Shield tunneling machine main drive transmission gear monitoring and fault diagnosis method and system
By laying multiple vibration sensors in the shield machine and using multi-stage feature encoder and fully connected network, the problem of difficult transmission gear failure is solved, and accurate fault diagnosis and stability of the shield machine transmission system is achieved.
Patent Information
- Application Number
- CN202411328596.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-24
- Publication Date
- 2024-11-08
AI Technical Summary
It is difficult to effectively monitor and diagnose failures such as wear, eccentricity, different axes, local abnormalities, tooth pitch errors and imbalances of the transmission gear in the shield machine, affecting the stability and efficiency of the entire drive system.
At least 3 vibration sensors are uniformly arranged on the bearing seat side. Through the processing of logarithmic Mel spectrogram and vibration signal intensity vector, combined with the pre-trained shield machine transmission gear fault diagnosis model, a multi-stage vibration feature encoder is used and Fully connected network, extract the characteristics of time, frequency and spatial scales, and predict the defect type and position of the transmission gear.
It realizes accurate identification and diagnosis of transmission gear failures, improves the transmission system stability and operating efficiency of the shield machine, and can better identify defect locations and types.
Smart Images

Figure CN118913689A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of shield machine main drive monitoring, and in particular to a shield machine main drive transmission gear monitoring and fault diagnosis method and system. Background Art
[0002] The main drive is the heart and power source of the shield machine, and the transmission gear and the main bearing rely on the motor and the reducer for gear transmission. The transmission relationship between the transmission gear and the main bearing determines the transmission of the entire shield. If the transmission gear is worn a lot or a gear is broken during the excavation of the shield machine, it will have a great impact on the entire drive system. Therefore, it is very necessary to add a vibration sensor to the bearing seat of the gear drive to monitor the transmission gear and diagnose the fault. Among them, the common gear failures are as follows: uniform wear of gears, gear eccentricity, gear misalignment, local gear abnormality, pitch error and gear imbalance. Uniform wear of gears refers to the wear of most of the tooth surface due to reasons such as the material and lubrication of the gear, or long-term work under high load. Gear eccentricity means that the center of the gear does not coincide with the center of the rotating shaft. Gear misalignment fault refers to the misalignment of the gear and the shaft caused by improper assembly of the gear and the shaft. The misalignment fault will cause local contact of the gear, resulting in some gear teeth bearing a large load. Local gear abnormalities include: large cracks at the root of the tooth, local tooth surface wear, gear tooth breakage, local tooth profile error, etc. Pitch error refers to the fact that the pitches of each gear are not equal and there is an error. Pitch error is caused by tooth shape error, and almost all gears have a small pitch error. Gear imbalance refers to the misalignment of the center of mass and the center of rotation of the gear, which leads to unstable operation and vibration of the gear pair. The failure of the transmission gear usually generates a vibration signal related to the transmission frequency. When the gear fails, the vibration signal generated will have different changes in the time domain and frequency domain. The characteristic changes of the signal will then reflect the operating status and fault type of the transmission gear in the shield machine. Summary of the invention
[0003] In order to solve the above technical problem or at least partially solve the above technical problem, the present invention provides a shield machine main drive transmission gear monitoring and fault diagnosis method and system.
[0004] In a first aspect, the present invention provides a method for monitoring and diagnosing faults of a main drive transmission gear of a shield machine, comprising: evenly distributing at least three vibration sensors on one side of a bearing seat in the main drive of the shield machine, wherein the bearing seat is connected to a main bearing of a transmission gear of the main drive of the shield machine;
[0005] Each group of vibration sensors extracts the vibration signal Vs k , k∈{1,2,......,K} represents the index of the vibration sensor, K is the total number of vibration sensors;
[0006] Each of the vibration signals is divided into a number of vibration frames using a time window of a set length. z∈{1,2,......,Z} represents the index of the vibration frame, and Z is the number of vibration frames in a batch of inputs;
[0007] Calculate each vibration frame The logarithmic Mel spectrogram and the vibration signal intensity vector of the vibration signal in the image are combined in the channel dimension, and the logarithmic Mel spectrogram and the intensity vector of the same frame of the same vibration sensor are sequentially spliced in order to obtain the frame input features of each frame; the set amount of frame input features are stacked in the channel dimension to obtain the input features;
[0008] The pre-trained shield machine transmission gear fault diagnosis model is used to process the input features to obtain the predicted type activation vector and predicted azimuth vector of the corresponding input features, and the predicted type activation vector and predicted azimuth vector are used to determine the transmission gear defect type and azimuth of the corresponding input features; wherein the shield machine transmission gear fault diagnosis model models the relationship between the input features and the type activation vector, and models the relationship between the input features and the azimuth vector.
[0009] Furthermore, the shield machine transmission gear fault diagnosis model includes: a multi-level vibration feature encoder, a multi-level vibration feature decoder and a fully connected network; the multi-level vibration feature encoder extracts three-level vibration features from different time, frequency and space scales step by step, and the multi-level vibration feature encoder models the state connection of the three-level vibration features of different time, frequency and space scales through a bidirectional Mamba state space model; the multi-level vibration feature decoder fuses the features that have established state connections of different time, frequency and space scales step by step; and the final fusion result is input into the first fully connected sub-network and the second fully connected sub-network in the fully connected network.
[0010] Furthermore, the multi-level vibration feature encoder includes: a dual convolution layer, the output of the dual convolution layer is input into a first-level convolution block after being processed by the convolution layer, the output of the first-level convolution block is divided into two branches, one branch is input into a second-level convolution block after downsampling processing, and the other branch is input into a first two-way Mamba state space model, and state modeling is performed at a scale using the first two-way Mamba state space model; the output of the second-level convolution block is divided into two branches, one branch is input into a third-level convolution block after downsampling processing, and the other branch is input into a second two-way Mamba state space model, and state modeling is performed at a scale using the second two-way Mamba state space model; the output of the third-level convolution block is input into a third two-way Mamba state space model, and state modeling is performed at a last scale using the third two-way Mamba state space model.
[0011] Furthermore, the multi-level vibration feature decoder includes: a third convolution block connected to a third bidirectional Mamba state space model, the output of the third convolution block is upsampled and added and fused with the output of the second bidirectional Mamba state space model, and the fusion result is input into the second convolution block; the output of the second convolution block is upsampled and added and fused with the output of the first bidirectional Mamba state space model, and the fusion result is input into the first convolution block.
[0012] Furthermore, in the multi-level vibration feature encoder and the multi-level vibration feature decoder, the structures of any type of convolution blocks are consistent. The convolution blocks include: several stacked point convolutions, depth convolutions and point convolutions. In the convolution blocks, the expansion rates of the depth convolutions are 1, 3, 5, 7...2n-1, respectively, where n is the total number of stacked point convolutions, depth convolutions and point convolutions.
[0013] Furthermore, the structural consistency of any bidirectional Mamba state space model includes: a forward Mamba state space model and a reverse Mamba state space model, wherein the reverse Mamba state space model reverses the order of input features by feature flipping before; the reverse Mamba state space model flips the output features to be consistent with the output order of the forward Mamba state space model by feature flipping after, and then the flipped reverse Mamba state space model output is fused with the forward Mamba state space model output by addition.
[0014] Furthermore, a training platform is built to collect standard vibration signals corresponding to each transmission gear defect, and the standard vibration signals are processed into training data:
[0015] Obtain several transmission gears with uniform wear, eccentricity, misalignment, local anomalies, pitch errors and imbalances, and use the transmission gears in conjunction with reducers, motors, main bearings and vibration sensors to build a training platform; traverse different gears of the reducer, control the output speed of the motor, and collect the defect vibration signal corresponding to each defect through each vibration sensor; align and normalize the defect vibration signals corresponding to each defect to obtain a normalized defect vibration signal; arbitrarily scale and phase-shift the normalized defect vibration signal, and add Gaussian noise to obtain a random defect vibration signal; arbitrarily superimpose the random defect vibration signal to expand it; use the signal containing the random defect vibration signal and its arbitrary superposition as the training vibration signal t Vs k , k∈{1,2,......,K} represents the index of the vibration sensor, K is the total number of vibration sensors; each of the training vibration signals is divided into a number of vibration frames using a time window of a set length z∈{1,2,......,Z} represents the index of the vibration frame of the training vibration signal, and Z is the number of vibration frames of a batch of input training vibration signals; the corresponding defects in each vibration frame in a batch of input vibration frames are marked by the DOA vector N;
[0016] The shield machine transmission gear fault diagnosis model is trained using training data, and the cross entropy between the predicted type activation vector and the actual type activation vector, and the weighted sum of the cross entropy between the predicted orientation vector and the actual predicted orientation vector are used as loss functions. Based on the loss function value, the shield machine transmission gear fault diagnosis model is trained using a gradient descent algorithm; so that the loss function of the shield machine transmission gear fault diagnosis model in the training data domain is less than a set threshold.
[0017] Furthermore, the DOA vector N is the product of the type activation vector and the range vector: N = AD, A∈M K×Z×C is the type activation vector of the DOA vector, C represents the total number of defect types, each vibration frame contains at most C types, when any defect exists, the dimension value of the corresponding class in the type activation vector is 1, and the subsequent orientation vector is meaningful, when any defect does not exist, the dimension value of the corresponding class in the type activation vector is 0, and the product of the subsequent orientation vector and the type activation vector is 0, K is the total number of vibration sensors; D∈R K×Z×C×3 is the azimuth vector of the DOA vector, R K×Z×C×3 The C dimension represents the total number of defect types, and the 3 dimensions correspond to the xyz axes of the K Cartesian coordinate system.
[0018] In the second aspect, the present invention provides a shield machine main drive transmission gear monitoring and fault diagnosis device, comprising: a processing unit, a storage unit and a collection unit interconnected by a bus, wherein the collection unit is used for vibration signals, wherein the storage unit stores a computer program and the vibration signal collected by the collection unit, and when the processing unit reads the vibration signal and executes the computer program, the shield machine main drive transmission gear monitoring and fault diagnosis method is implemented.
[0019] In a third aspect, the present invention provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, the shield machine main drive transmission gear monitoring and fault diagnosis method is implemented.
[0020] The above technical solution provided by the embodiment of the present invention has the following advantages compared with the prior art:
[0021] The present application evenly arranges at least three vibration sensors on one side of the bearing seat in the main drive of the shield machine, and the bearing seat is connected to the main bearing of the transmission gear of the main drive of the shield machine; each group of vibration sensors extracts vibration signals; each vibration signal is divided into a number of vibration frames using a time window of a set length; the logarithmic Mel spectrum and vibration signal intensity vector of the vibration signal in each vibration frame are calculated, the logarithmic Mel spectrum and intensity vector of the same frame of the same vibration sensor are combined in the channel dimension, and the logarithmic Mel spectrum and intensity vector of the same frame of different vibration sensors are spliced in sequence to obtain the frame input features of each frame; a set amount of frame input features are stacked in the channel dimension to obtain input features; the input features are processed using a pre-trained shield machine transmission gear fault diagnosis model to obtain the predicted type activation vector and predicted orientation vector of the corresponding input features, and the transmission gear defect type and orientation of the corresponding input features are determined using the predicted type activation vector and predicted orientation vector. In the fault diagnosis model of shield machine transmission gear, the multi-level vibration feature encoder extracts three-level vibration features from different time, frequency and space scales step by step. The multi-level vibration feature encoder models the state connection of three-level vibration features of different time, frequency and space scales through the bidirectional Mamba state space model to strengthen the connection between features; the multi-level vibration feature decoder integrates the features with different time, frequency and space scale state connections step by step; the final fusion result is input into the first fully connected sub-network and the second fully connected sub-network in the fully connected network, the first fully connected sub-network maps the final fusion result into a prediction type activation vector, and the second fully connected sub-network maps the final fusion result into a prediction orientation vector. The prediction type activation vector and the prediction orientation vector actually determine the prediction DOA vector. The DOA vector used to mark the vibration frame of the vibration signal not only marks the type of vibration frame defect collected by each vibration sensor, but also marks the orientation information of the defect in the vibration frame collected by each vibration sensor. At least three vibration sensors evenly distributed on one side of the bearing seat can better identify the defect location through the orientation information of the three. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0024] Figure 1 A flow chart of a method for monitoring and diagnosing faults of a main drive transmission gear of a shield machine provided by an embodiment of the present invention;
[0025] Figure 2 A schematic diagram of a training platform or a diagnostic platform provided by an embodiment of the present invention;
[0026] Figure 3 A flowchart of using a training platform to obtain training data provided by an embodiment of the present invention;
[0027] Figure 4 A schematic diagram of a fault diagnosis model for a shield machine transmission gear provided by an embodiment of the present invention;
[0028] Figure 5 A schematic diagram of a level vibration characteristic encoder provided by an embodiment of the present invention;
[0029] Figure 6 A schematic diagram of a bidirectional Mamba state space model provided in an embodiment disclosed in the present invention;
[0030] Figure 7 A schematic diagram of a Mamba state space model constituting a bidirectional Mamba state space model provided in an embodiment disclosed in the present invention;
[0031] Figure 8 A schematic diagram of a multi-level vibration feature decoder provided by an embodiment of the present invention;
[0032] Fig. 9 A schematic diagram of a shield machine main drive transmission gear monitoring and fault diagnosis device provided in an embodiment disclosed by the present invention. DETAILED DESCRIPTION
[0033] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments 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.
[0034] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.
[0035] Example 1
[0036] An embodiment of the present invention provides a shield machine main drive transmission gear monitoring and fault diagnosis method. The present application includes a shield machine transmission gear fault diagnosis model training phase and a shield machine transmission gear fault diagnosis model application diagnosis phase.
[0037] See also Figure 2 As shown, a platform with the following structure needs to be built in both the training stage and the diagnosis stage: at least three vibration sensors 4 are evenly arranged on one side of the bearing seat 3 in the main drive of the shield machine, and the bearing seat 3 is connected to the main bearing 2 of the transmission gear 1 of the main drive of the shield machine. In the example of this application, the number of the vibration sensors 4 is three, evenly distributed on one side of the bearing seat 3. The platform in the winning training stage of this application is called the training platform; the platform in the diagnosis stage is called the diagnosis platform.
[0038] The training phase includes:
[0039] A training platform is built to collect standard vibration signals corresponding to each transmission gear defect, and the standard vibration signals are processed into training data. In the specific implementation process, Figure 3 As shown, including:
[0040] Obtain transmission gears with several defects, including uniform wear, eccentricity, misalignment, local anomalies, pitch errors, and imbalance; use the transmission gears with the above defects in conjunction with reducers, motors, main bearings, and vibration sensors to build a training platform;
[0041] Traverse the different gears of the reducer, control the output speed of the motor, and collect the defect vibration signal corresponding to each defect through each vibration sensor;
[0042] Aligning and normalizing the defect vibration signals corresponding to each defect to obtain a normalized defect vibration signal;
[0043] The normalized defect vibration signal is arbitrarily scaled, arbitrarily phase-shifted, and Gaussian noise is added to obtain a random defect vibration signal;
[0044] Random defect vibration signals are arbitrarily superimposed and expanded;
[0045] The signal containing random defect vibration signal and its arbitrary superposition is used as the training vibration signal t Vs k , k∈{1,2,......,K} represents the index of the vibration sensor, K is the total number of vibration sensors;
[0046] Each of the training vibration signals is divided into a number of vibration frames using a time window of a set length. z∈{1,2,......,Z} represents the index of the vibration frame of the training vibration signal, and Z is the number of vibration frames of a batch of input training vibration signals.
[0047] The corresponding defects in each vibration frame in a batch of input vibration frames are annotated by the DOA vector N. The DOA vector N is the product of the type activation vector and the range vector: N = AD, A∈M K×Z×C is the type activation vector of the DOA vector, C represents the total number of defect types, each vibration frame contains at most C types, when any defect exists, the dimension value of the corresponding class in the type activation vector is 1, and the subsequent orientation vector is meaningful, when any defect does not exist, the dimension value of the corresponding class in the type activation vector is 0, and the product of the subsequent orientation vector and the type activation vector is 0, K is the total number of vibration sensors; D∈R K×Z×C×3 is the azimuth vector of the DOA vector, R K×Z×C×3 The C dimension represents the total number of defect types, and the 3 dimensions correspond to the xyz axes of the K Cartesian coordinate systems. The DOA vector not only marks the type of defects in the vibration frame collected by each vibration sensor, but also the orientation information of the defects in the vibration frame collected by each vibration sensor.
[0048] During the training process, the construction process of input features includes:
[0049] Calculate each vibration frame of the training vibration signal t Vs z k, combine the logarithmic Mel spectrogram and intensity vector of the vibration signal in the same frame of the same vibration sensor in the channel dimension, and splice the logarithmic Mel spectrogram and intensity vector of the same frame of different vibration sensors in sequence to obtain the frame input feature that combines the vibration frame information of all vibration sensors at the same time; stack the set amount of frame input features in the channel dimension to obtain the input feature.
[0050] like Figure 4 As shown, a fault diagnosis model for the transmission gear of a shield machine is constructed, and the fault diagnosis model for the transmission gear of the shield machine includes: a multi-level vibration feature encoder, a multi-level vibration feature decoder and a fully connected network; the multi-level vibration feature encoder extracts three-level vibration features from different time, frequency and space scales step by step, and the multi-level vibration feature encoder models the state connection of the three-level vibration features of different time, frequency and space scales through a bidirectional Mamba state space model; the multi-level vibration feature decoder fuses the features that have established state connections of different time, frequency and space scales step by step; and the final fusion result is input into the first fully connected sub-network and the second fully connected sub-network in the fully connected network.
[0051] Among them, Figure 5As shown, the multi-level vibration feature encoder includes: a dual convolution layer, the output of the dual convolution layer is input into a primary convolution block after being processed by the convolution layer, the output of the primary convolution block is divided into two branches, one branch is input into a secondary convolution block after downsampling, and the other branch is input into a first two-way Mamba state space model, and the first two-way Mamba state space model is used to perform state modeling at one scale; the output of the secondary convolution block is divided into two branches, one branch is input into a tertiary convolution block after downsampling, and the other branch is input into a second two-way Mamba state space model, and the second two-way Mamba state space model is used to perform state modeling at one scale; the output of the tertiary convolution block is input into a third two-way Mamba state space model, and state modeling is performed at the last scale using the third two-way Mamba state space model. In a multi-level vibration feature encoder, the structure of any type of convolution block is consistent. The convolution block contains: a number of stacked point convolutions, depth convolutions and point convolutions. In the convolution block, the expansion rate of the depth convolution is 1, 3, 5, 7...2n-1, respectively, where n is the total number of stacked point convolutions, depth convolutions and point convolutions. Since the stacking result of the frame input features in the channel dimension corresponds to the input features of a batch of vibration signals, the channel dimension of the input features actually corresponds to the time domain. The deep convolution of this application sets different expansion rates in the channel dimension. When the expansion rate is low, the convolution block can collect brief and detailed time context information; as the expansion rate increases, the convolution block can capture the extended and coarse-grained time context information. In summary, feature extraction from different time scales is achieved. The first-level convolution block, the second-level convolution block and the third-level convolution block are downsampled, and the frequency granularity and spatial granularity are changed by downsampling. The relationship between the logarithmic Mel spectrum features of different frequency granularities and the vibration signal intensity vectors of different spatial granularities are modeled, and features are extracted from different frequencies and spatial scales.
[0052] like Figure 6As shown, the structure of any bidirectional Mamba state space model is consistent, including: a forward Mamba state space model and a reverse Mamba state space model, the forward Mamba state space model and the reverse Mamba state space model are followed by layer normalization respectively, and the outputs of the forward Mamba state space model and the reverse Mamba state space model after layer normalization are added and fused with the inputs of the forward Mamba state space model and the reverse Mamba state space model respectively, wherein the feature order of the input is flipped by feature flipping before the reverse Mamba state space model; the output features are flipped by feature flipping after the reverse Mamba state space model to be consistent with the output order of the forward Mamba state space model, and then the flipped reverse Mamba state space model output is fused with the output of the forward Mamba state space model by addition. The structures of the forward Mamba state space model and the reverse Mamba state space model in any bidirectional Mamba state space model are consistent, as shown Figure 7 As shown in the figure, the input is divided into two branches. One branch is processed by a linear layer, a one-dimensional convolution and a SiLU activation function, and then the feature is expanded by selective scanning to obtain several feature sequences. The feature sequences are mapped into Δ, B, C through a linear layer. Δ, B, C are matrices used to construct the state equation in the state space model. In the state space model, the state equation of each feature sequence is: h(t) = ΔAh(t-1) + ΔBx(t), y(t) = Ch(t); where A is a learnable matrix, h(t) is the hidden state corresponding to the current input feature sequence x(t) and the previous hidden state h(t-1), and y(t) is the output of the state space model. The other branch is processed by a linear layer and a SiLU activation function, and then multiplied by the output of the state space model. Through the state modeling of time, frequency, and space features at different scales, the connection between features can be better established.
[0053] like Figure 8 As shown, the multi-level vibration feature decoder includes: a third convolution block connected to the third bidirectional Mamba state space model, the output of the third convolution block is added and fused with the output of the second bidirectional Mamba state space model after upsampling, and the fusion result is input to the second convolution block; the output of the second convolution block is added and fused with the output of the first bidirectional Mamba state space model after upsampling, and the fusion result is input to the first convolution block. The multi-level vibration feature decoder will gradually fuse the features that establish different time, frequency, and space scale state connections; the final fusion result is input to the first fully connected sub-network and the second fully connected sub-network in the fully connected network. The first fully connected sub-network maps the final fusion result into a predicted type activation vector, and the second fully connected sub-network maps the final fusion result into a predicted orientation vector.
[0054] The shield machine transmission gear fault diagnosis model is trained using training data, and the weighted sum of the cross entropy between the predicted type activation vector and the actual type activation vector, and the cross entropy between the predicted azimuth vector and the actual predicted azimuth vector is used as the loss function. Based on the loss function value, the shield machine transmission gear fault diagnosis model is trained using a gradient descent algorithm; so that the loss function of the shield machine transmission gear fault diagnosis model in the training data domain is less than a set threshold. Through training, the shield machine transmission gear fault diagnosis model models the relationship between the input features and the type activation vector, and models the relationship between the input features and the azimuth vector.
[0055] After training the fault diagnosis model of shield machine transmission gear, a diagnosis platform is built in the shield machine.
[0056] Each group of vibration sensors 4 of the diagnostic platform extracts the vibration signal Vs k , k∈{1,2,......,K} represents the index of the vibration sensor, and K is the total number of vibration sensors. Each of the vibration signals is divided into several vibration frames using a time window of a set length z∈{1,2,......,Z} represents the index of the vibration frame, and Z is the number of vibration frames input in a batch.
[0057] Calculate each vibration frame The logarithmic Mel spectrogram and the vibration signal intensity vector of the vibration signal in the image are combined in the channel dimension, and the logarithmic Mel spectrogram and the intensity vector of the same frame of the same vibration sensor are sequentially spliced in order to obtain the frame input features of each frame; the set amount of frame input features are stacked in the channel dimension to obtain the input features;
[0058] The pre-trained shield machine transmission gear fault diagnosis model is used to process the input features to obtain the prediction type activation vector and prediction orientation vector corresponding to the input features. The prediction type activation vector and prediction orientation vector are used to determine the transmission gear defect type and orientation corresponding to the input features.
[0059] Example 2
[0060] See also Fig. 9 As shown, the present invention provides a shield machine main drive transmission gear monitoring and fault diagnosis device, comprising: a processing unit, a storage unit and a collection unit interconnected via a bus, wherein the collection unit is used to collect vibration signals, wherein the storage unit stores a computer program and the vibration signal collected by the collection unit, and when the processing unit reads the vibration signal and executes the computer program, the shield machine main drive transmission gear monitoring and fault diagnosis method is implemented, comprising:
[0061] The unit acquisition includes at least 3 vibration sensors, and the at least 3 vibration sensors are evenly arranged on one side of the bearing seat in the main drive of the shield machine, and the bearing seat is connected to the main bearing of the transmission gear of the main drive of the shield machine;
[0062] Each group of vibration sensors extracts the vibration signal Vs k , k∈{1,2,......,K} represents the index of the vibration sensor, K is the total number of vibration sensors;
[0063] Each of the vibration signals is divided into a number of vibration frames using a time window of a set length. z∈{1,2,......,Z} represents the index of the vibration frame, and Z is the number of vibration frames in a batch of inputs;
[0064] Calculate each vibration frame The logarithmic Mel spectrogram and the vibration signal intensity vector of the vibration signal in the image are combined in the channel dimension, and the logarithmic Mel spectrogram and the intensity vector of the same frame of the same vibration sensor are sequentially spliced in order to obtain the frame input features of each frame; the set amount of frame input features are stacked in the channel dimension to obtain the input features;
[0065] The pre-trained shield machine transmission gear fault diagnosis model is used to process the input features to obtain the predicted type activation vector and predicted azimuth vector of the corresponding input features, and the predicted type activation vector and predicted azimuth vector are used to determine the transmission gear defect type and azimuth of the corresponding input features; wherein the shield machine transmission gear fault diagnosis model models the relationship between the input features and the type activation vector, and models the relationship between the input features and the azimuth vector.
[0066] Of course, the storage unit in the shield machine main drive transmission gear monitoring and fault diagnosis device provided in an embodiment of the present invention, the computer program stored in it is not limited to the method operation described above, and can also execute the relevant operations in the shield machine main drive transmission gear monitoring and fault diagnosis method provided in any embodiment of the present invention.
[0067] Example 3
[0068] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the shield machine main drive transmission gear monitoring and fault diagnosis method is implemented, including:
[0069] At least three vibration sensors are evenly arranged on one side of a bearing seat in the main drive of the shield machine, wherein the bearing seat is connected to a main bearing of a transmission gear of the main drive of the shield machine;
[0070] Each group of vibration sensors extracts the vibration signal Vs k, k∈{1,2,......,K} represents the index of the vibration sensor, K is the total number of vibration sensors;
[0071] Each of the vibration signals is divided into a number of vibration frames using a time window of a set length. z∈{1,2,......,Z} represents the index of the vibration frame, and Z is the number of vibration frames in a batch of inputs;
[0072] Calculate each vibration frame The logarithmic Mel spectrogram and the vibration signal intensity vector of the vibration signal in the image are combined in the channel dimension, and the logarithmic Mel spectrogram and the intensity vector of the same frame of the same vibration sensor are sequentially spliced in order to obtain the frame input features of each frame; the set amount of frame input features are stacked in the channel dimension to obtain the input features;
[0073] The pre-trained shield machine transmission gear fault diagnosis model is used to process the input features to obtain the predicted type activation vector and predicted azimuth vector of the corresponding input features, and the predicted type activation vector and predicted azimuth vector are used to determine the transmission gear defect type and azimuth of the corresponding input features; wherein the shield machine transmission gear fault diagnosis model models the relationship between the input features and the type activation vector, and models the relationship between the input features and the azimuth vector.
[0074] Of course, the computer-readable storage medium provided by an embodiment of the present invention stores a computer program which is not limited to the method operations described above, but can also execute related operations in a shield machine main drive transmission gear monitoring and fault diagnosis method provided by any embodiment of the present invention.
[0075] In the embodiments provided by the present invention, it should be understood that the disclosed structures and methods can be implemented in other ways. For example, the structural embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, structures or units, which can be electrical, mechanical or other forms.
[0076] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0077] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0078] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those 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 will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.
Claims
1. A shield machine main drive transmission gear monitoring and fault diagnosis method, characterized in that: include: At least three vibration sensors are evenly arranged on one side of a bearing seat in the main drive of the shield machine, wherein the bearing seat is connected to a main bearing of a transmission gear of the main drive of the shield machine; Each group of vibration sensors extracts the vibration signal Vs k , k∈{1,2,......,K} represents the index of the vibration sensor, K is the total number of vibration sensors; Each of the vibration signals is divided into a number of vibration frames using a time window of a set length. z∈{1,2,......,Z} represents the index of the vibration frame, and Z is the number of vibration frames in a batch of inputs; Calculate each vibration frame The logarithmic Mel spectrogram and the vibration signal intensity vector of the vibration signal in the image are combined in the channel dimension, and the logarithmic Mel spectrogram and the intensity vector of the same frame of the same vibration sensor are sequentially spliced in order to obtain the frame input features of each frame; the set amount of frame input features are stacked in the channel dimension to obtain the input features; The pre-trained shield machine transmission gear fault diagnosis model is used to process the input features to obtain the predicted type activation vector and predicted azimuth vector of the corresponding input features, and the predicted type activation vector and predicted azimuth vector are used to determine the transmission gear defect type and azimuth of the corresponding input features; wherein the shield machine transmission gear fault diagnosis model models the relationship between the input features and the type activation vector, and models the relationship between the input features and the azimuth vector.
2. The shield machine main drive transmission gear monitoring and fault diagnosis method according to claim 1 is characterized in that: The shield machine transmission gear fault diagnosis model includes: a multi-level vibration feature encoder, a multi-level vibration feature decoder and a fully connected network; the multi-level vibration feature encoder extracts three-level vibration features from different time, frequency and space scales step by step, and the multi-level vibration feature encoder models the state connection of the three-level vibration features of different time, frequency and space scales through a bidirectional Mamba state space model; the multi-level vibration feature decoder fuses the features that establish the state connection of different time, frequency and space scales step by step; and the final fusion result is input into the first fully connected sub-network and the second fully connected sub-network in the fully connected network.
3. The shield machine main drive transmission gear monitoring and fault diagnosis method according to claim 2 is characterized in that: The multi-level vibration feature encoder includes: a dual convolution layer, the output of the dual convolution layer is input into a primary convolution block after being processed by the convolution layer, the output of the primary convolution block is divided into two branches, one branch is input into a secondary convolution block after being down-sampled, and the other branch is input into a first two-way Mamba state space model, and the first two-way Mamba state space model is used to perform state modeling at a scale; the output of the secondary convolution block is divided into two branches, one branch is input into a tertiary convolution block after being down-sampled, and the other branch is input into a second two-way Mamba state space model, and the second two-way Mamba state space model is used to perform state modeling at a scale; the output of the tertiary convolution block is input into a third two-way Mamba state space model, and state modeling is performed at a last scale using the third two-way Mamba state space model.
4. The shield machine main drive transmission gear monitoring and fault diagnosis method according to claim 2 is characterized in that: The multi-level vibration feature decoder includes: a third convolution block connected to a third bidirectional Mamba state space model, the output of the third convolution block is added and fused with the output of the second bidirectional Mamba state space model after upsampling, and the fusion result is input into the second convolution block; the output of the second convolution block is added and fused with the output of the first bidirectional Mamba state space model after upsampling, and the fusion result is input into the first convolution block.
5. The shield machine main drive transmission gear monitoring and fault diagnosis method according to claim 3 or 4, characterized in that: In the multi-level vibration feature encoder and the multi-level vibration feature decoder, the structures of any type of convolution blocks are consistent. The convolution blocks include: several stacked point convolutions, depth convolutions and point convolutions. In the convolution blocks, the expansion rates of the depth convolutions are 1, 3, 5, 7...2n-1, respectively, where n is the total number of stacked point convolutions, depth convolutions and point convolutions.
6. The shield machine main drive transmission gear monitoring and fault diagnosis method according to claim 3 is characterized in that: The structural consistency of any bidirectional Mamba state space model includes: a forward Mamba state space model and a reverse Mamba state space model, wherein the reverse Mamba state space model reverses the order of input features by feature flipping before; the reverse Mamba state space model flips the output features to be consistent with the output order of the forward Mamba state space model by feature flipping after, and then the flipped reverse Mamba state space model output is fused with the forward Mamba state space model output by addition.
7. The shield machine main drive transmission gear monitoring and fault diagnosis method according to claim 1, characterized in that: Build a training platform to collect standard vibration signals corresponding to each transmission gear defect and process the standard vibration signals into training data: Obtain transmission gears with several defects, including: uniform wear, eccentricity, misalignment, local anomalies, pitch errors and imbalance; use the transmission gears with the above defects in conjunction with reducers, motors, main bearings and vibration sensors to build a training platform; traverse different gears of the reducer, control the output speed of the motor, and collect the defect vibration signal corresponding to each defect through each vibration sensor; align and normalize the defect vibration signals corresponding to each defect to obtain a normalized defect vibration signal; arbitrarily scale and arbitrarily phase-shift the normalized defect vibration signal, and add Gaussian noise to obtain a random defect vibration signal; arbitrarily superimpose the random defect vibration signal to expand it; use the signal containing the random defect vibration signal and its arbitrary superposition as the training vibration signal k∈{1,2,......,K} represents the index of the vibration sensor, K is the total number of vibration sensors; each of the training vibration signals is divided into a number of vibration frames using a time window of a set length z∈{1,2,......,Z} represents the index of the vibration frame of the training vibration signal, and Z is the number of vibration frames of a batch of input training vibration signals; the corresponding defects in each vibration frame in a batch of input vibration frames are marked by the DOA vector N; The shield machine transmission gear fault diagnosis model is trained using training data, and the cross entropy between the predicted type activation vector and the actual type activation vector, and the weighted sum of the cross entropy between the predicted orientation vector and the actual predicted orientation vector are used as loss functions. Based on the loss function value, the shield machine transmission gear fault diagnosis model is trained using a gradient descent algorithm; so that the loss function of the shield machine transmission gear fault diagnosis model in the training data domain is less than a set threshold.
8. The shield machine main drive transmission gear monitoring and fault diagnosis method according to claim 7, characterized in that: The DOA vector N is the product of the type activation vector and the range vector: N = AD, A∈M K×Z×C is the type activation vector of the DOA vector, C represents the total number of defect types, each vibration frame contains at most C types, when any defect exists, the dimension value of the corresponding class in the type activation vector is 1, and the subsequent orientation vector is meaningful, when any defect does not exist, the dimension value of the corresponding class in the type activation vector is 0, and the product of the subsequent orientation vector and the type activation vector is 0, K is the total number of vibration sensors; D∈R K×Z×C×3 is the azimuth vector of the DOA vector, R K×Z×C×3 The C dimension represents the total number of defect types, and the 3 dimensions correspond to the xyz axes of the K Cartesian coordinate system.
9. A shield machine main drive transmission gear monitoring and fault diagnosis device, characterized in that: include: A processing unit, a storage unit and a collection unit interconnected by a bus, wherein the collection unit is used for vibration signals, wherein the storage unit stores a computer program and the vibration signals collected by the collection unit, and when the processing unit reads the vibration signals and executes the computer program, the shield machine main drive transmission gear monitoring and fault diagnosis method as described in any one of claims 1-8 is implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by the processor, the shield machine main drive transmission gear monitoring and fault diagnosis method as described in any one of claims 1-8 is implemented.
