Tunnel boring machine main bearing fault simulation test bench and fault diagnosis method

Abstract

The invention discloses a tunnel boring machine main bearing fault simulation test bench and a fault diagnosis method, the tunnel boring machine main bearing fault simulation test bench comprises a rack and a main bearing rotatably arranged on the rack, the front end and the rear end of the main bearing are respectively provided with a first outer ring and a second outer ring, and one side of the second outer ring facing a main push roller is provided with a clamping groove; a fault simulation board is clamped in the clamping groove and is in interference fit with the second outer ring, and a fault detection module is installed on the outer side of the second outer ring. The rear end of the fault simulation board is provided with a screw hole, and a bolt passes through the second outer ring and extends into the screw hole so as to drag the fault simulation board in the clamping groove. Compared with the prior art, by arranging the fault simulation board and the second outer ring which are detachably connected, the difficulty of fault implantation can be greatly reduced, the manufacturing cost can be reduced, and meanwhile, the simulation test bed can more flexibly control the form, degree and number of faults.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing diagnosis, and in particular to a fault simulation test bench for a main bearing of a tunnel boring machine and a fault diagnosis method. Background Art

[0002] At present, most of the fault tests for bearings are for small high-speed bearings, and there are very few fault test benches for large low-speed bearings such as shield machine main bearings. In the test of small high-speed bearings, the simulation of faults is to directly replace the entire outer ring and implant different types of faults in multiple outer rings. However, for large-sized bearings, considering the difficulty and huge cost of manufacturing, if multiple faults are to be simulated, it is necessary to manufacture outer rings corresponding to the number of fault types, which is difficult to achieve. Summary of the invention

[0003] The invention provides a tunnel boring machine main bearing fault simulation test bench and a fault diagnosis method, which are used to solve the problem that the existing fault test bench is difficult to simulate various faults of large-size bearings.

[0004] The present invention provides a tunnel boring machine main bearing fault simulation test bench, comprising a frame and a main bearing rotatably arranged on the frame, wherein a first outer ring and a second outer ring are respectively arranged at the front end and the rear end of the main bearing, a slot is arranged on the side of the second outer ring facing the main push roller, a fault simulation board is clamped in the slot, the fault simulation board and the second outer ring are interference fit, and a fault detection module is installed on the outer side of the second outer ring at a position corresponding to the fault simulation board; a screw hole is arranged at the rear end of the fault simulation board, and a bolt passes through the second outer ring and extends into the screw hole to pull the fault simulation board into the slot; a loading device and an oil circulation device are arranged on the frame, the oil circulation device is used to provide lubricating fluid to the main bearing, and the loading device is used to drive the main bearing to rotate and provide axial thrust to the main bearing.

[0005] Preferably, the second outer ring and the first outer ring are fixed by bolts.

[0006] Preferably, a gap is provided between the fault simulation board and the bottom of the card slot.

[0007] Preferably, there are three fault simulation panels, which are distributed at three o'clock, six o'clock and nine o'clock on the second outer circle.

[0008] Preferably, the loading device comprises: a servo motor, a transmission shaft, a gear and an inner gear ring, the inner gear ring is fixed to the inner side of the main bearing, the gear is meshed with the inner gear ring, the gear is fixed on the transmission shaft, and the transmission shaft is fixedly connected to the output shaft of the servo motor.

[0009] The present invention also provides a method for diagnosing a fault of a main bearing of a tunnel boring machine, comprising the above-mentioned tunnel boring machine main bearing fault simulation test bench, and further comprising the following steps:

[0010] S1: The vibration data x of the main bearing is obtained by the fault detection module n (t), each fault detection module provides a channel of vibration data x n (t), and then adjust the data segment length in the data preprocessing process according to the fault characteristic frequency, and then normalize the data of each channel to obtain data J; where x n (t) represents the vibration data of the nth channel at time t;

[0011] S2: Use the dilated wide kernel convolution to preliminarily process the data J to obtain the data Y, perform batch normalization on the data Y to obtain the data G, then introduce nonlinear components into the data G and perform maximum pooling to obtain the channel feature U of the channel n ;

[0012] S3: Channel feature U n The feature vector U is obtained by element-by-element addition. Then the feature vector U is calculated through the full average pooling layer (GAP), the fully connected layer (FC), the BN and the ReLU activation function to obtain the feature Z. Finally, the feature Z is input into the fully connected layer to obtain the attention weight W of each channel. j , the sum of the attention weights of each channel is 1, that is,

[0013] S4: The calculated attention weight W j The feature vector S is assigned to obtain the weighted feature vector V, and then the weighted feature vector V is fused, and finally the fused feature is input into the final full connection to realize the mapping process of the fault type. The feature vector V is calculated by the following formula (3):

[0014]

[0015] Among them, W j is the attention weight of the jth channel, the feature vector S is the feature value obtained after the data Y is input into a series of convolutional layers, and s j is the eigenvalue of the jth channel.

[0016] Preferably, step S1 specifically includes the following steps:

[0017] S11, based on prior knowledge, solve the fault characteristic frequency of the corresponding area, combined with the sampling frequency f s , the required data segment length l is calculated by the following formula (4):

[0018]

[0019] Where, f is the fault characteristic frequency, f s is the sampling frequency, N is the number of cycles in the data segment, and the range of N is 5 to 10;

[0020] Calculate the fault characteristic frequency f according to the following table (1):

[0021] Table 1 Calculation formula for fault characteristic frequency at various locations

[0022]

[0023] Among them, D m The pitch diameter of the main push raceway roller, D a D is the pitch diameter of the auxiliary roller raceway. r is the radial raceway roller pitch diameter, d m Main push roller diameter, d a is the diameter of the auxiliary roller, d r is the radial roller diameter, f r is the inner ring rotation frequency, n m is the number of main push roller pairs, n a is the number of auxiliary rollers, n r is the number of radial rollers;

[0024] S12, combining the sliding window method to segment the time domain data according to the above-required sample length l:

[0025]

[0026] Where L is the total length of the data, l is the length of the data segment, stride is the step size, and num represents the number of data segments after segmentation;

[0027] S13, then normalize the data of each channel using formula (6):

[0028]

[0029] in Indicates the kth data segment in the nth channel data, that is, data J, represents the kth data segment of the nth channel, where k∈[0,num], Indicates the maximum value in the corresponding data segment. Indicates the minimum value in the corresponding data segment.

[0030] Preferably, step S2 specifically includes the following steps:

[0031] S21, firstly use the dilated wide kernel convolution to perform preliminary processing on the data J, and the calculation formula (7) of the dilated wide kernel convolution is as follows:

[0032]

[0033] where Y n [i] is the output at position i, which is data Y; W[m] is the weight in the convolution kernel, corresponding to the position m of the kernel; l is the dilation rate, which determines the spacing between the elements of the convolution kernel; M is half the size of the convolution kernel, For data J, is the [i+l·m]th data point of the corresponding data segment;

[0034] S22, output result Y n [i] Perform batch normalization, the formula is as follows:

[0035] Calculate the mean μ of the entire batch n :

[0036] Calculate the variance of the entire batch

[0037] Normalization:

[0038] Scaling and translation (reparameterization):

[0039] Among them, ∈ is a small number to prevent the denominator from being zero; γ and β are learnable parameters that allow the network to learn to recover any potentially useful feature distribution; x n [i] represents the data point of the nth channel, Represents the normalized data point, y n [i] represents the output of the data after batch normalization, which is data G;

[0040] S23, nonlinear components are introduced into the data through formula (8) and maximum pooling is performed:

[0041] U n =maxpool(ReLU(X)) (8)

[0042] Among them, U n It is the eigenvalue extracted from the nth channel, and ReLU(X) means the ReLU function is applied to the data G.

[0043] Preferably, step S3 specifically includes the following steps:

[0044] S31, the eigenvalues ​​obtained from each channel are added element by element using formula (9):

[0045] U=U1+U2+U3 (9)

[0046] Among them, the feature vector U integrates the information of multiple channels;

[0047] S32, U is calculated through the full average pooling layer (GAP), the fully connected layer (FC), the BN and the ReLU activation function to obtain the feature Z. The mathematical model of Z can be further expressed as:

[0048] Z=δ(ψ(U gap )) (10)

[0049] Among them, U gap represents the result of U after global pooling, ψ represents the BN batch normalization process, and δ represents the activation function ReLU;

[0050] It is then input into the fully connected layer to further compress the feature dimension:

[0051] Z=Dense(C / r,C) (11)

[0052] Where Dense represents the fully connected process, r is the feature compression rate, C is the number of input features, and C is usually r = 0.5;

[0053] S33, the compressed feature Z is input into three groups of fully connected layers again to generate three groups of attention weights, and then the weights are normalized through SoftMax. The specific process is as follows:

[0054]

[0055] Among them, A s , B s , C s is the sth row of the feature matrix, and the a generated by this part s 、b s 、c s Then is the attention weight of the three channels, formula (12), (13) and (14) satisfy the following relationship: a s +b s +c s =1.

[0056] Preferably, step S4 specifically includes the following steps:

[0057] S41, by convolving the feature Y obtained by the hollow wide kernel in step 2 n [i] Input a series of convolutional layers to further extract deep features:

[0058] S i =Convs(16,128) (15)

[0059] Among them, the number of feature channels obtained by the dilated wide kernel convolution is 16. After a series of convolutions, the number of channels becomes 128. i is the feature vector S;

[0060] S42, convolve the feature vector S obtained in step S41 i Multiply it with the attention weight obtained in step S3 to obtain the weighted feature vector V: V = a s ·S1+b s ·S2+c s ·S3.

[0061] Compared with the prior art, the present invention can greatly reduce the difficulty of fault implantation by providing a detachably connected fault simulation board and a second outer ring, which is beneficial to reducing the production cost, and also enables the simulation test bench to more flexibly control the form, degree and number of faults. On the other hand, the present invention first adds and fuses the data of the three channels, and then generates three attention weights through a fully connected layer corresponding to the number of channels. The fully connected layer reduces the channels with small correlation with the fault during the training process, so that the fault diagnosis method pays more attention to the channels with large correlation with the fault, alleviating the risk of overfitting caused by information redundancy in the multi-channel fault diagnosis process. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 It is a structural schematic diagram of the present invention;

[0064] Figure 2 It is a structural schematic diagram of the present invention without the box plate of the oil tank;

[0065] Figure 3 It is a partial structural schematic diagram of the present invention;

[0066] Figure 4 is a schematic structural diagram of the second outer ring of the present invention;

[0067] Figure 5 It is a structural schematic diagram of the fault simulation board of the present invention;

[0068] Figure 6 It is a schematic structural diagram of the main bearing of the present invention.

[0069] Reference numerals:

[0070] 1. Frame, 2. Main bearing, 3. First outer ring, 4. Second outer ring, 5. Fault simulation board, 6. Loading device, 7. Oil circulation device, 8. Fault detection module, 41. Slot, 51. Screw hole, 52. Notch, 61. Servo motor, 62. Transmission shaft, 63. Gear, 64. Inner ring gear, 65. Electric push rod, 66. Pressure sensor, 67. Roller, 68. Force ring, 69. Linear bearing, 651. Telescopic rod, 71. Oil outlet pipe. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0072] Refer to the attached Figure 1 , Attachment Figure 3 and attached Figure 4This embodiment provides a fault simulation test bench for the main bearing 2 of a tunnel boring machine, including a frame 1 and a main bearing 2 rotatably arranged on the frame 1, wherein the front and rear ends of the main bearing 2 are respectively provided with a first outer ring 3 and a second outer ring 4, and a slot 41 is provided on the side of the second outer ring 4 facing the main push roller, and a fault simulation board 5 is stuck in the slot 41, and the fault simulation board 5 is used to simulate fault defects, for example, a crack is engraved on the fault simulation board 5, and a vibration fault occurs when the roller slides over. The fault simulation board 5 and the second outer ring 4 are interference fit, and a fault detection module 8 is installed on the outer side of the second outer ring 4 at a position corresponding to the fault simulation board 5; one implementation of the fault detection module 8 is an acceleration sensor, and the acceleration signal is monitored by the acceleration sensor to complete the simulation of the vibration signal of the main bearing 2. The rear end of the fault simulation board 5 is provided with a screw hole 51, and the bolt passes through the second outer ring 4 and extends into the screw hole 51 to pull the fault simulation board 5 into the slot 41. The setting uses the pre-tightening force of the bolt to pull the fault simulation board 5 toward the second outer ring 4, so that the fault simulation board 5 is close to the second outer ring 4; the frame 1 is provided with a loading device 6 and an oil circulation device 7, the oil circulation device 7 is used to provide lubricating fluid to the main bearing 2, and the loading device 6 is used to drive the main bearing 2 to rotate and provide axial thrust to the main bearing 2. The present invention can greatly reduce the difficulty of fault implantation by providing a detachably connected fault simulation board 5 and the second outer ring 4, which is conducive to reducing the manufacturing cost, and also enables the simulation test bench to more flexibly control the form, degree and number of faults. The key point of fault simulation is the characterization of the fault impact. The present invention adopts the following means to reduce the impact attenuation caused by the detachable connection between the fault simulation board 5 and the second outer ring 4. The first is to insert the fault simulation board 5 into the second outer ring 4 by interference fit. Specifically, when installing the fault simulation board 5, liquid nitrogen is required for freezing assembly. Secondly, on the basis of interference fit assembly, a pre-tightening force is provided by means of bolts to increase the fitting stiffness of the fault simulation plate 5 and the second outer ring 4 .

[0073] A mounting method of the second outer ring 4 and the first outer ring 3: the second outer ring 4 and the first outer ring 3 are fixed by bolts. This structural design facilitates the disassembly of the second outer ring 4, thereby facilitating the installation and replacement of the fault simulation board 5.

[0074] As another embodiment of the present invention: Figure 5 A notch 52 is provided between the fault simulation board 5 and the bottom of the slot 41. Specifically, the notch 52 is provided on the fault simulation board 5. With this structural design, when disassembling, the fault simulation board 5 can be removed from the second outer ring 4 by simply prying at the notch 52 of the fault simulation board 5 with a flat-blade screwdriver.

[0075] As another embodiment of the present invention: Figure 2, there are three fault simulation boards 5, which are distributed at 3 o'clock, 6 o'clock and 9 o'clock on the second outer ring 4, and the corresponding three fault detection modules 8 are also distributed at 3 o'clock, 6 o'clock and 9 o'clock on the second outer ring 4. The positions on the second outer ring 4 are divided according to time, and the three fault simulation boards 5 are distributed at 3 o'clock, 6 o'clock and 9 o'clock. This is because of the influence of axial force, the outer ring main push part raceway (raceway on the second outer ring 4) and roller are loaded, the contact stress is large, and the outer ring is fixed, the load point is single, and the possibility of contact fatigue damage is greater than that of radial raceway and auxiliary push raceway (raceway on the first outer ring 3). Secondly, affected by the overturning force generated by factors such as uneven soil quality and shield body steering, the main raceway (second outer ring 4) will have large peak stresses at 3 o'clock, 6 o'clock and 9 o'clock, so the probability of damage such as crushing, cracking and pitting is relatively higher. At the same time, in order to make the selected points more representative as much as possible, the fault implantation points are set at the 3 o'clock, 6 o'clock and 9 o'clock positions on the second outer ring 4.

[0076] As another embodiment of the present invention: the second outer ring 4 is located at a side of the main bearing 2 away from the loading device 6 . This structural design facilitates the removal of the second outer ring 4 .

[0077] An implementation of the loading device 6: The loading device 6 includes: a servo motor 61, a transmission shaft 62, a gear 63 and an inner gear ring 64, the inner gear ring 64 is fixed to the inner side of the main bearing 2, the gear 63 is meshed with the inner gear ring 64, the gear 63 is fixed on the transmission shaft 62, and the transmission shaft 62 is fixedly connected to the output shaft of the servo motor 61. The servo motor 61 drives the transmission shaft 62 to rotate, thereby driving the gear 63 to rotate, and the rotation of the gear 63 drives the inner gear ring 64 to rotate, thereby driving the main bearing 2 to rotate.

[0078] As another embodiment of the present invention: the loading device 6 also includes: an electric push rod 65, a pressure sensor 66, a roller 67 and a force ring 68, the force ring 68 is fixed at the front end of the main bearing 2, the force ring 68 is arranged on the inner side of the first outer ring 3, the roller 67 rolls along the force ring 68, the roller 67 is rotatably set on the telescopic rod 651 of the electric push rod 65, and the telescopic rod 651 is installed on the frame 1 through a linear bearing 69. This structural design can push the roller 67 more smoothly, so that it applies a force on the force ring 68, thereby applying an axial load to the main bearing 2.

[0079] As another embodiment of the present invention, the oil circulation device 7 includes an oil pump, an oil outlet pipe 71 of the oil pump is connected to the first outer ring 3 and the second outer ring 4, an oil inlet pipe of the oil pump extends into the oil tank, and the main bearing 2 is located in the oil tank. The oil outlet pipe 71 provides lubricating oil to the main bearing 2, and the excess lubricating oil returns to the oil tank from the main bearing 2. This structural design can ensure that there is sufficient lubricating oil in the main bearing 2 at all times.

[0080] The present invention uses interference fit to install the fault simulation plate 5 on the second outer ring 4, which can not only simulate multiple fault types, but also simulate multiple fault locations and even multiple composite fault types, greatly increasing the freedom of fault simulation. Users can arbitrarily combine multiple fault types as needed. Secondly, the main push roller is in contact with the fault simulation plate 5, and the vibration is also transmitted from the fault simulation plate 5 to the second outer ring 4, and then to the fault detection module 8. The present invention greatly improves the fitting stiffness by applying bolt preload and using interference fit, thereby largely eliminating the vibration attenuation caused by the detachable design.

[0081] The present invention also provides a method for diagnosing a fault of a main bearing of a tunnel boring machine, comprising the above-mentioned tunnel boring machine main bearing fault simulation test bench, and further comprising the following steps:

[0082] S1: The fault detection module 8 obtains the vibration data x of the main bearing n (t), each fault detection module 8 provides a channel of vibration data x n (t), and then adjust the data segment length in the data preprocessing process according to the fault characteristic frequency, and then n (t) Perform normalization to obtain intermediate data J; x n (t) represents the vibration data of the nth channel at time t. In the present invention, there are three fault detection modules 8, that is, there are three channels. Since the shield main bearing is a large, low-speed, heavy-loaded bearing, its rotation speed is usually lower than 5rpm, and its rotation monitoring data often has the problems of large data volume and long monitoring time. Therefore, it is necessary to first calculate the fault characteristic frequency generated in the area where the fault is located based on prior knowledge, and then adjust the data segment length in the data preprocessing process according to the fault characteristic frequency so that a single data sample can contain enough fault cycles.

[0083] S2: Use the dilated wide kernel convolution to preliminarily process the intermediate data J to obtain the data Y, perform batch normalization on the data Y to obtain the data G, then introduce nonlinear components into the data G and perform maximum pooling to obtain the channel feature U of the channel n ; Since the shield main bearing is a low-speed heavy-load bearing, its speed is extremely low, generally 2 to 5 rpm. This extremely low speed makes the fault data mainly concentrated in the low-frequency part, so the intermediate data J is processed by the dilated wide kernel convolution to further improve the adaptability to the low-frequency part.

[0084] S3: Channel feature U nThe feature vector U is obtained by element-by-element addition, and then the feature vector U is calculated through the full average pooling layer (GAP), the fully connected layer (FC), the BN and the ReLU activation function to obtain the feature Z. Finally, the feature Z is input into the fully connected layer to obtain the attention weight W of each channel. j , the sum of the attention weights of each channel is 1, that is,

[0085] This process is inspired by the attention mechanism and applies the idea of ​​weighting features with attention mechanism learning weights to the features of different channels. Before classification learning based on the extracted features, a set of attention weights is learned with the help of the fully connected layer, and then the weights are converted into soft attention vectors through the SoftMax function. By multiplying the attention vector with the feature, the gain of channels with greater fault correlation and the loss of channels with less fault correlation are achieved.

[0086] S4: The calculated attention weight W j The feature vector S is assigned to obtain the weighted feature vector V, and then the weighted feature vector V is fused, and finally the fused feature is input into the final full connection to realize the mapping process of the fault type. The feature vector V is calculated by the following formula (3):

[0087]

[0088] Among them, W j is the attention weight of the jth channel, S is the feature value obtained after the data Y is input into a series of convolutional layers, and s j is the eigenvalue of the jth channel.

[0089] For bearings with large sizes and complex structures, single sensor monitoring can no longer achieve accurate monitoring of the overall fault status. Therefore, it is necessary to use the method of multi-source information fusion to integrate the monitoring data of multiple sensors and achieve mapping from local to overall. By using weighted operations on each channel, the diagnosis model can perceive the correlation between each channel and the fault. When the fault occurs in the area near the corresponding sensor, the data output by the sensor is more correlated with the fault, and the weight obtained through the attention mechanism is also greater. This method can enable the model to have a certain degree of perception of the fault location.

[0090] As another embodiment of the present invention, step S1 specifically includes the following steps:

[0091] S11, based on prior knowledge, solve the fault characteristic frequency of the corresponding area, combined with the sampling frequency f s , the required data segment length l is calculated by the following formula (4):

[0092]

[0093] Where, f is the fault characteristic frequency, f s is the sampling frequency, N is the number of cycles in the data segment, and the range of N is 5 to 10;

[0094] Calculate the fault characteristic frequency f according to Table 1 below:

[0095] Table 1 Calculation formula for fault characteristic frequency at various locations

[0096]

[0097] Among them, refer to the attached Figure 6 , D m The pitch diameter of the main push raceway roller, D a D is the pitch diameter of the auxiliary roller raceway. r is the radial raceway roller pitch diameter, d m Main push roller diameter, d a is the diameter of the auxiliary roller, d r is the radial roller diameter, f r is the inner ring rotation frequency, n m is the number of main push roller pairs, n a is the number of auxiliary rollers, n r is the number of radial rollers;

[0098] As can be seen from Table 1, there are three fault types for the outer ring, three fault types for the roller, and three fault types for the inner ring, which adds up to a total of 9 fault types.

[0099] The maximum value among the sample lengths corresponding to each fault type is selected as the final sample length, and all data are divided by the sample length;

[0100] S12, combining the sliding window method to segment the time domain data according to the above-required sample length l:

[0101]

[0102] Among them, L is the total length of the data, l is the length of the data segment, stride is the step size, and num represents the number of data segments after segmentation. Among them, num and l are necessary hyperparameters in the fault diagnosis training process and play a decisive role in the parameter results of fault training.

[0103] S13, then normalize the data of each channel using formula (6) to obtain the intermediate data J:

[0104]

[0105] in Indicates the kth data segment in the nth channel data, that is, the intermediate data J, represents the kth data segment of the nth channel, where k∈[0,num], Indicates the maximum value in the corresponding data segment. Indicates the minimum value in the corresponding data segment.

[0106] As another embodiment of the present invention: Step S2 specifically includes the following steps:

[0107] S21, firstly use the dilated wide kernel convolution to perform preliminary processing on the intermediate data J, and the calculation formula (7) of the dilated wide kernel convolution is as follows:

[0108]

[0109] where Y n [i] is the output at position i, which is data Y; W[m] is the weight in the convolution kernel, corresponding to the position m of the kernel; l is the dilation rate, which determines the spacing between the elements of the convolution kernel; M is half the size of the convolution kernel, is the intermediate data J, is the [i+l·m]th data point of the corresponding data segment;

[0110] This step can preliminarily filter out the high-frequency components in the data, thereby improving the adaptability to the low-frequency parts.

[0111] S22, output result Y n [i] Batch normalization is performed to prevent gradient explosion and gradient disappearance during training, and to initially prevent overfitting. The formula is as follows:

[0112] Calculate the mean μ of the entire batch n :

[0113] Calculate the variance of the entire batch

[0114] Normalization:

[0115] Scaling and translation (reparameterization):

[0116] Among them, ∈ is a small number to prevent the denominator from being zero; γ and β are learnable parameters that allow the network to learn to recover any potentially useful feature distribution; x n [i] represents the data point of the nth channel, Represents the normalized data point, y n[i] represents the output of the data after batch normalization, which is data G;

[0117] S23, nonlinear components are introduced into the data through formula (8) and maximum pooling is performed:

[0118] U n =maxpool(ReLU(X)) (8)

[0119] Among them, U n is the eigenvalue extracted from the nth channel. ReLU(X) represents the ReLU function processing of the data G, where ReLU is a nonlinear function. Its main function is to introduce nonlinear factors into the internal parameters of the model.

[0120] As another implementation of the present invention: Step S3 specifically includes the following steps:

[0121] S31, add the eigenvalues ​​obtained from each channel element by element using formula (9) to obtain the eigenvector U:

[0122] U=U1+U2+U3 (9)

[0123] Through this step, the dimension of U is not modified, and the information of multiple channels is fused at the same time.

[0124] S32, U is calculated through the full average pooling layer (GAP), the fully connected layer (FC), the BN and the ReLU activation function to obtain the feature Z. The mathematical model of Z can be further expressed as:

[0125] Z=δ(ψ(U gap )) (10)

[0126] Among them, U gap represents the result of U after global pooling, ψ represents the BN batch normalization process, and δ represents the activation function ReLU;

[0127] It is then input into the fully connected layer to further compress the feature dimension:

[0128] Z=Dense(C / r,C) (11)

[0129] Where Dense represents the fully connected process, r is the feature compression rate, C is the number of input features, and C is usually r = 0.5;

[0130] S33, the compressed feature Z is input into three groups of fully connected layers again to generate three groups of attention weights, and then the weights are normalized through SoftMax. The specific process is as follows:

[0131]

[0132] Among them, A s , B s , C s is the sth row of the feature matrix, and the a generated by this part s 、b s 、c s Then it is the attention weight of the three channels required, a s 、b s 、c s Equivalent to W1, W2, W3, formulas (12), (13) and (14) satisfy the following relationship: a s +b s +c s = 1. Since the number of channels used in this embodiment is 3, Z has 3 parts, namely A, B, and C, and each part is divided into 10 rows according to the 10 categories to be finally identified (the 9 fault categories in Table 1 in step S11 plus the no-fault category), and s is the index of each specific fault type.

[0133] As another implementation of the present invention: Step S4 specifically includes the following steps:

[0134] S41, by convolving the feature Y obtained by the hollow wide kernel in step 2 n [i] Input a series of convolutional layers to further extract deep features:

[0135] S i =Convs(16,128) (15)

[0136] Among them, the number of feature channels obtained by the dilated wide kernel convolution is 16. After a series of convolutions, the number of channels becomes 128. i is the feature vector S.

[0137] S42, multiply the feature vector S obtained by convolution in step S41 by the attention weight obtained in step S3 to obtain a weighted feature vector V: V = a s ·S1+b s ·S2+c s ·S3.

[0138] The present invention first adds and fuses the data of the three channels, and then generates three attention weights through a fully connected layer corresponding to the number of channels. The fully connected layer is used to reduce the channels with small fault correlation during the training process, so that the fault diagnosis method pays more attention to the channels with large fault correlation, alleviating the overfitting risk caused by information redundancy in the multi-channel fault diagnosis process.

[0139] The present invention first adds and fuses the data of the three channels, and then generates three attention weights through a fully connected layer corresponding to the number of channels. The fully connected layer is used to reduce the channels with small fault correlation during the training process, so that the fault diagnosis method pays more attention to the channels with large fault correlation, alleviating the overfitting risk caused by information redundancy in the multi-channel fault diagnosis process.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A tunnel boring machine main bearing fault simulation test bench, characterized in that: It includes a frame and a main bearing rotatably arranged on the frame, the front end and the rear end of the main bearing are respectively provided with a first outer ring and a second outer ring, a slot is provided on the side of the second outer ring facing the main push roller, a fault simulation board is clamped in the slot, the fault simulation board and the second outer ring are interference fit, and a fault detection module is installed on the outer side of the second outer ring at a position corresponding to the fault simulation board; a screw hole is provided at the rear end of the fault simulation board, and a bolt passes through the second outer ring and extends into the screw hole to pull the fault simulation board into the slot; a loading device and an oil circulation device are provided on the frame, the oil circulation device is used to provide lubricating fluid to the main bearing, and the loading device is used to drive the main bearing to rotate and provide axial thrust to the main bearing.

2. The tunnel boring machine main bearing fault simulation test bench according to claim 1, characterized in that: The second outer ring and the first outer ring are fixed by bolts.

3. The tunnel boring machine main bearing fault simulation test bench according to claim 2, characterized in that: A gap is arranged between the fault simulation board and the bottom of the card slot.

4. The tunnel boring machine main bearing fault simulation test bench according to claim 3 is characterized in that: There are three fault simulation panels, which are distributed at three o'clock, six o'clock and nine o'clock on the second outer circle.

5. The tunnel boring machine main bearing fault simulation test bench according to claim 1, characterized in that: The loading device comprises: a servo motor, a transmission shaft, a gear and an inner gear ring, wherein the inner gear ring is fixed on the inner side of the main bearing, the gear is meshed with the inner gear ring, the gear is fixed on the transmission shaft, and the transmission shaft is fixedly connected to the output shaft of the servo motor.

6. A method for diagnosing faults of a main bearing of a tunnel boring machine, characterized in that: The tunnel boring machine main bearing fault simulation test bench comprising any one of claims 1 to 5 further comprises the following steps: S1: The vibration data x of the main bearing is obtained by the fault detection module n (t), each fault detection module provides a channel of vibration data x n (t), and then adjust the data segment length in the data preprocessing process according to the fault characteristic frequency, and then normalize the data of each channel to obtain data J; where x n (t) represents the vibration data of the nth channel at time t; S2: Use the dilated wide kernel convolution to preliminarily process the data J to obtain the data Y, perform batch normalization on the data Y to obtain the data G, then introduce nonlinear components into the data G and perform maximum pooling to obtain the channel feature U of the channel n ; S3: Channel feature U n The feature vector U is obtained by element-by-element addition. Then the feature vector U is calculated through the full average pooling layer (GAP), the fully connected layer (FC), the BN and the ReLU activation function to obtain the feature Z. Finally, the feature Z is input into the fully connected layer to obtain the attention weight W of each channel. j , the sum of the attention weights of each channel is 1, that is, S4: The calculated attention weight W j The feature vector S is assigned to obtain the weighted feature vector V, and then the weighted feature vector V is fused, and finally the fused feature is input into the final full connection to realize the mapping process of the fault type. The feature vector V is calculated by the following formula (3): Among them, W j is the attention weight of the jth channel, the feature vector S is the feature value obtained after the data Y is input into a series of convolutional layers, and s j is the eigenvalue of the jth channel.

7. The fault diagnosis method for the main bearing of a tunnel boring machine according to claim 6, characterized in that: Step S1 specifically includes the following steps: S11, based on prior knowledge, solve the fault characteristic frequency of the corresponding area, combined with the sampling frequency f s , the required data segment length l is calculated by the following formula (4): Where, f is the fault characteristic frequency, f s is the sampling frequency, N is the number of cycles in the data segment, and the range of N is 5 to 10; Calculate the fault characteristic frequency f according to the following table (1): Table 1 Calculation formula for fault characteristic frequency at various locations Among them, D m The pitch diameter of the main push raceway roller, D a D is the pitch diameter of the auxiliary roller raceway. r is the radial raceway roller pitch diameter, d m Main push roller diameter, d a is the diameter of the auxiliary roller, d r is the radial roller diameter, f r is the inner ring rotation frequency, n m is the number of main push roller pairs, n a is the number of auxiliary rollers, n r is the number of radial rollers; S12, combining the sliding window method to segment the time domain data according to the above-required sample length l: Where L is the total length of the data, l is the length of the data segment, stride is the step size, and num represents the number of data segments after segmentation; S13, then normalize the data of each channel using formula (6): in Indicates the kth data segment in the nth channel data, that is, data J, represents the kth data segment of the nth channel, where k∈[0,num], Indicates the maximum value in the corresponding data segment. Indicates the minimum value in the corresponding data segment.

8. The fault diagnosis method for the main bearing of a tunnel boring machine according to claim 7, characterized in that: Step S2 specifically includes the following steps: S21, firstly use the dilated wide kernel convolution to perform preliminary processing on the data J, and the calculation formula (7) of the dilated wide kernel convolution is as follows: where Y n [i] is the output at position i, which is the data Y; W[m] is the weight in the convolution kernel, corresponding to the position m of the kernel; l is the dilation rate, which determines the spacing between the elements of the convolution kernel; M is half the size of the convolution kernel, For data J, is the [i+l·m]th data point of the corresponding data segment; S22, output result Y n [i] Perform batch normalization, the formula is as follows: Calculate the mean μ of the entire batch n : Calculate the variance of the entire batch Normalization: Scaling and translation (reparameterization): Among them, ∈ is a small number to prevent the denominator from being zero; γ and β are learnable parameters that allow the network to learn to recover any potentially useful feature distribution; x n [i] represents the data point of the nth channel, Represents the normalized data point, y n [i] represents the output of the data after batch normalization, which is data G; S23, nonlinear components are introduced into the data through formula (8) and maximum pooling is performed: HER n =maxpool(ReLU(X)) (8) Among them, U n It is the eigenvalue extracted from the nth channel, and ReLU(X) means the ReLU function is applied to the data G.

9. The fault diagnosis method for the main bearing of a tunnel boring machine according to claim 8, characterized in that: Step S3 specifically includes the following steps: S31, the eigenvalues ​​obtained from each channel are added element by element using formula (9): U=U1+U2+U3 (9) Among them, the feature vector U integrates the information of multiple channels; S32, U is calculated through the full average pooling layer (GAP), the fully connected layer (FC), the BN and the ReLU activation function to obtain the feature Z. The mathematical model of Z can be further expressed as: Z=δ(ψ(U gap )) (10) Among them, U gap represents the result of U after global pooling, ψ represents the BN batch normalization process, and δ represents the activation function ReLU; It is then input into the fully connected layer to further compress the feature dimension: Z=Dense(C / r,C) (11) Where Dense represents the fully connected process, r is the feature compression rate, C is the number of input features, and C is usually r = 0.5; S33, the compressed feature Z is input into three groups of fully connected layers again to generate three groups of attention weights, and then the weights are normalized through SoftMax. The specific process is as follows: Among them, A s , B s , C s is the sth row of the feature matrix, and the a generated by this part s , b s 、c s Then is the attention weight of the three channels, formula (12), (13) and (14) satisfy the following relationship: a s +b s +c s =1.

10. The fault diagnosis method of the main bearing of a tunnel boring machine according to claim 9, characterized in that: Step S4 specifically includes the following steps: S41, by convolving the feature Y obtained by the hollow wide kernel in step 2 n [i] Input a series of convolutional layers to further extract deep features: S i =Convs(16,128) (15) Among them, the number of feature channels obtained by the dilated wide kernel convolution is 16. After a series of convolutions, the number of channels becomes 128. i is the feature vector S; S42, convolve the feature vector S obtained in step S41 i Multiply it with the attention weight obtained in step S3 to obtain the weighted feature vector V: V = a s ·S1+b s ·S2+c s ·S3.

Images