A heavy load gearbox fault diagnosis method and system based on CNN model fusion

Abstract

The application discloses a heavy load gear box fault diagnosis method and system based on a CNN model fusion, and the method comprises the following steps: collecting sound signals of the heavy load gear box to form a training data set; extracting GFCC features to form a feature set, and inputting the feature set into a one-dimensional convolutional neural network model for training, and outputting a recognition result set; superimposing sound signals on sound signal spectrums to obtain a sound spectrum diagram, inputting the sound spectrum diagram into a two-dimensional convolutional neural network model for training, and outputting a recognition result set; fusing the recognition results of the two models to obtain a fusion CNN model; obtaining a fault sound signal, performing corresponding feature extraction on the fault sound signal, and constructing a sound spectrum diagram; inputting the feature set and the sound spectrum diagram into the fusion convolutional neural network model respectively to obtain a fault recognition result, and obtaining heavy load gear box fault diagnosis information. The scheme can more accurately detect the fault of the heavy load gear box.

Description

TECHNICAL FIELD

[0001] The present application relates to gear box fault diagnosis technology, in particular to a kind of based on CNN model fusion's heavy load gear box fault diagnosis method and system. BACKGROUND

[0002] Heavy load gear box is the important functional component of modern equipment manufacturing industry to transmit motion and deploy speed, heavy load gear box system generally includes large gear, bearing, shaft and box four parts, same as ordinary gear box, the fault of heavy load gear box can be divided into mechanical fault, electrical fault, auxiliary system fault.Mechanical fault is mainly gear fault, bearing fault, shaft fault, box fault, electrical fault and auxiliary system fault mainly have cooling fault, oil supply fault, sensor fault.In the three types of faults, electrical fault and auxiliary system fault although the frequency of occurrence is higher, but the consequence produced is relatively not serious, and processing is more convenient.Mechanical fault is less, but due to the complex structure of heavy load gear box and the huge size, therefore mechanical fault processing is difficult, and the cost is higher.According to statistics, 73% of fault processing time is in processing mechanical fault.

[0003] When heavy load gear box component fails, abnormal sound will be generated, and the amplitude and frequency components of the sound signal will change accordingly, and the sound signal of heavy load gear box contains a large amount of running state information of heavy load gear box internal components, so the sound signal generated by heavy load gear box can effectively reflect the running state of heavy load gear box.Through the analysis of heavy load gear box sound signal, accurate judgment can be made without stopping the machine, and the sound signal detection method has the advantages of non-contact, fast diagnosis speed, high precision, accurate fault position judgment, etc., therefore, sound detection method is widely used in heavy load gear box fault diagnosis at present. SUMMARY

[0004] The purpose of the present application is to provide a kind of based on CNN model fusion's heavy load gear box fault diagnosis method and system, can more accurately detect heavy load gear box fault.

[0005] Technical scheme: a kind of based on CNN model fusion's heavy load gear box fault diagnosis method of the present application, it includes the following steps:

[0006] Collect heavy load gear box sound signal, form training data set;

[0007] The GFCC feature set is extracted from the collected sound signal to form a feature set, which is input into the pre-constructed initial one-dimensional convolutional neural network model for training, to obtain the trained one-dimensional convolutional neural network model, and output the recognition result set;

[0008] The sound signal is superimposed on the sound signal spectrum to obtain a sound spectrogram, the sound spectrogram is input into an initial two-dimensional convolutional neural network model constructed in advance for training, a trained two-dimensional convolutional neural network model is obtained, and a recognition result set is output;

[0009] The recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model are input into a fusion model for fusion training, and finally a trained fusion convolutional neural network model including the one-dimensional convolutional neural network model, the two-dimensional convolutional neural network model, and the fusion model is obtained.

[0010] The fault sound signal is obtained, corresponding feature extraction and sound spectrogram construction are performed on the fault sound signal, the feature set and the sound spectrogram are input into the trained fusion convolutional neural network model respectively, a fault recognition result is obtained, and overload gearbox fault diagnosis information is obtained.

[0011] Further, the sound signal collected is extracted to form a feature set, including the following steps:

[0012] The collected sound signal is framed;

[0013] The framed sound signal is processed with a Hamming window;

[0014] The sound signal processed with the Hamming window is subjected to discrete Fourier transform to obtain an energy spectrum;

[0015] The energy spectrum is filtered by using a Gammatone filter bank;

[0016] The filtered energy spectrum is subjected to logarithmic compression to obtain a corresponding logarithmic energy signal;

[0017] The energy signal is subjected to discrete cosine transform;

[0018] Information entropy is introduced to measure the complexity of the energy signal after the discrete cosine transform, a threshold value is set, GFCC feature parameters are obtained, and a required feature set is formed.

[0019] Further, the sound signal collected is superimposed on the sound signal spectrum to obtain a sound spectrogram, including the following steps:

[0020] The collected sound signal is framed;

[0021] The framed sound signal is processed with a Hamming window;

[0022] The sound signal processed with the Hamming window is subjected to fast Fourier transform to obtain a frequency spectrum.

[0023] The frequency spectrum obtained by the fast Fourier transform is subjected to principal component analysis for dimension reduction;

[0024] The dimension-reduced spectrum is superimposed to obtain a spectrogram.

[0025] Further, the one-dimensional convolutional neural network model comprises a one-dimensional feature parameter input layer, two convolutional layers, two maximum pooling layers, a full connection layer, and an output layer.

[0026] Further, the convolutional neural network model comprises a two-dimensional image input layer, two convolutional layers, two maximum pooling layers, a full connection layer, and an output layer.

[0027] Further, the recognition results of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model are fused by using a weighted average method.

[0028] Further, when the collected acoustic signals are framed, each frame covers a time of 2-3 periods.

[0029] Based on the same inventive concept, a heavy load gearbox acoustic signal fault diagnosis system based on CNN model fusion provided by the application comprises:

[0030] A signal acquisition module is configured to collect acoustic signals of a heavy load gearbox to form a training data set.

[0031] A one-dimensional CNN model construction module is configured to extract GFCC features from the collected acoustic signals to form a feature set, input the feature set into a pre-constructed initial one-dimensional convolutional neural network model for training, obtain a trained one-dimensional convolutional neural network model, and output a recognition result set.

[0032] A two-dimensional CNN model construction module is configured to superimpose acoustic signal spectra on the collected acoustic signals to obtain a spectrogram, input the spectrogram into a pre-constructed initial two-dimensional convolutional neural network model for training, obtain a trained two-dimensional convolutional neural network model, and output a recognition result set.

[0033] A fusion CNN model construction module is configured to input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model into a fusion model for fusion training, and finally obtain a trained fusion convolutional neural network model comprising the one-dimensional convolutional neural network model, the two-dimensional convolutional neural network model, and the fusion model.

[0034] A fault recognition module is configured to acquire fault acoustic signals, perform corresponding feature extraction and spectrogram construction on the fault acoustic signals, input the feature set and the spectrogram into the trained fusion convolutional neural network model respectively, obtain fault recognition results, and obtain heavy load gearbox fault diagnosis information.

[0035] Based on the same inventive concept, the application discloses a heavy load gearbox fault diagnosis device based on CNN model fusion, which comprises a processor and a memory, the memory stores computer instructions, and the processor is used for executing the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the electronic device realizes the steps of the above-mentioned heavy load gearbox fault diagnosis method based on CNN model fusion.

[0036] Based on the same inventive concept, the application discloses a computer readable storage medium, which stores a computer program, and when the program is executed by a processor, the steps of the above-mentioned heavy load gearbox fault diagnosis method based on CNN model fusion are realized.

[0037] Beneficial effects: Compared with the prior art, the significant technical effects of the application are:

[0038] (1) The heavy load gearbox is large in size and complex in structure, so the sound signal data is large, the amplitude and frequency domain change quickly, the convolutional neural network has strong data representation and analysis capability, and the local features of the data can be mined, so the convolutional neural network fault diagnosis technology is adopted, and compared with other fault diagnosis technologies, the error rate is lower, the diagnosis time is shorter, and the efficiency is higher.

[0039] (2) The one-dimensional CNN model based on GFCC which can completely describe the joint distribution characteristics of the time domain and frequency domain of the sound signal and the two-dimensional CNN model based on the sound spectrum diagram which can express the signal intensity of different frequency bands of the sound signal are fused and weighted by using the model fusion technology, and finally a fusion CNN model including the one-dimensional CNN model, the two-dimensional CNN model and the fusion model is formed, the advantages of different models are fully utilized, the feature loss problem when a single model is used for diagnosis is effectively improved, and the classification accuracy and the diagnosis success rate of the convolutional neural network model are improved.

[0040] (3) In the case of improving the model accuracy, the method can light up the convolutional neural network model, can be well applied to industrial production to identify the fault categories of the heavy load gearbox, solves the problem of high error rate of the traditional fault diagnosis method, and has good application prospect in actual engineering. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a flowchart of a heavy load gearbox fault diagnosis method based on CNN model fusion disclosed by the embodiment of the application;

[0042] Figure 2 is a flowchart of extracting GFCC features disclosed by the embodiment of the application;

[0043] Figure 3 is a flowchart of constructing a sound spectrum diagram disclosed by the embodiment of the application;

[0044] Figure 4 This is a schematic diagram of the structure of the fusion of the one-dimensional CNN and two-dimensional CNN models disclosed in an embodiment of the present invention;

[0045] Figure 5 This is a structural diagram of a heavy-duty gearbox fault diagnosis system based on CNN model fusion disclosed in an embodiment of the present invention;

[0046] Figure 6 This is a structural schematic diagram of a heavy-duty gearbox fault diagnosis device based on CNN model fusion disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The technical solution of the present invention is described in detail below in conjunction with specific implementation methods and the accompanying drawings.

[0048] Example 1

[0049] like Figure 1 As shown, a heavy-duty gearbox fault diagnosis method based on CNN model fusion of the present invention can be applied to heavy-duty gearbox fault diagnosis, including the following steps:

[0050] S1. Collect the acoustic signals of the heavy-loaded gearbox to form a training data set.

[0051] In this step, the acoustic signal of the heavy-load gearbox can be collected by the acoustic pressure sensor when the gearbox is in operation.

[0052] In this embodiment, the sampling sensor is arranged at a distance of 1m from the surface of the gearbox to collect the gearbox working noise; the sampling frequency is 3KHz, the number of sampling points is 1000, and four types of signals are collected: normal rotation sound, gear fault sound, shaft fault sound, and bearing fault sound. The collected signals are saved as .wav files.

[0053] In this embodiment, before obtaining the acoustic signal of the heavy-loaded gearbox for detecting a heavy-loaded gearbox fault, the heavy-loaded gearbox convolutional neural network model should be trained in advance. Thus, when detecting a heavy-loaded gearbox fault, the pre-established heavy-loaded gearbox convolutional neural network model can be directly called.

[0054] S2. Extract GFCC Gammatone filter cepstral coefficients (CSFCs) features from the collected acoustic signal to form a feature set. The feature set is input into a pre-built initial one-dimensional convolutional neural network model for training to obtain a trained one-dimensional convolutional neural network model and output the recognition result.

[0055] In this step, combine Figure 2 , Figure 2 The steps of extracting features from acoustic signals shown here can be used to extract features from acoustic signals.Figure 1 The feature extraction of the sound signal involved in the actual detection of the heavy-duty gearbox fault process is shown in

[0056] As shown in Figure 2 In this embodiment, the GFCC features of the collected sound signal are extracted to form a feature set, including the following steps:

[0057] S2.1, frame the collected sound signal, and each frame covers about 2-3 periods of time;

[0058] S2.2, Hamming window processing is performed on the framed sound signal to increase the continuity of the two ends of the frame and reduce the spectral leakage;

[0059] S2.3, discrete Fourier transform (DFT transform) is performed on the Hamming window processed sound signal to obtain an energy spectrum;

[0060] S2.4, filter the energy spectrum using a Gammatone filter bank;

[0061] S2.5, log compression is performed on the filtered energy spectrum to obtain the corresponding logarithmic energy signal;

[0062] S2.6, discrete cosine transform (DCT) is performed on the energy signal;

[0063] S2.7, introduce information entropy to measure the complexity of the discrete cosine transformed energy signal, set a threshold, obtain the GFCC feature parameters, and form the required feature set.

[0064] S3, superimpose the sound signal spectrum on the collected sound signal to obtain a sound spectrogram, input the sound spectrogram into the pre-constructed initial two-dimensional convolutional neural network model for training, obtain the trained two-dimensional convolutional neural network model, and output the recognition result set.

[0065] In this step, the feature processing of the sound signal involved in the actual detection of the heavy-duty gearbox fault process is shown in Figure 3 , Figure 3 The flowchart for constructing the sound spectrogram is shown. The steps for constructing the sound spectrogram of the sound signal shown herein can be used for the feature processing of the sound signal involved in the actual detection of the heavy-duty gearbox fault process shown in Figure 1

[0066] As shown in Figure 3 In this embodiment, the sound spectrogram is obtained by superimposing the sound signal spectrum on the collected sound signal, including the following steps:

[0067] S3.1, frame the collected sound signal, and the overlap rate of each frame is set to 50%;

[0068] ​S3.2, Hamming window processing is performed on the framed sound signal to increase the continuity of both ends of the frame and reduce spectral leakage;

[0069] S3.3, fast Fourier transform (FFT transform) is performed on the sound signal after Hamming window processing to obtain a frequency spectrum;

[0070] S3.4, principal component analysis (PCA) dimension reduction is performed on the frequency spectrum obtained by the fast Fourier transform;

[0071] S3.5, superposition is performed on the frequency spectrum after dimension reduction to obtain a spectrogram.

[0072] S4, the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model are input into a fusion model for fusion training, and finally a trained fusion convolutional neural network model including the one-dimensional convolutional neural network model, the two-dimensional convolutional neural network model and the fusion model is obtained.

[0073] In this step, the weighted average method is used to fuse the recognition results of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model.

[0074] In this scheme, the training model algorithm is a black box model algorithm, that is, the specific working process cannot be known, and only enough data needs to be input to complete the training process.

[0075] Figure 4 The structure diagram of the overload gearbox fault diagnosis CNN fusion model is shown. In this embodiment, the gear one-dimensional convolutional neural network model includes a one-dimensional feature parameter input layer, two convolutional layers, two maximum pooling layers, a full connection layer and an output layer. As shown in FIG. 4, the gear two-dimensional convolutional neural network model includes a two-dimensional feature parameter input layer, two convolutional layers, two maximum pooling layers, a full connection layer and an output layer. Figure 4As shown, the input of the one-dimensional model is a 26x1-dimensional feature vector. The first convolutional layer has 16 convolutional kernels with a size of 8x1 and a step of 2. A non-linear factor is introduced by using a RELU activation function after convolution. The resulting feature vector is padded with 0 at the edge. After the first convolutional layer, a first pooling layer is connected to compress the feature vector after convolution, thereby simplifying the network calculation complexity. A maximum pooling is performed using a 2x1 convolutional kernel, thereby obtaining 16 feature vectors with a size of 8x1. The second convolutional layer has 32 convolutional kernels with a size of 8x1 and a step of 2. The activation function is RELU, and the edge is padded with 0. The second pooling layer uses a 2x1 convolutional kernel to perform maximum pooling, thereby obtaining 32 feature vectors with a size of 2x1. After the second pooling layer, a fully connected layer is connected to connect all features, and the output value is sent to a softmax classifier. After the fully connected layer, a 64x1-dimensional feature vector is obtained. The last one is an output layer with a size of 4x1. Figure 4 As shown, the input of the two-dimensional model is a spectrogram with a size of 32 pixels x 32 pixels x 3 channels (RGB). The first convolutional layer has 32 convolutional kernels with a size of 3x3 and a step of 2. A non-linear factor is introduced by using a RELU activation function after convolution. The resulting feature vector is padded with 0 at the edge. After the first convolutional layer, a first pooling layer is connected to compress the feature map after convolution, thereby simplifying the network calculation complexity. A maximum pooling is performed using a 2x2 convolutional kernel, thereby obtaining 32 feature maps with a size of 16x16. The second convolutional layer has 64 convolutional kernels with a size of 3x3 and a step of 2. The activation function is RELU, and the edge is padded with 0. The second pooling layer uses a 2x2 convolutional kernel to perform maximum pooling, thereby obtaining 64 feature vectors with a size of 8x8. After the second pooling layer, a fully connected layer is connected to connect all features, and the output value is sent to a softmax classifier. After the fully connected layer, a 4096-dimensional feature vector is obtained. The last one is an output layer with a size of 4x1. Finally, the output result set is input into the fusion model to assign weights, fuse the two CNN models, and output the fusion result.

[0076] S5, obtaining a fault sound signal, performing corresponding feature extraction and spectrogram construction on the fault sound signal, inputting the feature set and the spectrogram into the trained fusion convolutional neural network model respectively, obtaining a fault recognition result, and obtaining the overload gearbox fault diagnosis information.

[0077] When the heavy load gearbox is in operation, the detection device starts to work. First, the fault sound signal is acquired, and then the corresponding feature extraction and sound spectrogram construction are performed on the fault sound signal. The feature set and the sound spectrogram are input into the trained fusion convolutional neural network model to obtain the fault recognition result and the heavy load gearbox fault diagnosis information. With the technology for detecting the heavy load gearbox fault provided by the embodiments of the present application, the heavy load gearbox fault can be detected by means of the convolutional neural network model, and the accuracy is high. In addition, the convolutional neural network model for the heavy load gearbox proposed herein is objective and is not affected by subjective factors, so the reliability and stability are high.

[0078] Due to the large amount of sound signal data of the heavy load gearbox and the strong data representation and analysis capability of the convolutional neural network, the convolutional neural network fault diagnosis technology is adopted. When performing fault diagnosis, signal features need to be extracted in advance. The GFCC (Gammatone filter cepstrum coefficient) adopts the Gammatone filter to simulate the cochlear hearing model of the human ear, can more completely describe the joint distribution characteristics of the sound signal in the time domain and the frequency domain, and the spectral peak of the Gammatone filter is more flat, which can effectively improve the energy leakage problem of signal decomposition, so the GFCC features are extracted and input into the constructed one-dimensional CNN model for fault diagnosis. In order to improve the recognition accuracy, a two-dimensional CNN model based on the sound spectrogram capable of expressing the signal intensity of different frequency bands of the sound signal is constructed, and the two models are fused to output the fusion recognition result.

[0079] In the technology of mechanical fault diagnosis, professional sensor signal data is collected, and a heavy load gearbox fault convolutional neural network model is established by using a deep learning algorithm. Compared with traditional manual exploration and defect detection equipment exploration, the fault diagnosis time, manpower and material costs are reduced, and the heavy load gearbox fault diagnosis has great advantages in application prospect.

[0080] Embodiment 2

[0081] As shown in Figure 5 The heavy load gearbox fault diagnosis system based on the CNN model fusion provided by the present application can be applied to heavy load gearbox fault diagnosis, and specifically comprises:

[0082] A signal acquisition module is configured to collect sound signals of the heavy load gearbox to form a training data set.

[0083] A one-dimensional CNN model construction module is configured to extract GFCC features from the collected sound signals to form a feature set, input the feature set into a pre-constructed initial one-dimensional convolutional neural network model for training, obtain a trained one-dimensional convolutional neural network model, and output a recognition result set.

[0084] The two-dimensional CNN model construction module is configured to superimpose the sound signal spectrum on the collected sound signal to obtain a sound spectrum image, input the sound spectrum image into an initial two-dimensional convolutional neural network model constructed in advance for training, obtain a trained two-dimensional convolutional neural network model, and output a recognition result set.

[0085] The fusion CNN model construction module is configured to input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model into a fusion model for fusion training, and finally obtain a trained fusion convolutional neural network model including the one-dimensional convolutional neural network model, the two-dimensional convolutional neural network model and the fusion model.

[0086] The fault recognition module is configured to acquire a fault sound signal, perform corresponding feature extraction and sound spectrum image construction on the fault sound signal, input the feature set and the sound spectrum image into the trained fusion convolutional neural network model respectively, obtain a fault recognition result, and acquire fault diagnosis information of the heavy-duty gear box.

[0087] Embodiment 3

[0088] As shown in the Figure 6 A heavy-duty gear box fault diagnosis device based on CNN model fusion can be applied to heavy-duty gear box fault diagnosis. The device can include a processor and a memory, the memory stores computer instructions, and the processor is configured to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the method described in the above embodiments, and achieves the same technical effects as the above method.

[0089] The memory can include a computer system readable medium in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The device can further include other removable / non-removable, volatile / non-volatile computer system storage media. For example, the memory can be used to read and write non-removable, non-volatile magnetic media (commonly referred to as "hard disk drive"). The program / utility, having a set of (at least one) program modules, can be stored in, for example, memory, such as operating system, one or more application programs, other program modules, and program data, each of which or some combination thereof can include implementation of a network environment. The program modules typically perform the functions and / or methods described in the embodiments of the present application.

[0090] The processor performs various functional applications and data processing by running the program stored in the memory, such as implementing the method provided by the first embodiment of the present application.

[0091] Embodiment 4

[0092] The embodiment 4 of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize steps of the method in the above embodiment and achieve the same technical effects.

[0093] The computer storage medium of the embodiment of the present application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. More specific examples (non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus or device.

[0094] The computer readable signal medium can include a data signal propagated in a baseband or propagated as a carrier wave in a propagated data signal, in which the computer readable program code is contained. Such a propagated data signal can take many forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport program for use by or in connection with an instruction execution system, apparatus, or device.

[0095] The program code contained in the computer readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, cable, RF, etc., or any suitable combination thereof.

[0096] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0097] Of course, the storage medium provided by the embodiments of the present application includes computer executable instructions, and the computer executable instructions are not limited to the method operations described above, but can also perform related operations in the method provided by any embodiment of the present application.

[0098] The above specific embodiments further illustrate the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only for specific embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A heavy load gearbox fault diagnosis method based on CNN model fusion, characterized in that, The method comprises the following steps: Collecting the sound signals of the heavy-duty gearbox to form a training data set; Extracting GFCC features from the collected sound signals to form a feature set, inputting the feature set into a pre-constructed initial one-dimensional convolutional neural network model for training, obtaining the trained one-dimensional convolutional neural network model, and outputting a recognition result set; Superimposing the sound signal spectrum on the collected sound signals to obtain a sound spectrogram, inputting the sound spectrogram into a pre-constructed initial two-dimensional convolutional neural network model for training, obtaining the trained two-dimensional convolutional neural network model, and outputting a recognition result set; Inputting the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model into a fusion model for fusion training, and finally obtaining a trained fusion convolutional neural network model comprising the one-dimensional convolutional neural network model, the two-dimensional convolutional neural network model and the fusion model; Obtaining a fault sound signal, performing corresponding feature extraction and sound spectrogram construction on the fault sound signal, inputting the feature set and the sound spectrogram into the trained fusion convolutional neural network model respectively, obtaining a fault recognition result, and obtaining heavy-duty gearbox fault diagnosis information.

2. The CNN model fusion-based heavy-duty gearbox fault diagnosis method according to claim 1, characterized in that, The method for extracting GFCC features from the collected sound signals to form a feature set comprises the following steps: Frame the collected sound signals; Perform Hamming window processing on the framed sound signals; Perform discrete Fourier transform on the Hamming window processed sound signals to obtain an energy spectrum; Filter the energy spectrum by using a Gammatone filter bank; Logarithmically compress the filtered energy spectrum to obtain a corresponding logarithmic energy signal; Perform discrete cosine transform on the energy signal; Introduce information entropy to measure the complexity of the discrete cosine transformed energy signal, set a threshold, obtain GFCC feature parameters, and form the required feature set.

3. The CNN model fusion-based heavy-duty gearbox fault diagnosis method according to claim 1, characterized in that, The method for superimposing the sound signal spectrum on the collected sound signals to obtain a sound spectrogram comprises the following steps: Frame the collected sound signals; Perform Hamming window processing on the framed sound signals; Perform fast Fourier transform on the Hamming window processed sound signals to obtain a frequency spectrum; Perform principal component analysis dimension reduction on the frequency spectrum obtained by the fast Fourier transform; Superimpose the dimension-reduced frequency spectrum to obtain a sound spectrogram.

4. The CNN model fusion-based heavy-duty gearbox fault diagnosis method according to claim 1, characterized in that: The one-dimensional convolutional neural network model comprises a one-dimensional feature parameter input layer, two convolutional layers, two maximum pooling layers, a full connection layer and an output layer.

5. The CNN model fusion-based heavy-duty gearbox fault diagnosis method according to claim 1, characterized in that: The convolutional neural network model comprises a two-dimensional image input layer, two convolutional layers, two maximum pooling layers, a full connection layer and an output layer.

6. The CNN model fusion-based heavy-duty gearbox fault diagnosis method according to claim 1, characterized in that: The recognition results of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model are fused by using a weighted average method.

7. The CNN model fusion-based heavy-duty gearbox fault diagnosis method according to claim 2 or 3, characterized in that: When the collected sound signals are framed, each frame covers 2-3 periods of time.

8. A heavy load gearbox fault diagnosis system based on CNN model fusion, characterized in that, The method comprises the following steps: A signal acquisition module is configured to collect sound signals of the heavy-duty gearbox to form a training data set; A one-dimensional CNN model construction module is configured to extract GFCC features from the collected sound signals to form a feature set, input the feature set into a pre-constructed initial one-dimensional convolutional neural network model for training, obtain a trained one-dimensional convolutional neural network model, and output a recognition result set; The two-dimensional CNN model construction module is configured to superimpose the sound signal spectrum on the collected sound signal to obtain a sound spectrum image, input the sound spectrum image into an initially constructed two-dimensional convolutional neural network model for training, obtain a trained two-dimensional convolutional neural network model, and output a recognition result set; The fusion CNN model construction module is configured to input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network into a fusion model for fusion training, and finally obtain a trained fusion convolutional neural network model including the one-dimensional convolutional neural network model, the two-dimensional convolutional neural network model, and the fusion model; The fault recognition module is configured to acquire a fault sound signal, perform corresponding feature extraction and sound spectrum image construction on the fault sound signal, input the feature set and the sound spectrum image into the trained fusion convolutional neural network model respectively, obtain a fault recognition result, and acquire fault diagnosis information of the heavy-duty gear box.

9. A heavy load gearbox fault diagnosis device based on CNN model fusion, characterized in that, The device includes a processor and a memory, the memory stores computer instructions, and the processor is configured to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the device implements the steps of the heavy-duty gear box sound signal fault diagnosis method based on CNN model fusion according to 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, which is executed by the processor to implement the steps of the heavy-duty gear box fault diagnosis method based on CNN model fusion according to any one of claims 1 to 7.

Images