Search Results

Testing real-time vehicular systems challenges the tester to design test cases for concurrent and sequential input events, emulating unexpected user and usage profiles. The vehicle response should be robust to unexpected user actions. Sequence Covering Arrays (SCA) offer an approach which can emulate such unexpected user actions by generating an optimized set of test vectors which cover all possible t-way sequences of events. The objective of this research was to find an efficient nonfunctional sequence testing (NFST) strategy for testing the robustness of real-time automotive embedded systems measured by their ability to recover (prove-out test) after applying sequences of user and usage patterns generated by combinatorial test algorithms, considered as “noisy” inputs. The method was validated with a case study of an automotive embedded system tested at Hardware-In-the-Loop (HIL) level. The random sequences were able to alter the system functionality observed at the prove-out test.

A full vehicle NVH Simulator has been developed to provide a realistic interactive in-car environment where a subject can experience multi-modal stimuli ( accurately reproduced sound and vibration as well as visual ) whilst either driving or being driven. This paper describes its use in learning about the strategies subjects employ during sound evaluations, and how this information can help optimise decision making during product development. It is possible to understand how subjects assess the sound of vehicles, both in the way that they drive the vehicle and importantly which elements of the sound character have greatest influence on their evaluation of the vehicle. It is also possible to compare the strategies employed by NVH engineers, company decision-makers and non-experts such as customers.

Hybrid vehicles provide the most viable medium term solution to meet the demands of the automotive marketplace. Currently electric hybrids dominate the marketplace but mechanical hybrid systems, including flywheel, pneumatic and hydraulic systems all have the potential to compete with electrical systems. This paper provides an overview of a mechanical hybrid project created by the University of Warwick. The aim of the project is to assess alternative hybrid powertrains, in particular pneumatic, hydraulic and flywheel systems. This paper provides a description of a simulation tool to investigate and evaluate the mechanical alternatives to electric hybrid vehicles and the results from two fuel economy case studies using the simulation tool: 2.6 ton SUV and 17 ton bus applications. The paper also outlines a feasibility study of the mechanical hybrid options, including a cost benefit analysis of the different systems.

An automotive side door latch release mechanism has been modelled for the locking and unlocking vehicle functionality in Dymola. The performance of the developed door lock model is evaluated against an existing model of a similar door locking/unlocking system in Stateflow. The model performance is also compared with measurements from a real vehicle door latch. The model is converted into a Simulink model and built for a real-time environment such as the dSPACE target with a fixed step size solver. It is shown that a step size as small as 1 ms can be used for real-time simulation without task overrunning in the real-time target. The model is also benchmarked on a multiprocessor setup as multiprocessor simulators are common in system-level networked Electronic Controller Unit (ECU) testing facilities for implementing high fidelity closed loop models of integrated ECUs and actuators.

The popularity of CAN (Controller Area Network) in the production vehicles is well established. As a result, CAN has been developed for use in many non-automotive applications. This gave rise to the development of an open higher layer CAN protocol known as DeviceNet. With the popularity of DeviceNet for Automation Systems, this technology has drastically decreased in cost. Although DeviceNet is quite complex to develop, it easier to implement than SAE J1939 due to the large number of commercial off-the-shelf product that is available. Also, there are many configuration and diagnostic tools available by the same means. There are more than 300 vendors of DeviceNet product. Researchers at the University of Warwick have built a vehicle demonstrator using CAN/DeviceNet modules. This paper will illustrate the ease of vehicle system integration utilising this popular technology.

Sound quality targets for new vehicles are currently specified by jury evaluation techniques based upon listening studies in a sound laboratory. However, jury testing is costly, time consuming and at present there are no methods to include customer expectations or brand requirements. This paper describes a neural computing approach that is being developed to generate knowledge and tools to enable objective measures of a product's sound to be converted into a prediction of the subjective impression of potential customers without carrying out the traditional jury evaluation tests.

An on line method for verification of chassis dynamometer operation uses a neural network. During the testing of a vehicle, it is assumed that after a warm up period the parasitic losses remain stable. There is normally no provision for verification of correct dynamometer operation while the test is running. This technique will detect if a component wears or fails during the testing of a vehicle and thus avoid testing under erroneous conditions. A Learning Vector Quantization (LVQ) neural network is trained to recognise poor dynamometer operation in order to signal a fault condition to the operator.

All new European vehicles face strict drive-by noise regulations. It would help vehicle designers if they could predict drive-by noise given parameters available early in the design process. The large amount of data from previous tests suggests a new approach, using neural networks. This paper introduces neural networks and describes how to apply them to the prediction problem. The selection of suitable inputs and amount of data required is discussed. The problem can be simplified by first predicting vehicle performance. Interim results for a vehicle performance neural network are presented. Further work towards a drive-by noise neural network is proposed.