Shield tunneling machine or TBM hob associated characteristic analysis wear method and system

Abstract

The invention discloses a shield tunneling machine or TBM hob associated characteristic abrasion analysis method and system, and belongs to the technical field of associated characteristic analys.The method comprises the steps that parameter data related to cutter abrasion are collected in real time, and a parameter time sequence is constructed; calculating a change curve of the statistical characteristics, the derivative characteristics and the rock-soil layer, and updating a parameter sequential sequence; analyzing a dynamic change condition of a parameter time sequence, calculating a feature importance score by using a random forest algorithm, avoiding interference by unimportant information in an analysis process, and constructing an associated feature model; according to the method, three-dimensional point cloud data of a rock-soil layer is constructed, fused point cloud data containing a tool wear label is generated, rich dimension data is provided, a point cloud model used for judging the state of the tool is constructed, the risk of misjudgment and missed judgment is reduced, and the state of the tool is judged; once it is judged that the cutter is damaged, a cutter damage signal is generated, and a powerful guarantee is provided for safe and efficient operation of the shield tunneling machine or the TBM.

Description

Technical Field

[0001] The present invention belongs to the technical field of associated feature analysis, and relates to a method and system for analyzing the wear of associated features of a shield machine or a TBM hob. Background Art

[0002] A shield machine or a TBM is a core device in underground tunnel construction, integrating functions such as excavation, support, soil removal, and guidance, which greatly improves the efficiency and safety of tunnel construction; among them, the hob, as a key component of the shield machine or the TBM, undertakes the heavy responsibility of breaking rocks and soils, and its performance directly affects the tunneling speed of the shield machine or the TBM and the quality of the tunnel.

[0003] The existing Chinese patent with the authorized announcement number CN110119551B discloses a method for analyzing the associated features of the wear and degradation of shield machine cutters based on machine learning, which is used to solve the technical problem of narrow application range in the existing technology, and at the same time improve the accuracy of monitoring the health status and predicting the life of the cutters. The implementation steps include: obtaining the original feature dataset data3 under slightly weathered strata; obtaining the data resource set Data; constructing a LightGBM feature ranking model; obtaining the important feature subset Fea; discretizing the important feature subset Fea; performing association rule mining on the discretized important feature subset data_lsh; obtaining the associated features of the wear and degradation of the shield machine cutters.

[0004] Although the existing technology fully considers the influence of all collected data on the wear of shield machine cutters, has a wide application range and high accuracy, it does not consider the real-time and dynamic nature of the feature data. Especially during the actual operation of the shield machine, due to the continuous changes in working conditions and external environments, such as changes in strata conditions, adjustments in construction processes, and changes in the state of the equipment itself, such as the influence of cutters at different wear stages on other parameters, these factors may change over time and interfere with the judgment of cutter wear. Therefore, the present application provides a method and system for analyzing the wear of associated features of a shield machine or a TBM hob to address the problem of the lack of real-time and dynamic nature of feature data and improve the accuracy and real-time nature of wear analysis. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a method and system for analyzing the wear of associated features of a shield machine or a TBM hob. By constructing a parameter time series and updating the sequence content in combination with the rock and soil layer change curve, the real-time nature of the data is improved; and the parameter time series is dynamically analyzed to capture the fluctuations between the sequences, enhancing the adaptability to complex environments.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] Method for analyzing wear of associated features of shield machine or TBM cutter, comprising:

[0008] Step S1: Collect parameter data related to tool wear in real time, perform preprocessing, and construct a parameter time series;

[0009] Step S2: Calculate the change curves of statistical features, derivative features and rock and soil layers, and update the parameter time series;

[0010] Step S3: Analyze the dynamic changes of the parameter time series, calculate the feature importance scores using the random forest algorithm, and construct an associated feature model;

[0011] Step S4: Construct three-dimensional point cloud data of the rock and soil layer, generate fused point cloud data with tool wear labels, and construct a point cloud model for discriminating tool states;

[0012] Step S5: Calculate the prediction probability of the tool state and judge the tool state; once it is determined that the tool is damaged, a tool damage signal is generated.

[0013] S2.1: According to the cutter head torque collected in real time and the propulsion speed , calculate the ratio of the cutter head torque to the propulsion speed ;

[0014] S2.2: According to the jacking force collected in real time , calculate the product of the jacking force and the cutter head torque ;

[0015] S2.3: Continuously record the actual values of geotechnical parameters at different time points, and draw a curve of geotechnical parameters changing with time through data fitting;

[0016] S2.4: Save the calculated statistical features, derivative features and the change curve of the rock and soil layer to the parameter time series wherein the parameter time series includes a historical time series and a real-time time series .

[0017] Specifically, the specific steps of data fitting in S2.3 include:

[0018] During tunneling, according to the non-linear relationship between a certain geotechnical parameter and the tunneling distance, generate a fitting curve;

[0019] Construct a measurement sample group , calculate the sum of squares of errors ;

[0020] Respectively for , and Find the partial derivatives and set them to zero to obtain a system of partial derivative equations;

[0021] Solve the system of partial derivative equations to calculate the fitting coefficient group for the final value.

[0022] Specifically, the specific steps for solving the system of partial derivative equations include:

[0023] Let the functions in the system of partial derivative equations be partial derivative functions , , respectively, and use the approximation of the area of a curvilinear trapezoid to construct variable upper limit integral functions for each partial derivative function;

[0024] Calculate the gradients of each variable upper limit integral function with respect to each fitting coefficient respectively;

[0025] Let the weight coefficient be , and reconstruct and assign values to the variable upper limit integral function to generate a newly constructed assignment function;

[0026] Combine multiple assignment functions into a function vector, and calculate the Jacobian matrix of the function vector;

[0027] Set , , to their initial values as , and use the Newton iteration method to update , , according to the current fitting coefficient values, function vector, and Jacobian matrix;

[0028] Calculate the iteration error of the fitting coefficient group, set the iteration termination threshold to , and determine whether the iteration operation ends; when , it is determined that the iteration has not converged, and the iteration operation continues; when , it is determined that the iteration has converged, and output the latest iterated fitting coefficient values.

[0029] Specifically, the specific steps of step S3 include:

[0030] S3.1: For the parameter time series , use STL decomposition to decompose into a trend term, a seasonal term, and a residual term;

[0031] S3.2: Use the trend term in the historical time series to train the trend model and output the trend term prediction value;

[0032] S3.3: Train the residual model using the residual terms in the historical time series to fit the dynamic characteristics of the residual terms and generate predicted values of the residual terms;

[0033] S3.4: Combine the predicted trend values, the seasonal terms in the historical time series, and the predicted residual values to obtain historical predicted values.

[0034] Specifically, the specific steps of step S3 further include:

[0035] S3.5: Calculate the linear correlation degree between any two variables in the historical predicted values and construct a correlation matrix;

[0036] S3.6: Set that the actual data feature vector contains feature parameters, generate a -dimensional random vector

[0037] S3.7: Set the actual data feature vector as , and the weight vector as . Fuse the random vector with the actual data feature vector to generate a fused feature vector .

[0038] Specifically, the specific steps of step S3 further include:

[0039] S3.8: Use the random forest algorithm to construct multiple decision trees, calculate the feature importance scores, and generate the importance sequence of the fused feature vector ;

[0040] S3.9: Use the tool wear amount as the target variable, and establish an association function between various influencing factors and the tool wear amount through the feature importance scores and the fused feature vector to generate an association feature model.

[0041] Specifically, the specific steps of step S4 include:

[0042] S4.1: Use a ground penetrating radar and a laser scanning device to scan the rock and soil layers in front of and around the shield machine or TBM, and perform data verification to generate three-dimensional point cloud data of the rock and soil layers;

[0043] S4.2: Fuse the three-dimensional point cloud data with the features in the association feature model to generate fused point cloud data with tool wear labels;

[0044] S4.3: Preprocess the fused point cloud data, including grid processing and normalization processing;

[0045] S4.4: Construct a point cloud model, including an input layer, a multi-layer perceptron layer, a max pooling layer, and a fully connected layer, where the output is the predicted probability of the tool state;

[0046] S4.5: Randomly initialize the parameters in the point cloud model, and train the point cloud model based on the processed fused point cloud data.

[0047] Specifically, the specific steps of data verification in S4.1 include:

[0048] Taking the data point to be verified as the center, set an initial search area, obtain the number of surrounding points within the search area, and calculate the initial point distribution density ;

[0049] Set the search density range to , and determine whether the point distribution density is within the search density range;

[0050] If , the point distribution density is within the search density range;

[0051] If , expand the search range until the point distribution density is within the search density range;

[0052] If , reduce the search range until the point distribution density is within the search density range;

[0053] Obtain the position coordinates of multiple sampling points, generate a sample data set , and calculate the sample data average value ;

[0054] Calculate the sample data and the difference between the average value , and calculate the average value of the differences ;

[0055] Set the discrete threshold of the search area to , and determine whether the data point has an error; if , the data point has no error, and define the data point as a calibrated point; , the data point has an error, and define the data point as a preliminary error point;

[0056] In the search area of the preliminary error point, perform data verification on the surrounding points, obtain the number of surrounding points with errors and the average value , and calculate the average surrounding error ; where is the average value of the th error;

[0057] Set the geological change difference threshold to , and determine whether the preliminary error point is a boundary point; if , the preliminary error point is a boundary point; if , generate alternative data points without errors;

[0058] Generate the three-dimensional point cloud data of the rock and soil layer, including calibrated points and boundary points.

[0059] Specifically, the specific steps of step S5 include:

[0060] S5.1: Obtain the real-time time series , and calculate the predicted probability of the tool state ;

[0061] S5.2: Set the tool damage threshold to , and judge the damage situation of the current tool; if , the tool is in an undamaged state; if , the tool is damaged, and a tool damage signal is generated;

[0062] S5.3: After the tool change is completed, collect various parameter data of the new tool, and update the parameter time series and timestamp information at the same time;

[0063] S5.4: Use the updated parameter time series to optimize and adjust the correlation feature model and the point cloud model, and perform damage analysis on the new tool.

[0064] The wear system for analyzing the correlation features of the cutter head of a shield machine or a TBM includes: a data acquisition module, a feature analysis module, a prediction module, and an update module;

[0065] The data acquisition module uses various sensors to collect parameter data related to tool wear in real time;

[0066] The feature analysis module calculates statistical parameters and derivative parameters according to the initial acquisition parameters, and constructs a change curve of the rock and soil layer in combination with the tunneling situation of the shield machine or the TBM, and generates a parameter time series;

[0067] The prediction module includes a correlation unit and a judgment unit;

[0068] The correlation unit is used to dynamically analyze the parameter time series, calculate the importance scores of each feature, and construct a correlation feature model;

[0069] The judgment unit is used to construct the three-dimensional point cloud data of the rock and soil layer, combine it with the associated feature model to generate the fused point cloud data with tool wear labels, and construct a point cloud model for discriminating the tool state; output the predicted probability of the tool state and judge the tool state.

[0070] The update module is used to collect the parameter data of the new tool after the tool change is completed, update the parameter time series sequence, and optimize and adjust the associated feature model and the point cloud model with the new data.

[0071] Advantages of the present invention:

[0072] By calculating statistical and derivative features, information closely related to tool wear is mined from different dimensions, a rock and soil layer change curve is constructed, the grasp of the external environment dynamics is further refined, the parameter time series sequence is continuously updated to ensure the coherence and timeliness of the data, and the analysis deviation caused by data missing is avoided; at the same time, analyzing the dynamic changes of the parameter time series sequence helps to capture the subtle fluctuations and trend changes during the operation of the shield machine or TBM; fusing the three-dimensional point cloud data of the rock and soil layer with the constructed associated feature model to generate the fused point cloud data with tool wear labels improves the intuitiveness and accuracy of tool state discrimination and enhances the ability to cope with complex working conditions; when the model determines that the tool is damaged, not only can the tool change robot be started in time, but also the parameter data after the tool change will be used to optimize the model, enabling the model to continuously adapt to the ever-changing operation state and environment of the shield machine or TBM and always maintain a high level of discrimination ability, providing a strong guarantee for the safe and efficient operation of the shield machine or TBM. Description of the Drawings

[0073] Figure 1 Schematic diagram of the wear analysis method for the associated features of the hob of the shield machine or TBM;

[0074] Figure 2 Flow chart of updating the parameter time series sequence in the wear analysis method for the associated features of the hob of the shield machine or TBM;

[0075] Figure 3 Flow chart of constructing the associated feature model in the wear analysis method for the associated features of the hob of the shield machine or TBM;

[0076] Figure 4 Flow chart of judging tool damage in the wear analysis method for the associated features of the hob of the shield machine or TBM;

[0077] Figure 5 Structure diagram of the wear analysis system for the associated features of the hob of the shield machine or TBM. Detailed Implementation Modes

[0078] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. Without conflict, the technical features in the embodiments of the present invention and the embodiments can be combined with each other.

[0079] Embodiment 1

[0080] Reference Figures 1 to 4 As shown, this embodiment introduces a method for analyzing the wear of the associated features of the shield machine or TBM cutter, including the following steps:

[0081] Step S1: During the operation of the shield machine or TBM, use various sensors to collect parameter data related to tool wear in real time, including equipment operation parameters such as propulsion speed, cutterhead torque, cutterhead rotation speed, penetration, and jacking force, external environment parameters such as formation soil type, strength parameters, hardness, water content, mineral type and content, and construction process parameters such as tunneling mode and grouting volume. And preprocess the collected parameter data, including: filling the missing values in the data with the mean value, detecting and removing outliers using the Z-Score method, and performing data denoising using the Gaussian filtering method to construct a parameter time series to ensure the real-time nature of data collection and timely capture the dynamic changes of the operation state of the shield machine or TBM and the external environment; among them, the parameter sequence includes the parameters related to tool wear and the corresponding actual parameter values, and the actual parameter values are saved in chronological order, with the operation sampling frequency being , the environment sampling frequency is , the construction sampling frequency is ;

[0082] Step S2: According to the working principle of the shield machine or TBM and the tool wear theory, calculate the statistical features and derivative features related to tool wear, and at the same time construct the change curve of the rock and soil layer and update the parameter time series;

[0083] Step S3: Analyze the dynamic changes of the parameter time series, use the random forest algorithm to calculate the feature importance scores, clarify the influence rules of various features on tool wear, construct an associated feature model, and comprehensively consider the complex relationship between various features and tool wear to effectively describe the tool wear state;

[0084] Step S4: Construct the three-dimensional point cloud data of the rock and soil layer, combine it with the associated feature model to generate the fused point cloud data with tool wear labels, and construct a point cloud model for discriminating the tool state;

[0085] Step S5: Based on the real-time time series and the point cloud model, output the predicted probability of the tool state, and judge the tool state; once it is determined that the tool is damaged, generate a tool damage signal, and the control center starts the tool-changing robot; after the tool change is completed, collect the parameter data of the new tool, update the parameter time series, and optimize and adjust the associated feature model and the point cloud model with the new data to adapt to the new operating state and environmental changes of the shield machine or TBM, and improve the accuracy and reliability of model discrimination.

[0086] Specifically, the specific steps of step S2 include:

[0087] S2.1: The tool cutting process is essentially a mechanical action process. According to the relationship between force and motion, the torque reflects the resistance that needs to be overcome when the tool rotates, and the propulsion speed reflects the forward speed of the tool in the axial direction. In order to more directly understand the cutting efficiency and wear state of the tool, the ratio of the cutterhead torque to the propulsion speed is introduced as an evaluation index. According to the real-time collected cutterhead torque and the propulsion speed , calculate the ratio of the cutterhead torque to the propulsion speed . During the cutting process, the ratio represents the rotational resistance borne by the tool under the unit propulsion speed. From the perspective of mechanical principles, the larger the ratio , the greater the resistance that the tool needs to overcome during cutting, which will inevitably lead to an increase in the friction and acting force between the tool and the cutting medium, thereby accelerating tool wear. The expression is as follows:

[0088] ;

[0089] In the formula, by continuously monitoring and analyzing the ratio , the change in the cutting state of the tool can be detected in a timely manner; when the ratio is large, it indicates that at the same propulsion speed, the torque that the cutterhead needs to overcome is large, the resistance encountered by the tool during cutting is large, the cutting process is difficult, and the cutting efficiency is low. At this time, the friction and acting force between the tool and the medium increase, and the tool wear accelerates; on the contrary, the wear is relatively slow; for example, in hard rock formations, the ratio is large.

[0090] S2.2: In actual construction, when the shield machine or TBM is in an unstable cutting condition, such as encountering local geological mutations, the jacking force and the cutterhead torque will fluctuate greatly. The jacking force makes the tool closely contact the cutting surface and provides the axial force for cutting, while the cutterhead torque drives the tool to perform rotary cutting. According to the real-time collected jacking force and the cutterhead torque , calculate the product of the jacking force and the cutterhead torque , To a certain extent, it reflects the load borne by the cutting tool. Based on the principle of force composition and effect, the combined action of two forces affects the wear of the cutting tool. By comparing the values at different time periods, it is possible to determine whether the cutting condition of the shield machine or TBM is stable and whether the wear trend of the cutting tool is normal. The expression is as follows:

[0091] ;

[0092] In the formula, the jacking force is larger, indicating that the pressure on the cutting tool in the axial direction is greater. The cutter head torque is larger, indicating that the resistance torque on the cutting tool during the rotary cutting process is greater. When is relatively large, it indicates that the shield machine or TBM consumes more energy during the cutting process, the cutting tool bears a greater load during operation, and the wear will increase accordingly. Conversely, the wear is relatively light.

[0093] S2.3: As the shield machine or TBM advances, continuously record the actual values of the geotechnical parameters at different time points. Among them, the geotechnical parameters include but are not limited to: particle size distribution, void ratio, rock strength, and rock hardness. In order to better track the dynamic changes of the geotechnical parameters during the advancement of the shield machine or TBM, with the advancement distance as the abscissa and the geotechnical parameters as the ordinate, draw the change curve of the rock and soil layer over time, which clearly shows the change trend of the geotechnical parameters during the advancement process. Record the relevant parameter values every certain advancement distance (such as 5 meters), and draw the curve by means of data fitting.

[0094] S2.4: Save the calculated statistical features, derived features, and the change curve of the rock and soil layer to the parameter time series Among them, the parameter time series includes the historical time series and the real-time time series , is the value at the moment in the parameter time series, including the parameter data, statistical features, derived features, and the change curve of the rock and soil layer at the moment. The derived features include the ratio of the cutter head torque to the propulsion speed and the product of the jacking force and the cutter head torque. The statistical features include the mean, variance, maximum value, and minimum value of each parameter in the original parameter time series. The mean is used to measure the smoothness of the operation of the shield machine or TBM, reflecting the average level of each parameter of the shield machine or TBM during a specific time period. The variance is used to measure the degree of dispersion of the data, and the extreme values are used to measure the extreme operating state of the shield machine or TBM during a specific time period. The expressions are as follows:

[0095] ;

[0096] ;

[0097] ;

[0098] ;

[0099] In the formula, taking the parameter time sequence as an example, the parameters in it , , , are respectively the mean value, variance, maximum value and minimum value of the parameter at the moment , and , is the value of the parameter at the moment , ;

[0100] Specifically, the specific steps of data fitting in S2.3 include:

[0101] Taking the tunneling distance and a certain geotechnical parameter as an example, under the actual geological conditions, the situation of the stratum is often very complex. Therefore, the non-linear relationship between the geotechnical parameter and the tunneling distance is considered during the tunneling process to represent the changing trend of the geotechnical parameter during the tunneling process, and a fitting curve is generated. The expression of the fitting curve is as follows:

[0102] ;

[0103] In the formula, is the fitting coefficient group, is the initial value of the geotechnical parameter when the tunneling distance is , is the first-order change coefficient of the geotechnical parameter with the tunneling distance, is the second-order change coefficient of the geotechnical parameter with the tunneling distance, is the value of the geotechnical parameter at the moment , is

[0104] the value of the tunneling distance at the moment Based on the real-time obtained geotechnical parameters and tunneling distance, a measurement sample group is constructed, where and are respectively the geotechnical parameter value and the tunneling distance of the th sample group; due to the difference between the measured value and the theoretical value calculated through the expression, by calculating the sum of the squares of the errors between the measured value and the theoretical value , comprehensively measure the deviation degree between the theoretical values and the actual values of all measurement points, and comprehensively reflect the fitting effect based on the given expression. Among them, the smaller the sum of squared errors, the closer the fitting curve is to the actual measurement data, and the higher the fitting degree of the actual data. The expression is as follows:

[0105] ;

[0106] In the formula, is the number of measurement samples, is the actual geotechnical parameter value of the th measurement point at time is the corresponding tunneling distance, ;

[0107] In order to minimize the sum of squared errors, partial derivatives are taken with respect to , and respectively, to explore how the change of the fitting coefficients affects the sum of squared errors, and make the partial derivatives zero. At this time, the points where the partial derivatives are zero correspond to the extreme points on the fitting curve and are also the minimum points of the sum of squared errors. The following system of partial derivative equations is obtained:

[0108] ;

[0109] When using the Newton iteration method to solve the optimal solution, it has a second-order convergence rate. Compared with some first-order convergence methods, such as the gradient descent method, it can converge to the solution of the system of equations faster, improving the solution efficiency. At the same time, since the system of partial derivative equations is a non-linear system of equations, fast convergence can save a large amount of computing time and resources. Use the Newton iteration method to solve the system of partial derivative equations and calculate the final values of the fitting coefficient group .

[0110] Specifically, the specific steps for using the Newton iteration method to solve the system of partial derivative equations include:

[0111] Set the partial derivative functions as , , . Since the partial derivative functions are complex non-linear functions and are difficult to handle directly, use the approximation of the area of a curvilinear trapezoid to approximate the area under the curve as the sum of the areas of multiple small trapezoids, transform the complex non-linear functions into forms that are easy to handle, and by constructing a variable upper limit integral function, consider the cumulative effect of the partial derivative functions within a certain interval as a whole. For each partial derivative function , use the method of approximating the area of a curvilinear trapezoid to construct four variable upper limit integral functions , , , , where ;

[0112] Different variable-upper-limit integral functions have different degrees of influence on the final result. By introducing weight coefficients, each variable-upper-limit integral function is weighted and combined to highlight the important parts and balance the contributions of different functions. Calculate the gradient of each variable-upper-limit integral function with respect to 、 、 ; Among them, ; ;

[0113] Let the weight coefficient be , reconstruct and assign values to the variable-upper-limit integral function to construct a new function. The expression is as follows:

[0114] ;

[0115] Among them, is the newly constructed assignment function, and the weight coefficient is determined by those skilled in the art to ensure the convergence and speed of iteration;

[0116] By combining multiple assignment functions into a function vector, it helps to uniformly process and analyze these functions. Define the function vector as , and calculate the Jacobian matrix of ;

[0117] Set the initial values of , , , denoted as , and use the Newton iteration method to update , , according to the current fitting coefficient values, function vector, and Jacobian matrix to minimize the sum of squared errors and gradually converge. The expression is as follows:

[0118] ;

[0119] In the formula, is the number of iterations, is the fitting coefficient value at the -th iteration, is the inverse matrix of the Jacobian matrix at ;

[0120] Based on the fitting coefficient values of the latest iteration, calculate the iteration error from the previous iteration. Set the iteration termination threshold as , and determine whether the iterative operation ends; the expression is as follows:

[0121] ;

[0122] When happens, it is determined that the iteration has not converged, and the iterative operation continues; when happens, it is determined that the iteration has converged, and the fitting coefficient value of the latest iteration is output , and at this time is the final value of the fitting coefficient group .

[0123] Specifically, the specific steps of step S3 include:

[0124] S3.1: For the parameter time series , use STL decomposition to decompose into a trend term, a seasonal term, and a residual term. The trend term reflects the overall trend of the data over time, the seasonal term reflects the periodic pattern of the data over time, and the residual term reflects the random fluctuations and abnormal changes in the data. By decomposition, the complex time series data is simplified into three parts that are easier to analyze and predict; the expression is as follows:

[0125] ;

[0126] Among them, is the value of the trend term at time, is the value of the seasonal term at time, is the value of the residual term at time;

[0127] S3.2: Define the basic Informer model as the trend model, and use the trend term in the historical time series to train the trend model. Process the historical time series through the self-attention mechanism, continuously adjust the parameters of the model, mine the long-term dependencies in the trend term, and output the trend term prediction value . The expression of the self-attention mechanism is as follows:

[0128] ;

[0129] ;

[0130] ;

[0131] Among them, is the predicted value corresponding to the trend term at It is a multi - head attention function, which is used to parallelly use multiple attention heads, enabling the trend model to capture sequence features in different sub - spaces. It is a concatenation function, which is used to concatenate the output results of multiple attention heads along the dimension. It is the th attention head. , is the number of attention heads, forming a concatenated feature matrix. It is an attention calculation function. It is an activation function, which is used to convert the input numerical value into a probability distribution. , , are the query matrix, key matrix, and value matrix respectively. , , and are weight matrices. is the dimension of the key vector. is the number of attention heads in the multi - head attention mechanism; the self - attention mechanism can automatically focus on the importance of information at different positions in the sequence, discover the association between the wear trend and the working conditions (such as continuous cutting force, rotational speed) over a long period in the past, and thus predict the future wear trend.

[0132] S3.3: Define the ARIMA model as a residual model and the loss function as the mean squared error. Use the residual terms in the historical time series to train the residual model, fit the dynamic characteristics of the residual terms, and generate the predicted values of the residual terms, which helps to capture the impact of random fluctuations and anomalies in the data on the overall data, as well as the impact of unforeseen factors on tool wear, such as the sudden increase in wear caused by material inhomogeneity, and improve the accuracy of the overall prediction. The expression is as follows:

[0133] ;

[0134] In the formula, is the predicted value corresponding to the residual term at time , is the backward shift operator, is the autoregressive part, is the order of differencing, is the moving average part, is the white noise sequence.

[0135] S3.4: Combine the predicted trend values obtained from the trend model, the seasonal terms, and the predicted residual values obtained from the residual model, comprehensively consider the impacts of trends, seasons, and residuals, and fully consider the dynamic characteristics of time series data under different time scales and change patterns to generate more accurate predicted values and obtain the historical time series of the historical predicted values , enhancing the ability to cope with complex and changing construction environments and parameter variations. For example, for the prediction of formation pressure, while understanding the long-term trend changes, it is also possible to understand the periodic fluctuations and potential random anomalies; the expression is as follows:

[0136] ;

[0137] In the formula, is the historical predicted value at time

[0138] S3.5: Use the Pearson correlation coefficient to calculate the linear correlation degree between any two variables in the historical predicted values , construct a correlation matrix, and further understand the dynamic relationships between different variables;

[0139] S3.6: Set that the actual data feature vector contains feature parameters, generate a -dimensional random vector , where the actual data feature vector includes the feature time series, historical predicted values, and the correlation matrix, is the rd element of the random vector, ;

[0140] S3.7: Set the actual data feature vector as , the weight vector as , fuse the random vector with the actual data feature vector to generate a fused feature vector , so that the fused vector not only contains the key features of the actual data but also has some new features or dimensions, avoiding the limitations of analyzing only relying on a single type of data, and more comprehensively characterizing various states and relationships of the shield machine or TBM during construction. The expression is as follows:

[0141] ;

[0142] In the formula, is the th element of the fused feature vector, is the th element of the actual data feature vector, is the corresponding weight;

[0143] S3.8: Use the random forest algorithm to construct multiple decision trees. Each decision tree will select the optimal feature when splitting nodes to reduce impurity. Calculate the contribution of each feature to the reduction of impurity when splitting nodes in all decision trees, sum up the contributions of each feature in all decision trees and calculate the average value to obtain the feature importance score, and generate a fused feature vector importance sequence , so as to determine which features have a greater impact on the tool wear amount; where is the importance corresponding to the th element in the fused feature vector

[0144] S3.9: Use the multiple linear regression method, with the tool wear amount as the target variable, construct a correlation feature model, and establish a correlation function between various influencing factors and the tool wear amount through the feature importance score and the fused feature vector. The expression is as follows:

[0145] ;

[0146] In the formula, is the tool wear amount, is the constant term

[0147] Specifically, the specific steps of step S4 include:

[0148] S4.1: During the construction of a shield machine or TBM, in order to accurately grasp the detailed situation of the rock and soil layer, use external detection equipment to scan the rock and soil layer. Comprehensively use ground-penetrating radar and laser scanning equipment to scan the rock and soil layer in front of and around the shield machine or TBM. Use the ground-penetrating radar to generate two-dimensional image data of the stratum by processing the reflected waves of different stratum interfaces. At the same time, use the laser scanning equipment to measure the time difference from the emission to the reception of the laser beam, combine the principle of the speed of light, calculate the distance between the scanning point and the equipment, obtain the three-dimensional coordinate information of the spatial point, and combine the two-dimensional image data of the ground-penetrating radar with the coordinate information of the laser scanning to construct the initial three-dimensional point cloud data, visually display the spatial structure, undulation changes and the distribution of different strata of the rock and soil layer; and perform data verification on the initial three-dimensional point cloud data to generate the three-dimensional point cloud data of the rock and soil layer to ensure the accuracy of the collected three-dimensional point cloud data;

[0149] S4.2: Fuse the three-dimensional point cloud data with the features in the correlation feature model, map the feature values to the corresponding points in the three-dimensional point cloud model, and generate the fused point cloud data with tool wear labels, including spatial position information and environmental feature information related to tool wear;

[0150] S4.3: Preprocess the fused point cloud data and convert it into a format suitable for CNN input, including grid processing and normalization processing;

[0151] S4.4: Use a convolutional neural network to build a point cloud model, including an input layer, a multi-layer perceptron (MLP) layer, a max pooling layer, and a fully connected layer; among them, the input layer is used to receive the preprocessed point cloud data, the MLP layer extracts and transforms the features of each point through a multi-layer neural network, the max pooling layer is used to extract global features, and the fully connected layer maps the extracted features to the classification result and outputs the predicted probability of the tool state;

[0152] S4.5: Randomly initialize the parameters in the point cloud model and train the point cloud model based on the processed fused point cloud data; define the loss function of the point cloud model as the cross-entropy loss function, calculate the gradient of the loss function with respect to the network parameters through the backpropagation algorithm, and use stochastic gradient descent to update the network parameters, continuously adjusting the weights and biases of the model to minimize the loss function.

[0153] Specifically, the specific steps of data verification in S4.1 include:

[0154] Taking a single data point to be verified as the center, set an initial circular search area with a search radius of , quickly locate all surrounding data points within the search area, obtain the number of surrounding points , and calculate the initial point distribution density , and the expression is as follows:

[0155] ;

[0156] Set the search density range to , and determine whether the initial point distribution density is within the search density range; among them, is the high-density threshold of the search density range, is the low-density threshold of the search density range;

[0157] If , the point distribution density of the surrounding points is within the search density range, and keep the search radius as ;

[0158] If , then the surrounding points are sparsely distributed and the covered information is insufficient. Increase the search radius to , and recalculate the point distribution density within the new radius range until the point distribution density is within the search density range; among them, is the search coefficient;

[0159] If , the distribution of surrounding points is too dense, resulting in data redundancy or abnormal aggregation. The search radius is reduced to , and the point distribution density within the new radius is recalculated until the distribution density is within the search density range, ensuring that the verification range can cover sufficient valid information without including too many interfering points;

[0160] Use a pseudo-random number generator to randomly generate the position coordinates of multiple sampling points within the search area. For sampling points at different positions, data collection is performed separately to generate a sample data set , and calculate the average value of the sample data ; where is the number of samples in the sample data set, is the th sample data in the sample data set, ;

[0161] Calculate the difference between the sample data and the average value , and calculate the average value of the differences as the dispersion index of the search area; the expression is as follows:

[0162] ;

[0163] ;

[0164] Among them, the dispersion index is used to comprehensively measure the degree of data dispersion within the entire search area, which is conducive to judging the data fluctuation situation within the entire verification range and providing a basis for subsequent judgment of whether there are errors in data points or whether the geological conditions in the corresponding area are abnormal;

[0165] Set the dispersion threshold of the search area to , and judge whether there are errors in the data points to be verified; if , the data points to be verified have no errors, the verification is correct, and the data points are defined as verified points; , the data points to be verified have errors, and the data points are defined as preliminary error points;

[0166] Within the search area of the preliminary error points, data verification is performed on the surrounding points to obtain the number of surrounding points with errors and the specific average value , and calculate the average value of the surrounding errors , the expression is as follows:

[0167] ;

[0168] In the formula, is the The average value of the errors, ;

[0169] Set the geological change difference threshold to , and determine whether the preliminary error point is a boundary point; if , the preliminary error point is a true error, and define the preliminary error point as a boundary point; if , the preliminary error point is a false error, not a boundary point, and generate a substitute data point using the data interpolation algorithm based on the surrounding points so that the substitute data point has no error;

[0170] Generate the three-dimensional point cloud data of the rock and soil layer, including the calibrated points and the boundary points.

[0171] Specifically, the specific steps of step S5 include:

[0172] S5.1: Obtain the real-time time series , and after a series of processes, convert it into a format suitable for input to the point cloud model to obtain the predicted probability of the current tool state ;

[0173] S5.2: Set the tool damage threshold to , and judge the damage situation of the current tool; if , it is determined that the tool is in an undamaged state, and the shield machine or TBM continues to operate normally; if , it is determined that the tool is damaged, generate a tool damage signal, including relevant information about the tool damage, such as the approximate location of the tool damage and the degree of damage (predicted probability ), and transmit the tool damage signal to the tool-changing robot;

[0174] S5.3: After the tool change is completed, collect various parameter data of the new tool, add the parameter data of the new tool to the original parameter time series in chronological order, and update the timestamp information at the same time to ensure the temporal integrity of the data;

[0175] S5.4: Use the updated parameter time series to optimize and adjust the associated feature model. In an incremental learning manner, add the new point cloud data (including the point cloud data of the new tool under different working conditions) to the training set, fine-tune the point cloud model, update the parameters of the model, and analyze the damage situation of the new tool.

[0176] Embodiment 2

[0177] Please refer to Figure 5 , another embodiment provided by the present invention: a shield machine or TBM hob associated feature analysis wear system, including: a collection module, a feature analysis module, a prediction module, and an update module;

[0178] The acquisition module uses various sensors to collect parameter data related to tool wear in real time, including equipment operation parameters such as propulsion speed, cutterhead torque, and jacking force, external environment parameters such as formation soil type, hardness, and water content, and construction process parameters such as tunneling mode and grouting volume;

[0179] The feature analysis module extracts a series of statistical features of each parameter within the time period and introduces derivative parameters such as the ratio of cutterhead torque to propulsion speed and the product of jacking force and cutterhead torque. Combining with the tunneling situation of the shield machine or TBM, it constructs the change curve of the rock and soil layer by data fitting, generates and updates the parameter time series in real time;

[0180] The prediction module uses the STL decomposition method to dynamically analyze the parameter time series, decomposes the parameter time series into a trend term, a seasonal term, and a residual term, making the complex time series data structure clear, and uses the random forest algorithm to calculate the importance scores of each feature, constructing a correlation feature model; constructs the three-dimensional point cloud data of the rock and soil layer, combines with the correlation feature model, generates the fused point cloud data with tool wear labels, and constructs a point cloud model for judging the tool state, outputs the prediction probability of the tool state, and judges the tool state;

[0181] The update module is used to collect the parameter data of the new tool after the tool change is completed, update the parameter time series, and optimize and adjust the correlation feature model and the point cloud model with the new data to adapt to the new operating state and environmental changes of the shield machine or TBM, improving the accuracy and reliability of model discrimination.

[0182] Specifically, the prediction module includes a correlation unit and a judgment unit;

[0183] The correlation unit is used to dynamically analyze the parameter time series, calculate the importance scores of each feature, construct a correlation feature model, and comprehensively consider the complex relationship between various features and tool wear to effectively describe the tool wear state;

[0184] The judgment unit uses a ground penetrating radar and a laser scanning device to scan the rock and soil layer in front of and around the shield machine or TBM, constructs the three-dimensional point cloud data of the rock and soil layer, combines with the correlation feature model, generates the fused point cloud data with tool wear labels, and constructs a point cloud model specifically for judging the tool state; based on the real-time time series and the point cloud model, outputs the prediction probability of the tool state, and judges the tool state; once the tool is judged to be damaged, a tool damage signal is generated, and the control center activates the tool change robot.

[0185] In summary, in the above embodiments, the present invention utilizes various sensors to collect parameter data related to tool wear in real time, constructs a parameter time series that can reflect dynamic changes, collects and saves actual parameter values in chronological order according to different sampling frequencies; calculates relevant statistical and derivative features, constructs a change curve of the rock and soil layer, and updates the parameter time series in a timely manner. By analyzing the dynamic changes of the parameter time series, the random forest algorithm is used to clarify the influence of various features on tool wear, construct an associated feature model, and effectively describe the tool wear state; at the same time, the three-dimensional point cloud data of the rock and soil layer is fused with the associated feature model to generate fused point cloud data with tool wear labels, and a point cloud model is established; based on the real-time sequence and the point cloud model, the tool state is judged. Once the tool is damaged, the tool-changing robot is activated. After the tool is changed, new data is collected to optimize the associated features and the point cloud model, improving the adaptability and discrimination accuracy of the model to new working conditions of the shield machine or TBM.

[0186] The above is only the preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. Method for analyzing wear of associated features of shield machine or TBM hob, characterized in that, Including: Step S1: Collect parameter data related to tool wear in real time, perform preprocessing, and construct a parameter time series. Step S2: Calculate statistical features, derivative features, and the change curve of the rock and soil layer, and update the parameter time series. Step S3: Analyze the dynamic changes of the parameter time series, calculate the feature importance score using the random forest algorithm, and construct an associated feature model. Step S4: Construct three-dimensional point cloud data of the rock and soil layer, generate fused point cloud data with tool wear labels, and construct a point cloud model for discriminating tool states. Step S5: Calculate the prediction probability of the tool state and judge the tool state; once it is determined that the tool is damaged, generate a tool damage signal.

2. The method for analyzing wear of associated features of a shield machine or TBM cutter according to claim 1, wherein The specific steps of the said Step S2 include: S2.1: According to the cutter head torque collected in real time and the propulsion speed , calculate the ratio of the cutter head torque to the propulsion speed ; S2.2: Calculate the product of the jacking force and the cutterhead torque according to the real-time collected jacking force , ; S2.3: Continuously record the actual values of the rock and soil parameters at different time points, and draw the curve of the rock and soil parameters changing with time through data fitting. S2.4: Save the calculated statistical features, derivative features, and the change curves of the rock and soil layers to the parameter time series wherein the parameter time series includes a historical time series and a real-time time series .

3. The method for analyzing the wear of the shield machine or TBM hob correlation features according to claim 2, characterized in that, The specific steps of the data fitting in the said S2.3 include: During tunneling, according to the non-linear relationship between a certain rock and soil parameter and the tunneling distance, generate a fitting curve. Construct a measurement sample group , calculate the sum of squared errors ; Respectively, for , and take partial derivatives and set the partial derivatives to zero to obtain a system of partial derivative equations; Solve the partial derivative equations to calculate the final values of the fitting coefficient group of 4. The wear analysis method for shield machine or TBM hob correlation features according to claim 3, characterized in that The specific steps of solving the said partial derivative equations include: Let the functions in the partial derivative equations be partial derivative functions respectively , , , and use the approximation of the area of a curvilinear trapezoid to construct variable upper limit integral functions for each partial derivative function; Calculate the gradients of each variable upper limit integral function with respect to each fitting coefficient respectively. Let the weight coefficient be , and reconstruct and assign values to the integral function with a variable upper limit to generate a newly constructed assignment function; Combine multiple assignment functions into a function vector and calculate the Jacobian matrix of the function vector ; Settings , , The initial values are , and using the Newton iteration method, based on the current fitting coefficient values, function vectors, and Jacobian matrices, update , , as follows: Calculate the iterative error of the fitting coefficient group , set the iterative termination threshold as , and determine whether the iterative operation ends; when , it is determined that the iteration has not converged, and the iterative operation continues; when , it is determined that the iteration has converged, and the fitting coefficient value of the latest iteration is output.

5. The wear analysis method for shield machine or TBM hob correlation features according to claim 4, characterized in that, The specific steps of the said Step S3 include: S3.1: For the parameter time series , use STL decomposition to decompose it into a trend term, a seasonal term, and a residual term; S3.2: Use the trend term in the historical time series to train the trend model and output the trend term prediction value. S3.3: Use the residual term in the historical time series to train the residual model, fit the dynamic characteristics of the residual term, and generate the residual term prediction value. S3.4: Combine the trend term prediction value, the seasonal term in the historical time series, and the residual term prediction value to obtain the historical prediction value.

6. The wear analysis method for shield machine or TBM hob related features according to claim 5, characterized in that The specific steps of the said Step S3 also include: S3.5: Calculate the linear correlation degree between any two variables in the historical prediction value and construct a correlation matrix. S3.6: Set that the actual data feature vector contains feature parameters, and generate -dimensional random vector ; S3.7: Set the actual data feature vector as , the weight vector as , fuse the random vector with the actual data feature vector to generate a fused feature vector .

7. The method for analyzing the wear of the associated features of the shield machine or TBM hob according to claim 6, characterized in that The specific steps of the said Step S3 also include: S3.8: Using the random forest algorithm, construct multiple decision trees, calculate the feature importance scores, and generate the importance sequence of the fusion feature vector ; ; S3.9: Take the tool wear amount as the target variable, and establish an association function between various influencing factors and the tool wear amount through the feature importance score and the fused feature vector to generate an associated feature model.

8. The wear analysis method for shield machine or TBM hob correlation features according to claim 7, characterized in that, The specific steps of the said Step S4 include: S4.1: Use a ground penetrating radar and a laser scanning device to scan the rock and soil layer in front of and around the shield machine or TBM, and perform data verification to generate three-dimensional point cloud data of the rock and soil layer. S4.2: Fuse the three-dimensional point cloud data with the features in the associated feature model to generate fused point cloud data with tool wear labels. S4.3: Perform preprocessing on the fused point cloud data, including grid processing and normalization processing. S4.4: Construct a point cloud model, including an input layer, a multi-layer perceptron layer, a max pooling layer, and a fully connected layer, where the output is the prediction probability of the tool state. S4.5: Randomly initialize the parameters in the point cloud model, and train the point cloud model based on the processed fused point cloud data.

9. The method for analyzing the wear of the associated characteristics of the shield machine or TBM hob according to claim 8, wherein The specific steps of the data verification in the said S4.1 include: Taking the data point to be verified as the center, set an initial search area, obtain the number of surrounding points within the search area, and calculate the initial point distribution density ; Set the search density range to be , and determine whether the point distribution density is within the search density range; If , the point distribution density is within the search density range; If , expand the search range until the point distribution density is within the search density range; If the search range is reduced until the point distribution density is within the search density range; Obtain the position coordinates of multiple sampling points to generate a sample data set , and calculate the average value of the sample data ; Calculate the sample data and the said average value the difference between , and calculate the average value of the differences ; Set the discrete threshold of the search area to , and determine whether there is an error in the data point; if , the data point has no error, and define the data point as a calibrated point; , the data point has an error, and define the data point as a preliminary error point; Within the search area of the preliminary error points, data verification is performed on the surrounding points to obtain the number of points with errors among the surrounding points and the average value , and the average value of the surrounding errors is calculated ; where is the average value of the th error Set the geological change difference threshold to , and determine whether the preliminary error point is a boundary point; if , the preliminary error point is a boundary point; if , generate alternative data points without errors; Generate three-dimensional point cloud data of the rock and soil layer, including calibrated points and boundary points.

10. The wear analysis method for shield machine or TBM hob correlation features according to claim 9, characterized in that, The specific steps of the said Step S5 include: S5.1: Obtain the real-time time series , calculate the predicted probability of the tool state ; S5.2: Set the damage threshold of the tool to , and judge the damage condition of the current tool; if , the tool is in an undamaged state; if , the tool is damaged and a tool damage signal is generated; S5.3: After the tool change is completed, collect various parameter data of the new tool, and at the same time update the parameter time series and timestamp information; S5.4: Use the updated parameter time series to optimize and adjust the associated feature model and the point cloud model, and perform damage analysis on the new tool.

11. Shield machine or TBM hob associated feature analysis wear system, which is used to implement the shield machine or TBM hob associated feature analysis wear method as described in any one of claims 1-10, characterized in that, Including: Data acquisition module, feature analysis module, prediction module and update module; The data acquisition module uses various sensors to collect parameter data related to tool wear in real time; The feature analysis module calculates statistical parameters and derivative parameters according to the initially collected parameters, and combines with the tunneling situation of the shield machine or TBM to construct a change curve of the rock and soil layer, and generates a parameter time series; The prediction module includes an association unit and a judgment unit; The association unit is used to dynamically analyze the parameter time series, calculate the importance scores of each feature, and construct an associated feature model; The judgment unit is used to construct three-dimensional point cloud data of the rock and soil layer, combine it with the associated feature model to generate fused point cloud data with tool wear labels, and construct a point cloud model for discriminating tool states; output the prediction probability of the tool state, and judge the tool state; The update module is used to collect parameter data of the new tool after the tool change is completed, update the parameter time series, and optimize and adjust the associated feature model and the point cloud model with the new data.

Images