Browse Prior Art Database

A Gaussian-Gaussian bias correction-enhanced metamodeling algorithm for vehicle safety optimization considering multi-material choices Disclosure Number: IPCOM000246621D
Publication Date: 2016-Jun-21
Document File: 4 page(s) / 346K

Publishing Venue

The Prior Art Database

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 32% of the total text.

Page 01 of 4

A Gaussian-Gaussian bias correction-enhanced metamodeling algorithm for vehicle safety optimization considering multi-material choices

Optimization of vehicle design is a very challenging problem. Design of Experiments (DOE) & metamodeling (i.e. response surface model, RSM) technique is widely applied in modeling the vehicle performances. New design's responses can be efficiently predicted by metamodels, but the accuracy may not be satisfying.

The accuracy issue is even more challenging for vehicle safety design considering multi-material choices, as safety responses are highly non-linear and discontinuous, which is difficult to predict using metamodels. Three extraordinary challenges exist in current tools: (1) the existing software tools fail to converge for training some highly nonlinear metamodels. The safety engineers have to try multiple metamodels for certain responses until finding one that can converge. Such a trial-and-error process significantly lowers the engineering efficiency; (2) metamodel-based predictions have huge errors for safety performances; (3) if the selection of multiple materials (discrete and unordered design variable) is considered, the metamodel's convergency and accuracy will further deteriorate.

The inaccuracy of safety metamodels can significantly reduce the product qualities. First, it is difficult,
if not impossible, to find the optimal designs of high performances (e.g. lightest weight with NCAP 5- star rating). Inaccurate metamodels will mislead the optimization algorithm to inferior designs. Second, design feasibility is low. In some extreme cases, metamodel-based optimization fails to find any feasible designs which satisfying all design constraints. Such failure will drastically increase engineers' manual efforts for finding feasible designs.

In this work, a Gaussian-Gaussian bias correction-enhanced metamodeling method is developed. This technique employs Gaussian Process Regression methods to quantify and correct the predict bias of metamodels. The proposed method demonstrates consistent improvements of metamodels' accuracy.


In the process of training metamodel (i.e. RSM), the existing tools may fail to converge for certain highly nonlinear responses. Traditionally, the engineers have to manually try different metamodels until find one that converges. In addition, the traditional metamodels have very poor accuracy when used on safety responses. It will result in low performance and less feasibility for optimization search.

To address the aforementioned challenges, an industry first advanced metamodel-bias correction method is developed. The state-of-the-art Gaussian-Gaussian method is employed in this algorithm. Metamodels of the vehicle safety responses are established by Gaussian Process regression, and then its accuracy is further enhanced using Gaussian Process regression bias correction technique. Gaussian regression captures function nonlinearity and quantifies uncertainties su...