My research uses machine learning, image processing, computational techniques and optimization heuristics in structural engineering. Structural designers have to develop cost effective and efficient designs in spite of severe time and budget constraints. As a result of these constraints, it is difficult for a designer to make the best choices to yield an efficient structure. My research facilitates the exploration of the design space quickly in order to help the engineer make better design choices.
My research methodology comprises of two key steps: first develop a prototype structure and next, design a structrue based on the prototype and optimize it. The first stage of developing a prototype is done using topology optimization. The topology optimized structure is then post-processed via a novel algorithm to yield an appropriate structural design. This structural design is then optimized using evolutionary algorithms.
In addition to the development of this methodology, I explored the use of machine learning to determine the design type of bridges. In the preliminary design stage, decisions regarding design type are primarily based on the experience of the engineering team. Hence, a machine learning based methodology was developed which could capture the experience from past bridges, to predict the design type of a new bridge. The details of this study along with the studies for optimal structural design are presented below.
This study used the National Bridge Inventory (NBI) dataset to learn supervised learning models and predicted the bridge design type for any new bridge. Data from the NBI dataset was cleaned by removing irrelevant attributes and a subsequent step of attribute evaluation. The data was then integrated with other relevant datasets such as the United states Geological Survey seismic intensity dataset and historic concrete and steel cost dataset obtained from Engineering News Record. Decision Trees and Bayesian Networks were used to learn from this combined dataset and predictions for the design type were made. An average precision and recall of over 88% was obtained. A journal paper detailing this work is currently in press.
This study used topology optimization (TO) and evaluated its applicability in structural design. Topology optimization has been used to optimally design monolithic mechanical and aeronautical components. Due to a monolithic optimized structure, TO's applicability in structural design has been limited. In order to use TO in structural design, a methodology was developed, which uses TO to generate structures, converts it into a "joints and structural members" representation and finally optimizes it using an optimization heuristic. In this study, trusses were designed using topology optimization and the results obtained were compared with the results from the literature. Structural anlaysis showed that the results obtained from TO had lower deflections (by ~50%) but were slightly heavier (by ~7%) than the solutions in the literature. Impact of different parameters on the performance of TO was also analyzed. A journal paper based on this work is currently in preparation.
The result obtained via TO is a monolithic structure. However, a structure in practice is an assembly of different structural elements, and hence the TO result obtained needs to be processed before it can be converted to a structure. In this study, different algorithms were developed to detect the joints and structural members in a TO structure. Detecting the joints and members makes the representation of the structure suitable for futher optimization. The algorithms developed were used to identify the joints and members and the best performing algorithm was chosen to convert the structure into a representaiton suitable for optimization. A conference paper based on this work will be presented at IWCCE 2017. A journal paper based on this work is under preparation.
This study uses the representation of the TO structure obtained via the algorithm mentioned in the previous section and optimizes the structure. This study optimizes for the minimum weight of the structure while ensuring that the structure does not violate any design constraints imposed on it. This study is currently in progress and an optimization framework implementing evolutionary algorithms is being developed.