Complex geology shield cutter group layered arrangement and wear prediction model construction method

Abstract

A complex geological shield cutter group layered arrangement and wear prediction model construction method belongs to the field of shield machine cutter wear prediction, and comprises the following steps: obtaining cutter wear correlation factors and historical cutter wear loss, and preprocessing and normalizing the cutter wear correlation factors and the historical cutter wear loss; and by taking the tool wear related factors as input and the historical tool wear loss as output, constructing a tool wear model based on a BP neural network to predict the tool wear loss. A Jacobian matrix is constructed based on model parameters of a tool wear model, and an L-P iteration increment is calculated. And an MSE value is calculated according to the number of iterations and the error between the actual tool abrasion loss and the predicted tool abrasion loss, and iterative updating is conducted on the tool abrasion model based on the MSE value. According to the method, construction of the whole shield cutter group abrasion prediction model is achieved, and accurate prediction of the cutter abrasion loss can be achieved based on the cutter torque, the cutter rotating speed, the shield thrust, the average propelling speed, the penetration degree, the tunneling ring number, the cutter number and the number of the area where the cutter is located.

Description

Technical Field

[0001] This invention belongs to the field of shield machine cutter wear prediction, and more specifically, relates to a method for constructing a layered arrangement and wear prediction model for shield machine cutter groups in complex geological conditions. Background Technology

[0002] Currently, there are two main methods for monitoring tool wear: wear sensing devices and electrically powered tool wear monitoring devices. However, both methods require prior modification of the tool structure and can only be deployed on a subset of tools. The wear condition of these subsets is used to infer the overall wear condition of the tool shroud, relying on human experience and subjective judgment, lacking scientific guidance. Furthermore, modifying the tool structure to achieve wear monitoring functionality affects both the structural strength and service performance of the tool itself, and increases the tool manufacturing cost.

[0003] Therefore, it is essential to establish a wear prediction model for the entire shield cutterhead group based on historical cutter wear data, geological factors, shield tunneling construction parameters, and other factors. Furthermore, based on the cutter wear prediction results, optimization schemes for cutterhead group layout parameters should be proposed to achieve high-performance service of the cutterhead cutters. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to overcome the aforementioned deficiencies and propose a method for constructing a layered arrangement and wear prediction model for shield tunnel cutter groups in complex geological conditions.

[0005] The present invention adopts the following technical solution.

[0006] The first aspect of this invention discloses a method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions, the method comprising:

[0007] Obtain tool wear correlation factors and historical tool wear amounts, and preprocess and normalize the tool wear correlation factors and historical tool wear amounts;

[0008] Using the aforementioned tool wear correlation factors as input and historical tool wear amount as output, a tool wear model based on BP neural network is constructed to predict tool wear amount;

[0009] Based on the model parameters of the tool wear model, a Jacobian matrix is ​​constructed, and the LP iteration increment is calculated;

[0010] The MSE value is calculated based on the number of iterations and the error between the actual tool wear and the predicted tool wear, and the tool wear model is iteratively updated based on the MSE value.

[0011] Furthermore, the method also includes:

[0012] Obtain the number of cutters in the cutter head, and use each cutter as a node to determine the components of each node and the distance between each node and the root node, and construct a balanced binary tree of the cutter head.

[0013] Determine the difference in center distance between different tools on the tool turret, calculate the spatial coupling weight in combination with the spatial attenuation constant, and sort the nodes corresponding to each tool according to the magnitude of the spatial coupling weight;

[0014] The actual tool wear and tool penetration depth of different tools on the cutter head are obtained in sequence to determine the wear degree of each tool on the cutter head. The wear degree is directly proportional to the actual tool wear of the corresponding tool and inversely proportional to the tool penetration depth. It is the weighted result of the actual tool wear and tool penetration depth as a percentage of the ideal tool.

[0015] Based on the balanced binary tree of the cutter head and the wear degree of each tool, sample data and test data of tool wear correlation factors in the cutter head are screened according to the constraints, and the tool wear model is trained and verified through the sample data and test data.

[0016] In this system, the node number corresponds one-to-one with the tool number on the cutter head, and the order of the nodes is consistent with the tool number of each tool. The component is used to characterize the reaction coupling of the tool corresponding to each node from the current cutting load distribution to the local pressure. The expression of the component is:

[0017]

[0018]

[0019] In the formula, and These represent the tool wear when obtaining the actual tool wear amount. With knives Depth of entry and They represent cutting tools With knives The initial depth of cut, The order of representation is The components of the node, This represents the total number of cutting tools.

[0020] Furthermore, the BP neural network includes an input layer, a hidden layer, and an output layer, and the model parameters include a weight matrix from the input layer to the hidden layer, a bias vector of the hidden layer, a weight vector from the hidden layer to the output layer, and a bias vector of the output layer.

[0021] The tool wear correlation factors include multiple elements, namely cutterhead torque, cutterhead rotation speed, shield thrust, average advance speed, penetration depth, number of tunneling rings, and tool number, which are used to generate multiple input layer nodes of the BP neural network, and the historical tool wear amount is used to generate output layer nodes.

[0022] Furthermore, the step of constructing a tool wear model based on a BP neural network, using the tool wear correlation factors as input and historical tool wear data as output, to predict tool wear data includes:

[0023] Obtain the number of each tool wear-related factor, and determine the elements and element lengths corresponding to the number, the elements in the weight matrix, the elements in the bias vector, and the hidden layer number;

[0024] The net input of the hidden layer is calculated based on the elements and element lengths corresponding to the number, the elements in the weight matrix, the elements in the bias vector, and the hidden layer number.

[0025] The hidden layer activation is calculated and output by activating the net input of the hidden layer using the Sigmoid algorithm.

[0026] Furthermore, the step of constructing a tool wear model based on a BP neural network to predict tool wear, using the tool wear correlation factors as input and historical tool wear amount as output, also includes:

[0027] Obtain the elements in the weight vector and the number of hidden layers, and calculate the net input of the output layer based on the elements in the weight vector and the number of hidden layers;

[0028] The output layer net input is activated using the Sigmoid algorithm, and the predicted tool wear is calculated and output.

[0029] Obtain the actual tool wear amount corresponding to the predicted tool wear amount, and calculate the error between the predicted tool wear amount and the actual tool wear amount.

[0030] Furthermore, the construction of the Jacobian matrix based on the model parameters of the tool wear model and the calculation of the LP iteration increment include:

[0031] The cutterhead torque, cutterhead rotation speed, shield thrust, average advance speed, penetration depth, number of tunneling rings, and cutter number are concatenated into a model parameter vector, and each element in the Jacobian matrix is ​​determined.

[0032] Based on the elements in the Jacobian matrix, a damping diagonal matrix is ​​constructed, and the LP iteration increment is calculated according to the model parameter vector and the error between the predicted tool wear and the actual tool wear.

[0033] Furthermore, the step of calculating the MSE value based on the number of iterations and the error between the actual tool wear and the predicted tool wear, and iteratively updating the tool wear model based on the MSE value, includes:

[0034] The model parameter vector is reset based on the MSE value, and the reset model parameter vector is the sum of the model parameter vector before reset and the LP iteration increment;

[0035] When the tool wear model is updated for the first iteration, the weight matrix from the input layer to the hidden layer, the bias vector of the hidden layer, the weight vector from the hidden layer to the output layer, and the bias vector of the output layer are updated according to the reset model parameter vector, and the MSE value is calculated based on the updated model parameters.

[0036] When the tool wear model is not updated in the first iteration, it is determined whether the MSE value of the tool wear model meets the convergence condition. If so, the iteration is terminated.

[0037] The second aspect of this invention discloses a device for constructing a layered arrangement and wear prediction model for a complex geological shield tunnel cutter group, used to implement the method for constructing a layered arrangement and wear prediction model for a complex geological shield tunnel cutter group as described in the first aspect of claim, the device comprising:

[0038] The data preprocessing module is used to obtain tool wear correlation factors and historical tool wear amounts, and to preprocess and normalize the tool wear correlation factors and historical tool wear amounts;

[0039] The model training module is used to construct a tool wear model based on a BP neural network, using the tool wear correlation factors as input and historical tool wear amount as output, in order to predict tool wear amount.

[0040] The iterative increment calculation module is used to construct a Jacobian matrix based on the model parameters of the tool wear model and calculate the LP iterative increment.

[0041] The model iteration update module is used to calculate the MSE value based on the number of iterations and the error between the actual tool wear and the predicted tool wear, and to iteratively update the tool wear model based on the MSE value.

[0042] A third aspect of the present invention discloses a terminal, including a processor and a storage medium; characterized in that:

[0043] The storage medium is used to store instructions;

[0044] The processor is configured to operate according to the instructions to perform the steps of the method described in the first aspect.

[0045] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the method described in the first aspect.

[0046] The beneficial effects of the present invention are as follows: Compared with the prior art, the present invention has the following advantages:

[0047] (1) This invention preprocesses and normalizes the tool wear correlation factors and historical tool wear amount, takes the tool wear correlation factors as input and the historical tool wear amount as output, and constructs a tool wear model based on BP neural network to predict the tool wear amount. This realizes the construction of the entire shield tool group wear prediction model, so that the tool wear can be predicted based on shield tunneling parameters, geological parameters, tool position, etc.

[0048] (2) Based on the construction of the entire shield cutter group wear prediction model, the present invention constructs the Jacobian matrix based on the model parameters of the cutter wear model and calculates the LP iteration increment. Based on the iteration number and the error between the actual cutter wear amount and the predicted cutter wear amount, the MSE value is calculated to iteratively update the constructed cutter wear model. Combined with balanced binary trees and constraints, the sample data and test data are screened to train and verify the cutter wear model. This solves the problem of unbalanced sample data caused by the inability to continuously and quantitatively monitor the cutter wear in the prior art, thereby improving the prediction accuracy of the cutter wear amount of the trained cutter wear model. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the method for constructing a layered arrangement and wear prediction model for shield tunnel cutter groups in complex geological conditions provided by the present invention.

[0050] Figure 2A This is a schematic diagram of the tool arrangement before optimization;

[0051] Figure 2B This is a schematic diagram of the tool path before optimization;

[0052] Figure 3A This is a schematic diagram of the optimized tool arrangement;

[0053] Figure 3B This is a schematic diagram of the optimized toolpath. Detailed Implementation

[0054] The present application will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be construed as limiting the scope of protection of the present application.

[0055] Understandably, the interpretation of subscript-type variables in this invention should be bound to their corresponding variables, thereby limiting the possibility of duplicate definitions. For example: in variables In, variables of the subscript class The definition is the sample number; while the variable In, variables of the subscript class The definition is a matrix The Middle OK.

[0056] like Figure 1 As shown in one embodiment, a method for constructing a layered arrangement and wear prediction model for a complex geological shield cutterhead group includes the following steps:

[0057] Step S110: Obtain tool wear correlation factors and historical tool wear amount, and preprocess and normalize the tool wear correlation factors and historical tool wear amount.

[0058] Step S120: Using tool wear-related factors as input and historical tool wear amount as output, construct a tool wear model based on BP neural network to predict tool wear amount.

[0059] In some embodiments, the method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions provided by the present invention uses a BP neural network comprising an input layer, a hidden layer, and an output layer. Model parameters include a weight matrix from the input layer to the hidden layer, a bias vector for the hidden layer, a weight vector from the hidden layer to the output layer, and a bias vector for the output layer. The factors related to cutter wear include multiple elements, namely cutterhead torque, cutterhead rotation speed, shield thrust, average advance speed, penetration depth, number of tunneling rings, and cutter number, used to generate multiple input layer nodes of the BP neural network. Historical cutter wear data is used to generate output layer nodes.

[0060] In some embodiments, the method for constructing a layered arrangement and wear prediction model for complex geological shield tunnel cutterhead groups provided by the present invention includes the following steps in step S120:

[0061] Step S121: Obtain the number of each tool wear-related factor, and determine the elements corresponding to the number, the element length, the elements in the weight matrix, the elements in the bias vector, and the hidden layer number.

[0062] Step S122: Calculate the net input of the hidden layer based on the elements corresponding to the number, the element length, the elements in the weight matrix, the elements in the bias vector, and the hidden layer number.

[0063] Step S123: Activate the net input of the hidden layer using the Sigmoid algorithm, calculate and output the hidden layer activation.

[0064] In some embodiments, the method for constructing a layered arrangement and wear prediction model for complex geological shield tunnel cutterhead groups provided by the present invention further includes the following steps in step S120:

[0065] Step S124: Obtain the elements in the weight vector and the number of hidden layers, and calculate the net input of the output layer based on the elements in the weight vector and the number of hidden layers.

[0066] Step S125: Activate the net input of the output layer using the Sigmoid algorithm, calculate and output the predicted tool wear amount.

[0067] Step S126: Obtain the actual tool wear amount corresponding to the predicted tool wear amount, and calculate the error between the predicted tool wear amount and the actual tool wear amount.

[0068] In a specific embodiment, the method for constructing a layered arrangement and wear prediction model for shield tunnel cutter groups in complex geological formations provided by this invention, taking a sand-rock composite stratum scenario as an example, includes the following conventional prediction of shield tunnel cutter wear: establishing a cutter wear model using a BP neural network (Backpropagation Neural Network) under the sand stratum. The basic structure of the BP neural network includes three parts: an input layer, a hidden layer, and an output layer. First, it is necessary to establish cutter wear-related factors as the input layer. These factors include: cutterhead torque, cutterhead rotation speed, shield thrust, average advance speed, penetration depth, number of tunneling rings, and cutter number, forming seven input layer nodes for the BP neural network. Since the BP neural network only uses cutter wear as the sole prediction target, the output layer contains only one node, namely, the cutter wear amount. In this example, the LM algorithm (Levenberg–Marquardt Algorithm) is selected as the training algorithm for the BP neural network to construct the BP neural network model, using the cutter wear-related factors as input and the actual cutter wear amount as output; the LP algorithm is used to iteratively optimize the network weights and biases to output the BP network. In this example, the cutter wear-related factors are: describe, Indicates the sample number, and It is a vector of length 7.

[0069] In this embodiment, steps 1 to 4 are included:

[0070] Step 1: Data preprocessing and normalization.

[0071] Specifically, the acquired tool wear correlation factors are preprocessed and normalized to obtain a data format that meets the input requirements of the BP neural network.

[0072] Step 2: Initialization and forward propagation of the BP neural network structure.

[0073] Specifically, BP neural networks mainly include four types of parameters: the weight matrix from the input layer to the hidden layer. Hidden layer bias vector Weight vector from hidden layer to output layer and output bias .

[0074] The net input of the hidden layer is as follows:

[0075]

[0076] In the formula, Indicates the first The first sample Net input of one hidden layer The number indicating the factor associated with tool wear. express The One element, for Length; express The Middle Line number Column elements, express The One element, This indicates the number of the hidden layer.

[0077] Next, the sigmoid function is used to activate the net input, resulting in the hidden layer activation expression:

[0078]

[0079] In the formula, Indicates the first The first sample One hidden layer is activated.

[0080] Additionally, the net input of the output layer is as follows:

[0081]

[0082] In the formula, Indicates the net input of the output layer. express The Middle One element, This indicates the number of hidden layers.

[0083] Then, the Sigmoid function is used to activate the net input of the output layer to obtain the predicted tool wear amount, expressed as:

[0084]

[0085] In the formula, Indicates the first Predicted tool wear for each sample.

[0086] In this embodiment, it is also necessary to calculate the error between the predicted tool wear and the corresponding actual tool wear. The expression is:

[0087]

[0088] In the formula, For the first The actual tool wear of a sample.

[0089] Step S130: Construct the Jacobian matrix based on the model parameters of the tool wear model, and calculate the LP iteration increment.

[0090] In some embodiments, the method for constructing a layered arrangement and wear prediction model for complex geological shield tunnel cutterhead groups provided by the present invention includes the following steps in step S130:

[0091] Step S131: The cutterhead torque, cutterhead rotation speed, shield thrust, average advance speed, penetration depth, number of tunneling rings, and cutter number are concatenated into a model parameter vector, and each element in the Jacobian matrix is ​​determined.

[0092] Step S132: Based on the elements in the Jacobian matrix, construct the damping diagonal matrix, and calculate the LP iteration increment according to the model parameter vector and the error between the predicted tool wear and the actual tool wear.

[0093] In a specific embodiment, the method for constructing a layered arrangement and wear prediction model for complex geological shield tunnel cutterhead groups provided by the present invention, step 3, Jacobian matrix construction and LP iterative incremental calculation, is expressed as follows:

[0094]

[0095]

[0096]

[0097]

[0098] In the formula, Refers to the parameter vector The Each element is a parameter vector, which is a vector formed by concatenating elements of the four parameter types in the BP network. Represents the Jacobian matrix The Middle Line number The elements of the column.

[0099] Next, the damping diagonal matrix is ​​constructed, represented as:

[0100]

[0101] In the formula, Damping diagonal matrix The Middle Line number The elements of the column; understandably, the values ​​of non-diagonal elements are all 0; This represents the number of samples.

[0102] In this embodiment, parameter increment That is, the LP iteration increment, expressed as:

[0103]

[0104]

[0105] In the formula, The transpose symbol for a matrix. The learning rate, a coefficient controlling the parameter increment, is initially set to 0.001. If the error decreases after subsequent iterations, then... ,otherwise, .

[0106] Step S140: Calculate the MSE value based on the number of iterations and the error between the actual tool wear and the predicted tool wear, and iteratively update the tool wear model based on the MSE value.

[0107] In some embodiments, the method for constructing a layered arrangement and wear prediction model for complex geological shield tunnel cutterhead groups provided by the present invention includes the following steps in step S140:

[0108] Step S141: The model parameter vector is reset based on the MSE value. The reset model parameter vector is the sum of the model parameter vector before reset and the LP iteration increment.

[0109] Step S142: When the tool wear model is updated for the first iteration, the weight matrix from the input layer to the hidden layer, the bias vector of the hidden layer, the weight vector from the hidden layer to the output layer, and the bias vector of the output layer are updated according to the reset model parameter vector, and the MSE value is calculated based on the updated model parameters.

[0110] Step S143: When the tool wear model is not updated in the first iteration, determine whether the MSE value of the tool wear model meets the convergence condition. If so, terminate the iteration.

[0111] In a specific embodiment, the method for constructing a layered arrangement and wear prediction model for complex geological shield cutterhead groups provided by the present invention includes step 4: iterative update and convergence determination.

[0112] Specifically, first, calculate the MSE value, expressed as:

[0113]

[0114] In the formula, This represents the number of iterations, initially set to 1. In essence, it can be .

[0115] Then, reset the parameter vector, that is: If this is the first iteration, then update the parameters based on the reset parameter vector. , , and , and then Return to step S2 and recalculate. If it is not the first iteration, determine whether to accept or reject the update. Then accept the update:

[0116]

[0117]

[0118] Otherwise, refuse to update, and:

[0119]

[0120] After that, Return to step S2 and recalculate. .

[0121] In this embodiment, if If the numerical value is high but the convergence speed is slow, increase the number of hidden layers or adjust the number of neurons in a single hidden layer and retrain. If the convergence condition is met, the iteration terminates. The convergence condition can be set by setting a maximum number of iterations (e.g., 100 times) or a threshold range for parameter increments (e.g., ...). ).

[0122] In some embodiments, the method for constructing a layered arrangement and wear prediction model for complex geological shield cutterhead groups provided by the present invention further includes the following steps:

[0123] Step S210: Obtain the number of cutters in the cutter head, and use each cutter as a node to determine the components of each node and the distance between each node and the root node, and construct a balanced binary tree of the cutter head.

[0124] Step S220: Determine the difference in center distance between different tools on the tool turret, calculate the spatial coupling weight in combination with the spatial attenuation constant, and sort the nodes corresponding to each tool according to the magnitude of the spatial coupling weight.

[0125] Step S230: Obtain the actual tool wear and tool cutting depth of different tools on the cutter head in sequence to determine the wear degree of each tool on the cutter head.

[0126] The degree of wear is directly proportional to the actual wear of the corresponding tool and inversely proportional to the depth of cut. It is a weighted result of the actual wear and the depth of cut, expressed as a percentage of the ideal tool.

[0127] Step S240: Based on the balance binary tree of the cutter head and the wear degree of each tool, sample data and test data of tool wear correlation factors in the cutter head are screened according to the constraints, and the tool wear model is trained and verified by the sample data and test data.

[0128] In this system, the node number corresponds one-to-one with the tool number on the cutter head, and the order of the nodes is consistent with the tool number of each tool. The component is used to characterize the reaction coupling of the tool corresponding to each node from the current cutting load distribution to the local pressure. The expression of the component is as follows:

[0129]

[0130]

[0131] In the formula, and These represent the tool wear when obtaining the actual tool wear amount. With knives Depth of entry and They represent cutting tools With knives The initial depth of cut, The order of representation is The components of the node, This represents the total number of cutting tools.

[0132] In a specific embodiment, the method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological formations provided by this invention establishes a BP neural network-based cutter wear model in sand-rock composite strata. First, factors related to cutter wear are considered to determine the input layer of the neural network. Based on the correlation analysis of the damage amount of the toothed hob cutter teeth, the correlation coefficient between penetration depth, advance speed, and tooth damage amount is 1, indicating a direct correlation. The correlation coefficient between cutterhead rotation speed and tooth damage amount is 0.994, also indicating a direct correlation. The correlation coefficients between tunneling thrust, cutterhead torque, and tooth damage amount are -0.85, showing a similarly strong correlation. Therefore, the five parameters—penetration depth, advance speed, tunneling thrust, cutterhead torque, and cutterhead rotation speed—are used as the input layer parameters of the BP neural network.

[0133] In this example, tool wear increases with the distance traveled; therefore, the correlation between the number of tunneling rings and tool wear is also considered as one of the input layer parameters of the neural network. Since the rotation trajectory of the tools in this project is arranged in ascending order of tool number, installed sequentially from the inside of the cutterhead outwards, the tool number reflects the distance of the tool from the center of the cutterhead. With the same number of tunneling rings, tools closer to the center of the cutterhead have shorter rolling distances compared to tools farther from the center, resulting in less tool wear and even damage. Therefore, the tool number is also considered as one of the input layer parameters of the BP neural network.

[0134] Furthermore, the areas where the cutting tools are located need to be divided into a central cutting tool area, a frontal cutting tool area, and an edge cutting tool area. The central cutting tool area consists of tools numbered 1-12, the frontal cutting tool area consists of tools numbered 13-64, and the edge cutting tool area consists of tools numbered 65-76. The central cutting tool area is set to 1, the frontal cutting tool area to 2, and the edge cutting tool area to 3. These area numbers are used as one of the input layer parameters of the BP neural network. The BP neural network input layer contains eight nodes, and the factors ultimately determined to be related to tool wear are cutterhead torque, cutterhead rotation speed, shield thrust, average advance speed, penetration depth, number of tunneling rings, tool number, and the area number where the tool is located.

[0135] In this embodiment, the cutterhead is regarded as a whole to capture the coupling effect between regions. That is, the regions on the shield cutterhead do not work in isolation, but work as a whole to break and cut the strata simultaneously. The wear of a single cutter depends not only on its own parameters (such as torque, speed, etc.), but also on the cutting sequence, compression and debris discharge method of the surrounding cutters. For example, (1) when the central cutter cuts first, it weakens the local rock mass strength, which reduces the torque required for subsequent front cutters to cut; (2) assuming two cutters have the same torque and speed parameters, but one is located in the central region and the other in the edge region. Under the same number of advance rings, the central cutter cuts a smaller circumferential thickness and has a higher number of friction cycles, while the edge cutter cuts a larger single-ring cutting depth and has a stronger friction force, so the wear rate will be significantly different; (3) if two adjacent cutters cut simultaneously, one in the central region and one in the edge region, the debris generated by the central cutter is more likely to accumulate in the middle of the cutterhead and form a local blockage, which increases the friction load of the cutter in that region; while the debris generated by the edge cutter is more likely to be discharged directly, and the friction load is relatively smaller. Therefore, it is necessary to distinguish them by area numbering to reflect the above-mentioned complex coupling.

[0136] However, simply using the area number of the cutting tool as input is still insufficient to reflect the overall characteristics of the cutterhead. Therefore, in shield cutterhead wear monitoring, if a large amount of sample data is collected only for cutting tools "equipped with wear sensors," while only sparse or sporadic sample data is obtained for cutting tools "equipped with power-on monitoring devices," this results in a severe imbalance in the proportion of sample data at the cutting tool level. This imbalanced distribution of sample data will have a significant negative impact on subsequent wear prediction models, making it difficult to accurately reflect the balanced wear status of the entire cutterhead. Backpropagation (BP) neural networks essentially learn the mapping relationship between features and outputs based on existing sample data distribution. If the number of samples for certain cutting tools equipped with sensors is extremely large, the BP neural network will use the wear patterns of these tools as the primary reference, automatically focusing on them and increasing their weight. However, when the proportion of samples for cutting tools with power-on monitoring devices is unbalanced, the BP neural network will underlearn the relationship between their corresponding features and wear, creating a gap. Ultimately, this leads to the BP neural network overfitting to cutting tool types with more samples, while producing large prediction biases for cutting tool types with fewer samples, or even ignoring their actual wear trends. While the algorithm performs well on tools equipped with sensors, it struggles to accurately estimate the actual wear of the monitored tools when tasks require balanced wear prediction across the entire cutterhead. It also struggles to grasp the overall wear distribution of the cutterhead. Furthermore, the unbalanced sample data means that wear data for different tools only covers a portion of the geological formation or operating conditions. Sensors are often installed only locally, resulting in more local samples. However, the wear patterns in these local locations may only represent a "specific geological section" or "specific working condition." Tools without sensors are distributed in different locations, experiencing vastly different geological hardness, cutting loads, and debris accumulation. If the BP neural network focuses only on learning from "dense sample data in a specific area" during training, the trained BP neural network will have reduced sensitivity to wear trends in other areas under different working conditions, leading to a significant decrease in its generalization ability. In other words, when a joint evaluation of the entire cutterhead is required, the output lacks representative data, reducing the accuracy of tool wear prediction, and the trained model will struggle to adapt to diverse real-world production environments.

[0137] The following section will explain the previous paragraph using specific project examples.

[0138] In the first project, the cutterhead of the tunnel boring machine (TBM) from the starting shaft to the intermediate shaft was distributed in five layers, using 421 cutters of different heights, quantities, and types arranged in these five layers. Figure 2A As shown, the trajectory is as follows Figure 2BAs shown in Table 1, the types and quantities of cutting tools are statistically analyzed. The first layer of cutting tools is a welded lead-in tool with a tool height of 260mm; the second layer of cutting tools is a welded lead-in tool with a tool height of 230mm; the third layer of cutting tools is a replaceable atmospheric pressure lead-in tool with a tool height of 200mm; the fourth layer of cutting tools is a replaceable atmospheric pressure scraper with a tool height of 180mm; and the fifth layer of cutting tools is a bolt-type scraper with a tool height of 150mm.

[0139] Table 1

[0140] However, in the first project, a total of 202 replaceable atmospheric pressure cutting tools were replaced, including 149 replaceable atmospheric pressure leading cutters and 53 replaceable atmospheric pressure scrapers. The first replacement of the replaceable atmospheric pressure leading cutters occurred at ring 853, and the first replacement of the replaceable atmospheric pressure scrapers occurred at ring 1653. Cutters on various tracks were subsequently replaced during construction. During the tool replacement process, abnormal wear was discovered on the tools, including severe uneven wear and chipped edges.

[0141] In the second project, the cutterhead of the tunnel boring machine (TBM) from the starting shaft to the intermediate shaft is distributed in four layers, using 342 cutters of different heights, quantities, and types arranged in five layers, such as... Figure 3A As shown, the trajectory is as follows Figure 3B As shown in Table 2, the tool arrangement is as follows: the first layer of tools consists of welded lead-in tools with a tool height of 280mm; the second layer of tools is divided into two categories: the first category is welded lead-in tools with a tool height of 230mm; the second category is atmospheric pressure replaceable lead-in tools with a tool height of 230mm; the third layer of tools consists of atmospheric pressure replaceable scrapers with a tool height of 180mm; and the fourth layer of tools consists of bolt-type scrapers with a tool height of 150mm.

[0142] Table 2

[0143] It's easy to see that in actual projects, such as the five-layer arrangement of a tunnel boring machine cutterhead, almost all cutters in each layer use the same type. This "same type in the same layer, different between layers" leads to extreme bias in the monitoring data: once wear sensors or power-off devices are used to collect data on a certain type of cutter, a large number of wear samples will be obtained for that type, while almost no usable data will be available for cutters in other layers or of other types. First, the uneven distribution of samples caused by this isomorphic distribution means that any statistical or machine learning-based method will preferentially "focus" on the cutter type with the largest sample size during training, resulting in severe class imbalance. As a result, the model will "overfit" to the wear patterns of that type of cutter, and when encountering cutter types with sparse samples, it will be unable to give accurate predictions due to a lack of sufficient training samples, or even fail completely.

[0144] Before optimization, the cutter arrangement required 42 cutter replacements per kilometer when tunneling through sand and silty clay layers. After optimization, the cutter arrangement enabled the tunnel boring machine to continuously tunnel 2.6km through sand and silty clay layers with a maximum cutter wear of only 6mm. No abnormal wear occurred, and no downtime was required for cutter replacement, saving a significant amount of cutter replacement and material costs.

[0145] The optimized arrangement of the blade group is mainly reflected in reducing the number of blade layers and increasing the blade height difference between the advance blade and the scraper blade, so that the functions of blades at different layer heights can be more rationally performed. The scraper blade can play a scraping role after the advance blade has fully plowed and loosened the soil, and the wear rate of the scraper blade is reduced.

[0146] Therefore, to address the issue of unbalanced sample data proportions, in some embodiments, the method for constructing a layered arrangement and wear prediction model for complex geological shield tunnel cutter groups provided by this invention further includes steps A1-A2:

[0147] Step A1, construct a balanced binary tree for the cutter head: After determining the components and order of each node, the following constraints must be satisfied. :

[0148]

[0149] In the formula, Represents a node The amount, Represents a node Distance from the root node This represents the number of cutting tools in the cutter head.

[0150] In this embodiment, each node is used to characterize the tool on the cutter head. The order of the nodes determines the node number, and the node number corresponds one-to-one with the tool number. The order of the nodes actually reflects the hierarchical position of the tool. The component is used to characterize the coupling of the tool corresponding to the node from the perspective of "current cutting load distribution to local high pressure". That is, the higher the proportion of the total cutting force of the tools around the tool, the more load is concentrated in the adjacent area, and the more easily it is subjected to rock cutting squeeze or cutting impact, thus aggravating wear. It can be understood that the magnitude of the component is actually proportional to the tool height.

[0151] The order of the nodes is represented as follows:

[0152]

[0153]

[0154] In the formula, Indicates the first The center distance of the tool and the first The difference in the center distance of the cutting tools, the first The center distance of the tool refers to the Euclidean distance between the tool and a certain center point on the axis of the tool head. The center point can usually be taken as the center of gravity of the tool head. This is the spatial attenuation constant, used to control the attenuation rate of coupling at different distances; Represents spatial coupling weights; function Used for all nodes Sort them in ascending order to determine the corresponding values ​​in sequence. The number, that is, from .therefore, and The sizes are positively correlated, and the node with the smaller number value always precedes the node with the larger number value.

[0155] Components of a node As shown in the following formula:

[0156]

[0157]

[0158] In the formula, and These represent the tool wear when obtaining the actual tool wear amount. With knives Depth of entry and They represent cutting tools With knives The initial depth of cut, The order of representation is The components of the node.

[0159] In this embodiment, when the order With weight Once all constraints have been determined, in order to ensure that... The value is the smallest, therefore It is also the only certainty. The essence of the constraint is to move nodes with high wear closer to the root node, and vice versa, to move them closer to leaf nodes. The significance of a balanced binary tree lies in maximizing the use of sample data from different cutterheads for prediction and learning, and in this example, only all historical sample data from one cutterhead are typically selected for prediction.

[0160] Step A2: Filter sample data and test data based on a balanced binary tree.

[0161] Tool wear correlation factors increase constraints Meanwhile, the tool numbers are in sequence. Alternatively, the following constraints should be met during the selection of sample and test data (e.g., by deleting some sample data to meet the following constraints, in order to solve the problem of unbalanced sample data ratio):

[0162]

[0163]

[0164] In the formula, This indicates the sample data number, since the sample data number actually corresponds to each specific tool number. , and These represent the distances between node i and node b and the root node, respectively. Indicates the first The tool number for each sample data; and These represent the number of sample data under the cutter head and the number of cutters in the cutter head, respectively. All of these are limiting parameters, among which, ,generally It is acceptable ; ,generally It is acceptable Understandable. The closer the value is to 1, the fewer the number of samples, but the higher the training accuracy. This means that as the amount of sample data increases, The value can keep rising.

[0165] The following describes the device for constructing a layered arrangement and wear prediction model for a complex geological shield tunnel cutter group provided by the present invention. The device for constructing a layered arrangement and wear prediction model for a complex geological shield tunnel cutter group described below can be referred to in correspondence with the method for constructing a layered arrangement and wear prediction model for a complex geological shield tunnel cutter group described above.

[0166] In one embodiment, a device for constructing a layered arrangement and wear prediction model for a complex geological shield cutterhead group includes a data preprocessing module, a model training module, an iterative incremental calculation module, and a model iterative update module.

[0167] The data preprocessing module is used to obtain the tool wear correlation factors and historical tool wear amounts, and to preprocess and normalize the tool wear correlation factors and historical tool wear amounts.

[0168] The model training module is used to construct a tool wear model based on a BP neural network, using tool wear-related factors as input and historical tool wear as output, in order to predict tool wear.

[0169] The iterative increment calculation module is used to construct the Jacobian matrix based on the model parameters of the tool wear model and to calculate the LP iterative increment.

[0170] The model iteration update module is used to calculate the MSE value based on the number of iterations and the error between the actual tool wear and the predicted tool wear, and to iteratively update the tool wear model based on the MSE value.

[0171] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0172] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0173] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0174] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0175] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0176] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0177] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0178] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions, characterized in that, The method includes: Obtain tool wear correlation factors and historical tool wear amounts, and preprocess and normalize the tool wear correlation factors and historical tool wear amounts; Using the aforementioned tool wear correlation factors as input and historical tool wear amount as output, a tool wear model based on BP neural network is constructed to predict tool wear amount; Based on the model parameters of the tool wear model, a Jacobian matrix is ​​constructed, and the LP iteration increment is calculated; The MSE value is calculated based on the number of iterations and the error between the actual tool wear and the predicted tool wear, and the tool wear model is iteratively updated based on the MSE value.

2. The method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions according to claim 1, characterized in that, The method further includes: Obtain the number of cutters in the cutter head, and use each cutter as a node to determine the components of each node and the distance between each node and the root node, and construct a balanced binary tree of the cutter head. Determine the difference in center distance between different tools on the tool turret, calculate the spatial coupling weight in combination with the spatial attenuation constant, and sort the nodes corresponding to each tool according to the magnitude of the spatial coupling weight; The actual tool wear and cutting depth of different tools on the cutter head are obtained in sequence to determine the degree of wear of each tool on the cutter head. Based on the balanced binary tree of the cutter head and the wear degree of each tool, sample data and test data of tool wear correlation factors in the cutter head are screened according to the constraints, and the tool wear model is trained and verified through the sample data and test data. The node number corresponds one-to-one with the number of each tool on the cutter head, and the order of the nodes is consistent with the tool number of each tool. The component is used to characterize the reaction coupling of the tool corresponding to each node from the current cutting load distribution to the local pressure.

3. The method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions according to claim 1, characterized in that, The BP neural network includes an input layer, a hidden layer, and an output layer. The model parameters include the weight matrix from the input layer to the hidden layer, the bias vector of the hidden layer, the weight vector from the hidden layer to the output layer, and the bias vector of the output layer. The tool wear correlation factors include multiple elements, namely cutterhead torque, cutterhead rotation speed, shield thrust, average advance speed, penetration depth, number of tunneling rings, and tool number, which are used to generate multiple input layer nodes of the BP neural network, and the historical tool wear amount is used to generate output layer nodes.

4. The method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions according to claim 1, characterized in that, The step of constructing a tool wear model based on a backpropagation neural network, using the tool wear correlation factors as input and historical tool wear data as output, to predict tool wear data includes: Obtain the number of each tool wear-related factor, and determine the elements and element lengths corresponding to the number, the elements in the weight matrix, the elements in the bias vector, and the hidden layer number; The net input of the hidden layer is calculated based on the elements and element lengths corresponding to the number, the elements in the weight matrix, the elements in the bias vector, and the hidden layer number. The hidden layer activation is calculated and output by activating the net input of the hidden layer using the Sigmoid algorithm.

5. The method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions according to claim 4, characterized in that, The method of constructing a tool wear model based on a BP neural network to predict tool wear, using the tool wear correlation factors as input and historical tool wear amount as output, further includes: Obtain the elements in the weight vector and the number of hidden layers, and calculate the net input of the output layer based on the elements in the weight vector and the number of hidden layers; The output layer net input is activated using the Sigmoid algorithm, and the predicted tool wear is calculated and output. Obtain the actual tool wear amount corresponding to the predicted tool wear amount, and calculate the error between the predicted tool wear amount and the actual tool wear amount.

6. The method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions according to claim 3, characterized in that, The construction of the Jacobian matrix based on the model parameters of the tool wear model and the calculation of the LP iteration increment include: The cutterhead torque, cutterhead rotation speed, shield thrust, average advance speed, penetration depth, number of tunneling rings, and cutter number are concatenated into a model parameter vector, and each element in the Jacobian matrix is ​​determined. Based on the elements in the Jacobian matrix, a damping diagonal matrix is ​​constructed, and the LP iteration increment is calculated according to the model parameter vector and the error between the predicted tool wear and the actual tool wear.

7. The method for constructing a layered arrangement and wear prediction model for shield tunnel cutterheads in complex geological conditions according to claim 6, characterized in that, The step of calculating the MSE value based on the number of iterations and the error between the actual tool wear and the predicted tool wear, and iteratively updating the tool wear model based on the MSE value, includes: The model parameter vector is reset based on the MSE value, and the reset model parameter vector is the sum of the model parameter vector before reset and the LP iteration increment; When the tool wear model is updated for the first iteration, the weight matrix from the input layer to the hidden layer, the bias vector of the hidden layer, the weight vector from the hidden layer to the output layer, and the bias vector of the output layer are updated according to the reset model parameter vector, and the MSE value is calculated based on the updated model parameters. When the tool wear model is not updated in the first iteration, it is determined whether the MSE value of the tool wear model meets the convergence condition. If so, the iteration is terminated.

8. A device for constructing a layered arrangement and wear prediction model for complex geological shield tunnel cutterheads, characterized in that, The apparatus for implementing the method for constructing a layered arrangement and wear prediction model of complex geological shield cutterhead groups according to any one of claims 1 to 7 includes: The data preprocessing module is used to obtain tool wear correlation factors and historical tool wear amounts, and to preprocess and normalize the tool wear correlation factors and historical tool wear amounts; The model training module is used to construct a tool wear model based on a BP neural network, using the tool wear correlation factors as input and historical tool wear amount as output, in order to predict tool wear amount. The iterative increment calculation module is used to construct a Jacobian matrix based on the model parameters of the tool wear model and calculate the LP iterative increment. The model iteration update module is used to calculate the MSE value based on the number of iterations and the error between the actual tool wear and the predicted tool wear, and to iteratively update the tool wear model based on the MSE value.

9. A terminal, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

Images