Search Results

Deep learning based approaches for object detection are heavily dependent on the nature of data used for training, especially for vehicles driving in cluttered urban environments. Consequently, the performance of Convolutional Neural Network (CNN) architectures designed and trained using data captured under clear weather and favorable conditions, could degrade rather significantly when tested under cloudy and rainy conditions. This naturally becomes a major safety issue for emerging autonomous vehicle platforms relying on CNN based object detection methods. Furthermore, despite a noticeable progress in the development of advanced visual deraining algorithms, they still have inherent limitations for improving the performance of state-of-the-art object detection. In this paper, we address this problem area by make the following contributions.

With increasingly stringent emissions regulations and concurrent requirements for enhanced engine thermal efficiency, a comprehensive characterization of the automotive gasoline fuel spray has become essential. The acquisition of accurate and repeatable spray data is even more critical when a combustion strategy such as gasoline direct injection is to be utilized. Without industry-wide standardization of testing procedures, large variablilities have been experienced in attempts to verify the claimed spray performance values for the Sauter mean diameter, Dv90, tip penetration and cone angle of many types of fuel sprays. A new SAE Recommended Practice document, J2715, has been developed by the SAE Gasoline Fuel Injection Standards Committee (GFISC) and is now available for the measurement and characterization of the fuel sprays from both gasoline direct injection and port fuel injection injectors.

This paper presents an assessment of competing algorithms for normalizing volume levels between tracks and/or sources in an automotive infotainment system. Portable media players such as smartphones and iPod® devices are extremely popular for listening to music collections or streaming content from the Internet. The lack of normalization is a source of dissatisfaction if the user experiences significant changes in audio level between tracks. Several commercially available algorithms exist to solve this problem. This research includes a double-blind listening test comparing an audio sample processed with the different leveling algorithms to an unprocessed reference. The listener preference rating is recorded and results indicate which algorithm is preferred.

Human intraspecific variation is a complex problem, but may be better understood by using computational models in tandem with knowledge about the genetic bases of phenotypic traits. These results can be used in a multitude of settings. To move closer to this goal, biologically-realistic mappings between genotype and phenotype are constructed using genetic algorithm and neural network-like models. These models allow for gene-gene and gene-environment interactions to be characterized in the resulting phenotype. Two types of model are introduced: a simple, two-layer model, and a more complex model. The final section will focus on trends of growth and development in relation to relationship to modeling anthropometric traits and other morphological phenomena.

Lower extremity (knee) injury prediction resulting from impact trauma is currently based on a bone fracture criterion derived from experiments on predominantly aged cadavers. Subsequent experimental and theoretical studies indicate that more aged, pathological specimens require higher, not lower, loads to initiate bone fracture. This suggests that a bone fracture criterion based solely on aged specimens may not be representative of the current driving population. In the current study, we sought to determine if cadaver age, physical size, sex, baseline joint pathology, or patellar geometry correlated with fracture load. An analysis was made of data from previous impact experiments conducted on fifteen isolated cadaver knees using a consistent impact protocol. The protocol consisted of sequentially increasing the impact energy with a rigid interface until gross fracture. Gross bone fractures occurred at loads of 6.9±2.0 kN (range 3.2 to 10.6 kN) using this protocol.

Electro-pneumatic valve actuators are used to eliminate the cam shaft of a traditional internal combustion engine. They are used to control the opening timing, duration, and lift of both intake and exhaust valves. A physics based nonlinear mathematical model called the level one model was built using Newton's law, mass conservation and thermodynamic principles. A control oriented model, the level two model, was created by partially linearizing the level one model for model reference parameter identification. This model reduces computational throughput and enables real-time implementation. A model reference adaptive control system was used to identify the nonlinear parameters that were needed for generating a feedforward control signal. The closed-loop valve lift tracking, valve opening and closing timing control strategies were proposed.

The traditional method of engine knock detection is to compare the knock intensity with a predetermined threshold. The calibration of this threshold is complex and difficult. A statistical knock detection method is proposed in this paper to reduce the effort of calibration. This method dynamically calculates the knock threshold to determine the knock event. Theoretically, this method will not only adapt to different fuels but also cope with engine aging and engine-to-engine variation without re-calibration. This method is demonstrated by modeling and evaluation using real-time engine dynamometer test data.

The requirement of reduced emissions and improved fuel economy led the introduction of direct-injection (DI) spark-ignited (SI) engines. Dual-fuel injection system (direct-injection and port-fuel-injection (PFI)) was also used to improve engine performance at high load and speed. Ethanol is one of the several alternative transportation fuels considered for replacing fossil fuels such as gasoline and diesel. Ethanol offers high octane quality but with lower energy density than fossil fuels. This paper presents the combustion characteristics of a single cylinder dual-fuel injection SI engine with the following fueling cases: a) gasoline for PFI and DI, b) PFI gasoline and DI ethanol, and c) PFI ethanol and DI gasoline. For this study, the DI fueling portion varied from 0 to 100 percentage of the total fueling over different engine operational conditions while the engine air-to-fuel ratio remained at a constant level.

Injuries to the lower extremities are common during car accidents because the lower extremity is typically the first point of contact between the occupant and the car interior. While injuries to the knee, ankle and hip are usually not life threatening, they can represent a large societal burden through treatment costs, lost work days and a reduced quality of life. The aim of the current study was to specifically study injuries associated with the knee and to propose a methodology which could be used to prevent future knee injuries. To understand the scope of this problem, a study was designed to identify injury trends in car crashes for the years 1979-1995. The NASS (National Accident Sampling System) showed that 10% of all injuries were to the knee, second only to head and neck injuries. Most knee injuries resulted from knee-to-instrument panel contact. Subfracture injuries were most common (contusions, abrasions, lacerations) followed by gross fracture injuries.

A practical and low cost Blind Spot Monitoring system is proposed. By using a single camera, the range and azimuth position of a vehicle in a blind spot are measured. The algorithm is based on the proposed RWA (Range Window Algorithm). The camera is installed on the door mirror and monitoring the side and rear of the host vehicle. The algorithm processes the image and identifies range and azimuth angle of the vehicle in the adjacent lane. This algorithm is applied to real situations. The 388 images including several kinds of vehicles are analyzed. The detection rate is 86% and the range accuracy is 1.6[m]. The maximum detection range is about 30[m].

This paper focuses on the numerical investigation of the mixing and combustion of ethanol and gasoline in a single-cylinder 3-valve direct-injection spark-ignition engine. The numerical simulations are conducted with the KIVA code with global reaction models. However, an ignition delay model mitigates some of the deficiencies of the global one-step reaction model and is implemented via a two-dimensional look-up table, which was created using available detailed kinetics models. Simulations demonstrate the problems faced by ethanol operated engines and indicate that some of the strategies used for emission control and downsizing of gasoline engines can be employed for enhancing the combustion efficiency of ethanol operated engines.

Because combustion engine equipped vehicles must conform to stringent hydrocarbon (HC) emission requirements, many of them on the road today are equipped with an engine air intake system that utilizes a hydrocarbon adsorber. Also known as HC traps, these devices capture environmentally dangerous gasoline vapors before they can enter the atmosphere. A majority of these adsorbers use activated carbon as it is cost effective and has excellent adsorption characteristics. Many of the procedures for evaluating the adsorbtive performance of these emissions devices use mass gain as the measurand. It is well known that activated carbon also has an affinity for water vapor; therefore it is useful to understand how well humidity must be controlled in a laboratory environment. This paper outlines investigations that were conducted to study how relative humidity levels affect an activated carbon hydrocarbon adsorber.

Internal combustion engines are designed to maximize power subject to meeting exhaust emission requirements and minimizing fuel consumption. Maximizing engine power and fuel economy is limited by engine knock for a given air-to-fuel charge. Therefore, the ability to detect engine knock and run the engine at its knock limit is a key for the best power and fuel economy. This paper shows inaudible knock detection ability using in-cylinder ionization signals over the entire engine speed and load map. This is especially important at high engine speed and high EGR rates. The knock detection ability is compared between three sensors: production knock (accelerometer) sensor, in-cylinder pressure and ionization sensors. The test data shows that the ionization signals can be used to detect inaudible engine knock while the conventional knock sensor cannot under some engine operational conditions.

Generalized predictive control (GPC) is a discrete time control strategy proposed by Clark et al [1]. The controller tries to predict the future output of a system or plant and then takes control action at present time based on future output error. Such a predictive control algorithm is presented in this paper for deceleration slip regulation in an automobile. Most of the existing literature on the anti-lock brake control systems lacks the effectiveness of the wheel lockup prevention when the automobile is in a skid condition (in a low friction coefficient surface with panic braking situation). Simulation results show that the predictive feature of the proposed controller provides an effective way to prevent wheel lock-up in a braking event.

The major objective of this paper is to address how the actual force versus extension relationship for a seat belt during a collision is different from the one obtained at typical low rate (static) conditions. We also look at what features of the tensile stress-strain characteristic are important for the optimal performance of a seat belt. To answer these questions experimentally we use our high rate Instron -1331. We also designed an experimental set up that required special grips and contact sensors for characterizing samples of belt and yarn. In the theoretical part we demonstrate the selected rates for the tensile testing as relevant to the collisions. We also discuss the importance of the energy absorbing capacity of the belts as the most relevant characteristic of the tensile curves for this application. We then show the effect of visco-elastic factors on energy absorbing properties of fibers during collisions and the role of weaving and dyeing the belt.

The process for accurately estimating product reliability early in the development process can be a difficult and costly task. Traditional methods like Reliability Prediction Models and Life Testing Strategies yield beneficial results when relative information is known about the product that is to be analyzed. When there is minimal information known (prior failure rates…) such a new concept design these above reliability methods have limitations. For these cases a Virtual Testing Strategies have proven to yield valuable results. This paper will demonstrate a reliability analysis procedure for a new automotive concept design. This analysis procedure composes of a mathematical model, model validation, parameter diagram, design of experiment (DOE), response surface, and optimization.

Application of Ultra Thin Wall (UTW) ceramic substrates in the catalytic converter system requires the canner and component manufacturers to better understand the root cause and physics behind substrate breakage during the canning process. For this purpose, a ceramic substrate strength study for shoebox design has been conducted within Visteon Corporation. Computer Numerical Control (CNC) machined top and bottom fixtures, with identical inner surfaces as shoebox converter upper and lower shells, were used to crush mat wrapped substrates. Thin film pressure sensor technology enables the recording of substrate surface pressure during the compression process. Shell rib, washcoat, canning speed and cell density effects on substrate failure have been experimentally investigated. The development of a mathematical model helps to identify a better indicator to evaluate the substrate strength in the canning process and establish the strength for uncoated & coated substrates.

A new revolutionary fiber (SECURUS™) that is offered by Honeywell to the automotive industry suitable for use in safety belts is able to absorb a passenger's kinetic energy without the passenger being subjected to excessive forces. This paper analyses thoroughly the mechanisms of the energy dissipation during an automotive accident. Such analysis shows the role of the belt and its characteristics in the process of energy dissipation. We have shown that there are two interacting mechanisms: one by the vehicle (energy absorbing elements) and another by the belt itself. We built simple and useful mathematical models of the process and showed quantitatively (using the belts of real characteristics and dimensions) how the remarkable characteristics of SECURUS™ fiber allow the force reduction. In this paper the models are simple enough to keep the major effects transparent to the observer.

Generalized predictive control (GPC) is a discrete time control strategy proposed by Clark et al [1]. The controller tries to predict the future output of a system or plant and then takes control action at present time based on future output error. Such a predictive control algorithm is presented in this paper for yaw stability management of an automobile. Most of the existing literature on the yaw stability management systems lacks the insight into the yaw rate error growth when the automobile is in a understeer or oversteer condition on a low friction coefficient surface in a handling maneuver. Simulation results show that the predictive feature of the proposed controller provides an effective way to control the yaw stability of a vehicle.

In an effort to understand steering systems performance and properties at the microscopic level, we developed Multibody simulations that include multiple three-dimensional gear surfaces that are in a dynamic state of contact and separation. These validated simulations capture the dynamics of high-speed impact of gears traveling small distances of 50 microns in less than 10 milliseconds. We exploited newly developed analytic, numeric, and computer tools to gain insight into steering gear forces, specifically, the mechanism behind the inception of mechanical knock in steering gear. The results provided a three dimensional geometric view of the sequence of events, in terms of gear surfaces in motion, their sudden contact, and subsequent force generation that lead to steering gear mechanical knock. First we briefly present results that show the sequence of events that lead to knock.