Design Optimization of Safety Critical Component for Fatigue and Strength Using Simulation and Data Analytics

This presentation was made at CAASE18, The Conference on Advancing Analysis & Simulation in Engineering. CAASE18 brought together the leading visionaries, developers, and practitioners of CAE-related technologies in an open forum, to share experiences, discuss relevant trends, discover common themes, and explore future issues.

Resource Abstract

The current work focuses on simulation based optimization of a complex, safety critical component where it is prohibitively expensive to carry out finite element analysis (FEA) simulations for all possible sample realizations and therefore requires statistical or machine learning techniques for a timely yet accurate solution. The applicability of machine learning further brings the opportunity of performing in-service monitoring using sensor data and thereby performing predictive maintenance.

A deterministic design optimization of Blowout Preventer (BOP) for operation in high pressure subsea environment is performed with the objective of maximizing the fatigue life of the equipment while reducing the weight and displacement at a critical location. BOPs are mechanical devices designed to seal off wellbore, safely control and monitor oil and gas well in case of blowout.

The optimization process requires exploration of the design space, creating response surface functions to represent the complex input-output relationship, and running the optimization algorithm based to response surface model. The process requires many sample evaluations and therefore, a workflow automating is done using Simulia portfolio software Isight that performs parametric optimization and automation. The design input parameters with their respective realistic lower and upper bounds are defined to study the design space. Based on input-output data traditional methods for response surface generation are used.

A separate algorithm based on the principles of machine learning is also used to generate the response surface function using the design space exploration matrix of input-output variables & parameters. Machine learning can help in design simulations by generating predictive models to estimate output given initial parameters. The output of a simulation is approximated using deep network architectures for regression. This approach generally requires less sampling of design space compared to traditional methods such as radial basis function, krigging, etc. This framework can be included within the Isight environment easily which may reduce the overall burden of individual FEA runs.

Document Details

ReferenceCAASE_Jun_18_27
AuthorChakraborty. A
LanguageEnglish
TypePresentation
Date 7th June 2018
OrganisationVIAS
RegionAmericas

Download


Back to Search Results