Soil testing method to predict ground improvement quality and ground improvement state prediction method

Abstract

To provide a soil testing method that can improve predicting accuracy of possibility of poor mixing of target soil with a solidification material or quality of an improved soil.SOLUTION: A first test is carried out, which is at least one of an adhesion soil amount test that measures an amount of target soil adhering to a mixer's stirring blade and an adhesion strength test that measures adhesion strength when target soil adhering to steel or glass material peels off. A second test consisting of a vane shear test on the target soil and a fall cone test on the target soil will be conducted. The results of the first and second tests are combined to predict the ground improvement state corresponding to the target soil.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a soil testing method for a ground improvement method using a solidification material. [Background technology]

[0002] Mechanical mixing is a ground improvement method that is used to improve soft ground for the construction of large-scale civil engineering facilities. The biggest challenge in ground improvement methods is how to ensure consistent improvement strength, and uniform mixing is confirmed based on construction records as indirect data, and strength and core information from check borings after construction as verification data (collection rate: approximately 90% for clayey soils, plus consolidation rate and RQD) (see Non-Patent Document 1).

[0003] It is known that these poor mixing conditions are generally more likely to occur when improving highly plastic clay. To prevent poor mixing, for example, single-shaft mechanical mixing methods are equipped with anti-rotation blades that mechanically prevent co-rotation. Meanwhile, a method has been proposed for predicting co-rotation conditions during preliminary mix testing using a relationship between the number of drops in vane shear tests and liquid limit tests (see Patent Document 1). Other methods have been proposed, such as using neural networks to learn construction data and finished form data (see Patent Document 2). A method has also been published that estimates the extent of poor mixing based on the original soil properties and moisture content, using tests to examine the moisture content of the target soil and the adhesion properties of model mixing blades to determine the conditions for poor mixing (see Patent Document 3). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 6648332 [Patent Document 2] Patent Publication No. 7269137 [Patent Document 3] Japanese Patent Publication No. 2022-126955 [Non-patent literature]

[0005] [Non-Patent Document 1] Revised edition: Guidelines for design and quality control of improved ground for buildings - Deep and shallow mixed treatment method using cement-based solidification materials -, The Building Center of Japan. pp.175-188, 2002 [Non-patent document 2] Satoshi Omine and Hidetoshi Ochiai, Determination of Liquid Limit Based on Soil Adhesion, 49th Annual Conference of the Japan Society of Civil Engineers, III-51, pp.96-97, 1994 [Non-patent document 3] Geotechnical Society Standard, JGS 0142-2020, Liquid Limit Test Method Using Fall Cone [Non-patent document 4] Nobutaka Nunoya, Takashi Tsuchida, and Hiroki Abe, Flow characteristics of marine clay at high water content, Journal of the Japan Society of Civil Engineers, B2, Vol. 68, No. 2, I_551-555, 2012 [Non-Patent Document 5] Jin-jun Zhang, Mie University Graduate School of Bioresource Science, Dissertation, September 2013, pp.43-57 Summary of the Invention [Problem to be solved by the invention]

[0006] The quality of the soil to be improved varies widely, and based on the accumulated data from preliminary soil tests and on-site quality check results, it can be inferred that poor mixing occurs due to the following mechanism.

[0007] The target soil adheres to the impeller, forming lumps that rotate together with the impeller, resulting in a "co-rotation phenomenon." This prevents the target soil from being uniformly mixed with the solidification material slurry, resulting in poor mixing (primarily due to the high adhesion of the target soil to the impeller). If the target soil has high shear strength, the impeller will not break up the lumps of soil, but will instead cut them in chunks, filling the gaps around them with the solidification material slurry, resulting in poor mixing (primarily due to the high shear strength of the soil). If the target soil is marine clay and / or volcanic ash clayey soil, the reaction between the solidification material and salts or amorphous clay minerals in the soil will likely cause a sudden increase in the viscosity of the soil, leading to co-rotation or lumpy cutting (primarily due to the increase in viscosity of the soil due to the reaction between the soil and the solidification material).

[0008] However, at the testing stage, it is not possible to accurately predict the possibility of co-rotation occurring, and ultimately the risk of poor mixing between the target soil and the solidification material.

[0009] Therefore, an object of the present invention is to provide a soil testing method that can improve the accuracy of predicting the possibility of poor mixing of the target soil and solidification material or the quality of the improved soil. [Means for solving the problem]

[0010] The soil testing method of the present invention includes: a first test step of conducting a first test which is at least one of an adhesion soil amount test for measuring the amount of the target soil adhering to the mixing blades of the mixer and an adhesion strength test for measuring the adhesion strength when the target soil adhering to the steel material or glass material peels off; A second test step of conducting a second test consisting of a vane shear test on the target soil and a fall cone test on the target soil; a prediction step of predicting the ground improvement form corresponding to the target soil by combining the first test result obtained by implementing the first test step and the second test result obtained by implementing the second test step; Contains:

[0011] "Ground improvement form" includes the risk of poor mixing of the target soil with the solidification material and / or the quality of the improved soil obtained after mixing the target soil with the solidification material.

[0012] In the soil testing method, a testing step of carrying out the first test and the second test for each of three or more levels of primary solidification treated soil with different amounts of solidification material added to the target soil while keeping the solidification material water ratio constant; Based on the test results obtained by implementing the test process, if there is a high risk of poor mixing of the solidified treated soil (if the predicted result of the ground improvement form is poor), the first test, the second test and the prediction process will be carried out on the secondary solidified treated soil obtained by adding a solidification material adjusted to an appropriate mix to the target soil in order to reduce the risk of poor mixing (improve the ground improvement form) and achieve the indoor mix target strength. It is preferable.

[0013] The ground improvement shape prediction method of the present invention comprises: A process of generating, by machine learning, a ground improvement form prediction model that expresses the correlation between the first test results and the second test results of the target soil and the solidification-treated soil obtained by adding a solidification material to the target soil, respectively, and the ground improvement form, using multiple sets of learning data including first test results obtained by conducting at least one of an adhesion volume test and an adhesion strength test, and second test results obtained by conducting a vane shear test and a fall cone test, for the target soil and the solidification-treated soil obtained by adding a solidification material to the target soil; The method includes a step of inputting prediction input data, including the first test results and the second test results for the new target soil and the solidification treated soil, into the ground improvement form prediction model, and predicting the ground improvement form of the new target soil as prediction output data from the ground improvement form prediction model.

[0014] In the ground improvement shape prediction method, a mix determination process for determining an appropriate mix to reduce the risk of poor mixing of the solidified soil and achieve the indoor target mix strength based on the first test results and the second test results for each of three or more levels of primary solidification treated soil with different amounts of solidification material added to the target soil while keeping the solidification material water ratio constant; and a solidification-treated soil testing step for obtaining the first test results and the second test results constituting the plurality of sets of learning data, using the secondary solidification-treated soil obtained by adding the solidification material adjusted to the appropriate mix determined in the mix determination step to the target soil. It is preferable. [Brief explanation of the drawings]

[0015] [Figure 1] 1 is a diagram illustrating the configuration of a ground improvement shape prediction device according to one embodiment of the present invention. [Figure 2] 1 is a flowchart showing the procedure of a soil testing method according to one embodiment of the present invention. [Figure 3] 1 is a flowchart showing the steps of a ground improvement shape prediction method according to one embodiment of the present invention. [Figure 4] An explanatory diagram of the relationship between moisture content and the proportion of adhering soil. [Figure 5] FIG. 1 is an explanatory diagram of the relationship between water content and vane shear strength. [Figure 6] An explanatory diagram of the relationship between moisture content and fall cone penetration. [Figure 7] An explanatory diagram showing the relationship between the amount of solidification material added and unconfined compressive strength. [Figure 8] An explanatory diagram of the relationship between C / Wt and uniaxial compressive strength. [Figure 9] An explanatory diagram of the test results for the ratio of adhering soil volume AS according to the liquid index. [Figure 10] FIG. 1 is an explanatory diagram showing test results of adhesive strength AD according to liquid index. [Figure 11] An explanatory diagram of the test results of vane shear strength BS according to liquid index. [Figure 12] An explanatory diagram showing the test results of fall cone penetration CP according to liquid index. [Figure 13] An explanatory diagram showing the test results of the proportion of adhering soil depending on the natural water content and liquid index of the first clayey soil. [Figure 14] An explanatory diagram showing the test results of the proportion of adhering soil depending on the natural water content and liquid index of the second clayey soil. DETAILED DESCRIPTION OF THE INVENTION

[0016] (Configuration of the device for assessing the risk of poor mixing) The ground improvement form prediction device 10 according to one embodiment of the present invention shown in FIG. 1 includes an input interface 11, a ground improvement form prediction model generation unit 12, a ground improvement form prediction unit 13, and an output interface 14. The ground improvement form prediction device 10 is configured with a computer such as a server computer (including a cloud server), a personal computer, a smartphone, or a tablet computer. Each component of the ground improvement form prediction device 10 is configured with an arithmetic processing unit (comprised of a CPU and a processor (core)), a storage device (comprised of memory (ROM, RAM, etc.)), an I / O circuit, etc. The components of the ground improvement form prediction device 10 have the function of reading the necessary programs (software) and data from the storage device and performing specified arithmetic processing in accordance with the programs based on the data.

[0017] The input interface 11 includes a training input data acquisition unit 111, a training output data acquisition unit 112, and a prediction input data acquisition unit 113. The input interface 11 may include touch panel buttons, a keyboard, and / or a voice input device.

[0018] The learning input data acquisition unit 111 is configured to acquire learning input data. The "learning input data" includes the water content of the target soil (raw soil), the first and second test results, the type, amount, and water-solidification ratio of the solidification material added to the solidification-treated soil, the type and amount of admixture added, and the first and second test results of the solidification-treated soil. The "learning input data" may also include the type and specifications of the ground improvement method and the specifications of the mixing blade. The learning output data acquisition unit 112 is configured to acquire learning output data. The "learning output data" includes a time series of the rotation speed of the co-rotation prevention blade and / or boring core data (core collection rate and solidification rate) representing the quality of the improved soil, depending on the type and specifications of the ground improvement method and the specifications of the mixing blade.

[0019] The prediction input data acquisition unit 113 is configured to acquire prediction input data. The "prediction input data" includes the water content of the target soil (raw soil) to be worked on, the first test results and the second test results, the type, amount, and water-to-solidification agent ratio of the solidification-treated soil added to the solidification-treated soil, the type and amount of admixture added, and the first test results and the second test results of the solidification-treated soil.

[0020] The ground improvement form prediction model generation unit 12 generates a trained model by machine learning the correlation between the training input data and the ground improvement form based on a plurality of training data including at least the training input data, using a training model with "supervised learning," "semi-supervised learning," or "unsupervised learning." The training model is stored and held in a storage device, or is read from a database and then stored in a storage device, and is then read from the storage device.

[0021] The ground improvement form prediction unit 13 is configured to input the prediction input data acquired by the prediction input data acquisition unit 113 into the trained model generated by the ground improvement form prediction model generation unit 12, thereby outputting a predicted result of the ground improvement form. The ground improvement form is defined by the risk of poor mixing between the target soil and the solidification material and / or the quality of the improved body obtained by mixing the target soil and the solidification material. The risk of poor mixing may be set by setting a risk value based on the results of the first and second tests depending on the type and specifications of the ground improvement method and the specifications of the mixing impeller, or may be defined by measuring the resistivity of the ground, evaluating by sounding, or by color changes using an underground camera. In addition, in the case of an improvement method using a single-shaft machine, the risk of poor mixing may be defined by the time series of the rotation speed of the anti-co-rotation impeller. The quality of the improved body may be defined by boring core data (core recovery rate, solidification rate).

[0022] The output interface 14 is configured to include an image output device, or an image output device and an audio output device, and is configured to output the predicted results of the ground improvement form output by the ground improvement form prediction unit 13.

[0023] (Target soil test) It was known that when the target soil (raw soil) has high shear strength, the soil lumps cannot be broken down during the mixing process in mechanical mixing methods, which can easily result in poor mixing with the solidification material slurry.In addition, the fact that the adhesion characteristics of the target soil to the mixing blades are important as a factor that can easily cause poor mixing between the target soil and solidification material slurry was made clear by using a model device that models the mixing mechanism to measure the amount of soil that adheres to the mixing blades as they penetrate and are pulled out of the target soil.

[0024] According to a soil testing method in one embodiment of the present invention, first, a first test and a second test are conducted on the target soil (raw soil) (FIG. 2 / STEP 10). Specifically, in order to simultaneously evaluate poor mixing of the target soil and the solidification material due to the two mechanisms, a small soil mixer (such as a commonly used Hobart mixer) used in blending tests is used to mix the target soil for a predetermined period of time. Either the first test or the second test may be conducted first, or they may be conducted simultaneously.

[0025] The "first test" is at least one of the following tests: an "adhered soil amount test," in which the ratio of the amount of target soil adhering to a mixing blade (a pre-selected hook type or beater type) to the total amount of soil is measured, and an "adhesion strength test" (see non-patent document 2), in which the adhesion strength of the target soil to a steel plate or glass plate is measured.

[0026] The "second test" consists of a "vane shear test" to determine the shear strength of the target soil and a "fall cone penetration test," which is a type of consistency test conducted to investigate the liquid limit of the target soil (see Non-Patent Document 3). Alternatively, the liquid limit test (see JIS A1205) may be used as the consistency test.

[0027] For the soil adhesion test, a Hobart mixer impeller shape suitable for mixing is selected, taking into consideration the impeller shape used in the actual construction method. For the adhesion strength test, a test method using a direct shear tester to determine the adhesion force and / or friction coefficient to steel (see Non-Patent Document 5) or a test method using a tensile tester may be adopted. Vane shear strength can also be determined as the convergence value in the high shear strain rate region of the flow curve (see Non-Patent Document 4). Consistency is generally referred to as resistance to deformation or fluidity. Vane shear strength corresponds to the convergence value in the high shear strain rate region of the flow curve and is the deformation resistance at that time. Fall cone penetration corresponds to the yield value when the strain rate is close to zero and is the flow resistance. Therefore, in the second testing step, consistency evaluation was performed using both tests.

[0028] Next, the first and second tests are conducted on the target soil that has been hydrated to three or more levels (Figure 2 / STEP 11). Based on the test results, the amount of water that will be used as a guideline for the degree of poor mixing when adding the solidification material is determined.

[0029] (Setting mix specifications for solidification-treated soil tests) The mix specifications for the solidification-treated soil test, i.e., the indoor mix target strength (q ul ) is set (Figure 2 / STEP 12). The purpose of this is to determine the indoor mix target strength (q ul The objective of this study is to obtain solidification material mix specifications that have a low risk of poor mixing and that can verify the effects of reactions between the raw soil and solidification material slurry, and to plan and verify countermeasure specifications in cases where poor mixing is a concern.

[0030] (1) Method for selecting the solidification material mix level with W / C setting For the target soil (raw soil), the solidification material slurry w / c is set to 1.0, and the mixing amount according to the soil properties is 1 m 3 100, 200, 300 kg / m 3 The solidification material and water equivalent to 1 m are mixed in a mixer to prepare a specimen for the unconfined compressive strength test. If the test results indicate that the soil is likely to be poorly mixed, for example, the mixture amount is 1 m 3 150, 250, 350 kg / m 3 The solidification material and water equivalent to the above may be mixed in a mixer to prepare a specimen for the unconfined compressive strength test.

[0031] During this process, the risk of poor mixing is assessed by conducting (1A) soil adhesion test and / or (1B) adhesion strength test, (2A) vane shear test, and (2B) fall cone penetration test on the solidified soil (primary solidified soil) (Figure 2 / STEP 20).

[0032] If the risk of poor mixing is assessed as high, a mix test will be conducted according to the test level of specifications that pose a low risk of poor mixing. The solidified soil (secondary solidified soil) obtained by adding solidification material slurry to the target soil will be subjected to (1A) soil adhesion volume test and / or (1B) adhesion strength test, (2A) fall cone penetration test, and (2B) vane shear strength test (Figure 2 / STEP 20).

[0033] Based on the results of the mix test and the investigation of the test construction, the proportion of adhering soil AS (%), the adhesive strength AD (kN / m 2 ), vane sheath strength BS (kN / m 2 For each of the CP (mm) and fall cone penetration, the risk of poor mixing was classified into three levels (low, medium, and high) as shown in Table 1. The risk of poor mixing may also be classified into two or four or more levels. The criteria for determining the risk of poor mixing (AS1, AS2, AD1, AD2, BS1, BS2, CP1, and CP2) are affected by the type of construction method and / or the shape of the impeller, and may be changed as appropriate depending on the construction conditions. The risk values listed in Table 1 are for cases where a beater-type impeller and a 10-liter mixing vessel are used.

[0034] [Table 1]

[0035] When improvements are made using an actual mixing machine in test construction or actual construction, the occurrence of poor mixing is judged based on the rotation data monitoring the rotation status of the anti-rotation blades in the single-shaft mechanical mixing method and / or the results of confirmation using boring samples.Then, the values of the adhered soil volume ratio AS, adhered strength AD, vane shear strength BS, and fall cone penetration CP when poor mixing occurs are defined as the warning adhered soil volume ratio AS2, warning adhered strength AD2, warning vane shear strength BS2, and warning fall cone penetration CP2, respectively.

[0036] Furthermore, after the assumed countermeasures (increasing the amount of water and / or slurry and / or adding admixtures) are implemented, the rotation data and / or verification results are used to confirm whether the risk of poor mixing has been reduced. The values of the adhering soil volume ratio AS, adhering strength AD, vane shear strength BS, and fall cone penetration CP when the risk of poor mixing is reduced are defined as the allowable adhering soil volume ratio AS1, allowable adhering strength AD1, allowable vane shear strength BS1, and allowable fall cone penetration CP1. These allowable values, the initial classification of the risk of poor mixing, and the type of construction method and / or shape of the mixing blades are associated and cumulatively stored in a storage device. The criterion value for determining the risk of poor mixing is then updated, and by adopting this criterion value, the accuracy of predicting the risk of poor mixing is improved.

[0037] (Risk assessment of poor mixing due to the proportion of soil adhering) Based on the results of the solidification-treated soil test, the risk of poor mixing of the solidification-treated soil is evaluated (Figure 2 / STEP 21).

[0038] Figures 4 to 6 show the results of the target soil (original soil) with three selected moisture content levels and the results of the soil at 100 kg / m under the condition of w / c -1.0. 3 , 200 kg / m 3 , 300 kg / m 3 This plots the relationship between the moisture content and characteristic values of the added solidified soil. As shown in Figure 4, the test results are plotted on a two-dimensional coordinate system with the moisture content on the horizontal axis and the percentage of adhering soil on the vertical axis. The moisture content of solidified soil is difficult to measure because hydrates are formed over time, and the calculated apparent moisture content can be understood by following the relationship (1).

[0039] Apparent water content of solidified soil = w n +k·C / 1000ρ t (1+w n )‥(1).

[0040] For example, if the wet density is ρ t (g / cm3 ) and the water content w n {(%) / 100} of saturated raw soil 1m 3 When a slurry with a water-to-solidification material ratio of k = (w / c) is added, C (kg) of solidification material is added. Therefore, the increase in water volume due to the addition of the slurry is k·C (kg), and the apparent increase in water content Δw is Δw = k·C / 1000ρ, ignoring the amount of solidification material. t (1+w n )

[0041] As shown in Figure 4, the amount of water brought in usually increases as the amount of solidification material increases, so the percentage of adhering soil is often low. When the risk of poor mixing is evaluated based on the provisional evaluation criteria, AS≦30 is considered a low risk, so the moisture content of the target soil must be 82% or more.

[0042] (Risk assessment of poor mixing due to vane shear strength) For example, as shown in Figure 5, the test results are plotted on a two-dimensional coordinate system with the water content on the horizontal axis and the vane shear strength on the vertical axis. The water content of the solidified soil is calculated in the same way as described above (assessment of the risk of poor mixing related to the proportion of adhering soil).

[0043] From Figure 5, when the risk of poor mixing is evaluated based on the provisional evaluation criteria, a BS of 4 or less is considered a low risk, so the moisture content of the target soil must be 81% or more.

[0044] (Assessing the risk of poor mixing based on fall cone penetration) The fluidity of the target soil and the solidification-treated soil is evaluated by the fall cone penetration. When the risk of poor mixing is evaluated based on the provisional evaluation criteria from Figure 6, a CP of 10 or less is considered a low risk, so the water content of the target soil must be 85% or more.

[0045] However, Kanto loam or marine clay containing a lot of salts, etc., begin to react with the solidifying agent early on, and conversely, the vane shear strength may increase over time. Therefore, in the case of special soil types, it is necessary to pay attention to the relationship with the elapsed time after mixing the solidifying agent.

[0046] Increasing (adjusting) the water ratio (W / C) generally reduces the risk of poor mixing. However, if the risk of poor mixing is still high, the addition of a separate admixture with a dispersing effect, such as a naphthalene sulfonic acid or polycarboxylic acid admixture, is considered, and the mix level is added again to reduce the risk of poor mixing, and then the mixture is evaluated.

[0047] From the relationship between the moisture content and physical properties of the target soil (raw soil) in Figures 4 to 6, for example, according to the risk classification of poor mixing in Table 1, the minimum moisture content to determine the risk as "low" is 64% for the adhering soil volume ratio, 66% for the vane shear strength, and 73% or more for the fall cone penetration. However, in the case of solidified treated soil, it is recognized that the risk of poor mixing cannot be reduced unless the moisture content is further increased. According to the same risk classification for the solidified treated soil in Figures 4 to 6, the solidification material is set to w / c = 1.0 and the amount of solidification material added is 100, 200, or 300 kg / m 3 This is an example of a case where the adhering soil volume ratio was 82%, the vane shear strength was 81%, and the fall cone penetration was 85% or more. For the target soil, the moisture content of the solidified soil had to be set 12 to 18% higher than that of the original soil. As such, the moisture content that should be set for the solidified soil will vary depending on the soil quality of the target soil (original soil) and the type of solidification material, but it is possible to roughly predict it by accumulating data.

[0048] It is determined whether the risk of poor mixing of the solidified treated soil is low (Figure 2 / STEP 22). If the result is negative (Figure 2 / STEP 22...NO), the mix specifications for the solidified treated soil test are reset, including increasing the amount of solidifying agent and water (Figure 2 / STEP 12). If the result is positive (Figure 2 / STEP 22...YES), the 28-day compressive strength of the test specimen derived from the solidified treated soil is measured (Figure 2 / STEP 40). In addition to the 28-day compressive strength, 7-day compressive strength, 90-day compressive strength, etc. may also be measured as compressive strength.

[0049] It is determined whether the 28-day compressive strength of the specimen is equal to or greater than the target strength for the indoor mix (Figure 2 / STEP 42). The amount of solidification agent to be added that satisfies the target strength for the indoor mix is determined, for example, from the relationship diagram of unconfined compressive strength vs. amount of solidification agent added, as shown in Figure 7. If the determination result is negative (Figure 2 / STEP 42...NO), the mix specifications for the solidification-treated soil test are reset, including changes to the amount of solidification agent and water (Figure 2 / STEP 12). If the determination result is positive (Figure 2 / STEP 42...YES), the process ends.

[0050] For example, if the target indoor strength is 1500kN / m 2 If w / c=1.0, the minimum is approximately 200 kg / m 3 As shown in Figures 4 to 6, 200 kg / m 3 In this case, it is estimated that the risk of poor mixing is near the border between low and medium. If we take the safe side in terms of the risk of poor mixing, the amount of solidification material added should be 220 kg / m 3 It is advisable to take measures such as setting w / c=1.0.

[0051] In other words, to determine the amount of water (w / c) to be added for the three levels of solidification material determined based on the target soil properties, tests are usually conducted with w / c = 1.0, and the physical properties of the first and second test processes are determined.The w / c is then selected from 1.5, 2.0, etc. (0.67, 0.5, etc. for c / w) as necessary to achieve a mix that satisfies both the strength and risk of poor mixing.

[0052] (2) Method for selecting the solidification material mix level based on the soil moisture content and solidification material water ratio calculation If the strength of the test specimen has been measured, the water-to-solidification ratio of the treated soil (including the water content in the soil) that satisfies the target strength in the laboratory is selected. For example, if the target strength in the laboratory is 1500 kN / m 2 If so, C / W according to Figure 8 t =22%.

[0053] The amount of solidification material to be added, C, to satisfy the target strength for the indoor mix for any water-cement ratio, k, in the mix test can be calculated according to the relation (2) for saturated soil.

[0054] C / W t =C / {1000ρ t ·w n / (1+w n )+k·C} ‥(2). Here, "C" is the target soil 1m 3 Amount of solidification material added per unit (kg / m 3 ) is "W t " is the total mass (kg) of water in the target soil and the solidification material slurry. n " is the natural moisture content of the target soil (%) / 100. t " is the wet density of the target soil (g / cm 3 )=10,00ρ t (kg / m 3 ) where "k" is the water-cement ratio (w / c) of the solidification material slurry. Based on the amount of solidifying agent added for an arbitrary water-cement ratio k, the amount of solidifying agent is set to at least three levels as in the above "(1) Method for selecting the solidifying agent mixture level with W / C setting," or the amount of solidifying agent is set so that there are at least three levels of water-to-solidifying ratio, and the corresponding amount of solidifying agent is calculated. Then, a specimen for the unconfined compressive strength test is prepared.

[0055] During this process, the risk of poor mixing is assessed by conducting (1A) soil adhesion volume test and / or (1B) adhesion strength test, (2A) vane shear test and (2B) fall cone penetration test on the solidified soil (primary solidified soil).

[0056] (Ground improvement shape prediction method (first embodiment)) In the first embodiment, the learning model learns the correlation between the learning input data and the learning output data through "supervised learning." To this end, a plurality of pieces of learning data are acquired by each of the learning input data acquisition unit 111 and the learning output data acquisition unit 112 (FIG. 3 / STEP 60). That is, each of the plurality of pieces of learning data includes the learning input data and the corresponding learning output data. A plurality of sets of learning data are stored in a storage device and / or a database. The number of pieces of learning data is appropriately set in consideration of the inference accuracy required for the learning model.

[0057] The ground improvement form prediction model generation unit 12 generates a ground improvement form prediction model (trained model) by having the learning model learn the correlation between the input data and the ground improvement form (risk of poor mixing and / or quality of the improved body) based on multiple learning data (combination of input data and corresponding output data) (Figure 3 / STEP 61).

[0058] When "supervised learning" is performed, for example, a neural network model is used as the learning model to generate a trained model. According to the neural network model, training input data included in the training data is input to the input layer, and model output data output from the output layer as a prediction result or inference result is compared with training output data (teacher data) included in the training data, thereby learning the correlation between the training input data and the training output data. The ground improvement shape prediction model generation unit 12 determines whether or not machine learning needs to continue based on at least one of the error between the model output data and the teacher data, the number of times learning has been performed, and the remaining amount of untrained training data. When it is determined that continuation of machine learning is unnecessary, a trained model is established.

[0059] Then, the prediction input data acquisition unit 113 acquires prediction input data containing the first test results and the second test results of the target soil and the solidification treated soil, whose ground improvement form is unknown (Figure 3 / STEP 62).

[0060] Next, the ground improvement form prediction unit 13 inputs the prediction input data into the input layer of the ground improvement form prediction model, and obtains the predicted result of the ground improvement form of the target soil as model output data output from the output layer of the ground improvement form prediction model (Figure 3 / STEP 63).

[0061] Then, the predicted results of the ground improvement form of the solidification-treated soil are output as the model output data through the output interface 14 (FIG. 3 / STEP 64).

[0062] (Ground improvement shape prediction method (second embodiment)) The second embodiment differs from the first embodiment in that "unsupervised learning" is adopted as the machine learning method instead of "supervised learning." The following mainly describes the differences from the first embodiment.

[0063] A plurality of pieces of learning input data are acquired by the learning input data acquisition unit 111 as a plurality of pieces of learning data (FIG. 3 / STEP 60). That is, the learning data acquisition unit is not configured with both the learning input data acquisition unit 111 and the learning output data acquisition unit 112, but is configured with only the learning input data acquisition unit 111.

[0064] The ground improvement form prediction model generation unit 12 inputs learning data into the learning model, and the unsupervised learning model learns the correlation between the learning input data contained in the learning data and data indicating that the ground improvement form is normal (no poor mixing has occurred and / or the quality of the improved body is good), and a ground improvement form prediction model (trained model) is generated (Figure 3 / STEP 61).

[0065] For example, an autoencoder (AE) model is employed as the learning model. According to the autoencoder model, learning input data relating to target soil and solidified soil without poor mixing (normal) is input to the input layer as learning data. Next, in the middle layer, scores are assigned (weighted) according to the importance of the data, and data with low scores are eliminated (encoded). Weighting is then applied again when moving to the output layer, and the sum of the data received from multiple edges is output (decoded) as the final value. This procedure learns patterns and / or trends in the learning input data so that the model output data matches the learning input data. This series of steps is repeated, and when specified conditions are met, learning by the learning model ends, and a trained model is generated.

[0066] As the autoencoder (AE), a stacked autoencoder (SAE), a convolutional autoencoder (CAE), a variational autoencoder (VAE) or a conditional autoencoder may be employed.

[0067] Then, the prediction input data acquisition unit 113 acquires prediction input data containing the first test results and the second test results of the target soil and the solidification treated soil, whose ground improvement form is unknown (Figure 3 / STEP 62).

[0068] Next, the ground improvement form prediction unit 13 inputs the prediction input data into the input layer of the ground improvement form prediction model, and obtains the predicted result of the ground improvement form of the solidified treated soil as model output data output from the output layer of the ground improvement form prediction model (Fig. 3 / STEP 63). The deviation between the prediction input data input into the ground improvement form prediction model or the feature values based thereon and the model output data output from the ground improvement form prediction model or the feature values based thereon is calculated. Furthermore, if the deviation is less than a threshold, the designated function is judged to be "1: Normal (risk of poor mixing is low and the quality of the improved soil is good)," and if the deviation is equal to or greater than the threshold, the designated function is judged to be "2: Abnormal (risk of poor mixing is not low or high and the quality of the improved soil is poor)."

[0069] Then, the predicted results of the ground improvement form of the target soil are output as the model output data through the output interface 14 (FIG. 3 / STEP 64).

[0070] (Other embodiments) In the above-described embodiments, a neural network (first embodiment) or an autoencoder (second embodiment) is used as the learning method, but any other machine learning method may be used. For example, tree-type models such as decision trees and regression trees, ensemble learning such as bagging and boosting, neural network-type models (including deep learning) such as recurrent neural networks and convolutional neural networks, clustering-type models such as hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, and k-means, multivariate analysis such as principal component analysis (PCA), factor analysis, and logistic regression, and support vector machines (SVM) may be used as learning models. In addition to supervised learning and unsupervised learning, semi-supervised learning may also be used.

[0071] Example 1 Figures 9 to 12 show the test results for tuffaceous clayey soil, with respect to the liquidity index, for the proportion of adhered soil AS, adhesive strength AD, vane shear strength BS, and fall cone penetration CP. The solid lines in the figures show the liquidity index at the natural moisture content and the corresponding measured values. Black circles (●) represent test results when the moisture content of the target soil (original soil) was changed, and white circles (○) represent test results when the amount of solidification material added was 100, 200, and 300 kg / m3 for the original soil with a natural moisture content. 3 The figures show the test results for solidified soil with three levels of W / C of the solidification material slurry: 0.8, 1.0. The dashed line in the figure indicates the value of 300 kg / m 3 , and for the case of w / c=0.8, the measured values are shown relative to the corresponding liquid index.

[0072] As is clear from Figure 11, as the liquid index IL (= (natural water content wn - plastic limit wp) / plastic index Ip) increases, the vane shear strength BS tends to decrease. As is clear from Figure 12, as the liquid index IL increases, the fall cone penetration CP tends to increase. On the other hand, as is clear from Figure 9, the adhered soil volume ratio does not necessarily have a constant trend, and may show a maximum value (maximum value) at a certain liquid index. As is clear from Figure 10, the adhesion strength may also show a maximum value (maximum value) at a certain liquid index. For this reason, it cannot be said that an increase in the liquid index IL will always result in a good mixture of the target soil and solidification material.

[0073] The test results for the original soil with a liquid index (IL) of 0.42 under natural water content conditions were: (1A) Adherence test: Adherence ratio of soil: 0.53 (= 53%); (1B) Adhesion strength: 18.9 kN / m 2 ], (2A) Vane shear strength 12.1 [kN / m 2 ], (2B) Fall cone penetration was 3.1 [mm]. Therefore, according to the criteria shown in Table 1, the risk of poor mixing was judged to be "high" in both cases.

[0074] Therefore, we moved on to the test of solidified soil, and under the condition of solidification material water ratio (c / w) = 0.8, the solidification material was 300 [k / m 3 ] was added (see Figures 9 to 12), the apparent liquid index IL, which includes the amount of water added together with the solidification material, increased to 0.77, the proportion of solidified soil adhering to the soil was 0.27 (= 27%), and the adhesive strength was 0.73 [kN / m 2 ], and the adhesive strength in the vane shear test was 2.7 [kN / m 2 ], and the fall cone penetration was 9.5 mm. According to the criteria in Table 1, the risk of poor mixing for the three items other than the fall cone penetration changed to "low." The fall cone penetration was also 9.5 mm, compared to the lower limit of 10 mm for low risk, so although the risk was "medium," it was very close to "low."

[0075] Figure 13 shows the effect of the liquid index on the proportion of adhering soil for the first clayey soil (original soil) (wn = 0.726 (= 72.6%), IL = 0.76), and Figure 14 shows the effect of the liquid index on the proportion of adhering soil for the second clayey soil (original soil) (wn = 0.404 (= 40.4%), IL = 0.12). The arrows in the figures indicate the liquid index corresponding to the proportion of adhering soil at the natural moisture content. For ordinary clayey soils like those in Figure 13, adding solidification agent slurry increases the apparent liquid index, which includes the water content of the solidification agent slurry, and reduces the proportion of adhering soil. However, for special soils like those in Figure 14, the liquid index at the natural moisture content is small, and as the moisture content increases, the proportion of adhering soil increases. Therefore, it is possible to predict trends through original soil testing and then confirm them through solidification-treated soil testing.

[0076] Therefore, the trend can be predicted through the target soil test (raw soil test), and the predicted results can be confirmed through the solidification-treated soil test.

[0077] Regarding the setting of the liquid index IL, for raw soil, it is set to several values within a range of approximately 1.2 to 1.5 based on the natural water content, but in cases such as Kanto loam where a reaction between the solidification material and clay minerals and / or salts is predicted, it is necessary to expand to even larger values. For solidified treated soil, it is sufficient to plot the apparent liquid index, which includes the amount of water contained in the slurry as part of the water content, at approximately three levels, including the amount of addition that will give normal design strength.

[0078] For the soil to be improved (original soil) and the mixed soil (solidified soil), if there is a high risk of poor mixing in at least one of the adhering soil volume ratio and adhesion strength, as well as shear strength and fall cone penetration, countermeasures include increasing the amount of solidification material slurry added and / or increasing the water-solidification material ratio under conditions where the mix strength exceeds the design value, or determining the type of appropriate admixture and the amount to be added through testing, and setting the specifications for preventing poor mixing.

[0079] In test construction and actual work, the mixing condition of the target soil changes depending on the combination of the type of mechanical mixing method and the shape of the mixing blade, but the presence or absence of poor mixing can be evaluated by monitoring the rotation speed of the anti-rotation blades and / or the core collection rate using boring samples to determine the mixing condition using the actual machine.The physical property value data from the preliminary mix test is used as input data for learning, and the quality data collected during test construction or actual work is used as output data for learning, and a machine learning model is constructed from these data sets to predict the mixing condition using the actual machine.

[0080] By inputting the data from the upcoming pre-mixing test into the ground improvement form prediction model as input data for prediction, the monitoring rotation speed of the improvement body co-rotation prevention blades at the site and the boring core collection rate can be obtained as output data for prediction. By accumulating learning data from the machine learning-based ground improvement form prediction model, it is possible to improve the accuracy of predicting the risk of poor mixing according to the mixing blade shape of each mixing method, and by being able to predict the risk of poor mixing at the site from the results of the pre-mixing test and taking measures in advance, it will lead to improved quality of ground improvement.

[0081] When predicting whether or not co-rotation will occur, machine learning methods such as discriminant analysis based on multiple regression or clustering techniques are used. When boring core data (core recovery rate and solidification rate) are output and the frequency of poor mixing of the target soil and solidification material is predicted to be high, other machine learning models (e.g., random forests) may be used in addition to ground improvement shape prediction models using neural networks or regression analysis.

[0082] Example 2 Table 2 shows the results of the solidification-treated soil tests using several different types of solidification-treated soil. Using these test results, the proportion of soil adhering to the mixer impeller was set as the objective variable, and multiple regression analysis was carried out with x1: liquid index, x2: adhesive strength, x3: vane shear strength, and x4: fall cone penetration selected as explanatory variables. The regression equation does not necessarily fit well because there were two types of soil and three types of solidification material data, but it was significant at F<0.05, and explanatory variable x4 was a significant explanatory variable with p<0.05 and t>|2|. Regarding shear strength, 6 [kN / m 2 ] or more, it is known that clump cuttings will occur and poor mixing will likely occur, but in this case, the standard value for determining poor mixing was set at 40 (%) for the proportion of target soil adhering to the mixing blades.

[0083] A predicted value for the percentage (%) of the target soil adhering to the mixer's impellers was obtained based on the regression equation in the multiple regression analysis. If the predicted value exceeded the judgment standard value (=40), the risk of poor mixing was predicted as "A" (poor mixing (Abnormal)), and if the predicted value was below the judgment standard value of 40, the risk was predicted as "N" (Normal). The prediction results are also summarized in Table 2. A discriminant analysis of N or A was performed on the combined data of target soil types A and B and solidification material types A and B, and on the "unknown" clayey soil C.

[0084] The "adhesion rate" in Table 2 is the measured value of the "adhesion rate," and in the "prediction result" column, "*" indicates an incorrect prediction. The correct answer rate was 25 / 32 = 0.78.

[0085] [Table 2]

[0086] As training data, for unknown clayey soil C, explanatory variables x1 to x4 were read from the test data of solidified treated soil in Table 3, and a discriminant analysis was performed to predict the objective variable. All predictions for the six tests were correct.

[0087] [Table 3]

[0088] Conventionally, the cause of poor mixing of the target soil and solidification material has been considered to be the large shear strength of the target soil and the large deformation resistance, which is the influence of the consistency. However, when the vane shear strength BS shown in Table 2 is 6 (kN / m 2 ), 16 samples (samples 9-12, 14-16, 20, 22-26, 29, 30, 32) showed an adhesion ratio of 40% or more, and were judged to be at risk of poor mixing (N), indicating that there is a high possibility of poor mixing occurring due to factors related to the adhesion soil ratio. This supports the need to combine the results of the first and second tests to predict the risk of poor mixing between the target soil and solidification material.

[0089] In this example, the proportion of adhering soil is predicted as a predictor of the risk of poor mixing of the target soil and solidification material, and a ground improvement form prediction model is constructed from learning data from laboratory tests, making it possible to predict the risk of poor mixing even if the proportion of adhering soil has not been measured in unknown laboratory test data.Similarly, a ground improvement form prediction model is constructed to predict vane shear strength, making it possible to predict the risk of poor mixing according to the ground improvement form prediction model.

[0090] Based on the above, a ground improvement form prediction model created by machine learning using multiple pieces of learning data, which is a combination of learning input data from pre-mixing tests and learning output data consisting of the occurrence or non-occurrence of co-rotation determined from the rotation speed of the anti-co-rotation blades, which is the result of on-site construction, can be used to input the results of pre-mixing tests of the target soil for which construction is to begin as prediction input data into the ground improvement form prediction model, and the prediction output data from the ground improvement form prediction model can be used to predict poor mixing of the improvement body based on the occurrence or non-occurrence of co-rotation determined from the rotation speed of the anti-co-rotation blades.

[0091] (Effects of the Invention) The soil testing method of the present invention improves the accuracy of predicting the risk of poor mixing between the target soil and the solidification material (solidification material slurry) based on the results of the first and second tests. When the risk is high, specific countermeasures can be considered, and the risk of poor mixing between the target soil and the solidification material (solidification material slurry) when the countermeasures are adopted can be predicted with high accuracy. Furthermore, the more data collected at the construction site by confirming the presence or absence of co-rotation and / or the mixing state through boring, the more accurate the ground improvement form prediction model can be, using the data and machine learning to predict the risk of poor mixing between the target soil and the solidification material. Furthermore, by using the ground improvement form prediction model, the accuracy of predicting the risk of poor mixing between the target soil and the solidification material can be further improved.

[0092] (Another embodiment of the present invention) Based on the results of the first and second tests on the target soil (raw soil), the risk of poor mixing may be predicted in accordance with the risk assessment criteria (see Table 1). Based on the results of the first and second tests on the solidification-treated soil (primary solidification-treated soil and / or secondary solidification-treated soil), the risk of poor mixing may be predicted in accordance with the risk assessment criteria (see Table 1). [Explanation of symbols]

[0093] 10. Ground improvement shape prediction device 11. Input interface 111...Learning input data acquisition unit 112...Learning output data acquisition unit 113. Prediction input data acquisition unit 12. Ground improvement shape prediction model generation section 13. Ground improvement shape prediction section 14. Output interface.

Claims

1. a first test step of conducting a first test which is at least one of an adhesion soil amount test for measuring the amount of adhesion of the target soil to the mixing blades of the mixer, and an adhesion strength test for measuring the adhesion strength when the target soil adhered to a steel material or a glass material peels off; A second test step of conducting a second test consisting of a vane shear test on the target soil and a fall cone test on the target soil; A prediction process of predicting the ground improvement form corresponding to the target soil by combining the first test result obtained by implementing the first test process and the second test result obtained by implementing the second test process; A soil testing method including:

2. The soil testing method according to claim 1, a testing step of carrying out the first test and the second test for each of three or more levels of primary solidification treated soil with different amounts of solidification material added to the target soil while keeping the solidification material-water ratio constant; Based on the test results obtained by carrying out the test process, if there is a high risk of poor mixing of the solidified treated soil, the first test, the second test and the prediction process are carried out on the secondary solidified treated soil obtained by adding a solidification material adjusted to an appropriate mix to the target soil in order to reduce the risk of poor mixing and achieve the indoor target mix strength. Soil testing methods.

3. A process of generating, by machine learning, a ground improvement form prediction model that expresses the correlation between the first test results and the second test results of the target soil and the solidification-treated soil obtained by adding a solidification material to the target soil, using multiple sets of learning data including first test results obtained by conducting at least one of an adhesion volume test and an adhesion strength test, and second test results obtained by conducting a vane shear test and a fall cone test, for each of the target soil and the solidification-treated soil obtained by adding a solidification material to the target soil; and inputting prediction input data including the first test result and the second test result for new target soil and solidification-treated soil, respectively, into the ground improvement form prediction model, and predicting the ground improvement form of the new target soil as prediction output data from the ground improvement form prediction model. Method for predicting ground improvement form.

4. In the ground improvement shape prediction method according to claim 3, a mix determination process for determining an appropriate mix to reduce the risk of poor mixing of the solidified treated soil and achieve the indoor target mix strength based on the first test results and the second test results for each of three or more levels of primary solidification treated soil with different amounts of solidification material added to the target soil while keeping the solidification material water ratio constant; and a solidification-treated soil testing step for obtaining the first test results and the second test results constituting the plurality of sets of learning data, using the secondary solidification-treated soil obtained by adding the solidification material adjusted to the appropriate mix determined in the mix determination step to the target soil. Method for predicting ground improvement form.

5. The ground improvement shape prediction method according to claim 4, As a form of ground improvement targeting the new target soil, predict the risk of poor mixing between the new target soil and the solidification material. Method for predicting ground improvement form.

6. The ground improvement shape prediction method according to claim 4, As a form of ground improvement targeting the new target soil, the quality of the improved body obtained after the new target soil and solidification material are mixed is predicted. Method for predicting ground improvement form.

Images