eArtius - ANSYS Optimization Add-in webinar

©Copyright eArtius Inc 2012 All Rights Reserved April 30, 2012

eArtius and ANSYS Stretch the Limits of Multi-Objective Design Optimization

Vladimir SevastyanoveArtius, Inc., Irvine, CA 92614, USA

[email protected]

Boosting Optimization Standards


Agenda:

eArtius Technology Overview

eArtius-ANSYS Optimization Add-in

Getting Started

Demo

Optimization Results

eArtius Optimization Technology in Detail


eArtius Technology Overview

• eArtius is a company dedicated to development of a new innovative Design Optimization Technology over last 14 years (US patents #6,417,852, #7,593,834, #8,041,545):

• Software to Optimize Complex Designs that have predictive mathematical models– Hours/days to optimize high value designs

• Aerospace, automotive, turbo machinery, electronics, chemical processing, ship design, weapons systems

– Main product—a PIDO application Pareto Explorer– Other products: plug-ins for Simulia Isight, Noesis OPTIMUS, ESTECO

modeFrontier, ANSYS Workbench

• Breakthrough Optimization Algorithms– Orders of magnitude faster than other algorithms– Directed optimization on Pareto frontier– Thousands of design variables

• Partnership with ANSYS since middle of 2011– Two new products eArtius-ANSYS Optimization Add-in (local and remote)

have been developed for ANSYS users since then


Intevac Inc, Santa Clara, CA served as first beta site for eArtius ANSYS Add- in. Dr. V. Kudriavtsev (Intevac) provided demo problems and valuable inputs to make plug-in more suitable for ANSYS multi-physics users. Intevac utilized plug-in for the development of optimal high power heating equipment for its new c-Si solar cell manufacturing (Lean Solar) and for its hard disk media deposition product lines.

http://www.intevac.com

Acknowledgement

Intevac c-Si Technology


Current Computational Design Process

Comput er

8 threads i7 CPU

240 cores TESLA Graphic Processing Unit GPU (x2)

slowest component

(meetings, reviews, alignments, cancellations)

Ingenious

Solutions

Human Thinking

and Analysis

fastest component and grows exponentially faster

D E L A Y


eArtius Optimization Add-in is Now Available among Workbench Components

eArtius Optimization Add-in is now available as an add- in to ANSYS Workbench.

It can be found in the Design Exploration toolbox

With little more effort than for a single run, you can use eArtius to drive ANSYS Workbench

Leverage the parametric and persistent power of ANSYS Workbench with the eArtius Optimization Add-in


Synergy of ANSYS & eArtius Technologies

According to a Survey performed by ANSYS, there are some obstacles to optimization (see the diagram)

ANSYS and eArtius are focused on removing the most significant obstacles:– ANSYS Workbench:

• Allows easily create parametric models,

• and integrate add-ins with WB

– eArtius Optimization Add-in:• Removes all integration issues; looks and behaves like part of WB• Reduces integration cycle and learning curve to minutes• Significantly reduces the number of design points required for a given

number of parameters

Optimization Add-in is designed specifically for ANSYS users, and has excellent ROI value


Multi-Objective OptimizationeArtius optimization algorithms solve optimization tasks with multiple objectives. For

instance, we need to minimize weight and cost, and maximize an engine efficiency. The solution of the optimization task is a set of trade-offs (set of Pareto optimal points) found by

an optimization algorithm.Why it is so important to use Pareto optimal designs?

– Because for any non-Pareto optimal design we can find at least one optimal design which is better with respect to all objectives. It does not make any sense to use non-Pareto optimal designs.

Point C is not on the Pareto Frontier because it is dominated by both point A and point B.

Points A and B are not dominated by any other, and hence do lie on the frontier.

Example of a Pareto frontier. Smaller values are preferred to larger ones.


There are two separate eArtius-ANSYS Optimization Add-ins—local and remote

Local Add-in is build into ANSYS Workbench GUI, and performs optimization locally

Remote Add-in is also integrated with ANSYS WB in the same way as local one, but optimization is performed remotely by eArtius stand alone software Pareto Explorer (PE)

PE communicates with remote eArtius-ANSYS Optimization Add-in via Internet, and allows to monitor and plot design points in real time

Remote Add-in is designed for advanced users

This presentation gives and idea about both add-ins, but it is focused more on the local one

eArtius-ANSYS Optimization Add-ins


eArtius Consol Optimizers

(consoleapplications)

eArtius Pareto Explorer(Desktop Windows

application)

eArti

us A

dd-in

ANSYS Workbench(GUI application)

eArtiusLocal Add-in

eArtiusRemote Add-in

Overall Scheme of Interaction


We will demonstrate use of add-in using Stress and Deformation of Cantilever

Beam problem as a backdrop

Min Deformation, Min Max Stress, min Weight These are conflicting objectives.

deformation

Force

length

height

width

Fixed


Getting Started

1. Install the Add-in

2. Add to project3. Transfer i/o parameters

After installation eArtius Optimization Add-in appears in the Design Exploration section


Model SetupWorkbench Parameter Set ScreenWorkbench Main

Screen


4. Define a simulation model and select an optimization algorithm

Optimization Model Setup

As follows from the Model Properties screenshot, a few types of design variables are supported: Constant/Double/Integer/Shortcut

Output variables can be set as constraints, minimized/maximized objectives, or as ignored (No Action) variables

There is an option to formulate an output variable as an algebraic expression based on existent input/output parameters—see Formula Editor


5. Specify parameters of the algorithm

6. Start optimization by clicking on ‘Update’, and watch logs and a progress bar

Optimization Properties


7. Check the optimization results in the local mode

Grid allows to see all evaluated solutions and optimal designs (green):

Charts allow to click on a marker, and see all properties of the selected point:

Plot Results

Selected point can be set as an initial point for further improvement in the following optimization session

deformation

stressMass

deformation


Safety Factor

Force

Von Mises Stress

deformation deformation

deformation

deformation

length

height

width

Force

deformation

Selected Results


8. Observe runtime optimization results in the remote modeeArtius Pareto Explorer is a full featured design optimization tool with a library of optimization algorithms and powerful post-processing capabilities

Pareto Explorer is a part of remote eArtius-ANSYS Add-in. It performs optimization and exchanges data with ANSYS WB via an HTTP connection

Pareto Explorer has OpenGL-based interactive 2D/3D graphics, and allows observing and analyzing optimization results in runtime.

Getting Started


OPTIMIZATION EXAMPLES


Flat Sheet of Aluminum serves as radiator heat sink, dumps heat via natural convection and radiation to ambient. Heated locally in a small area by

electronics heat source with load from 50 to 5000W. Need to minimize heat sink area (mass), determine if it is a suitable solution for high power range.

Max allowable temperature is limited in 70…150 deg. C

Case Study 1: Heat Sink Thermal Optimization

Heat Flux

Convective and Radiative Cooling

emissivity1emissivity2

Alpha, heat transfer coefficient


Model Setup


Model Setup

Selected design can be set as an initial point for further improvement by an optimization algorithm

COMBI optimization algorithm builds the optimization strategy based on available time resource. It does not require any knowledge or training from users.


VIDEO 1

How to setup a model for optimization



Tmax

Tmin

deltaTmax

Heat Flux



HMGE optimization algorithm has been used for optimization of the model. Pareto front looks like an almost linear curve in the 3-dimensional criteria space. It is filled by Pareto optimal points evenly, which allows to choose the best trade-off precisely.

Pareto front includes at least two disjoint areas in the design space


VIDEO 2

How to observe optimization results locally


Case Study 2: Multi-Physics Steady State

Thermoelectric Simulation coupled with Solid Works Shape Optimization and Transient Radiative Heat Transfer for

Substrate Heat-up

Solid WorksDesign Modeler (imports geometry parameters from Solid Works, modifies model adding symmetry

Workbench+eArtius

substrateleft

right


Complex Multi Physics Problem

Optimization Log output

Optimization Messages updates (# of data points computed)

Optimization Method selection (MGE)

WorkBench status Bar (stop button)

Steady State Thermo- Electric with Surface to Surface Radiation

Transient Radiation with Surface to Surface

Design Modeler Parametric Geometry interface with Solid Works

Project folders


Optimization Parameters

Heater Geometry dimension 2, from Solid Works

Heater Geometry dimension 1, from Solid Works

P23=P24 (heating on left side=heating on right)

Electrical current runs through 3 separate heating elements creating temperature distribution. Electrical power in each heater equals I*V and to minimize P18 we need to find optimal ratio of power between center and left/Right elements.

Substrate Temperature after short term transie exposure to heater

F1= Tmax-350 F2=Tmin-350350 =>desired process temperature we want to reach



Heater Geometry dimension 1

Heater Geometry dimension 2

dT,Deg. C

dT,Deg. C

dT,Deg. C

dT,Deg. C

(Tmax-350), Deg. C

(Tmin-350), Deg. C

Want to pick bestValues for geometry dimensions 1 and 2


Most Essential Result

These are demo results of overnight –run, so study is not complete. However, we instantly see relationship between key conflicting variables (P18- maximum temperature difference in substrate) vs F1 – deviation from desired maximum temperature. The larger F1 the lower maximum temperature during heat-up, that means lower thermal ramp (gradient), lower power and thus lower temperature difference P18.

It is easy to have low temperature difference if you heat less, it means you loose less heat as well and thermal uniformity is better. In this problem we need to heat more, thus we are interested in Pareto frontier distribution looking for multiple trade- offs.

dT,Deg. C

(Tmax-350), Deg. C

3135Optimal Range of

interest found


Summary Result

Global computational optimizationof heating module and element designs to minimize temperature difference on substrate surface (DeltaT).

Optimization uses state-of-the art hybrid genetic-multi-gradient optimization methodology.

Optimal power ratio ~2Increase in power reducesuniformity

Plotting by EXCEL using CSV export from eArtius

Dimension 1Dimension 2


Conclusions to the ANSYS Add-in Section

New design improvement technology is now available in ANSYS

It is simple to use: - Removes all integration issues; behaves like part of Workbench- Reduces learning curve to minutes

eArtius optimization technology finds better designs faster because it is based on a multi-gradient analysis

eArtius optimization algorithms:- COMBI—simple to use, one parameter—builds optimization

strategy based on available time resource- MGE/MGP/HMGE/HMGE—for advanced users

Evaluation license is available for all webinar participants—for 2 months, no restrictions


Thank You!

Vladimir [email protected] phone: 949-375-7647


Installation package for the Windows version of eArtius-ANSYS Optimization Add-in can be downloaded from http://www.eartius.com/download.html

mailto:[email protected]

http://www.eartius.com/download.html


Questions regarding ANSYS Add-in?

Next section is about eArtius design optimization technology


Part II eArtius Optimization Technology


The biggest issues of current design optimization algorithms:

Low computational efficiency

Low scalability

Reasons:

Absence of efficient algorithms for estimating gradients

Curse of Dimensionality Phenomenon

Searching for optimal solutions in the entire design space while the search space can be reduced

Approximating the entire Pareto frontier while the user only needs a small part of it

Consequences:

Artificially reduced task dimensions by arbitrarily excluding design variables

Overhead in use of global response surfaces and sensitivity analysis

Have to rely only on use of brute-force methods such as algorithms’ parallelization

Fundamental Design Optimization Issues Study Motivation


- Sample points necessary to build an adequate global surrogate model - Pareto optimal points to maintain equal distance between neighboring optimal points in

the design space

How eArtius Addresses the Issues:For Response Surface Methods: - eArtius DDRSM spends just 0-7 points for local approximations—no global Response

SurfacesFor Approximation of the Entire Pareto Frontier:- eArtius performs directed search on Pareto Frontier—no global approximation of the entire

Pareto frontier

Curse of Dimensionality Phenomenon and Design Optimization

Example of uniformly distributed points:

Unit interval—0.01 distance between points—100 points

10-dimensional unit hypercube, a lattice with 0.01 between neighboring points—1020 sample points (Richard Bellman)

Adding extra dimensions to the design space requires an exponential increase in the number of:


DDRSM Benefits:

Equally efficient and accurate for any task dimension

Requires just 0-7 model evaluations regardless of task dimension

Fast— it builds a local approximation in 10-30 milliseconds

Automatic and hidden from users

Eliminates necessity in global response surface methods

Eliminates necessity in a sensitivity analysis

DDRSM evaluates gradients necessary for any gradient based optimization algorithms.

How DDRSM operates:Start iteration:

Determines the most significant design variables

for each responsevariable separately

Start iteration:Determines the most

significant design variablesfor each responsevariable separately

Builds local approximations for each response based

only on the most significant design variables

Builds local approximations for each response based

only on the most significant design variables

Analytically estimates gradients based on local

approximations

Analytically estimates gradients based on local

approximations

Performs a gradient based step

Performs a gradient based step

Dynamically Dimensioned Response Surface Method (DDRSM)


2225 green points visualize Pareto frontier for the above task

Pareto frontier is located on the flat x3=1in the design space

Search in the Entire Design Space

15.0 3 x10 2 x

65.00 1 x

)2/sin()2/cos()1(3 1133 xxxfMinimize

)2/sin()2/cos()1(3 2132 xxxfMinimize

)2/cos()2/cos()1(3 2131 xxxfMinimize

Why do we need to search in the entire design space? The search on the plane x3=1 would be more efficient


On the first half-step MGP improvespreferable objective (F2 )—green arrows

On the second half-step MGP improvesALL objectives—blue arrows—to maintaina short distance to Pareto frontier

Then MGP starts the next step from the newly found Pareto optimal point

Multi-Gradient Pathfinder (MGP) Method

F1

F2


MGP started optimization three times from the same start point {x1=1; x2=1; x3=1},but with different preferable objectives. Green trajectory:Min f1Min f2Min+ f3Red trajectory: Min+ f1; Min f2Min f3Blue trajectory:Min+ f1 Min f2Min+ f3

Light-green small markers visualize entire Pareto frontier, which is located on the plane x3=1 in the design space

Directed Optimization on Pareto Frontier


MGP—18 global Pareto optimal points out of 38 model evaluations Pointer—5 optimal points out of 1500 evaluationsNSGA-II & AMGA—FAILED to find a single Pareto optimal point after

1500 evaluations!

Searching the Entire Design Space is Not Productive!

30,..1,10

191

1

2

2

12

11

nnix

xn

g

gFgFMinimize

xFMinimize

i

n

ii

ZDT2 Benchmark Problem: multiple Pareto frontiers


MGP spent 185 evaluations, and found exact solutionsPointer, NSGA-II, AMGA spent 2000 evaluations each, and failed

Searching the Entire Design Space is Not Productive!

hgFMinimizexFMinimize

XFgFgFh

nxxxxxxng nn

2

11

111

3222

322

]1;0[][);10sin(//1

10)],4cos(...)4cos()4[cos(10)...()1(101


Hybrid Multi-Gradient Explorer (HMGE) Optimization Algorithm

Synergy of the features brings HMGE on unparalleled level of efficiency and scalability

HMGE is believed to be the first global multi-objective optimization algorithm which provides:

- Efficiency in finding the global Pareto frontier

- High convergence typical for gradient- based methods

- Scalability: Equal efficiency optimizing models with dozens, hundreds, and even thousands of design variables

Genetic Algorithm Framework

Random Mutation Gradient Mutation

DDRSM – Super Fast Gradient Estimation


COMBI Optimization Algorithm

MGE

DDRSM – Super Fast Gradient Estimation

HMGE

MGP

Output all Optimal Designs Found over 8 hours

ANSYS ModelTime Resource = 8 hours

COMBI – takes just one parameter – time resource available for optimization, and dramatically simplifies using the optimization technology

COMBI is a smart wrapper for eArtius optimization algorithms MGE, MGP, and HMGE

COMBI decides which algorithm to use based on a model analysis and available time resource

COMBI is designed for the users that need benefits of optimization, but do not have time to learn optimization technology


Thank You!

Vladimir [email protected] phone: 949-375-7647


Installation package for the Windows version of eArtius-ANSYS Optimization Add-in can be downloaded from http://www.eartius.com/download.html

mailto:[email protected]

http://www.eartius.com/download.html

eArtius - ANSYS Optimization Add-in webinar

Design

eartius ansys

detailcopyright eartius

thencopyright eartius

incopyright eartius

cancellationscopyright

local onecopyright eartius

optimization tasks

optimization standardseartius