Prediction of the Unconfined Compressive Strength of Biopolymer-based Soil Treatment via Machine Learning Approaches
기계학습 분석을 통한 바이오폴리머 처리 흙의 강도 예측
- 주제(키워드) Support vector regression , Biopolymer-based soil treatment , Unconfined compressive strength , Prediction
- 주제(DDC) 690
- 발행기관 아주대학교
- 지도교수 장일한
- 발행년도 2023
- 학위수여년월 2023. 2
- 학위명 석사
- 학과 및 전공 일반대학원 건설시스템공학과
- 실제URI http://www.dcollection.net/handler/ajou/000000032700
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
Bio-based materials have been recently introduced in the field of soil improvement in geotechnical engineering to reduce hazardous substances such as heavy metal and greenhouse gases. Among the bio-based soil improvement techniques, there are biopolymer-based soil treatment (BPST), microbially-induced calcite precipitation (MICP), and enzyme-induced carbonate precipitation (EICP). These novel approaches, however, have not been studied in a long time, thus their experience is insufficient and a lot of uncertainty exists. Hence, laboratory tests must be accompanied for field application, consuming many manpower and time. In this study, a machine learning technique is adopted to predict behavior and analyze mechanisms of BPST that can obtain more unified experiment results than other bio-based materials. For model training, the database with 479 records was organized and compiled by collecting experimental data about the unconfined compressive strength (UCS) of BPST. Support vector regression (SVR) algorithm was used for modeling. SVR has been recognized for its capability to predict non-linear data in many fields, and it has also been widely used to predict the strength of various soil cementation materials such as concrete, soil cement, and geopolymer. The data set was split into a training set (90%) and a testing set (10%) to validate the model. The kernel function of SVR used a radial basis function (RBF). The three hyperparameters of SVR, i.e., the penalty factor (C), the RBF kernel function parameter (γ), and with of the ε-tube (ε), were optimized through grid search and 5-fold cross-validation. Root mean squared error (RMSE) and coefficient of determination (R2) were used as prediction performance criteria. The prediction result of the optimized SVR model showed high prediction performance, with R2 = 0.9747, RMSE = 411.082 on the training set, and R2 = 0.9785, RMSE = 374.591 on the test set. Permutation feature importance (PFI) was used to detect influence features in UCS prediction. As a result, the water content and the average particle size (i.e., soil type) were derived as dominant in the UCS prediction of BPST. These results matched the mechanism of BPST, which was dealt with in previous studies, and the reliability of the model can be obtained. To confirm the applicability of the designed prediction model, a new data set was contrived and the value of UCS was predicted. Under dry conditions, it was confirmed that the UCS decreased as the size of soil particles increased and increased as the biopolymer content increased. It is thought that the UCS can be predicted if we know only the type of soil, type of biopolymer, and water content. As a result, it can present guidelines prior to practical laboratory tests, saving us a significant amount of time and money.
more목차
CHAPTER I INTRODUCTION 1
1.1 Background 1
1.2 Literature Review 2
1.3 Scope - Organization 6
CHAPTER II BIO-BASED SOIL IMPROVEMENT METHODS 7
2.1 Introduction 7
2.2 Biopolymer-based Soil Treatment (BPST) 9
2.2.1 Background 9
2.2.2 Mechanism of Strength Enhancement 9
2.3 Microbially-Induced Calcite Precipitation (MICP) 13
2.3.1 Background 13
2.3.2 Mechanism of Strength Enhancement 14
2.4 Enzyme-Induced Calcite Precipitation (EICP) 17
2.4.1 Background 17
2.4.2 Mechanism of Strength Enhancement 17
2.5 Summary and Conclusions 19
CHAPTER III DATABASE CONFIGURATION 20
3.1 Introduction 20
3.2 UCS Database Configuration 21
3.2.1 Retrieved Data from Literature 21
3.2.2 Unconfined Compressive Strength (UCS) 23
3.2.3 Parameters for UCS 23
3.3 Feature Selection 29
3.4 Data Preprocessing 34
3.5 Summary and Conclusions 35
CHAPTER IV METHODOLOGY AND MODELING 36
4.1 Introduction 36
4.2 Support Vector Regression 37
4.2.1 Concepts of Support Vector Regression 37
4.2.2 Evaluation Criteria 40
4.2.3 Modeling for UCS Prediction 41
4.3 Feature Importance 45
4.3.1 Permutation Feature Importance 45
4.3.2 Ranking of Feature Importance 46
4.4 Summary and Conclusions 47
CHAPTER V GEOTECHNICAL ENGINEERING IMPLEMENTATIONS 48
5.1 Introduction 48
5.2 Dominant Parameters of Strength Enhancement 49
5.2.1 The Most Dominant Variable: Water Content 49
5.2.2. The Second Dominant Variable: Soil Type 50
5.2.3 The Most Insignificant Variable: Curing Time 52
5.3 Applications of UCS Prediction Model 54
5.4 Summary and Conclusions 58
CHAPTER VI CONCLUSION AND RECOMMENDATIONS 60
6.1 Conclusion 60
6.2 Recommendations for Future Study 62
REFERENCES 63
