30 May – 1 June, 2018
International Design Centre
Singapore University of Technology and Design
SUTD Think Tank 21 (Building 2, Level 3)
The design and control of next-generation engineered systems pose many exciting challenges that require innovations in computational methods. This workshop highlights a range of work in methods for inference, prediction and decision-making under uncertainty in both design and control settings, with a particular focus on approaches that exploit the synergies of models and data.
9.00am–9.15am Karen Willcox Welcome and overview
9.15am–10.15am Douglas Allaire Invited Talk: What next? Sequentially value-optimal engineering tasking for analysis and design
10.45am–11.30am Victor Singh Data-driven design using digital thread
1.30pm–2.30pm Laura Mainini Invited Talk: Data-driven methods for self-aware aerospace systems and structures
2.30pm–3.15pm Renee Swischuk Projection-based model reduction: Formulations for physics-based machine learning
3.30pm–4.30pm Discussion Future directions in data-driven modeling and design
9.15am–10.15am Nguyen Van Bo Invited Talk: Model order reduction for continuous chemical processing and control
10.45am–11.30am Boris Kramer Lifting nonlinear systems: More structure, more opportunities for ROM?
11.30am–12.15pm Elizabeth Qian Challenges in data-driven reduced order modeling for combustion problems
1.15pm–2.15pm Tiangang Cui Invited Talk: Subspace acceleration for large-scale Bayesian inverse problems
2.15pm–3.00pm Harriet Li Nonlinear goal-oriented inference
3.15pm–4.15pm Discussion Future directions in reduced modeling and large-scale inverse problems
Friday 1 June – Design under uncertainty
9.15am–10.15am Rhea Liem Invited Talk: Data-enhanced modeling for aircraft design and air transportation
10.45am–11.30am Laurence Cook Optimization under uncertainty using multiple dominance criteria for aerospace design
11.30am–12.15pm Alex Feldstein Multi-fidelity methodologies for conceptual design under uncertainty
12.15pm–1.30pm Lunch and Discussion Future directions in design under uncertainty
Design of a complex system involves a rich set of information sources, including such diverse sources as physics-based models, experiments, historical data, and expert opinions. There is an increasing recognition of the value of mathematical methods that provide systematic guidance in managing these information sources in support of decisions across the design phase.
Douglas Allaire, Assistant Professor, TAMU
Victor Singh, PhD student, MIT
Ng Jia Yi, PhD student, SUTD
Model reduction has in recent years become a widespread approach to mitigate the computational cost of solving large-scale systems such as those resulting from discretization of partial differential equations. Recent developments in combining traditional projection-based model reduction methods with machine learning methods have led to new approaches that tune and adapt reduced models in the face of changing system properties and dynamic data. This provides new opportunities to discover and learn models, informed by physics-based first principles but guided by data.
Data-driven methods for self-aware aerospace systems and structures
Laura Mainini, Visiting Professor, Politecnico di Torino
Projection-based model reduction: Formulations for physics-based machine learning
Renee Swischuk, SM student, MIT
Model reduction has in recent years become a widespread approach to mitigate the computational cost of solving large-scale systems such as those resulting from discretization of partial differential equations. Considerable academic advances have been made in model reduction methods, yet impact in industry has been limited. Several key challenges must be addressed to make model reduction applicable – and applied – to a broader range of problems. First, research is needed to address the challenges of nonlinear systems; second, methods are needed that overcome the intrusive nature of existing model reduction approaches yet retain the rigor of the approaches.
Nguyen Van Bo, A*STAR
Boris Kramer, Postdoc, MIT
Elizabeth Qian, PhD student, MIT
An inverse problem seeks to estimate an unknown and unobservable parameter from (often indirect and noisy) measurements. Inverse problems are pervasive in science and engineering, yet solving them at scale for systems governed by complex nonlinear dynamics is a significant challenge. This challenge can be addressed through approaches that leverage low-dimensional approximations that exploit system structure.
Tiangang Cui, Lecturer, Monash
Harriet Li, PhD student, MIT
The many sources of uncertainty in the early stages of design pose critical risks to program success. Uncertainty arises due to the uncertain models used to make critical decisions in the design process, programmatic uncertainties in cost and schedule, the uncertainty surrounding advanced technologies that are often employed to push the envelope in system performance, the uncertain manufacturing processes that result in as-built systems that may deviate from as-designed ones, and the uncertain external environment in which these systems must operate.
Rhea Liem, Assistant Professor, HKUST
Laurence Cook, Postdoc, MIT
Alex Feldstein, SM student, MIT