MS006 - Uncertainty quantification and design optimization of complex structures and innovative construction process using machine learning
Keywords: engineering design, geotechnics, industrial applications of AI, inverse problems, scientific machine learning, surrogate modeling, uncertainty quantification
The last two decades have been marked by the constant development of new technologies in the construction industry. For instance, taking advantage of the evolution of digitalization and the robotics industry, 3D concrete printing (3DCP) has revealed its great potential for automating the construction of concrete structures with great geometric freedom. In addition, the rapid evolution of efficient characterization methods and intelligent data analysis provides a great opportunity to help optimize innovative construction process as well as the complex structures design.
Besides, the limitation of the conventional approach that uses different partial safety factors to account for the uncertainty of input parameters in the design procedure has been observed in many contexts. For example, based on a limited number of tests in the laboratory to characterize the material properties, this approach seems inapplicable for the 3DCP process because of the lack of standardized tests. The design of underground structures (e.g., tunnels constructed in long and complex geological formations that contain highly heterogeneous soil or rock layers) with this conventional approach encounters also lots of difficulties.
By considering explicitly the uncertainty in the analysis and design, the probabilistic approach has been considered as a performant method for the uncertainty quantification and optimization design of complex structures and the innovative construction techniques like 3DCP process. According to this method, the uncertainty of input parameters is quantified in the first step using data provided by laboratory tests or in-situ measurements. Then the reliability analysis can be combined with an efficient optimization algorithm to determine the optimal result of the design parameters. Nevertheless, in most cases, the evaluation of the complex structure behaviour or the response during 3D printing of concrete structure can only be assessed numerically which is highly expensive and presents an obstacle to the reliability-based optimization design. Machine learning gives an efficient and useful tool when it can be chosen as the surrogate to replace the direct numerical simulation of structures. The aim of this symposium is therefore to bring together experts and to present the state of the art, advances and trends in the uncertainty quantification and optimization design of engineering structures and construction process using machine learning.
Besides, the limitation of the conventional approach that uses different partial safety factors to account for the uncertainty of input parameters in the design procedure has been observed in many contexts. For example, based on a limited number of tests in the laboratory to characterize the material properties, this approach seems inapplicable for the 3DCP process because of the lack of standardized tests. The design of underground structures (e.g., tunnels constructed in long and complex geological formations that contain highly heterogeneous soil or rock layers) with this conventional approach encounters also lots of difficulties.
By considering explicitly the uncertainty in the analysis and design, the probabilistic approach has been considered as a performant method for the uncertainty quantification and optimization design of complex structures and the innovative construction techniques like 3DCP process. According to this method, the uncertainty of input parameters is quantified in the first step using data provided by laboratory tests or in-situ measurements. Then the reliability analysis can be combined with an efficient optimization algorithm to determine the optimal result of the design parameters. Nevertheless, in most cases, the evaluation of the complex structure behaviour or the response during 3D printing of concrete structure can only be assessed numerically which is highly expensive and presents an obstacle to the reliability-based optimization design. Machine learning gives an efficient and useful tool when it can be chosen as the surrogate to replace the direct numerical simulation of structures. The aim of this symposium is therefore to bring together experts and to present the state of the art, advances and trends in the uncertainty quantification and optimization design of engineering structures and construction process using machine learning.
