CN117235668A - CNN model fusion-based fault diagnosis method and system for heavy-duty gearbox - Google Patents

Abstract

The invention discloses a fault diagnosis method and system for a heavy-duty gearbox based on CNN model fusion, wherein the method comprises the following steps: collecting acoustic signals of the heavy-duty gearbox to form a training data set; extracting GFCC features to form a feature set, inputting the feature set into a one-dimensional convolutional neural network model for training, and outputting an identification result set; superposing acoustic signal spectrum on the acoustic signal to obtain a spectrogram, inputting the spectrogram into a two-dimensional convolutional neural network model for training, and outputting an identification result set; fusing the recognition results of the two models to obtain a fused CNN model; and obtaining a fault sound signal, carrying out corresponding feature extraction and spectrogram construction on the fault sound signal, respectively inputting a feature set and the spectrogram into the fusion convolutional neural network model to obtain a fault identification result and obtain fault diagnosis information of the heavy-load gearbox. The scheme can more accurately detect the fault of the heavy-load gearbox.

Description

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

Technical field

The invention relates to gearbox fault diagnosis technology, and specifically relates to a heavy-duty gearbox fault diagnosis method and system based on CNN model fusion.

Background technique

Heavy-duty gearboxes are important functional components for transmitting motion and adjusting speed in modern equipment manufacturing. Heavy-duty gearbox systems generally include four parts: large gears, bearings, shafts and boxes. They are the same as ordinary gearboxes. Faults can be divided into mechanical faults, electrical faults, and auxiliary system faults. Mechanical failures are mainly gear failures, bearing failures, shaft failures, and box failures. Electrical failures and auxiliary system failures include cooling failures, oil supply failures, and sensor failures. Among these three types of faults, although electrical faults and auxiliary system faults occur more frequently, their consequences are relatively less serious and they are easier to deal with. The number of mechanical failures is low, but due to the complex structure and large size of the heavy-duty gearbox, mechanical failure treatment is difficult and costly. According to statistics, 73% of troubleshooting time is spent dealing with mechanical failures.

When a heavy-duty gearbox component fails, abnormal sound will be produced, and the amplitude and frequency components of the sound signal will change accordingly. The sound signal of the heavy-duty gearbox contains a large number of operating conditions of the internal components of the heavy-duty gearbox. information, so the acoustic signal generated by the heavy-duty gearbox can effectively reflect the operating status of the heavy-duty gearbox. By analyzing the acoustic signal of the heavy-loaded gearbox, the fault can be accurately judged without stopping the machine. The acoustic signal detection method has the advantages of non-contact, fast diagnosis speed, high accuracy, and more accurate fault location judgment. Therefore, , the acoustic detection method is currently widely used in fault diagnosis of heavy-duty gearboxes.

Contents of the invention

Purpose of the invention: The purpose of the present invention is to provide a heavy-load gearbox fault diagnosis method and system based on CNN model fusion, which can detect heavy-load gearbox faults more accurately.

Technical solution: A heavy-duty gearbox fault diagnosis method based on CNN model fusion of the present invention includes the following steps:

Collect the acoustic signals of the heavy-duty gearbox to form a training data set;

Extract GFCC features from the collected acoustic signals to form a feature set, input the feature set into the pre-built initial one-dimensional convolutional neural network model for training, obtain the trained one-dimensional convolutional neural network model, and output the recognition results set;

Add the acoustic signal spectrum to the collected acoustic signal to obtain a spectrogram, input the spectrogram into the pre-constructed initial two-dimensional convolutional neural network model for training, and obtain the trained two-dimensional convolutional neural network model, and Output the recognition result set;

Input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model into the fusion model for fusion training, and finally obtain the trained one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model. , fusion convolutional neural network model including fusion model;

Obtain the fault sound signal, extract the corresponding features and construct the spectrogram of the fault sound signal, input the feature set and spectrogram respectively into the trained fusion convolutional neural network model, obtain the fault identification result, and obtain the heavy-duty gearbox Troubleshooting information.

Further, extracting GFCC features from the collected acoustic signals to form a feature set includes the following steps:

Divide the collected acoustic signals into frames;

Add Hamming window processing to the framed acoustic signal;

Perform discrete Fourier transform on the acoustic signal processed by the Hamming window to obtain the energy spectrum;

Use Gammatone filter bank to filter the energy spectrum;

Perform logarithmic compression on the filtered energy spectrum to obtain the corresponding logarithmic energy signal;

Perform discrete cosine transform on the energy signal;

Information entropy is introduced to measure the complexity of the energy signal after discrete cosine transform, and the threshold is set to obtain the GFCC characteristic parameters to form the required feature set.

Further, superimposing the acoustic signal spectrum on the collected acoustic signal to obtain the spectrogram includes the following steps:

Divide the collected acoustic signals into frames;

Add Hamming window processing to the framed acoustic signal;

The fast Fourier transform is performed on the acoustic signal processed by the Hamming window to obtain the spectrum.

Perform principal component analysis on the spectrum obtained by fast Fourier transform to reduce dimensionality;

The spectrum after dimensionality reduction is superimposed to obtain the spectrogram.

Further, the one-dimensional convolutional neural network model includes a one-dimensional feature parameter input layer, two convolutional layers, two maximum pooling layers, a fully connected layer and an output layer.

Further, the convolutional neural network model includes a two-dimensional image input layer, two convolutional layers, two maximum pooling layers, a fully connected layer and an output layer.

Furthermore, 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.

Furthermore, when the collected acoustic signals are divided into frames, each frame covers 2 to 3 cycles.

Based on the same inventive concept, the invention's heavy-duty gearbox acoustic signal fault diagnosis system based on CNN model fusion includes:

The signal acquisition module is used to collect the acoustic signals of the heavy-duty gearbox to form a training data set;

The one-dimensional CNN model building module is used to extract GFCC features from the collected acoustic signals to form a feature set. The feature set is input into the pre-built initial one-dimensional convolutional neural network model for training, and the trained one-dimensional volume is obtained. Accumulate the neural network model and output the recognition result set;

The two-dimensional CNN model building module is used to superimpose the acoustic signal spectrum on the collected acoustic signal to obtain a spectrogram. The spectrogram is input into the pre-built initial two-dimensional convolutional neural network model for training, and the trained result is obtained. Two-dimensional convolutional neural network model and output a recognition result set;

The fusion CNN model building module is used to input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model into the fusion model, perform fusion training, and finally obtain the trained one-dimensional convolutional neural network model. , two-dimensional convolutional neural network model, fusion model including fusion convolutional neural network model;

The fault identification module is used to obtain the fault acoustic signal, extract the corresponding features and construct the spectrogram of the fault acoustic signal, and input the feature set and spectrogram into the trained fusion convolutional neural network model to obtain the fault identification result. , obtain heavy-duty gearbox fault diagnosis information.

Based on the same inventive concept, the present invention provides a heavy-duty gearbox fault diagnosis equipment based on CNN model fusion, including a processor and a memory. Computer instructions are stored in the memory, and the processor is used to execute the stored information in the memory. Computer instructions, when the computer instructions are executed by the processor, the electronic device implements the steps of the overloaded gearbox fault diagnosis method based on CNN model fusion.

Based on the same inventive concept, the present invention provides a computer-readable storage medium on which a computer program is stored. When the program is executed by the processor, the steps of the overloaded gearbox fault diagnosis method based on CNN model fusion are implemented.

Beneficial effects: Compared with the existing technology, the significant technical effects of the present invention are:

(1) The heavy-duty gearbox has a huge size and complex structure, so the acoustic signal data is huge and the amplitude and frequency domain change rapidly. However, the convolutional neural network has strong data representation and analysis capabilities and can mine local characteristics of the data, so it is used Compared with other fault diagnosis technologies, convolutional neural network fault diagnosis technology has lower error rate, shorter diagnosis time, higher efficiency, and saves time and effort.

(2) Use model fusion technology to fuse the one-dimensional CNN model based on GFCC, which can completely describe the joint distribution characteristics of the acoustic signal in the time domain and frequency domain, and the two-dimensional CNN model based on the spectrogram, which can express the signal strength of the acoustic signal in different frequency bands. Weighted, the fusion CNN model including one-dimensional CNN model, two-dimensional CNN model and fusion model is finally formed, which fully utilizes the advantages of different models, effectively improves the problem of feature loss when a single model is used for diagnosis, and improves the convolutional neural network. Classification accuracy and diagnostic success rate of network models.

(3) With the improvement of model accuracy, this method can lightweight the convolutional neural network model and can be well applied to industrial production to identify fault categories of heavy-duty gearboxes, solving the problem of high identification error rate of traditional fault diagnosis methods. , which has good application prospects in actual projects.

Description of drawings

Figure 1 is a schematic flow chart of a heavy-duty gearbox fault diagnosis method based on CNN model fusion disclosed in an embodiment of the present invention;

Figure 2 is a schematic flowchart of extracting GFCC features disclosed in the embodiment of the present invention;

Figure 3 is a schematic flow chart of constructing a spectrogram disclosed in an embodiment of the present invention;

Figure 4 is a schematic structural diagram of the fusion of one-dimensional CNN and two-dimensional CNN models disclosed in the embodiment of the present invention;

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

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

Detailed ways

The technical solution of the present invention will be introduced in detail below with reference to the specific embodiments and the accompanying drawings.

Example 1

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

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

In this step, the heavy-duty gearbox acoustic signal can be collected by the sound pressure sensor while the gearbox is in operation.

In this embodiment, the sampling sensor is arranged 1m away from the surface of the gearbox to collect the operating noise of the gearbox; the sampling frequency is 3KHz, the number of sampling points is 1000, and there are four types of collected signals: normal rotation sound, gear failure sound, and shaft failure sound. , bearing failure sound, and save the collected signal as a .wav file.

In this embodiment, the heavy-load gearbox convolutional neural network model should be trained in advance before acquiring the acoustic signal of the heavy-load gear box for detecting the fault of the heavy-load gear box. Therefore, when detecting the fault of the heavy-load gear box, the model can be directly Call the pre-built overloaded gearbox convolutional neural network model.

S2. Extract the GFCC Gammatone filter cepstrum coefficient) features from the collected acoustic signals to form a feature set. The feature set is input into the pre-constructed initial one-dimensional convolutional neural network model for training, and the trained one-dimensional Convolutional neural network model and output recognition results.

In this step, combined with Figure 2, Figure 2 shows a flow chart for extracting GFCC features. The step of feature extraction of the acoustic signal shown here can be used for feature extraction of the acoustic signal involved in the actual process of detecting the fault of the heavy-duty gearbox shown in FIG. 1 .

As shown in Figure 2, in this embodiment, extracting GFCC features from the collected acoustic signals to form a feature set includes the following steps:

S2.1. Divide the collected acoustic signals into frames, and each frame covers about 2 to 3 cycles;

S2.2. Add Hamming window processing to the framed acoustic signal to increase the continuity at both ends of the frame and reduce spectrum leakage;

S2.3. Perform discrete Fourier transform (DFT transform) on the acoustic signal processed by the Hamming window to obtain the energy spectrum;

S2.4. Use Gammatone filter bank to filter the energy spectrum;

S2.5. Perform logarithmic compression on the filtered energy spectrum to obtain the corresponding logarithmic energy signal;

S2.6. Perform discrete cosine transform (DCT) on the energy signal;

S2.7. Introduce information entropy to measure the complexity of the energy signal after discrete cosine transform, set the threshold, obtain the GFCC characteristic parameters, and form the required feature set.

S3. Add the acoustic signal spectrum to the collected acoustic signal to obtain a spectrogram, input the spectrogram into the pre-constructed initial two-dimensional convolutional neural network model for training, and obtain the trained two-dimensional convolutional neural network model. , and output the recognition result set.

In this step, combined with Figure 3, Figure 3 shows a flow chart for constructing a spectrogram. The step of spectrogram construction of the acoustic signal shown here can be used for the characteristic processing of the acoustic signal involved in the actual detection of a fault in a heavy-duty gearbox as shown in FIG. 1 .

As shown in Figure 3, in this embodiment, superimposing the acoustic signal spectrum on the collected acoustic signal to obtain a spectrogram includes the following steps:

S3.1. Divide the collected acoustic signals into frames, and set the overlap rate of each frame to 50%;

S3.2. Add Hamming window processing to the framed acoustic signal to increase the continuity at both ends of the frame and reduce spectrum leakage;

S3.3. Perform fast Fourier transform (FFT transform) on the acoustic signal processed by the Hamming window to obtain the spectrum;

S3.4. Perform dimensionality reduction using principal component analysis (PCA) on the spectrum obtained by fast Fourier transform;

S3.5. Superpose the dimensionally reduced spectra to obtain a spectrogram.

S4. Input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model into the fusion model, perform fusion training, and finally obtain the trained model including the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model. Fusion convolutional neural network model including network model and fusion model.

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.

In this solution, the training model algorithm is a black box model algorithm, that is, its specific working process cannot be known, and the training process can be completed by inputting enough data.

Figure 4 shows the structural diagram of the CNN fusion model for heavy-duty gearbox fault diagnosis. In this implementation, the gear one-dimensional convolutional neural network model includes a one-dimensional feature parameter input layer, two layers of convolutional layers, two layers of maximum pooling layers, a fully connected layer and an output layer. As shown in Figure 4, the input of this one-dimensional model is a feature vector of size 26x1. The number of convolution kernels in the first convolution layer is 16, the convolution kernel size is 8x1, and the step size is 2. After convolution, the RELU activation function is used to introduce nonlinear factors. The feature vector obtained after convolution fills the edge part with 0 ; After the first convolution layer, the first pooling layer is connected to compress the convolved feature vector to simplify the network calculation complexity. A convolution kernel of size 2x1 is used for maximum pooling, resulting in 16 pixels of size 8x1 feature vector; the number of convolution kernels in the second convolutional layer is 32, the convolution kernel size is 8x1, the stride is 2, the activation function is RELU, and the edge part is filled with 0; the second pooling layer uses size Maximum pooling is performed for the 2x1 convolution kernel to obtain 32 feature vectors of size 2x1; the second pooling layer is followed by a fully connected layer, which is used to connect all features and send the output value to the softmax classifier. The 64x1 dimensional feature vector obtained after the fully connected layer; the last one is the output layer, with a size of 4x1. The gear two-dimensional convolutional neural network model includes a two-dimensional image input layer, two convolutional layers, two maximum pooling layers, a fully connected layer and an output layer. As shown in Figure 4, the input to this 2D model is a spectrogram with a size of 32 pixels x 32 pixels x 3 channels (RGB). The number of convolution kernels in the first convolution layer is 32, the convolution kernel size is 3x3, and the step size is 2. After convolution, the RELU activation function is used to introduce nonlinear factors. The feature vector obtained after convolution fills the edge part with 0 ; After the first convolution layer, the first pooling layer is connected to compress the convolved feature map to simplify the network calculation complexity. A convolution kernel of size 2x2 is used for maximum pooling, and 32 pixels of size 2 are obtained. 16x16 feature map; the number of convolution kernels in the second convolution layer is 64, the convolution kernel size is 3x3, the stride is 2, the activation function is; RELU, the edge part is filled with; 0; the second pooling layer A convolution kernel of size 2x2 is used for maximum pooling, and 64 feature vectors of size 8x8 are obtained; the second pooling layer is followed by a fully connected layer, which is used to connect all features and send the output value to the softmax classifier. , the 4096-dimensional feature vector obtained after the fully connected layer; the last one is the output layer, with a size of 4x1. Finally, the output result set is input into the fusion model and weighted, the two CNN models are fused, and the fusion result is output.

S5. Obtain the fault acoustic signal, extract the corresponding features and construct the spectrogram of the fault acoustic signal, input the feature set and spectrogram respectively into the trained fusion convolutional neural network model, obtain the fault identification result, and obtain the overload Gearbox troubleshooting information.

When the heavy-duty gearbox is in operation, the detection equipment starts to work. First, the fault sound signal is obtained. Secondly, the corresponding feature extraction and spectrogram construction of the fault sound signal are performed. The feature set and spectrogram are input into the trained fusion convolutional neural network model respectively to obtain the fault identification result. Obtain Heavy duty gearbox troubleshooting information. With the technology for detecting heavy-load gearbox faults provided according to embodiments of this article, heavy-load gearbox faults can be detected with high accuracy with the help of a convolutional neural network model. In addition, the heavy-duty gearbox convolutional neural network model proposed here is objective and not affected by subjective factors, so it has high reliability and stability.

Since the acoustic signal data of the heavy-duty gearbox is huge, and the convolutional neural network has strong data representation and analysis capabilities, the convolutional neural network fault diagnosis technology is used. When performing fault diagnosis, signal features need to be extracted in advance. GFCC (Gammatone Filter Cepstral Coefficient) uses Gammatone filters to simulate the human cochlear hearing model, which can more completely describe the joint distribution characteristics of the acoustic signal in the time domain and frequency domain, and Gammatone filtering The spectrum peak of the detector is flatter, which can effectively improve the energy leakage problem of signal decomposition. Therefore, 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 spectrogram that can express the signal intensity of different frequency bands of the acoustic signal is constructed, and the two models are fused to output the fusion recognition result.

In the mechanical fault diagnosis technology, professional sensor signal data is collected, and deep learning algorithms are used to establish a convolutional neural network model of heavy-duty gearbox faults. Compared with traditional manual detection and flaw detection equipment detection, fault diagnosis time is reduced. The cost of human and material resources has great advantages in the application prospects of fault diagnosis of heavy-duty gearboxes.

Example 2

As shown in Figure 5, a heavy-duty gearbox fault diagnosis system based on CNN model fusion of the present invention can be applied to heavy-duty gearbox fault diagnosis, specifically including:

The signal acquisition module is used to collect the acoustic signals of the heavy-duty gearbox to form a training data set;

The one-dimensional CNN model building module is used to extract GFCC features from the collected acoustic signals to form a feature set. The feature set is input into the pre-built initial one-dimensional convolutional neural network model for training, and the trained one-dimensional volume is obtained. Accumulate the neural network model and output the recognition result set;

The two-dimensional CNN model building module is used to superimpose the acoustic signal spectrum on the collected acoustic signal to obtain a spectrogram. The spectrogram is input into the pre-built initial two-dimensional convolutional neural network model for training, and the trained result is obtained. Two-dimensional convolutional neural network model and output a recognition result set;

The fusion CNN model building module is used to input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model into the fusion model, perform fusion training, and finally obtain the trained one-dimensional convolutional neural network model. , two-dimensional convolutional neural network model, fusion model including fusion convolutional neural network model.

The fault identification module is used to obtain the fault acoustic signal, extract the corresponding features and construct the spectrogram of the fault acoustic signal, and input the feature set and spectrogram into the trained fusion convolutional neural network model to obtain the fault identification result. , obtain heavy-duty gearbox fault diagnosis information.

Example 3

As shown in Figure 6, a heavy-duty gearbox fault diagnosis equipment based on CNN model fusion of the present invention can be applied to heavy-duty gearbox fault diagnosis. The device may include a processor and a memory. Computer instructions are stored in the memory. 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 above implementation. steps of the method described in the example, and can achieve the same technical effect as the above method.

Memory may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and/or cache memory. The device may further include other removable/non-removable, volatile/non-volatile computer system storage media. By way of example only, memory may be used to read and write to non-removable, non-volatile magnetic media (commonly referred to as "hard drives"). A program/utility having a set of (at least one) program modules, which may be stored, for example, in memory. Such program modules include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. In these examples Each of these, or some combination thereof, may include the implementation of a network environment. Program modules generally perform functions and/or methods in the described embodiments of the invention.

The processor executes various functional applications and data processing by running programs stored in the memory, for example, implementing the method provided in Embodiment 1 of the present invention.

Example 4

Embodiment 4 of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method described in the above embodiment are implemented, and the results consistent with the above method can be achieved. technical effects.

The computer storage medium in this embodiment of the present invention may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples (non-exhaustive list) of computer readable storage media include: electrical connections having one or more conductors, portable computer disks, hard drives, random access memory (RAM), read only memory (ROM), Erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. As used herein, a computer-readable storage medium may be any tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device.

A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave carrying computer-readable program code therein. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device .

Program code embodied on a computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for performing the operations of the present invention may be written in one or more programming languages, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages, or a combination thereof. A programming language, such as "C" or a similar programming language. The program code may 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 situations involving remote computers, the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer, such as through the Internet using an Internet service provider. ).

Of course, embodiments of the present invention provide a storage medium containing computer-executable instructions, and the computer-executable instructions are not limited to the above method operations, and can also perform related operations in the method provided by any embodiment of the present invention.

The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Within the spirit and principles of the present invention, any modifications, equivalent substitutions, improvements, etc. shall be included in the protection scope of the present invention.

Claims (10)

1. A heavy-duty gearbox fault diagnosis method based on CNN model fusion, which is characterized by including the following steps: Acoustic signals from heavy-duty gearboxes are collected to form a training data set; Extract GFCC features from the collected acoustic signals to form a feature set, input the feature set into the pre-built initial one-dimensional convolutional neural network model for training, obtain the trained one-dimensional convolutional neural network model, and output the recognition results set; Add the acoustic signal spectrum to the collected acoustic signal to obtain a spectrogram, input the spectrogram into the pre-constructed initial two-dimensional convolutional neural network model for training, and obtain the trained two-dimensional convolutional neural network model, and Output the recognition result set; Input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model into the fusion model for fusion training, and finally obtain the trained one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model. , fusion convolutional neural network model including fusion model; Obtain the fault sound signal, extract the corresponding features and construct the spectrogram of the fault sound signal, input the feature set and spectrogram respectively into the trained fusion convolutional neural network model, obtain the fault identification result, and obtain the heavy-duty gearbox Troubleshooting information. 2. The heavy-duty gearbox fault diagnosis method based on CNN model fusion according to claim 1, characterized in that extracting GFCC features from the collected acoustic signals to form a feature set includes the following steps: Divide the collected acoustic signals into frames; Add Hamming window processing to the framed acoustic signal; Perform discrete Fourier transform on the acoustic signal processed by the Hamming window to obtain the energy spectrum; Use Gammatone filter bank to filter the energy spectrum; Perform logarithmic compression on the filtered energy spectrum to obtain the corresponding logarithmic energy signal; Perform discrete cosine transform on the energy signal; Information entropy is introduced to measure the complexity of the energy signal after discrete cosine transform, and the threshold is set to obtain the GFCC characteristic parameters to form the required feature set. 3. The heavy-duty gearbox fault diagnosis method based on CNN model fusion according to claim 1, characterized in that, superimposing the acoustic signal spectrum on the collected acoustic signal to obtain a spectrogram, including the following steps: Divide the collected acoustic signals into frames; Add Hamming window processing to the framed acoustic signal; The fast Fourier transform is performed on the acoustic signal processed by the Hamming window to obtain the spectrum. Perform principal component analysis on the spectrum obtained by fast Fourier transform to reduce dimensionality; The spectrum after dimensionality reduction is superimposed to obtain the spectrogram. 4. The heavy-duty gearbox fault diagnosis method based on CNN model fusion according to claim 1, characterized in that: the one-dimensional convolutional neural network model includes a one-dimensional feature parameter input layer, two convolution layers, two layer max pooling layer, fully connected layer and output layer. 5. The heavy-duty gearbox fault diagnosis method based on CNN model fusion according to claim 1, characterized in that: the convolutional neural network model includes a two-dimensional image input layer, two layers of convolution layers, and two layers of maximum pooling. layer, fully connected layer and output layer. 6. The heavy-duty gearbox fault diagnosis method based on CNN model fusion according to claim 1, characterized in that: the weighted average method is used to identify the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network model. Fusion. 7. The heavy-duty gearbox fault diagnosis method based on CNN model fusion according to claim 2 or 3, characterized in that: when the collected acoustic signals are divided into frames, the time covered by each frame is 2 to 3 cycles. 8. A heavy-duty gearbox fault diagnosis system based on CNN model fusion, which is characterized by: The signal acquisition module is used to collect the acoustic signals of the heavy-duty gearbox to form a training data set; The one-dimensional CNN model building module is used to extract GFCC features from the collected acoustic signals to form a feature set. The feature set is input into the pre-built initial one-dimensional convolutional neural network model for training, and the trained one-dimensional volume is obtained. Accumulate the neural network model and output the recognition result set; The two-dimensional CNN model building module is used to superimpose the acoustic signal spectrum on the collected acoustic signal to obtain a spectrogram. The spectrogram is input into the pre-built initial two-dimensional convolutional neural network model for training, and the trained result is obtained. Two-dimensional convolutional neural network model and output a recognition result set; The fusion CNN model building module is used to input the recognition result sets of the one-dimensional convolutional neural network model and the two-dimensional convolutional neural network into the fusion model for fusion training, and finally obtain the trained one-dimensional convolutional neural network model, Fusion convolutional neural network models including two-dimensional convolutional neural network models and fusion models; The fault identification module is used to obtain the fault acoustic signal, extract the corresponding features and construct the spectrogram of the fault acoustic signal, and input the feature set and spectrogram into the trained fusion convolutional neural network model to obtain the fault identification result. , obtain heavy-duty gearbox fault diagnosis information. 9. A heavy-duty gearbox fault diagnosis equipment based on CNN model fusion, characterized in that it includes a processor and a memory, the memory stores computer instructions, and the processor is used 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 heavy-duty gearbox acoustic 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. When the program is executed by a processor, it implements the CNN model based on any one of claims 1 to 7. Steps of a converged heavy-duty gearbox fault diagnosis method.