state, S(i-1) is the number of nodes in the previous state, and C(S(i)) is the gear fault signal classification accuracy when taking this number of nodes. If ΔC is greater than zero, S(i) is accepted as a new solution; if it is less than zero, the following probability formula is used to select whether to accept this solution.

[0073] The probability formula is as follows:

[0074]

[0075] If i reaches the required number of iterations M or the temperature T(i) is less than the final temperature e, then the iteration stops. The result S(i) is the final optimal number of hidden layer nodes.

[0076] In summary, this invention proposes a gear fault diagnosis method based on an adaptive extreme learning machine (ELM). This method uses interval search for the number of hidden layer nodes and simulated annealing to select the optimal number of hidden layer nodes for fault diagnosis. The proposed method requires minimal experimentation or complex algorithms, and can quickly select a suitable number of hidden layer nodes. An ELM with an appropriate number of hidden layer nodes exhibits fast learning speed and good generalization performance in gear fault diagnosis. Experimental results show that the gear fault diagnosis accuracy for 1800 samples with 16-dimensional features and 4 classifications is 97.75%, representing an improvement in both speed and accuracy compared to similar methods using the same database.

[0077] The present invention also provides a gear fault diagnosis system based on an adaptive extreme learning machine, comprising a memory and a processor; wherein the memory stores a computer program, which, when executed by the processor, can implement the aforementioned gear fault diagnosis method based on an adaptive extreme learning machine.

[0078] The present invention also provides a computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the above-described gear fault diagnosis method based on an adaptive extreme learning machine.

[0079] All parts not covered in this invention are the same as or implemented using existing technologies.

[0080] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A gear fault diagnosis method based on adaptive limit learning machine, characterized in that: include Step S100: Obtain the vibration signal of the gear in the fault type state, and identify the gear fault state based on the vibration signal; Step S200: Extract fault features from vibration signals to form feature samples, and divide the feature samples into training set and test set; Step S300: Train an adaptive extreme learning machine network based on the training set. First, use the method of segmenting the interval of the number of hidden layer nodes to search for the optimal number of nodes to obtain the optimal number of nodes search region. Then, use the simulated annealing algorithm to iterate the obtained optimal number of nodes in the adaptive hidden layer to obtain the optimal number of nodes. include: Step S310: Determine the initial search interval range using empirical formulas and set the termination threshold condition for interval segmentation; The number of nodes is calculated using an empirical formula. Using this number as a baseline, the upper and lower limits of the initial search interval are determined proportionally. A search termination threshold is then set proportionally based on the length of the search interval, serving as the termination condition for the slope-based interval segmentation method iteration. The empirical formula and parameters are as follows: ； in, For the number of nodes, For the sample size, For classification categories, The golden ratio, For feature dimensions; Step S320: Select the segmentation position based on the search interval range; The two dividing points are located at the two golden ratio points near the upper and lower boundaries of the search interval. The selection rules for the dividing point positions are as follows: ； ； in, The upper and lower boundaries of the search interval are the endpoints. To get closer The dividing point at the location , To get closer The dividing point at the location , The golden ratio; Step S330: Calculate the slope of the accuracy change of each interval of the five-segment segmentation; Using the gear fault diagnosis and identification accuracy change curve as the target object, a parabolic model is established. The independent variable is the number of hidden layer nodes, and the dependent variable is the identification accuracy corresponding to the number of nodes. The slope is used to analyze the changing trend of the parabolic model to obtain the next search interval. The slope parameter is calculated as follows: ； in, The slope of the interval. It is the number of hidden layer nodes in the Extreme Learning Machine. yes The accuracy rate of gear fault diagnosis; Step S340: The interval segmentation is actually dividing the interval [a, b] into three segments, and then studying the slope changes of the five intervals contained therein to predict the new search interval; analyze the slope change trend of the interval to obtain the interval with gentle changes and containing extreme values, and then perform the next interval search on the obtained node number area; perform iterative search on the interval, and the iterative search ends according to the set termination threshold, and then obtain the optimal node number search area. Step S350: After obtaining the optimal number of nodes search area, use the simulated annealing algorithm to iterate and find the optimal number of hidden layer nodes for gear fault diagnosis; Step S400: Using the test sample set as input, construct an extreme learning machine network with the obtained optimal number of nodes in the adaptive hidden layer and perform gear fault diagnosis.

2. The gear fault diagnosis method based on adaptive limit learning machine according to claim 1, characterized in that: In step S100, the vibration signal of the gear running state is measured by an accelerometer installed on the outer surface of the input shaft gearbox housing; the gear fault state types include four types: normal, minor damage, moderate damage, and broken teeth.

3. The gear fault diagnosis method based on adaptive limit learning machine according to claim 1, characterized in that: In step S200, the method for extracting the fault features of the vibration signal is as follows: The vibration signal is decomposed into wavelet packets, and the features are extracted as gear state features. The features are the wavelet packet entropy corresponding to the vibration signal.

4. The gear fault diagnosis method based on adaptive limit learning machine according to claim 1, characterized in that: Before extracting the fault characteristics of the vibration signal in step S200, the following steps are also included: Wavelet decomposition and reconstruction method is used to perform noise reduction preprocessing on vibration signals in order to remove noise from the vibration signals.

5. The gear fault diagnosis method based on adaptive limit learning machine according to claim 1, characterized in that: In step S350, the simulated annealing algorithm method is as follows: In a state As the original solution, and to generate the current solution according to a certain prescribed method. An evaluation function is used to evaluate the merits of the current solution and the original solution; If the current solution is better, then accept the current solution; If the original solution is better, then there is a certain probability of choosing whether to accept the current solution.

6. A gear fault diagnosis system based on adaptive extreme learning machine, characterized in that: Including memory and processor; The memory stores a computer program that, when executed by a processor, can implement the gear fault diagnosis method based on an adaptive extreme learning machine as described in any one of claims 1 to 5.

7. A computer storage medium storing a computer program thereon, characterized in that: When the computer program is executed by the processor, it implements the gear fault diagnosis method based on adaptive extreme learning machine as described in any one of claims 1 to 5.

Images