Multi-signal input fault diagnosis method for fan main shaft based on Glow model

Abstract

This invention discloses a multi-signal input fault diagnosis method for a fan main shaft based on a Glow model. This method relates to the field of fan main shaft fault diagnosis and includes steps of acquiring data, building a Glow model, building a fault diagnosis model, training the fault diagnosis model, and diagnosing the fault. This invention addresses the issues of prolonged downtime caused by fan faults and the waste of resources resulting from inspection and maintenance, thereby improving the operational reliability of fan equipment.

Description

Technical Field

[0001] The present invention relates to the field of fan main shaft fault diagnosis, and in particular to a fan main shaft multi-signal input fault diagnosis method based on a Glow model. Background Art

[0002] Bearings are key components of wind turbine drivetrains, and drivetrains are crucial components of any rotating machine. Drivetrain failures unrelated to bearings, such as those in gears and blades, can be directly or indirectly caused by bearing failures. Planned maintenance and after-sales service have long been implemented in wind turbines. Real-time monitoring of bearing health is crucial for overall drivetrain fault diagnosis and wind turbine operation and maintenance.

[0003] A wind turbine main bearing primarily consists of an outer ring, inner ring, ball bearings, and a cage. The main shaft is connected to the turbine blades at one end and to the turbine drive system at the other. The inner ring of the main shaft is connected to the shaft, while the outer ring is connected to the cage. The ball bearings are key components for bearing rotation. Therefore, the inner ring, outer ring, and ball bearings of a wind turbine main bearing are all susceptible to failure. To ensure effective fault diagnosis of a wind turbine main shaft, it is necessary to first study the theory of wind turbine main shaft failures. During operation, the main bearing area is easily affected by external factors and can cause failures. Therefore, the failure mechanisms of wind turbine main bearings should be thoroughly analyzed to improve the accuracy of fault diagnosis.

[0004] Fault diagnosis technology originated from the sensory perception of temperature, sound, odor, etc. by operation and maintenance personnel to determine the presence of faults. Bearing vibration signals are currently the most widely used fault diagnosis method. Traditional fault diagnosis methods mainly involve signal acquisition and processing, and the establishment of a fan fault sample feature library. The main solution is to extract the fault attributes from the input signal and then diagnose by classifying the fault attributes. In recent years, based on observing the fault characteristics in the time and frequency domain by detecting partial or overall vibration of the equipment, researchers have begun to conduct in-depth research on combining artificial intelligence algorithms for fault diagnosis, using deep learning algorithms to realize fault diagnosis of fan bearings.

[0005] The emergence of artificial intelligence has brought new research directions to the field of wind turbine bearing fault diagnosis. Compared to traditional methods, deep learning models trained on vibration signals can automatically learn and extract fault characteristics, enabling "end-to-end" fault diagnosis. When fault samples are limited, fault diagnosis based on deep learning algorithms offers significant advantages.

[0006] Therefore, it is an urgent problem for those skilled in the art to propose a multi-signal input fault diagnosis method for the fan main shaft based on the Glow model to solve the difficulties existing in the existing technology. Summary of the Invention

[0007] In view of this, the present invention provides a fan main shaft multi-signal input fault diagnosis method based on the Glow model, which improves the problems of long-term shutdown caused by fan failure and waste of resources caused by inspection and maintenance, and improves the operational reliability of fan equipment.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] Data acquisition steps: obtain data sets through sensors;

[0010] Glow model construction steps: Build the Glow model using real sample data X and random variables Z;

[0011] Fault diagnosis model construction steps: Based on convolutional neural network, a multi-image input fault diagnosis model for fan main shaft is proposed;

[0012] Fault diagnosis model training steps: Use the Glow model to generate additional fan main shaft fault samples under different scenarios, increase the number of unbalanced fault category samples, and add different levels of Gaussian white noise to real samples to simulate the noise scenarios in actual fan operation. The fault sample set is input into the fault diagnosis model through the fan main shaft multi-image input. Based on the fault diagnosis accuracy, a trained fan main shaft multi-image input fault diagnosis model is obtained;

[0013] Fault diagnosis steps: Use the trained fan main shaft multi-image input fault diagnosis model to diagnose the fan main shaft and determine whether the fan main shaft is faulty.

[0014] In the above method, optionally, in the data acquisition step, a data set is collected by an acceleration sensor, and the data set includes but is not limited to a normal state signal of a wind turbine bearing, a ball body fault vibration signal, an inner ring fault vibration signal, and an outer ring fault signal.

[0015] The above method optionally includes the real sample data X and random variable Z in the Glow model construction step, where Z obeys a known simple prior distribution π(Z) and the sample data X obeys a complex distribution p(X). There exists a transformation function f that satisfies the mapping from Z to X.

[0016] f:Z→X (1)

[0017] For every sampling point in π(Z), there can be a new sample point corresponding to it in p(X) to obtain the generated sample. In the standardized flow model, the prior distribution π(Z) of the random variable Z is usually Gaussian distribution. The standardized flow converts the simple distribution into a complex distribution by applying a series of reversible transformation functions, and repeatedly replaces the new variable according to the variable substitution theorem to finally obtain the probability distribution of the final target variable.

[0018] Alternatively, the generation process of the standardized flow model can be defined by the following formula:

[0019] Z~π(Z) (2)

[0020] X=g θ (Z) (3)

[0021] In the formula, Z represents the latent variable; p θ (Z) is the sample distribution of latent variable Z; g θ Is a reversible function, the latent variable Z can be expressed as Among them, f θ It consists of a series of conversion functions: θ is the generative model parameter, and the relationship between sample X and latent variable Z0 can be written as:

[0022]

[0023] By outputting x until tracing back to the initial distribution z, the model probability density function given a sample data x can be expressed as:

[0024]

[0025] The training loss function of the flow-based generative model is the negative log-likelihood on the training dataset:

[0026]

[0027] In the above method, optionally, in the Glow model construction step, the Glow model consists of a series of repeated layers named scales, each scale includes a squeeze function and a flow step, followed by a split function; the split function divides the input into two equal parts in the channel dimension; half of the parts enter the subsequent layer, and the other half enters the loss function; the split is to reduce the impact of gradient disappearance, which occurs when the model is trained in an end-to-end manner; the flow step includes three parts: an activation constant layer, a 1x1 reversible convolution layer, and an affine coupling layer.

[0028] In the above method, an optional activation constant layer is used for activation normalization, which uses the scale and bias parameters of each channel to perform an affine transformation on the activation, similar to batch normalization, initializing these parameters so that given a small batch of initial data, the subsequent behavior of each channel has zero mean and unit variance. After initialization, the scale and bias are treated as regular trainable parameters independent of the data;

[0029] The 1x1 reversible convolution layer is used to reverse the order of the channels, where the weight matrix is initialized to a random rotation matrix, and the number of input and output channels of the convolution layer is the same;

[0030] The affine coupling layer builds a bijective function by superimposing a series of simple bijections. In each simple bijection, part of the input vector is updated using a simple inverted function, but it depends on the remainder of the input vector in a complex way. The affine coupling layer can be divided into three parts: zero initialization, splitting and connection, and permutation.

[0031] In the above method, optionally, in the fault diagnosis model construction step, the fan main shaft multi-image input fault diagnosis model realizes the fault diagnosis function by learning the fault features of multiple image inputs: the input is 2 grayscale images of size 28*28, each input is convolved through multiple convolution layers, the output is connected to the fully connected layer, and the multiplication layer is used to multiply the inputs from the two fully connected layers; the output is the classification output layer, and the activation function is softmax.

[0032] In the above method, optionally, in the fault diagnosis model training step, the fault samples are classified into 19 categories according to the fault type, fault location and damage diameter.

[0033] In the above method, optionally, the vibration signal in the fan main shaft vibration data set is a one-dimensional time series signal. The one-dimensional time series signals from the fan end and the drive end of the fan main shaft are converted into two-dimensional image signals respectively, and fault samples are generated and fault diagnosis is performed through a two-dimensional convolutional network; the original one-dimensional time domain signal needs to be normalized before sample construction.

[0034] It can be seen from the above technical solution that compared with the existing technology, the present invention provides a multi-signal input fault diagnosis method for a wind turbine main shaft based on the Glow model, and the beneficial effects include: through online fault diagnosis and analysis, it can help operation and maintenance personnel discover early signs of wind turbine failure, determine the fault type and degree of failure, so that the wind farm can adopt an effective maintenance plan, avoid the occurrence of major safety accidents and the waste of resources caused by inspection and maintenance, optimize the equipment inspection and maintenance work plan, and improve the operational reliability of wind turbine equipment; it can effectively avoid long-term shutdowns caused by wind turbine failures, reduce power generation losses, and can reduce the failure rate and maintenance costs of gearboxes by at least 80%, increase the utilization rate of wind turbine lubricating oil by about 10%, and save about 9% of operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0036] Figure 1Flowchart of the blower main shaft multi-signal input fault diagnosis method based on the Glow model provided by the present invention;

[0037] Figure 2 This is a structural diagram of the Glow model provided by the present invention;

[0038] Figure 3 This is a flow step structure diagram of the Glow model provided by the present invention;

[0039] Figure 4 A diagram of the fault sample construction process provided by the present invention;

[0040] Figure 5 This is a multi-signal input fault diagnosis model diagram provided by the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] In this application, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. The terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or apparatus comprising the element.

[0043] Reference Figure 1 As shown, the present invention discloses a fan main shaft multi-signal input fault diagnosis method based on the Glow model, comprising the following steps:

[0044] A multi-signal input fault diagnosis method for a fan main shaft based on a Glow model is characterized by comprising the following steps:

[0045] Data acquisition steps: obtain data sets through sensors;

[0046] Glow model construction steps: Build the Glow model using real sample data X and random variables Z;

[0047] Fault diagnosis model construction steps: Based on convolutional neural network, a multi-image input fault diagnosis model for fan main shaft is proposed;

[0048] Fault diagnosis model training steps: Use the Glow model to generate additional fan main shaft fault samples under different scenarios, increase the number of unbalanced fault category samples, and add different levels of Gaussian white noise to real samples to simulate the noise scenarios in actual fan operation. The fault sample set is input into the fault diagnosis model through the fan main shaft multi-image input. Based on the fault diagnosis accuracy, a trained fan main shaft multi-image input fault diagnosis model is obtained;

[0049] Fault diagnosis steps: Use the trained fan main shaft multi-image input fault diagnosis model to diagnose the fan main shaft and determine whether the fan main shaft is faulty.

[0050] Furthermore, in the data acquisition step, a data set is collected by an acceleration sensor, and the data set includes but is not limited to a normal state signal of a wind turbine bearing, a vibration signal of a ball body fault, a vibration signal of an inner ring fault, and a signal of an outer ring fault.

[0051] Further, such as Figure 2 As shown, in the Glow model construction step, the real sample data X and the random variable Z, where Z obeys the known simple prior distribution π(Z), and the sample data X obeys the complex distribution p(X), there exists a transformation function f that satisfies the establishment of a mapping from Z to X

[0052] f:Z→X (1)

[0053] For every sampling point in π(Z), there can be a new sample point corresponding to it in p(X) to obtain the generated sample. In the standardized flow model, the prior distribution π(Z) of the random variable Z is usually Gaussian distribution. The standardized flow converts the simple distribution into a complex distribution by applying a series of reversible transformation functions, and repeatedly replaces the new variable according to the variable substitution theorem to finally obtain the probability distribution of the final target variable.

[0054] Furthermore, the generation process of the standardized flow model can be defined by the following formula:

[0055] Z~π(Z) (2)

[0056] X=g θ (Z) (3)

[0057] In the formula, Z represents the latent variable; p θ (Z) is the sample distribution of latent variable Z; g θ Is a reversible function, the latent variable Z can be expressed as Among them, f θ It consists of a series of conversion functions: θ is the generative model parameter, and the relationship between sample X and latent variable Z0 can be written as:

[0058]

[0059] By outputting x until it is traced back to the initial distribution z , given a sample data x, the model probability density function can be expressed as:

[0060]

[0061] The training loss function of the flow-based generative model is the negative log-likelihood on the training dataset:

[0062]

[0063] Further, such as Figure 3 In the Glow model construction steps shown, the Glow model consists of a series of repeated layers named scales. Each scale includes a squeeze function and a flow step, followed by a split function. The split function divides the input into two equal parts in the channel dimension. Half of the parts enter the subsequent layer, and the other half enters the loss function. The split is to reduce the impact of gradient disappearance, which occurs when the model is trained in an end-to-end manner. The flow step contains three parts: an activation constant layer, a 1x1 reversible convolution layer, and an affine coupling layer.

[0064] Furthermore, an activation constant layer is used for activation normalization, which performs an affine transformation on the activation using the scale and bias parameters of each channel, similar to batch normalization, initializing these parameters so that given a small batch of initial data, the subsequent actions of each channel have zero mean and unit variance. After initialization, the scale and bias are treated as regular trainable parameters independent of the data;

[0065] The 1x1 reversible convolution layer is used to reverse the order of the channels, where the weight matrix is initialized to a random rotation matrix, and the number of input and output channels of the convolution layer is the same;

[0066] The affine coupling layer builds a bijective function by superimposing a series of simple bijections. In each simple bijection, part of the input vector is updated using a simple inverted function, but it depends on the remainder of the input vector in a complex way. The affine coupling layer can be divided into three parts: zero initialization, splitting and connection, and permutation.

[0067] Further, such as Figure 4As shown in the figure, in the fault diagnosis model construction steps, the fan main shaft multi-image input fault diagnosis model realizes the fault diagnosis function by learning the fault features of multiple image inputs: the input is two grayscale images of size 28*28, each input is convolved through multiple convolution layers, the output is connected to the fully connected layer, and the multiplication layer is used to multiply the inputs from the two fully connected layers; the output is the classification output layer, and the activation function is softmax.

[0068] Furthermore, in the fault diagnosis model training step, the fault samples are divided into 19 categories according to the fault type, fault location and damage diameter. The classification of fan main shaft faults is shown in Table 1:

[0069] Table 1 Fan main shaft fault classification

[0070]

[0071] Furthermore, the vibration signals in the fan main shaft vibration data set are one-dimensional time series signals. The one-dimensional time series signals from the fan end and the drive end of the fan main shaft are converted into two-dimensional image signals respectively, and fault samples are generated and fault diagnosis is performed through a two-dimensional convolutional network; the original one-dimensional time domain signal needs to be normalized before sample construction.

[0072] Furthermore, during the fault diagnosis model training step, to verify the performance of the generated model, we use image quality metrics such as Maximum Mean Difference (MMD), Peak Signal-to-Noise Ratio (PSNR), and Feature Similarity (FSIM) to evaluate the authenticity of the generated fault samples. Maximum Mean Difference uses the sum of the projections of each image to determine the distribution difference between the two images; smaller values indicate smaller image distribution differences. Peak Signal-to-Noise Ratio is a relative value based on the mean squared error between the original and generated images and the square of the maximum possible signal value of the image. Larger PSNR values indicate higher quality generated samples. Feature Similarity uses feature similarity to evaluate image quality; higher values indicate greater similarity.

[0073] Further, refer to Figure 5 As shown, in the fault diagnosis step, the multi-signal input fault diagnosis model determines the fault category by image quality.

[0074] In a specific embodiment, the model disclosed in the present invention is simulated.

[0075] Table 2 evaluates the quality of generated fault samples under different noise intensities. Based on the three metrics, the Glow model's image quality improves when supplementing the original fault samples with the noise. The generated drive-end fault samples also have better image quality than the fan-end fault samples. Comparing different image quality metrics, we found no significant degradation in the quality of model-generated fault samples under noise interference. These results demonstrate that the proposed model has excellent capabilities in generating fault samples for fan main shafts.

[0076] Table 2 Quality evaluation of generated fault samples

[0077]

[0078] 2. The Glow model was used to generate new samples based on two sets of fan main shaft vibration data, supplementing the number of samples in the unbalanced fault category. The balanced fault sample set was then passed through the fan main shaft multi-input fault diagnosis model to obtain the fault diagnosis accuracy. The sample set construction for the sample imbalance scenario is shown in Table 3. The number of samples in the balanced category is 2000 per category, and the number of samples in the unbalanced category is 1000 per category. Table 4 shows the fault diagnosis accuracy of different sample sets for the sample imbalance scenario. When the sample imbalance is relatively low, the model can diagnose fan main shaft faults with 100% accuracy. As the sample imbalance increases, the fault diagnosis accuracy of the sample set decreases. However, the model can still effectively diagnose the presence of fan main shaft faults, indicating that the proposed model has good fault diagnosis capabilities in the sample imbalance scenario.

[0079] Table 3 Sample set construction under sample imbalance scenario

[0080]

[0081] Table 4 Fault diagnosis results of different sample sets under sample imbalance scenario

[0082]

[0083] 3. To verify the fault diagnosis performance of the proposed method in complex scenarios, we constructed sample sets based on different numbers of training samples. The number of samples was supplemented using the Glow model until the number of samples in each sample set was the same. The diagnostic performance of the proposed method was then tested on the same test set. The sample set construction for the sample-deficient scenario is shown in Table 5.

[0084] Table 6 shows the fault diagnosis accuracy of the original and supplemented sample sets in the sample-sufficiency scenario. As shown in the figure, insufficient samples can lead to a decrease in the fault diagnosis accuracy of the sample set. By supplementing the sample set with fault samples generated by the generative model, the fault diagnosis accuracy is improved. Although the fault diagnosis effect of the sample set supplemented with fault samples generated by the deep learning model is not satisfactory, it does verify that the proposed fault diagnosis model has certain fault diagnosis performance when samples are insufficient, and the fan main shaft fault samples generated by Glow can effectively improve the fault diagnosis accuracy.

[0085] Table 5 Sample set construction in the sample shortage scenario

[0086]

[0087] Table 6 Fault diagnosis results of the original sample set and the supplementary sample set in the sample shortage scenario

[0088]

[0089] 4. Considering that the collection of vibration data during actual operation of the fan may be interfered by noise, in order to verify the effectiveness of the proposed fault diagnosis model in practical applications, different degrees of Gaussian white noise are added to the real sample data to simulate the noise scenario during actual operation of the fan.

[0090] Table 7 shows the fault diagnosis accuracy of different sample sets under noise interference. When the number of fault samples is balanced and sufficient, the fault diagnosis accuracy of the sample set is significantly higher than that of other sample sets. When the number of samples is insufficient and unbalanced, the fault diagnosis accuracy of the sample set decreases. While the fault diagnosis accuracy of individual sample sets is relatively low under specific scenarios, the fault diagnosis accuracy of each sample set remains above 97.5%. This study demonstrates that the proposed model has excellent performance in diagnosing fan main shaft faults under noisy conditions.

[0091] Table 7 Fault diagnosis results of different sample sets under different noise intensities

[0092]

[0093] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0094] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-signal input fault diagnosis method for fan main shaft based on Glow model, characterized in that: The following steps are involved: Data acquisition steps: obtain data sets through sensors; Glow model construction steps: Build the Glow model using real sample data X and random variables Z; Fault diagnosis model construction steps: Based on convolutional neural network, a multi-image input fault diagnosis model for fan main shaft is proposed; Fault diagnosis model training steps: Use the Glow model to generate additional fan main shaft fault samples under different scenarios, increase the number of unbalanced fault category samples, and add different levels of Gaussian white noise to real samples to simulate the noise scenarios in actual fan operation. The fault sample set is input into the fault diagnosis model through the fan main shaft multi-image input. Based on the fault diagnosis accuracy, a trained fan main shaft multi-image input fault diagnosis model is obtained; Fault diagnosis steps: Use the trained fan main shaft multi-image input fault diagnosis model to diagnose the fan main shaft and determine whether the fan main shaft is faulty; In the Glow model construction step, the real sample data X and the random variable Z, where Z obeys the known simple prior distribution π(Z), and the sample data X obeys the complex distribution p(X), and there exists a transformation function f that satisfies the establishment of a mapping from Z to X f:Z→X (1) For every sampling point in π(Z), there can be a new sample point corresponding to it in p(X) to obtain the generated sample. In the standardized flow model, the prior distribution π(Z) of the random variable Z is usually Gaussian distribution. The standardized flow transforms the simple distribution into a complex distribution by applying a series of reversible transformation functions. According to the variable substitution theorem, the new variable is repeatedly replaced to finally obtain the probability distribution of the final target variable. The generation process of the normalized flow model is defined by the following formula: Z~π(Z) (2) X=g θ (Z) (3) In the formula, Z represents the latent variable; p θ (Z) is the sample distribution of latent variable Z; g θ Is a reversible function, the latent variable Z can be expressed as Among them, f θ It consists of a series of conversion functions: θ is the generative model parameter, and the relationship between sample X and latent variable Z0 is: By outputting x until tracing back to the initial distribution z, the model probability density function given a sample data x can be expressed as: The training loss function of the flow-based generative model is the negative log-likelihood on the training dataset: In the Glow model construction step, the Glow model consists of a series of repeated layers named scales. Each scale includes a squeeze function and a flow step, followed by a split function. The split function divides the input into two equal parts in the channel dimension; half of the parts enter the subsequent layer, and the other half enters the loss function. The flow step consists of three parts: an activation constant layer, a 1x1 reversible convolution layer, and an affine coupling layer.

2. The fan main shaft multi-signal input fault diagnosis method based on the Glow model according to claim 1 is characterized in that: In the data acquisition step, a data set is collected through an acceleration sensor. The data set includes but is not limited to a normal state signal of a wind turbine bearing, a vibration signal of a ball bearing fault, a vibration signal of an inner ring fault, and a signal of an outer ring fault.

3. The fan main shaft multi-signal input fault diagnosis method based on the Glow model according to claim 1 is characterized in that: The activation constant layer is used for activation normalization, which uses the scale and bias parameters of each channel to perform an affine transformation on the activation. Similar to batch normalization, these parameters are initialized so that the subsequent action of each channel has zero mean and unit variance given a small batch of initial data. After initialization, the scale and bias are treated as regular trainable parameters independent of the data; The 1x1 reversible convolution layer is used to reverse the order of the channels, where the weight matrix is initialized to a random rotation matrix, and the number of input and output channels of the convolution layer is the same; The affine coupling layer builds a bijective function by superimposing a series of simple bijections. In each simple bijection, part of the input vector is updated using a simple inverted function, but it depends on the remainder of the input vector in a complex way. The affine coupling layer can be divided into three parts: zero initialization, splitting and connection, and permutation.

4. The fan main shaft multi-signal input fault diagnosis method based on the Glow model according to claim 1 is characterized in that: In the fault diagnosis model construction step, the fan main shaft multi-image input fault diagnosis model realizes the fault diagnosis function by learning the fault features of multiple image inputs: the input is two grayscale images of size 28*28, each input is convolved through multiple convolution layers, the output is connected to the fully connected layer, and the multiplication layer is used to multiply the inputs from the two fully connected layers; the output is the classification output layer, and the activation function is softmax.

5. The fan main shaft multi-signal input fault diagnosis method based on the Glow model according to claim 1 is characterized in that: In the fault diagnosis model training step, fault samples are divided into 19 categories according to fault type, fault location and damage diameter.

6. The fan main shaft multi-signal input fault diagnosis method based on the Glow model according to claim 5 is characterized in that: The vibration signals in the fan main shaft vibration dataset are one-dimensional time series signals. The one-dimensional time series signals from the fan end and the drive end of the fan main shaft are converted into two-dimensional image signals respectively. Fault samples are generated and fault diagnosis is performed through a two-dimensional convolutional network. Before sample construction, the original one-dimensional time domain signal needs to be normalized.

Images