Search Results

Design optimization often relies on computational models, which are subjected to a validation process to ensure their accuracy. Because validation of computer models in the entire design space can be costly, we have previously proposed an approach where design optimization and model validation, are concurrently performed using a sequential approach with variable-size local domains. We used test data and statistical bootstrap methods to size each local domain where the prediction model is considered validated and where design optimization is performed. The method proceeds iteratively until the optimum design is obtained. This method however, requires test data to be available in each local domain along the optimization path. In this paper, we refine our methodology by using polynomial regression to predict the size and shape of a local domain at some steps along the optimization process without using test data.

Design optimization often relies on computational models, which are subjected to a validation process to ensure their accuracy. Because validation of computer models in the entire design space can be costly, a recent approach was proposed where design optimization and model validation were concurrently performed using a sequential approach with both fixed and variable-size local domains. The variable-size approach used parametric distributions such as Gaussian to quantify the variability in test data and model predictions, and a maximum likelihood estimation to calibrate the prediction model. Also, a parametric bootstrap method was used to size each local domain. In this article, we generalize the variable-size approach, by not assuming any distribution such as Gaussian. A nonparametric bootstrap methodology is instead used to size the local domains. We expect its generality to be useful in applications where distributional assumptions are difficult to verify, or not met at all.

A common approach to the validation of simulation models focuses on validation throughout the entire design space. A more recent methodology validates designs as they are generated during a simulation-based optimization process. The latter method relies on validating the simulation model in a sequence of local domains. To improve its computational efficiency, this paper proposes an iterative process, where the size and shape of local domains at the current step are determined from a parametric bootstrap methodology involving maximum likelihood estimators of unknown model parameters from the previous step. Validation is carried out in the local domain at each step. The iterative process continues until the local domain does not change from iteration to iteration during the optimization process ensuring that a converged design optimum has been obtained.

Existing methods for design optimization under uncertainty assume that a high level of information is available, typically in the form of data. In reality, however, insufficient data prevents correct inference of probability distributions, membership functions, or interval ranges. In this article we use an engine design example to show that optimal design decisions and reliability estimations depend strongly on uncertainty characterization. We contrast the reliability-based optimal designs to the ones obtained using worst-case optimization, and ask the question of how to obtain non-conservative designs with incomplete uncertainty information. We propose an answer to this question through the use of Bayesian statistics. We estimate the truck's engine reliability based only on available samples, and demonstrate that the accuracy of our estimates increases as more samples become available.

Medium trucks constitute a large market segment of the commercial transportation sector, and are also used widely for military tactical operations. Recent technological advances in hybrid powertrains and fuel cell auxiliary power units have enabled design alternatives that can improve fuel economy and reduce emissions dramatically. However, deterministic design optimization of these configurations may yield designs that are optimal with respect to performance but raise concerns regarding the reliability of achieving that performance over lifetime. In this article we identify and quantify uncertainties due to modeling approximations or incomplete information. We then model their propagation using Monte Carlo simulation and perform sensitivity analysis to isolate statistically significant uncertainties. Finally, we formulate and solve a series of reliability-based optimization problems and quantify tradeoffs between optimality and reliability.

This paper proposes an approach for optimal design of multilevel systems under uncertainty. The approach utilizes the stochastic extension of the analytical target cascading formulation. The reliability of satisfying the probabilistic constraints is computed by means of the most probable point method using the hybrid mean value algorithm. A linearization technique is employed for estimating the propagation of uncertainties throughout the problem hierarchy. The proposed methodology is applied to a piston-ring/cylinder-liner engine subassembly design problem. Specifically, we assess the impact of variations in manufacturing-related properties such as surface roughness on engine attributes such as brake-specific fuel consumption. Results are compared to the ones obtained using Monte Carlo simulation.

This paper investigates the effects of manufacturing variations in fuel injectors on the engine performance with emphasis on emissions. The variations are taken into consideration within a Reliability-Based Design Optimization (RBDO) framework. A reduced version of Multi-Zone Diesel engine Simulation (MZDS), MZDS-lite, is used to enable the optimization study. The numerical noise of MZDS-lite prohibits the use of gradient-based optimization methods. Therefore, surrogate models are developed to filter out the noise and to reduce computational cost. Three multi-objective optimization problems are formulated, solved and compared: deterministic optimization using MZDS-lite, deterministic optimization using surrogate models and RBDO using surrogate models. The obtained results confirm that manufacturing variation effects must be taken into account in the early product development stages.

Developing a new technology requires decision-makers to understand the technology's implications on an organization's objectives, which depend on user needs targeted by the technology. If these needs are common between two organizations, collaboration could result in more efficient technology development. For hybrid truck design, both commercial manufacturers and the military have similar performance needs. As the new technology penetrates the truck market, the commercial enterprise must quantify how the hybrid's superior fuel efficiency will impact consumer purchasing and, thus, future enterprise profits. The Army is also interested in hybrid technology as it continues its transformation to a more fuel-efficient force. Despite having different objectives, maximizing profit and battlefield performance, respectively, the commercial enterprise and Army can take advantage of their mutual needs.

The target cascading methodology is applied to the conceptual design of an advanced heavy tactical truck. Two levels are defined: an integrated truck model is represented at the top (vehicle) level and four independent suspension arms are represented at the lower (system) level. Necessary analysis models are developed, and design problems are formulated and solved iteratively at both levels. Hence, vehicle design variables and system specifications are determined in a consistent manner. Two different target sets and two different propulsion systems are considered. Trade-offs between conflicting targets are identified. It is demonstrated that target cascading can be useful in avoiding costly design iterations late in the product development process.