Gearbox fault detection method, device, electronic device and storage medium

Abstract

The present application provides a gearbox fault detection method, device, electronic device, and storage medium. The method includes: obtaining the original vibration signals of the gearbox under test when the gears of the gearbox under test are in different states; extracting the features of the original vibration signals sampled with a fixed sampling length from all the original vibration signals to obtain a feature vector corresponding to each sampling under the fixed sampling length that characterizes the working condition of the gearbox under test; selecting different parameters from a preset parameter range as smoothing factors of a preset PNN model, and cross-validating the preset PNN model using the smoothing factors and feature vectors to obtain the accuracy of the preset PNN model under different smoothing factors, and determining the target parameters based on the accuracy; constructing a gearbox fault detection model based on the preset PNN model and the target parameters; and determining the fault type of the gearbox under test using the gearbox fault detection model. In this way, the problem of insufficient accuracy in current gearbox fault detection methods can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to a gearbox fault detection method, device, electronic equipment and storage medium. Background Art

[0002] Gearboxes are widely used in numerous mechanical devices, as they transmit torque, change transmission direction, and distribute power. The condition of a gearbox significantly impacts the proper functioning of the equipment. Therefore, strict monitoring of gearbox failures is crucial to prevent undetected gearbox failures from causing major production accidents.

[0003] Current gearbox fault detection typically uses a neural network model to analyze characteristic data representing the gearbox's operating conditions to determine the fault. This approach suffers from inaccurate results and is prone to false positives, which increases maintenance costs for mechanical equipment. Summary of the Invention

[0004] In view of this, the purpose of the embodiments of the present application is to provide a gearbox fault detection method, device, electronic device and storage medium, which can improve the problem of insufficient accuracy of current gearbox fault detection methods.

[0005] To achieve the above technical objectives, the technical solutions adopted in this application are as follows:

[0006] In a first aspect, an embodiment of the present application provides a gearbox fault detection method, characterized in that the method includes:

[0007] Acquire original vibration signals of the gearbox under test when the gears of the gearbox under test are in different states;

[0008] Extracting features of the original vibration signals sampled with a fixed sampling length from all original vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be tested;

[0009] Selecting different parameters from a preset parameter range as smoothing factors of a preset PNN model, and cross-validating the preset PNN model using the smoothing factors and the feature vector to obtain the accuracy of the preset PNN model under different smoothing factors, and determining the smoothing factor corresponding to the highest accuracy as the target parameter;

[0010] Constructing a gearbox fault detection model according to the preset PNN model and the target parameters;

[0011] The gearbox fault detection model is used to determine the fault type of the gearbox to be tested.

[0012] In combination with the first aspect, in some optional embodiments, the features of the original vibration signals sampled with a fixed sampling length are extracted from all original vibration signals to obtain a feature vector corresponding to each sampling at the fixed sampling length that characterizes the working condition of the gearbox to be measured, including:

[0013] For each group of the original vibration signals sampled with a fixed sampling length, constructing a third-order tensor according to the original vibration signals through continuous wavelet transformation;

[0014] determining the number of principal components of the third-order tensor by a kernel consistency diagnostic method;

[0015] Decomposing the third-order tensor by parallel factor analysis to obtain a corresponding trilinear model according to the third-order tensor and the number of principal components;

[0016] Extracting a frequency signal and a time domain signal from the trilinear model, and determining a centroid frequency corresponding to the frequency signal according to the frequency signal, and determining a kurtosis index corresponding to the time domain signal according to the time domain signal;

[0017] The feature vector is constructed according to the center of gravity frequency and the kurtosis index.

[0018] In conjunction with the first aspect, in some optional implementations, different parameters are selected from a preset parameter range as smoothing factors of a preset PNN model, and the preset PNN model is cross-validated using the smoothing factors and the feature vector to obtain the accuracy of the preset PNN model under different smoothing factors, and the smoothing factor corresponding to the highest accuracy is determined as the target parameter, including:

[0019] Initializing relevant parameters, including the number of vectors used to search for the target parameter from the preset parameter range, a maximum number of iterations, an initial position of the vector, and an initial global optimal position;

[0020] For each of the vectors, updating the initial position of the vector based on a first preset rule to obtain a vector with an updated position;

[0021] Merging the vector after the position update with the vector that meets the preset conditions to obtain a first new vector;

[0022] Bringing the parameter corresponding to the position of the first new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor;

[0023] Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the initial global optimal position, determining the position with the higher accuracy as the new optimal position;

[0024] According to the new optimal position, updating the first new vector based on a second preset rule to obtain a second new vector;

[0025] Bringing the parameter corresponding to the position of the second new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor;

[0026] Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the new optimal position, determining the position with the higher accuracy as the target optimal position;

[0027] Taking the target optimal position as the initial global optimal position, and updating the initial position of the vector based on the first preset rule to obtain the updated vector, repeating the above steps until the number of iterations is greater than or equal to the maximum number of iterations;

[0028] The smoothing factor corresponding to the target optimal position obtained in the last iteration is determined as the target parameter.

[0029] In conjunction with the first aspect, in some optional implementations, the method further includes:

[0030] Performing noise reduction processing on the original vibration signal by wavelet packet transform to obtain a noise-reduced vibration signal;

[0031] The features of the original vibration signals sampled with a fixed sampling length are extracted from all the original vibration signals to obtain a feature vector corresponding to each sampling under the fixed sampling length, which characterizes the working condition of the gearbox to be tested, including:

[0032] The features of the noise-reduced vibration signals sampled with a fixed sampling length are extracted from all noise-reduced vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be measured.

[0033] In a second aspect, an embodiment of the present application further provides a gearbox fault detection device, the device comprising:

[0034] an acquisition unit, configured to acquire original vibration signals of the gearbox under test when the gears of the gearbox under test are in different states;

[0035] An extraction unit is used to extract features of the original vibration signals sampled with a fixed sampling length from all original vibration signals, and obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be tested;

[0036] a first determining unit, configured to select different parameters from a preset parameter range as smoothing factors of a preset PNN model, and cross-validate the preset PNN model using the smoothing factors and the feature vector to obtain accuracy rates of the preset PNN model under different smoothing factors, and determine the smoothing factor corresponding to the highest accuracy rate as a target parameter;

[0037] A construction unit, configured to construct a gearbox fault detection model according to the preset PNN model and the target parameters;

[0038] The second determining unit is configured to determine the fault type of the gearbox to be tested by using the gearbox fault detection model.

[0039] In conjunction with the second aspect, in some optional implementations, the extraction unit is further configured to:

[0040] For each group of the original vibration signals sampled with a fixed sampling length, constructing a third-order tensor according to the original vibration signals through continuous wavelet transformation;

[0041] determining the number of principal components of the third-order tensor by a kernel consistency diagnostic method;

[0042] Decomposing the third-order tensor by parallel factor analysis to obtain a corresponding trilinear model according to the third-order tensor and the number of principal components;

[0043] Extracting a frequency signal and a time domain signal from the trilinear model, and determining a centroid frequency corresponding to the frequency signal according to the frequency signal, and determining a kurtosis index corresponding to the time domain signal according to the time domain signal;

[0044] The feature vector is constructed according to the center of gravity frequency and the kurtosis index.

[0045] In conjunction with the second aspect, in some optional implementations, the first determining unit is further configured to:

[0046] Initializing relevant parameters, including the number of vectors used to search for the target parameter from the preset parameter range, a maximum number of iterations, an initial position of the vector, and an initial global optimal position;

[0047] For each of the vectors, updating the initial position of the vector based on a first preset rule to obtain a vector with an updated position;

[0048] Merging the vector after the position update with the vector that meets the preset conditions to obtain a first new vector;

[0049] Bringing the parameter corresponding to the position of the first new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor;

[0050] Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the initial global optimal position, determining the position with the higher accuracy as the new optimal position;

[0051] According to the new optimal position, updating the first new vector based on a second preset rule to obtain a second new vector;

[0052] Bringing the parameter corresponding to the position of the second new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor;

[0053] Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the new optimal position, determining the position with the higher accuracy as the target optimal position;

[0054] Taking the target optimal position as the initial global optimal position, and updating the initial position of the vector based on the first preset rule to obtain the updated vector, repeating the above steps until the number of iterations is greater than or equal to the maximum number of iterations;

[0055] The smoothing factor corresponding to the target optimal position obtained in the last iteration is determined as the target parameter.

[0056] In conjunction with the second aspect, in some optional implementations, the device further includes:

[0057] a noise reduction unit, configured to perform noise reduction processing on the original vibration signal by wavelet packet transform to obtain a noise-reduced vibration signal;

[0058] The extraction unit is also used to:

[0059] The features of the noise-reduced vibration signals sampled with a fixed sampling length are extracted from all noise-reduced vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be measured.

[0060] In a third aspect, an embodiment of the present application further provides an electronic device, characterized in that the electronic device includes a processor and a memory coupled to each other, and a computer program is stored in the memory. When the computer program is executed by the processor, the electronic device executes the above method.

[0061] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium. When the computer program is run on a computer, the computer executes the above method.

[0062] The invention adopting the above technical solution has the following advantages:

[0063] In the technical solution provided by the present application, the original vibration signals of the gearbox to be tested are first obtained when the gears of the gearbox to be tested are in different states; then, the features of the original vibration signals sampled with a fixed sampling length are extracted from all the original vibration signals to obtain a feature vector corresponding to each sampling under the fixed sampling length, which characterizes the working condition of the gearbox to be tested; then, different parameters are selected from the preset parameter range as the smoothing factors of the preset PNN model, and the preset PNN model is cross-validated by the smoothing factors and the feature vectors to obtain the accuracy of the preset PNN model under different smoothing factors, and the corresponding smoothing factor with the highest accuracy is determined as the target parameter; finally, a gearbox fault detection model is constructed based on the preset PNN model and the target parameters, and the fault type of the gearbox to be tested is determined by the gearbox fault detection model. In this way, the problem of insufficient accuracy of the current gearbox fault detection method can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The present application may be further illustrated by the non-limiting embodiments provided in the accompanying drawings. It should be understood that the following drawings illustrate only certain embodiments of the present application and are therefore not to be construed as limiting the scope of the present application. It is understood that a person skilled in the art can derive other relevant drawings from these drawings without inventive effort.

[0065] Figure 1 A block diagram of an electronic device provided in an embodiment of the present application.

[0066] Figure 2 A schematic flow chart of a gearbox fault detection method provided in an embodiment of the present application.

[0067] Figure 3 This is a block diagram of a gearbox fault detection device provided in an embodiment of the present application.

[0068] Icon: 100 - electronic device; 101 - processor; 102 - memory; 200 - gearbox fault detection device; 210 - acquisition unit; 2210 - extraction unit; 230 - first determination unit; 240 - construction unit; 250 - second determination unit. DETAILED DESCRIPTION

[0069] The present application will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that similar or identical parts in the drawings or descriptions are numbered the same. Implementations not shown or described in the drawings are known to those of ordinary skill in the art. In the description of this application, the terms "first," "second," etc. are used solely to distinguish descriptions and are not to be construed as indicating or implying relative importance.

[0070] Please refer to Figure 1 In an embodiment of the present application, an electronic device 100 may include a processor 101 and a memory 102. The memory 102 stores a computer program, which, when executed by the processor 101, enables the electronic device 100 to perform corresponding steps in the following gearbox fault detection method.

[0071] In this embodiment, the electronic device 100 can be a personal computer, laptop, cloud server, etc. It is used to obtain the raw vibration signal of the gearbox under test and extract features from the raw vibration signal to obtain a feature vector. A preset PNN model is then cross-validated using various parameters within a preset parameter range as smoothing factors. The parameter with the highest accuracy in the cross-validation is determined as the target parameter. A gearbox fault detection model is then constructed based on the target parameter and the preset PNN model. Finally, the gearbox fault detection model is used to determine the fault type of the gearbox under test.

[0072] Please refer to Figure 2 The present application also provides a gearbox fault detection method, which can be applied to the electronic device 100 described above, and each step of the method is executed or implemented by the electronic device 100. The gearbox fault detection method can include the following steps:

[0073] Step 110, obtaining original vibration signals of the gearbox under test when the gears of the gearbox under test are in different states;

[0074] Step 120: extracting features of the original vibration signals sampled with a fixed sampling length from all original vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be tested;

[0075] Step 130: Select different parameters from a preset parameter range as smoothing factors of a preset PNN model, and cross-validate the preset PNN model using the smoothing factors and the feature vector to obtain the accuracy of the preset PNN model under different smoothing factors, and determine the smoothing factor corresponding to the highest accuracy as the target parameter;

[0076] Step 140: constructing a gearbox fault detection model based on the preset PNN model and the target parameters;

[0077] Step 150: Determine the fault type of the gearbox to be tested by using the gearbox fault detection model.

[0078] In the above-mentioned embodiment, the original vibration signals of the gearbox to be tested are first obtained when the gears of the gearbox to be tested are in different states; then, the features of the original vibration signals sampled with a fixed sampling length are extracted from all the original vibration signals to obtain a feature vector corresponding to each sampling under the fixed sampling length, which represents the working condition of the gearbox to be tested; then, different parameters are selected from the preset parameter range as the smoothing factors of the preset PNN model, and the preset PNN model is cross-validated using the smoothing factors and feature vectors to obtain the accuracy of the preset PNN model under different smoothing factors, and the corresponding smoothing factor with the highest accuracy is determined as the target parameter; then, a gearbox fault detection model is constructed based on the preset PNN model and the target parameters; finally, the fault type of the gearbox to be tested is determined using the gearbox fault detection model. In this way, the problem of insufficient accuracy in the current gearbox fault detection method can be improved.

[0079] The following will describe in detail the various steps of the gearbox fault detection method:

[0080] In step 110, the original vibration signal acquisition of this embodiment requires equipment such as the gearbox to be tested, a normal gear, four gears with different fault types, a dynamic simulator, an acceleration sensor, a signal analyzer, and an electronic device 100 (a laptop computer is used as an example in this embodiment). Specifically, the acceleration sensors are respectively installed at the vertical and horizontal positions of the corresponding gears of the gearbox housing to be tested, and each device is connected. Initially, a normal gear F1 is used, and the engine speed is set to 800 rpm. After the gearbox system is running stably, the gearbox dynamic simulator is used to collect the original vibration signal and send it to the laptop computer via the signal analyzer. After a set of data is collected (for example, the data volume reaches the user's ideal standard, or the collection time reaches a pre-set time), the speed is set to 1000 rpm, 1200 rpm, ..., 3450 rpm, and the above steps are repeated to collect data. After the original vibration signal of the normal gear F1 is collected, the four faulty gears F2, F3, F4, and F5 are replaced in sequence, and the above steps are repeated to collect the original vibration signal and send it to the laptop computer.

[0081] In this embodiment, the original vibration signal can be acquired based on the user's operation quality from a real-time experimental environment or the operating environment of the gearbox to be tested, and then sent to the electronic device 100. Alternatively, the acquired historical data (i.e., the original vibration signal) can be stored in the memory 102 of the electronic device 100, and based on subsequent user operation instructions, the data can be retrieved from the memory 102 to facilitate subsequent processing. The method for acquiring the original vibration signal is not specifically limited herein.

[0082] In step 120, extracting features of the original vibration signals sampled with a fixed sampling length from all original vibration signals to obtain a feature vector corresponding to each sampling at the fixed sampling length that characterizes the working condition of the gearbox to be measured may include:

[0083] For each group of the original vibration signals sampled with a fixed sampling length, constructing a third-order tensor according to the original vibration signals through continuous wavelet transformation;

[0084] determining the number of principal components of the third-order tensor by a kernel consistency diagnostic method;

[0085] Decomposing the third-order tensor by parallel factor analysis to obtain a corresponding trilinear model according to the third-order tensor and the number of principal components;

[0086] Extracting a frequency signal and a time domain signal from the trilinear model, and determining a centroid frequency corresponding to the frequency signal according to the frequency signal, and determining a kurtosis index corresponding to the time domain signal according to the time domain signal;

[0087] The feature vector is constructed according to the center of gravity frequency and the kurtosis index.

[0088] In this embodiment, the original vibration signal is passed through a wavelet basis function (cmor3-3 in this embodiment) with a fixed scale factor, and then compared with the original vibration signal. By changing the translation parameter and the scale parameter, the original vibration signal is transformed into a third-order tensor of 2*126*4096. Here, 2 represents two channels (low frequency and high frequency, respectively), 126 represents the scale of the continuous wavelet transform, and 4096 represents the length of the signal.

[0089] Then the main city fraction of the third-order tensor is determined by the kernel consistency diagnostic method. The kernel consistency value expression is as follows:

[0090]

[0091] Where ξ represents the kernel consistency value, t def represents the elements in the superdiagonal matrix T in the trilinear model, g def Represents the elements of the core three-dimensional matrix G in the trilinear model.

[0092] It is understandable that in practical applications, when the kernel consistency value is greater than or equal to 60%, the third-order tensor conforms to the trilinear model, and when the kernel consistency value is less than 60%, the third-order tensor deviates from the trilinear model. In this embodiment, the number of principal components F can be ultimately determined to be 4 based on the change in the kernel consistency value ξ.

[0093] Then, the third-order tensor is decomposed by parallel factor analysis to obtain the following trilinear model:

[0094]

[0095] Where, represents the data value in the trilinear model, a if represents the value of the fth component related to sample information i, b jf represents the response value of the jth measurement time point in the fth component, c kf represents the value associated with the kth sensor in the fth component, e ijk Represents the error between the actual observed value and the model's predicted value.

[0096] In this embodiment, first initialize the I×F matrix A=[a1, a2…a I ], the J×F matrix B=[b1,b2…b j ], K×F matrix C=[c1,c2…c k]. Then, the matrix A, matrix B and matrix C are iteratively calculated by the least squares optimization algorithm, and during the iteration process, the sum of the residual squares of the difference between the observed data and the model prediction value is less than a preset value (in this embodiment, the preset value is 1×10 -6 ), extract the resulting matrix A as the spatial signal, matrix B as the spectral signal, and matrix C as the time-domain signal. Because the fault characteristics represented by the spectral and time-domain signals differ significantly, the centroid frequency of the spectral signal and the kurtosis index of the time-domain signal are calculated to construct the eigenvector. Given a principal component of 4, four centroid frequencies and four kurtosis branches are calculated for each data set, meaning each data set corresponds to eight features, from which the eigenvector is constructed.

[0097] In this embodiment, the residual sum of squares is calculated as follows:

[0098]

[0099] Where SSR represents the residual sum of squares, x ijk Represents the data values ​​in the trilinear model.

[0100] In step 130, different parameters are selected from a preset parameter range as smoothing factors of a preset PNN model, and the preset PNN model is cross-validated using the smoothing factors and the feature vector to obtain the accuracy of the preset PNN model under different smoothing factors, and the smoothing factor corresponding to the highest accuracy is determined as the target parameter, which may include:

[0101] Initializing relevant parameters, including the number of vectors used to search for the target parameter from the preset parameter range, a maximum number of iterations, an initial position of the vector, and an initial global optimal position;

[0102] For each of the vectors, updating the initial position of the vector based on a first preset rule to obtain a vector with an updated position;

[0103] Merging the vector after the position update with the vector that meets the preset conditions to obtain a first new vector;

[0104] Bringing the parameter corresponding to the position of the first new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor;

[0105] Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the initial global optimal position, determining the position with the higher accuracy as the new optimal position;

[0106] According to the new optimal position, updating the first new vector based on a second preset rule to obtain a second new vector;

[0107] Bringing the parameter corresponding to the position of the second new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor;

[0108] Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the new optimal position, determining the position with the higher accuracy as the target optimal position;

[0109] Taking the target optimal position as the initial global optimal position, and updating the initial position of the vector based on the first preset rule to obtain the updated vector, repeating the above steps until the number of iterations is greater than or equal to the maximum number of iterations;

[0110] The smoothing factor corresponding to the target optimal position obtained in the last iteration is determined as the target parameter.

[0111] In this embodiment, by selecting a target parameter from a preset parameter range as a smoothing factor of a preset PNN model, the parameter tuning of the model is achieved, which can improve the accuracy and stability of the model in fault classification.

[0112] Specifically, first, the number of vectors used to search for target parameters, the maximum number of iterations in the iterative search process, the initial position of the vectors, and the initial global optimal position are initialized;

[0113] For each vector, the initial position of the vector is updated based on the following first preset rule to obtain the updated vector and

[0114]

[0115]

[0116] Where, and represents the updated vector of the position obtained in the g-th iteration, f(·) represents the fitness function, x bs represents the optimal solution of all vectors in the g-th iteration population (i.e. within the preset parameter range), x bt represents the optimal solution of all vectors in the g-th iteration population, σ is the scaling factor of the vector, σ=2α×rand-α, where α can be calculated based on Updated with the number of iterations, Maxg represents the maximum number of iterations, a1≠a2≠a3≠l is a different integer randomly selected from [1,NP], and denote the a1-th, a2-th, a3-th, b-th, and l-th dimension vectors of the g-th iteration, respectively. Where randn∈{rand1,rand2,…,randn} is a standard normal distribution random value, and MeanRule is defined as follows:

[0117]

[0118] Where r is a random number between [0,0.5], and The definition is as follows:

[0119]

[0120] l=1,2,…,NP (8)

[0121] in:

[0122]

[0123]

[0124]

[0125] w a =max(f(x a1 ),f(x a2 ),f(x a3 )) (12)

[0126]

[0127] l=1,2,…,NP (14)

[0128] in:

[0129]

[0130]

[0131]

[0132] w b =f(x ws ) (18)

[0133] Where w1, w2, w3, w4, w5 and w6 are weighted functions used to calculate the weighted average of the vector, which is beneficial to the global search of the vector within the preset parameter range. a2 and xa3 are the a2th and a3th dimensional vectors obtained in the gth iteration, x ws It represents the worst parameter solution vector (i.e., the smallest fitness function value) in the g-th iteration population, and ε is a small constant.

[0134] In the above formula (7) and formula (13), δ is the weighted average factor, δ = 2β × rand - β, That is, δ = 2α × rand - α. In this way, the weighted average factor δ is updated, and the updated vector is calculated based on the above formula (4) and formula (5) and

[0135] After the updated vector is calculated, the updated vector is compared with the vector that meets the preset condition rand < 0.5. Merge to generate the first new vector as follows:

[0136]

[0137] Where, represents the first new vector obtained in the vector merging phase in the g-th iteration, rand1 and rand2 are the random numbers generated after the first and second updated vectors, respectively, and μ = 0.05 × randn.

[0138] After determining the first new vector, the first new vector is used as a smoothing factor for the preset PNN model, and the aforementioned original feature vector is used as a data set to cross-validate the preset PNN model. The accuracy of the preset PNN model after validation is determined (which can be understood as using the accuracy of the preset PNN model as a fitness function for parameter tuning). Since cross-validation and determining model accuracy through cross-validation are conventional technical means in the field of computer processing technology, they are not further described here.

[0139] Then, the preset PNN model calculates the accuracy corresponding to the first new vector in this iteration process, compares it with the accuracy obtained by cross-validation of the parameters represented by the global optimal position, and takes the vector position corresponding to the one with higher accuracy as the new optimal position.

[0140] Then, the first new vector is updated based on the following second preset rule, that is, when rand < 0.5, the optimal solution x in all vectors is updated. bs , that is, the new optimal position generates the second new vector

[0141]

[0142] x rnd=φ×x avg +(1-φ)×(φ×x bt +(1-φ)×x bs ) (twenty one)

[0143]

[0144] Where x a 、x b 、x c is the loading vector of the latent factor, which is used to explain the structure and pattern in the observed data. avg is the average value of the three load vectors, used to ensure the balance of the second new vector during the search process. v1 and v2 are two random numbers. p represents a random number in the range (0,1).

[0145] In the above formula, φ represents a random number in the range of (0,1), and the random combination x avg 、x bt and x bs , and get the new solution x rnd , so that the optimal solution x is obtained through the above formula (20) bs Generate a second new vector near

[0146] After determining the second new vector, the second new vector is used as a smoothing factor of the preset PNN model, the original feature vector is used as a data set, the preset PNN model is cross-validated, and the accuracy of the preset PNN model after verification is determined.

[0147] Then, the preset PNN model is used to calculate the accuracy of the second new vector in this iteration process, and the accuracy obtained by cross-validation of the parameters represented by the global optimal position is compared with the accuracy, and the vector position corresponding to the one with higher accuracy is used as the target optimal position.

[0148] This target optimal position is then used as the initial global optimal position. Starting from the initial position of the vector updated based on the first preset rule, the above steps are repeated until the number of iterations is greater than or equal to the maximum number of iterations. The smoothing factor corresponding to the target optimal position obtained in the last iteration (i.e., the parameter within the preset parameter range corresponding to the target optimal position in the last iteration) is used as the target parameter to complete the tuning of the smoothing factor.

[0149] In step 140, after the smoothing factor is tuned, the target parameter is used as the smoothing factor of the preset PNN model and brought into the preset PNN model for training. Specifically, the model training process is as follows:

[0150] The corresponding labels were added to the aforementioned feature vectors. The five feature vectors were labelled 0 for normal gears, label 1 for crack levels of 25%, label 2 for crack levels of 50%, label 3 for crack levels of 75%, and label 4 for crack levels of 100%. The labeled feature vectors were then divided into training and test sets in a ratio of 7:3. The training and test sets were then fed into the input layer of a pre-set PNN model, whose number of neurons was equal to the length of the input feature vector.

[0151] The hidden layer of the preset PNN model receives the samples of the input layer (i.e. the samples in the training set or test set mentioned above) and calculates the distance between the input vector and the center, and finally returns a scalar value. The number of neurons is the same as the number of input samples. The jth neuron of the i-th pattern in the hidden layer is given by the formula Definition. Where i = 1, 2, ..., m, m represents the total number of sample categories, d represents the dimension of the sample space data, x represents the input, x represents the ij represents the jth center of the i-th class sample, and λ represents the smoothing factor of the preset PNN model, which is usually between 0 and 1 and plays a crucial role in model performance. In this embodiment, the selection of the smoothing factor refers to the above step 130.

[0152] The summation layer then takes a weighted average of the outputs of neurons in the hidden layer that belong to the same class: Among them, v i Represents the output of the i-th category, and L represents the number of neurons in the i-th category.

[0153] Finally, the output layer takes the largest one in the sum layer as the final output category: y = argmax (v i ).

[0154] In the above training process, the number of nodes in the input layer and hidden layer of the preset PNN model are 8 and 140 respectively, and the number of nodes in the summation layer is 5, the number of gearbox fault modes to be tested.

[0155] The aforementioned target parameters are brought into the trained preset PNN model to complete the construction of the gearbox fault detection model.

[0156] In step 150, the measured raw vibration signal is subjected to feature extraction to obtain a feature vector. This feature vector is then input into the gearbox fault detection model constructed in steps 130 and 140, ultimately outputting the gearbox fault type. This completes the diagnosis and monitoring of the gearbox fault type being tested.

[0157] As an optional implementation, between step 110 and step 120, the method may further include:

[0158] Performing noise reduction processing on the original vibration signal by wavelet packet transform to obtain a noise-reduced vibration signal;

[0159] Then step 120 may include:

[0160] The features of the noise-reduced vibration signals sampled with a fixed sampling length are extracted from all noise-reduced vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be measured.

[0161] In this example, a three-layer wavelet packet decomposition operation is performed on the original vibration signal to obtain signals with eight mutually orthogonal frequency bands. This analysis then identifies the optimal wavelet packet basis. The optimal characteristic information for each frequency band is then obtained, and a threshold value is determined. Finally, the decomposed and processed signal is reconstructed using wavelet packets, selecting appropriate frequency signals to obtain a de-noised vibration signal.

[0162] In this way, by performing noise reduction processing on the original signal through wavelet packet transform, both the high-frequency and low-frequency parts of the signal can be decomposed at the same time, thereby improving the time-frequency resolution of the signal, reducing the amplitude of the signal after noise reduction, and making the signal curve clearer.

[0163] Please refer to Figure 3 The present application also provides a gearbox fault detection device 200, which includes at least one software function module that can be stored in the form of software or firmware in the memory 102 or embedded in the operating system (OS) of the electronic device 100. The processor 101 is configured to execute the executable modules stored in the memory 102, such as the software function modules and computer programs included in the gearbox fault detection device 200.

[0164] The gearbox fault detection device 200 includes an acquisition unit 210, an extraction unit 220, a first determination unit 230, a construction unit 240, and a second determination unit 250. The functions of each unit may be as follows:

[0165] An acquisition unit 210 is configured to acquire original vibration signals of the gearbox under test when the gears of the gearbox under test are in different states;

[0166] An extraction unit 220 is configured to extract features of the original vibration signals sampled with a fixed sampling length from all original vibration signals, and obtain a feature vector corresponding to each sampling with the fixed sampling length that represents the working condition of the gearbox to be tested;

[0167] A first determining unit 230 is configured to select different parameters from a preset parameter range as smoothing factors of a preset PNN model, and cross-validate the preset PNN model using the smoothing factors and the feature vector to obtain accuracy rates of the preset PNN model under different smoothing factors, and determine the smoothing factor corresponding to the highest accuracy rate as a target parameter;

[0168] A construction unit 240 is configured to construct a gearbox fault detection model based on the preset PNN model and the target parameters;

[0169] The second determining unit 250 is configured to determine the fault type of the gearbox to be tested by using the gearbox fault detection model.

[0170] Optionally, the extraction unit 220 is further configured to:

[0171] For each group of the original vibration signals sampled with a fixed sampling length, constructing a third-order tensor according to the original vibration signals through continuous wavelet transformation;

[0172] determining the number of principal components of the third-order tensor by a kernel consistency diagnostic method;

[0173] Decomposing the third-order tensor by parallel factor analysis to obtain a corresponding trilinear model according to the third-order tensor and the number of principal components;

[0174] Extracting a frequency signal and a time domain signal from the trilinear model, and determining a centroid frequency corresponding to the frequency signal according to the frequency signal, and determining a kurtosis index corresponding to the time domain signal according to the time domain signal;

[0175] The feature vector is constructed according to the center of gravity frequency and the kurtosis index.

[0176] Optionally, the first determining unit 230 is further configured to:

[0177] Initializing relevant parameters, including the number of vectors used to search for the target parameter from the preset parameter range, a maximum number of iterations, an initial position of the vector, and an initial global optimal position;

[0178] For each of the vectors, updating the initial position of the vector based on a first preset rule to obtain a vector with an updated position;

[0179] Merging the vector after the position update with the vector that meets the preset conditions to obtain a first new vector;

[0180] Bringing the parameter corresponding to the position of the first new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor;

[0181] Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the initial global optimal position, determining the position with the higher accuracy as the new optimal position;

[0182] According to the new optimal position, updating the first new vector based on a second preset rule to obtain a second new vector;

[0183] Bringing the parameter corresponding to the position of the second new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor;

[0184] Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the new optimal position, determining the position with the higher accuracy as the target optimal position;

[0185] Taking the target optimal position as the initial global optimal position, and updating the initial position of the vector based on the first preset rule to obtain the updated vector, repeating the above steps until the number of iterations is greater than or equal to the maximum number of iterations;

[0186] The smoothing factor corresponding to the target optimal position obtained in the last iteration is determined as the target parameter.

[0187] Optionally, the device further includes:

[0188] a noise reduction unit, configured to perform noise reduction processing on the original vibration signal by wavelet packet transform to obtain a noise-reduced vibration signal;

[0189] The extraction unit 220 is further configured to:

[0190] The features of the noise-reduced vibration signals sampled with a fixed sampling length are extracted from all noise-reduced vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be measured.

[0191] In this embodiment, the processor 101 may be an integrated circuit chip having signal processing capabilities. The processor 101 may be a general-purpose processor. For example, the processor 101 may be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components, and may implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present application.

[0192] The memory 102 may be, but is not limited to, a random access memory, a read-only memory, a programmable read-only memory, an erasable programmable read-only memory, an electrically erasable programmable read-only memory, etc. In this embodiment, the memory 102 may be used to store raw vibration signals, feature vectors, a preset PNN model, the smoothing factor, accuracy, target parameters of the preset PNN model, a gearbox fault detection model, etc. Of course, the memory 102 may also be used to store programs, which the processor 101 executes upon receiving an execution instruction.

[0193] It is understandable that Figure 1 The structure of the electronic device 100 shown in FIG is only a schematic diagram of a structure. The electronic device 100 may also include Figure 1 More components shown. Figure 1 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0194] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the electronic device 100 described above can refer to the corresponding processes of each step in the aforementioned method, and will not be elaborated here.

[0195] The present application also provides a computer-readable storage medium that stores a computer program, which, when executed on a computer, causes the computer to execute the gearbox fault detection method described in the above embodiment.

[0196] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented through hardware or by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0197] In summary, the embodiments of the present application provide a gearbox fault detection method, device, electronic device, and storage medium. In this technical solution, the original vibration signals of the gearbox to be tested are first obtained when the gears of the gearbox to be tested are in different states; then, the features of the original vibration signals sampled with a fixed sampling length are extracted from all the original vibration signals to obtain a feature vector corresponding to each sampling under the fixed sampling length that characterizes the working condition of the gearbox to be tested; then, different parameters are selected from the preset parameter range as the smoothing factors of the preset PNN model, and the preset PNN model is cross-validated by the smoothing factors and the feature vector to obtain the accuracy of the preset PNN model under different smoothing factors, and the corresponding smoothing factor with the highest accuracy is determined as the target parameter; then, a gearbox fault detection model is constructed based on the preset PNN model and the target parameters; finally, the fault type of the gearbox to be tested is determined by the gearbox fault detection model. In this way, the problem of insufficient accuracy of the current gearbox fault detection method can be improved.

[0198] In the embodiments provided in the present application, it should be understood that the disclosed devices, systems and methods can also be implemented in other ways. The device, system and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of code, and a part of the module, program segment or code includes one or more executable instructions for implementing the specified logical function. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. In addition, the functional modules in the various embodiments of the present application can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0199] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A gearbox fault detection method, characterized in that: The method comprises: Acquire original vibration signals of the gearbox under test when the gears of the gearbox under test are in different states; Extracting features of the original vibration signals sampled with a fixed sampling length from all original vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be tested; Selecting different parameters from a preset parameter range as smoothing factors of a preset PNN model, and cross-validating the preset PNN model using the smoothing factors and the feature vector to obtain the accuracy of the preset PNN model under different smoothing factors, and determining the smoothing factor corresponding to the highest accuracy as the target parameter; Constructing a gearbox fault detection model according to the preset PNN model and the target parameters; Determining the fault type of the gearbox to be tested by using the gearbox fault detection model; Extracting features of the original vibration signals sampled with a fixed sampling length from all original vibration signals to obtain a feature vector corresponding to each sampling at the fixed sampling length that characterizes the working condition of the gearbox to be tested, including: For each group of the original vibration signals sampled with a fixed sampling length, constructing a third-order tensor according to the original vibration signals through continuous wavelet transformation; determining the number of principal components of the third-order tensor by a kernel consistency diagnostic method; Decomposing the third-order tensor by parallel factor analysis to obtain a corresponding trilinear model according to the third-order tensor and the number of principal components; Extracting a frequency signal and a time domain signal from the trilinear model, and determining a centroid frequency corresponding to the frequency signal according to the frequency signal, and determining a kurtosis index corresponding to the time domain signal according to the time domain signal; constructing the feature vector according to the center of gravity frequency and the kurtosis index; Selecting different parameters as smoothing factors of a preset PNN model, cross-validating the preset PNN model using the smoothing factors and the feature vector, obtaining the accuracy of the preset PNN model under different smoothing factors, and determining the smoothing factor corresponding to the highest accuracy as the target parameter, including: Initializing relevant parameters, including the number of vectors used to search for the target parameter from the preset parameter range, a maximum number of iterations, an initial position of the vector, and an initial global optimal position; For each of the vectors, updating the initial position of the vector based on a first preset rule to obtain a vector with an updated position; Merging the vector after the position update with the vector that meets the preset conditions to obtain a first new vector; Bringing the parameter corresponding to the position of the first new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor; Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the initial global optimal position, determining the position with the higher accuracy as the new optimal position; According to the new optimal position, updating the first new vector based on a second preset rule to obtain a second new vector; Bringing the parameter corresponding to the position of the second new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor; Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the new optimal position, determining the position with the higher accuracy as the target optimal position; Taking the target optimal position as the initial global optimal position, and updating the initial position of the vector based on the first preset rule to obtain the updated vector, repeating the above steps until the number of iterations is greater than or equal to the maximum number of iterations; The smoothing factor corresponding to the target optimal position obtained in the last iteration is determined as the target parameter.

2. The method according to claim 1, characterized in that The method further comprises: Performing noise reduction processing on the original vibration signal by wavelet packet transform to obtain a noise-reduced vibration signal; The features of the original vibration signals sampled with a fixed sampling length are extracted from all the original vibration signals to obtain a feature vector corresponding to each sampling under the fixed sampling length, which characterizes the working condition of the gearbox to be tested, including: The features of the noise-reduced vibration signals sampled with a fixed sampling length are extracted from all noise-reduced vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be measured.

3. A gearbox fault detection device, characterized in that: For implementing the method according to claim 1 or 2, the device comprises: an acquisition unit, configured to acquire original vibration signals of the gearbox under test when the gears of the gearbox under test are in different states; An extraction unit is used to extract features of the original vibration signals sampled with a fixed sampling length from all original vibration signals, and obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be tested; a first determining unit, configured to select different parameters from a preset parameter range as smoothing factors of a preset PNN model, and cross-validate the preset PNN model using the smoothing factors and the feature vector to obtain accuracy rates of the preset PNN model under different smoothing factors, and determine the smoothing factor corresponding to the highest accuracy rate as a target parameter; A construction unit, configured to construct a gearbox fault detection model according to the preset PNN model and the target parameters; The second determining unit is configured to determine the fault type of the gearbox to be tested by using the gearbox fault detection model.

4. The device according to claim 3, characterized in that The extraction unit is further configured to: For each group of the original vibration signals sampled with a fixed sampling length, constructing a third-order tensor according to the original vibration signals through continuous wavelet transformation; determining the number of principal components of the third-order tensor by a kernel consistency diagnostic method; Decomposing the third-order tensor by parallel factor analysis to obtain a corresponding trilinear model according to the third-order tensor and the number of principal components; Extracting a frequency signal and a time domain signal from the trilinear model, and determining a centroid frequency corresponding to the frequency signal according to the frequency signal, and determining a kurtosis index corresponding to the time domain signal according to the time domain signal; The feature vector is constructed according to the center of gravity frequency and the kurtosis index.

5. The method according to claim 3, characterized in that The first determining unit is further configured to: Initializing relevant parameters, including the number of vectors used to search for the target parameter from the preset parameter range, a maximum number of iterations, an initial position of the vector, and an initial global optimal position; For each of the vectors, updating the initial position of the vector based on a first preset rule to obtain a vector with an updated position; Merging the vector after the position update with the vector that meets the preset conditions to obtain a first new vector; Bringing the parameter corresponding to the position of the first new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor; Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the initial global optimal position, determining the position with the higher accuracy as the new optimal position; According to the new optimal position, updating the first new vector based on a second preset rule to obtain a second new vector; Bringing the parameter corresponding to the position of the second new vector as the smoothing factor into the preset PNN model for cross-validation, and determining the accuracy of the preset PNN model under the current smoothing factor; Comparing the accuracy of the preset PNN model under the current smoothing factor with the accuracy corresponding to the new optimal position, determining the position with the higher accuracy as the target optimal position; Taking the target optimal position as the initial global optimal position, and updating the initial position of the vector based on the first preset rule to obtain the updated vector, repeating the above steps until the number of iterations is greater than or equal to the maximum number of iterations; The smoothing factor corresponding to the target optimal position obtained in the last iteration is determined as the target parameter.

6. The device according to claim 3, characterized in that The device further comprises: a noise reduction unit, configured to perform noise reduction processing on the original vibration signal by wavelet packet transform to obtain a noise-reduced vibration signal; The extraction unit is also used to: The features of the noise-reduced vibration signals sampled with a fixed sampling length are extracted from all noise-reduced vibration signals to obtain a feature vector corresponding to each sampling with the fixed sampling length that characterizes the working condition of the gearbox to be measured.

7. An electronic device, characterized in that: The electronic device includes a processor and a memory coupled to each other, wherein the memory stores a computer program. When the computer program is executed by the processor, the electronic device executes the method according to claim 1 or 2.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed on a computer, enables the computer to execute the method according to claim 1 or 2.

Images