Research | CP Andriotis

Decision-making optimization

Engineered and natural environments are exposed to adverse operating conditions during their life-cycle, with

deterioration and hazards posing a constant threat to their reliability and resilience. Optimal decision-

making for such systems is concerned with controlling the extent and consequences of this exposure through

scheduling of efficient strategies, from preventive planning to post-disaster recovery, with the objective to minimize the involved socioeconomic and environmental risks. To be successful in this objective,

decision optimization needs to holistically alleviate multi-faceted complexities arising, among others,

by the curse of dimensionality related to large-scale multi-component systems; the curse of history

related to long-term sequential decisions; the model and data uncertainties; the operational constraints;

the presence of multiple agents and decision layers; and, finally, the sheer connection of physics-based

engineering models with the optimization process. This line of research studies solutions to the above

challenges through novel conceptual and computational frameworks within the contexts of stochastic optimization, systems reliability & control, and artificial intelligence. read more

Risk & reliability assessment

Quantification of uncertainties in models and data constitutes the fundamental basis for quantifying risk, thus

being able to make informed engineering decisions. Towards this, uncertainties associated with chronic

deteriorating stressors (e.g. corrosion or fatigue) as well as with recurrent hazards (e.g. earthquakes

or hurricanes) need to be efficiently processed, either directly, based on data, or indirectly, propagated

through physics-based models. The goal is to efficiently estimate and infer certain probabilistic quan- tities of interest, but also to learn statistical structures that are able to forecast the dynamics of these

quantities in the absence of physics-based simulators. Probabilistic modeling in this regard has to

accommodate the presence of high dimensional feature and random variable spaces, as well as the

presence of correlations and dependencies at the spatial and temporal scales. This research is involved in

the study and development of such risk & reliability methods as these pertain to probabilistic performance-

based engineering, structural fragility analysis, Bayesian networks inference and learning, and supervised or unsupervised probabilistic machine learning. read more

Computational structural mechanics

Extreme events come with extreme loads, which force structural systems to their limit states. These states are often

manifested through large inelastic displacements and rotations that utterly defy linear predictions. Numerically

accurate and computationally efficient assessment of responses in materially and geometrically nonlinear

regimes is thus indispensable, especially in cases where statistical estimation and stochastic control are

the ultimate objectives. This line of research is involved in the study and development of computational

approaches, broadly pertaining to constitutive modeling of materials with coupled plasticity and

damage considerations, Bouc-Wen phenomenological hysteretic simulators, and structural element

formulations based on hybridized variational principles. In this direction, a major objective of this

research is to advance the joint platforms of computational structural mechanics and optimization,

as these relate to how nonlinear programming concepts can be integrated with the analysis process to

tame sources of computational complexity, and, reversely, to how highly nonlinear simulators can be integrated with optimization algorithms to drive dynamic design and structural intervention decisions. read more

Contact

Faculty of Architecture & the Built Environment

Delft University of Technology

Julianalaan 134, 2628 BL, Delft

email: c.andriotis [at] tudelft [dot] nl