QstatLab features
Optimization
8 Constrained and unconstrained – fully flexible definition of constraints
8 Single objective – Genetic algorithms, Gradient descent, Exhaustive search, Monte Carlo simulation
8 Multiobjective optimization – construction and visual exploration of Pareto front of optimal solutions, using a method based on genetic algorithm (NSGA2)
8 Pre-build and customized sequences of optimization methods
8 Optimization using discrete variables
8 Parallel function evaluations
8 Mixtures and linearly dependant variables
Response surface modeling
8 Radial basis functions – six types of bases, computationally fast approximation of irregular and non-linear data relationships
8 Kriging – automatic tuning or manual definition of full kriging model, which works both for simulated and real (noised or repeated) data sets. Numerical stability and efficiency of tuning processes provided by established scientific achievements and tested in industry. Provided are three types of kriging based models:
o Prediction – requires comparatively low number of unstructured data. Provides fast and accurate prediction
o Mean squared prediction error – peaks of this quantity indicate regions of the design space that would benefit of further experiments. It is used for construction of sequential design of experiments
o Expected improvement – a model of the probability of a design being optimal. Used for early detection of optimal regions. This knowledge can be used in conjunction with an optimization algorithm, by steering it towards the optimum, reducing significantly the number of required function evaluations.
8 Multiple regression analysis
o Construction of multi-factorial regression polynomials of up to 4th order
o Significance testing and contribution analysis
o Analysis of residuals
o ANOVA
o Normal and half-normal plots of effects
o Stepwise regression, backward elimination and forward selection
8 Kohonen Neural networks
8 Number of uni-variate curve fitting methods, including splines, non-linear equations, polynomials.
8 Analysis of variance – one way and multi-way ANOVA. Can be used to investigate the effect of variables and interactions.
Robust optimal design
8 Rapid stochastic model construction. Models include – mean, standard deviation and variance
8 Variation of variables can be defined as standard deviations or percentage of current value
8 Four methods of mean and variance computation:
o Analytical – uses mathematical formulae to compute mean and variance. This is the fastest and most accurate method that can be applied to 2nd order regression polynomials.
o Simulation – mean and variance are computed using a cloud of data with user-defined distribution and size. Depending on the problem, the method may require large size of variation cloud.
o Taylor – 1 – using 1st order Taylor series
o Taylor – 2 – using 2nd order Taylor series
o In combination with multiobjective optimization methods, designers can easily search for both optimal and robust (minimal variance) solutions.
Flexible model definition
Models used for optimization can be constructed using various mixture of:
8 Spreadsheet functions
8 Regression polynomials
8 Radial basis functioins
8 Kriging models
8 Kohonen neural networks
8 Curvefit models
8 Scripts – VBscript, JavaScript, any script supported by WHS
8 External executables – allows integration of any commercial or custom built software packages.
8 Stochastic models of mean, variance and standard deviation for the purpose of robust designs
Design of experiments (DOE)
8 Full factorial design with customized number of factor levels
8 Fractional factorial design – uses a specified fraction of the full factorial design, reducing the number of experiments but preserving the orthogonality of the design, which ensures uncorrelated parameter estimates
8 Composite and rotatable designs
8 LPtau – a repeatable Sobol sequence that effectively fills up the design space
8 D-optimal designs – used to construct designs that actively reduce the uncertainty of the design space. Uses a customizable selection of powerful optimization algorithms and design methods.
8 Random designs
8 Build upon existing designs or start from scratch
8 Works with both continuous and discrete variables
8 Graphical visualization of design points and prediction variance
8 Mixtures and linearly dependant variables
Quality improvement
8 Control charts – variables and attributes
8 R&R analysis – gage capability studies
8 Statistical data analysis – hypothesis testing and cross-correlation analysis
8 Discrete values, proportions and chi-squared tests
8 Assessment agreement analysis
Additional tools
8 Spreadsheet pre-loaded with high number of functions.
8 Search, replace and filter functionality
8 Random number generator, using several distributions
8 Automatic updates
See also
other help topics