Simulation-Based Optimization of a Series Hydraulic Hybrid Vehicle using Additional Optimization Algorithms
Title: Simulation-Based Optimization of a Series Hydraulic Hybrid Vehicle using Additional Optimization Algorithms
DNr: SNIC 2019/3-373
Project Type: SNIC Medium Compute
Principal Investigator: Liselott Ericson <liselott.ericson@liu.se>
Affiliation: Linköpings universitet
Duration: 2019-07-01 – 2019-10-01
Classification: 20399
Keywords:

Abstract

Simulation-based design optimization for complex technical products enables comprehensive design space explorations in the conceptual phase of the product design process, i.e. when design changes have far-reaching consequences, but do not entail the same cost of changes as later on. The simulation-based approach is of special interest where the system operating behaviour depends on interdependent subsystem characteristics, including the usage profile. To use simulation-based optimization (SBO) effectively in the design process, efficient algorithms are required that yield close-to-globally-optimal solutions, even for complex problem formulations, with limited computational resource usage. Researchers at the Division of Fluid and Mechatronic Systems (Flumes) explored a number of algorithms in a previous project (SNIC 2018/3-573): a sequential direct search algorithm, limited parallelized variants (up to four threads), a standard evolutionary algorithm as well as two state-of-the-art evolutionary algorithms. The design problem for this was the simultaneous component and control design optimization for a series hydraulic hybrid vehicle (SHHV): a hydraulic hybrid transmission includes a hydraulic energy storage component for energy recuperation and temporary energy storage from engine operating point modulation. Its discharge profile and limited energy density leads to a highly transient energy reserve for the transmission, requiring a well-tuned control strategy to ensure sufficient performance. The system was modelled in Hopsan, the open-source multi-disciplinary simulation tool developed at Flumes. The previous work illustrated the costs and benefits of using evolutionary algorithms over sequential and limited parallel direct-search algorithms (the latter of which can just about be handled by standalone computers, though not with sufficient statistical reliability). The results have since been published in “Engineering Optimization”. This follow-up study aims to extend the knowledge base on algorithms for SBO by - evaluating further standard evolutionary algorithms (Genetic Algorithm, alternative parameterization of the previous algorithms), - exploring the performance of highly parallelized direct-search algorithms (8/16/32 threads), - and potentially applying both a robust-design-optimization (RDO) variant of the direct-search algorithm and a gradient-based algorithm. Results gained from the use of the NSC resources will serve first and foremost as a knowledge for future simulation-model-based design of complex technical products by providing for the SHHV example case a performance comparison for different optimization algorithms. This allows in future research endeavours to refer to the conducted set of optimizations for algorithm selection based on a simulation case instead of mathematical test functions. The possible inclusion of an RDO-algorithm points towards the necessity of addressing the potentially highly sensitive results that a numerical optimization yields in order to make the optimization results implementable.