A real-time prediction method and system for TBM cutter wear based on neural network

Abstract

The present invention provides a real-time prediction method and system for TBM cutter wear based on a neural network, which obtains field data of each cutter; obtains a time series prediction value of each cutter wear amount based on the obtained field data and a preset TBM cutter wear time series prediction model; within each cutter life cycle, accumulates the historical cutter wear amount and the predicted cutter wear amount in each time period to obtain the total wear amount; determines the predicted cutter wear state and cutter replacement time based on the total wear amount; based on LSTM, the present invention comprehensively considers the field information and historical information of field excavation data, effectively extracts the sequence change information of the data, establishes a time series prediction model for predicting TBM cutter wear amount, uses field excavation data to realize intelligent prediction and information management of cutter wear, predicts the cutter wear state and cutter replacement time, avoids the shield machine from stopping under a complex rock formation, and circumvents the occurrence of adverse geological problems.

Description

Technical Field

[0001] The present invention relates to the technical field related to TBM excavation, and in particular to a real-time prediction method and system for TBM cutter wear based on a neural network. Background Art

[0002] TBM tunnel construction is characterized by high construction speed, high quality and low pollution, but its adaptability to complex and changing strata is insufficient. During TBM excavation in hard rock strata, the cutterhead disc wears constantly and requires regular inspection and replacement of the discs. On the one hand, cutter inspection and replacement takes up a lot of construction time and hinders construction progress. On the other hand, the large amount of manpower and material resources invested is also a major factor in increasing project costs. Therefore, in order to better utilize the efficiency advantages of TBM work, reduce cutterhead inspection time, and improve the intelligence of TBM disc inspection and replacement management, it is urgent to predict the wear degree of TBM discs in real time.

[0003] Currently, TBM cutter wear is monitored using instruments mounted on the cutterhead, such as odor additives, lasers, and eddy current sensors. However, these methods are difficult to implement on complex construction sites and rely primarily on the judgment of on-site engineers. This results in low equipment efficiency and a long construction cycle. Furthermore, in complex and changing geological environments, if cutter wear necessitates cutter replacement, resulting in TBM downtime, the likelihood of causing adverse geological disasters increases, seriously endangering the safety of the TBM and construction personnel. TBM excavation parameters vary continuously over time, and field excavation data can effectively reflect rock formation information. Therefore, field excavation data can be used to analyze and study TBM cutter wear.

[0004] The inventors found that there are currently few methods for predicting cutterhead cutter wear based on excavation data. These methods do not fully utilize the field information and historical information of on-site TBM excavation data, making it difficult to achieve real-time prediction. Summary of the Invention

[0005] To address these issues, the present invention proposes a real-time prediction method and system for TBM cutter wear based on a neural network. This method comprehensively considers both the domain and historical information of on-site TBM excavation data, effectively extracting information about sequential changes in the data to accurately and timely predict cutter wear, reducing cutter maintenance time and further leveraging the TBM's operational efficiency. With the continuous expansion of the database, the accuracy of cutter wear prediction will also be improved. This allows for timely cutter replacement before traversing complex and changing geological environments, thus preventing the occurrence of adverse geological issues.

[0006] In a first aspect, the present invention provides a real-time prediction method for TBM cutter wear based on a neural network, comprising:

[0007] Get on-site data of each hob;

[0008] Based on the acquired field data and a preset TBM cutter wear time series prediction model, a time series prediction value of each cutter wear amount is obtained; wherein the TBM cutter wear time series prediction model is obtained based on long short-term memory neural network training;

[0009] During each hob life cycle, the historical hob wear and predicted hob wear in each period are accumulated to obtain the total wear; based on the total wear, the predicted hob wear state and hob replacement time are determined.

[0010] Furthermore, the training process of the TBM cutter wear time series prediction model is as follows:

[0011] Obtaining time series data of tunneling parameters and wear data of each cutterhead cutter of the TBM during tunneling; performing time series processing on each cutterhead wear data to obtain a time series value of wear of each cutterhead cutter, wherein the time series value of wear of each cutterhead cutter corresponds to the time series data of tunneling parameters;

[0012] Establish a data sample library including the time series data of tunneling parameters and the time series value of each cutter wear of the cutterhead;

[0013] Based on the established data sample library and long short-term memory neural network, a TBM cutter wear time series prediction model was trained.

[0014] Furthermore, the excavation parameters include at least one or more of excavation speed, cutterhead torque, cutterhead thrust, cutterhead rotation speed and cutter penetration;

[0015] The wear data of each hob cutter of the cutter disc includes at least one or more of a hob cutter installation radius, a hob cutter wear height, a hob cutter replacement and maintenance position, and a hob cutter replacement time.

[0016] Furthermore, time series processing is performed on each disc cutter wear data, which is to make the disc cutter wear data correspond to the excavation parameter data one by one through the pile number and the disc cutter replacement and maintenance time.

[0017] Furthermore, in the data sample library, the time intervals for obtaining the excavation parameters and the disc cutter wear data are the same; the missing disc cutter wear data are supplemented by interpolation.

[0018] Furthermore, the effectiveness of the model was evaluated using three indicators: relative error rate between the predicted value and the true value, mean absolute percentage error, and goodness of fit.

[0019] Furthermore, the wear status of the hob is determined based on the total wear of each hob, which is based on the correspondence between the wear amount and the wear height of the hob; for hobs of different types and different positions, when the predicted wear height exceeds its defined wear height, it can be determined that the hob has failed and needs to be replaced.

[0020] In a second aspect, the present invention further provides a real-time prediction system for TBM cutter wear based on a neural network, comprising:

[0021] The data acquisition module is configured to: obtain on-site data of each hob;

[0022] The prediction module is configured to obtain a time series prediction value of each cutter wear amount based on the acquired field data and a preset TBM cutter wear time series prediction model; wherein the TBM cutter wear time series prediction model is obtained based on long short-term memory neural network training;

[0023] The wear determination module is configured to: accumulate the historical hob wear and the predicted hob wear in each period during each hob life cycle to obtain the total wear; and determine the predicted hob wear state and hob replacement time based on the total wear.

[0024] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the real-time prediction method for TBM cutter wear based on a neural network described in the first aspect are implemented.

[0025] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the neural network-based TBM cutter wear real-time prediction method described in the first aspect.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] The present invention is based on a long short-term memory (LSTM) neural network, comprehensively considers the domain information and historical information of on-site excavation data, effectively extracts the sequence change information of the data, and establishes a time series prediction model for predicting the wear of TBM cutters. It uses on-site excavation data to realize intelligent prediction and information management of cutter wear, predicts the wear state of the cutters and the time for cutter replacement, avoids the shield machine from shutting down under a complex rock formation, and circumvents the occurrence of adverse geological problems. This method can improve the TBM construction progress and reduce construction costs, and give full play to the excavation efficiency of the TBM. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings constituting a part of the specification of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments and descriptions of this embodiment are used to explain this embodiment and do not constitute an improper limitation on this embodiment.

[0029] Figure 1 This is a flow chart of Example 1 of the present invention;

[0030] Figure 2 This is a model framework diagram of Example 1 of the present invention. DETAILED DESCRIPTION

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present application belongs.

[0033] Example 1:

[0034] like Figure 1 As shown, the present invention provides a real-time prediction method for TBM cutter wear based on a neural network, comprising:

[0035] Get on-site data of each hob;

[0036] Based on the acquired field data and a preset TBM cutter wear time series prediction model, a time series prediction value of each cutter wear amount is obtained; wherein the TBM cutter wear time series prediction model is obtained based on long short-term memory neural network training;

[0037] During each hob life cycle, the historical hob wear and predicted hob wear in each period are accumulated to obtain the total wear; based on the total wear, the predicted hob wear state and hob replacement time are determined.

[0038] In this embodiment, the neural network uses a long short-term memory (LSTM) neural network. First, the main influencing parameters of cutter wear are collected and screened, and the time series data of the excavation parameters of the TBM during tunnel excavation are collected to reflect the rock formation conditions and characterize the rolling distance of the cutter on the tunnel face. The wear of each cutter on the cutterhead is collected and time series processing is performed on it to make the cutter wear data correspond to the excavation parameters in time. A data sample library is established based on the cutter wear time series data and the TBM excavation parameter time series data, and the data sample library is divided into a training set and a test set. A TBM cutter wear time series prediction model is established based on the LSTM network. The prediction model is trained using the training and test sets in the sample database. The prediction model parameters are updated and debugged using the back propagation through time (BPTT) algorithm until the accuracy and goodness of fit of the model prediction results meet the error requirements of the engineering example. The model evaluation indicators are relative error rate, mean absolute percentage error, and goodness of fit. The field measured data of the disc cutter to be predicted is input into the TBM disc cutter wear time series prediction model to predict the cutterhead disc cutter wear value time series data, that is, the disc cutter wear amount in the prediction period. The disc cutter wear amount in each period of each disc cutter life cycle is accumulated to obtain the total wear amount, and the disc cutter wear status is determined based on the total wear amount.

[0039] like Figure 1 As shown, the specific contents of this embodiment are as follows:

[0040] Analyze and screen the main influencing parameters of cutterhead cutter wear to reflect rock formation conditions and characterize the rolling distance of the cutter on the tunnel face. Tunneling parameters are derived from TBM operating parameters, including at least tunneling speed, cutterhead torque, cutterhead thrust, cutterhead speed, and cutter penetration. Wear data is derived from cutter parameters, including at least cutter installation radius, cutter wear, cutter replacement and maintenance location, and cutter replacement time.

[0041] Collect time series data on the excavation parameters of the TBM during tunnel excavation; collect wear data on each cutter on the cutterhead, perform time series processing on the cutter wear data, and ensure that the cutter wear data corresponds to the excavation parameter data in time;

[0042] Establish a sample database. Create a data sample library based on the cutter wear time series data and TBM excavation parameter time series data, and divide the data sample library into training sets and test sets.

[0043] A time series prediction model for TBM cutter wear is established based on the LSTM neural network. The prediction model uses the future value of each cutter wear amount as the output variable; the historical values ​​of TBM excavation speed, cutter torque, cutter thrust, cutter speed, cutter penetration, cutter installation radius and cutter wear amount as the input variables. The training set and test set in the sample database are used to train the prediction model, update the parameters and debug it.

[0044] The cutter wear time series data prediction is to input the field measured data of each cutter to be predicted into the TBM cutter wear time series prediction model to obtain the cutter wear time series data, that is, the cutter wear data for the prediction period;

[0045] Evaluation and determination of TBM cutter wear: The cutter wear in each period (historical period and predicted period) is accumulated to obtain the total wear within each roller life cycle. The cutter wear status is determined based on the total wear, and the cutter replacement time is determined at the same time.

[0046] In this embodiment, when analyzing and screening the main influencing parameters of cutterhead cutter wear, factors that have little impact on TBM cutter wear may be ignored, and the analysis is mainly focused on the parameters that have a greater impact on cutter wear.

[0047] In this embodiment, the amount of cutter wear is calculated using the cutter wear height, which can be collected by on-site construction personnel during cutter replacement or maintenance, or obtained through an online cutter wear monitoring device;

[0048] In this embodiment, the time series processing of the cutter wear refers to the one-to-one correspondence between the cutter wear data and the excavation parameter data through the pile number and the cutter replacement and maintenance time.

[0049] In this embodiment, when establishing a sample database, the TBM excavation data and the cutter wear data in the database should be collected at the same time interval; the same time interval can be set by using an online cutter wear monitoring device; when manually collecting on-site, the collected cutter wear data needs to be divided into the same time interval for processing, and the missing cutter wear data in the middle is supplemented by interpolation to make them correspond one-to-one with the excavation data in time.

[0050] In this embodiment, the ratio of the training set to the test set in the database can be 8:2, or other ratios, to obtain the best model output results.

[0051] In this embodiment, if Figure 2 As shown in the figure, the prediction model parameters are updated and debugged through the back propagation through time (BPTT) algorithm until the accuracy and goodness of fit of the model prediction results meet the error requirements of the engineering example. The model evaluation indicators are relative error rate, mean absolute percentage error and goodness of fit. In the figure, X1, X2 and X L Represents the input time series sample; LSTM L Represents the L-th LSTM cell structure in the hidden layer; C L-1 Indicates the state of the L-1th cell; H L-1 represents the cell output at time L-1; P1, P2 and P L Represents the corresponding rolling wear prediction results.

[0052] In this embodiment, the wear state of the hob and the time to replace the hob are determined based on the total wear of each hob, which is achieved based on the correspondence between the wear amount and the wear height of the hob. For hobs of different types and positions, when the wear height exceeds the defined wear height, it can be determined that the hob has failed and needs to be replaced; the predicted moment corresponding to the wear height exceeding the defined wear height is the time to replace the hob. It can be understood that the total wear amount is obtained by accumulating the hob wear amount of each time period (historical time period and predicted time period) within each hob life cycle, and the time period corresponding to the time when the hob is determined to be failed is the time to replace the hob.

[0053] In this embodiment, since the cutterhead wear is affected by another factor, namely the quartz content in the rock, in order to obtain accurate prediction results, the model is best applied to hard rock. For complex and changeable rock formations, a large amount of data is required to train the model and debug the parameters to obtain good prediction results.

[0054] Example 2:

[0055] This embodiment provides a real-time prediction system for TBM cutter wear based on a neural network, including:

[0056] The data acquisition module is configured to: obtain on-site data of each hob;

[0057] The prediction module is configured to obtain a time series prediction value of each cutter wear amount based on the acquired field data and a preset TBM cutter wear time series prediction model; wherein the TBM cutter wear time series prediction model is obtained based on long short-term memory neural network training;

[0058] The wear determination module is configured to: accumulate the historical hob wear and the predicted hob wear in each period during each hob life cycle to obtain the total wear; and determine the predicted hob wear state and hob replacement time based on the total wear.

[0059] Example 3:

[0060] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the method for real-time prediction of TBM cutter wear based on a neural network described in Example 1 are implemented.

[0061] Example 4:

[0062] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the method for real-time prediction of TBM cutter wear based on a neural network described in Example 1 are implemented.

[0063] The above description is merely a preferred embodiment of this embodiment and is not intended to limit this embodiment. Those skilled in the art will readily appreciate that this embodiment may be modified and varied in various ways. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this embodiment shall be within the scope of protection of this embodiment.

Claims

1. A real-time prediction method for TBM cutter wear based on neural network, characterized in that: include: Get on-site data of each hob; Based on the acquired field data and a preset TBM cutter wear time series prediction model, a time series prediction value of each cutter wear amount is obtained; wherein the TBM cutter wear time series prediction model is obtained based on long short-term memory neural network training; The training process of the TBM cutter wear time series prediction model is as follows: obtaining the time series data of the excavation parameters of the TBM during tunnel excavation, and the wear data of each cutter of the cutter head; The wear data for each cutter is processed in time series to obtain the wear time series value for each cutterhead cutter. The wear time series value for each cutterhead cutter corresponds to the tunneling parameter time series data. A data sample library containing the tunneling parameter time series data and the wear time series value for each cutterhead cutter is established. Based on the established data sample library and a long-short-term memory neural network, a TBM cutter wear time series prediction model is trained. The excavation parameters include at least one or more of excavation speed, cutterhead torque, cutterhead thrust, cutterhead rotation speed, and cutter penetration; the wear data of each cutter of the cutterhead includes at least one or more of cutter installation radius, cutter wear height, cutter replacement and maintenance position, and cutter replacement time; During each hob life cycle, the historical hob wear and predicted hob wear in each period are accumulated to obtain the total wear; based on the total wear, the predicted hob wear state and hob replacement time are determined.

2. The method for real-time prediction of TBM cutter wear based on neural network according to claim 1, characterized in that: The time series processing of each disc cutter wear data is to make the disc cutter wear data correspond to the excavation parameter data one by one through the pile number and the disc cutter replacement and maintenance time.

3. The method for real-time prediction of TBM cutter wear based on neural network according to claim 1, characterized in that: In the data sample library, the time intervals for obtaining the excavation parameters and the disc cutter wear data are the same; the missing disc cutter wear data are supplemented by interpolation.

4. The method for real-time prediction of TBM cutter wear based on neural network according to claim 1, characterized in that: The effectiveness of the model was evaluated using three indicators: relative error rate between the predicted value and the true value, mean absolute percentage error, and goodness of fit.

5. The method for real-time prediction of TBM cutter wear based on neural network according to claim 1, characterized in that: The wear status of each hob is determined based on the total wear of each hob, which is based on the corresponding relationship between the wear amount and the wear height of the hob. For hobs of different types and different positions, when the predicted wear height exceeds the defined wear height, the hob can be determined to have failed and needs to be replaced.

6. A real-time prediction system for TBM cutter wear based on neural network, characterized in that: include: The data acquisition module is configured to: obtain on-site data of each hob; The prediction module is configured to obtain a time series prediction value of each cutter wear amount based on the acquired field data and a preset TBM cutter wear time series prediction model; wherein the TBM cutter wear time series prediction model is obtained based on long short-term memory neural network training; the training process of the TBM cutter wear time series prediction model is to obtain time series data of tunneling parameters of the TBM during tunneling and wear data of each cutter of the cutterhead; The wear data for each cutter is processed in time series to obtain the wear time series value for each cutterhead cutter. The wear time series value for each cutterhead cutter corresponds to the tunneling parameter time series data. A data sample library containing the tunneling parameter time series data and the wear time series value for each cutterhead cutter is established. Based on the established data sample library and a long-short-term memory neural network, a TBM cutter wear time series prediction model is trained. The excavation parameters include at least one or more of excavation speed, cutterhead torque, cutterhead thrust, cutterhead rotation speed, and cutter penetration; the wear data of each cutter of the cutterhead includes at least one or more of cutter installation radius, cutter wear height, cutter replacement and maintenance position, and cutter replacement time; The wear determination module is configured to: accumulate the historical hob wear and the predicted hob wear in each period during each hob life cycle to obtain the total wear; and determine the predicted hob wear state and hob replacement time based on the total wear.

7. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for real-time prediction of TBM cutter wear based on a neural network as described in any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium, characterized in that A computer program is stored thereon, which, when executed by a processor, implements the steps of the method for real-time prediction of TBM cutter wear based on a neural network as claimed in any one of claims 1 to 5.

Images