Search Results

In structural design optimization, it is challenging to determine the optimal dimensions and material for each component simultaneously. Material selection of each part is always formulated as a categorical design variable in structural optimization problems. However, it is difficult to solve such mixed-variable problems using the metamodelbased strategy, because the prediction accuracy of metamodels deteriorates significantly when categorical variables exist. This paper investigates two different strategies of mixed-variable metamodeling: the “feature separating” strategy and the “all-in-one” strategy. A supervised learning-enhanced cokriging method is proposed, which fuses multi-fidelity information to predict new designs’ responses. The proposed method is compared with several existing mixed-variable metamodeling methods to understand their pros and cons. These methods include Neural Network (NN) regression, Classification and Regression Tree (CART) and Gaussian Process (GP).

Design optimization methods are commonly used for weight reduction subjecting to multiple constraints in automotive industry. One of the major challenges remained is to deal with a large number of design variables for large-scale design optimization problems effectively. In this paper, a new approach based on fuzzy rough set is proposed to address this issue. The concept of rough set theory is to deal with redundant information and seek for a reduced design variable set. The proposed method first exploits fuzzy rough set to screen out the insignificant or redundant design variables with regard to the output functions, then uses the reduced design variable set for design optimization. A vehicle body structure is used to demonstrate the effectiveness of the proposed method and compare with a traditional weighted sensitivity based main effect approach.

One of the major challenges in multiobjective, multidisciplinary design optimization (MDO) is the long computational time required in evaluating the new designs' performances. To shorten the cycle time of product design, a data mining-based strategy is developed to improve the efficiency of heuristic optimization algorithms. Based on the historical information of the optimization process, clustering and classification techniques are employed to identify and eliminate the low quality and repetitive designs before operating the time-consuming design evaluations. The proposed method improves design performances within the same computation budget. Two case studies, one mathematical benchmark problem and one vehicle side impact design problem, are conducted as demonstration.

Particle swarm optimization (PSO) is a relatively new stochastic optimization algorithm and has gained much attention in recent years because of its fast convergence speed and strong optimization ability. However, PSO suffers from premature convergence problem for quick losing of diversity. That is to say, if no particle discovers a new superiority position than its previous best location, PSO algorithm will fall into stagnation and output local optimum result. In order to improve the diversity of basic PSO, design of experiment technique is used to initialize the particle swarm in consideration of its space-filling property which guarantees covering the design space comprehensively. And the optimization procedure of PSO is divided into two stages, optimization stage and improving stage. In the optimization stage, the basic PSO initialized by Optimal Latin hypercube technique is conducted.

Vehicle restraint system design is a difficult optimization problem to solve because (1) the nature of the problem is highly nonlinear, non-convex, noisy, and discontinuous; (2) there are large numbers of discrete and continuous design variables; (3) a design has to meet safety performance requirements for multiple crash modes simultaneously, hence there are a large number of design constraints. Based on the above knowledge of the problem, it is understandable why design of experiment (DOE) does not produce a high-percentage of feasible solutions, and it is difficult for response surface methods (RSM) to capture the true landscape of the problem. Furthermore, in order to keep the restraint system more robust, the complexity of restraint system content needs to be minimized in addition to minimizing the relative risk score to achieve New Car Assessment Program (NCAP) 5-star rating.

In vehicle design, response surface model (RSM) is commonly used as a surrogate of the high fidelity Finite Element (FE) model to reduce the computational time and improve the efficiency of design process. However, RSM introduces additional sources of uncertainty, such as model bias, which largely affect the reliability and robustness of the prediction results. The bias of RSM need to be addressed before the model is ready for extrapolation and design optimization. This paper further investigates the Bayesian inference based model extrapolation method which is previously proposed by the authors, and provides a systematic and integrated stochastic bias corrected model extrapolation and robustness design process under uncertainty. A real world vehicle design example is used to demonstrate the validity of the proposed method.

Response Surface Model (RSM)-based optimization is widely used in engineering design. The major strength of RSM-based optimization is its short computational time. The expensive real simulation models are replaced with fast surrogate models. However, this method may have some difficulties to reach the full potential due to the errors between RSM and the real simulations. RSM's accuracy is limited by the insufficient number of Design of Experiments (DOE) points and the inherent randomness of DOE. With recent developments in advanced optimization algorithms and High Performance Computing (HPC) capability, Direct Multidisciplinary Design Optimization (DMDO) receives more attention as a promising future optimization strategy. Advanced optimization algorithm reduces the number of function evaluations, and HPC cut down the computational turnaround time of function evaluations through fully utilizing parallel computation.

Sampling-based methods are general but time consuming for solving a Reliability-Based Design Optimization (RBDO) problem. In order to alleviate the computation burden, score function together with the Monte Carlo method was used to compute the stochastic sensitivities of reliability functions. In literature, re-weighting schemes were shown to converge faster than the regular Monte Carlo method. In this paper, a reweighting scheme together with score function is employed to perform sampling-based stochastic sensitivity analysis to improve the computational efficiency and accuracy. An analytical example is used to show the advantages of the proposed method. Comparisons to the conventional methods are made and discussed. Two RBDO problems are solved to demonstrate the use of the proposed method.

Reliability-based design optimization (RBDO) has been widely used to obtain a reliable design via an existing CAE model considering the variations of input variables. However, most RBDO approaches do not consider the CAE model bias and uncertainty, which may largely affect the reliability assessment of the final design and result in risky design decisions. In this paper, the Gaussian Process Modeling (GPM) approach is applied to statistically correct the model discrepancy which is represented as a bias function, and to quantify model uncertainty based on collected data from either real tests or high-fidelity CAE simulations. After the corrected model is validated by extra sets of test data, it is integrated into the RBDO formulation to obtain a reliable solution that meets the overall reliability targets while considering both model and parameter uncertainties.

Finite Element (FE) models are widely used in automotive for vehicle design. Even with increasing speed of computers, the simulation of high fidelity FE models is still too time-consuming to perform direct design optimization. As a result, response surface models (RSMs) are commonly used as surrogates of the FE models to reduce the turn-around time. However, RSM may introduce additional sources of uncertainty, such as model bias, and so on. The uncertainty and model bias will affect the trustworthiness of design decisions in design processes. This calls for the development of stochastic model interpolation and extrapolation methods that can address the discrepancy between the RSM and the FE results, and provide prediction intervals of model responses under uncertainty.

Multidisciplinary design optimization (MDO) methods are commonly used for weight reduction in automotive industry. The design variables for MDO are often selected based on critical parts, which usually are close to optimal after many design iterations. As a result, the real weight reduction benefit may not be fully realized due to poor selection of design parameters. In addition, most applications require running design of experiments (DOE) to explore the full design space and to build response surfaces for optimization. This approach is often too costly if too many design variables are simultaneously considered. In this research, an alternative approach to address these issues is presented. It includes two optimization phases. The first phase uses critical parts for design iterations and the second phase use non-critical for weight reduction. A vehicle body structure is used to demonstrate the proposed strategy to show its effectiveness.

Computer Aided Engineering (CAE) has become a vital tool for product development in automotive industry. Various computer models for occupant restraint systems are developed. The models simulate the vehicle interior, restraint system, and occupants in different crash scenarios. In order to improve the efficiency during the product development process, the model quality and its predictive capabilities must be ensured. In this research, an objective model validation metric is developed to evaluate the model validity and its predictive capabilities when multiple occupant injury responses are simultaneously compared with test curves. This validation metric is based on the probabilistic principal component analysis method and Bayesian statistics approach for multivariate model assessment. It first quantifies the uncertainties in both test and simulation results, extracts key features, and then evaluates the model quality.