Why Use Design of Experiments (DOE)

Manufacturing Industry Title SlideAn Introduction
© 2011 Autodesk
Why Use DOE
Use DOE to understand The sensitivity of results to processing parameters or thickness changes
What influences the warpage of the part
How critical dimensions are influenced by processing
DOE can be used for a Wide scope
To get an overall understanding of processing variable sensitivity
Narrow scope Determine what needs to be done to solve a problem
DOE Objectives Must have a defined set of objectives to use DOE effectively
Without objectives the DOE analysis may not be set up correctly and will be difficult to interpret the results
© 2011 Autodesk
Why Use DOE?
Look at how processing parameters influence results Possible uses for DOE include:
Example Parameters Results Why
Volumetric shrinkage Sink mark depth Injection pressure
Determine the sensitivity of the processing parameters on sink marks and how uniformly the part can be packed out
2 Melt temperature Injection time Pack profile
Clamp force Injection pressure Volumetric shrinkage
Determine if the mold will fit in the desired press size and uniformly pack out the part
3 Nominal wall thickness Rib thickness Inj. & pack pressure
Shear stress Bulk temperature at EOF Sink mark depth
Look at how thin the nominal wall and ribs can be to make the part and how the thickness will influence the moldability of the part
4 Melt temperature Each coolant inlet temp
Cycle time Mold surface temperature Volumetric shrinkage
Determine how the coolant inlets need to be set to make the mold surface temperatures more uniform and how reducing hot spots improves shrinkage
5 Injection time Melt temperature Packing profile
Volumetric shrinkage Flatness of an edge Dimension between 2 pts
What is the sensitivity of the deflection (warpage and shrinkage) to process inputs? Will the molding window be small? Should the part be stiffened?
© 2011 Autodesk
DOE Inputs
Any numerical input into the analysis sequence Processing parameters Boundary conditions Thickness
{ } Melt temperature [+120, +160]°C
{ } Mold temperature [+60, +80 ]°C
{ } Hydraulic pressure [+20, +40]MPa
{ } Pack Time [+5, +10]s
{ } Flow rate [+250, +300]mm3/s
{ } Cooling inlet Ø [+5, +10] mm
© 2011 Autodesk
- Single analysis
Mold temperature
Melt temperature
Flow rate
Moldflow® runs
Runs a series of simulations based on
statistical models
simulations
Model types: All available, (Midplane, Dual-domain, 3D)
Input variables: Common processing parameters (single values and profiles)
Output quality criteria: Processed common analysis results
Results generated: Variable % of influences on quality criteria Quality response surfaces Variable values for targeted quality optimum
Midplane Dual-domain 3D
DOE Analysis Modes
1. Influence analysis Influence of each variable Taguchi design based on orthogonal design
2. Response analysis Quality response surfaces Optimum variable values
3. Combined Influence then Response analysis Filter ‘N’ most influential variables before determination of the quality responses.
Optimum Solution
Mold temperature
Melt temperature
Flow rate
© 2011 Autodesk
Observed Quality Criteria
Require one numerical value from each DOE run Method in a simulation environment: Combine results over the entire model or by zone (group, layer, …) Simple arithmetic Difference between single properties Average Minimum Maximum Range Standard deviation
Geometric arithmetic Distance between two points
© 2011 Autodesk
Quality range and estimate of optimum (using the response equation)
DOE Log Results
DOE CSV Result File
The CSV (comma separated values) file is located in the project folder
Written by default but can be turned off using: ‘DoE Builder’ / ’Options’ , uncheck: “Save results in CSV files”
The file includes: All raw results (variable design pattern, quality values) Quality response equations Range and optimum results Individual file location for each design run when the runs are saved using ‘DoE Builder’/’Options’, check: ’Keep analysis result in temporary folder’
© 2011 Autodesk
© 2011 Autodesk
© 2011 Autodesk
DOE Results – Contour
© 2011 Autodesk
Special Components
For variables – “Dimension scale factor” Affect beam diameters and shell thicknesses for midplane and dual-domain meshes The selection of elements is based on custom ‘layers’ This method allows the mixing of element types
For qualities – “New Warp Criteria”
Custom quality criteria based on nodal warpage deflections
Save list Mesh menu
Node list With custom axis & anchor plane
Boundary conditions menu
Handling of Short-Shot
The DoE solver will automatically try to reduce the variable range.
The number of retry / range reduction is controlled using: “DoE Builder” / “Options”, Fill: “Max number of relaunch”
Original design Modified design
Design of experiment Directed optimum search
Primary objective Analyze the process environment Find the optimum
Analysis runs Predefined before execution Defined step by step
Favored execution Parallel Sequential
End of execution Completion of initial set of runs Accuracy criteria or maximum number of runs
Process mapping Homogeneous coverage Defined by step path
Optimum Estimate based on fitted equation Accurate
© 2011 Autodesk
Design Selection
Variable range are mapped to [-1,+1] Design type is set by the result objective: influences or response The most basic design is called “Factorial” For analysis of influences Use the two values {-1},{+1} The Factorial design requires 2variables runs ( 8 runs for 3 variables ).
AMI DoE use the Taguchi design for analysis of influence Orthogonal sets are used to maximize the number of variables for a minimum of runs ( 8 runs for 7 variables)
Taguchi design L8
Factorial Orthogonals
© 2011 Autodesk
Design Selection
For analysis of response The fitting of a quadratic equation (x2) requires three values per variable: {-1}, {0}, {+1} Factorial design for response use: nRuns = 3variables
27 runs for 3 variables 81 runs for 4 variables
However, a direct fit of a quadratic equation (x2) requires 10 runs. Factorial is usually replaced by designs with lower run counts.
AMI DoE use the face centered design which uses: nRuns = 2nVariables + 2* nVariables + 1 15 runs for 3 variables 25 runs for 4 variables
Face-centered design – 4 variables
© 2011 Autodesk
A variable with very low direct influence on a quality criteria but large interactions with other variable(s) may sometimes receive a large influence result
Taguchi design is composed of a Factorial design and Orthogonal sets
Taguchi design maximize the number of variables for the minimum number of runs
The price is that interactions between variables are mixed within the influence results However, this is a low probability risk
Limitation on Analysis of Influence
Taguchi design L8
© 2011 Autodesk
Limitation on Analysis of Response
The response equation is an approximated fit of the quality criterion behavior
1. DOE uses more runs than mathematically required for a fitting to cover the experimental range For two variables, X & Y:
2. The “Least square” method is used to compensate but provides an approximate fitting
6 points 9 points
To cover entire range
{ least square method - http://en.wikipedia.org/wiki/Least_squares}
coefficients
Least square method = Fit a mathematical equation to fit the behavior of scattered points
© 2011 Autodesk
Impact of the Response Limitation
One extreme point like a near short-shot may destabilize the fitting operation.
The result of a destabilize fit can be a negative value when the real value is positive only. For example, a negative fill-time (illustrated above).
Time
Why Use DOE
Why Use DOE?
DOE Results – Contour
Design Selection
Design Selection
Slide Number 23