Engine fault diagnosis method and device based on wavelet neural network, computer equipment and medium

Abstract

The embodiment of the invention provides an engine fault diagnosis method and device based on a wavelet neural network, computer equipment and a medium, and the method comprises the following steps: constructing the wavelet neural network comprising an input layer, a hidden layer and an output layer, and determining the number of nodes of the input layer and the number of nodes of the output layer; performing normalization processing on a data set comprising a training set sample and a test set sample, inputting the normalized training set sample into a wavelet neural network, training the wavelet neural network by using an L-M algorithm, determining an optimal hidden layer node number p according to a node number of an input layer and a node number of an output layer, and generating a trained wavelet neural network; and inputting the normalized test set sample into the trained wavelet neural network to generate a diagnosis result of the engine fault. According to the scheme, the fault is diagnosed through the wavelet neural network, and the accuracy and efficiency of diagnosis of the rotor-rotor / rotor-static rub-impact fault of the aero-engine are improved.

Description

Technical Field

[0001] This invention relates to the field of aero-engine fault diagnosis technology, and in particular to a method, apparatus, computer equipment, and medium for diagnosing engine faults based on wavelet neural networks. Background Technology

[0002] As a high-precision aero-thermodynamic rotating machine, modern aero-engines have undergone continuous refinement to achieve high thrust-to-weight ratios, high efficiency, and low fuel consumption. One crucial measure taken in this process is reducing the design clearances between the high-pressure and low-pressure rotors, and between the rotor and stators (rotor and brush seals, blades and casing, etc.). This significantly increases the probability of rotor-to-rotor and rotor-to-stationary contact failures, making the engine highly susceptible to problems such as increased overall vibration and blade breakage, seriously threatening the operational safety and reliability of aero-engines.

[0003] In the field of aero-engine health management, statistics show that over 70% of engine mechanical failures can be manifested through vibration. The amplitude and frequency information contained in the casing vibration signal can directly reflect the working status of the engine rotation and transmission system. By identifying and diagnosing early abnormal vibration signals of aero-engines, the location and severity of engine failures can be determined in a timely manner, thereby preventing further deterioration of the failure and reducing potential risks and losses.

[0004] For the intelligent diagnosis of rotor-to-rotor and rotor-to-static rubbing faults in aero-engines, existing fault diagnosis methods mainly include: rubbing dynamics model, which mainly studies the complex vibration characteristics of various rubbing processes in aero-engines through theoretical modeling and simulation; rubbing numerical simulation, which mainly analyzes the transient process and dynamic characteristics of rubbing faults by establishing a rotor-to-rotor / rotor-to-static (e.g., blade-casing) finite element model of the aero-engine; and rotor-to-rotor / rotor-to-static rubbing fault bench simulation and whole-engine experiment, which mainly relies on bench simulation experiments and real test runs to acquire casing vibration data during engine failure through sensors, and performs frequency domain analysis based on methods such as Fast Fourier Transform (FFT) to obtain the typical diagnostic frequency characteristics of rotor-to-rotor / rotor-to-static rubbing faults in aero-engines.

[0005] However, the aforementioned fault diagnosis methods have two significant drawbacks: Firstly, these methods primarily rely on the accurate identification of vibration frequency characteristics of aero-engine rubbing faults and depend on expert experience to determine the fault type and severity of the engine system. Furthermore, the method's fault diagnosis efficiency is low and cannot meet the demands of rapid fault diagnosis for modern aero-engines. Secondly, due to the complex internal mechanical structure of aero-engines and their variable operating conditions, the vibration frequency of a non-faulty component within the engine system may coincide with the vibration characteristics of a rubbing fault, potentially leading to misdiagnosis of engine faults. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a method for diagnosing engine faults based on wavelet neural networks, to solve the technical problems of low fault diagnosis efficiency and susceptibility to misdiagnosis in the prior art. The method includes: A wavelet neural network comprising an input layer, a hidden layer, and an output layer is constructed, and the number of nodes in the input layer and the number of nodes in the output layer are determined. The input layer is used to receive vibration input signals generated by the engine, the hidden layer is used to perform nonlinear transformation on the vibration input signals, and the output layer is used to output the diagnostic results of engine faults. The dataset, including training and test samples, is normalized. The normalized training samples are then input into the wavelet neural network. The wavelet neural network is trained using the LM algorithm. The optimal number of hidden layer nodes is determined based on the number of nodes in the input layer and the number of nodes in the output layer. p The optimal number of hidden layer nodes p The number of nodes in the hidden layer is used to generate the trained wavelet neural network; The normalized test set samples are input into the trained wavelet neural network to generate engine fault diagnosis results. The types of diagnosis results include normal, rotary-rotational contact fault, and rotary-static contact fault.

[0007] This invention also provides a diagnostic device for engine faults based on wavelet neural networks, to address the technical problems of low fault diagnosis efficiency and susceptibility to misdiagnosis in existing technologies. The device includes: A wavelet neural network module is constructed to build a wavelet neural network including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer and the number of nodes in the output layer are determined. The input layer is used to receive the vibration input signal generated by the engine, the hidden layer is used to perform nonlinear transformation on the vibration input signal, and the output layer is used to output the diagnostic results of engine faults. The wavelet neural network training module is used to normalize the dataset, which includes training and test samples. The normalized training samples are then input into the wavelet neural network, which is trained using the LM algorithm. Finally, the optimal number of hidden layer nodes is determined based on the number of nodes in the input and output layers. p The optimal number of hidden layer nodes p The number of nodes in the hidden layer is used to generate the trained wavelet neural network; The fault diagnosis module is used to input the normalized test set samples into the trained wavelet neural network to generate engine fault diagnosis results. The types of diagnosis results include normal, rotary-rotational contact fault, and rotary-static contact fault.

[0008] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned wavelet neural network-based engine fault diagnosis methods, thereby solving the technical problems of low fault diagnosis efficiency and easy misdiagnosis in the prior art.

[0009] This invention also provides a computer-readable storage medium storing a computer program that executes any of the above-described wavelet neural network-based engine fault diagnosis methods, in order to solve the technical problems of low fault diagnosis efficiency and easy misdiagnosis in the prior art.

[0010] Compared with the prior art, the beneficial effects that at least one technical solution adopted in the embodiments of this specification can achieve include at least: By using wavelet neural networks to diagnose faults, the accuracy and efficiency of diagnosing rotor-to-rotor / rotor-to-static rubbing faults in aero-engines have been improved. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of an engine fault diagnosis method based on wavelet neural network provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the wavelet neural network model provided in an embodiment of the present invention; Figure 3 This is a Morlet wavelet basis function image provided in an embodiment of the present invention; Figure 4 This is a linear Purelin activation function image provided in an embodiment of the present invention; Figure 5 This is a flowchart of a rotation-rotation / rotation-static collision fault diagnosis based on a wavelet neural network provided in an embodiment of the present invention; Figure 6 This is a comparison chart of the output errors of the wavelet neural network with different numbers of hidden layer nodes provided in the embodiments of the present invention; Figure 7 This is a comparison chart of the training set prediction results of the wavelet neural network with different numbers of hidden layer nodes provided in the embodiments of the present invention; Figure 8 This is a schematic diagram of the optimal network structure for wavelet neural network rotation-rotation / rotation-static collision fault diagnosis provided in an embodiment of the present invention; Figure 9 This is a comparison chart and confusion matrix of the training set prediction results for the optimal number of hidden layer nodes (12) provided in this embodiment of the invention; Figure 10 This is a comparison chart and confusion matrix of the test set prediction results for the optimal number of hidden layer nodes (12) in this embodiment of the invention.

[0013] Figure 11 This is a structural block diagram of a computer device provided in an embodiment of the present invention; Figure 12 This is a structural block diagram of an engine fault diagnosis method and device based on wavelet neural network provided in an embodiment of the present invention. Detailed Implementation

[0014] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0015] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0016] In this embodiment of the invention, a method for diagnosing engine faults based on wavelet neural networks is provided, such as... Figure 1 As shown, the method includes: Step S101: Construct a wavelet neural network including an input layer, a hidden layer, and an output layer, and determine the number of nodes in the input layer and the number of nodes in the output layer. The input layer is used to receive the vibration input signal generated by the engine, the hidden layer is used to perform nonlinear transformation on the vibration input signal, and the output layer is used to output the diagnostic results of engine faults. Step S102: Normalize the dataset including training set samples and test set samples. Input the normalized training set samples into the wavelet neural network, train the wavelet neural network using the LM algorithm, and determine the optimal number of hidden layer nodes based on the number of nodes in the input layer and the number of nodes in the output layer. p The optimal number of hidden layer nodes p The number of nodes in the hidden layer is used to generate the trained wavelet neural network; Step S103: Input the normalized test set samples into the trained wavelet neural network to generate engine fault diagnosis results. The types of diagnosis results include normal, rotary-rotational contact fault, and rotary-static contact fault.

[0017] In practice, the wavelet neural network consisting of an input layer, a hidden layer, and an output layer is constructed through the following steps: The linear Purelin function is used as the activation function of the output layer; the Morlet wavelet basis function is used as the activation function of each node in the hidden layer, wherein the Morlet wavelet basis function... , x The horizontal axis represents the time variable. e It is a natural constant.

[0018] In specific implementation, the number of nodes in the input layer and the number of nodes in the output layer are determined through the following steps: The vibration signal time-domain features are determined, including peak factor, kurtosis factor, impulse factor, waveform factor, and margin factor; the frequency-domain features are determined, including mean square frequency and frequency standard deviation; the sum of the number of vibration signal time-domain features and the number of vibration signal frequency-domain features is set as the number of nodes in the wavelet neural network input layer, and the solution formulas for the vibration signal time-domain features and the vibration signal frequency-domain features are constructed respectively; the number of diagnostic result types is set as the number of nodes in the output layer.

[0019] The specific implementation of constructing the solution formulas for the time-domain features and frequency-domain features of the vibration signal is achieved through the following steps: Peak factor ,in ，X max The peak value of the vibration signal ，x ( n() represents the discrete form of the time-domain vibration signal. n The index of the discrete point. N The length of the discrete signal, RMS x The root mean square value of the signal; kurtosis factor. ,in, β The kurtosis of the signal, σ x The standard deviation of the signal. The mean of the signal; impulse factor. ,in, The average amplitude of the signal; waveform factor ; margin factor ,in, X r The root square amplitude of the signal; the mean square frequency. ,in, k For frequency component index, Δ f For frequency resolution, P [ k ] is the first k Power of each frequency component; frequency standard deviation ,in, The center frequency.

[0020] In specific implementation, the optimal number of hidden layer nodes is determined based on the number of nodes in the input layer and the number of nodes in the output layer through the following steps. p : The range of hidden layer node counts is determined based on the number of nodes in the input layer and the number of nodes in the output layer; the number of training sample groups is set; within the range of the hidden layer node count, the normalized training set sample data of the set number of groups are sequentially input into the wavelet neural networks with different hidden layer node counts to obtain the output error value. e The prediction results of the neural network; based on the error value e The optimal number of hidden layer nodes is determined by the prediction results of the training set of the neural network. p。

[0021] In specific implementation, the following steps are used to achieve the effect based on the error value. e The optimal number of hidden layer nodes is determined by the prediction results of the training set of the neural network. p : The prediction results of the training set of the neural network are cross-compared with the actual results to generate a confusion matrix; the accuracy of the prediction results is determined based on the confusion matrix; and the error value is... e Using the minimum and the correctness of the prediction result as conditions, determine whether the current number of hidden layer nodes is the optimal number of hidden layer nodes. p .

[0022] In specific implementation, the wavelet neural network is trained using the LM algorithm through the following steps: Define the weight update formula ,in, W Let be a vector composed of all values ​​and biases of the wavelet neural network. J Let be the Jacobian matrix of the differential of the error with respect to the weights. μ For variable parameter factors, I The identity matrix is ​​used; the weights are updated based on the update formula until the error output of the wavelet neural network is updated. e The requirements are met.

[0023] The wavelet neural network model structure is as follows: Figure 2 As shown, a wavelet neural network is a three-layer network consisting of an input layer, a hidden layer, and an output layer. The input layer receives external vibration input signals; the hidden layer performs nonlinear transformations on the vibration input signals and is the core of the wavelet neural network; the output layer outputs the engine fault diagnosis results.

[0024] The following three steps will detail the design of each layer of the wavelet neural network: Step 1: Design the input layer of the wavelet neural network.

[0025] For the design of the input layer of a wavelet neural network, since a single statistical index of vibration signal fault characteristics cannot simultaneously cover the typical time-domain and frequency-domain features of the vibration signal, if only time-domain or frequency-domain statistical indices are used as input to the wavelet neural network, it is easy to cause deviations and misdiagnoses in the output diagnostic results of the neural network, thereby affecting the accuracy of the final fault diagnosis of the wavelet neural network. Therefore, in order to more comprehensively capture and identify the time-domain and frequency-domain features of engine casing vibration signals, this invention proposes a method based on the fusion of time-domain and frequency-domain feature vector input. This method uses 5 time-domain statistical indices and 2 frequency-domain statistical indices to form a 7-dimensional feature vector, that is, each sample of the wavelet neural network contains 7 different feature values, which are used as the input signal of the neural network.

[0026] The five time-domain features of the vibration signal used in this embodiment of the invention are all dimensionless statistical indices: peak factor, kurtosis factor, impulse factor, waveform factor, and margin factor, which are defined as feature values ​​1 to 5 of the wavelet neural network input. The two frequency-domain features are the mean square frequency and frequency standard deviation, which are defined as feature values ​​6 to 7 of the wavelet neural network input. Thus, each set of wavelet neural network input samples forms a 7-dimensional feature vector formed by the fusion of the time and frequency domains; therefore, the number of nodes in the wavelet neural network input layer in this embodiment should be set to 7.

[0027] The formulas for solving the statistical indices in the time and frequency domains are as follows:

[0028] In the formula, X max The peak value of the vibration signal. x ( n ) represents the discrete form of the time-domain vibration signal. n The index of the discrete point. N Indicates the length of the discrete signal. The mean of the signal. The average amplitude of the signal. RMS x This is the root mean square value of the signal. β The kurtosis of the signal, σ x The standard deviation of the signal. X r The square root amplitude of the signal. k For frequency component indices, values ​​range from 0 to N-1, Δ f It is the frequency resolution, which is equal to the signal sampling frequency. f s Divide by the discrete signal length N, For the center frequency, P [ k ] is the first k The power of each frequency component.

[0029] Step 2: Design of hidden layers in wavelet neural networks.

[0030] For the hidden layers of a wavelet neural network, the core lies in the selection and setting of the activation function. Compared to traditional backpropagation neural networks, the activation function of each node in the hidden layer of the wavelet neural network in this embodiment of the invention is the Morlet wavelet basis function, and its function graph is shown below. Figure 3 As shown. This function can better capture the time-frequency characteristics in vibration signals and fully utilize the generalization ability of wavelet analysis. Additionally, regarding the number of hidden layer nodes... p The settings will be explained in detail in the grid search strategy section below.

[0031] The mathematical expression for the Morlet wavelet basis function is:

[0032] In the formula, x The horizontal axis represents the time variable. e It is a natural constant.

[0033] Step 3: Design of the wavelet neural network output layer.

[0034] For the wavelet neural network output layer, it should correspond to the different fault output types of the rotating aero-engine, namely, type 1 normal signal, type 2 rotation-rotation rubbing, and type 3 rotation-static rubbing fault, that is, a total of three fault types are output. Therefore, in this embodiment, the number of nodes in the wavelet neural network output layer should be set to 3. As for the activation function of the wavelet neural network output layer, this invention selects the linear Purelin function, the graph of which is shown below. Figure 4 As shown.

[0035] Therefore, the basic structure of the rotation-rotation / rotation-static collision fault diagnosis model based on wavelet neural networks is "7- p -3".

[0036] The following is combined with, for example Figure 5 The flowchart shown here illustrates the basic process of fault diagnosis using wavelet neural networks, specifically the rotation-rotation / rotation-static collision fault diagnosis process.

[0037] Step 1: First, during the test run of the rotating aero-engine, a total of 3*48 sets of vibration fault data were acquired from sensors at multiple cross-sectional measuring points of the casing, and wavelet denoising preprocessing was performed. Here, the 3*48 sets of data represent three types of vibration signal states of the engine: Type 1 normal vibration signal, Type 2 rotation-rotation rubbing, and Type 3 rotation-static rubbing.

[0038] Step 2: Divide all 3*48 sets of vibration sample data after wavelet denoising into two categories: a 3*40 set of wavelet neural network training samples and a 3*8 set of neural network test samples. The training set is used for the learning and training process of the wavelet neural network, while the test set is used to verify and test the fault diagnosis effect of the wavelet neural network.

[0039] Step 3: Based on the time-domain and frequency-domain fusion feature vector solution method of vibration signal mentioned above, calculate the fusion feature vector of 3*40 sets of neural network training set samples and 3*8 sets of test set samples respectively, and perform input data normalization processing to obtain fusion feature vectors where the input data are all in the interval [0,1].

[0040] Step 4, according to Figure 2 The diagram shown illustrates the structure of the wavelet neural network model, which is constructed as "7- p A wavelet neural network with a depth of -3” was then fed with 3*40 sets of training samples, and the Levenberg-Marquardt (LM) algorithm was used instead of the traditional gradient descent method for network training. Here, the optimal number of nodes in the hidden layer of the wavelet neural network is determined. pThe following grid search strategy will be used to calculate the impact of different numbers of hidden layer nodes on the output error of the neural network training set. The number of hidden layer nodes corresponding to the minimum output error value is the optimal number of hidden layer nodes for the wavelet neural network. p .

[0041] The weight update formula for the LM algorithm is as follows:

[0042] In the formula, W It is a vector consisting of all the values ​​and biases of the network. J Let be the Jacobian matrix of the differential of the error with respect to the weights. μ For a variable parameter factor, I It is an identity matrix.

[0043] The connection process between layers in a neural network requires the calculation of weights. Throughout the network training process, these connection weights are continuously calculated and updated until the network outputs the error. e If the requirements are met, the weights of the connections between the network layers will also be determined at this point. These are the optimal weights; that is, the optimal output error corresponds to the optimal connection weights. The completion of neural network training means that the connection weights are also fixed and will no longer change. This is equivalent to the network forming a stable connection relationship, which is used for subsequent diagnostic processes on the network test set.

[0044] Step 5: After the above solution, the wavelet neural network hidden layer is set to the optimal number of nodes. p If the wavelet neural network training set output error e If the target output error value is met, it means that the neural network has met the training index requirements, and a well-trained wavelet neural network fault diagnosis model can be obtained.

[0045] Step Six: After the wavelet neural network is trained, the time-domain and frequency-domain fused feature vectors of the 3*8 test set sample data normalized in Step Three are input into the trained wavelet neural network fault diagnosis model to classify the 3*8 test set sample data for fault diagnosis. Finally, the fault diagnosis accuracy is calculated by comparing the output value of the wavelet neural network with the true value, thereby effectively evaluating the overall performance of the wavelet neural network rotation-rotation / rotation-static collision fault diagnosis method proposed in this invention.

[0046] At this point, the basic structure and fault diagnosis process of the wavelet neural network model have been completed. The next step will be to determine the number of nodes in the hidden layers of the neural network in this embodiment based on a grid search strategy. p .

[0047] First, for a 3-layer neural network structure, the number of hidden layer nodes... pThe preliminary determination can be made based on the empirical formula in Table 1, where... m The number of input layer nodes is 7. n The number of output layer nodes is 3.

[0048] Table 1 Number of hidden layer nodes in neural networks p Reference range

[0049] in, a It is a constant, and its value range is [1, 10].

[0050] As shown in Table 1, in this embodiment, the number of hidden layer nodes in the neural network is... p The initial range can be set to 5-13. Further, to determine the optimal number of hidden layer nodes for the wavelet neural network rotation-rotation / rotation-static collision fault diagnosis model in this embodiment, the present invention employs a grid search strategy to traverse and calculate the output error value of the wavelet neural network training set when the number of hidden layer nodes is 5-13. e Furthermore, the number of hidden layer nodes corresponding to the minimum output error value is the optimal number of hidden layer nodes in the wavelet neural network. p .

[0051] Subsequently, the normalized 3*40 sets of training sample data were input into wavelet neural networks with different numbers of hidden layer nodes, respectively. That is, a grid search strategy was used to calculate the output error values ​​of the wavelet neural networks with 5 to 13 hidden layer nodes. e After training with a wavelet neural network, the following can be obtained: Figure 6 and Figure 7 The comparison charts show the output errors and training set prediction results for different numbers of hidden layer nodes.

[0052] according to Figure 6 The comparison results show that when the number of hidden layer nodes in the neural network is 12, the training set output error of the wavelet neural network is... e The minimum value of 0% was reached. Therefore, in this embodiment, the optimal number of hidden layer nodes in the wavelet neural network is... p It should be set to 12, that is, the number of nodes in each layer of the wavelet neural network rotation-rotation / rotation-collision fault diagnosis model is "7-12-3", and its structure is as follows: Figure 8 As shown.

[0053] In this wavelet neural network structure (number of hidden layer nodes) p (12 samples), for the 3*40 training set samples in this embodiment, the comparison chart and confusion matrix of the fault diagnosis prediction results are as follows: Figure 9As shown in the figure, with the optimal number of hidden layer nodes, no misdiagnosis occurred in the 3*40 training set samples, and the network prediction values ​​of the 3*40 training set samples were consistent with the actual fault values. This indicates that the wavelet neural network-based rotation-rotation / rotation-static collision fault diagnosis model has good training performance.

[0054] Finally, the normalized 3*8 test set sample data is input into the trained wavelet neural network to verify and analyze the fault diagnosis model. Figure 10 The image shows a comparison of prediction results and a confusion matrix for 3*8 test set samples. Figure 10 As can be seen from the data, for the wavelet neural network collision fault diagnosis model trained in this embodiment, when the number of hidden layer nodes is set to the optimal value (12), the network prediction values ​​of the 3*8 test set samples are completely consistent with the actual fault values, that is, no misdiagnosis occurs in the 3*8 test set samples. Moreover, in this process, the total fault diagnosis time of the wavelet neural network is less than 200 ms, which is significantly better than the traditional backpropagation neural network.

[0055] Therefore, the wavelet neural network fault diagnosis method proposed in this invention has good diagnostic accuracy and efficiency for rotary aero-engine rotation-rotation / rotation-static contact faults.

[0056] In this embodiment, a computer device is provided, such as... Figure 11 As shown, it includes a memory 1101, a processor 1102, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned wavelet neural network-based engine fault diagnosis methods.

[0057] Specifically, the computer device can be a computer terminal, a server, or a similar computing device.

[0058] In this embodiment, a computer-readable storage medium is provided, which stores a computer program that executes any of the above-described wavelet neural network-based engine fault diagnosis methods.

[0059] Specifically, computer-readable storage media include both permanent and non-permanent, removable and non-removable media, which can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer-readable storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable storage media do not include transient media, such as modulated data signals and carrier waves.

[0060] Based on the same inventive concept, this invention also provides a diagnostic device for engine faults based on wavelet neural networks, as described in the following embodiments. Since the principle of the engine fault diagnosis device based on wavelet neural networks is similar to that of the engine fault diagnosis method based on wavelet neural networks, the implementation of the engine fault diagnosis device based on wavelet neural networks can refer to the implementation of the engine fault diagnosis method based on wavelet neural networks, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0061] Figure 12 This is a structural block diagram of an engine fault diagnosis device based on a wavelet neural network according to an embodiment of the present invention, such as... Figure 12 As shown, it includes: a wavelet neural network construction module 1201, a wavelet neural network training module 1202, and a fault diagnosis module 1203. The structure is described below.

[0062] A wavelet neural network module 1201 is constructed to construct a wavelet neural network including an input layer, a hidden layer, and an output layer, and to determine the number of nodes in the input layer and the number of nodes in the output layer. The input layer is used to receive vibration input signals generated by the engine, the hidden layer is used to perform nonlinear transformation on the vibration input signals, and the output layer is used to output the diagnostic results of engine faults. The wavelet neural network training module 1202 is used to normalize the dataset including training set samples and test set samples, input the normalized training set samples into the wavelet neural network, train the wavelet neural network using the LM algorithm, and determine the optimal number of hidden layer nodes based on the number of nodes in the input layer and the number of nodes in the output layer. p The optimal number of hidden layer nodes p The number of nodes in the hidden layer is used to generate the trained wavelet neural network; The fault diagnosis module 1203 is used to input the normalized test set samples into the trained wavelet neural network to generate engine fault diagnosis results. The types of diagnosis results include normal, rotary-rotational contact fault and rotary-static contact fault.

[0063] In one embodiment, constructing a wavelet neural network module includes: Construct an output layer activation function unit to use a linear Purelin function as the activation function of the output layer; Construct hidden layer activation function units to use Morlet wavelet basis functions as activation functions for each node of the hidden layer, wherein the Morlet wavelet basis functions... , x The horizontal axis represents the time variable. e It is a natural constant.

[0064] In one embodiment, constructing the wavelet neural network module further includes: A vibration signal time-domain feature unit is defined to determine the vibration signal time-domain features, which include peak factor, kurtosis factor, impulse factor, waveform factor, and margin factor. A frequency domain feature unit is defined to determine frequency domain features, wherein the frequency domain features include mean square frequency and frequency standard deviation; A solution formula unit is constructed to set the sum of the number of time-domain features and the number of frequency-domain features of the vibration signal as the number of nodes in the input layer of the wavelet neural network, and to construct solution formulas for the time-domain features and the frequency-domain features of the vibration signal respectively. The output layer node number unit is used to set the number of diagnostic result types to the number of nodes in the output layer.

[0065] In one embodiment, the construction of the solution formula unit is also used for peak factor. ,in ，X max The peak value of the vibration signal ，x ( n () represents the discrete form of the time-domain vibration signal. n The index of the discrete point. NThe length of the discrete signal, RMS x The root mean square value of the signal; kurtosis factor. ,in, β The kurtosis of the signal, σ x The standard deviation of the signal. The mean of the signal; impulse factor. ,in, The average amplitude of the signal; waveform factor ; margin factor ,in, X r The root square amplitude of the signal; the mean square frequency. ,in, k For frequency component index, Δ f For frequency resolution, P [ k ] is the first k Power of each frequency component; frequency standard deviation ,in, The center frequency.

[0066] In one embodiment, the wavelet neural network training module includes: A range determination unit is used to determine the range of the number of hidden layer nodes based on the number of nodes in the input layer and the number of nodes in the output layer. The "Set Number of Groups" unit is used to set the number of training sample groups. The network training unit is used to sequentially input normalized training set sample data of a predetermined number of groups into the wavelet neural network with different numbers of hidden layer nodes within the range of the number of hidden layer nodes, and obtain the output error value. e And the prediction results of neural networks; The optimal hidden layer node number calculation unit is used to calculate the number of nodes based on the error value. e The optimal number of hidden layer nodes is determined by the prediction results of the training set of the neural network. p。

[0067] In one embodiment, the optimal hidden layer node count calculation unit is further configured to cross-compare the training set prediction results of the neural network with the actual results to generate a confusion matrix; determine whether the prediction results are correct based on the confusion matrix; and set the error value... e Using the minimum and the correctness of the prediction result as conditions, determine whether the current number of hidden layer nodes is the optimal number of hidden layer nodes. p .

[0068] In one embodiment, the wavelet neural network training module further includes: Define a weight update formula unit, used to define the weight update formula. ,in, W Let be a vector composed of all values ​​and biases of the wavelet neural network. J Let be the Jacobian matrix of the differential of the error with respect to the weights. μ For variable parameter factors, I It is the identity matrix; The weight update unit is used to update the weights based on the update formula until the error output by the wavelet neural network is updated. e The requirements are met.

[0069] The embodiments of the present invention achieve the following technical effects: The aircraft engine collision fault diagnosis method of this invention combines wavelet analysis with artificial neural networks. To avoid the inability of a single statistical index to simultaneously cover the typical time-domain and frequency-domain features of the vibration signal, thus affecting the accuracy of the final fault diagnosis by the neural network, this method fuses five time-domain features and two frequency-domain features to form a 7-dimensional feature vector. This proposes a wavelet neural network collision fault diagnosis method based on the input of the fused time-domain and frequency-domain feature vector. Furthermore, the aircraft engine collision fault diagnosis method of this invention addresses the issue that traditional neural network algorithms mainly use gradient descent to train the network, and gradient descent becomes less effective near minimum values. Its convergence speed is slower, and multiple iterations are required to find the optimal solution. Therefore, this fault diagnosis method uses the Levenberg-Marquardt (LM) algorithm instead of the traditional gradient descent method for wavelet neural network training, effectively improving the fault diagnosis efficiency and network stability of the wavelet neural network. In order to optimize the hidden layer structure of the wavelet neural network, the aero-engine collision fault diagnosis method of this invention employs a wavelet neural network structure optimization method based on a grid search strategy. This method can search for the optimal number of hidden layer nodes p in the wavelet neural network within a local range, thereby significantly improving the fault diagnosis accuracy of the wavelet neural network.

[0070] Obviously, those skilled in the art should understand that the modules or steps of the above-described embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

[0071] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing engine faults based on wavelet neural networks, characterized in that, include: A wavelet neural network comprising an input layer, a hidden layer, and an output layer is constructed, and the number of nodes in the input layer and the number of nodes in the output layer are determined. The input layer is used to receive vibration input signals generated by the engine, the hidden layer is used to perform nonlinear transformation on the vibration input signals, and the output layer is used to output the diagnostic results of engine faults. The dataset, including training and test samples, is normalized. The normalized training samples are then input into the wavelet neural network. The wavelet neural network is trained using the LM algorithm. The optimal number of hidden layer nodes is determined based on the number of nodes in the input layer and the number of nodes in the output layer. p The optimal number of hidden layer nodes p The number of nodes in the hidden layer is used to generate the trained wavelet neural network; The normalized test set samples are input into the trained wavelet neural network to generate engine fault diagnosis results. The types of diagnosis results include normal, rotary-rotational contact fault, and rotary-static contact fault.

2. The engine fault diagnosis method based on wavelet neural network as described in claim 1, characterized in that, Determining the number of nodes in the input layer and the number of nodes in the output layer includes: The time-domain characteristics of the vibration signal are determined, including peak factor, kurtosis factor, impulse factor, waveform factor, and margin factor. Determine the frequency domain characteristics, which include the mean square frequency and the frequency standard deviation; The sum of the number of time-domain features and the number of frequency-domain features of the vibration signal is set as the number of nodes in the input layer of the wavelet neural network, and the solution formulas for the time-domain features and the frequency-domain features of the vibration signal are constructed respectively. Set the number of diagnostic result types to the number of nodes in the output layer.

3. The engine fault diagnosis method based on wavelet neural network as described in claim 2, characterized in that, Formulas for solving the time-domain and frequency-domain features of the vibration signal are constructed respectively, including: Peak factor ,in ，X max The peak value of the vibration signal ，x ( n () represents the discrete form of the time-domain vibration signal. n The index of the discrete point. N The length of the discrete signal, RMS x This is the root mean square value of the signal; kurtosis factor ,in, β The kurtosis of the signal, σ x The standard deviation of the signal. The mean of the signal; Pulse factor ,in, The average amplitude of the signal; Waveform factor ; margin factor ,in, X r This represents the root square amplitude of the signal. Mean square frequency ,in, k For frequency component index, Δ f For frequency resolution, P [ k ] is the first k Power of each frequency component; Frequency standard deviation ,in, The center frequency.

4. The engine fault diagnosis method based on wavelet neural network as described in claim 1, characterized in that, The optimal number of hidden layer nodes is determined based on the number of nodes in the input layer and the number of nodes in the output layer. p ,include: The range of the number of hidden layer nodes is determined based on the number of nodes in the input layer and the number of nodes in the output layer. Set the number of training sample groups; Within the range of the number of hidden layer nodes, normalized training set sample data of a predetermined number of groups are sequentially input into the wavelet neural networks with different numbers of hidden layer nodes to obtain the output error values. e And the prediction results of neural networks; Based on the error value e The optimal number of hidden layer nodes is determined by the prediction results of the training set of the neural network. p。 5. The engine fault diagnosis method based on wavelet neural network as described in claim 4, characterized in that, Based on the error value e The optimal number of hidden layer nodes is determined by the prediction results of the training set of the neural network. p ,include: The prediction results of the training set of the neural network are cross-compared with the actual results to generate a confusion matrix; Determine whether the prediction result is correct based on the confusion matrix; The error value e Using the minimum and the correctness of the prediction result as conditions, determine whether the current number of hidden layer nodes is the optimal number of hidden layer nodes. p .

6. The engine fault diagnosis method based on wavelet neural network as described in claim 1, characterized in that, The wavelet neural network is trained using the LM algorithm, including: Define the weight update formula ,in, W Let be a vector composed of all values ​​and biases of the wavelet neural network. J Let be the Jacobian matrix of the differential of the error with respect to the weights. μ For variable parameter factors, I It is the identity matrix; The weights are updated based on the update formula until the error output by the wavelet neural network is updated. e The requirements are met.

7. The engine fault diagnosis method based on wavelet neural network as described in any one of claims 1 to 6, characterized in that, Constructing a wavelet neural network including an input layer, hidden layers, and an output layer, including: Use the linear Purelin function as the activation function for the output layer; The Morlet wavelet basis function is used as the activation function for each node of the hidden layer. , x The horizontal axis represents the time variable. e It is a natural constant.

8. A diagnostic device for engine faults based on wavelet neural networks, characterized in that, include: A wavelet neural network module is constructed to build a wavelet neural network including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer and the number of nodes in the output layer are determined. The input layer is used to receive the vibration input signal generated by the engine, the hidden layer is used to perform nonlinear transformation on the vibration input signal, and the output layer is used to output the diagnostic results of engine faults. The wavelet neural network training module is used to normalize the dataset, which includes training and test samples. The normalized training samples are then input into the wavelet neural network, which is trained using the LM algorithm. Finally, the optimal number of hidden layer nodes is determined based on the number of nodes in the input and output layers. p The optimal number of hidden layer nodes p The number of nodes in the hidden layer is used to generate the trained wavelet neural network; The fault diagnosis module is used to input the normalized test set samples into the trained wavelet neural network to generate engine fault diagnosis results. The types of diagnosis results include normal, rotary-rotational contact fault, and rotary-static contact fault.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the engine fault diagnosis method based on wavelet neural network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that performs the engine fault diagnosis method based on wavelet neural networks according to any one of claims 1 to 7.

Images