Biologically-inspired algorithms for selecting and scaling ground motions and predicting the seismic response of buildings

Performance-based earthquake engineering (PBEE) is a framework that provides the tools to assess structures under seismic actions in order to quantify the seismic risk (i.e., seismic damage, loss, downtime and casualties expressed through the probability or frequency of occurrence in a time frame). The need to communicate seismic risk and to compare it with other sources of risk is essential for informed decisions to be taken and reduce the economical and societal impact. Even though the PBEE framework has been introduced more than two decades and it is widely appreciated, engineering practice is not able to employ it extensively because of the increased resources required in terms of computation time and expertise. In particular, the processes of ground motion (GM) selection and nonlinear response history analysis (NLRHA) are notably challenging because of the involved computational cost and complex formulation. Moreover, if amplitude scaling is also employed, then the integrity of the results is questioned under the notion that bias is introduced. The advances in these three areas proposed in this work are supposed to reduce their hardships and promote PBEE in engineering practice.

A GM record selection scheme typically ensures that the input excitation to be used in NLRHA is appropriate in terms of spectral compatibility, hazard and intensity measure (IM) consistency, seismological and site-specific criteria (with respect to corresponding targets). Simultaneously, it is expected that it performs in a computationally efficient manner. A GM selection scheme utilizing genetic algorithms is proposed here, able to select multi-component GM and satisfying simultaneously all the typically required selection objectives of earthquake engineering applications while ensuring increased ef- ficiency. Multi-objective optimization is performed, claimed to be superior in delivering robust results that account for spectral compatibility in first and second order statistics (i.e., mean and standard deviation) in a wide range of spectral values, while satisfying seismological and site-specific criteria. A unique contribution of the proposed scheme is the ability to include probability distribution targets in specific ordinates of the spectrum, on top of the mean and standard deviation. This delivers more refined ground motion sets that can be used to reduce the number of GM required in NLRHA. Additionally, a novel benchmarking process to assess the efficiency of GM record selection methodologies is introduced. Instead of assessing the resulting quality in spectral compatibility, it is claimed that GM selection efficiency should be investigated in providing GM sets that are globally-optimal solutions to the optimization problem. Through this benchmarking algorithm, the proposed methodology appears to be impeccable in extracting the best possible GM sets.

The application of amplitude scaling of GM is controversially discussed in earthquake engineering. In this study, bias is questioned in determining an engineering demand parameter (EDP) as a result of NLRHA when using scaled rather than unscaled GM that have the same level of intensity as described through IM. To this end, 10 planar steel- frame building models are analyzed ranging from low- to high-rise. The EDP of interest is the maximum interstory drift ratio (MIDR) and the structural responses range from linear to collapse. For an in-depth investigation of the research question, a vast number of more than 17,000 recorded GM are collected from the NGA-West2 database. Performing incremental dynamic analyses in all structural models subjected to all GM resulted in approximately 3.4 million NLRHA thus creating a rich database of structural responses. The importance of well-known IM is discussed and by considering them together with newly-introduced spectra describing the sustained vibration amplitude, the introduction of bias is examined from different points of view. Firstly, simple and intuitive statistical methods are employed, then machine learning techniques, and finally the GM selection approach proposed in this work is applied. In the numerous investigations, no bias could be detected under the inherent uncertainty of the calculations. The results indicate that scaled records can be safely used in NLRHA to assess the seismic structural behaviour if spectral and scenario compatibility are ensured and it is verified that the sustained amplitude is also consistent.

To circumvent the state-of-the-art NLRHA and reduce the computation time, simplified procedures have long been pursued. To this end, this study investigates artificial neural networks (NN) as prediction models to bypass the NLRHA and quickly and reliably determine the MIDR of building structures. First, the possible designs of such prediction models are discussed in terms of their scope, implementation and corresponding database. The database of structural responses mentioned above is utilized again for this investigation. Based on this database, a designated prediction model is developed for each building, capable of predicting the outcome of NLRHA under a GM excitation that is not included in the database. In addition to the ten building-specific (i.e., record- to-record) prediction models, another one is developed that can predict the outcome of NLRHA for a building and GM excitation, both of which are not included in the database. These investigations indicate that NN is an excellent tool to capture the record-to-record uncertainty and reliably predict MIDR ranging from linear responses to the collapse limit state without resorting to the tedious NLRHA. It is shown that in addition to record-to- record predictions, building-to-building predictions are also feasible if the database used to create the prediction models consists of building structures that are comparable to the building of interest.

Nach oben scrollen