Parametric. Practices. Patrick Cunningham. CAE Associates Inc. and ANSYS Inc. Proprietary 2012 CAE Associates Inc. and ANSYS Inc. All rights reserved.

Transcription

1 Parametric Modeling Best Practices Patrick Cunningham July, 2012 CAE Associates Inc. and ANSYS Inc. Proprietary 2012 CAE Associates Inc. and ANSYS Inc. All rights reserved.

2 E-Learning Webinar Series This the first of a series of E-Learning webinars to be offered by CAE Associates. If you are a New Jersey or New York resident you will earn continuing education credit for attending the full webinar, participating in the polls and completing a survey at the conclusion of the webinar. Other E-Learning webinar topics coming up are: CAD Cleanup August 14 th and August 16 th Meshing Part I - 3D Solids August 28 th Meshing Part II - Shells and Beams September 11 th Visit for details. 2

3 Topics 1. Benefits of a parametric modeling approach to analysis. 2. Defining input and output parameters. 3. Using the Parameter Set 4. Parameter Correlation 5. DesignXploration DOE Matrix Response Surfaces Goal Based Optimization Six Sigma Analysis 3

4 Why Model Parametrically? Reasons for creating models that are parametric: 1. If you are using a CAD system, you are probably already doing it. 2. You can gain a deeper understanding of your design at very little cost. 3. You can eliminate unwanted surprises in the design process. 4

5 Parametric Modeling Steps The setup of a parametric modeling environment involves the following steps: 1. Identify the input parameters. These are quantities that may alter the response. Input parameters can consist of the following: CAD parameters Material coefficients Loading magnitudes Mesh controls 2. Identify the output parameters. These are the responses of the system that you are interested in such as: Mass and volume Deflection, strain and stress Reaction force Contact t gap and pressure User Defined Results 5

6 Parametric Modeling Steps (continued) 3. Evaluate parameter sensitivity. 4. Perform Design of Experiments analysis using the critical the design parameters. 5. Curve fit the DOE results. 6. Find the optimum design configuration. 7. Use a statistical approach to determine the robustness of your design. 6

7 Parametric Modeling Example Consider the lower jaw of a vice grip. We would like to determine the optimum shape of the jaw based on the stress limit of the material, the amount of deflection under load, and the force required to close the jaw. 7

8 Demo Geometry For this example we will be using SolidWorks as the CAD modeling tool. A simplified representation of the lower jaw will be analyzed. Pivot_horz Jaw_length Jaw_thickness Pivot_vert Jaw_OD 8

9 Parametric Modeling The parametric modeling approach begins with the CAD model. Identify the features on the CAD side that may affect the design. Locate the dimensional parameters of the features and make them identifiable to the ANSYS Workbench environment. Select the parameter from the sketch. Rename the parameter with a prefix (Example MP ). 9

10 Parametric Modeling Named Selections can also be created at the CAD level. Named Selections can be used for model setup in the Mechanical environment. 10

11 Loading the Parametric Model into ANSYS There are two ways to launch ANSYS and connect to the CAD geometry: 1. Select Workbench from the plug in menu in the CAD tool. The connection to the active CAD model will me made automatically. 2. Run Workbench as a stand alone process and drag and drop a Geometry component or an Analysis System onto the project page. RMB click on the Geometry row to select the active CAD model. 11

12 Loading the Parametric Model into ANSYS With the geometry connected to the active CAD file, define the Properties to include CAD parameters as well as other items such as Named Selections and Coordinate Systems. Use the Parameter and key to filter out the specific parameters of interest. 12

13 Defining Geometry Input Parameters Double click (or RMB > Edit) the Model row open the Mechanical tool. Select the part in the Geometry section and look for the shared CAD parameter data in the Details menu. Click on the box to the left of each CAD parameter to add it to the Workbench Parameter Tool (indicated by the capital P in the box). Note: If DesignModeler is used this step is accomplished there. 13

14 Model Setup Apply all loads and supports in the Mechanical interface. If you have Named Selections that were imported from the CAD model you can use them to define loads and supports. 14

15 Input Parameters for Loads If any of the load magnitudes will be varied parametrically add them to the Parameter Set by clicking on the box to the left of item. 15

16 Output Parameters Output parameters consist of the result quantities that you want to keep track off. Output quantities can consist of the following: Mass and Volume from the Details menu of a body (in Geometry). Minimum or Maximum quantities for standard result items in the Solution folder (deformation, strain, stress, reaction force, User Defined Results, contact pressure, etc.). Scalar parameters defined in a post-processing command block. Pulling scalar parameters from an APDL postprocessing command block. 16

17 Material Input Parameters To identify material coefficients as input parameters return to the Project page and double click on the Engineering Data row to open it. Identify the material coefficient that you would like to define parametrically by clicking on the box in column E of the Properties window. 17

18 Working with Parameters Return to the Project Page and double click on the Parameter Set. 18

19 Working with Parameters You can set up design points and generate solutions by filling in the Table of Design Points and solving. The Output Parameters will be updated as each design point solution is completed. A table of design points can be pasted into the Parameter Set. 19

20 Solving Design Points Using ANSYS RSM The Remote Solve Manager is an ANSYS utility with which you can identify other computers on your network that can be used to generate design point solutions. You can identify single remote solver machines or a cluster of machines. 20

21 Solving Design Points Using RSM If your CAD program is not installed on the remote machines you will need to generate design point updates locally. Set the Update Option in the Properties of the Parameter Set to Run in Foreground. With this configuration the design point files are generated in series and submitted to the Remote Solve Manager. 21

22 Submitting Design Points to RSM To use an RSM cluster to generate each of the design point solutions select the Solution row of the Analysis System and set the Update Option to Submit to Remote Solve Manager. With a cluster of machines you can solve generate the design point solutions simultaneously. 22

23 Which Parameters are Important? The Parameters Correlation tool in ANSYS Workbench can be used to evaluate importance of an input parameter to the response. Drag and drop the Parameters Correlation tool onto the project page and edit. 23

24 Parameters Correlation Identify which parameters you would like to include in the Parameter Correlation and define a value range for each parameter. 24

25 Parameters Correlation Select the Parameters Correlation (A2) and define the properties in the window below: Preserve Design Points Correlation Type : Spearman (quadratic) or Pearson (linear) Number of samples Auto Stop Enabled (based on the defined accuracy criteria) 25

26 Parameters Correlation Use the Preview button on the tool bar to review the correlation design points and the Update button to generate the design point solutions. 26

27 Parameters Correlation When the design point solutions are complete review the correlation matrix to evaluate the relationship between the input and output parameters. The closer the value is to 1 (or -1) the stronger the relationship. 27

28 Parameters Correlation The Correlation Matrix has four quadrants: Q1 and Q3 are the input to output strengths and are symmetric. Q2 is the correlation between input parameters. Q4 is the correlation between output parameters Input to Input Input to Output Output to Input Output to Output 28

29 Parameters Correlation The Correlation Matrix indicates the following: Strong coupling of input P1 to P8, P2 to P7 and P9, and P4 to P6. Weak coupling of P3 and P5 to all output parameters. 29

30 Determination Matrix The quadratic coefficient of determination (also called R-square) tells you what proportion of the variation between the data points is explained or accounted for by the best fit curve (quadratic) fitted to the points. It indicates how close the points are to the curve. It provides a measure of how well future outcomes are likely to be predicted by the model. 30

31 Goal Driven Optimization Return to the project page and drag and drop the Goal Driven Optimization tool onto the project. Double click in the Design of Experiments row to open it. 31

32 Goal Driven Optimization Uncheck the input parameters that displayed a weak coupling with the output in the Parameters Correlation (P3 and P5). Select a Design of Experiments Type from the pull down menu. 32

33 Goal Driven Optimization Select the active parameters and set the desired parameter range for each parameter. Use the Preview button on the Toolbar to view the design points. 33

34 Goal Driven Optimization Select the Update button to generate the design point solutions. If you have specified the Remote Solve Manager as the update option for Mechanical the design points will be solved simultaneously. 34

35 Goal Driven Optimization When the design point solutions are complete you can perform a cursory check of the response by viewing input versus output parameters as line plots. 35

36 Goal Driven Optimization For a more detailed evaluation return to the project page and double click on the Response Surface row. 36

37 Response Surfaces Select Response Surface (A2 of the Outline view) Pick a response surface type under Meta Model in the Properties view. Click on the Update button in the toolbar to generate the response surface. 37

38 Response Surface Types There are several choices for Response Surface types: Standard Response Surface 2nd-Order Polynomial (default) Effective when the variation of the output is smooth with regard to the input parameters. Manual refine is available. Kriging Efficient in a large number of cases. Suited to highly nonlinear responses. Do NOT use when results are noisy; Kriging is an interpolation that matches the points exactly. Always use verification points to check Goodness of Fit. Auto and manual refinement is available. Non-Parametric Regression Suited to nonlinear responses. Use when results are noisy. Manual refine is available. Neural Network Suited to highly nonlinear responses. Use when results are noisy. Control over the algorithm is very limited. Manual refine is available. Sparse Grid Suited for studies containing discontinuities. Use when solve is fast. Auto and manual refinement is available. 38

39 Response Surfaces Goodness of Fit Check the Goodness of Fit for the response surface type (note: Kriging fits to all design points by definition). 39

40 Response Surfaces Set the response surface mode to 2D or 3D and select the variables to be plotted. Display the design points on the response as a further evaluation of goodness of fit. 40

41 Local Parameter Sensitivities You can plot local parameter sensitivities in the Response Surface window. Local parameter sensitivities differ from Global Correlation sensitivities in that they are based on the difference between the minimum and maximum value obtained by varying one input parameter while holding all other input parameters constant. 41

42 GDO Sensitivity vs Parameters Correlation The sensitivities available under the Six Sigma Analysis and the Goal Driven Optimization views are statistical sensitivities. Statistical sensitivities are global sensitivities, whereas the single parameter sensitivities available under the Responses view are local sensitivities. Global statistical sensitivities are based on a correlation analysis using the generated sample points, which are located throughout the entire space of input parameters. Global statistical sensitivities do not depend on the values of the input parameters, because all possible values for the input parameters are already taken into account when determining the sensitivities. Local parameter sensitivities are based on the difference between the minimum and maximum value obtained by varying one input parameter while holding all other input parameters constant. As such, the values obtained for local parameter sensitivities depend on the values of the input parameters that are held constant. 42

43 Goal Driven Optimization Return to the project page and double click on the Goal Driven Optimization row. 43

44 Goal Driven Optimization Goal Driven Optimization is a technique in used to identify the best possible designs are obtained from a sample set. The results are relative to the goals you set for parameters as well as the goodness of fit of the response surfaces. 44

45 Goal Driven Optimization Candidate designs are curve fits to the response functions. RMB click on a candidate and choose Verify by Design Point Update to generate a deterministic result. 45

46 Goal Driven Optimization Compared the verified result to the candidate result when the solution is complete. 46

47 Six Sigma Analysis To determine the design robustness return to the project page and drag and drop a Six Sigma Analysis onto the project. 47

48 What is Six Sigma Analysis? Typical analyses assume a fixed value for each input quantity. Design For Six Sigma provides a mechanism to include and account for scatter in input and provide insight into how they affect the system response from a probabilistic standpoint. A product has Six Sigma quality if only 3.4 parts out of every 1 million fail to meet design criteria for stress limits safety factors, etc.. The Six Sigma designation includes a15sigmashifttotakeintoaccount 1.5 to take into account variability in manufacturing processes over time. 48

49 Six Sigma Analysis Double click on the Design of Experiments row to edit the Six Sigma analysis setup. Click on each input variable and define the Distribution Type. 49

50 Six Sigma Input Data There are several distribution types that are used depending on the type of data available: 50

51 Six Sigma Input Data Distribution Types: 1. Uniform Used when only the limits of the input data are known (such as a min-max tolerance on a manufacturing dimension). Uniform 2. Ti Triangular Used when additional background information is available. Example: the machinist also knew from experience the upper and lower bounds as well as the most common value. Triangular 3. Normal Describes a Gaussian Normal distribution (bell curve). Normal 4. Truncated Normal Gaussian distribution with the outer extremes removed. 5. Log Normal - Typically used to describe the scatter of the measurement data of a physical phenomena. 6. Exponential used where there is a physical reason that the probability density strictly decreases as the uncertainty variable value increases. Lognormal Truncated Normal Exponential 7. Beta Useful for random variables that are bounded at both sides. 8. Weibull - Most often used for strength or strength-related life cycle parameters. Beta Weibull 51

52 Six Sigma Input When distribution type and limits for each of the input parameters are defined click Update on the toolbar to generate the new DOE results. 52

53 Six Sigma Output Once the design point solutions are complete update the Six Sigma analysis row. Double click to open it and review the probability distributions of the output parameters. 53

54 Six Sigma Compliance A product has Six Sigma quality if only 3.4 parts out of every 1 million fail to meet design criteria. To access the results for each output parameter select the parameter and set the Probability Table to Percentile-Quantile 54

55 Six Sigma Compliance Scroll down to the bottom of the Probability Table and enter in a value of (1-(3.4/1e6)). Note that in this case 3.4 out of 1,000,000 parts will experience an equivalent stress of PSI or greater. If a PSI stress is acceptable based on design criteria a Six Sigma quality criteria is satisfied. 55

56 Thank you! Thank you for attending CAE Associates webinar on parametric modeling. You will receive via a survey to fill out and return. We welcome any comments or additional questions on the content. A transcript of this presentation can be downloaded from our website: com 56