Sign-Constrained Surrogate Learning for Reliability-Based Design Optimization Under Limited Objective Information

Reliability-based design optimization (RBDO) needs an explicit objective function, which early-stage design rarely offers, because only a few noisy evaluations exist. Engineers usually do know the direction in which each variable acts; for example, a longer weld cannot lower cost. We encode this knowledge as sign constraints on a Ridge regression surrogate of the objective. Variables with competing physical pathways are split into virtual variables the net effect of which is left to the data. The surrogate replaces only the objective. The reliability constraints stay on the original limit-state functions and are evaluated with the First-Order Reliability Method (FORM), so surrogate error never enters the reliability analysis. On the welded beam and the Golinski speed reducer, the signs help only for variables the effect of which is small relative to the noise, which an effect-to-noise ratio from a first unconstrained fit identifies. On Golinski the paired gain in test R2 reaches 0.01 in 8 of 17 cells, though single datasets can lose accuracy, and with 20 to 100 labels the full-information design is returned in 47% to 86% of the replicates, against 11% to 29%, although mean mass differs by only 0.2% to 0.3%. A wrong sign on a strong variable lowers R2 by 0.42 to 0.83, so doubted signs should be left free.

Authors

Publication Details

Journal
Machines
Published
2026-11-17
DOI
https://doi.org/10.3390/machines14111364
Primary Topic
Probabilistic and Robust Engineering Design
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Sign-Constrained Surrogate Learning for Reliability-Based Design Optimization Under Limited Objective Information

Ahmed Azab, Khaldon T. Meselhy
1 links
Machines
Probabilistic and Robust Engineering Design
article

Sign-Constrained Surrogate Learning for Reliability-Based Design Optimization Under Limited Objective Information

Ahmed Azab, Khaldon T. Meselhy
article en

Abstract

Reliability-based design optimization (RBDO) needs an explicit objective function, which early-stage design rarely offers, because only a few noisy evaluations exist. Engineers usually do know the direction in which each variable acts; for example, a longer weld cannot lower cost. We encode this knowledge as sign constraints on a Ridge regression surrogate of the objective. Variables with competing physical pathways are split into virtual variables the net effect of which is left to the data. The surrogate replaces only the objective. The reliability constraints stay on the original limit-state functions and are evaluated with the First-Order Reliability Method (FORM), so surrogate error never enters the reliability analysis. On the welded beam and the Golinski speed reducer, the signs help only for variables the effect of which is small relative to the noise, which an effect-to-noise ratio from a first unconstrained fit identifies. On Golinski the paired gain in test R2 reaches 0.01 in 8 of 17 cells, though single datasets can lose accuracy, and with 20 to 100 labels the full-information design is returned in 47% to 86% of the replicates, against 11% to 29%, although mean mass differs by only 0.2% to 0.3%. A wrong sign on a strong variable lowers R2 by 0.42 to 0.83, so doubted signs should be left free.

MachinesVol. 14(11)
Industry, innovation and infrastructure
Probabilistic and Robust Engineering Design
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon