A machine learning procedure for seismic qualitative assessment and design of structures considering safety and serviceability

In this study, a new machine learning-based procedure is proposed for seismic design and qualitative assessment of structures considering safety and serviceability aspects. The procedure is composed of three different intelligent computational techniques, which are artificial neural network (ANN), fuzzy inference system (FIS), and ensemble bagged tree (EBT) classification algorithm. The procedure provides a qualitative assessment of structures based on their basic structural characteristics, which are the natural period and the strength ratio. Numerous nonlinear time history analyses (NLTHA) (more than 60,000 runs) are conducted to prepare the required dataset for training, validation, and testing. Many case-study examples with different characteristics under different ground excitations are used to validate the proposed procedure. The procedure results showed a very good agreement with NLTHA results and proved to be superior compared to conventional methods. After that, the implementation potential of the proposed procedure is shown through different applications in seismic design and assessment. Firstly, two case-study models are qualitatively classified and then the required design stiffness and strength for seismic retrofitting are quantified. Secondly, the median collapse capacity and seismic fragility of a case-study model are estimated and compared with those obtained from the incremental dynamic analysis (IDA) and fragility curves using NLTHA. Besides being a quick assessment tool, the results indicated that the proposed procedure is reliable and accurate with far less computational cost compared to the traditional NLTHA method.

Automated summary

This study proposes a new machine learning-based procedure for seismic design and qualitative assessment of structures. The procedure combines artificial neural network, fuzzy inference system, and ensemble bagged tree classification algorithm. The procedure provides qualitative assessment of structures based on their basic structural characteristics. The results of the procedure showed good agreement with nonlinear time history analyses and proved to be superior compared to conventional methods. The procedure also demonstrated potential in seismic design and assessment applications, with lower computational cost compared to traditional methods.