A TBM cutter wear prediction method driven by mechanism and data

Abstract

The present invention discloses a TBM cutter wear prediction method driven jointly by mechanism and data, belonging to the field of cutter wear prediction. On the one hand, the present invention derives a cutter wear calculation formula from wear mechanisms such as abrasive wear, adhesive wear and fatigue wear between the cutter and the rock mass. On the other hand, a data model is used to model the residual part of the mechanism wear and the actual wear, and the TBM excavation parameters are also used as factors affecting the cutter wear. The prediction method of the present invention fully combines the respective advantages of the theoretical mechanism model and the data model, and can accurately predict the actual wear degree of each cutter based on TBM specification parameters and on-site engineering data. It has better generalization characteristics than pure data models and better fitting accuracy than pure mechanism models, provides assistance to on-site operators in mastering cutter information, provides data basis for timely cutter replacement, and improves the utilization rate of the cutter and the working efficiency of the tunnel boring machine.

Description

Technical Field

[0001] The present invention belongs to the field of cutter wear prediction in the technical field of TBM excavation, and in particular relates to a TBM cutter wear prediction method jointly driven by mechanism and data. Background Art

[0002] As an important component of TBM, the cutterhead tool has the characteristics of huge load, wide range of variation, and strong randomness when working. Therefore, it is very easy to be damaged during construction. Excessive wear of the cutter seriously affects construction efficiency, and opening the cutter to change the cutter can easily cause engineering accidents such as instability of the excavation face and tunnel collapse, increasing construction risks and costs. Relevant data shows that the cost of tool consumption accounts for 20% to 30% of the total cost, which is the largest proportion of the consumption of accessories. The time spent on tool maintenance and replacement accounts for about two-thirds of the downtime maintenance time. Therefore, it is particularly important to understand the wear status of the tool during the excavation process. At present, the wear detection method of the tool has not been automated. In actual projects, most still use the method of regular shutdown for manual inspection. The sensor monitoring method is limited by conditions such as high cost, limited installation space and harsh working environment. Therefore, current research focuses on tool wear prediction methods, which are mainly divided into mechanism models and data models. The mechanism model starts from the mechanical analysis or friction energy of the tool when breaking rock, and has good interpretability and extrapolation properties. The data model starts from the perspective of geological parameters or excavation parameters (propulsion speed, cutterhead speed, cutterhead torque, propulsion force, etc.) to fully explore the information in the excavation data.

[0003] Because the interaction between tools and rock is quite complex, most mechanistic models only consider a single wear mechanism, ignoring other wear mechanisms and their interactions during the wear process. This often results in low prediction accuracy and poor application in actual engineering. The extensive geological information required in data models often relies on geological survey reports and manual testing, lacking consideration of theoretical mechanisms. The relationship between model input and output often contradicts actual physical relationships. Furthermore, data models are entirely dependent on the performance of the acquired data, resulting in poor interpretability and extrapolation. Summary of the Invention

[0004] The present invention aims to overcome the shortcomings of existing technologies by proposing a TBM cutter wear prediction method driven by both mechanisms and data. This proposed method combines the advantages of both existing models, providing a reliable reference for accurately identifying cutter failure points.

[0005] The technical solutions of the present invention are as follows:

[0006] The present invention provides a TBM cutter wear prediction method driven by a combination of mechanism and data, which comprises the following steps:

[0007] 1) Calculate the vertical load F on the cutterhead hob n n ;

[0008] 2) Calculate the wear amount Q caused by the wear mechanism based on the vertical load on the hob t The wear caused by the wear mechanism includes the abrasive wear caused by the friction between the hob and the hard rock δ Abr , Adhesive wear caused by shedding of hard abrasive particles δ Adh , fatigue wear of the cutter and rock under alternating contact stress δ Fat ;

[0009] 3) Construct and train a data model; the data model is based on propulsion speed v, cutter head speed n, cutter head torque T c , total propulsion force F, hob speed n s As input, the actual wear amount and the wear amount caused by the wear mechanism Q t The difference is the output, and historical data is used for training;

[0010] 4) Establish a TBM cutter wear prediction model to predict the actual wear degree of each cutter on the TBM cutterhead based on real-time data.

[0011] As a preferred embodiment of the present invention, in step 3), the training of the data model includes the following steps:

[0012] Step 1: Process the actual construction history data, including filtering out outliers, using a binary function to filter out non-working points, and converting the time series data into distance series data;

[0013] Step 2: Select the data model to propel the speed v, cutter head speed n, and cutter head torque T c , total propulsion force F, hob speed n s As input, the actual wear amount and the wear amount Q caused by the wear mechanism calculated in step 2) are t The difference is the output for model training;

[0014] Step 3: Use the distance sequence data obtained in step 1 to train the data model in step 2.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] The hybrid prediction method proposed in this invention fully combines the respective advantages of the theoretical mechanism model and the data model. It can accurately predict the actual wear degree of each disc cutter based on TBM specification parameters and on-site engineering data. It has better generalization characteristics than pure data models and better fitting accuracy than pure mechanism models. It helps on-site operators to grasp tool information and provides data basis for timely tool change, thereby improving the utilization rate of the disc cutter and the working efficiency of the tunnel boring machine. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is the overall flow chart of the method of the present invention;

[0018] Figure 2 The wear prediction results of the 420-444 ring No. 49 hob in the embodiment are shown. DETAILED DESCRIPTION

[0019] The present invention will be further described and illustrated below in conjunction with specific embodiments. The technical features of each embodiment of the present invention can be combined accordingly without conflict.

[0020] like Figure 1 The figure shows a schematic diagram of the overall process of the method of the present invention, which uses the constructed TBM cutter wear prediction model to predict wear, wherein the TBM cutter wear prediction model includes two parts: a mechanism model and a data model. The mechanism model comprehensively considers the wear mechanism under the influence of factors such as abrasive wear, adhesive wear and fatigue wear. The data model mainly uses on-site TBM excavation parameters as input and is constructed based on a machine learning model to supplement the theoretical calculation formula with detailed geological and rock-machine interaction information. The method proposed in the present invention fully considers the actual theoretical mechanism of cutter wear and on-site data information, has high prediction accuracy, and the prediction results are consistent with the actual physical relationship. The interpretability and extrapolation are also greatly improved.

[0021] like Figure 1 As shown, the process of the method of the present invention is as follows:

[0022] First, calculate the vertical load F on the hob n n .

[0023] Secondly, calculate the hob wear Q caused by various wear mechanisms respectively t , including the abrasive wear caused by friction between the hob and hard rock δ Abr , adhesive wear caused by the shedding of hard abrasive particles δ Adh , fatigue wear of the cutter and rock under alternating contact stress δ Fat , whose formula is as follows:

[0024] Q t=aδ Abr +bδ Adh +cδ Fat

[0025] Where Q t is the wear amount caused by the wear mechanism, a, b, and c are the weights of abrasive wear, adhesive wear, and fatigue wear, respectively, which are identified from field data.

[0026] Finally, train the propulsion speed v, cutter head speed n, and cutter head torque T c , total propulsion force F, hob speed n s The data model with the difference between actual wear amount and mechanism wear amount as input is output, which serves as an effective supplement to the calculation model.

[0027] The TBM cutter wear prediction model of the present invention is shown as follows:

[0028] Q i =aδ Abr +bδ Adh +cδ Fat +Δ

[0029] Where Q i is the total wear of the i-th disc cutter on the cutterhead after excavation L, in mm; Δ is the residual term calculated by the data model, in mm.

[0030] In this embodiment, The calculation formula is as follows. This calculation model is based on the mechanical model (CSM) proposed by the Colorado School of Mines and is simplified using similar calculations:

[0031]

[0032] Where C is a dimensionless parameter; ψ is the pressure distribution coefficient on the hob blade, usually in the range of -0.2 to 0.2; S is the distance between adjacent hobs, in mm; R is the radius of the disc hob, in mm; σ c is the rock compressive strength, unit: MPa; σ t is the tensile strength of rock, in MPa; p is the penetration rate, in mm; T is the tool tip width, in mm; β is the distribution angle of the tool, in rad.

[0033] Furthermore, the abrasive wear amount δ Abr The calculation formula is as follows:

[0034]

[0035] Where N i is the number of circles that the i-th knife has turned, Where L is the TBM excavation distance, unit is mm, Ri is the installation radius of the i-th cutter, in mm, R is the radius of the disc hob, in mm; K Abr is the abrasive wear coefficient, Where θ is the half angle of the cone in the micro-cutting hypothesis, unit is rad, K is the probability constant, which depends on the abrasive size and material properties, etc.; σ s is the compressive yield limit, in MPa; l is the distance traveled by the tool in one circle, Unit: mm.

[0036] The adhesive wear amount δ Adh The calculation formula is as follows:

[0037]

[0038] Where K Adh is the adhesive wear coefficient, which depends on the material and friction conditions.

[0039] The fatigue wear amount δ Fat The calculation formula is as follows:

[0040]

[0041] Where K Fat is the fatigue wear coefficient, K Fat =1 / n Fat , where n Fat It is the number of stress cycles that produce fatigue failure, that is, the number of hob circles.

[0042] The data model of the present invention primarily uses tunneling parameters to calculate residual wear, i.e., the residual term in the TBM cutter wear prediction model. The data model can be constructed using machine learning models such as multivariate regression (MLR), support vector regression (SVR), and back propagation neural networks (BPNN). Model training can be divided into the following steps:

[0043] Step 1: Process the actual construction data, including filtering out outliers (this embodiment uses isolation forest to filter out outliers), using a binary function to filter out non-working points, and converting the time series data into distance series data.

[0044] The binary function is used to determine whether a point is in the excavation working state, and the expression is as follows:

[0045]

[0046] D=f(F)·f(v)·f(T c )·f(n)

[0047]

[0048] Where f(x) is a binary function, F is the total propulsion force, v is the propulsion speed, T c is the cutter head torque, n is the cutter head speed, and D is the state discriminant function.

[0049] The sequence conversion uses the following expression:

[0050]

[0051] Where, L j is the excavation mileage within a certain sampling period, v j Indicates the advancement speed at a certain sampling moment, t j+1 -t j Indicates a sampling period.

[0052] Step 2: Select the data model to propel the speed v, cutter head speed n, and cutter head torque T c , total propulsion force F, hob speed n s As input, the difference between actual wear amount and mechanism wear amount is used as output for model training.

[0053] Step 3: Use the trained data model to calculate the corresponding residual value.

[0054] The following describes the specific use of the cutter wear calculation method of the present invention in conjunction with an example to demonstrate the practicality and accuracy of the present invention. The excavation state of cutter No. 49 in the 529-547 ring of a subway project is used as an example.

[0055] Step 1: Substitute the above equations into the simplified formula to get the wear amount Q of the hob mechanism. t It can be calculated by the following expression:

[0056]

[0057]

[0058] Where C is a dimensionless parameter, which is taken as 0.02 in this example; L is the excavation distance, in mm; ψ is the pressure distribution coefficient on the cutter blade, which is usually in the range of -0.2 to 0.2, and is taken as 0.1 in this example; S is the distance between adjacent cutters, in mm, and is taken as 100 in this example; σ s is the compressive yield strength, in MPa, in this case it is 1853.85; σ c is the rock compressive strength, in MPa. In this example, rings 529 to 547 contain three types of rocks: slightly weathered crushed rock, slightly weathered slate, and moderately slightly weathered crushed rock, with compressive strengths of 46.4, 54.6, and 41.9, respectively; σ t is the tensile strength of rock, unit MPa, which is 1 / 10 of the compressive strength; β is the installation angle of the tool, unit rad; KAbr is the abrasive wear coefficient, in this case it is 4.5×10 -3 ; R i is the installation radius of the i-th cutter, in mm, in this example, it is 4970; R is the radius of the disc cutter, in mm, in this example, it is 241.5; p is the penetration, in mm; T is the tip width, in mm, in this example, it is 20; K Adh is the adhesive wear coefficient, which depends on the material and friction conditions. In this example, K Adh ;K Fat is the fatigue wear coefficient, K Fat =1 / n Fat , where n Fat is the number of stress cycles that produce fatigue failure, that is, the number of hob revolutions; a, b, and c are the weight coefficients of abrasive wear, adhesive wear, and fatigue wear, respectively, obtained by fitting and identifying field data. In this example, a is 0.8, b is 0.14, and c is 0.06;

[0059] After substituting each determined parameter, the wear amount Q can be obtained. t The numerical relationship between the excavation distance L and the penetration degree p is used to calculate the mechanical wear amount at each point of the 529-547 rings.

[0060] Step 2: Using propulsion speed v, cutter head speed n, cutter head torque T c , total propulsion force F, tool speed n s As input, the difference between the actual wear amount and the mechanical wear amount is used as the output training data model. In this embodiment, a back propagation neural network (BPNN) is selected as the training model for the residual term.

[0061] 1) For on-site construction data, the isolation forest method is used to screen out outliers, and the binary function is used to remove non-working points, and the time series data is converted into distance series data.

[0062] 2) Determine the input and output vectors. The input features include propulsion speed v, cutter head speed n, and cutter head torque T. c , total propulsion force F, tool speed n s The difference between actual wear and mechanical wear is used as the model output. The model was trained using 80% of the data from the No. 49 tool as the training set, and 20% of the data as the validation set. The model was fine-tuned to include four hidden layers, with 8, 10, 12, and 14 neurons in each layer. Using the Adam optimizer, the model achieved MAE = 0.39 and RMSE = 0.53 on the training set, and MAE = 1.19 and RMSE = 1.60 on the validation set.

[0063] 3) Using the trained data model, by inputting propulsion speed v, cutter head speed n, cutter head torque T c , total propulsion force F, tool speed ns Calculate the residual value of each point;

[0064] Step 3: Add the values ​​obtained in steps 1 and 2 to calculate the wear amount at that point, and realize the prediction of the wear amount. The final prediction result of tool No. 49 at rings 420-444 is as follows: Figure 2 As shown, the wear at other excavation locations in the project can be calculated analogously to this embodiment. This invention fully combines the advantages of both theoretical and data-based models, enabling high-precision prediction of the actual wear of each cutter based on TBM specifications and field engineering data. It offers better generalization than purely data-based models and greater fitting accuracy than purely mechanistic models.

[0065] The above-described embodiments merely represent several implementation methods of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims.

Claims

1. A TBM cutter wear prediction method driven by both mechanism and data, characterized in that: The steps include: 1) Calculate the vertical load F on the cutterhead hob n n ; Where C is a dimensionless parameter; ψ is the pressure distribution coefficient on the hob blade; S is the distance between the hobs, in mm; R is the radius of the disc hob, in mm; σ c is the uniaxial compressive strength of rock, unit: MPa; σ t is the rock tensile strength, unit MPa; p is the penetration, unit mm; T is the tool tip width, unit mm; β is the tool distribution angle, unit rad; 2) Calculate the wear amount Q caused by the wear mechanism based on the vertical load on the hob t The wear caused by the wear mechanism includes the abrasive wear caused by the friction between the hob and the hard rock δ Abr , Adhesive wear caused by shedding of hard abrasive particles δ Adh , fatigue wear of the cutter and rock under alternating contact stress δ Fat ; 3) Construct and train a data model; the data model is based on propulsion speed v, cutter head speed n, cutter head torque T c , total propulsion force F, hob speed n s As input, the actual wear amount and the wear amount caused by the wear mechanism Q t The difference is the output, and historical data is used for training; Data model training includes the following steps: Step 1: Process the actual construction history data, including filtering out outliers, using a binary function to filter out non-working points, and converting the time series data into distance series data; Step 2: Select the data model to propel the speed v, cutter head speed n, and cutter head torque T c , total propulsion force F, hob speed n s As input, the actual wear amount and the wear amount Q caused by the wear mechanism calculated in step 2) t The difference is the output for model training; Step 3: Use the distance sequence data obtained in step 1 to train the data model of step 2; 4) Establish a TBM cutter wear prediction model to predict the actual wear degree of each cutter on the TBM cutterhead based on real-time data.

2. The TBM cutter wear prediction method driven by mechanism and data according to claim 1 is characterized in that: In the step 2), After the i-th cutter in the cutter head has penetrated L, the abrasive wear caused by friction with the hard rock δ Abr , the calculation formula is as follows: Where N i is the number of circles that the i-th knife has turned, Where L is the TBM excavation distance, unit is mm, R i is the installation radius of the i-th cutter, in mm, R is the radius of the disc hob, in mm; K Abr is the abrasive wear coefficient, Where θ is the half angle of the cone in the micro-cutting hypothesis, unit is rad, K is the probability constant; σ s is the compressive yield limit, unit is MPa; l is the distance traveled by the hob in one circle, unit is mm.

3. The TBM cutter wear prediction method driven by mechanism and data according to claim 2 is characterized in that: In step 2), after the i-th cutter in the cutter head has penetrated L, the adhesive wear amount δ caused by the shedding of hard abrasive particles is Adh , the calculation formula is as follows: Where K Adh is the adhesive wear coefficient.

4. The TBM cutter wear prediction method driven by mechanism and data according to claim 3 is characterized in that: In step 2), after the cutter head i has excavated L, the fatigue wear amount δ of the cutter head i under the alternating contact stress with the rock and soil is Fat , the calculation formula is as follows: Where K Fat is the fatigue wear coefficient, K Fat =1 / n Fat , where n Fat It is the number of stress cycles that produce fatigue failure, that is, the number of hob circles.

5. The TBM cutter wear prediction method driven by mechanism and data according to claim 4 is characterized in that: In the step 3), the data model is a machine learning model selected from a multiple linear regression model, a support vector machine regression model, and a back propagation neural network.

6. The TBM cutter wear prediction method driven by mechanism and data according to claim 5 is characterized in that: The use of a binary function to screen out non-working points is to remove non-working state data in the historical data. The binary function is used to determine whether a point is in the excavation working state. The expression is as follows: D=f(F)·f(v)·f(T c )·f(n) Where f(x) is a binary function, F is the total propulsion force, v is the propulsion speed, T c is the cutter head torque, n is the cutter head speed, and D is the state discriminant function.

7. The TBM cutter wear prediction method driven by a combination of mechanism and data according to claim 6 is characterized in that: The following expression is used to convert time series data into distance series data: Where, L j is the excavation mileage within a certain sampling period, v j Indicates the advancement speed at a certain sampling moment, t j+1 -t j Indicates a sampling period.

8. The TBM cutter wear prediction method driven by mechanism and data according to claim 7 is characterized in that: In step 4), the TBM cutter wear prediction model is as follows: Q i =aδ Abr +bδ Adh +cδ Fat +D Where Q i is the total wear of the i-th disc cutter on the cutterhead after excavation L, in mm; a, b, c are the weights of the corresponding items, respectively; Δ is the residual term calculated by the data model, in mm.

Images