The info (main) panel has fields for entering problem description and author information.
In the Strategy Panel you may choose the sampling strategies for tasks like metamodel-based optimization and RBDO.
- Single Stage: Sampling points are chosen only once and the whole design space is taken as the region of interest. Therefore no iteration is needed.
- Sequential: Sampling is done sequentially, only a small number of points are chosen for each iteration.
- Sequential Strategy with Domain Reduction: It is the same as the Sequential Strategy, but in each iteration the domain reduction strategy is used to reduce the size of the region of interest.
- Choose a strategy to apply in the task.
- For the latter two strategies, choose a tolerance strategy for termination and specify the tolerance values.
The Solver Panel allows the user to define:
- Pre-Processor Package
- Change of input files by a parametric Pre-Processor controlled by design variables from LS-OPT (ANSA, HyperMorph, DEP-Morpher, TrueGrid, User-Defined).
- Solver Package
- Solver Command: LS-DYNA or others (User-Defined)
- Input File: Will be appended to the Command in the Command Line
- Post-Processor Package
- Interface to METAPost for result extraction
The probabilistic component of a design variable is described using a probabilistic distribution. The distributions are created without referring to a variable. Many design variables can refer to a single distribution. As statistical distribution you can choose between these types:
Normal, Uniform, Lognormal, Weibull, Beta, User PDF (probability density function), User CDF (cumulative distribution function), Binomial
Add a Distribution
In the Variables Panel you can define or modify variables and choose between different types:
Variable, Constant, Dependent, Noise Variable, Discrete Variable
Add a Variable
Variable means a standard design variable which will be integrated in the optimization process.
Add a Constant
Constants are used:
In the sampling panel you can choose between these metamodels:
Polynomial, Sensitivity, Feedforward Neural Network, Radial Basis Function Network
Furthermore, several options and point selections can be selected, e.g. "Compute Global Sensitivities" and "Approximate Histories" (calculation of metamodels for curve data).
In the Histories panel you can specify time history curves to be extracted from the solver database.
Create a response in the Responses Panel. A response can be a result of a mathematical operation applied to a time history curve, or can be directly extracted using the standard LS-DYNA interface or a user-defined interface.
Specification of the objective (function to minimize) for the optimization problem.
Soecitication of constraints (functions to satisfy) of the optimization problem.
In the Algorithms panel you can specify an optimization algorithms for the task. There are three core solver and two hybrid algorithms to choose.
- LFOP: Leap Frog Optimizer
- GA: Genetic Algorithm
- Hybrid GA: Hybrid Genetic Algorithm
- ASA: Adaptive Simulated Annealing
- Hybrid ASA: Hybrid Adaptive Simulated Annealing
- Specify an optimization algorithm.
In the Run panel you can start the optimization, specify the number of concurrent jobs or adjust the convergence criteria. The state of completion is monitored.
The View panel starts the LS-OPT Viewer for the visualization of the optimization results. For example Metamodel Surface, Sensitivity Analysis, Optimization History....
You can plot the metamodel surface you have built. The plot can be zoomed and rotated. You may also add the sampling points (both feasible and infeasible) and the optimum points (predicted and computed) to the plot. This is a visual way to check the results.
Which variable is the most important for a response? The sensitivity analysis shows the influence of all the variables on the selected response. There are two analysis methods to choose.
- Linear ANOVA (Analysis of Variance)
- GSA/Sobol (Sobol's Global Sensitivity Analysis)
Progress of a variable, a response or other entities during the optimization process. If the strategy Sequential with Domain Reduction is applied, the blue curves show the shrink of the domain of a variable.
In the DYNA stats panel you can compute various statistics of the LS-DYNA d3plot results and LS-OPT history data for viewing in LS-PREPOST on the FE model. These statistics show:
- The variation of the LS-DYNA results due to the variation of the design parameters.
- The variation of the LS-DYNA results due to bifurcations and other stochastic process events.
Fringe Plot of the z-displacement of a steel tube being crushed