Anomaly detection device and method for tunnel boring machines

Abstract

To detect signs of abnormalities in tunnel boring machines early and with high accuracy. [Solution] The abnormality diagnosis device 20 for the tunnel boring machine 1 includes an analysis unit 21 that analyzes fluctuations in the correlation between multiple types of measurement items MI_k related to the tunnel boring machine 1, and a diagnosis unit 22 that diagnoses abnormalities in the tunnel boring machine 1 based on the results of the analysis.

Description

Technical Field

[0001] The present invention relates to an abnormality diagnosis device and an abnormality diagnosis method for a tunnel boring machine.

Background Art

[0002] Generally, a tunnel boring machine excavates a tunnel by rotating a cutter head, and a plurality of cutter bits attached to the front surface of the cutter head excavate the ground ahead to form a face. The cutter head is attached to the front end of a cylindrical boring machine body, and the tunnel is excavated as the boring machine body is propelled forward.

[0003] In tunnel excavation work using a tunnel boring machine, it is necessary to detect abnormalities of the tunnel boring machine and take measures against the abnormalities when they occur. For example, a failure of a cutter drive unit (for example, a cutter rotation motor) that rotationally drives the cutter head has a great impact on the progress of the work. Therefore, for example, Patent Document 1 discloses a technique for determining that an abnormality has occurred when the torque of a motor that rotationally drives a cutter head exceeds a threshold value.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, in the prior art, it has been difficult to grasp the signs of abnormalities of a tunnel boring machine at an early stage. Also, it has been desired to grasp the abnormalities of a tunnel boring machine more accurately. Thus, new proposals regarding the diagnosis of abnormalities of a tunnel boring machine are in demand.

[0006] Therefore, in view of these problems, the present invention aims to provide a tunnel boring machine abnormality diagnosis device and abnormality diagnosis method that can detect signs of abnormality in the tunnel boring machine early and with high accuracy. [Means for solving the problem]

[0007] To solve the above problems, the tunnel boring machine abnormality diagnosis device of the present invention comprises an analysis unit that analyzes fluctuations in the correlation between multiple types of measurement items related to the tunnel boring machine, and a diagnosis unit that diagnoses abnormalities in the tunnel boring machine based on the results of the analysis.

[0008] The analysis unit may pre-train a model of standard values ​​for multiple types of measurement items of a tunnel boring machine, and in the analysis, analyze the fluctuations in correlation and the factors causing such fluctuations based on the model and the actual measured values ​​of the multiple types of measurement items currently in use.

[0009] The analysis department may perform the analysis using the Mahalanobis-Daguchi method.

[0010] The analysis unit may pre-train a model that predicts the predicted value of one of several types of measurement items using the measured values ​​of the other types of measurement items. In the analysis, it may use the model to predict the current predicted value of one type of measurement item, and then analyze the fluctuations in the correlation and the factors causing those fluctuations based on the difference between the current predicted value of one type of measurement item and the current measured value of that item.

[0011] The analysis department may perform the analysis using the multiple regression method.

[0012] The analysis department may use multiple regression methods in addition to the Mahalanobis-Daguchi method for its analysis.

[0013] The analysis unit may exclude abnormal data from the measurement data of actual values ​​of multiple types of measurement items during the operation of the tunnel boring machine, and then train a model using the measurement data after the exclusion of abnormal data.

[0014] The analysis department may use an isolation forest to remove abnormal data from the measurement data.

[0015] The analysis unit may pre-train multiple models and perform the analysis using one model selected from among them.

[0016] It may also be equipped with a notification unit that notifies the results of the diagnosis.

[0017] To solve the above problems, the present invention provides a method for diagnosing abnormalities in a tunnel boring machine, which includes an analysis step of analyzing fluctuations in the correlation between multiple types of measurement items related to the tunnel boring machine, and a diagnosis step of diagnosing abnormalities in the tunnel boring machine based on the results of the analysis. [Effects of the Invention]

[0018] According to the present invention, it becomes possible to detect signs of abnormality in a tunnel boring machine early and with high accuracy. [Brief explanation of the drawing]

[0019] [Figure 1] This is a schematic cross-sectional view showing the overall configuration of a tunnel boring machine according to an embodiment of the present invention. [Figure 2] This is a block diagram showing an example of the functional configuration of an abnormality diagnosis device according to an embodiment of the present invention. [Figure 3] This figure schematically shows the correlation between multiple types of measurement items according to the embodiment of the present invention. [Figure 4] This flowchart shows an example of a general flow of the process related to the diagnosis of an abnormality performed by the abnormality diagnosis device according to an embodiment of the present invention. [Figure 5] This is a diagram illustrating the analysis using the Mahalanobis-Daguchi method. [Figure 6] This is a diagram for explaining the analysis using the regression method. [Figure 7] This is a flowchart showing an example of the general flow of processing related to the learning of a model performed by the abnormality diagnosis apparatus according to an embodiment of the present invention. [Figure 8] This is a diagram for explaining the cluster analysis.

Embodiments for Carrying Out the Invention

[0020] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. Dimensions, materials, and other specific numerical values shown in such embodiments are merely examples for facilitating the understanding of the invention, and do not limit the present invention unless otherwise specified. In the present specification and drawings, elements having substantially the same functions and configurations are denoted by the same reference numerals to omit redundant descriptions, and elements not directly related to the present invention are not shown.

[0021] First, the overall configuration of the tunnel boring machine 1 according to an embodiment of the present invention will be described. FIG. 1 is a cross-sectional schematic view showing the overall configuration of the tunnel boring machine 1. Note that the arrow A1 in FIG. 1 indicates the traveling direction of the tunnel boring machine 1. Hereinafter, the traveling direction of the tunnel boring machine 1 (the left direction in FIG. 1) will also be referred to as the forward direction, and the direction opposite to the traveling direction (the right direction in FIG. 1) will also be referred to as the backward direction.

[0022] The tunnel boring machine 1 is an earth pressure type (including the earth pressure balance type) shield boring machine capable of excavating the ground. As shown in FIG. 1, the tunnel boring machine 1 includes a boring machine main body 10. The boring machine main body 10 has a cylindrical shape (for example, a cylindrical shape or a rectangular cylindrical shape, etc.). The axial direction of the boring machine main body 10 coincides with the tunnel excavation direction. Hereinafter, the axial direction of the boring machine main body 10 will also be simply referred to as the axial direction, the radial direction of the boring machine main body 10 will also be simply referred to as the radial direction, and the circumferential direction of the boring machine main body 10 will also be simply referred to as the circumferential direction.

[0023] The excavator body 10 includes a skin plate 10a. The skin plate 10a is the part of the excavator body 10 that is in contact with the inner wall surface (pit wall) of the ground formed by excavation by the tunnel boring machine 1. The skin plate 10a is cylindrical (for example, cylindrical or rectangular) and forms the outer circumference of the excavator body 10.

[0024] A cutter head 11 is provided at the front end of the excavator body 10. The cutter head 11 is a roughly disc-shaped rotating body. The front end of the cutter central shaft 12 is fitted into the center of the cutter head 11, and the cutter head 11 is pivotally supported so as to be rotatable around the cutter central shaft 12.

[0025] The cutter head 11 includes an outer ring 11a, cutter spokes 11b, a fishtail cutter 11c, and cutter bits 11d. Of these, the outer ring 11a forms the outer circumference of the cutter head 11. The multiple cutter spokes 11b are arranged radially on the front surface of the cutter head 11, centered on the cutter central axis 12. The fishtail cutter 11c is mounted in the center of the front surface of the cutter head 11. Furthermore, numerous cutter bits 11d are mounted on the front surface of the cutter spokes 11b. The fishtail cutter 11c and cutter bits 11d may or may not be detachable.

[0026] Furthermore, the cutter head 11 has multiple openings formed between the outer ring 11a and the cutter spokes 11b. These openings function as intake ports for excavated soil generated when the cutter head 11 excavates the ground (face) and take it into the excavator body 10 (into the chamber 17, which will be described later).

[0027] A partition wall 13 is positioned behind the cutter head 11 in the excavator body 10. The partition wall 13 is a plate-shaped (for example, disc-shaped) wall positioned perpendicular to the axial direction (tunnel extension direction), and its outer edge is attached to the inner surface of the excavator body 10. The cutter head 11 and the partition wall 13 are positioned at a predetermined distance apart in the axial direction (tunnel extension direction). Various equipment of the tunnel excavator 1 is positioned behind the partition wall 13, and the partition wall 13 isolates this equipment from the excavated soil generated at the tunnel face. An outlet 13a, which is an opening for discharging excavated soil, is formed at the bottom of the partition wall 13.

[0028] A cutter central shaft 12 is rotatably supported at the center of the partition wall 13. Furthermore, an annular rotating ring 14 is rotatably supported on the partition wall 13 about the cutter central shaft 12. Multiple connecting beams 15 are provided at predetermined intervals in the circumferential direction at the front of the rotating ring 14. The multiple connecting beams 15 connect the cutter head 11 and the rotating ring 14. The front ends of the connecting beams 15 are connected to the cutter spokes 11b of the cutter head 11. On the other hand, a ring gear 14a is provided at the rear of the rotating ring 14. The ring gear 14a may be an external gear type or an internal gear type. Furthermore, a cutter rotation motor 16 is provided at the rear of the partition wall 13. The drive gear 16a of this cutter rotation motor 16 meshes with the ring gear 14a of the rotating ring 14.

[0029] By driving the cutter rotation motor 16, the rotation of its drive gear 16a is transmitted from the ring gear 14a to the rotating ring 14 and the connecting beam 15. This allows the cutter head 11 to rotate around the cutter central axis 12. As a result, the front surface of the rotating cutter head 11 can be pressed against the ground (work face) to excavate the ground. Thus, the cutter rotation motor 16 corresponds to the cutter drive unit that rotates the cutter head 11.

[0030] A chamber 17 is defined between the cutter head 11 and the partition wall 13. The chamber 17 is a space (for example, a roughly cylindrical space) partitioned by the rear surface of the cutter head 11, the front surface of the partition wall 13, and the inner circumferential surface of the excavator body 10. Excavated soil generated as a result of excavating the ground by the cutter head 11 is taken into the chamber 17 through the opening (excavated soil intake port) formed through the cutter head 11. The chamber 17 functions as a space (chamber) for temporarily storing the excavated soil. The excavated soil taken into the chamber 17 is discharged from the chamber 17 into the screw conveyor 18 through the discharge port 13a located at the bottom of the partition wall 13.

[0031] The screw conveyor 18 is installed on the rear side of the partition wall 13 within the excavator body 10. Within the excavator body 10, the screw conveyor 18 is positioned at an upward incline as it approaches the rear. The opening at the front end of the screw conveyor 18 is connected to the discharge port 13a of the partition wall 13. As a result, the internal space of the screw conveyor 18 communicates with the chamber 17 through the discharge port 13a of the partition wall 13. Inside the screw conveyor 18 is a screw-shaped rotating body with helical blades, called a screw blade 18a. By rotating the screw blade 18a, excavated soil stored in the chamber 17 can be taken into the screw conveyor 18, transported toward the rear of the excavator body 10, and discharged.

[0032] Furthermore, an erector device (not shown) is provided behind the bulkhead 13 of the excavator body 10. The erector device is provided so as to be movable in the axial, radial, and circumferential directions (i.e., the tunnel extension direction, the tunnel radial direction, and the tunnel circumferential direction) of the excavator body 10. The erector device is capable of gripping the segments S, which are lining members, and assembling the gripped segments S along the inner wall surface of the ground.

[0033] Segment S is a ring-shaped piece with a curved shape that conforms to the inner wall surface of the excavated ground. By driving the erector device, multiple segments S can be assembled in a ring shape along the circumferential direction. As a result, the tunnel is lined with multiple segments S, preventing the collapse of the inner wall surface of the ground.

[0034] Multiple shield jacks 19 are provided inside the excavator body 10, spaced apart from each other in the circumferential direction. Each shield jack 19 is provided to extend along the inner circumferential surface of the excavator body 10 in the direction of tunnel extension. The shield jacks 19 are, for example, hydraulic jacks, but other types of jacks, actuators, etc., may be used as long as they can generate thrust for the tunnel excavator 1. A retractable drive rod 19a is provided at the rear end of each shield jack 19. The tip of the drive rod 19a faces the front end surface of the existing segment S. By extending the drive rod 19a of the shield jack 19 backward and pressing against the segment S, a propulsive reaction force (i.e., thrust) can be applied to the excavator body 10. That is, the thrust generated when the shield jack 19 presses against the segment S allows the excavator body 10 to move forward.

[0035] The tunnel boring machine 1 shown in Figure 1 is a type of tunnel boring machine in which thrust is transmitted from the front end of the shield jack 19 to the boring machine body 10, but the tunnel boring machine according to the present invention is not limited to this example. For example, the tunnel boring machine according to the present invention may be a type of tunnel boring machine in which thrust is transmitted from the rear part of the shield jack 19 to the boring machine body 10. Furthermore, the tunnel boring machine according to the present invention may be a type of tunnel boring machine that has a folding function and is propelled by pushing the front body, or a type of tunnel boring machine that has a folding function and is propelled by pushing the rear body. In addition, the tunnel boring machine according to the present invention may be a tunnel boring machine in which the drive method of the cutter head 11 is a method other than the intermediate support method shown in Figure 1 (for example, a center shaft method, a central axis support method, or an outer circumference support method).

[0036] Incidentally, in tunnel excavation work using the tunnel boring machine 1, it is necessary to detect abnormalities in the tunnel boring machine 1 and take action when abnormalities occur. For example, a failure of the cutter rotation motor 16, which is the cutter drive unit, will have a significant impact on the progress of the work. Also, abnormalities in the earth pressure of the chamber 17, the thrust speed, and the thrust force, which are related to the excavation thrust, will have a significant impact on the stability of the ground (face) and the load acting on the tunnel boring machine 1. Therefore, it is desirable to grasp signs of various abnormalities, such as abnormalities in the cutter rotation motor 16 and the excavation thrust system, early and with accuracy. In this embodiment, by making improvements to the processing related to the diagnosis of abnormalities by the abnormality diagnosis device (specifically, the abnormality diagnosis device 20 described later), it is possible to grasp signs of abnormalities in the tunnel boring machine 1 early and with accuracy.

[0037] Figure 2 is a block diagram showing an example of the functional configuration of an abnormality diagnosis device 20 according to an embodiment of the present invention. The abnormality diagnosis device 20 is, for example, located on the rear side of the partition wall 13 of the tunnel boring machine 1.

[0038] The abnormality diagnosis device 20 includes a CPU (Central Processing Unit), which is an arithmetic processing unit; a ROM (Read Only Memory), which is a memory element that stores programs and calculation parameters used by the CPU; and a RAM (Random Access Memory), which is a memory element that temporarily stores parameters that change as appropriate during CPU execution.

[0039] As shown in Figure 2, the anomaly diagnosis device 20 is capable of communicating with various sensors 30_k (k=1, 2, 3...n) and the worker terminal 40.

[0040] The various sensors 30_k detect various measurement items MI_k (k=1, 2, 3...n) and output them to the anomaly diagnosis device 20. The number of sensors 30_k is not particularly limited, but it is sufficient to have at least two or more.

[0041] As will be described later, in the abnormality diagnosis performed by the abnormality diagnosis device 20 according to this embodiment, abnormalities in the tunnel boring machine 1 are diagnosed by focusing on the fluctuations in the correlation between multiple types of measurement items MI_k output to the abnormality diagnosis device 20 from various sensors 30_k. The measurement items MI_k are various parameters that represent the excavation state by the tunnel boring machine 1 and are measured during excavation work by the tunnel boring machine 1. Examples of measurement items MI_k include cutter torque (i.e., torque of the cutter rotation motor 16), thrust speed of the tunnel boring machine 1, thrust force of the tunnel boring machine 1, earth pressure of the chamber 17, rotation speed of the screw conveyor 18, pressure of the screw conveyor 18, backfill injection pressure, backfill injection flow rate, mud pressure, mud flow rate, excavation distance, ring number (i.e., the number of rings consisting of segments S installed), etc.

[0042] Figure 3 schematically illustrates the correlation between multiple types of measurement items MI_k. Note that while Figure 3 shows only five types of measurement items MI_k (MI_1, MI_2, MI_3, MI_4, and MI_5), in reality, there may be fewer than five types of measurement items MI_k, or six or more. As shown in Figure 3, each measurement item MI_k not only influences other single measurement items MI_k, but also influences each other in complex ways.

[0043] The worker terminal 40 is an information processing terminal used by a worker. For example, the worker terminal 40 may be carried by the worker or may be used while installed on a workbench or the like. The worker terminal 40 has various functions, such as a function to display information and a function to output sound.

[0044] The abnormality diagnosis device 20 comprises, for example, an analysis unit 21, a diagnosis unit 22, and a notification unit 23. The analysis unit 21 analyzes the fluctuations in the correlation between multiple types of measurement items MI_k related to the tunnel boring machine 1. The diagnosis unit 22 diagnoses abnormalities in the tunnel boring machine 1 based on the results of the analysis by the analysis unit 21. The notification unit 23 notifies the results of the diagnosis by the diagnosis unit 22. Details of each process performed by the abnormality diagnosis device 20 will be described later.

[0045] Figure 4 is a flowchart illustrating an example of a schematic flow of the abnormality diagnosis process performed by the abnormality diagnosis device 20 according to an embodiment of the present invention. The processing flow shown in Figure 4 is repeated at predetermined time intervals, for example, in tunnel excavation work.

[0046] When the processing flow shown in Figure 4 begins, in step S101, the analysis unit 21 analyzes the fluctuations in the correlation between multiple types of measurement items MI_k. Specifically, in the analysis of step S101, the analysis unit 21 analyzes the fluctuations in the correlation and the factors causing those fluctuations, as will be described later. Details of the analysis performed by the analysis unit 21 will be described later.

[0047] Following step S101, in step S102, the diagnostic unit 22 diagnoses any abnormalities in the tunnel boring machine 1 based on the results of the analysis by the analysis unit 21. In the diagnosis of step S102, the diagnostic unit 22 diagnoses the degree of the abnormality in the tunnel boring machine 1 and the cause of the abnormality. Details of the diagnosis by the diagnostic unit 22 will be described later, along with details of the analysis by the analysis unit 21.

[0048] Following step S102, in step S103, the notification unit 23 notifies the result of the diagnosis performed by the diagnostic unit 22, and the processing flow shown in Figure 4 ends. In step S103, the notification unit 23 outputs, for example, the result of the diagnosis in step S102 to the worker terminal 40. As a result, the result of the diagnosis in step S102 is notified to the worker, for example, by display and sound output on the worker terminal 40.

[0049] For example, in step S103, if the diagnostic unit 22 diagnoses an abnormality in the tunnel boring machine 1, the system can automatically notify the operator of the abnormality using email or other means, enabling a quick response.

[0050] Here, we will explain the details of the analysis process performed by the analysis unit 21 (step S101 in Figure 4). For example, the analysis unit 21 performs analyses using the Mahalanobis-Daguchi method and analyses using the multiple regression method. Below, we will explain the analysis using the Mahalanobis-Daguchi method and the analysis using the multiple regression method in order.

[0051] The analysis unit 21 may perform both the Mahalanobis-Daguchi method and the multiple regression method. Alternatively, the analysis unit 21 may perform only one of the two methods. Alternatively, the analysis unit 21 may perform analysis using methods other than the Mahalanobis-Daguchi method and the multiple regression method.

[0052] Figure 5 is a diagram illustrating the analysis using the Mahalanobis-Daguchi method. In performing the analysis using the Mahalanobis-Daguchi method, the analysis unit 21 pre-learns a model MO_1 of standard values ​​for multiple types of measurement items MI_k under normal operation of the tunnel boring machine 1. Then, based on the pre-learned model MO_1 and the current measured values ​​of the multiple types of measurement items MI_k, the analysis unit 21 analyzes the fluctuations in the correlation between the multiple types of measurement items MI_k and the factors causing these fluctuations. As explained above with reference to Figure 3, each measurement item MI_k influences the others in complex ways. The analysis using the Mahalanobis-Daguchi method analyzes the fluctuations in the correlation between such multiple types of measurement items MI_k and the factors causing these fluctuations.

[0053] Figure 5 conceptually shows the Mahalanobis-Daguchi model MO_1 when only two types of measurement items, MI_1 and MI_2, are used as measurement item MI_k, for ease of understanding. In this case, as shown in Figure 5, model MO_1 is represented on a two-dimensional plane with axes for measurement item MI_1 and measurement item MI_2.

[0054] In the example shown in Figure 5, for example, the analysis unit 21 continuously collects measured values ​​of measurement items MI_1 and MI_2 over a predetermined period while the tunnel boring machine 1 is in operation. The analysis unit 21 then defines model MO_1 as the region in the two-dimensional plane where the measured values ​​of measurement items MI_1 and MI_2 under normal conditions are densely distributed among the collected measured values. In other words, model MO_1 corresponds to an index that shows the standard values ​​of measurement items MI_1 and MI_2 under normal conditions. As described above, model MO_1 is pre-trained. Pre-training of model MO_1 allows for flexible selection of the learning range of the learning data used to train model MO_1, sequential addition of learning data according to the progress of excavation, and learning that includes the most recent learning data. These measures can lead to the evolution of the model itself. Note that this pre-training may be performed using data from other machines (sites) as well as data from the tunnel boring machine 1 in question (site).

[0055] The analysis unit 21 then calculates the Mahalanobis distance D in the two-dimensional plane described above. The Mahalanobis distance D is the distance between model MO_1 (for example, the center of model MO_1) and point P, which represents the actual measured values ​​of measurement items MI_1 and MI_2 at the present time. Here, the Mahalanobis distance D is an indicator of how much the correlation between measurement items MI_1 and MI_2 deviates from the normal state. In other words, the analysis unit 21 can determine that the longer the Mahalanobis distance D, the greater the fluctuation in the correlation between measurement items MI_1 and MI_2. If the analysis unit 21 determines that the fluctuation in the correlation between measurement items MI_1 and MI_2 is large (for example, if the value calculated by the Mahalanobis-Daguchi method is greater than or equal to a predetermined value), the diagnosis unit 22 diagnoses that there is a high possibility or degree of abnormality (i.e., the severity of the abnormality) of the tunnel boring machine 1.

[0056] Furthermore, if the analysis unit 21 determines that there is a large fluctuation in the correlation between measurement item MI_1 and measurement item MI_2, it can analyze how much each measurement item MI_k contributes to the fluctuation in the correlation (i.e., how much it contributes to the degree of abnormality) based on the positional relationship between model MO_1 (for example, the center of model MO_1) and point P, which shows the current measured values ​​of measurement item MI_1 and measurement item MI_2, in the two-dimensional plane described above. The diagnosis unit 22 can then diagnose, for example, the type of measurement item MI_k related to the cause of the abnormality in the tunnel boring machine 1 based on such analysis results.

[0057] For the sake of ease of understanding, the above explanation described the case where only two types of measurement items, MI_1 and MI_2, are used as measurement item MI_k. However, in reality, it is conceivable that there may be three or more measurement items MI_k. In this case, model MO_1 can be defined on a space of dimensions (i.e., n dimensions) corresponding to the number of measurement items MI_k. Even in that case, similar to the example above, the analysis unit 21 can analyze the fluctuations in the correlation between multiple types of measurement items MI_k, and the factors causing those fluctuations, by focusing on the Mahalanobis distance D.

[0058] According to the analysis using the Mahalanobis-Daguchi method described above, multivariate data consisting of multiple types of measurement items MI_k can be aggregated into a single index, and the degree of anomaly can be evaluated to determine how much each measurement item MI_k contributes to the degree of anomaly. Therefore, by using the Mahalanobis-Daguchi method, it is possible to analyze data during drilling, quantify changes in drilling conditions, and identify the measurement items MI_k that are causing those changes.

[0059] For the sake of ease of understanding, the above explanation describes an example in which the Mahalanobis-Daguchi model MO_1 corresponds to a region within a space having an axis for the measurement item MI_k. However, the Mahalanobis-Daguchi model MO_1 may also correspond to a region within a space having axes for other state items in addition to the axis for the measurement item MI_k representing the excavation state by the tunnel boring machine 1. Here, the other state items may be, for example, items representing the soil type of the ground excavated by the tunnel boring machine 1 (e.g., whether it is alluvial cohesive soil, diluvial cohesive soil, granite soil, sandy soil, gravel, bedrock, etc.). In that case, the distance between the model MO_1 in the above space and the point representing the current state is specified as the Mahalanobis distance D.

[0060] Furthermore, the analysis unit 21 may perform analysis using methods other than the Mahalanobis-Daguchi method to analyze the fluctuations in the correlation between multiple types of measurement items MI_k and the factors causing such fluctuations, based on the model MO_1 of the standard values ​​of multiple types of measurement items MI_k when the tunnel boring machine 1 is operating normally and the actual measured values ​​of multiple types of measurement items MI_k at present.

[0061] Figure 6 is a diagram illustrating analysis using the multiple regression method. When performing analysis using the multiple regression method, the analysis unit 21 pre-trains a model MO_2 that predicts the predicted value of one type of measurement item MI_k out of multiple types of measurement items MI_k using the measured values ​​of the other types of measurement items MI_k. Then, using model MO_2, the analysis unit 21 predicts the current predicted value of the one type of measurement item MI_k, and analyzes the fluctuations in the correlation between the multiple types of measurement items MI_k and the factors causing those fluctuations based on the difference between the current predicted value of the one type of measurement item MI_k and the current measured value of the one type of measurement item MI_k. As explained above with reference to Figure 3, each measurement item MI_k influences each other in a complex way. Analysis using the multiple regression method analyzes the fluctuations in the correlation between such multiple types of measurement items MI_k and the factors causing those fluctuations.

[0062] Figure 6 shows model MO_2, which predicts the value of measurement item MI_n from the measured values ​​of n-1 types of measurement item MI_k, namely MI_1, 2, 3...n-1. However, as will be discussed later, in practice, models that predict the value of measurement item MI_k of types other than measurement item MI_n can also be used as model MO_2.

[0063] In the example shown in Figure 6, for example, the analysis unit 21 continuously collects actual values ​​of all types of measurement items MI_k (i.e., n types of measurement items MI_1, 2, 3...n) over a predetermined period while the tunnel boring machine 1 is in operation. Then, using the normal operation data from the collected actual values, the analysis unit 21 learns a multiple regression equation as model MO_2, with n-1 types of measurement items MI_k (MI_1, 2, 3...n-1) as explanatory variables and measurement item MI_n as the dependent variable. As described above, model MO_2 is pre-trained. Pre-training of model MO_2 allows for flexible selection of the learning range of the learning data used to train model MO_2, sequential addition of learning data according to the progress of excavation, and learning that includes the most recent learning data. These measures can lead to the evolution of the model itself. Note that this pre-training may be done using data from other machines (sites) as well as data from the tunnel boring machine 1 in question (site).

[0064] The analysis unit 21 then inputs the current measured values ​​of n-1 types of measurement items MI_k (MI_1, 2, 3...n-1) into model MO_2 to predict the current value of measurement item MI_n. Next, the analysis unit 21 calculates the difference between the current predicted value of measurement item MI_n obtained in this way and the current measured value of measurement item MI_n. The analysis unit 21 then determines that the larger the above difference, the greater the fluctuation in the correlation between multiple types of measurement items MI_k. Furthermore, the analysis unit 21 determines that if the above difference is large, there is a high possibility that measurement item MI_n is related to the cause of the abnormality of the tunnel boring machine 1.

[0065] In the above explanation, for ease of understanding, one model MO_2 that predicts the value of one measurement item MI_n was described as an example. However, in reality, the analysis unit 21 pre-trains multiple models MO_2 that predict the value of each measurement item MI_k, for example. In other words, in this case, n types of models MO_2 are prepared in advance. The analysis unit 21 then calculates the difference between the current predicted value and the actual value of each measurement item MI_k, and based on the above difference for each measurement item MI_k, it analyzes the fluctuations in the correlation between multiple types of measurement items MI_k and the factors causing those fluctuations.

[0066] For example, if the above difference is small for any of the measurement items MI_k, the analysis unit 21 determines that the fluctuation in the correlation between multiple types of measurement items MI_k is small. On the other hand, if the above difference is large for at least one measurement item MI_k, the analysis unit 21 determines that the fluctuation in the correlation between multiple types of measurement items MI_k is large. Then, if the analysis unit 21 determines that the fluctuation in the correlation between multiple types of measurement items MI_k is large, the diagnosis unit 22 diagnoses that there is a high possibility or degree of abnormality (i.e., the severity of the abnormality) of the tunnel boring machine 1.

[0067] Furthermore, if the analysis unit 21 determines that there is a large variation in the correlation between multiple types of measurement items MI_k, it can analyze that the measurement item MI_k analyzed as having a large difference contributes significantly to the variation in the correlation (i.e., contributes significantly to the degree of abnormality). The diagnosis unit 22 can then, for example, diagnose the type of measurement item MI_k related to the cause of the abnormality in the tunnel boring machine 1 based on such analysis results.

[0068] The analysis using the multiple regression method described above allows for a quantitative evaluation of how much each explanatory variable influences the dependent variable by using a statistical method that predicts the dependent variable from multiple explanatory variables. Therefore, by using the multiple regression method, it is possible to calculate the predicted value of each measurement item MI_k, output the measurement item MI_k that is considered to be a factor in the change in the excavation state, and the difference between the predicted value and the actual value of that measurement item MI_k.

[0069] For the sake of clarity, the above explanation describes an example where the input to the multiple regression model MO_2 is only the measurement item MI_k. However, the input to the multiple regression model MO_2 may include other condition items in addition to the measurement item MI_k. Here, the other condition items may be, for example, items that represent the soil type of the ground excavated by the tunnel boring machine 1 (e.g., whether it is alluvial cohesive soil, diluvial cohesive soil, granite soil, sandy soil, gravel, bedrock, etc.). In that case, when predicting the current measurement item MI_k, data from the other condition items will be input to the multiple regression model MO_2 in addition to the measurement item MI_k.

[0070] Furthermore, the analysis unit 21 may use model MO_2, which predicts the predicted value of one of the multiple types of measurement items MI_k using the measured values ​​of the other measurement items MI_k, to predict the current predicted value of the one type of measurement item MI_k. Based on the difference between the current predicted value of the one type of measurement item MI_k and the current measured value of the one type of measurement item MI_k, the analysis unit 21 may perform analysis using a method other than multiple regression to analyze the changes in the correlation between the multiple types of measurement items MI_k and the factors causing such changes.

[0071] Here, the Mahalanobis-Daguchi method alone has a problem in that it is difficult to determine whether the increase in anomaly is due to a change in the correlation between the measurement items MI_k, or simply a result of the value of one measurement item MI_k increasing. Therefore, by performing the analysis using the multiple regression method described above in addition to the Mahalanobis-Daguchi method described above, the above problem can be solved, and it becomes possible to detect changes in drilling conditions early and identify the causes. In this case, the Mahalanobis-Daguchi method detects changes in the correlation between the measurement items MI_k (changes in drilling conditions) and identifies the causes (measurement items MI_k), while the multiple regression method calculates the predicted value of each measurement item MI_k and outputs the difference between the predicted value and the measured value. On the other hand, when the Mahalanobis-Daguchi method and the multiple regression method are used in combination, the changes in drilling conditions can be quantified, the causes of the changes in drilling conditions (measurement items MI_k) can be identified, the predicted value of each measurement item MI_k can be calculated, and the difference between the predicted value and the measured value of the measurement item MI_k that is considered to be the cause of the changes in drilling conditions can be output.

[0072] As described above, the analysis unit 21 pre-trains model MO_1 when performing analysis using the Mahalanobis-Daguchi method. Furthermore, the analysis unit 21 pre-trains model MO_2 when performing analysis using the multiple regression method. Hereafter, these models will be collectively referred to as model MO_k. The following describes improvements to the training method of model MO_k as a way to enhance the accuracy of the analysis performed by the analysis unit 21.

[0073] Figure 7 is a flowchart illustrating an example of the schematic flow of the process related to learning the model MO_k performed by the anomaly diagnosis device 20 according to an embodiment of the present invention. The processing flow shown in Figure 7 is performed, for example, after the collection of measurement data of multiple types of measurement items MI_k during the operation of the tunnel boring machine 1 (i.e., data necessary for learning the model MO_k) is completed.

[0074] When the processing flow shown in Figure 7 begins, in step S201, the analysis unit 21 uses an isolation forest to remove abnormal data from the measurement data. In learning, measurement data from past (or most recent) drilling performance becomes the learning data, but this includes data from when anomalies occurred, and it is difficult to manually extract and remove such data, so a mechanical (automatic) operation to remove abnormal data is necessary.

[0075] An isolation forest is an algorithm that infers outliers from multiple randomly generated decision trees. For example, in an isolation forest, the decision tree is repeatedly divided until each data point is isolated, and outliers are estimated from the distance (depth) to isolation. The analysis unit 21 excludes data that the isolation forest has estimated to be outliers as anomalous data.

[0076] Following step S201, in step S202, the analysis unit 21 performs cluster analysis and learns a model MO_k for each cluster, and the processing flow shown in Figure 7 is completed. In step S202, the analysis unit 21 learns a model MO_k for each cluster using the measurement data after the anomalous data has been excluded in step S201.

[0077] Figure 8 is a diagram illustrating cluster analysis. For ease of understanding, Figure 8 shows a case where only three types of measurement items, MI_1, MI_2, and MI_3, are used as measurement item MI_k. In Figure 8, the measurement data is represented by dots, which are placed in a three-dimensional space with axes for measurement item MI_1, MI_2, and MI_3.

[0078] For example, the analysis unit 21 classifies a group of measurement data that are close to each other in the three-dimensional space described above as a cluster Ck. In the example in Figure 8, the measurement data is classified into three clusters Ck: cluster C1, cluster C2, and cluster C3. The analysis unit 21 then separately learns a model MO_k for cluster C1, a model MO_k for cluster C2, and a model MO_k for cluster C3. In this case, in step S101 in Figure 4, the analysis unit 21 determines, for example, which cluster Ck the current measured values ​​of multiple types of measurement items MI_k belong to, and performs the analysis using the model MO_k for the determined cluster Ck.

[0079] For the sake of ease of understanding, the above explanation described the case where only three types of measurement items, MI_1, MI_2, and MI_3, are used as measurement item MI_k. However, in reality, it is expected that there will be four or more measurement items MI_k. In this case, cluster Ck can be defined in a space of dimensions corresponding to the number of measurement items MI_k (i.e., n dimensions). In this case as well, similar to the example above, the analysis unit 21 classifies data groups consisting of measurement data that are close to each other in the above n-dimensional space as cluster Ck. Cluster analysis generates cluster Ck (groups), which is a statistical and automatic analysis of some factor, such as "by soil type," "by operator," or "by machine size." Analyzing which cluster Ck the current event belongs to means that higher accuracy can be obtained when applying the learning model.

[0080] In the above, an example was described in which the analysis unit 21 learns multiple models MO_k based on the distribution of measurement data (for example, the distribution in three-dimensional space in the example of Figure 8). However, the analysis unit 21 may also learn a model MO_k for each type of item representing a state other than the measurement item MI_k. Here, the item representing a state other than MI_k may be, for example, an item representing the soil type of the ground excavated by the tunnel boring machine 1 (for example, whether it is alluvial cohesive soil, diluvial cohesive soil, granite soil, sandy soil, gravel, or bedrock). Alternatively, the item representing a state other than MI_k may be, for example, the type of tunnel boring machine 1. In that case, in step S101 of Figure 4, the analysis unit 21 performs the analysis using, for example, a model MO_k corresponding to the item representing the state other than MI_k at present (for example, a model MO_k corresponding to the current soil type).

[0081] As described above, by using isolation forests, it becomes possible to detect and separate data that is considered abnormal from drilling data that cannot be clearly classified as normal or abnormal. This allows only the data that should be considered normal to be used as training data, thereby improving the reliability of the diagnostic results.

[0082] Furthermore, the analysis unit 21 may exclude abnormal data from the measurement data of multiple types of measurement items MI_k during the operation of the tunnel boring machine 1 using a method other than isolation forest, and then train the model MO_k using the measurement data after the exclusion of abnormal data.

[0083] Furthermore, as described above, classifying training data and creating models using cluster analysis allows for the absorption of differences in construction sites, soil types, and operator characteristics, enabling more accurate anomaly detection and failure prevention.

[0084] The analysis unit 21 may also pre-train multiple models MO_k using methods other than cluster analysis, and then perform the analysis using one model MO_k selected from among the multiple models MO_k.

[0085] As described above, the abnormality diagnosis device 20 according to this embodiment includes an analysis unit 21 that analyzes fluctuations in the correlation between multiple types of measurement items MI_k related to the tunnel boring machine 1, and a diagnosis unit 22 that diagnoses abnormalities in the tunnel boring machine 1 based on the results of the analysis. By focusing on the fluctuations in the correlation described above, it becomes possible to grasp signs of abnormalities in measurement items MI_k, such as abnormalities in the cutter rotation motor 16, as well as signs of abnormalities in the measurement items MI_k shown in Figure 3, early and with good accuracy, as described above.

[0086] Preferred embodiments of the present invention have been described above with reference to the attached drawings. However, it goes without saying that the present invention is not limited to the embodiments described above, and that various modifications or alterations within the scope of the claims also fall within the technical scope of the present invention.

[0087] For example, although the above describes a tunnel boring machine 1 of the earth pressure type (including the mud pressure type), the tunnel boring machine according to the present invention may also be of the slurry type.

[0088] Furthermore, although the above description of the components of the tunnel boring machine 1 was given with reference to the drawings, the dimensions and positional relationships of the components in the drawings are merely illustrative examples, and the dimensions and positional relationships of the components of the tunnel boring machine 1 are not limited to the examples shown in the drawings. In addition, components may be added, deleted, or modified as appropriate for the tunnel boring machine 1 illustrated in the drawings. [Explanation of Symbols]

[0089] 1. Tunnel boring machine 10 Excavator body 11 cutter heads 12 Cutter central axis 13 Bulkhead 14 Rotating Rings 15 Linked beams 16 Cutter Swivel Motor 17 Chambers 18 Screw conveyor 19 Shield Jack 20 Anomaly Diagnosis Device 21 Analysis Department 22 Diagnostic Department 23 Hochi Department 30k sensor 40 Worker terminals Ck cluster D. Distance of Mahalanobis MI_k Measurement Items MO_k Model S segment

Claims

1. An abnormality diagnosis device for a tunnel boring machine, An analysis unit that analyzes the fluctuations in the correlation between multiple types of measurement items related to the tunnel boring machine, Based on the results of the above analysis, a diagnostic unit is provided to diagnose any abnormalities in the tunnel boring machine, Equipped with, An anomaly detection device for tunnel boring machines.

2. The aforementioned analysis unit is The model of standard values ​​for the multiple types of measurement items of the tunnel boring machine is learned in advance, In the above analysis, based on the model and the actual measured values ​​of the multiple types of measurement items at present, the fluctuations in the correlation and the factors causing such fluctuations are analyzed. An abnormality diagnosis device for a tunnel boring machine according to claim 1.

3. The analysis unit performs the analysis using the Mahalanobis-Daguchi method. An abnormality diagnosis device for a tunnel boring machine according to claim 2.

4. The aforementioned analysis unit is A model is pre-trained to predict the predicted value of one of the multiple types of measurement items using the measured values ​​of the other types of measurement items. In the above analysis, The current predicted value of one type of measurement item is predicted using the model, Based on the difference between the predicted value of the aforementioned one type of measurement item at present and the actual measured value of the aforementioned one type of measurement item at present, the fluctuations in the correlation and the factors causing such fluctuations are analyzed. An abnormality diagnosis device for a tunnel boring machine according to claim 1.

5. The analysis unit performs the analysis using the multiple regression method. An abnormality diagnosis device for a tunnel boring machine according to claim 4.

6. The analysis unit performs the analysis using the Mahalanobis-Daguchi method in addition to the multiple regression method. An abnormality diagnosis device for a tunnel boring machine according to claim 3.

7. The aforementioned analysis unit is Abnormal data is excluded from the measurement data of the actual values ​​of the multiple types of measurement items during the operation of the tunnel boring machine. The model is trained using the measurement data after the removal of the abnormal data. An abnormality diagnosis device for a tunnel boring machine according to any one of claims 2 to 6.

8. The analysis unit uses an isolation forest to exclude the abnormal data from the measurement data. An abnormality diagnosis device for a tunnel boring machine according to claim 7.

9. The aforementioned analysis unit is Multiple of the aforementioned models are pre-trained, The analysis is performed using one of the aforementioned models selected from among the multiple aforementioned models. An abnormality diagnosis device for a tunnel boring machine according to any one of claims 2 to 6.

10. The system includes a notification unit that notifies the results of the aforementioned diagnosis. An abnormality diagnosis device for a tunnel boring machine according to any one of claims 1 to 6.

11. A method for diagnosing abnormalities in a tunnel boring machine, An analysis step for analyzing the fluctuations in the correlation between multiple types of measurement items related to the tunnel boring machine, Based on the results of the above analysis, a diagnostic process is performed to diagnose any abnormalities in the tunnel boring machine, including, Methods for diagnosing abnormalities in tunnel boring machines.

Images