JP7832069B2 - Soil condition determination system and soil condition determination method - Google Patents

Description

This invention relates to a technology for determining the properties of the target ground during boring, and more specifically, to a ground property determination system that can determine the properties of the target ground in real time during construction using a pre-generated, trained model.

Boring is a construction method that creates relatively small, relatively long, cylindrical boreholes in the ground. It is usually performed by drilling using a rod and a bit attached to its end. Boring is sometimes performed for ground investigation purposes, such as core sampling or standard penetration testing, but it can also be part of the main construction work. For example, in blast tunnel excavation, a drill jumbo is used to load explosives (dynamite), and in NATM (New Australian Tunneling Method) tunnel construction, a drill jumbo is also used to insert rock bolts. Furthermore, when performing ground improvement through deep mixing, such as mechanical agitation or high-pressure injection, boring is performed vertically downwards relative to the ground.

In recent years, heavy rains associated with extreme weather events have led to deep-seated landslides and large-scale mudslides on slopes and embankments, making stabilization measures for slopes and other areas more important than ever. Anchor systems have become indispensable for preventing deep-seated landslides and mudslides on slopes and other areas. This anchor system, which involves installing one fixed end of a tendon into the ground, was formerly called an "earth anchor" and is now known as a "ground anchor." Ground anchors were first introduced in Japan in 1957 as a stabilization measure for dams and have been used for nearly 60 years. Because they are economical and easy to install, and coupled with the acceleration of infrastructure development accompanying high economic growth, they have been widely used, with an estimated 24,000 installations. Ground anchors are also installed using boring machines such as rotary percussion boring machines.

When conducting permanent boring work (hereinafter referred to as "boring work"), a ground survey is generally conducted beforehand, and drilling is carried out according to the survey results. For example, in boring work to install ground anchors, a ground survey is conducted to determine the location of the anchorage layer and its physical properties (such as surface friction resistance). While drilling is carried out according to the design plan based on the results of this ground survey, it is naturally impossible to visually inspect the conditions inside the ground during drilling, making it extremely difficult to accurately determine whether the work will proceed as planned according to the design.

On the other hand, it is known that understanding the properties of the ground in real time while boring work is being carried out is extremely beneficial. For example, if the location and physical properties of the anchorage layer can be determined in real time while drilling for ground anchors, it may be found that a shorter drilling length (free length or anchorage length) than planned is sufficient, or conversely, that a longer drilling length is necessary for the anchor to function. As a result, it becomes possible to install ground anchors efficiently and effectively.

Therefore, various efforts have been made to understand the properties of the ground in real time during boring work. Representative methods include techniques for estimating the hardness or softness of the ground from electrical signals (torque, thrust, impact pressure, etc.) from construction machinery, as disclosed in Patent Document 1, and techniques for determining geological and soil classifications by visually inspecting the appearance of the slime brought to the surface during drilling, or by conducting chemical tests using chemicals.

Japanese Patent Publication No. 2021-127655

However, conventional methods for understanding the properties of the ground in real time all had their problems. For example, methods that estimate the hardness or softness of the ground from the electrical signals of construction machinery (hereinafter referred to as "machine information estimation methods") can determine the hardness or softness of the ground and the presence or absence of voids, but are unsuitable for determining geological or soil classifications. On the other hand, methods that judge the appearance of slime (hereinafter referred to as "slime appearance judgment methods") can determine geological or soil classifications, but it is difficult to determine the hardness or softness of the ground and the presence or absence of voids.

Furthermore, in machine information estimation methods, information may be acquired from various measuring instruments and devices attached to construction machinery. However, this information is independent, differing in format and dimensions (1- to 3-dimensional). Even if uploaded and aggregated into a single central system, it is not easy to comprehensively interpret this information. In other words, judging the properties from such disparate, unorganized information relied on human knowledge, experience, and technical skills. Consequently, the results varied depending on the judgment of the person making the decision, making it impossible to obtain a stable judgment. Moreover, because the analysis was conducted by humans with careful consideration, it was sometimes impossible to determine the ground properties (geological classification, soil classification, etc.) in real time, and the acquired information could not be fed back into construction in a timely manner.

As described above, the slime appearance assessment method involves a skilled technician visually inspecting the slime's appearance to determine the geological and soil classifications. Therefore, conducting boring work requires securing and stationing a specialized technician, and this personnel cost must be included in the budget. In particular, when multiple construction teams are deployed for drilling, several specialized technicians must be stationed there, which is costly and can be difficult to secure the necessary number. Furthermore, it is difficult to quickly consolidate the assessment results from multiple technicians, resulting in the inability to assess ground properties in real time and feed the obtained information back into construction in a timely manner. Additionally, differences in assessment results due to the skill level of the technicians are unavoidable, making it difficult to obtain consistent assessments.

The objective of this invention is to solve the problems of the prior art, namely, to provide a ground condition determination system that can grasp the ground conditions during drilling in real time without relying on human judgment.

The ground property determination system of the present invention focuses on generating a trained model by learning construction data with type labels, and then determining geological and soil classifications in real time by inputting construction data into the trained model during construction. This invention is based on a completely new concept.

The ground property determination system of the present invention is a system for determining the properties of the target ground to be drilled or excavated, and comprises a trained model generation means and a ground property determination means. Of these, the trained model generation means is a means for generating a trained model by machine learning on training data to which geological classification labels and soil classification labels are attached, and the ground property determination means is a means for determining the geological classification and soil classification of the target ground by inputting construction data acquired during drilling (or excavation) into the trained model. The training data is generated by attaching geological classification labels and soil classification labels to one or more data selected from among construction machine data related to construction obtained from a drilling machine (or excavation machine), camera image data obtained by photographing ground slime obtained by drilling (or excavation), spectral image data obtained by acquiring ground slime with a spectral camera, and borehole image data obtained by photographing with a borehole camera after drilling (or excavation). Furthermore, the construction data input to the trained model consists of one or more data sets selected from construction machine data, camera image data, spectral image data, and borehole image data, and is of the same type as the training data used in machine learning.

The ground property determination system of the present invention can also generate a trained model by machine learning using training data with ground hardness/softness labels. In this case, the ground property determination means also determines the hardness/softness of the target ground when construction data is input to the trained model.

The ground property determination system of the present invention can also generate a trained model by machine learning using training data with labels indicating ground cavities. In this case, the ground property determination means also determines the condition of cavities in the target ground when construction data is input to the trained model.

The ground condition determination system of the present invention may further comprise one or more terminal devices and a central device capable of communicating with these terminal devices. In this case, construction data input using the terminal devices is transmitted to the central device. The central device receives the construction data and, by inputting the construction data into the ground condition determination means, determines the geological classification and soil classification of the target ground. The terminal devices can also output the results determined by the ground condition determination means in real time.

The ground condition determination system of the present invention can also be configured to display in three dimensions (or two dimensions) by overlaying the actual drilling (or excavation) results as construction progresses with the planned drilling (or excavation) shape.

The ground condition determination system of the present invention has the following effects:
(1) It is possible to determine information about the ground that is invisible (geological classification, soil classification, hardness of the ground, presence or absence of voids) in real time, and as a result, necessary information can be quickly fed back to the construction team, enabling more appropriate boring work to be carried out.
(2) Because it does not rely on human knowledge, human error is eliminated, and stable and consistent judgment results can be obtained.
(3) Even when drilling is carried out by deploying multiple construction teams, the construction data from each team can be collected in a central device, allowing various stakeholders to review and share it. This enables appropriate advice from all parties involved, such as predictions regarding uncertainties in the construction environment and appropriate construction management.
(4) By overlaying the actual drilling (or excavation) results as construction progresses, the planned drilling (or excavation) shape, and the determined ground information (geological classification, soil classification, ground hardness, presence or absence of voids) and displaying them in three dimensions (or two dimensions), the properties of the ground can be grasped more clearly, and for example, it becomes possible to confirm the planned anchorage layer and bearing layer.
(5) By using the information obtained during construction as training data and continuously performing deep learning, it is possible to obtain a trained model with further improved estimation accuracy.

A model diagram schematically showing the ground conditions determined based on construction data. A block diagram showing the main components of the ground condition determination system of the present invention. A block diagram showing a ground condition assessment system with a central unit and terminal equipment distributed amongst the others. An overview diagram showing an example of construction machine data obtained from a drilling machine. (a) is a schematic model image showing the ground surface and ground anchor head in 3D, and (b) is a schematic model image showing the ground cross section and ground anchor in 3D. (a) is a schematic model image showing the planned borehole displayed by the output control means, and (b) is a schematic model image showing the overlapping display of the planned borehole and the completed borehole displayed by the output control means. This flowchart shows the main processing steps from conducting test construction to generating a trained model using a ground condition assessment system. This flowchart shows the main process flow from the time of construction until the ground properties are determined using the ground properties determination system.

An example of the ground condition determination system of the present invention will be explained with reference to the diagram.

1. Overall Overview The present invention is a technology that can determine the properties of the target ground in real time while boring work or ground excavation is being carried out. More specifically, as shown in Figure 1, the ground properties are determined based on data obtained during construction (hereinafter referred to as "construction data"). The present invention can also determine the ground properties at each drilling depth during boring work, or at each excavation depth during ground excavation work, but for convenience, the example of boring work will be explained here.

Construction data includes, for example, "construction machine data," "slime photographs," "spectral data," and "borehole photographs," as shown in Figure 1. Of these, construction machine data, as described later, is various information obtained from drilling machinery (such as boring machines and drill jumbos) used in boring work. Slime photographs are image data obtained by photographing the slime brought to the surface during drilling with a digital camera, and spectral data is obtained by photographing the slime with a hyperspectral camera or multispectral camera. However, it is desirable to obtain slime photographs and spectral data together with a color sample. Borehole photographs are image data obtained by photographing the inner wall of the borehole with a borehole camera. Construction data is acquired periodically (or intermittently) during construction. For example, construction machine data is acquired every 15 cm of drilling, slime photographs and spectral data are acquired every 1-2 m of drilling, and borehole photographs are acquired when drilling is completed.

On the other hand, ground properties can be described as follows, for example, as shown in Figure 1: "geological classification," "soil classification," "ground hardness/softness," and "void conditions." Geological classification refers to classifications such as sandstone, granite, and slate, while soil classification refers to classifications such as sandy soil, gravelly soil, and cohesive soil. Ground hardness/softness is an indicator of its strength, such as the rock class classification, and void conditions refer to the presence, size, and scale of voids—literally, the condition of any cavities.

When determining ground properties based on construction data, a "trained model" generated by machine learning is used. In other words, when construction data is input, the trained model outputs the ground properties. Various conventional machine learning techniques, including deep learning, can be used to generate the trained model.

To generate a trained model using machine learning, training data is used. This training data consists of construction data to which correct labels, i.e., known ground properties, are attached (hereinafter simply referred to as "ground property labels"). In order to obtain training data, it is necessary to conduct a trial construction (hereinafter simply referred to as "trial construction") beforehand. In other words, after conducting trial construction, the ground properties are determined during the actual construction (hereinafter simply referred to as "main construction"). It is desirable to use the same machinery as in the main construction (especially drilling machinery) in the trial construction, and at least the same construction data as that obtained in the main construction should be obtained. Therefore, in order to distinguish between the two types of construction data, the data obtained in trial construction will be called "trial construction data," and the data obtained in main construction will be called "main construction data." Alternatively, main construction carried out in the past can be treated as trial construction. In short, training data is created by attaching ground property labels to construction data based on various data obtained in main construction carried out in the past, and a trained model is generated by training on this training data. Of course, training data can also be created by utilizing information obtained from both test construction and past full-scale construction projects.

2. Ground Condition Determination System The ground condition determination system of the present invention will be described in detail. Figure 2 is a block diagram showing the main configuration of the ground condition determination system 100 of the present invention. As shown in this figure, the ground condition determination system 100 of the present invention is configured to include a trained model generation means 101 and a ground condition determination means 102. It can also be configured to include a construction machine data acquisition means 103, an output control means 104, an output means 105, a trained model storage means 108, a test boring data storage means 109, a test construction data storage means 110, and a main construction data storage means 111. Furthermore, it can also be configured to include a terminal-side transmitting and receiving means 106, a central-side transmitting and receiving means 107, terminal equipment, and a central device, which will be described later.

The main components of the ground condition determination system 100—the trained model generation means 101, the ground condition determination means 102, and the output control means 104—can be manufactured as dedicated components or utilize general-purpose computer equipment. This computer equipment includes a processor such as a CPU, memory such as ROM or RAM, input means such as a mouse or keyboard, and a display. It can be comprised of personal computers (PCs), servers, tablet PCs such as iPads (registered trademark), and mobile terminals including smartphones. When using computer equipment with a display, it is preferable to use that display as the output means 105.

The trained model storage means 108, the test boring data storage means 109, the test construction data storage means 110, and the actual construction data storage means 111 can utilize the storage devices of a general-purpose computer or be built on a database server. When built on a database server, it can be located on a local network (LAN) or on a cloud server accessed via the internet. These storage means can be configured as separate units or two or more storage means can be combined into a single unit.

While some elements of the ground condition determination system 100 can utilize general-purpose computer equipment, as explained earlier, it can also be further composed of terminal equipment and a central device, as shown in Figure 3. The terminal equipment should be placed near the construction team actually performing the boring work, and may utilize personal computers, tablet PCs, or smartphones. The central device, on the other hand, should be located in a management office or supervisor's office, separate from the construction team, and may utilize servers or personal computers. Furthermore, by providing terminal-side transmission/reception means 106 in the terminal equipment and central-side transmission/reception means 107 in the central device, communication between the terminal equipment and the central device becomes possible. This allows for the transmission of various information (e.g., construction data) from the terminal equipment to the central device, and from the central device to the terminal equipment (e.g., determination results). Figure 3 shows an example where terminal equipment located in two construction teams is connected to the central device, but the system is not limited to this; it can also be operated with one terminal device, or three or more terminal devices connected to the central device.

The following will provide a detailed explanation of each of the main elements constituting the ground condition determination system 100 of the present invention.

(Means for generating pre-trained models)
The pre-trained model generation means 101 is a means for generating a pre-trained model. Specifically, this pre-trained model is generated by training the model on training data (construction data with ground condition labels attached) using machine learning techniques such as deep learning. The training data can be obtained by conducting test construction as described above. Below, an example of the procedure for obtaining the training data will be described.

First, a test borehole (hereinafter simply referred to as the "test borehole") is formed. For example, if the main construction is for the installation of ground anchors and the plan is to drill using a rotary percussion boring machine, then the test borehole should be drilled using the planned (or equivalent) rotary percussion boring machine. During this process, drilling should be carried out while acquiring the construction data planned for the main construction. For example, if the main construction plans to acquire construction machine data, camera image data, spectral image data, and borehole image data, then the test construction should also be carried out while acquiring construction machine data, camera image data, spectral image data, and borehole image data.

Here, construction machine data refers to various information obtained from drilling machines (such as boring machines and drill jumbos) used in boring work, as previously described, and is acquired by the construction machine data acquisition means 103 mounted (or attached) to the drilling machine. For example, as shown in Figure 4, construction machine data can include raw data such as the drilling speed, feed force, torque, water pressure, impact pressure, and water flow rate of the boring machine, as well as impact energy, number of impact blows, fracture energy value, and drilling energy value derived from this raw data. Of course, it is not limited to these; various information related to the drilling machine, such as measured values like the bit position and rod insertion angle during drilling, can also be acquired as construction machine data. During test construction, the acquired construction machine data, as well as construction data such as camera image data, spectral image data, and borehole image data (i.e., test construction data), are stored in the test construction data storage means 110 (Figure 2).

Once the test borehole is formed, the acquired borehole core is observed. Furthermore, physical test data of the ground is obtained by performing uniaxial (or triaxial) compression tests, in-situ horizontal loading tests, standard penetration tests during drilling, and other desired physical tests on the obtained borehole core. Based on the observation results of the borehole core and the physical test data, a person (e.g., a specialist engineer) evaluates the ground properties of the test borehole. Specifically, the test borehole is divided into blocks based on depth, and the geological classification and soil classification (or one of them) are evaluated for each block. At this time, in addition to the geological and soil classifications, the hardness and softness of the ground and the presence of voids can also be evaluated. The physical test data obtained here, as well as the ground properties for each block (geological and soil classifications, etc.), are stored in the test borehole data storage means 109 (Figure 2).

Meanwhile, during the drilling of the test boreholes, test construction data such as construction machine data, camera image data, spectral image data, and borehole image data are obtained. In particular, multiple data sets of construction machine data, camera image data, and spectral image data are obtained for each drilling depth. Based on the depth at which this test construction data was acquired, ground properties (such as geological classification and soil classification) are associated with each test construction data set. Training data is then created by assigning ground property labels such as geological classification labels, soil classification labels, hardness/softness labels, and cavity labels to each test construction data set.

Training data should be created according to the type of ground properties evaluated. For example, when a specialist evaluates geological classification and soil classification, training data should be created by adding geological classification labels and soil classification labels to the test construction data. When a specialist evaluates geological classification, soil classification, ground hardness/softness, and void conditions, training data should be created by adding geological classification labels, soil classification labels, hardness/softness labels, and void labels to the test construction data.

Furthermore, training data should be created according to the type of test construction data acquired. For example, in cases where only construction machine data is acquired as test construction data, training data is created by assigning ground property labels to that construction machine data. However, in cases where construction machine data, camera image data, spectral image data, and borehole image data are acquired as test construction data, training data is created by assigning ground property labels to the construction machine data, camera image data, spectral image data, and borehole image data, respectively. Alternatively, in cases where construction machine data, camera image data, and spectral image data are acquired as test construction data, training data can also be created by assigning ground property labels to a dataset consisting of construction machine data, camera image data, and spectral image data (a dataset corresponding to the drilling depth), that is, to each dataset.

As previously mentioned, training data does not necessarily need to be created after conducting a test construction (meaning not the actual construction in this case). Training data can also be created based on various information obtained from past actual constructions, or it can be created based on various information obtained from test constructions and various information obtained from past actual constructions.

The trained model generation means 101 generates a trained model by learning from a large amount of training data obtained through the above procedure. The training data can be created from one test borehole, or from two or more test boreholes. The trained model generated by the trained model generation means 101 is stored in the trained model storage means 108 (Figure 2).

(Means for determining ground properties)
During construction, construction data such as construction machine data, camera image data, spectral image data, and borehole image data (i.e., actual construction data) is stored in the actual construction data storage means 111 (Figure 2). The ground property determination means 102 reads the actual construction data from the actual construction data storage means 111 and outputs the ground properties (geological classification, soil classification, etc.) by inputting the actual construction data into a trained model. As described above, construction data is acquired periodically (or intermittently) during construction. Therefore, the ground property determination means 102 can also output the ground properties each time actual construction data is acquired, in other words, for each drilling depth. This makes it possible to grasp the location and physical properties of the anchoring layer in real time, for example, when drilling ground anchors, and as a result it is possible to find out that a shorter drilling length than planned is sufficient, or conversely, that it is necessary to drill longer than planned for it to function, thereby realizing efficient and effective construction.

(Output control means)
The ground properties (such as geological classification and soil classification) determined by the ground property determination means 102 are output to an output means 105, such as a display or printer. At this time, the ground properties for each drilling depth can be output in list format or in drawing format. Furthermore, the output control means 104 can display the ground properties for each drilling depth, along with the shape of the borehole formed by the main construction (hereinafter simply referred to as "main borehole"), in two dimensions (2D) or three dimensions (3D).

In cases where measurement values such as the position of the drill bit and the insertion angle of the rod are acquired as construction machine data, the shape of the drilled hole (i.e., the actual completed shape) can be grasped according to the progress of construction. Therefore, the output control means 104 can display the actual borehole in 3D (or 2D) on the output means 105 , as shown in Figure 5. Furthermore , as shown in Figure 6, the output control means 104 can also display the shape of the planned borehole (hereinafter referred to as "planned borehole DP") (drilled shape) and the shape of the actual completed borehole (hereinafter referred to as "completed borehole DA") overlapping on the output means 105 in 3D (or 2D). Note that in Figure 6(a), only the planned borehole DP is shown with a dashed line, and in Figure 6(b), the overlapping display of the planned borehole DP and the completed borehole DA is shown.

Furthermore, in Figure 6(b), the ground conditions for each drilling depth are displayed on the completed borehole DA. Different ground conditions can be displayed by changing the color, shade, or pattern. Also, as shown in Figure 3, when the system is configured with terminal equipment and a central unit, the construction team performing the boring work can also check the information displayed on the output means 105 of the terminal equipment in real time. For example, by using the terminal-side transmission/reception means 106 of the terminal equipment to transmit the actual construction data to the central unit, which is then received by the central-side transmission/reception means 107. The ground condition determination means 102 then determines the ground conditions for each drilling depth and transmits this information to the terminal equipment. This allows the construction team to sequentially check the ground conditions for each drilling depth along with the planned borehole DP and the completed borehole DA.

(Example of use)
An example of using the ground condition determination system 100 of the present invention will be described with reference to Figures 7 and 8. Figure 7 is a flowchart showing the main processing flow from conducting test construction to generating a trained model using the ground condition determination system 100. Figure 8 is a flowchart showing the main processing flow of determining the ground conditions using the ground condition determination system 100 and displaying the results during actual construction. In Figures 7 and 8, the central column shows the processing to be performed, the left column shows the information necessary for that processing, and the right column shows the information generated from that processing.

To generate the trained model, drilling is first performed to create a test borehole, as shown in Figure 7 (Step 211 in Figure 7). At this time, it is preferable to use drilling equipment similar to that planned for the actual construction. During drilling of the test borehole, a bore core is collected (Step 212 in Figure 7), construction data (test construction data) similar to that planned for the actual construction is obtained (Step 213 in Figure 7), and a standard penetration test is performed as needed.

Once the test borehole is formed, physical tests such as in-bore horizontal loading tests and boring core compression tests are conducted to acquire physical test data of the ground (Step 214 in Figure 7), and the acquired boring cores are observed. Then, for example, a specialist engineer evaluates the ground properties for each depth block of the test borehole based on the physical test data and the observation results of the boring cores. Here, the ground properties can be evaluated by geological classification and soil classification (or one of them), or in addition to geological classification and soil classification, the hardness and softness of the ground and the presence of voids can also be evaluated (Step 215 in Figure 7).

When ground properties are evaluated, training data is created by associating the ground properties with each test construction data based on the acquisition depth, and then assigning ground property labels to each test construction data (Step 216 in Figure 7). It is advisable to create training data according to the type of ground properties evaluated. For example, if a specialist evaluates geological classification and soil classification, training data is created by assigning geological classification labels and soil classification labels to the test construction data. If a specialist evaluates geological classification, soil classification, ground hardness/softness, and void conditions, training data is created by assigning geological classification labels, soil classification labels, hardness/softness labels, and void labels to the test construction data.

Furthermore, training data should be created according to the type of test construction data acquired. For example, in cases where only construction machinery data is acquired as test construction data, training data should be created by assigning ground property labels to that construction machinery data. In cases where construction machinery data, camera image data, spectral image data, and borehole image data are acquired as test construction data, training data should be created by assigning ground property labels to each of the construction machinery data, camera image data, spectral image data, and borehole image data. Alternatively, in cases where construction machinery data, camera image data, and spectral image data are acquired as test construction data, training data can also be created by assigning ground property labels to each dataset consisting of construction machinery data, camera image data, and spectral image data.

Once training data is created, the trained model generation means 101 generates a trained model by training on this training data using machine learning techniques such as deep learning (Step 217 in Figure 7).

After conducting a test construction and generating a trained model, the main construction will be carried out. For the main construction, as shown in Figure 3, it is advisable to place the central device in the management area and terminal equipment near the construction area. Once the central device and terminal equipment are in place and other necessary temporary construction and preparatory work is completed, drilling to form the main borehole will be carried out as shown in Figure 8 (Step 221 in Figure 8). At this time, it is advisable to use the same drilling machinery as that used in the test construction.

During drilling of the main borehole, construction data similar to that obtained in the test construction (main construction data), i.e., construction data of the same type as the training data, is acquired. For example, in Figure 8, construction machine data is acquired by the construction machine data acquisition means 103 as drilling progresses (Step 222 in Figure 8), and camera image data and spectral image data are acquired each time slime is obtained (Step 223 in Figure 8). In this example, in addition to this main construction data, measured values such as the position of the drill bit and the insertion angle of the rod are acquired as drilling progresses, meaning that the actual shape of the main borehole (drilled hole shape) is understood according to the progress of drilling (Step 224 in Figure 8).

Once the construction data is acquired, it is input to the ground condition determination means 102, which then outputs the ground conditions related to the construction data (Step 225 in Figure 8). At this time, since construction data of the same type as the training data has been acquired, construction data of the same type as the training data is input to the ground condition determination means 102. For example, in a case where training data with ground condition labels attached to construction machine data, camera image data, and spectral image data has been machine-learned, the construction machine data, camera image data, and spectral image data are input to the ground condition determination means 102.

Furthermore, the ground property determination means 102 naturally outputs the same type of ground property as the training data used to generate the trained model. For example, in the case where training data with geological classification labels and soil classification labels has been machine-learned, the ground property determination means 102 outputs the geological classification and soil classification as ground properties. In the case where training data with geological classification labels, soil classification labels, hardness/softness labels, and cavity labels has been machine-learned, the ground property determination means 102 outputs the geological classification, soil classification, ground hardness/softness, and cavity status as ground properties. Note that, for example, if training data is created by assigning ground property labels to construction machine data, camera image data, and spectral image data, the ground properties will be output for each of the input construction machine data, camera image data, and spectral image data. On the other hand, if training data is created by assigning ground property labels to a dataset consisting of construction machine data, camera image data, and spectral image data, the ground properties will be output for each dataset.

As shown in Figure 3, when the system is configured with terminal equipment and a central unit, it is also possible to configure it so that construction data is input using the terminal equipment, and the ground condition determination means 102 of the central unit outputs the ground condition. That is, when construction data is input to terminal equipment located near the construction team, that construction data is transmitted to the central unit transmission/reception means 107 by the terminal-side transmission/reception means 106, and the ground condition determination means 102 of the central unit receives that construction data and outputs the ground condition. In this case, the ground condition output by the ground condition determination means 102 should be transmitted to the terminal equipment and displayed on its output means 105.

When the ground properties are output by the ground properties determination means 102, the output control means 104 displays the borehole in 3D (or 2D) on the output means 105 as shown in Figure 5 (Step 226 in Figure 8). Alternatively, as shown in Figure 6, the shape of the planned borehole DP (drilling shape) and the shape of the completed borehole DA can be displayed together, and the type of ground properties can be indicated using color, shade, and pattern .

The ground property determination system of the present invention can be used in various construction projects involving boring, such as ground anchor work as a countermeasure against slopes, tunnel construction, and ground improvement work by deep mixing. According to the present invention, high-precision information-based construction is possible, and as a result, it is possible to construct high-quality social infrastructure such as tunnel structures and foundation grounds, and to implement appropriate disaster countermeasures on slopes, etc. Therefore, this invention is not only industrially applicable but also has the potential to make a significant contribution to society.

100 Ground condition determination system of the present invention 101 Means for generating a trained model (of the ground condition determination system) 102 Means for determining ground conditions (of the ground condition determination system) 103 Means for acquiring construction machine data (of the ground condition determination system) 104 Means for controlling the output (of the ground condition determination system) 105 Means for outputting (of the ground condition determination system) 106 Means for transmitting and receiving data from the terminal side (of the ground condition determination system) 107 Means for transmitting and receiving data from the central side (of the ground condition determination system) 108 Means for storing a trained model (of the ground condition determination system) 109 Means for storing test boring data (of the ground condition determination system) 110 Means for storing test construction data (of the ground condition determination system) 111 Means for storing actual construction data (of the ground condition determination system) DA Completed borehole DP Planned borehole

Claims (5)

A system for determining the properties of the target ground to be drilled or excavated,
A means for generating a trained model that generates a trained model by machine learning on training data that has geological classification labels and/or soil classification labels attached,
The system includes a ground property determination means that determines the geological classification and/or soil classification of the target ground by inputting construction data acquired during drilling or excavation into the trained model,
The aforementioned training data is generated by attaching the geological classification labels and/or soil classification labels to a dataset consisting of construction machine data related to construction obtained from a drilling machine or excavating machine, camera image data obtained by photographing ground slime obtained by drilling or excavating, and spectral image data obtained by acquiring the ground slime with a spectral camera.
The construction data input to the trained model is the construction machine data, the camera image data, and the spectral image data , and is of the same type as the training data used in machine learning.
A ground condition determination system characterized by the following features. The aforementioned camera image data is image data obtained by photographing the ground slime together with a color sample.
The spectral image data is image data obtained by photographing the ground slime together with a color sample.
The ground condition determination system according to claim 1. The pre-trained model generation means generates the pre-trained model by machine learning the training data to which the hardness and softness of the ground are labeled.
When the construction data is input to the learned model, the ground condition determination means determines the hardness or softness of the target ground.
A ground condition determination system according to claim 1 or 2, characterized by the features described above. One or more terminal devices,
The system further comprises a central device capable of communicating with the aforementioned terminal devices,
The construction data entered using the terminal device is transmitted to the central device.
The central device receives the construction data and inputs the construction data into the ground condition determination means to determine the geological classification and/or soil classification of the target ground.
The terminal device outputs the results determined by the ground condition determination means in real time.
Furthermore, the terminal equipment can overlay the actual completed drilling or excavation results as construction progresses with the planned drilling or excavation shape in a two-dimensional or three-dimensional display.
A ground condition determination system according to claim 1 or 2, characterized by the features described above. A method for determining the properties of the target ground for drilling ground anchors, using the ground property determination system described in claim 1,
The process of determining the geological classification and/or soil classification of the target ground by inputting the construction data acquired during drilling of the ground anchor into the trained model,
The system includes a step of determining whether to use a shorter or longer drilling length than planned by understanding the position and physical properties of the anchoring layer in real time during ground anchor drilling.
A method for determining ground properties, characterized by the features described above.