Estimation device and estimation method

Abstract

To make it possible to reach target coordinates without an error as much as possible, by quickly performing correction of a power point.SOLUTION: An estimation device 30 for estimating a power point position to be recommended for progress of a shield machine 10 comprises: an estimation device data acquisition unit 41 for acquiring a plurality of pieces of control data for the shield machine 10 and ring reach target coordinates to be a reach target at a ring end; and a power point recommendation position estimation unit 44 for estimating a power point recommendation position to be recommended in progress to the ring reach target coordinates. The power point recommendation position estimation unit 44 sequentially estimates, for each predetermined time interval or predetermined excavation length during excavation, the power point recommendation position for making reach the reach target coordinates in remaining excavation length.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an estimation device and an estimation method. [Background technology]

[0002] One tunnel construction method uses a shield machine. The shield machine is operated to excavate according to a planned excavation instruction. The excavation instruction contains tunnel excavation instruction values, which indicate (1) the horizontal direction (instructed direction) or the difference in the left and right shield jack strokes at the end of each ring excavation, and (2) the vertical direction (instructed pitching) or the difference in the upper and lower shield jack strokes. The shield machine is controlled to excavate along the planned tunnel line. The shield machine is equipped with multiple shield jacks located at the rear, and the excavation direction is changed by controlling these shield jacks. In other words, the direction is controlled by changing the force point acting on the shield machine by switching each shield jack on or off (or by controlling the pressure of each shield jack). For example, if the thrust of all shield jacks is equalized, the force point will be centered horizontally and vertically, so in theory the shield machine will move in a straight line.

[0003] Conventionally, there has been an estimation device that estimates the position of the force point of a shield machine (see, for example, Patent Document 1). This estimation device includes a direction data acquisition unit that acquires directional data related to the excavation direction of the shield machine, and a force point position estimation unit that estimates the position of the force point at which the shield machine should excavate by inputting data including at least the directional data into an estimation model. The estimation model is a model created by performing machine learning using training data in which the position of the force point at which the shield machine should excavate is associated with input data including data corresponding to the directional data. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2019-143385 Summary of the Invention [Problem to be solved by the invention]

[0005] However, depending on the structure of the shield machine (for example, whether or not the shield jack is equipped with a hydraulic control valve), the position of the force point that can be set may be restricted, which may result in a discrepancy between the actual force point and the recommended force point. If the force point cannot be aligned with the recommended force point, it will be difficult to get the shield machine to reach the target coordinates. Furthermore, due to various factors such as the quirks of the shield tunneling machine (such as a tendency to turn left even when digging straight), the center of gravity, uneven ground conditions, and the amount of overexcavation, it may not be possible to reach the target coordinates even if the force point is set to the recommended force point and tunneling continues.

[0006] In these cases, the deviation of the reached coordinates from the target coordinates is recognized only after the target coordinates are approached, so the progress of the shield tunneling machine is not stable and subsequent trajectory correction may be difficult.

[0007] From this perspective, the present invention provides an estimation device and an estimation method that can quickly correct the point of force to reach the target coordinates with as little error as possible. [Means for solving the problem]

[0008] The estimation device according to the present invention is a device that estimates a recommended force point position for the advancement of a shield tunneling machine. This estimation device includes an estimation device data acquisition unit that acquires multiple pieces of control data for the shield tunneling machine and ring target coordinates that are the targets to be reached at the end of the ring, and a recommended force point position estimation unit that estimates a recommended force point position for advancement to the ring target coordinates. The recommended force point position estimation unit estimates the recommended force point position at predetermined time intervals or for each predetermined excavation length during excavation.

[0009] In the estimation device according to the present invention, the point of force can be quickly corrected to reach the target coordinates with as little error as possible.

[0010] The shield machine may further include a ring arrival coordinate estimating unit that estimates a predicted ring arrival coordinate that is predicted to be reached at the end of the ring, or that estimates a predicted arrival deviation amount of the predicted ring arrival coordinate from an inference reference line that is set based on the direction of the shield machine at its current position. The ring arrival coordinate estimating unit estimates the predicted ring arrival coordinate or the predicted arrival deviation amount of the predicted ring arrival coordinate from the inference reference line at predetermined time intervals or for each predetermined excavation length during excavation.

[0011] In this way, it is possible to check the deviation in direction relative to the target.

[0012] An inference reference length indicating a predetermined distance from the shield machine on the inference reference line may be used as the reference. The recommended force point position estimating unit may output the recommended force point position by inputting at least one or more pieces of control data excluding the force point position and a planned deviation amount between the reference length planned target coordinates and the inference reference length end point after moving along a planned travel path geometrically determined from the current position by the inference reference length into a learning model for estimating the recommended force point position.

[0013] The learning model for estimating the recommended force point position is a model created based on data acquired during the past progress of the shield tunneling machine, and is created by learning using learning data in which one or more control data excluding the force point position and the amount of reach deviation between the reference length arrival coordinate, which is the actual position after moving the inference reference length from the current position, and the inference reference length end point are used as explanatory variable data, and the force point position actually controlled while the shield tunneling machine moves the inference reference length from the current position is used as objective variable data.

[0014] The ring arrival predicted coordinate estimating section may include a reference length arrival predicted coordinate estimating section and a ring arrival predicted coordinate output section. The reference length arrival predicted coordinate estimation unit inputs at least a plurality of control data including the force point position into a learning model for arrival coordinate estimation, and outputs the amount of arrival predicted deviation between the reference length arrival predicted coordinate at which the shield tunneling machine is predicted to arrive after traveling the inferred reference length and the inferred reference length end point. The ring arrival predicted coordinate output unit determines the ring arrival predicted coordinate based on the arrival predicted deviation amount, or determines the arrival predicted deviation amount from the inference reference line to the ring arrival predicted coordinate.

[0015] The learning model for estimating arrival coordinates is a model created based on data acquired during the past progress of the shield tunneling machine, and is a model created by learning using learning data in which a plurality of control data including the force point position actually controlled while the shield tunneling machine moved the inference reference length from its current position is used as explanatory variable data, and the arrival deviation amount between the reference length arrival coordinate, which is the actual position after moving the inference reference length from the current position, and the inference reference length end point is used as objective variable data.

[0016] The estimation method according to the present invention is a method for estimating a recommended force point position for the advancement of a shield tunneling machine. This estimation method comprises an estimation device data acquisition step of acquiring multiple pieces of control data for the shield tunneling machine and ring target coordinates that are the targets to be reached at the end of the ring, and a recommended force point position estimation step of estimating a recommended force point position for the advancement to the ring target coordinates. The recommended force point position estimation step is performed at predetermined time intervals or for each predetermined excavation length during excavation.

[0017] In the estimation method according to the present invention, the point of force can be quickly corrected to reach the target coordinates with as little error as possible. [Effects of the Invention]

[0018] According to the present invention, the point of force can be quickly corrected to reach the target coordinates with as little error as possible. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a schematic configuration diagram of a shield automatic driving system according to an embodiment of the present invention. [Figure 2] FIG. 1 is a diagram illustrating the configuration of a shield tunneling machine. [Figure 3] This is an example of a system screen used to manage the excavation of a shield machine. [Figure 4] A diagram for explaining explanatory variables and target variables in learning a trained model for estimating a recommended force point position and a trained model for estimating a reached coordinate. [Figure 5] 1 is an example of a training dataset used to train an AI model for estimating recommended force point positions. [Figure 6] FIG. 10 is a diagram for explaining normalization of the point of force position. [Figure 7] 10A and 10B are diagrams for explaining weighted averaging processing of the position of the point of force; [Figure 8] This is an example of a training dataset used to train an AI model for estimating predicted arrival coordinates. [Figure 9] This is an illustration of learning that uses information measured when a shield tunneling machine actually excavated at a past shield construction site. [Figure 10] 10 is an example of an estimation dataset used for estimation using a trained model for estimating recommended force point positions. [Figure 11] FIG. 10 is a diagram for explaining explanatory variable data of estimation data used for estimation by a trained model for estimating a recommended force point position. [Figure 12] 10 is an example of an estimation dataset used for estimation using a trained model for estimating predicted arrival coordinates. [Figure 13] 10A and 10B are diagrams for explaining a process of estimating predicted arrival coordinates; [Figure 14] The shield machine is shown at the start of the first ring. [Figure 15] This shows the state when the shield tunneling machine has excavated 20 mm from the start position of the first ring. [Figure 16] This shows the state when the shield tunneling machine has excavated 520 mm from the start position of the first ring. [Figure 17] 1 is an example of a flowchart for setting a level of an autonomous driving system. [Figure 18] FIG. 10 is an example of a flow diagram of an estimation method in a shield automatic driving system. [Figure 19] This is a diagram to explain variations in the inference reference length, and is an image of learning performed using information measured when a shield tunneling machine actually excavated at a past shield construction site. DETAILED DESCRIPTION OF THE INVENTION

[0020] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Each drawing is merely a schematic illustration to allow a sufficient understanding of the present invention. Therefore, the present invention is not limited to the illustrated examples. In each drawing, common or similar components are designated by the same reference numerals, and redundant explanations thereof will be omitted.

[0021] <Configuration of shield automated driving system according to the embodiment> The configuration of the shield automatic driving system 1 will be described with reference to Fig. 1. Fig. 1 is a schematic configuration diagram of the shield automatic driving system 1. Note that arrows shown in Fig. 1 indicate the flow of data.

[0022] The automatic shield tunneling operation system 1 is a system for realizing automatic operation of a shield tunneling machine 10. The automatic shield tunneling operation system 1 mainly comprises the shield tunneling machine 10, an excavation management device 20, and an estimation device 30. Note that the automatic shield tunneling operation system 1 can be operated manually as necessary.

[0023] The configuration of the shield machine 10 will be described with reference to Figure 2. Figure 2 is a configuration diagram of the shield machine 10. Figure 2 shows a longitudinal cross section of the shield machine 10. In this embodiment, the description will be based on the assumption that the shield machine 10 is an earth pressure type, but the type of shield machine 10 is not limited to this. In other words, the automatic shield driving system 1 can also control shield machines that use construction methods other than the earth pressure type. In the description of the shield machine 10, the tunnel excavation direction is defined as the forward direction (forward), and up, down, left, and right are defined.

[0024] As shown in Figure 2, a shield tunneling machine 10 rotates a cutter 11 attached to its front end and presses a shield jack 12 against the joint surface 9a of a segment 9 attached to its rear, thereby obtaining thrust and excavating the natural ground. Multiple shield jacks 12 are arranged at a predetermined pitch around the circumferential direction of the tunnel (only those arranged on the lower side of the shield tunneling machine 10 are shown in Figure 2).

[0025] The main body (machine main body) of the shield tunneling machine 10 is composed of a cylindrical front body 13 and a rear body 14. Inside the front body 13, a gyro sensor 15, a two-axis inclinometer 16, a center-bending jack 17, etc. are provided. The two-axis inclinometer 16 measures pitching and rolling.

[0026] The excavation management device 20 shown in Figure 1 is a device that controls and manages excavation by the shield machine 10. The excavation management device 20 shown in Figure 1 is a computer comprising, for example, a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), etc. As shown in Figure 1, the excavation management device 20 comprises a various data storage unit 21, a system display and operation unit 22, an excavation instruction data acquisition unit 23, a measurement data acquisition unit 24, an arithmetic processing unit 25, a shield machine control unit 26, an estimation result acquisition unit 27, and an estimation device data transfer unit 28.

[0027] The various data storage unit 21 is a unit that stores all data acquired and calculated within the excavation management device 20. The various data storage unit 21 is, for example, an HDD (hard disk drive) or SSD (solid state drive) for storing data. The various data storage unit 21 stores various data as time-series data. Time-series data is data that is recorded in association with time at regular time intervals (for example, every 5 seconds).

[0028] The system display and operation unit 22 is a section where the operator of the shield tunneling machine 10 and the responsible on-site staff perform monitoring, operation, and settings for construction management. The system display and operation unit 22 is, for example, a display, keyboard, mouse, etc. The system display and operation unit 22 can display the information required for monitoring not only numerically but also in layouts and graphs that make it easier to distinguish. An example of a screen displayed on the system display and operation unit 22 is shown in Fig. 3. Fig. 3 is an example of a system screen 60 used to manage the excavation of the shield tunneling machine 10. Fig. 3 shows the excavation status (system screen 60 displayed during excavation).

[0029] The excavation instruction data acquisition unit 23, measurement data acquisition unit 24, calculation processing unit 25, shield tunneling machine control unit 26, estimation result acquisition unit 27 and estimation device data transfer unit 28 shown in Figure 1 are realized, for example, by program execution processing by a CPU.

[0030] The excavation instruction data acquisition unit 23 shown in Figure 1 is a part that acquires design information, excavation instruction data from excavation instructions, etc. In principle, this data is acquired before excavation begins, but it is also possible to acquire it during excavation, in which case it is updated with new data.

[0031] The design information includes information such as the alignment that serves as the basis for the shield machine's excavation. For example, the design information includes linear information such as the tunnel's center coordinates, direction, and gradient, as well as the shape and dimensions of the segments that make up the tunnel, their combination, and the position of the K-segments.

[0032] The excavation instructions are documents or data that contain the target jack stroke difference, target direction, target pitching, and various control instruction values for the shield machine for the upcoming excavation. The site staff leader (the person responsible for creating the excavation instructions) decides the content to be written in the excavation instructions based on the current position and attitude information of the shield machine. The excavation instruction data acquisition unit 23 acquires the excavation instruction values entered into the system by the site staff leader.

[0033] The target jack stroke difference, target direction, and target pitching were decided by the site staff in charge to determine where the shield machine should be advanced from the next ring onwards.The target jack stroke difference, target direction, and target pitching are not necessarily adopted as is from the design information, but are determined by comparing the design information with the survey results of the position coordinates, direction (yawing), pitching, rolling, and other attitudes of the current shield machine that has already excavated, and taking into account any deviations to the left, right, up, or down.

[0034] Furthermore, the site staff in charge can also decide the instructions that control the direction of the shield machine in order to excavate toward the target.The instructions in the excavation instructions include the design alignment for each ring, segment assembly instructions, shield machine instructions, earth pressure, backfill, additive material, tail sealer injection, etc.The instructions that control the direction of the shield machine include the left-right stroke difference, up-down stroke difference, front body direction (gyro), pitching, and center-bending angle (only if there is a center-bending mechanism, same below).

[0035] The measurement data acquisition unit 24 shown in Figure 1 communicates with the shield tunneling machine 10 and acquires data such as measurement values of the various sensors equipped in the shield tunneling machine 10, their calculated values, and control signals for the shield tunneling machine 10. The measurement data acquisition unit 24 acquires data at predetermined intervals (for example, once per second). The various sensors in the shield tunneling machine 10 have converters that convert their measurement values into digital values, and the measurement data acquisition unit 24 may acquire data by communicating with these converters.

[0036] Items that can be obtained from the main sensors include net stroke, jack stroke, jack speed, shield jack pressure, pressure control valve pressure, total thrust, articulation jack stroke, articulation angle, gyro direction, pitching, rolling, level, tail clearance, cutter rotation speed, cutter torque, copy cutter stroke, copy cutter position, screw rotation speed, gate opening, screw pressure, screw torque, earth pressure at the face, and pressure inside the chamber. Note that it is not necessary to use all of the items listed here for automatic operation of the shield tunneling machine 10; only some of the items may be used.

[0037] The control signals for the shield machine 10 are signals that correspond to the ON / OFF of various switches on the control panel used by the operator of the shield machine 10, and volume knobs that adjust the size of various indicators. The main control signals for the shield machine 10 include excavation mode ON / OFF, assembly mode ON / OFF, shield jack selection ON / OFF, excavation speed adjustment, cutter rotation direction left / right switching, cutter rotation speed adjustment, center bending angle adjustment, screw rotation speed adjustment, copy cutter stroke adjustment, screw rotation direction forward / reverse switching, and gate opening adjustment.

[0038] The data acquired by the measurement data acquisition unit 24 is written and stored in the various data storage unit 21. In addition, data that requires calculation is transferred to the calculation processing unit 25, and the calculation results are written and stored in the various data storage unit 21.

[0039] The calculation processing unit 25 shown in Figure 1 is a part that performs calculations on various data used within the excavation management device 20. The various data refers to data recorded in the various data storage unit 21, and includes excavation instruction data, various instrument measurement data, survey result data, etc. The calculation processing unit 25 may also perform calculations on the recommended force point position and predicted ring arrival coordinates estimated by the estimation device 30.

[0040] The excavation management device 20 has functions (e.g., application programs) for realizing, for example, "excavation management," "survey management," "shield excavation planning," and "segment assembly planning," and performs calculations on data appropriate for these functions.

[0041] As part of the calculation process for creating data for excavation management, the calculation processing unit 25 performs, for example, unit conversion of the obtained raw data and conversion processing into physical quantities, and performs statistical processing on multiple pieces of data to output average values, maximum values, minimum values, standard deviations, and integrated values. The calculation processing unit 25 also performs statistical processing such as time moving averages on time series data, and outputs the data as data for each fixed distance of excavation distance. Furthermore, when one ring of excavation is completed, the calculation processing unit 25 performs statistical processing for that ring and outputs a ring report as a ring representative value.

[0042] As a calculation process for creating data for survey management, the calculation processing unit 25 performs, for example, a survey calculation process for the position and attitude of the shield machine 10 based on the input survey results. The survey calculation process calculates the latest position, orientation, and attitude of the shield machine 10 based on the jack stroke difference, gyro orientation value, pitching value, etc., using the position, orientation, and attitude information of the shield machine 10 at the start of excavation as a reference. For survey-related data, a relative coordinate conversion process is performed to compare it with the design alignment. For data for which a control reference value or other setting has been made in advance for control reference value determination, the calculation processing unit 25 compares the control reference value with the latest data and issues an alarm, etc. The control reference value is set by the site staff manager and is a standard value required for safe and accurate excavation. Depending on the management method, multiple control reference values can be set for a single piece of data.

[0043] As part of the computational processing for creating data for segment assembly planning, the computational processing unit 25 inputs the center linear coordinates of the shield tunnel design, for example, before shield construction begins. After shield construction begins, the computational processing unit 25 also inputs the coordinates of the current position of the segment at the end of the tunnel face based on the most recent survey results (face center position coordinate, face orientation, face pitching). Furthermore, the computational processing unit 25 compares the face center position coordinate of the segment at the end of the tunnel face with the center linear coordinates of the design to calculate horizontal and vertical deviations. These are designated as the "segment reference position." On-site personnel determine and input the "segment assembly settings" for each subsequent ring from the "segment reference position."

[0044] As part of the calculation process for creating data for shield tunneling planning, the calculation processor 25 inputs the center linear coordinates of the shield tunnel design, for example, before shield construction begins. After shield construction begins, the calculation processor 25 also inputs the coordinates of the shield machine 10's current position (leading edge, center bend, and trailing edge) based on the most recent survey results. Field personnel determine and input the target ring coordinates for each ring. Instead of the target ring coordinates, the horizontal and vertical deviations from the design center linear coordinates may be input. The calculation processor 25 may also automatically set the target ring coordinates without relying on input from the field personnel. For example, if the current horizontal deviation is 20 mm and the deviation is desired to be 0 mm 10 rings ahead, the calculation processor 25 may determine a proportional adjustment (correction) of the deviation amount by 2 mm per ring, and set the target ring coordinates based on that determination. Once the target ring coordinates are determined, the calculation processor 25 calculates the leading edge coordinates and the direction of the shield machine 10 at each ring it passes from its reference position.

[0045] "Segment assembly settings" refers to selecting the segment type and the insertion position of the K segment. Segment types include straight, single-tapered, and double-tapered. For tapered segments, the amount of change in orientation and pitching for that segment can be adjusted depending on the insertion position of the K segment. Using the "segment reference position" as the starting point, the "segment assembly settings" for the next ring are used to calculate the center position coordinates, face orientation, and face pitching on the face side of the next ring. Furthermore, calculations are performed for the next and subsequent rings using the same procedure as above to calculate the center coordinates, face orientation, and face pitching. The results of these calculations are called the "segment assembly plan."

[0046] The shield machine control unit 26 shown in Figure 1 is the part that controls the shield machine 10. Control by the shield machine control unit 26 includes control of the force point of the shield machine 10. The force point is the position where the total thrust of the shield machine 10 acts, and by adjusting this position up, down, left, or right, the excavation direction of the shield machine 10 can be controlled. For example, if the force point is set to the right of a vertical line passing through the center of the shield machine, the shield machine 10 will displace to the left, and if the force point is set to the left of the vertical line, the shield machine 10 will displace to the right.

[0047] In the automatic shield driving system 1, it is possible to set and select control levels 1, 2, and 3 depending on the specifications of the shield tunneling machine 10 and the construction conditions on site. At level 1, it is assumed that the shield jack can be selected to be ON or OFF. Levels 2 and 3 assume that hydraulic control valves are installed on the shield jacks. At levels 2 and 3, the hydraulic pressure of each shield jack can be adjusted with the hydraulic control valve while all shield jacks are being propelled. The difference between levels 2 and 3 is whether directional control is performed manually or automatically.

[0048] (Level 1) At level 1, as described above, the shield machine 10 is not equipped with a hydraulic control valve for the shield jack, and the direction is controlled by manually changing the position of the force point of the thrust by controlling the ON / OFF of the shield jack. At level 1, the shield machine control unit 26 calculates which shield jack to turn ON / OFF according to the force point at which the shield machine 10 is desired to excavate, and outputs the calculation results. For example, it refers to the recommended force point position estimated by the estimation device 30 to determine which shield jack should be turned ON / OFF. The shield operator refers to the recommended ON / OFF of the shield jack and manually sets the ON / OFF of the shield jack to advance the shield machine 10.

[0049] For example, an ON jack "opens" the hydraulic pump circuit, thereby extending the jack and transmitting the propulsive force to the segment, and an OFF jack "closes" the circuit, thereby not extending the jack and not transmitting the propulsive force to the segment. By controlling the ON / OFF of the shield jacks in this way, the point of force can be adjusted, and the direction of the shield machine 10 can be controlled. For example, by having more ON shield jacks on the right side than on the left side, the point of force will be on the right, and the direction of the shield machine 10 can be controlled to the left. Furthermore, the difference in the number of jacks on the left and right sides determines the "size" of the point of force, and therefore the "size" of the amount of directional deviation.

[0050] (Level 2) At level 2, as described above, the shield machine 10 is equipped with hydraulic control valves for the shield jacks, and by controlling the hydraulic pressure of each shield jack, the position of the thrust force point is manually changed to control direction. At level 2, the shield machine control unit 26 calculates a command pressure value for the shield jack's hydraulic control valve according to the force point at which the shield machine 10 is desired to excavate, and outputs the calculation result. For example, the shield operator inputs the force point position with reference to the recommended force point position estimated by the estimation device 30, and the shield machine control unit 26 calculates a command pressure value for the hydraulic control valve of each shield jack. The shield machine control unit 26 outputs each command pressure value to the shield machine 10, and the shield machine 10 advances at the input command pressure value.

[0051] Controlling this hydraulic control valve adjusts the point of force, making it possible to control the direction of the shield machine 10. For example, by making the right-side jack pressure higher than the left-side jack pressure, the point of force moves to the right, and the direction of the shield machine 10 can be controlled to the left. Furthermore, the total difference between the left and right jack pressures determines the "size" of the point of force, and therefore the "size" of the amount of directional deviation.

[0052] (Level 3) At Level 3, as described above, the shield machine 10 is equipped with hydraulic control valves for the shield jacks, and by controlling the hydraulic pressure of each shield jack, the position of the thrust force point is automatically changed to control the direction. At Level 3, the shield machine control unit 26 calculates a command pressure value for the shield jack's hydraulic control valve according to the thrust force point at which the shield machine 10 is desired to excavate, and automatically controls the pressure of each shield jack based on the calculation results. For example, the command pressure value for the shield jack's hydraulic control valve is calculated based on the recommended thrust point position estimated by the estimation device 30, and the pressure of each shield jack is automatically controlled based on the calculation results to propel the shield machine 10. Even at Level 3, if the operator desires manual operation, such as when the shield machine 10 exhibits behavior that differs from the progress toward the ring's target coordinates, the shield operator can adjust the thrust force position by operating a manual button.

[0053] The conventional method of controlling the direction of a shield tunneling machine by turning the shield jacks on and off was to have the shield operator select which shield jacks to turn on and off at his discretion based on the excavation instructions (referred to as "Level 0"). This conventional method was highly dependent on the skill of the shield operator, and was prone to problems such as deviations from the instructions and meandering of the tunneling.

[0054] The estimation result acquisition unit 27 shown in FIG. 1 acquires the recommended force point position and predicted ring arrival coordinates obtained through data processing by the estimation device 30 (described later). The estimation result acquisition unit 27 acquires, for example, the recommended force point position (α, β) in association with the ring number and net stroke (mm). The net stroke is the excavation distance of the shield machine 10 from the start of excavation for the next segment ring after the assembly of the segment rings is completed, and is calculated, for example, by averaging the net stroke lengths of the upper, lower, left, and right shield jacks. The estimation result acquisition unit 27 also acquires, for example, predicted arrival coordinates (X, Y, Z) in association with the ring number and net stroke (mm). In this way, the ring number and net stroke at the time of the estimation may be written into the output data from the estimation device 30. In this way, if there is a difference of a certain value or more between the ring number and net stroke at the current position of the shield machine 10 when the data was acquired and the net stroke, it is possible to determine that the estimation device 30 is not providing accurate output due to an abnormality or the like, and to display an alarm. If you are operating at level 3, you can automatically transition to level 2, where you manually enter the emphasis.

[0055] The recommended force point position is transferred to the shield machine control unit 26 and the various data storage unit 21, where it is processed for system display. In addition, the predicted arrival coordinate value is transferred to the various data storage unit 21 and the calculation processing unit 25, where it is processed for system display.

[0056] The data transfer unit for estimation device 28 shown in Figure 1 is a part that converts data required by the estimation device 30 (described later) into a format that can be read by the estimation device 30 and transfers the converted data. The data required by the estimation device 30 is set in advance from the various data storage unit 21. There are no particular restrictions on the timing or frequency of data transfer; for example, when the excavation distance (net stroke value as data) of the shield tunneling machine 10 is a multiple of a set distance (20 mm), the data transfer unit for estimation device 28 transfers the latest value.

[0057] The data transfer unit 28 for the estimation device transfers to the estimation device 30 at least a plurality of control data for the shield tunneling machine 10, including the force point position, and the ring target coordinates that are the target to be reached at the end of the ring (the target to be reached at the time when excavation of one ring of the segment is completed).

[0058] The data transfer items by the estimation device data transfer unit 28 include, for example, ring number, net stroke, shield tunneling machine tip coordinates (X, Y, Z), shield tunneling machine center bend coordinates (X, Y, Z), shield tunneling machine rear end coordinates (X, Y, Z), upper shield jack stroke, right shield jack stroke, lower shield jack stroke, left shield jack stroke, upper and lower stroke difference (difference between upper shield jack stroke and lower jack stroke), left and right stroke difference (difference between right shield jack stroke and left jack stroke), upper shield jack speed, right shield jack speed, lower shield jack speed, left shield jack speed, front body Heading (gyro), front barrel pitching, front barrel rolling, rear barrel heading (gyro), rear barrel pitching, rear barrel rolling, left face earth pressure, right face earth pressure, copy cutter stroke, up and down center bending angle, left and right center bending angle, cutter torque, cutter right rotation, cutter left rotation, cutter rotation speed, screw rotation speed, gate opening, upper tail clearance, right tail clearance, lower tail clearance, left tail clearance, shield jack pressure, total thrust, horizontal force point position, vertical force point position, horizontal moment, vertical moment, horizontal deviation of the tip of the shield tunneling machine, vertical deviation of the shield tunneling machine, segment length, target coordinates reached by the ring (X, Y, Z), radius of curvature, etc.

[0059] The configuration of the estimation device 30 will be described with reference to Fig. 1. The estimation device 30 shown in Fig. 1 is a device that estimates a recommended position of the point of force (recommended force point position) for progression to a ring arrival target coordinate, and a predicted ring arrival coordinate that is predicted to be reached at the end of the ring. The estimation device 30 is, for example, a computer including a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), etc.

[0060] The installation location of the estimation device 30 is not particularly limited. The estimation device 30 may be, for example, an edge device installed at a construction site, or a server (including a cloud system configuration) installed on the Internet that communicates via the Internet. When the estimation device 30 is an edge device, communication does not go via the Internet, which reduces communication costs and achieves high security. Furthermore, in the automatic operation of the shield tunneling machine 10, a time lag between judgment and control can affect excavation accuracy, so high real-time performance is required. In this regard, an edge device has a stable communication environment, making real-time processing possible. Furthermore, since communication does not go via the Internet, the risk of delaying the process due to problems such as communication failures can be avoided.

[0061] As shown in FIG. 1, the estimation device 30 includes a trained model recording unit 31 for estimating a recommended force point position and a trained model recording unit 32 for estimating a destination coordinate. The estimation device 30 also includes an estimation device data acquisition unit 41, an estimation data preprocessing unit 42, an estimation data selection unit 43, a recommended force point position estimation unit 44, a recommended force point position output unit 45, and a predicted ring arrival coordinate estimation unit 46. The predicted ring arrival coordinate estimation unit 46 includes a predicted reference length arrival coordinate estimation unit 47 and a predicted ring arrival coordinate output unit 48. The estimation device 30 also includes a learning data selection unit 51, a learning data preprocessing unit 52, a force point position learning unit 53, and a destination coordinate learning unit .

[0062] The trained model recording unit 31 for estimating a recommended force point position and the trained model recording unit 32 for estimating a destination coordinate are units that store trained models. These components are, for example, a hard disk drive (HDD) or a solid state drive (SSD) for storing data.

[0063] The estimation device data acquisition unit 41, estimation data preprocessing unit 42, estimation data selection unit 43, recommended force point position estimating unit 44, recommended force point position output unit 45, predicted ring arrival coordinate estimating unit 46, predicted reference length arrival coordinate estimating unit 47, and predicted ring arrival coordinate output unit 48 are realized, for example, by a CPU executing a program. Similarly, the learning data selection unit 51, learning data preprocessing unit 52, force point position learning unit 53, and arrival coordinate learning unit 54 are realized, for example, by a CPU executing a program.

[0064] Note that the functions of the estimation device 30 shown in FIG. 1 are categorized for the sake of convenience, and the categorization of the functions is merely an example. Therefore, the functions of the estimation device 30 can be integrated or divided as appropriate. For example, the estimation data preprocessing unit 42 and the estimation data selection unit 43 can be configured as a single function, or the learning data selection unit 51 and the learning data preprocessing unit 52 can be configured as a single function. Furthermore, for example, the estimation device data acquisition unit 41 can be divided into a function for sending data to the estimation data preprocessing unit 42 and a function for sending data to the learning data selection unit 51.

[0065] The trained model recording unit 31 for estimating recommended force point positions is a unit that stores trained models for estimating recommended force point positions. The trained model for estimating recommended force point positions is an AI (Artificial Intelligence) model that has learned the correspondence between the force point position, which serves as the objective variable, and its explanatory variables. The force point position, which serves as the objective variable, is preferably normalized. The trained model recording unit 31 for estimating recommended force point positions may store multiple trained models for estimating recommended force point positions. The trained model recording unit 31 for estimating recommended force point positions may store multiple trained models according to, for example, an inference reference length. The inference reference length M (see FIG. 4) will be described later. The trained model recording unit 31 for estimating recommended force point positions may also store multiple trained models according to different cross-sectional diameters (small cross section, medium cross section, large cross section).

[0066] The inference reference length M will be described with reference to Fig. 4. Fig. 4 is a diagram for explaining explanatory variables and objective variables in learning of a trained model for estimating a recommended force point position and a trained model for estimating arrival coordinates.

[0067] In this embodiment, an inference reference line B is virtually set as a means for learning and estimating the AI model. The inference reference line B is set based on the direction of the shield tunneling machine 10 at its current position. The inference reference line B is, for example, a line extending forward from the position reference point (one example is the tip) of the shield tunneling machine 10 (a line coaxial with the rotation center line of the cutter 11), and in this case the inference reference line B is the travel path of the shield tunneling machine 10 when it is assumed to move straight ahead.

[0068] An inference reference length M indicating a predetermined distance from the position reference point of the shield tunneling machine 10 (for example, the tip) is set on the inference reference line B. The inference reference length start point F is the position reference point of the shield tunneling machine 10, and the inference reference length end point C is the position when the shield tunneling machine 10 moves straight forward the inference reference length M. The inference reference line B and the inference reference length M are used as references for identifying the travel path of the shield tunneling machine 10 and its position after traveling for a predetermined time. Multiple inference reference lengths M of different lengths may be used as references.

[0069] The training data for the AI model for estimating recommended force point positions will be described with reference to Figure 5. Figure 5 shows an example of a training dataset used for training the AI model for estimating recommended force point positions. The training dataset is created using information measured when a shield tunneling machine 10 (or a shield tunneling machine other than the shield tunneling machine 10 used for estimation) actually excavated at a past shield construction site. A trained model for estimating recommended force point positions is created by training the AI model with the training dataset of Figure 5.

[0070] The learning data shown in FIG. 5 includes the cross-sectional diameter of the shield machine, ring No., inference reference length, inference reference length start position, inference reference length end position, explanatory variable data, and objective variable data. The cross-sectional diameter of the shield tunneling machine is information relating to the cross-sectional diameter of the shield tunneling machine 10, and may be information that classifies the cross-sectional diameter into small, medium, or large cross-sectional diameters. The ring number is information for identifying the ring, and is assigned an ascending number in the excavation direction. The ring length is, for example, "1000 mm."

[0071] The inference reference length is a distance reference set on the inference reference line B, and in this embodiment, it is assumed that inference reference lengths M of "100 mm" and "500 mm" are used as the reference. The inference reference length start point position is the coordinate of the start point of the inference reference length M (that is, the coordinate of the position reference point of the shield machine 10 at the time the position is measured). The inference reference length end point position is the coordinate of the end point of the inference reference length M (that is, the coordinate of a position that is the inference reference length M forward of the position reference point of the shield machine 10 at the time the position is measured).

[0072] The explanatory variable data is input data for the AI model, and includes at least one or more pieces of control data excluding the force point position, and information regarding the coordinates to be reached after moving the inference reference length M from the current position.

[0073] In this embodiment, it is assumed that common data is used as one or more pieces of control data excluding the point of effort position. The common data is data that is common to the explanatory variable data of the learning data for the AI model for estimating the arrival coordinates, which will be described later. Examples of common data include net stroke, front barrel direction (gyro), front barrel pitching, front barrel rolling, left-right earth pressure difference, left-right earth pressure average, copy cutter stroke, top-bottom bend angle, left-right bend angle, shield jack pressure, total thrust, remaining stroke to the end of the ring (i.e., remaining excavation length until excavation of one ring of segment is completed), etc. The common data may also include the cross-sectional diameter of the shield machine and the ring number.

[0074] In addition, in this embodiment, it is assumed that the reach deviation amount h after inference reference length excavation (see Figure 4) is used as information regarding the reach coordinate after advancing the inference reference length M from the current position. As shown in Figure 4, the reach deviation amount h is the deviation amount between the reference length reach position E, which is the actual position after advancing the inference reference length M from the current position, and the inference reference length end point C. The inference reference length is, for example, "100 mm" or "500 mm".

[0075] The objective variable data is data on the output side of the AI model, and includes at least information regarding the force point positions that were actually controlled while the shield tunneling machine 10 moved the inference reference length M from its current position. The force point positions used in the objective variable data may include a horizontal force point position that indicates the position in the horizontal direction, and a vertical force point position that indicates the position in the vertical direction. In this embodiment, a case will be described in which normalized and weighted averaged force point positions are used as objective variable data.

[0076] Data that depends on the size of the shield machine 10 (for example, the force point position) should be converted into an equivalent evaluation form (normalized) by dividing it by the radius of the shield machine 10. An image of normalizing the force point position is shown in Figure 6. For example, if the force point position of a shield machine 10 with a diameter of 4 m is 500 mm to the right, then the normalized force point position is obtained by dividing it by the radius r of 2 m, or +0.5 m / 2 m = +0.25. Here, the center position is represented as 0 (zero), and the right side of the left and right sides are represented by + and -, respectively, and the upper and lower sides are represented by + and -.

[0077] Alternatively, the force point position may be calculated as a weighted average of multiple force point positions from the start of inference reference length excavation until the end of excavation. In this case, the weighted average force point position may be further divided by the radius r of the shield tunneling machine 10 for normalization. The weighted averaging process for force point positions will be explained with reference to Figure 7. Because it is considered that the force point position in the initial section of the inference reference length M has a large influence on the final deviation amount, the force point position is multiplied by the proportion of the remaining section (for example, "0.9" at the point where excavation has progressed "10 mm" since the start of the section) as a weight, and the average value of those values within the section is calculated.

[0078] If the history of the force point position is as shown in the table in Figure 7, the calculation is as follows: The calculated average value is multiplied by "2" to match the scale with the actual force point and obtain the value after weighted average processing. "(0.1 × 1.0 + 0.3 × 0.8 + 0.5 × 0.6 + 0.6 × 0.4 + 0.7 × 0.2) / 5 × 2 = 0.408" In addition, when averaging the force point positions over the inference reference length M, a simple averaging process may be used.

[0079] The trained model recording unit 32 for estimating reached coordinates is a unit that stores trained models for estimating reached coordinates. The trained model for estimating reached coordinates is an AI model that has learned the correspondence between the reached deviation amount h after inference reference length excavation, which is the objective variable, and its explanatory variables. The trained model recording unit 32 for estimating reached coordinates may store multiple trained models for estimating reached coordinates. The trained model recording unit 32 for estimating reached coordinates may store multiple trained models according to, for example, the inference reference length. Furthermore, the trained model recording unit 32 for estimating reached coordinates may store multiple trained models according to different cross-sectional diameters (small cross-section - medium cross-section - large cross-section).

[0080] The training data for the AI model for estimating predicted arrival coordinates will be described with reference to Figure 8. Figure 8 shows an example of a training dataset used to train the AI model for estimating predicted arrival coordinates. The training dataset is created using information measured when a shield tunneling machine 10 (or a shield tunneling machine other than the shield tunneling machine 10 used for estimation) actually excavated at a past shield construction site. By training the AI model with the training dataset of Figure 8, a trained model for estimating arrival coordinates is created.

[0081] The learning data shown in Figure 8 includes the cross-sectional diameter of the shield machine, ring number, inference reference length, inference reference length start position, inference reference length end position, explanatory variable data, and target variable data. The cross-sectional diameter of the shield machine, ring number, inference reference length, inference reference length start position, and inference reference length end position are the same as those in the learning data for the AI model for estimating the recommended force point position (see Figure 5), so a description thereof will be omitted.

[0082] The explanatory variable data is input data for the AI model, and includes at least a plurality of control data including the force point positions that were actually controlled while the shield tunneling machine 10 advanced the inference reference length M.

[0083] Information about the force point positions actually controlled while the shield tunneling machine 10 travels the inference reference length M may be the same as the objective variable data of the learning data (see FIG. 5) for the AI model used to estimate the recommended force point position described above. The force point positions used in the explanatory variable data may include a horizontal force point position indicating the position in the horizontal direction, and a vertical force point position indicating the position in the vertical direction. Normalized and weighted average force point positions may also be used as explanatory variable data.

[0084] Control data other than the force point position used as an explanatory variable may be, for example, common data. The common data is data that is common to the explanatory variable data of the learning data for the AI model for estimating the recommended force point position described above. Examples of common data include net stroke, front barrel direction (gyro), front barrel pitching, front barrel rolling, left-right earth pressure difference, left-right earth pressure average, copy cutter stroke, up-down center bending angle, left-right center bending angle, shield jack pressure, total thrust, and remaining stroke to the end of the ring. The common data may also include the cross-sectional diameter of the shield tunneling machine and the ring number.

[0085] The objective variable data is data on the output side of the AI model, and includes at least information regarding the coordinate to be reached after advancing the inference reference length M from the current position. The information regarding the coordinate to be reached used in the objective variable data may be the same as the explanatory variable data of the learning data (see Figure 5) of the AI model for estimating the recommended force point position described above. In this embodiment, it is assumed that the amount of deviation h reached after excavating the inference reference length (see Figure 4) is used as information regarding the coordinate to be reached after advancing the inference reference length M from the current position. As shown in Figure 4, the amount of deviation h reached is the amount of deviation between the reference length reached position E, which is the actual position after advancing the inference reference length M from the current position, and the inference reference length end point C. The inference reference length is, for example, "100 mm" or "500 mm".

[0086] The learning of the AI model will be explained with reference to Figure 9 (and also with reference to Figures 1 to 8 as appropriate). Figure 9 is an image diagram of learning that is performed using information measured when a shield tunneling machine 10 actually excavated at a past shield construction site. For the sake of convenience, only movement in a two-dimensional plane (front-back and left-right) will be considered here, and movement in the up-down direction will not be taken into account.

[0087] Assume that the current position of the shield machine 10 is the "start position P1s of the first ring." In this case, the amount of deviation h reached after the inference reference long excavation (for example, after excavation of "100 mm") is the amount of deviation h reached (P1s + 0 to 100 mm) at "position P1s at excavation of 100 mm + 100 mm" relative to the inference reference straight line B (P1s) when the shield machine 10 is at the "start position P1s of the first ring." Furthermore, the amount of deviation h reached after the inference reference long excavation (for example, after excavation of "500 mm") is the amount of deviation h reached (P1s + 0 to 500 mm) at "position P1s at excavation of 500 mm + 500 mm" relative to the inference reference straight line B (P1s) when the shield machine 10 is at the start position P1s of the first ring.

[0088] Assume that the current position of shield machine 10 is "position P1s + 500 mm when excavating 500 mm." In this case, the amount of deviation h reached after the inference reference long excavation (for example, after excavating "100 mm") is the amount of deviation h reached at "position P1s + 600 mm when excavating 600 mm" (P1s + 500 to 600 mm) relative to inference reference line B (P1s + 500 mm) when shield machine 10 is at "position P1s + 500 mm when excavating 500 mm." Furthermore, the achieved deviation amount h after the inference reference long excavation (for example, after excavation of "500 mm") is the achieved deviation amount h (P1s + 500 to 1000 mm) at "position P1s + 1000 mm when excavating 1000 mm" relative to the inference reference straight line B (P1s + 500 mm) when the shield tunneling machine 10 is at "position P1s + 500 mm when excavating 500 mm".

[0089] The estimation device data acquisition unit 41 shown in Figure 1 is a part that receives and stores the excavation data required for the estimation device 30 from the excavation management device 20, and transfers it to the estimation data preprocessing unit 42. It is also a part that transfers data to the learning data selection unit 51 for additional learning after one ring of excavation.

[0090] The data acquired by the estimation device data acquisition unit 41 during shield tunneling includes ring number, net stroke, shield tunneling machine tip coordinates (X, Y, Z), shield tunneling machine center bend coordinates (X, Y, Z), shield tunneling machine rear end coordinates (X, Y, Z), upper shield jack stroke, right shield jack stroke, lower shield jack stroke, left shield jack stroke, upper and lower stroke difference, left and right stroke difference, upper shield jack speed, right shield jack speed, lower shield jack speed, left shield jack speed, front body direction (gyro), front body pitching, front body rolling, rear body direction. (Gyro), rear body pitching, rear body rolling, left face earth pressure, right face earth pressure, copy cutter stroke, up and down center bending angle, left and right center bending angle, cutter torque, cutter right rotation, cutter left rotation, cutter rotation speed, screw rotation speed, gate opening, upper tail clearance, right tail clearance, lower tail clearance, left tail clearance, shield jack pressure, total thrust, horizontal force point position, vertical force point position, horizontal moment, vertical moment, horizontal deviation of the tip of the shield tunneling machine, vertical deviation of the shield tunneling machine, segment length, target coordinates reached by the ring (X, Y, Z), radius of curvature, etc.

[0091] The estimation data pre-processing unit 42 shown in Figure 1 is a part that performs calculations for data items that require pre-processing among the excavation data transferred from the estimation device data acquisition unit 41, to convert them into values suitable for estimating the recommended force point position and predicted ring arrival coordinates.

[0092] Data that requires preprocessing include, for example, the shield tunneling machine tip coordinates (X, Y, Z), the shield tunneling machine center bend coordinates (X, Y, Z), the left face earth pressure, the right face earth pressure, the horizontal position of the force point, the vertical position of the force point, the segment length, and the ring target coordinates (X, Y, Z).

[0093] The data output after preprocessing is the remaining stroke to the end of the ring, the difference between the left and right earth pressures, the average left and right earth pressures, the normalized horizontal and vertical positions of the force point, and the planned horizontal and vertical deviation amounts at the inferred reference length. The remaining stroke to the end of the ring (i.e., the remaining excavation length until excavation of one segment ring is completed) is calculated from the difference between the segment length and the net stroke (the excavation distance of the shield machine 10 from the start of excavation for the next segment ring after assembly of the segment ring is completed).

[0094] Taking into consideration the influence that the left and right earth pressure values have on the shield machine 10, the average left and right earth pressures are calculated as the force applied to the shield machine 10 in the direction of travel, and the left and right earth pressure difference applied to the shield machine 10 as a couple (left and right direction). Here, the left and right earth pressure difference is calculated as "left side earth pressure value - right side earth pressure value", where + means that the left side earth pressure is greater and - means that the right side earth pressure is greater.

[0095] Data that depends on the size of the shield machine 10, i.e., the force point position, should be converted into an equivalent evaluation form (normalized) by dividing it by the radius r of the shield machine 10. For example, the normalized horizontal and vertical force point positions are calculated by dividing the horizontal and vertical force point positions by the radius r of the shield machine 10, respectively.

[0096] 1 selects data items (explanatory variables) necessary for prediction by each estimation unit (recommended force point position estimation unit 44 and predicted ring arrival coordinate estimation unit 46). Specifically, it selects data that is considered to be common and useful for estimating the recommended force point position and predicted ring arrival coordinate from among the preprocessed data sent from the estimation data preprocessing unit 42 and other data.

[0097] The common data required for each estimation section include the cross-sectional diameter of the shield tunneling machine, ring number, net stroke, front barrel direction (gyro), front barrel pitching, front barrel rolling, left and right earth pressure difference, left and right earth pressure average, copy cutter stroke, upper and lower center bend angle, left and right center bend angle, shield jack pressure, total thrust, remaining stroke to the end of the ring, etc.

[0098] These common data items and data items that have a large impact specific to each prediction are input together to each estimation unit. That is, "data common to each estimation unit + normalized horizontal and vertical force point positions" are input to the ring arrival predicted coordinate estimation unit 46. Also, "data common to each estimation unit + planned horizontal and vertical deviation amounts at inference reference length M" is input to the recommended force point position estimation unit 44.

[0099] The recommended force point position estimating unit 44 shown in Fig. 1 is a unit that estimates a recommended force point position that is recommended for progress up to the ring's target coordinates. The recommended force point position estimating unit 44 estimates a recommended force point position, for example, at predetermined time intervals or for each predetermined excavation length during excavation. The recommended force point position estimating unit 44 inputs explanatory variable data into a trained model for estimating recommended force point positions, thereby outputting a recommended force point position that serves as the objective variable. The recommended force point position that serves as the objective variable may be normalized.

[0100] Here, to obtain a highly accurate force point recommended position, it is advisable to select one of the on-site AI models that most closely resembles the cross-sectional size of the shield tunneling machine 10 currently in operation. Alternatively, it is advisable to select one of the on-site AI models that most closely resembles the tendency of the force point position relative to the amount of deflection. Also, it is advisable to determine whether to select an inference reference length of "100 mm" for a short excavation length or an inference reference length of "500 mm" for a long excavation length based on the remaining stroke to the end of the ring.

[0101] The estimation data used for estimation using the trained model for estimating recommended force point positions will be described with reference to Fig. 10. Fig. 10 shows an example of an estimation dataset used for estimation using the trained model for estimating recommended force point positions. The estimation dataset is created using information measured when the shield tunneling machine 10 actually excavates at a shield construction site during construction.

[0102] 10 includes the current position, explanatory variable data, and response variable data. Note that the response variable data is an item that is output as a result of estimation. The current position is the position (coordinates) of the shield machine 10 at the present time. The explanatory variable data is input data for the AI model, and includes one or more control data excluding the force point position, as well as information about the virtual target coordinates after moving the inference reference length M from the current position.

[0103] In this embodiment, it is assumed that common data will be used as one or more pieces of control data excluding the point of effort position. The common data is data that is common to the explanatory variable data of the estimation data for estimating the arrival coordinates, which will be described later. Examples of the common data include the cross-sectional diameter of the shield machine, front barrel direction (gyro), front barrel pitching, front barrel rolling, left-right earth pressure difference, left-right earth pressure average, copy cutter stroke, up-down center bending angle, left-right center bending angle, shield jack pressure, total thrust, and remaining stroke to the end of the ring.

[0104] Furthermore, in this embodiment, it is assumed that the planned deviation amount d' after inference reference length excavation (see Figure 11) is used as information regarding the virtual target coordinate to be reached after traveling the inference reference length M from the current position. Figure 11 is a diagram for explaining the explanatory variable data of the estimation data used for estimation by the trained model for estimating the recommended force point position. The planned deviation amount d' shown in Figure 11 is the deviation amount between the reference length planned target coordinate qa and the inference reference length end point C after traveling the inference reference length M along a planned progression path (an example is a parabolic or circular function) geometrically determined from the current position.

[0105] The planned progression path is calculated, for example, by obtaining the planned deviation amount D' at the end of the ring (how far the shield machine 10 must reach the coordinates curved from its current straight direction) from the current position of the shield machine 10 (start point F of the inference reference length) and the ring's target coordinate Qa.The planned deviation amount d' after excavating the inference reference length is then calculated by multiplying the planned deviation amount D' by the square of (inference reference length M / remaining stroke L to the ring end).The inference reference length is, for example, "100 mm" or "500 mm".

[0106] For example, suppose one ring is 1000 mm long and the planned deviation D' at the end of the ring is 50 mm to the right. If, during 100 mm of excavation, there is 900 mm of excavation remaining and the planned deviation D' at the end of the ring is 45 mm to the right, then an inference base length of 500 mm is adopted. The square of 500 / 900 is 0.056, so +45 mm multiplied by 0.056 equals +2.5 mm, which is the planned deviation d' after 500 mm of excavation. For example, during 800 mm of excavation, there is 200 mm of excavation remaining and the planned deviation D' at the end of the ring is 5 mm to the right. Then, an inference base length of 100 mm is adopted for the short excavation length. The square of "100 / 200" is "0.25", and "+5mm" multiplied by "0.25" is "+1.25mm", which is the planned deviation amount d' after excavating the inferred reference length of "100mm". Here, deviation to the right from the current direction is expressed as +, and deviation to the left as -.

[0107] The objective variable data is obtained by a trained model for estimating recommended force point positions. The objective variable data is output data of the AI model, and in this embodiment, is composed of recommended force point positions from the current position to the reference length planning target coordinates at the inference reference length destination. The recommended force point positions used in the objective variable data may include a recommended horizontal force point position indicating the position in the horizontal direction, and a recommended vertical force point position indicating the position in the vertical direction. The recommended force point positions, which are objective variable data, are preferably normalized values.

[0108] The recommended force point position output unit 45 shown in Fig. 1 is a part that converts the normalized recommended force point position into a recommended force point position for use in the excavation management device 20 by multiplying it by the radius r of the shield tunneling machine 10, and outputs the converted recommended force point position. The input data are, for example, the cross-sectional diameter of the shield tunneling machine, the ring number, the net stroke, and the normalized recommended force point position. The output data are the ring number, the net stroke, and the converted recommended force point position.

[0109] The predicted ring arrival coordinate estimating unit 46 shown in FIG. 1 is a unit that estimates predicted ring arrival coordinates that are predicted to be reached at the end of the ring. The predicted ring arrival coordinate estimating unit 46 estimates predicted ring arrival coordinates, for example, at predetermined time intervals or for each predetermined excavation length during excavation. Note that the predicted ring arrival coordinate estimating unit 46 may estimate a predicted arrival deviation amount from the inference reference line B to the predicted ring arrival coordinates instead of the predicted ring arrival coordinates. The predicted ring arrival coordinate estimating unit 46 has a reference length predicted arrival coordinate estimating unit 47 and a predicted ring arrival coordinate output unit 48.

[0110] The reference length arrival predicted coordinate estimating unit 47 shown in Figure 1 is a part that estimates the reference length arrival predicted coordinate that is predicted to be reached after excavation of the inferred reference length. Note that the reference length arrival predicted coordinate estimating unit 47 may estimate the amount of predicted arrival deviation from the inferred reference line B to the reference length arrival predicted coordinate instead of the reference length arrival predicted coordinate. The reference length arrival predicted coordinate estimating unit 47 inputs explanatory variable data into a trained model for estimating predicted arrival coordinate, and outputs the predicted reference length arrival coordinate after excavation of the inferred reference length or the predicted arrival deviation amount to the reference length arrival predicted coordinate, which is the objective variable.

[0111] Here, to obtain highly accurate predicted arrival coordinates, it is advisable to select one of the on-site AI models that most closely resembles the cross-sectional size of the shield tunneling machine 10 currently in operation. Alternatively, it is advisable to select one of the on-site AI models that most closely resembles the tendency of the amount of deviation relative to the force point position. Also, it is advisable to determine whether to select an inference reference length of "100 mm" for a short excavation length or an inference reference length of "500 mm" for a long excavation length based on the remaining stroke to the end of the ring.

[0112] The estimation data used for estimation using the trained model for estimating predicted arrival coordinates will be described with reference to Fig. 12. Fig. 12 shows an example of an estimation dataset used for estimation using the trained model for estimating predicted arrival coordinates. The estimation dataset is created using information measured when the shield tunneling machine 10 actually excavates at a shield construction site under construction.

[0113] 12 includes the current position, explanatory variable data, and response variable data. Note that the response variable data is an item that is output as a result of estimation. The current position is the position (coordinates) of the shield machine 10 at the present time. The explanatory variable data is input data for the AI model and includes multiple control data including the force point position.

[0114] In this embodiment, it is assumed that the explanatory variable data uses the force point position and common data. The force point position used in the explanatory variable data is the force point position at which control was actually performed at the time of inference. This force point position is preferably a normalized value.

[0115] The common data is data that is common to the explanatory variable data of the estimation data for estimating the recommended force point position described above. Examples of the common data include the cross-sectional diameter of the shield machine, front barrel direction (gyro), front barrel pitching, front barrel rolling, left-right earth pressure difference, left-right earth pressure average, copy cutter stroke, up-down center bending angle, left-right center bending angle, shield jack pressure, total thrust, and remaining stroke to the end of the ring.

[0116] The objective variable data is obtained by a trained model for estimating the recommended force point position. The objective variable data is data on the output side of the AI model, and in this embodiment is composed of the predicted arrival deviation amount d (see FIG. 13) after inference reference long excavation. The predicted arrival deviation amount d (see FIG. 13), which is the objective variable data, is the deviation amount after inference reference long excavation when it is assumed that the state including the current force point position will continue. FIG. 13 is a diagram for explaining the process of estimating the predicted arrival coordinate. The predicted reference length arrival coordinate estimation unit 47 may output the predicted reference length arrival coordinate qb (see FIG. 13) instead of the predicted arrival deviation amount d.

[0117] The ring arrival prediction coordinate output unit 48 shown in Figure 1 obtains the arrival prediction deviation amount D (see Figure 13) at the ring end by multiplying the arrival prediction deviation amount d after excavation of the inference reference length by the square of (remaining stroke L to the ring end / inference reference length M). The inference reference length is, for example, "100 mm" or "500 mm." Here, a parabola is assumed to derive the arrival prediction deviation amount D at the ring end, but a circular function may also be used.

[0118] Furthermore, the predicted ring arrival coordinate output unit 48 calculates the orientation and pitching of the shield tunneling machine 10 from the tip coordinates (X, Y, Z) and center bend coordinates (X, Y, Z) of the shield tunneling machine 10. Furthermore, the predicted ring arrival coordinate Qb (see Figure 13) at the ring end is converted into a predicted ring arrival coordinate at the ring end to be used by the excavation management device 20 from the predicted arrival deviation amount D at the ring end, the current orientation and pitching of the shield tunneling machine 10, and the remaining stroke to the ring end.

[0119] The data input to the predicted ring arrival coordinate output unit 48 is, for example, the cross-sectional diameter of the shield tunneling machine, the ring number, the net stroke, the shield tunneling machine tip coordinates (X, Y, Z), the shield tunneling machine center bend coordinates (X, Y, Z), and the predicted arrival deviation amount d after inference reference long excavation. The data output is, for example, the ring number, the net stroke, and the predicted ring arrival coordinate Qb(X, Y, Z) at the ring end.

[0120] The progress of the shield machine 10 will be described with reference to Figures 14 to 16 (and Figures 1 to 15 as appropriate). Figure 14 shows the shield machine 10 at the start position P1s of the first ring. Figure 15 shows the state when the shield machine 10 has excavated 20 mm from the start position P1s of the first ring. Figure 16 shows the state when the shield machine 10 has excavated 520 mm from the start position P1s of the first ring.

[0121] 14, the estimation device 30 estimates a recommended force point position from the current position to a reference length planned target coordinate qa that is 500 mm ahead in the inference reference length, based on the planned deviation d' (P1s + 0 to 500 mm) after excavation of the inference reference length of 500 mm from the start position P1s of the first ring. The estimation device 30 also estimates a predicted ring arrival coordinate Qb of the first ring based on the force point position that was actually controlled at the start position P1s of the first ring.

[0122] 15, the estimation device 30 estimates the recommended force point position from the current position to the reference length planned target coordinate qa, an inference reference length 500 mm ahead, based on the planned deviation d' (P1s+20 to 520 mm) after excavation of an inference reference length of 500 mm from the position P1s+20 mm, an excavation advance of 20 mm. The estimation device 30 also estimates the predicted ring arrival coordinate Qb of the first ring based on the force point position where control was actually performed at the position P1s+20 mm, an excavation advance of 20 mm.

[0123] 16, the estimation device 30 estimates the recommended force point position from the current position to the reference length planned target coordinate qa, an inference reference length 100 mm ahead, based on the planned deviation d' (P1s + 520 to 620 mm) after excavation of the inference reference length 100 mm from the position P1s + 520 mm, an excavation advance of 520 mm. The estimation device 30 also estimates the predicted ring arrival coordinate Qb of the first ring based on the force point position where control was actually performed at the position P1s + 520 mm, an excavation advance of 520 mm.

[0124] Through the processing described so far, the estimation device 30 ultimately outputs data such as the ring number, net stroke, recommended force point position, and predicted ring arrival coordinate Qb(X, Y, Z) at the end of the ring to the estimation result acquisition unit 27 of the excavation management device 20.

[0125] The learning data selection unit 51 shown in Figure 1 is a part that selects the data items required for each learning unit (force point position learning unit 53 and arrival coordinate learning unit 54) to learn the AI model from the excavation data transferred from the estimation device data acquisition unit 41. After one ring of excavation is completed, the necessary data is transferred from the estimation device data acquisition unit 41 when it has been collected.

[0126] The data required for each learning unit includes, for example, the cross-sectional diameter of the shield machine, ring number, net stroke, shield machine tip coordinates (X, Y, Z), shield machine center bending coordinates (X, Y, Z), front barrel direction (gyro), front barrel pitching, front barrel rolling, left face earth pressure, right face earth pressure, copy cutter stroke, vertical center bending angle, left and right center bending angle, shield jack pressure, total thrust, horizontal force point position, vertical force point position, etc. This data is transferred from the excavation management device every 20 mm when one ring is excavated.

[0127] The learning data pre-processing unit 52 is a unit that performs arithmetic processing on the excavation data transferred from the learning data selection unit 51 to convert it into values suitable for machine learning. The excavation data transferred from the learning data selection unit 51 is excavation data for strokes of "20 mm", and since machine learning requires teacher data for each inference reference length, it performs arithmetic processing on each piece of data.

[0128] The data transferred to the learning data pre-processing unit 52 includes, for example, the cross-sectional diameter of the shield tunneling machine, ring number, net stroke, shield tunneling machine tip coordinates (X, Y, Z), shield tunneling machine bend coordinates (X, Y, Z), front barrel direction (gyro), front barrel pitching, front barrel rolling, left face earth pressure, right face earth pressure, copy cutter stroke, upper and lower bend angles, left and right bend angles, shield jack pressure, total thrust, horizontal force point position, vertical force point position, etc.

[0129] After preprocessing, the data output from the learning data preprocessing unit 52 includes, for example, the cross-sectional diameter of the shield tunneling machine, ring number, section name of the inferred reference length (100 mm and 500 mm), the achieved deviation amount at the inferred reference length, average front barrel direction (gyro), average front barrel pitching, average front barrel rolling, average left and right earth pressure difference, average left and right earth pressure averages, average copy cutter stroke, average upper and lower center bend angle, average left and right center bend angle, average shield jack pressure, average total thrust, weighted average and normalized horizontal and vertical force point positions at the inferred reference length, etc.

[0130] The section names for the inference reference lengths (100 mm and 500 mm) are determined from the net stroke from the start of excavation at the inference reference length to the end of excavation. The average values for the front body heading (gyro), front body pitching, front body rolling, copy cutter stroke, top-bottom bending angle, left-right bending angle, shield jack pressure, and total thrust are calculated by adding the data for every 20 mm and dividing by the number of data points.

[0131] The average value of the left-right earth pressure difference and the average value of the left-right earth pressure average are calculated by determining the left-right earth pressure difference and the average left-right earth pressure from the left face earth pressure and right face earth pressure within the data for every 20 mm, and then calculating the average value in the same way as above.

[0132] The force point position learning unit 53 shown in Figure 1 is a part that performs additional learning on the trained model for estimating recommended force point positions about the correspondence between the selected explanatory variables and the objective variable. By performing additional learning, it becomes possible to estimate recommended force point positions that are more in line with the characteristics of the shield tunneling machine at the site. The additional learning data used in the additional learning may be the same as the previously described learning data (see Figure 5) of the trained model for estimating recommended force point positions.

[0133] Among the data transferred from the learning data preprocessing unit 52, explanatory variables can include, for example, the cross-sectional diameter of the shield machine, ring number, section name of the inference reference length, the amount of horizontal deviation and the amount of vertical deviation achieved at the inference reference length, the average front barrel direction (gyro), the average front barrel pitching, the average front barrel rolling, the average left-right earth pressure difference, the average left-right earth pressure averages, the average copy cutter stroke, the average up-down center bending angle, the average left-right center bending angle, the average shield jack pressure, the average total thrust, etc. Furthermore, the weighted average and normalized horizontal and vertical force point positions at the inference reference length can be used as response variables.

[0134] The arrival coordinate learning unit 54 shown in Figure 1 is a part that performs additional learning on the trained model for estimating arrival coordinates about the correspondence between the selected explanatory variables and the objective variable. By performing additional learning, it becomes possible to estimate predicted arrival coordinates that are more in line with the characteristics of the shield tunneling machine at the site. The additional learning data used in the additional learning may be the same as the above-mentioned learning data (see Figure 8) of the trained model for estimating predicted arrival coordinates.

[0135] Among the data transferred from the learning data preprocessing unit 52, explanatory variables can include, for example, the cross-sectional diameter of the shield machine, ring number, section name of the inference reference length, weighted average and normalized horizontal and vertical force point positions over the inference reference length, average front barrel direction (gyro), average front barrel pitching, average front barrel rolling, average left-right earth pressure difference, average left-right earth pressure averages, average copy cutter stroke, average top-bottom bend angle, average left-right bend angle, average shield jack pressure, average total thrust, etc. Furthermore, the target variables can be the amount of horizontal deviation and the amount of vertical deviation achieved over the inference reference length.

[0136] Setting the level of the autonomous driving system will be described with reference to Fig. 17. Fig. 17 is an example of a flowchart for setting the level of the autonomous driving system. For example, it is determined whether a shield jack hydraulic control valve is equipped (step S1), and if not, it is set to level 1, and the shield automatic driving system 1 is operated as a driving assistance system that recommends jack selection.

[0137] If the shield jack hydraulic control valve is equipped, it is determined whether or not automatic operation will be performed (step S2), and if automatic operation will not be performed, it is set to level 2, and the shield automatic operation system 1 is operated as an operation assistance system that recommends the force point position. On the other hand, when performing autonomous driving, the level is set to level 3, and the shield autonomous driving system 1 is operated as an autonomous driving system based on the recommended force point position.

[0138] The cycle of the estimation method in the shield road automatic driving system 1 will be described with reference to Fig. 18. Fig. 18 is an example of a flow diagram of the estimation method in the shield road automatic driving system 1. Before starting excavation, the current position and attitude of the shield machine 10 are measured using surveying equipment (step S11). The survey results are then used to perform necessary calculations (step S12). Next, the calculated values for the shield excavation plan and segment assembly plan are calculated (step S12), and an excavation instruction sheet is created (step S13).

[0139] During excavation, measurements are taken using various instruments on the shield machine 10 (step S21), and excavation management of the shield machine 10 is performed based on the measured information (step S22). The estimation device 30 also estimates the recommended force point position and predicted ring arrival coordinates at intervals of "0 mm" or "20 mm" for the excavation length, and outputs the results to the excavation management device 20 (steps S23, S24).

[0140] The excavation management device 20 checks whether the recommended force point position or predicted arrival coordinates have been updated (step S25), and if so, determines whether a command to control the hydraulic pressure of the jack is necessary (step S26). If a command to control the hydraulic pressure of the jack is not necessary, the device returns to step S25 and continues to monitor for updates to the recommended force point position or predicted arrival coordinates. On the other hand, if a command to control the hydraulic pressure of the jack is necessary, the device carries out hydraulic control of the jack (step S27).

[0141] As described above, the estimation device 30 and estimation method according to this embodiment estimate the recommended force point position using a trained model for estimating the recommended force point position, and then set the force point position of the shield tunneling machine 10 manually (for levels 1 and 2) or automatically (for level 3) based on the estimated recommended force point position. The shield tunneling machine 10 controls the shield jack based on the set force point position and excavates.

[0142] Furthermore, the trained model for estimating the target coordinates is used to predict the target coordinates that will be reached by controlling the actual force point position that has been set based on the recommended force point position, and the predicted target coordinates are displayed.The display of the target coordinates is then checked, and if the force point position needs to be corrected, the force point position is reset.

[0143] In this way, the estimation device 30 and estimation method according to this embodiment estimate the recommended force point position, allowing the force point to be set quickly. Also, because the predicted arrival coordinates are estimated, it is possible to quickly determine the need to review the force point even if the direction with respect to the target deviates for some reason.

[0144] For example, depending on the structure of the shield tunneling machine 10 (for example, whether or not the shield jack is equipped with a hydraulic control valve), the position of the force point that can be set may be restricted, which may result in a discrepancy between the actual force point and the recommended force point (for example, in the case of level 1). If the force point cannot be matched with the recommended force point, it becomes difficult to make the shield tunneling machine 10 reach the target coordinates. Even in such cases, it is possible to make the shield tunneling machine 10 reach the initial target coordinates with as little error as possible.

[0145] Furthermore, due to the quirks of the shield tunneling machine 10 (differences in soil quality, manufacturing precision, size, shape, center of gravity position, etc.), it will not advance in exactly the same direction even when advanced with the same force point. Therefore, even when control is performed using a learning model trained on training data from past shield construction sites, there will be a discrepancy between the target coordinates and the coordinates actually reached. Even in such cases, it is possible to reach the initial target coordinates with as little error as possible.

[0146] Furthermore, if a field worker makes a calculation or input error and the operator excavates based on those instruction values, problems may occur, such as deviation from the design alignment, meandering beyond the allowable value, or the inability to assemble segments. In this embodiment, by displaying the target coordinates to be reached, if a mistake such as the one described above is made, it becomes clear that the position has deviated from the design alignment, and the excavation instruction values can be rechecked or corrected before excavation begins (checking is possible by visualizing them).

[0147] Furthermore, if there is no tail clearance between the tail and the segments being assembled within the tail section of the shield tunneling machine 10, it is impossible to assemble the segments of a new ring. Furthermore, if there is no tail clearance, problems such as damage can occur due to contact with the assembled segments at the rear end of the tail. Therefore, directional control is sometimes changed to ensure tail clearance, but in the past, it was not possible to confirm the extent to which the shield tunneling machine 10 would deviate from the alignment. However, in this embodiment, if an excavation instruction manual is created that takes tail clearance into account, it is possible to confirm in advance how much the predicted coordinates will deviate from the design alignment. Furthermore, even if the clearance becomes even smaller during excavation and it becomes necessary to further deviate from the design alignment, it is possible to excavate while checking the predicted coordinates when the recommended force point position is shifted, thereby enabling alignment management that prevents significant meandering from the alignment.

[0148] Furthermore, with the automatic shield driving system 1 according to this embodiment, data on the excavation site can be acquired for each inference reference length M, allowing for the acquisition of more data and for learning to be performed using that data. Furthermore, additional learning can be performed to build a learning model that reflects the characteristics of the site at an earlier stage.

[0149] Although the embodiment of the present invention has been described above, the present invention is not limited to this and can be practiced within the scope of the claims.

[0150] For example, in the embodiment, the inference reference length M is set to "100 mm" for a short excavation length and "500 mm" for a long excavation length. However, for example, the inference reference length M may be set in increments of "20 mm" between "100 mm and 500 mm" (see FIG. 19). FIG. 19 is a diagram for explaining variations in the inference reference length M, and is an image diagram of learning performed using information measured when the shield tunneling machine 10 actually excavated at a past shield construction site. This results in 21 types of inference reference length M. When additional learning is performed on a trained model, this inference reference length M is used to create an AI model and is used as an explanatory variable and a target variable. Furthermore, this inference reference length itself is used as an explanatory variable.

[0151] For example, when one ring is excavated 800mm, if there is 200mm of excavation remaining and the planned deviation at the end of the ring is 5mm to the right, the inference reference length M will remain 200mm, and the planned deviation at the end of the ring = planned deviation after excavating the inference reference length (because the square of the inference reference length / remaining stroke to the end of the ring = 1). Therefore, the planned deviation after excavating 200mm of the inference reference length is +5mm. Here, deviation to the right from the current direction is expressed as +, and deviation to the left as -. [Explanation of symbols]

[0152] 1. Shield Autonomous Driving System 10 Shield tunneling machine 20 Excavation control device 21 Various data storage unit 22 System display and operation section 23 Excavation instruction data acquisition unit 24 Measurement data acquisition unit 25 Processing unit 26 Shield tunneling machine control unit 27 Estimation result acquisition part 28 Data transfer unit for estimation device 30 Estimation device 31 Recording unit of trained model for estimating recommended force point position 32 Recording unit of trained model for estimating arrival coordinates 41 Data acquisition unit for estimation device 42 Estimation data preprocessing section 43 Estimation data selection unit 44 Recommended force point position estimation unit 45 Recommended force point position output section 46 Ring arrival prediction coordinate estimation unit 47 Reference length arrival prediction coordinate estimation unit 48 Ring arrival prediction coordinate output unit 51 Learning data selection unit 52 Learning data preprocessing section 53 Force position learning section 54 Arrival coordinate learning section 60 System Screen

Claims

1. An estimation device that estimates a force point position recommended for the advancement of a shield tunneling machine, a data acquisition unit for an estimation device that acquires a plurality of control data for the shield tunneling machine and ring destination coordinates that are destinations at the end of the ring; a recommended force point position estimation unit that estimates a recommended force point position that is recommended in the progression to the ring reach target coordinates, the recommended force point position estimation unit estimates the recommended force point position at predetermined time intervals or for each predetermined excavation length during excavation. An estimation device characterized by:

2. a ring arrival predicted coordinate estimating unit that estimates a ring arrival predicted coordinate that is predicted to be reached at the end of the ring, or that estimates an amount of predicted arrival deviation from an inference reference line that is set based on the direction of the shield machine at its current position to the ring arrival predicted coordinate, the ring arrival predicted coordinate estimating unit estimates the ring arrival predicted coordinate or estimates the arrival predicted deviation amount from the inference reference straight line to the ring arrival predicted coordinate at predetermined time intervals or for each predetermined excavation length during excavation; 2. The estimation device according to claim 1 .

3. an inference reference length indicating a predetermined distance from the shield machine on the inference reference straight line is used as a reference; the recommended force point position estimation unit outputs the recommended force point position by inputting at least one or more pieces of control data excluding the force point position and a planned deviation amount between a reference length planned target coordinate and an inference reference length end point after moving along a planned progress path geometrically determined from the current position by an inference reference length into a learning model for estimating the recommended force point position, The learning model for estimating a recommended force point position is a model created based on data acquired during the past progress of the shield tunneling machine, and is created by learning using learning data in which one or more control data excluding the force point position and the amount of reach deviation between the reference length reach coordinate, which is the actual position after moving the inference reference length from the current position, and the inference reference length end point are used as explanatory variable data, and the force point position actually controlled while the shield tunneling machine moves the inference reference length from the current position is used as objective variable data, The ring arrival predicted coordinate estimating unit a reference length predicted arrival coordinate estimating unit that outputs a predicted arrival deviation amount between a reference length predicted arrival coordinate at which the shield machine is predicted to arrive after traveling the inference reference length and the inference reference length end point by inputting at least a plurality of control data including the force point position into a learning model for estimating arrival coordinates; a ring arrival predicted coordinate output unit that determines the ring arrival predicted coordinate based on the arrival predicted deviation amount, or that determines the arrival predicted deviation amount from the inference reference line to the ring arrival predicted coordinate, The learning model for estimating arrival coordinates is a model created based on data acquired during the past progress of the shield tunneling machine, and is created by learning using learning data in which a plurality of control data, including the force point positions actually controlled while the shield tunneling machine moved the inference reference length from its current position, is used as explanatory variable data, and the amount of arrival deviation between the reference length arrival coordinates, which is the actual position after moving the inference reference length from the current position, and the inference reference length end point is used as objective variable data.

3. The estimation device according to claim 2.

4. A method for estimating a force point position recommended for the advancement of a shield machine, comprising: a data acquisition process for an estimation device for acquiring a plurality of control data for the shield tunneling machine and ring destination coordinates that are destinations at the end of the ring; a recommended force point position estimating step of estimating a recommended force point position recommended in the progression to the ring reach target coordinates, the step of estimating the recommended force point position is executed at predetermined time intervals or for each predetermined excavation length during excavation. An estimation method characterized by:

Images