Experimental Modeling (Identification)
- Building models from data, not from first principles (physical, chemical, ... laws)
- Improving models from first principles with the help of measurement data (grey-box models)
- Metamodeling: Fast models than approximate complex, time-consuming calculations. Suited for dynamic simulation, optimization, real-time application
- Nonlinear and/or dynamic models
Design of Experiments
- Where to measure?(not location but which operating points?)
- How to describe the operating regime boundaries?
- Which inputs are how relevant for modeling?
- Which inputs have strongly nonlinear effects?
Classification, novelty detection, extrapolation detection, ...
Models: What For?
- Models are the basis for most advanced techniques in many disciplines.
- Model-based techniques can be divided into a (sometimes iterative) two-step procedure:
1. Building a model.
2. Using a technique based on this model.
- Applications of models:
Models: What From?
1. Experimental Modeling (Identification)
- Local model networks.
2. Design of Experiments (DoE)
- Optimal maximin latin hypercubes (unsupervised).
- Optimization of next measurement point to gather most information with HILOMOT-DoE (supervised).
3. Automatic Input Selection
- For CFD, FEM, and Look-up Tables.
5. Nonlinear Dynamic Models
- Local model networks with OBF and FIR.
- Least squares support vector machine.
7. Design of Excitation Signals for Identification
- Combustion engines
- Driveability calibration
- Automatic transmission modeling
- Structural health monitoring
- Metamodels for CFD simulations
- Metamodels for FEM simulations
- Location estimation for inductive charging