the open-source approach for computational …...the open-source approach for computational modeling...

The Open-Source Approach for Computational Modeling and Simulation

for Earthquake Engineering: History, Accomplishments, and Future Needs

Gregory L. Fenves

Cockrell School of EngineeringThe University of Texas at Austin

National Autonomous University of MexicoMexico City, Feb. 5, 2015

Fundamentals Are Timeless

Ray Clough, UC Berkeley, 1960s

But Software Evolves Slowly

Compute Technology

Base Code

Elements

Input Language

User Interface

Simulation State-of-the-Art

• “The Good”

– Linear structural analysis routine

– Good commercial software widelyused and integrated with BIM

– Nonlinear static analysis becoming important

– Performance methods becoming more widelyused, e.g. ATC 58

• “The Bad”

– Linear analysis with equivalent lateral loads is not simulation

– Nonlinear static analysis uses very simplified models;it is not simulation

– “Performance criteria” not thoroughly investigated, e.g. FEMA 356, nor models adequately developed

– Long way to go in including uncertainty quantification

... and the Ugly

• Many nonlinear analysis methods basedon concepts from 1980s at often softwarearchitecture from 1970-1980s

• Underinvestment in research in simulation,and what is done is not well organized

• Poor linkages between fundamental experimental studies and modeling; insufficient validation and verification of models

• Simulation models, methods, and computational procedures in earthquake engineering have not kept up with rapid advances in computing hardware, software engineering, databases, network communications

– Limited interaction with computer science

– Inadequate education of students in computing

Observations on Historical Situationwith Simulation Software

• Tight binding of models in research and commercial codes

is an impediment to new research and implementation of

models for professional practice.

• Embedding of computational procedures in codes makes it

difficult to experiment and take advantage of computing

technology:

– Parallel and distributed computers

– Computational grids

– Now, cloud computing

• “Closed-source” is the norm, whereas other fields have

adopted “open-source” software for communities

users.

Simulation Needs inEarthquake Engineering

• Performance-based engineering depends on evaluation of damage and estimate of consequences

• Rational, validated models of behavior of structural and geotechnical materials, components and systems are needed for simulating performance

• Simulation applications:

– Assessment of performance

– Design using parameterized models, including optimization with performance constraints

– Reliability-based design

– Regional loss estimation and disaster planning

• Additional applications include structural and system health monitoring for control and operations

Simulation Has TransformedOther Engineering and Science Fields

• Computational chemistry,computational biology

• Material science, particularly atnano-scale

• Computational fluid dynamics

– Aerodynamics

– Building interior environment

– Virtual wind tunnels

• Aircraft design

• Automotive design

• U.S. nuclear weapons stewardship (ASCI, PSAAP)

Vision for Earthquake Engineering Simulation (2007)

Computational modeling and simulation is central to the vision of NEES to transform the development of new earthquake engineering solutions from being primarily based on experiments to a balanced use of simulation and experimentation using computational models validated by experimental data.

A close integration of modern computational models and simulation software with other NEES applicationsand services will provide the earthquake engineering community, and broad engineering users, new capabilities for developing innovative and cost-effective solutions.

Software Framework

• A framework is a set of cooperating software components

for building applications in a specific domain

• A framework dictates the architecture of the application – it

represents the design decisions common to the application

domain

• A frameworks is based on the assumption that an

architecture will work for most applications within the

domain

• Loose-coupling of components within the framework is

essential for extensibility and re-usability for applications

• Examples: Visualization (GLUT), Hadoop, Google Apps, …

• A framework is not a “code”

http://opensees.berkeley.edu• OpenSees has been under

development by PEER since 1998. NEES supported 2005-2014. PEER since then.

• Windows application downloaded over 10,000 times a year.

• Parallel Applications utilize over 1,000,000 CPU hours on NSF XSEDE compute resources yearly.

• Open-source and royalty free license for non-commercial use and and internal commercial use.

• License must be obtained for software developers including OpenSees code in their applications.

• Written in C++, C and Fortran

(C++ being the main language)

OpenSees Worldwide Usage (2014)

OpenSees Approach to Simulation

• Basic approach:

– Modular software design for implementing and integrating modeling, numerical methods, and IT for scalable, robust simulation

– Focus on capabilities needed for performance-based engineering

– Programmable interfaces

• Most users: a “code” for nonlinearanalysis. Fully scriptable.

• Generally: a software framework for developing simulation applications.

Structural Modeling and Simulation

Joints with both bond-slip springs and shear springs

Column base bond-slip springs

s

s

s

e

e

e

Non-ductile RC frames and calibration of

building code provisions

C. Haselton, G. Deierlein, Stanford

Corotational geometric

transformations

Bilinear

Soil-Structure Modeling

Ahmed Elgamal

UC San Diego, 2008

Examples of OpenSees Applications

• Parametric studies to examine relationship

between intensity and damage for PBEE and

design procedures

• Computational reliability for PBEE

• Soil-structure-foundation interaction

• Spatial distribution of damage

• Simulation of bridge performance

Open-Source CommunitySimulation Framework

Conceptual Approach for Simulation

InformationTechnology Software framework,

Databases, Visualization,Internet/grid computation

ComputationAlgorithms,Solvers,Parallel/distributedcomputing

Models

Material, component, system models

Simulation models,Performance models,Limit state models

Simulation Software Architecture

Compute Technology

Base Code

Elements

Input Language

User Interface

Traditional Code

Model Domain

Materials

Framework of Components

Solvers

Compute Technology

Elements

Databases

Solution Procedures

Other

Vis

ual

izat

ion

Application Program Interface (API)

Software Depends on Expressiveness of Language and Power of Processor(s)

What is Object-OrientedProgramming?

• Object-oriented programs are composed of objects that bind data and operators on data

• Objects are operated upon by sending messages to it. The public interface defines the operations on an object

• Object’s internal state is encapsulated in the object. The implementation is private

• Classes define the software behavior of objects.

• Classes and their objects are designed to represent key abstractions

• A programmer should be able to use a class through the interface independently of the implementation

Object-Oriented Finite Element Framework

1997

Structural Models as Aggregation Pattern

AnalysisModelBuilder

Builder - separate representationFrom construction

Domain

LoadPattern ElementNodeConstraints

Aggregation

Analysis Class for Simulation

DO_Numberer AnalysisModel

SolutionAlgSystemofEqs

ConstraintHandler Integrator

Analysis DomainAnalysis class is responsiblefor performing an analysis on adomain and is formed byAggregation.

Example of Analysis Class

Analysis

analyze(n)

StaticAnalysis

analyze(n)

theModel

theAlgorithm

theIntegrator

EquiSolAlgo

solveCurrentStep()

StaticIntegrator

newStep()

commitDomain()

AnalysisModel

for i=1:n

theIntegrator->newStep()

theAlgorithm->solveCurrentStep()

theModel->commitDomain()

Strategy Pattern

Beam-Column Models I

Element

Basic System

GeometricTran

LinearLinearPDCorotational

Sectione, s

Material

e s

No assumptions are made on section or material behavior; each level in the hierarchy can be defined independently of other levels

s

e

Beam-Column Models II

Basic System

Displacement Force

Material

ForceDeformationUniaxialMaterial

BeamWithHinges

2

FiberSectionFiber

UniaxialFiber

Concrete01 Steel01

s

e s

e

as = [1 –y z]

Form Follows Mechanics

Types of Behavior

• Ductile and brittle

modes represented

• Solution method

converges rapidly

even with strong

softening

PH

PElastic

Concentrated

Plasticity

A A

Domain decomposition implemented with Actors–distributed computational objects with a communication protocol.

OpenSees Parallel Processing

Large-Scale Computing and Visualization

Ahmed Elgamal

UC San Diego, 2008

• 30,237 nodes

• 1,140/280 linear/nonlinear BC elements

• 81 linear shell elements

• 23,556 solid brick elements

• 1,806 zero-length elements

OpenSees as Open-Source Software

• Roadmap

• Architecture

• Program Interfaces

(API’s)

• Code repository

– Checkout/in

– Branch/merge

– Versioning

• Release engineering

• Testing

• Validation & Verification

Open Source is…

# set some variables set gMotion el_centroset scale 1.set roofWeight [expr 80*120.*72./1000.]; #kips; set floorWeight [expr 95*120.*72./1000.];set numFrameResisting 2.0; #load resisting frames set percentLoadFrame [expr 15./120.]set dampRatio 0.03set mode1 1set mode2 3set Fy 60.set E 30000.set b 0.03

# set up my lists set floorOffsets {216. 150. 150. 150. 150. 150.}; # inchesset colOffsets {288. 288. 288.}; #inchesset colSizes {W30X173 W30X173 W27X146 W27X146 W24X104 W24X104};set colExtSizes {W14X193 W14X193 W14X159 W14X159 W14X109 W14X109};set beamSizes {W30X99 W30X99 W27X94 W27X94 W24X76 W24X76};# build and run the model using standard template filesource SteelMomentFrame2d_UniformExcitation.tcl

2D Steel Moment FrameLoad: 95psf typical, roof 80psf

E=29,000, Fy=50.0, b =0.003

3% Rayleigh Damping 1st and 3rd Modes

18 ft

12.5 ft

24 ft

Concrete Building Study113 records, 4 intensities

3 hour a record, 1356

hours or 56.5 days.

Ran on 452 processors

On XSEDE in less than 5

hours.

How Does OpenSees Compare With Commercial Software?

Prof. Xin-Zheng LuTsinghua University

Similar Results -(Results with commercial applications the same iff model and analysis are the same)

Lessons and Observations

• Many see the benefits of exchanging research and ideas through software

• Success depends on ability of developers to understand abstractions in the software design

• “Not invented here” is sometimes an issue

• Computing education and experience of civil engineers makes the learning curve look steep

• Many users just want the code and are not interested in open-source

• Documentation is never good enough

• Long-term support of an organization is necessary

• Innovation is possible, but it takes long-term commitment

Using the Internet for Simulation (2001)

Model Domain

Materials

Solvers

Compute Technology

Elements

Databases

Solution Procedures

Other

Vis

ual

izat

ion

InternetAPI’s

Schematic of Simulation in the Future (2002)myCompany.building.com

modelBuilders.com

EngineeringTeam

aci.models.org

usgs.gov

PERFORMANCECODE

buildingcode.org computation.com

Experiments

Cloud as Deployed Services

Key Question for the Future

• How will nonlinear simulation models based on fundamentals be developed for use in performance-based design?

• How will validation, verification, and uncertainty quantification (VVUQ) be incorporated in earthquake engineering simulation?

• How will the earthquake engineering industry use transformational cloud-based services?