Predicting formula 1 tire aerodynamics : sensitivities, uncertainties and optimization
- The efficient design of complex engineering devices requires detailed understanding of the physical processes. These may either negatively or positively affect performance even in the presence of variability in the operating environment or uncertainties in manufacturing. The focus of this work is the prediction of the aerodynamics associated with an isolated, stationary and rotating Formula 1 tire and brake assembly. The end goal is to design a better performing tire assembly in the presence of uncertainties, while reducing as much as possible the computational effort required. This required innovations in computational modeling, uncertainty analysis, and optimization techniques. This application is extremely challenging due to the complex fluid dynamics processes such as impingement, jetting, large-scale separation, and wake recovery, inherent in the problem. Previous literature has indicated differing mechanisms explaining the dominant features such as the pressure loading, wake structures and unsteadiness. Limited work has been published on the aerodynamics of realistic tire geometry with specific emphasis on advanced turbulence closures (i.e. large eddy simulation - LES) and optimization under uncertainty; this work provides the first comprehensive reference for this problem. In this work we use large-scale simulations to validate turbulence closures against experimental particle image velocimetry results. We then for the first time quantify the uncertainty associated with Formula 1 tires through sensitivity analysis, and finally optimize the shape of a brake duct under these uncertainties using novel algorithms. Part I of this thesis focuses on a deterministic analysis of a realistic tire and brake assembly geometry. The numerical simulations were setup to reproduce a controlled experiment performed at Stanford, in a low-speed close-loop wind tunnel. Multiple turbulence closures are evaluated to determine the closure that most accurately matches experimental results. The results show that LES is more accurate than Reynolds averaged Navier-Stokes (RANS) equations at predicting experimental results. The bluff body flow around an isolated tire in a closed wind tunnel is shown to be inherently unsteady. Time accurate methods (i.e. Unsteady RANS, LES) that capture the unsteadiness of the flow specifically in the near wake, contact patch, and hub regions are more precise at predicting macro-scale structures. A novel RANS wall treatment for moving walls is introduced to correctly represent situations in which the tangential velocity of the wall does not match the freestream velocity; this allows the use of coarser meshes without penalizing dramatically the prediction quality. Part II of this thesis focuses on sensitivity, uncertainty quantification and optimization. This is a very important issue in computational engineering, especially in highly competitive environments such as in Formula 1. Current design practice based on a nominal scenario is gradually being replaced with an optimization process that targets realistic race conditions where temperatures, track condition, and 'dirty' air all impact the overall performance of the car. Sensitivities associated with inflow conditions, treatment of rotating components, geometrical tire uncertainties, contact patch assumptions, and turbulence model closure are quantified and ranked in order of importance. In this work, multi-dimensional response surfaces are constructed to relate input variables to output quantities of interest and to compute confidence intervals on the numerical predictions. A classic non-dominated sorting based multi-objective evolutionary algorithm is used to optimize the brake cooling duct to achieve the highest air flow mass capture with the least possible tire drag. We continued this analysis by considering the effect of uncertainties in the optimization framework. An extension of the optimization algorithm, the Probabilistic Non-dominated Sorting Genetic Algorithm (P-NSGA) is proposed and tested. The result is a tight coupling between the uncertainty quantification procedure and the optimization procedure giving the most robust brake duct design in the presence of uncertainty. With the availability of high-performance computing clusters, the design, optimization, and aerodynamic prediction of extremely complex configurations is becoming increasingly prevalent. In order to efficiently explore the full parameter space for design purposes or to assess the effect of uncertainties, multiple deterministic simulations (or an 'ensemble') need to be performed. In this work, a simulations environment, Leland, has also been developed to schedule, monitor and stir the calculation ensemble and extract runtime information as well as simulations results and statistics on the fly. Leland is a dynamic scheduler that starting from a small ensemble automatically selects the new candidate simulations to be performed to increase the efficiency of both the optimization procedure and the uncertainty quantification method. Leland is equipped with an auto-tuning strategy for optimal load balancing and fault tolerance checks to ensure that a simulation or a processor stall is detected and does not impact the overall ensemble.
|Type of resource
|electronic; electronic resource; remote
|1 online resource.
|Axerio-Cilies, John Anthony
|Stanford University, Department of Mechanical Engineering
|Eaton, John K
|Jameson, Antony, 1934-
|Eaton, John K
|Jameson, Antony, 1934-
|Statement of responsibility
|Submitted to the Department of Mechanical Engineering.
|Thesis (Ph.D.)--Stanford University, 2012.
- © 2012 by John Anthony Axerio-Cilies
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...