Publicação
A Bayesian approach to NDT Data Fusion for St. Torcato Church
| Resumo: | The main objective of this thesis is to combine information gathered from different Non Destructive tests (NDT) (direct and indirect) and fuse it by using Bayesian approach. Many time practitioners working with NDT data want to choose parameters based on results of different NDT tests with different levels of reliability and uncertainty quantification. As suggested by literature the use of a single technique might not suffice to gain information and the combination of different techniques is recommended. Also for the case of masonry structures it might not be possible to perform destructive tests but since the parameter has to be estimated based on information provided by various NDT data sources coupled with literature information. NDT data from San Torcato Church was used in this thesis to test a Methodology to transform the data into a single and uniform format by the help of Bayesian approach. A simple Matlab Toolbox NDT_FUSION was developed and tested with different models available and modified later by using a Trust Factor which takes into account the weightage of different NDT tests. The developed toolbox is very easy to use since it has Graphical user interface (GUI) and does not required practitioner to learn the complex mathematics involved in calculation behind the Bayesian black box. The data fusion was done at different levels and steps so every time an updating takes place we arrive to a more realistic value of parameter. Two geomechanical parameters namely the Elastic modulus (E) and compressive strength ( fc) of granite blocks from St. Torcato Church were studied in this thesis. The normal probability distribution function for the parameter of interest was calculated by using Jeffrey’s Prior and Conjugate Prior, considering different levels of initial knowledge. The Elastic modulus (E) was updated by using data from Literature knowledge, sonic, ultrasonic and direct compressive strength tests to arrive to a more certain value in form of a posterior distribution. In both the cases the raw data from direct and indirect sources was processed and combined with data fusion toolbox to transform values into statistical distribution. The reliability confidence intervals of parameters were updated every time a new data becomes available providing more broad information. Different levels of uncertainty are present in data fusion system proposed in this report starting from the literature knowledge to direct compression test core data which were quantified and addressed in this thesis. The tests of different reliability levels were weighed by circulating a survey form among professors and graduate students experts in the field to take their opinion. The results of the surveys come was the calculation of Trust Factor to update the spread of the parameters and incorporate in the model to obtain better predication of the parameters. The application developed comes with a Matlab compiler runtime (MCR) installer which allows the application to run on computers without the prerequisite of having Matlab installed. |
|---|---|
| Autores principais: | Mishra, Mayank |
| Assunto: | NDT data fusion Bayesian updating Uncertainty Mechanical parameter Fusão de dados Análise bayesiana Incertezas Estimativa de parâmetros mecânicos |
| Ano: | 2013 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | The main objective of this thesis is to combine information gathered from different Non Destructive tests (NDT) (direct and indirect) and fuse it by using Bayesian approach. Many time practitioners working with NDT data want to choose parameters based on results of different NDT tests with different levels of reliability and uncertainty quantification. As suggested by literature the use of a single technique might not suffice to gain information and the combination of different techniques is recommended. Also for the case of masonry structures it might not be possible to perform destructive tests but since the parameter has to be estimated based on information provided by various NDT data sources coupled with literature information. NDT data from San Torcato Church was used in this thesis to test a Methodology to transform the data into a single and uniform format by the help of Bayesian approach. A simple Matlab Toolbox NDT_FUSION was developed and tested with different models available and modified later by using a Trust Factor which takes into account the weightage of different NDT tests. The developed toolbox is very easy to use since it has Graphical user interface (GUI) and does not required practitioner to learn the complex mathematics involved in calculation behind the Bayesian black box. The data fusion was done at different levels and steps so every time an updating takes place we arrive to a more realistic value of parameter. Two geomechanical parameters namely the Elastic modulus (E) and compressive strength ( fc) of granite blocks from St. Torcato Church were studied in this thesis. The normal probability distribution function for the parameter of interest was calculated by using Jeffrey’s Prior and Conjugate Prior, considering different levels of initial knowledge. The Elastic modulus (E) was updated by using data from Literature knowledge, sonic, ultrasonic and direct compressive strength tests to arrive to a more certain value in form of a posterior distribution. In both the cases the raw data from direct and indirect sources was processed and combined with data fusion toolbox to transform values into statistical distribution. The reliability confidence intervals of parameters were updated every time a new data becomes available providing more broad information. Different levels of uncertainty are present in data fusion system proposed in this report starting from the literature knowledge to direct compression test core data which were quantified and addressed in this thesis. The tests of different reliability levels were weighed by circulating a survey form among professors and graduate students experts in the field to take their opinion. The results of the surveys come was the calculation of Trust Factor to update the spread of the parameters and incorporate in the model to obtain better predication of the parameters. The application developed comes with a Matlab compiler runtime (MCR) installer which allows the application to run on computers without the prerequisite of having Matlab installed. |
|---|