Main Research Topic: Uncertainty Quantification (UQ)
Why do we need to consider UNCERTAINTIES ?
Topic 1: Deterministic Input-Output Model Construction
(Data-Driven & Physics-Based)
Topic 2-1: Forward Uncertainty Quantification
(Forward UQ; Forward Uncertainty Propagation)
Topic 2-2: Inverse Uncertainty Quantification
(Inverse UQ; Inverse Uncertainty Propagation)
Topic 2-3: Model Uncertainty Quantification
#1. Finite Element (FE) Model Updating using Deep Reinforcement Learning Agent
- Finite element models may have errors, due to various reasons (Sehgal and Kumar 2016):
- Wrong inputs of material properties (e.g., mass density, Young's modulus, etc.)
- Geometric simplification in modeling complex shapes
- Non-consideration of non-linear properties
- Incorrect modeling of boundary conditions and joints
- Poor quality mesh design
- Computational approximation (e.g., rounding off)
- These factors need to be carefully corrected for diminishing the model errors. Here, this correction process is an optimization problem with multiple design variables, which may generate expensive computational cost.
- I'm interested in solving this expensive optimization problem using an OPTIMIZATION ROBOT, well-trained using the deep reinforcement learning algorithm.
#2. Finite Element Reliability Analysis
- It will be updated soon.
#3. Structural Condition Monitoring by Introducing Bayesian Inference
#4. Probabilistic Finite Element Analysis