The purpose of this paper is to propose a new Set-based concurrent engineering method using Bayesian active learning and to show an application to a multi-disciplinary design optimization problem.

To solve multi-disciplinary design problem, many experts in different discipline fields need to corroborate each other. Concurrent engineering (CE) is an efficient work methodology emphasizing the parallelization of tasks. Based on the CE, development time and costs can be reduced. A requirement to implement CE is how to decouple the multi-disciplinary design problem into set of discipline sub-problems.

Multi-disciplinary design optimization (MDO) is a method to find optimal solution of multi-disciplinary phenomena. By using the MDO, a large, complex system design problem can be decomposed into sets of simpler design problems. In practice, one difficulty of applying the MDO to early stage of product development is how to hold the optimal cascaded design throughout design process against uncertainty of optimization conditions such as change of number of constraint functions or its boundary conditions. These uncertainties can be caused by considering change of trend in market demand or competitors in latter design stage. As a result of these changes, a single solution may not hold. Moreover, critical rework is inevitable if it causes in latter process.

To avoid this issue, it is important in early design stage to solve not only a single target solution but also feasible design solutions that satisfy all multi-disciplinary requirements. In this study, we introduce a “*set-based design approach*”, in which feasible set are solved instead of solving single feasible solution to overcome the critical rework due to uncertainty of design condition.

The set-based design can enhance to maintain feasibility and robustness of cascaded solutions throughout design process. Moreover, the set-based design approach has an advantage in the decoupling ability of multi-disciplinary design problem because iterations between optimization problems of each discipline sub-problems, which is required in some MDO methods are unnecessary. As a result, the set-based design approach makes it possible to study each discipline sub-problem concurrently

In order to discover the feasible region with limited resources, efficient sampling strategy using CAE simulation is necessary. In this study, a sampling strategy based on Bayesian active learning is proposed to discover feasible region of multi-disciplinary constraints concurrently. In the proposed method, Gaussian Process models of the multi-disciplinary constraints are trained. Based on posterior distributions of trained Gaussian Processes, two different type of acquisition functions, Probability of Feasibility and Entropy Search are concurrently evaluated and maximized to generate new sampling point to improve prediction accuracy of feasible region effectively.

To show validity of proposed method, a multi-disciplinary design problem of torsion-beam type suspension was solved. The feasible region where 15 inequality constraints from 3 multidiscipline were satisfied was efficiently and concurrently discovered using the proposed active learning method.


A Set Based Design Method Using Bayesian Active Learning

Division / Sponsors: 
Design Engineering Division and Computers & Information In Engineering Division


Paper Proceedings: https://asmedigitalcollection.asme.org/IDETC-CIE/IDETC-CIE2020/volume/83907

IDETC2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.