A systematic model identification procedure in geotechnical prediction based on the information criteria | |
| Author | Liu, Wen-tsung |
| Call Number | AIT Diss. no. GT-93-01 |
| Subject(s) | Soils--Analysis Soil mechanics--Mathematical models |
| Note | A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Engineering, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Dissertation ; no. GT-93-01 |
| Abstract | Inverse analysis was applied to estimate pa:ameters based on the observed data and other information. The difference between this method and the conventional trial and error procedure is that it employs statistical formulations and optimizational techniques to execute the calibration in a much more systematic and rational way. The difficulties encountered in geotechnical inverse analysis can be classified in terms of instability, non-uniqueness and multicollinearity. The objective of this dissertation is to propose a new statistical formulation which may be termed as the extended Bayesian method for the inverse problem. This solved the difficulties encountered in the geotechnical inverse analysis of this research. The key point of this method is to establish a way of matching two different types of information, i.e. the observed data and the prior information based on a new view of Bayesian statistics proposed by Akaike. The concept is exactly the same as the one he developed in his Akaike Information Criterion. The methodology is illustrated by a case study of a three-meter high control embankment on untreated marine clay at Muar Flat, Malaysia. This embankment provides the laboratory test data and the field instrumentation data. For analyzing the data, the FEM program CRISP was used in the calculation of this study. Based on the analysis of the laboratory data, it was obvious that multicollinearity exists in the estimated parameters. It was also found that the multicollinearity comes from two sources, namely the model structure and the sampling scheme. The former suggests that the dimension of the estimated parameters should be reduced to a working limit based on the parameter sensitivity, while the latter suggests that too much available data employed in the case could reduce the parameters' uncertainty effectively. Analyzing the field cases, it was proved that the extended Bayesian formulation can overcome the following critical problems in geotechnical inverse analysis: (1) The instability and non-uniqueness of the solution encountered in the formulation based on the maximum likelihood method. (2) The appropriate matching of two different types of information, i.e. the objective information (the observed data) and the subjective information (the prior information). (3) The model identification problem: to choose the most appropriate model for prediction which has a moderate degree of sophistication based on the various amounts of data. |
| Year | 1994 |
| Corresponding Series Added Entry | Asian Institute of Technology. Dissertation ; no. GT-93-01 |
| Type | Dissertation |
| School | School of Engineering and Technology (SET) |
| Department | Other Field of Studies (No Department) |
| Chairperson(s) | Sugimoto, M. |
| Examination Committee(s) | Balasubramaniam, A.S. ;Bergado, Dennes T. ;Noppadol Phien-Wej ;Fujiwara, Okitsugu ;Honjo, Y. |
| Scholarship Donor(s) | Republic of China; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 1994 |