Chapter 2 Using DesignXplorer 12.0 ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-1 May 28, 2009 Inventory #002670 Parameters from CAD Training Manual • DesignXplorer can utilize user-selected parameters from supported CAD systems. To expose CAD parameters, a special naming convention should be used* (the same conventions apply to DesignModeler). – A “personal parameter key” is used to flag CAD parameters to be exported to Workbench. The default parameter key is “ds”. • Example: – To expose a CAD parameter named “Length” in Workbench, it could be renamed as “dslength”, Lengthds”, “ds_Length”, “Length_ds”, etc. (the order is arbitrary as is the underscore and the case). – The parameter key is user controlled and can be modified by the user from the “Geometry Import” section of the options. * Note: leaving the personal parameter key field blank will cause all CAD parameters to be passed to Workbench. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-2 May 28, 2009 Inventory #002670 Parameter Set… Training Manual • Parameter Set is global throughout the WB applications that support parameters. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-3 May 28, 2009 Inventory #002670 Parameter Set … Training Manual Table of Design Points for What-If analysis • To build “What-If” studies double click on Parameter Set and the Table of Design Points becomes available. • The initial values from the simulation run are listed and the user can enter new values by clicking on the “*” button. • Alternatively, users can copy and paste values from Excel spread sheet. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-4 May 28, 2009 Inventory #002670 Parameter Set … Training Manual Each new “Add” becomes a design point. When all design points are entered choose to “Update all Design Pints.” Status bar gives more info about design points update. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-5 May 28, 2009 Inventory #002670 Parameter Set … Training Manual • Users can also choose to run a selected design point from the list by RMB on the design point “Update Selected Design Point.” • Users can set up selected design point as Current by “Copy input to Current”, automatically Mechanical application will be updated to selected design point. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-6 May 28, 2009 Inventory #002670 Parameter Set … Training Manual • Parameters Charts allow users to configure and plot Input vs. Output in XY plots. • Select input and output parameter and on the Toolbox select Parameter Chart to show the chart output vs. input parameter. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-7 May 28, 2009 Inventory #002670 Design Points Training Manual • Design Points: – A Design Point is a location within the response space which is a result of a specific combination of input parameters. – A simple example: • What is the effect on the first natural frequency of a 25 x 25 mm cantilever beam when the length is varied from 360 to 440 mm? 400 mm • Intuition tells us that as the length increases, the natural frequency will decrease, and the results of a DX solution confirms this (next page). ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-8 May 28, 2009 Inventory #002670 Design Points Training Manual • Here the graph from DX shows 1st frequency decreasing as length increases. • To develop this response curve DX ran 5 complete solutions using 5 different values for the “Length_ds” parameter (360, 380, 400, 420 and 440 mm). • These solutions are indicated by the symbol in the figure at the right. • The graph between these solutions is curve fit using a regression analysis technique (see the documentation for more details). • Thus, choosing any value within the stated range for Length_ds and evaluating the response (frequency), represents a “Design Point”. Design point (typical) Automatic solutions ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-9 May 28, 2009 Inventory #002670 Design Points Training Manual • Any design point that is chosen from the response chart (other than the automatic design point) is a response design point. – Response design points represent approximations based on the response charts. It is good practice to generate a hard design point before proceeding to develop or modify a design based on a response design point. Response Design point (typical) Automatic design point ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-10 May 28, 2009 Inventory #002670 Design Points Training Manual • Design points are generated based on the number of input parameters in the system. Design points can be previewed by clicking the corresponding Preview button in the toolbar. • Output parameter can be calculated for the generated design points in several ways: – From the workspace, click the corresponding Update button in the toolbar to solve for all of the generated design points. – From the Project Schematic, right click the cell and select Update from the menu. – From the Project Schematic, click the Update Project button in the toolbar to update all of the systems in the project. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-11 May 28, 2009 Inventory #002670 Design points Training Manual Additional Sampling Methods Optimal Space Filling Design allows you to fill the design space efficiently with as few number of points as possible while keeping the discrepancy as low as possible. Tools>Options Options to provide a comprehensive goodness of fit report of a standard response surface to determine the adequacy to be applied in Goal Driven Optimization (GDO), Design for Six Sigma (DFSS), and Robust Design (RD) samples. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-12 May 28, 2009 Inventory #002670 Design points Training Manual Central Composite Design For deterministic method the locations of the automatic design points are determine according to a DOE method that is by default CCD with fractional factorial design. •If N is the number of input parameters then CCD consists of: 1. One center point. 2. 2*N axis point located at -a and +a position on each axis of the selected input parameter. 3. 2^(N-f) factorial points located at the -1 and +1 position along the diagonals of the input parameter space. Number of Automatic design points as a function of the number of input parameters Optimal Space-Filling Design •You can create optimal space filling Design of Experiment (DOE) plans according to specified criteria. The option allows you to fill the design space efficiently with as few number of points as possible. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-13 May 28, 2009 Inventory #002670 Design points Training Manual • For the deterministic method, the locations of the generated design points are determined according to a design of experiments method that is by default a central composite design with a fractional factorial design. Response Surfaces will be generated from the associated DOE solution. • Generated design points are temporarily written to the Table of Design Points in the project's Parameter Set Bar during the solution. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-14 May 28, 2009 Inventory #002670 Meta Model Type Training Manual Standard Response Surface – Full 2nd Order Polynomials (default) Assumes n sampling points and for each sampling point the corresponding values of the response parameters are known. Determines the relationship between the input and output parameters and the response parameters based on these sampling points. The resulting approximation of the output parameters as a function of input parameters is called response surface. Second order polynomial is preferred. Kriging Comprehensive goodness of the fit of standard response surface Y(x)=f(x)+Z(x). Combination of polynomial model plus Z(x)-realization of a normally distributed Gaussian random process with mean zero, variance s^2 and non zero variance. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-15 May 28, 2009 Inventory #002670 Meta Model Type Training Manual Non Parametric Regression Support vector method. Used hyperplane to categorize a subset of the input sample vectors which are deemed sufficient to represent the output in question. Neutral Network Mathematical technique is based on the natural network in the human brain. Weighted functions are issued from the algorithm which minimizes the distance between the interpolation and the known values (design points) – learning process. The error is check in every iteration. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-16 May 28, 2009 Inventory #002670 Goal Driven Optimization Training Manual • Preserve Design Points After DX Run in the cell's Properties view, also this can be set up in Tools>Options>Design Exploration dialog. • Preserving design points has the advantage of letting the user see the exact design points used by a Design Exploration cell. In a Parameters Correlation analysis for instance, it is the only way to see the actual design points used to calculate the correlation since there is no preview for that cell. • Preserving design points will allow subsequent Design Exploration update to reuse the existing Design Points, which will reduce the CPU cost. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-17 May 28, 2009 Inventory #002670 Goal Driven Optimization Training Manual • The Screening approach is a non-iterative direct sampling method by a quasi-random number generator. • The MOGA approach is an iterative Multi-objective Genetic Algorithm, which can optimize problems with continuous input parameters. • NLPQL is a gradient based single objective optimizer which is based on quasi-Newton methods. • MOGA is better for calculating the global optima while NLPQL is a gradient-based algorithm ideally suited for local optimization. So you can start with Screening or MOGA to locate the multiple tentative optima and then refine with NLPQL to zoom in on the individual local maximum or minimum value. Problems with mixed parameter types (i.e., usability, discrete, or scenario parameters with continuous parameters) or discrete problems cannot currently be handled by the MOGA or NLPQL techniques and in these cases you will only be able to use the Screening technique. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-18 May 28, 2009 Inventory #002670 Goal Driven Optimization ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. Training Manual 2-19 May 28, 2009 Inventory #002670 Goal Driven Optimization Training Manual • The screen below shows various parametric preferences and their importance specified on the Goals and Candidate Designs screen. • Once goals are stated, a set of samples is generated from which candidate designs will be obtained. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-20 May 28, 2009 Inventory #002670 Goal Driven Optimization Training Manual • At least one of the output parameters should have an Objective of Maximize, Minimize, or Seek Target in order to do optimization with the MOGA or NLPQL methods (only one output can have an objective for the NLPQL method). • If this is not done, then the optimization problem is either undefined (No Objective) or is merely a constraint satisfaction problem (Objective set to >= Target or <= Target or = Target . When the problem is not defined, the MOGA or NLPQL analysis cannot be run. • Screening method does not depend on any parameter settings and can be used to perform preliminary design studies. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-21 May 28, 2009 Inventory #002670 Goal Driven Optimization Training Manual • After generating a sample set, candidate designs are generated and displayed along with parameter rankings. • Parameter values are displayed along with rankings corresponding to stated goals. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-22 May 28, 2009 Inventory #002670 Goal Driven Optimization Training Manual • Another useful feature of GDO is the ability to visualize the impact that variations in one parameter may have on another. • These are called “Trade Off Plots” and are available in either 2D or 3D, depending on the number of parameters. – Trade off plots use the concept of Pareto optimal points which are discussed later. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-23 May 28, 2009 Inventory #002670 Goal Driven Optimization Training Manual • Regardless of how a candidate design is chosen (from response charts or goal driven optimization), it can be stored for future study as a design point. • From Goal Driven Optimization: ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-24 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual • We will walk through the process of setting up a scenario in DX. • In this example we’ll use DesignModeler as our geometry source. There is no difference to the process when using a commercial CAD system. • We begin by specifying the geometry parameter(s) in our DM model. Geometry parameters ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-25 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual Input and output parameters from Mechanical application. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-26 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual • Once the analysis model is prepared, DX is started from the Project Schematic. Double click on Goal Driven Optimization in the Design Exploration Toolbox. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-27 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual • At this point, only design points and parameter definitions and ranges are specified. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-28 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual We can Preview all design points at this point, no outputs are calculated. By updating design of experiments, output parameters are calculated. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-29 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual • From the views section, “Responses” can be chosen to review the calculated data. • The results information can be viewed in several formats listed under Response Charts. • The response charts allow potential design decisions to be evaluated visually. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-30 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual • As seen, response charts are means of exploring the entirety of a design and selecting desired design points visually. Goal Driven Optimization is a continuation of generating a response surface. Goal Driven Optimization is a three step process: – Generate samples – Specify design goals – Generate candidate designs • Before generating candidates, DX must generate a group of sample design points. – The number of samples is user specified. The more samples generated, the more refined the candidate designs will be. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-31 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual • The second step is specifying the design goals. – The values as well as the importance of each input and response parameter is set by clicking in the appropriate field. • The third step is to generate candidate designs for evaluation. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-32 May 28, 2009 Inventory #002670 Creating a DX Scenario - example Training Manual • Three candidate designs are generated based on the stated goals. • Once a suitable design is chosen it can be stored as a response design for further study. • Numerous candidate designs can be saved as response designs. Before finalizing a design based on this data, a design point should be generated to insure the validity of the choice. ANSYS, Inc. Proprietary © 2009 ANSYS, Inc. All rights reserved. 2-33 May 28, 2009 Inventory #002670