dynamic software & engineering

CAE-Software

Model Calibration Sensitivity Analysis Optimization Robustness Evaluation Robust Design Optimization Metamodeling optiSLang Product Overview

PRODUCT OVERVIEW

Since its market launch in 2002, optiSLang has developed into one of the leading universal software platforms for CAE-based optimization in virtual prototyping. Based on design variations or measurement and observation points, effi cient variation analyses can be performed with minimal user input and few Typical workfl ow using sensitivity analysis, optimization, meta-models and validation of the best design solver calls.

optiSLang supports the engineer with: Metamodeling with optiSLang • Calibration of virtual models to physical tests signs to the boundaries of tolerable stresses, deformations optiSLang’s expands parametric design studies to impact of input variability to the responses. For the identifi ca- • Analysis of parameter sensitivity and importance or other critical responses. As a result, the product behavior metamodeling based on a set of virtual design points, experi- tion process, optiSLang’s Coeffi cient of Prognosis (CoP) quanti- • Metamodeling may become sensitive to scatter with regard to material, mental design responses or fi eld observations. The software fi es the prognosis accuracy of the meta-model regarding the • Optimization of product performance geometric or environmental conditions. Subsequently, a automatically identifi es the corresponding input parameters response variation of a data set. The result is the Metamodel • Quantifi cation of product robustness and reliability also robustness evaluation has to be implemented as a Robust for every response variation. Furthermore, the best possible of Optimal Prognosis (MOP). This procedure fulfi lls the most referred as to Uncertainty Quantifi cation (UQ) Design Optimization (RDO) strategy consisting of: functional meta-model is identifi ed to adequately describe the important tasks in metamodeling: • Robust Design Optimization (RDO) also referred as to Design for Six Sigma (DFSS) • Sensitivity analyses to identify the most affecting opti- • Avoidance of overfi tting mization parameters regarding the optimization task • Identifi cation of the most corresponding meta-model In the past, one of the largest bottle necks for a success- • Multi-disciplinary and multi-objective optimizations to • Reduction of complexity in large dimensions of variability ful simulation driven product development was to create a determine the optimal design workfl ow including all necessary components for paramet- • Robustness evaluations to verify design robustness and By automatic reduction to important parameter spaces, ric design and system evaluation in one standardized pro- reliability the procedure ensures the creation of meta-models with a cess. Therefore, since version 4, optiSLang has been expand- minimum of virtual design points. Consequently, even tasks ing to a powerful simulation workfl ow environment. This optiSLang’s Best-Practice-Management supports this strat- involving a large number of optimization variables, scatter- makes the software the perfect tool for simulation driven egy by an automatic selection of the appropriate sensitivity, ing parameter as well as non-linear system behavior can be workfl ow generation using parametric models for sensitiv- optimization and robustness algorithms and their settings. effi ciently solved. ity analysis, optimization and robustness evaluation. The procedures are guided by intuitive drag & drop-work- fl ows and powerful postprocessing tools. Within optiS- Lang’s workfl ow building environment, any parametric CAD Robustness Evaluation and Reliability Analysis Simulation based product optimization in or CAE model can be easily integrated. Thus, the engineer optiSLang provides powerful sets of stochastic analysis al- virtual prototyping extensively benefi ts from the capabilities of parametric gorithms. It enables the user to conduct a reliable determi- The predominant aim of CAE-based optimization in virtual modeling and design studies in order to innovate and ac- nation of failure probabilities by evaluating the result value prototyping is often to achieve an optimal product perfor- celerate virtual product development. variation, which includes the identifi cation and consider- mance with a minimal usage of resources. This pushes de- 3D visualization of the Metamodel of Optimal Prognosis ation of relevant scatter input parameters.

1 www.dynardo.de 2 Variation Analysis

OPTISLANG – GENERAL PURPOSE TOOL FOR VARIATION ANALYSIS

Sensitivity analysis, optimization and robustness evaluation based on multiple designs, measurement or observation points with a minimum of user input and solver runs for an effective virtual product development.

A U T L T S O M A F A U T E • Identifi cation of the relevant input parameters and D E D L M response values based on sensitivity analysis A O • Stochastic sampling (LHS) for optimized scanning M D • Pre-optimization of the parameter sets with MOP T I U of multi-dimensional parameter spaces L P A without additional solver runs O R • Quantifi cation of prognosis quality (CoP) of meta-models • Further optimization of the parameter sets with the most

• Generation of the Metamodel of Optimal Prognosis (MOP) appropriate algorithms (Best-Practice-Management)

A R I SENSITIVITY V A T I E R O N OPTIMIZATION T E ANALYSIS M A R A P

L CAE-Data A

M I Optimal Design

N I

M

L

E

S

S

S

O

L

Measurement V

Data E

R

-

MODEL CALIBRATION R

U

N • Effi cient methods of stochastic analysis for the

• Find the best fi t for simulation and measurement S determination of failure probabilities

• Evaluation of result value variation

• Identifi cation of the relevant scatter input parameter

Coeffi cient of Prognosis (CoP)

The CoP quantifi es the forecast quality of a meta-model (re-

gression model) for the prognosis of a result value.

ROBUSTNESS EVALUATION

Metamodel of Optimal Prognosis (MOP) variables are evaluated with different meta-models. Thus,

The MOP represents the meta-model with the best prog- a No Run Too Much-strategy is implemented with a maxi- nosis quality of the result value variation. For the determi- mum of prognosis quality regarding the given number of nation of the MOP, different subspaces of important input design evaluations.

3 www.dynardo.de 4 S Process Integration and Automation S E C O ITIVIT R S Y N A O P E N - S A P E L T Y I A S M C I I S Z A T IO N

DoE

Virtual Product

manual HiL

Calibration

Simulation = Hardware

ES S Connect PLM Connect PLM

TN E S VA Digital Twin U L B U autom. complex CAx workfl ow O A R T Laboratory I Development O parametric studies, RDO, …

N

CAx 3rd party Product CAE-Data Production

PROCESS INTEGRATION AND AUTOMATION

Interactive process automation and integration as well as access to best possible parametric simulation Statistical Data-based ROM models are the key for successful CAE - based parametric studies. In optiSLang, this procedure is guided Analysis and supported by wizards and default settings.

Process integration optiSLang supports the interfacing to any software tool used are displayed at the same time. This enables full access and custom applications is secured and optiSLang projects can be • Meta-models in virtual product development which can be externally traceability of the complete workfl ow. The user can connect any integrated into customized platforms. Repetitive and perva- • Tool integrations called. Interfacing is either offered by text fi le based or pre- complex simulation process of CAE solvers, pre- and postproces- sive tasks can be standardized and automatized. • Database connections defi ned tool integrations. Nowadays, more than 100 differ- sors in heterogeneous networks or clusters. They are automa- ent CAx/PLM software solutions are coupled with optiSLang. tized either in a single solver process chain or in very complex Current requirements for fl exibility and upcoming requests The new generation of optiSLang gives access to: multidisciplinary / multidomain fl ows. Even performance maps Extensibility for extensibility are satisfi ed by those interfaces. Therefore, and their appraisal can be part of standardized projects. The openness of optiSLang also enables users to plug: optiSLang is the platform to address future needs of para- • CAD (Catia, Nx, Creo, Solidworks …) • Algorithms for DOE, Optimization, Robustness etc. metric and simulation driven virtual product development. • CAE (ANSYS, Abaqus, AMESim …) • MS Excel, Matlab, Python … Integration of optiSLang into parametric model- • PLM (EKM, Teamcenter, Subversion ...) ing environments • In-house solver The modular structure of optiSLang supports the direct in- tegration of its modules into standard parametric modeling Different parametric environments can be collected and environments. This framework allows the seamless integra- combined to one automatized parametric workfl ow for tion of optiSLang into, e.g., ANSYS Workbench/AIM, Excel or simulation driven product development. SimulationX. Here, users do not have to leave their paramet- ric modeling environment and can access optiSLang mod- ules through a minimum of user input. Defi nition of CAx Workfl ows The graphical user interface supports the workfl ow approach Interfaces and automation visually by single building blocks and algorithms which are optiSLang provides Python, C++ and command line interfac- graphically coupled in order to show dependencies and schedul- es to allow the automatic creation, modifi cation, execution ing. The relationships can be determined and controlled in one and remote control of projects within optiSLang or from ex- context. Easily understandable charts as well as control panels ternal tool integrations. As a consequence, the usage within Fully automatized optimization workfl ow in optiSLang considering structural costs and metric of performance map, running several solvers and using HPC

5 www.dynardo.de 6 ANSYS optiSLang

ANSYS Workbench plugin Workbench node for optiSLang integration for ANSYS integration in ANSYS in optiSLang optiSLang ® CAx-Data Workflow Combining powerful parametric model Management Sensitivity capabilities with Robust Design Optimization Robustness Model Calibration Sensitivity Analysis Optimization Optimizer Robustness Evaluation Reliability Measurement Data

ANSYS OPTISLANG CAx 3rd party Simulation Process and Data Management (SPDM) and Data Process Simulation Management (SPDM) and Data Process Simulation

ANSYS optiSLang combines leading simulation technology of ANSYS with optiSLang’s powerful capabili- Excel® Add-in Other Solver Post Processing ties in variation analysis and process automation. By building CAx-workfl ows for automatic design varia- tion studies (sensitivity analysis, optimization and robustness evaluation), virtual product development becomes more effi cient. Thus, better products can be developed in shorter time. ANSYS optiSLang connecting simulation and optiSLang’s effi cient algorithms and tools for workfl ows of variation analysis

Modes of interoperability with ANSYS tionality. The user only needs to set up the variability space, The process integration and workfl ow building capabilities the constraints and the objectives. optiSLang automatically via ANSYS via optiSLang of optiSLang offer a distinguished support of ANSYS simu- identifi es the most important parameter and generates the 1 Workbench plugin process integration node lation tools. There exist 3 major ways to integrate ANSYS MOP. A Best-Practice-Management selects the appropriate 2 tools into optiSLang workfl ows and/or to use ANSYS’ pow- methods for optimization. erful parametric modeling with optiSLang’s algorithms: The options for parallel computing at several cores with ANSYS Remote Solve Manager and the use of ANSYS HPC • In ANSYS Workbench: the optiSLang Workbench plugin Pack Parametric Licenses for simultaneous computing of makes optiSLang technology available inside the power- different designs are supported. The plugin also includes ful simulation platform robust handling of design failures and any source of noise, • Text-based integration (ANSYS APDL) for combining AN- like solver accuracy. All successful designs are stored in op- SYS with other CAE / CAD tools for process automation tiSLang’s database and can be used and postprocessed in- and workfl ow generation dependently from the ANSYS Workbench design table. via optiSLang • ANSYS Workbench node: The direct integration node en- 3 ANSYS Workbench node ables the use of ANSYS Workbench projects in optiSLang General features of the Workbench Plugin are, e.g.: workfl ows • Reduction of the number of CAE solver runs by optiSLang’s minimalist philosophy • Automatic identifi cation of important parameters dur- ANSYS Workbench plugin ing sensitivity analysis optiSLang’s ANSYS Workbench Plugin toolbox includes all • Automatic buildup of best possible regression functions modules for Robust Design Optimization namely sensi- (MOP) tivity analysis, optimization and robustness evaluation. A • Multidisciplinary and multiobjective optimization wizard guides the user through the defi nition of the differ- • Robustness evaluation Three ways of integrating ANSYS simulation and optiSLang using the example of an electric machine ent modules and gives an optimal solution strategy while • Insightful and effi cient result post-processing module minimal user input is required. Therefore, the optiSLang • Support of non-scalar responses (vectors, curves, signals, integration can be easily started with drag and drop func- matrices)

7 www.dynardo.de 8 Sensitivity Analysis

Metamodel of 2 Optimal Prognosis

Latin Hypercube Parameter 1 Sampling 3 importance

SENSITIVITY ANALYSIS By means of a global sensitivity analysis and the automatic generation of the Metamodel of Optimal Understand the most important input variables Prognosis (MOP), optimization potential and the corresponding important variables are identifi ed. This previous knowledge enables the formulation of task-related objective functions and constraints as well as the selection of suitable optimization algorithms.

Practical Application • Identifi cation of the most important input variables Design variables are defi ned by their lower and upper related to each response value, constraint and objective bounds or by several possible discrete values. In industrial • Minimization of solver runs by MOP workfl ow optimization tasks, the number of design variables can of- ten be very large. With the help of a sensitivity analysis, engineers can accurately identify those variables which ef- Methods fectively contribute to a possible improvement of the opti- • Defi nition of optimization variables with upper and mization goal. Based on this identifi cation, the number of lower bounds or discrete values design variables will be decisively reduced and an effi cient • Defi nition and creation of the Design of Experiments optimization can be conducted. Additionally, a sensitivity (full factorial, central composite, D-optimal, customized analysis helps to formulate the optimization task appropri- DoE); Latin Hypercube Sampling for optimal scanning of ately concerning the choice and number of objectives, their multi-dimensional parameter spaces weighting or possible constraints. Furthermore, it is used to • Automated generation of the MOP by testing a library of estimate the numerical noise of the CAE solver as well as the approximation methods proper physical formulation of the design problem. • Quantifi cation of the prognosis quality by the model independent CoP

Best Practice • Coverage of the entire design space and minimization of Postprocessing & Visualization correlation errors among input variables by optimized • Histograms Latin Hypercube Sampling (LHS) • Correlation matrix / parallel coordinate plots • Automated identifi cation of the meta-model with the • 2/3D Anthill plots and 2/3D surface plots of the MOP best prognosis quality of the response value in the sub • CoP matrix with all responses space of important variables • Prognosis of residual plots and local error estimates CoP matrix of a work holding device analyzing mass and maximum displacements: The mass is mainly infl uenced by the upper plate thickness and the displacements by the cross section parameters of the lower beams. • Quantifi cation of the forecast quality of each meta-model • Integration of images and processes for the prognosis of response values by the CoP • Customized plots

9 www.dynardo.de 10 Optimization

Optimization 2 using MOP

Sensitivity analysis Direct optimization 1 for pre-optimization 3 with algorithms OOPOPTPTT

MULTIDISCIPLINARY OPTIMIZATION Pre-optimize on meta model and leading edge optiSLang provides powerful optimization algorithms and automated workflows for an efficient determination of optimal design parameters regarding various multidisciplinary, nonlinear and optimization algorithms multicriteria optimization tasks.

Practical Application Methods Structures and sub-systems often need to be designed to • Gradient-based methods (NLPQL) withstand multidisciplinary load cases. For example, vehicle • Nature-inspired Optimization Algorithms (NOA) for body structures are exposed to crash (nonlinear transient), single and multiobjective optimization Noise Vibration Harshness (frequency domain), stiffness • Adaptive Response Surface Method (ARSM) in case of (linear static), durability (linear static) and aerodynamics less than 20 important optimization variables (CFD). The structural requirements to meet loads in one dis- • Customized optimization algorithms cipline are very often different to requirements for loads in the other. Unless loads from all disciplines are considered simultaneously during the optimization process, the result- Postprocessing & Visualization ing design will not be well balanced for structural perfor- • Interactive postprocessing adapted to the optimization mance. Multidisciplinary optimization considering single algorithm and multiple goals is essential to achieve this objective. • Fast investigation of optimization performance using different visualization options • Parallel coordinate plots and cluster analysis for selec- Best Practice tion of best design candidates • Identifi cation of the most relevant input parameters and re- • Selection of individual designs and easy visualization of sponse values with the help of sensitivity analysis and MOP nonscalar results as time-series, 3D-fi eld data, auto- • Pre-optimization of parameter sets and studying possible mated animation objective confl icts using the MOP approximation • Objective history plot for single objectives, 2D and 3D • Optimization wizard for automatic selection of suitable Pareto plots for multiple objectives algorithms for optimization • Easy defi nition of parameter ranges, single and multiple objectives and constraints

Optimization of a tuning fork regarding mass and frequency

11 www.dynardo.de 12 Model Calibration

Identifi cation of 2 parameters

1 Signal processing 3 Match of curves

MODEL CALIBRATION Match experimental data with simulation to Model calibration identifi es unknown parameters of CAE models to obtain the best possible agreement with available test results. By combining sensitivity analysis and optimization, model parameters which increase simulation quality are not directly measurable can be consistently identifi ed.

Practical Application Methods Measurement data represents characteristic system respons- • Consideration of scalar response values es that are critical to validate and to improve the physical • Defi nition of multi-channel signals, e.g. time-displace- model of the system. In the context of parameter identifi ca- ment curves tion, model calibration means using experimental observa- • Extensive library of functions, e.g. local values as tions and simulation runs to estimate unknown simulation maximum and minimum amplitudes, global values as model parameters. By means of sensitivity analyses, fi rst, pa- integrals of certain properties and more complex signal rameters will be detected which actually have an infl uence calculations on the simulation results and the calibration procedure. Sec- • Defi nition of individual objective functions ond, the analysis helps to defi ne suitable measures to quan- • MOP based sensitivity analysis of different signal prop- tify the difference between measurement and simulation. Fi- erties and pre-evaluation nally, it can be analyzed whether the inverse problem will be • Several optimization algorithms (e.g. gradient-based or solved non-ambiguously, which means a unique parameter nature-inspired) combination exists that allows optimal matching between measurement and simulation. Postprocessing & Visualization • Illustration of statistical evaluations Best Practice • Visualization of signal functions and the corresponding • Sensitivity analysis to check unknown parameters for reference value for each design signifi cant infl uence on the model response • Sensitivities and approximation of signal function values/ • CoP supports the identifi cation of the best possible response parameter sensitivities extraction by comparing model and measured values • Interactive evaluation of curve fi tting and corresponding • CoP verifi es the uniqueness of the best possible correla- design images tion model between parameter and response variation • Parallel coordinates plot for uniqueness evaluation Parameter identifi cation of a spring steel model based on a tension bar experiment: The load displacement curves of experiment and corresponding simulation are • Identifi cation of non-unique (multiple) parameter sets compared in the signal processing and the minimum deviation is found in the least squares minimization. by coupling of parameters

13 www.dynardo.de 14 Automated generation ofoptimized Latin Hypercube • 15 Predefi • Defi • Best Practice Best relative variation variables. oftheinput the relative variations ofthecriticalmodelresponses tothe effi besides standard variation orSigmaLevels, thevariance co- on stochasticanalysis.Ourconsulting yieldsthat, expertise propriate robustness evaluation andmeasurement based manufacturing processes, isessential tointroduce it anap- avoidable inoperating uncertainties conditions aswellin boundary conditions orloads. Inorder tocope withtheun- ables onthesedesigns, geometry, e.g., material parameters, necessary toinvestigate vari- ofscattering input theimpact boundaries, regarding e.g., material strength.istherefore It Optimized designsare often pushedtotheirperformance Practical Application variation andcorrelation measures withaminimumofrequireddesignvariants. variability variationinput ontheresult by theCoeffi variables.basis ofscattering input OptimizedLatin HypercubeSamplingandthequantifi optiSLang quantifi ROBUSTNESS EVALUATION

mal input correlationmal input error Samples (LHS)toscantherobustness spacewith mini- variablesscattering input correlation possibledefi best matrix tosupport crucial input ofarobustness analysis crucial input isalsoasuitablerobustnesscient measure comparing nition ofallpossibly infl ned distribution function typesandaninput ned distributionfunction ROBUST

es therobustness ofdesignsby generating ofsuitabledesignvariations aset onthe NESS uencing uncertainties astheuencing uncertainties nition of cient ofPrognosis (CoP)cient ensuresthereliabilityof Quantifi • Identifi • Traffi • Distributionfi • Linearcorrelations matrix, nonlinearCoP basedcorrela- • Histograms toillustrate scatter ofresultvalues • Approximation ofviolation probability • Approximation ofSigmamargins • Fitting inthehistogram ofdistribution function ofresult • Optimized Latin Hypercube Sampling • variables Stochasticinput withdistributiontypesand • Postprocessing & Visualization Methods

mation ofsigmalevel andviolation probability values includingfi ence usingtheCoeffi every response andquantifi critical responses tion matrix, MOP andCoP plots values correlationinput c light plot to check the violation of limit values tocheckthe violation oflimit plot c light of cation of the most affectingcation ofthemost scatter for input cation ofrobustness bythehistogram ofresult tting, Sigmavalues,tting, violation probabilities tting of distribution function, approxi-tting ofdistributionfunction, cient ofPrognosiscient (CoP) cation of input variablecation ofinput infl cation ofevery u- www.dynardo.de 1 Identify the most important scattering variables important to Identify themost Sampling Latin Hypercube ensure designquality 2 variation parameterOutput more, theMOP gives more information onthesources ofthescatter. mean value andvariance thesafety margin canbeobserved. Further- ter incomparison togiven failure limits. With thehelpofestimated robustness analysis: The traffi Analysis ofthescatter oftheresponse values withinavariance-based 3 c light plot indicates theresponse plot c light scat- importance Parameter Robustness Evaluation 16 Reliability Analysis

Reliability 2 algorithms

Defi nition of Estimation of 1 failure limit 3 failure probability

RELIABILITY ANALYSIS Estimate the failure probability to ensure design quality For reliability analysis, optiSLang provides powerful numerical algorithms for the determination of small event violation probabilities. Thus, it allows the essential fi nal verifi cation of such failure scenarios subsequent to a robustness evaluation or Robust Design Optimizations (RDO).

Practical Application Methods If designs need to meet high safety or quality requirements • First Order Reliability Method (FORM) and Adaptive Im- with low event probabilities of less than 1 out of 1000, a portance Sampling for continuously differentiable limit reliability analysis is necessary to investigate how these de- state functions signs are affected by scattering input variables, e.g., geom- • Directional Sampling and Adaptive Sampling (AS) for a etry, material parameters, boundary conditions or loads. As moderate number of random variables, multiple failure an alternative to the estimation of safety distances by using mechanisms and small probabilities of failure standard deviations in robustness evaluations, a reliability • Adaptive Response Surface Method (ARSM) as the most analysis calculates the probability of exceeding a certain effi cient method for up to 15 important random variables limit by using stochastic algorithms. Thus, rare event viola- tions can be quantifi ed and proven to be less than the ac- cepted values. Postprocessing & Visualization • Histograms • 2D/3D anthill plots Best Practice • History plots • Robustness evaluation for the approximation of viola- • Violation probabilities tion probabilities and for the identifi cation of important random variables as the basis for an appropriate selec- tion of methods regarding a reliability analysis • Defi nition of one or various failure mechanisms using

limit state functions The reliability analysis as fi nal proof of the results of a variance-based robustness evaluation: The identifi cation and more accurate integration of the • Recommendation of verifying low probabilities of failure failure domain by directional sampling enables a qualifi ed reliability assessment that is independent of the distribution type of inputs and outputs. with two alternative algorithms of reliability analysis

17 www.dynardo.de 18 Robust Design Optimization

Robustness 1 Optimization 2 evaluation

ROBUST DESIGN OPTIMIZATION (RDO) Combine leading edge algorithms of optimization and RDO combines methods of design optimization with robustness evaluation. It allows a product improvement with a corresponding quantifi cation and assurance of quality. optiSLang provides techniques for variance- robustness for effi cient RDO based and reliability-based RDO in compliance to the Taguchi method or Design For Six Sigma (DFSS).

Practical Application Quality is one of the most important product properties. If safety distances strongly vary in the design space, simul- Providing it in an optimal dose means to reduce costs for taneous RDO might become the method of choice. Here the rework, scrap, recall or even legal actions while satisfying workfl ow building capabilities allow nesting a robustness costumers demand for reliability. During the product de- or reliability analysis into an optimization algorithm. velopment, the common approach to achieve this goal is to apply RDO. The method uses results of stochastic analysis as constraints or objectives to accomplish the optimization. Methods Iterative and simultaneous approach for: Best Practice • Variance-based RDO - tasks with low sigma level From our consulting expertise, it is recommended to start (≤ 2-3sigma) with an iterative RDO. In this procedure, the optimization • Reliability-based RDO - tasks with high sigma level has to consider safety factors to assure a certain safety (≥ 3sigma) margin of critical responses. In iteration with optimiza- tion and succeeding robustness or reliability analysis these safety factors are adapted until the fi nal proof of reliability: Postprocessing & Visualization • Interactive postprocessing adapted to the optimization • Defi nition of the design space of optimization variables as algorithm well as of the robustness space of all scattering variables • Fast investigation of optimization performance using • Initial sensitivity analysis within the design space as well different visualization options as initial robustness evaluation within the space of scat- • Histograms to illustrate scatter of result values tering variables in order to identify important parame- • Distribution fi tting, Sigma values, probability of failure 20% 1:1.000.000 ters, optimization potential, initial violation probabilities • Traffi c light plot to check the violation of limit values of and safety margins critical responses • Recommendation of fi nal reliability proof for tasks with a sigma level higher than three Simultaneous Robust Design Optimization of a steel hook: The failure rate was minimized from 22% to less than 1:1.000.000

19 www.dynardo.de 20 Data-Based ROM

DATA-BASED Input Field Response Field Variation Variation FMOP

„Sensor“ CAE scans provide simulation information provides USING MOP & FMOP on input Statistics on Structures information variation. on response Complex CAE models are replaced by variation highly accurate and resource-effi cient FMOPs for real-time approximation of signals, FEM solutions or geometric deviations EXTENDED METAMODELING – FROM SCALAR VALUES TO FIELDS IN TIME AND SPACE optiSLang’s MOP reveals how scalar input variation affects scalar output variation. To analyze how fi eld inputs affect fi eld responses, Dynardo has developed Statistics on Structures (SoS). Worklow for the generation of a Field Metamodel of Optimal Prognosis (FMOP) connecting input and response fi eld variation to be integrated in digital twins

Using metamodeling for robust design optimization, in sys- posed. This helps to explain their correlation to any other based on a few real-life measurements, a statistical model However, these reductions are often restricted to linear tem simulation or on customer’s hardware to operate and scalar or fi eld variable. Successful applications include a re- can be automatically created, which is capable of generat- systems. The alternative for non-linear systems are data- maintain products in an optimal way, customer ask for ap- sponse data decomposition distributed in time or frequen- ing hundreds of new virtual random samples representing based ROMs. They use functional models to approximate proximating responses in time or space. Using multiple sca- cy (1D signals, e.g. load-displacement curves), response on the statistics of the measured data. This is particularly help- response surfaces considering the effect of input variations lar metamodels for discrete time and space support points, 2D grids (e.g. surface stress) or in 3D space (e.g. 3D temper- ful when dealing with geometric imperfections from laser on the response variation based on the given sample set. however, is often not successful, because existing correla- ature distribution). Similar to modal shapes, the variation scans. With SoS, these laser scans can be statistically anal- For fi eld data as input or response, SoS provides the Field tions between individual points in time and space are miss- patterns are sorted by their importance. ysed directly on the FEM mesh. The parameterization how Metamodel of Optimal Prognosis (FMOP) which can be ing. To extend correlation analysis to fi eld variables Dynar- In addition to the application for correlation analyses, geometry deviates is also given. By generating and simulat- used to approximate signals, FEM solutions or geometric do developed the software Statistics on Structure (SoS). SoS the automatically identifi ed variation patterns are param- ing a set of possible geometries, the user can quantify the deviations. provides models to automatically identify relations in time eterizations of nearly arbitrary input data and can be used effect of the variations on the structural performance. or space. to generate scattering design realizations. For example, SoS is specifi cally designed for the automatic identifi ca- tion and analysis of data relations between individual points Data-based Reduced Order Model (ROM) in time and space. Thus, the coupling of optiSLang and SoS ROMs are very important in system simulation and are will extend correlation analysis and metamodeling from sca- expected to become a key technology for digital twins. In lar values to input and output variables in time and space. typical applications a detailed product simulation needs to be linked to sensor data in order to predict product param- eters (e.g. life of turbine blades) accurately enough to be SoS decomposition and analysis of variation capable of optimizing the maintenance and operation. To patterns fulfi ll the reaction time requirements from digital twin, the SoS analyzes variations from a given DOE or from mea- detailed simulation models need to be reduced. The classi- surements and it automatically identifi es the dominant cal approach of ROMs uses a matrix condensation which is variation patterns including their “scatter shapes” and am- called “physics-based” ROMs, because the formula still con- plitudes. Thus, the variations in time or space are decom- tains the physics of how input variation affects response. Standard deviation of response thickness variation after forming simulation (left) Picture of scatter shape parametric model of measured brake pad surfaces and the fi rst scatter mode already representing 84% of the total variation (right)

21 www.dynardo.de 22 Fields of Application

FIELDS OF APPLICATION & CUSTOMER STORIES Workfl ow of sensitivity analysis, MOP building and working point optimization for an electronic motor As a general purpose tool for variation analysis based on multiple designs, measurement or observation points, optiSLang serves an extensive range of application fi elds across various industries.

Robust Design Optimization Product Lifecycle Management (PLM) In the uncomplicated and fl exible cooperation with Dynardo, Since 2010, Shell has been cooperating with Dynardo to de- Initially introduced in the automotive industry in 2002, Due to a powerful graphical environment for CAE process it is a great advantage that the company is not only a soft- velop and to implement successfully a simulation based work- optiSLang has been extensively used for design optimiza- integration and automation as well as for workfl ow build- ware developer but also an engineering service provider. Di- fl ow for the optimization of oil and gas production at uncon- tion, robustness evaluation based on CAE models, reliabil- ing, optiSLang is widely used for the integration and stan- rect communication with the programers and individual li- ventional reservoirs. Key factor for success is the excellent and ity analysis as well as parametric CAD-modeling or system dardization of RDO workfl ows regarding PLM-Systems. cense agreements ensure a rapid adaptation and extension reliable consulting service of Dynardo as well as the company’s simulation in various industries. of the software optiSLang to specifi c technical requirements powerful software tools. The FEM based hydraulic stimulation of Robert Bosch GmbH. simulator as well as the software optiSLang are sophisticated Metamodeling Roland Schirrmacher | Robert Bosch GmbH workfl ow environments to calibrate simulation models and to Parameter and System Identifi cation The metamodeling functionality of optiSLang enables the Corporate Sector Research and Advance Engineering generate metamodels for connecting hydraulic fracturing sim- optiSLang provides key functionality in the automatic iden- user not only to identify the optimal design confi guration, Future Mechanical and Fluid Components (CR/ARF1) ulations with related production costs and risk management. tifi cation of sensitive parameters and the quantifi cation of but also implements the optimal design within a meta- The workfl ow has been effectively applied worldwide at major the forecast quality of response variable variation. model. This includes the pre-investigation of the expected unconventional oil and gas assets of Shell. variation window of the product to understand how the Taixu Bai | Shell Exploration & Production Company USA design will operate under this condition. Thus, customized In the framework of the virtual product development pro- Completion & Technology Effectiveness Team Data Mining optimizations of product performance, confi guration or cess of the Daimler AG, optiSLang variation analysis using optiSLang’s leading technology of processing large numbers maintenance can be conducted. parametric simulation models are employed for the evalua- of parameter and generating the best possible metamod- tion and optimization of different functional requirements els has become key technology for data mining or machine like driving comfort and crashworthiness behavior. In order learning. The methods can be applied to sets of virtual de- For detailed information please visit the library section of to ensure design robustness within the virtual prototyping, sign points as well as available experimental data or obser- our website www.dynardo.de. There you will fi nd customer in 2002, Daimler started implementing optiSLang for NVH vation points. That opens up new fi elds of application like stories chronologically ordered by methodology and indus- analysis of driving comfort. Since then, applications have generating metamodels based on fi eld data for optimiza- trie application. been extended to many functions such as crashworthiness, tions in the oil and gas production. brake squeal, forming or joining simulation.

23 www.dynardo.de 24 WELCOME TO

CONSULTING & TRAINING

Dynardo provides simulation services and customized solutions for your FE analyses and CAE optimization in virtual product development. Due to the combination of being a CAE consulting company and software developer, Dynardo is your competent and fl exible partner for complex tasks in the CAE fi eld.

Technology Implementation Projects RDO Consulting Service Especially for the introduction of CAE-based Robust Design If customers have not yet implemented a CAE-based de- Optimization in product development processes, a pilot velopment process but would like to benefi t from the po- project based on the customer‘s product knowledge and tentials for their product lines, we offer the generation and Dynardo’s consulting experience would be an adequate verifi cation of virtual product models as well as the con- initial cooperation. Dynardo offers expertise in various in- duction of a CAE-based optimization as a consulting service. dustrial fi elds helping you to conduct a realistic safety and The metamodeling methodology will show you possible op- reliability analysis. We will further support you in proper timized product confi gurations and will explain how input assessments of material behavior, the prediction of failure variation affects the design responses. evolution, design optimization or simulation of FEM based ANNUAL WEIMAR limit load analysis. Getting Started and Advanced Training OPTIMIZATION AND STOCHASTIC DAYS For a competent and customized introduction to our software Customization products, visit our basic or expert training clearly explaining Your conference for CAE-based parametric optimization, stochastic analysis and Robust Design Optimization You want to economize your virtual product development? theory and application of a sensitivity analysis, multidisci- in virtual product development. Dynardo provides you with customized solutions based on plinary optimization and robustness evaluation. The training our software optiSLang and Statistics on Structures (SoS). addresses all engineers and decision makers involved in the We integrate your in-house software into optiSLang or im- development process and product life cycle. plement optiSLang as part of your company SPDM (Simula- Furthermore, our internet library is the perfect source for tion Process & Data Management) solution. Even fully au- your research on CAE-topics and applications of CAE-based The annual conference aims at promoting successful appli- Take the opportunity to obtain and exchange knowledge tomatized workfl ows to optimize your products regarding RDO. There you will fi nd practical references and state-of- cations of parametric optimization and CAE-based stochas- with recognized experts from science and industry. specifi c customer requirements can be generated. We help the-art case studies matched to the different fi elds of meth- tic analysis in virtual product design. The conference offers you to establish a company-wide standard workfl ow and ods and industrial applications. focused information and training in practical seminars and You will fi nd more information and current dates at: make your products benefi t from consistent and effi cient interdisciplinary lectures. Users can talk about their experi- www.dynardo.de CAE processes. For detailed information please visit our website: ences in parametric optimization, service providers present www.dynardo.de. their new developments and scientifi c research institutions We are looking forward to welcoming you to the next Weimar inform about state-of-the-art RDO methodology. Optimization and Stochastic Days.

25 Contact & Distributors

Worldwide Austria India CADFEM (Austria) GmbH CADFEM Engineering Services India Dynardo GmbH Vienna Hyderabad Steubenstraße 25 www.cadfem.at www.cadfem.in 99423 Weimar Switzerland Phone: +49 (0)3643 9008-30 CADFEM (Suisse) AG USA Fax.: +49 (0)3643 9008-39 Aadorf www.dynardo.de www.cadfem.ch CADFEM Americas, Inc. [email protected] Farmington Hills, MI Czech Republic, Slovakia, Hungary www.cadfem-americas.com Dynardo Austria GmbH SVS FEM s.r.o. Offi ce Vienna Brno-Židenice Ozen Engineering Inc. Wagenseilgasse 14 www.svsfem.cz Sunnyvale, CA 1120 Vienna www.ozeninc.com www.dynardo.at Sweden, Denmark, Finland, Norway [email protected] EDR & Medeso AB USA/Canada Västerås SimuTech Group Inc. Dynardo US, Inc. www.edrmedeso.com Rochester, NY 315 Montgomery Street www.simutechgroup.com 9th and 10th Floor United Kingdom of Great Britain and San Francisco, CA 94104 Northern Ireland Japan USA CADFEM UK CAE Ltd TECOSIM Japan Limited www.dynardo.com Croydon, Surrey Saitama [email protected] www.cadfemukandireland.com www.tecosim.co.jp

Ireland Korea CADFEM Ireland Ltd TaeSung S&E Inc. Worldwide distribution of Dublin Seoul ANSYS optiSLang www.cadfemukandireland.com www.tsne.co.kr ANSYS, Inc. Canonsburg Turkey China www.ansys.com FIGES A.S. PERA-CADFEM Consulting Inc. Istanbul Beijing Worldwide distribution of www.fi ges.com.tr www.peraglobal.com optiSLang LightTrans GmbH North Africa Jena CADFEM Afrique du Nord s.a.r.l. www.lighttrans.de Sousse www.cadfem-an.com

Germany Russia CADFEM GmbH CADFEM CIS Grafi ng b. München Moscow www.cadfem.de www.cadfem-cis.ru

Publication details

Publisher Dynardo GmbH Steubenstraße 25 99423 Weimar www.dynardo.de [email protected] Publication worldwide Executive Editor & Layout Henning Schwarz © Images [email protected] SIEMENS AG: p. 22 (top) | Christian Meyer, Constantin Beyer: p. 25 Fotolia, Jörg Vollmer (turbine): p. 4/6/8 Registration Local court Jena: HRB 111784 Copyright © Dynardo GmbH. All rights reserved VAT Registration Number The Dynardo GmbH does not guarantee or warrant accuracy or DE 214626029 completeness of the material contained in this publication.