Search Results

This study aims at building efficient and robust artificial neural networks (ANN) able to reconstruct the in-cylinder pressure of Diesel engines and to identify engine conditions starting from the signal of a low-cost accelerometer placed on the engine block. The accelerometer is a perfect non-intrusive replacement for expensive probes and is prospectively suitable for production vehicles. In this view, the artificial neural network is meant to be efficient in terms of response time, i.e. fast enough for on-line use. In addition, robustness is sought in order to provide flexibility in terms of operation parameters. Here we consider a feed-forward neural network based on radial basis functions (RBF) for signal reconstruction, and a feed-forward multi-layer perceptron network with tan-sigmoid transfer function for signal classification. The networks are trained using measurements from a three-cylinder real engine for various operating conditions.

This study aims at building an efficient and robust radial basis function (RBF) artificial neural network (ANN), to reconstruct the in-cylinder pressure of a diesel engine starting from the signal of a low-cost accelerometer placed on the engine block. The accelerometer is a perfect non-intrusive replacement for expensive probes and is prospectively suitable for production vehicles. The RBF network is trained using measurements from different engine operating conditions. Training data are composed of time series from the accelerometer and corresponding measured in-cylinder pressure signals. The RBF network is then validated using data not included in training and the results show good correspondence between measured and reconstructed pressure signal. Various network parameters are used to optimize the network quality.

In order to meet the increasingly strict emission regulations, several solutions for NOx and PM emissions reduction have been studied. Exhaust gas recirculation (EGR) technology has become one of the more used methods to accomplish the NOx emissions reduction. However, actual control strategies do not consider, in the definition of optimal EGR, its effect on particle size and density. These latter have a great importance both for the optimal functioning of after-treatment systems, but also for the adverse effects that small particles have on human health. Epidemiological studies, in fact, highlighted that the toxicity of particulate particles increases as the particle size decreases. The aim of this paper is to present a Neural Network model able to provide real time information about the characteristics of exhaust particles emitted by a Diesel engine.

In recent years, great interest on NVH characteristics of vehicles has been paid by all the big automotive manufacturers. Interior acoustic comfort is now one of the main key factors in vehicle development process, since it contributes to improved product overall quality. Therefore, in automotive industry advanced NVH refinement needs to work in synergy with all research activities. Assessing the level of experienced noise in interior cabin requires particular arrangements for ensuring adequate measurement accuracy (AC system off, closed window, etc.). The use of parameters such as the level of seat vibration, not affected by the acoustic field conditions inside the vehicle, could facilitate experiments in parallel with engine/vehicle calibration activities.

Advanced numerical techniques, such as fuzzy logic and neural networks have been applied in this work to digital images acquired on a mono-component fuel spray (iso-octane), in order to define, in a stochastic way, the gas-liquid interface evolution. The image is a numerical matrix and so it is possible to characterize geometrical parameters and the time evolution of the jet by using deterministic, statistical stochastic and other several kinds of approach. The algorithm used works with the fuzzy logic concept to binarize the shades gray of the pixel, depending them, by using the schlieren technique, on the gas density. Starting from a primary fixed threshold, the applied technique, can select the ‘gas’ pixel from the ‘liquid’ pixel and so it is possible define the first most probably boundary lines of the spray.

The paper reports an innovative method of analysis based on an advanced statistical techniques applied to images captured by a high-speed camera that allows highlighting phenomena and anomalies hardly detectable by conventional optical diagnostic techniques. The images, previously elaborated by neural network tools in order for clearly identifying the contours, have been analyzed in their time evolution as pseudo-chaotic variables that may have internal periodic components. In addition to the Fourier analysis, tools as Lyapunov and Hurst exponents and average Kω permitted to detect the chaos level of the signals. The use of this technique has permitted to distinguish periodic oscillations from chaotic variations and to detect those parameters that actually determine the spray behavior.

The analysis of a spray behavior is confined to study the fluid dynamic parameters such as axial and radial velocity of the droplets, size distribution of the droplets, and geometrical aspect as the penetration length. In this paper, the spray is considered like a dynamic system and consequently it can be described by a number of parameters that characterize its dynamic behavior. The parameter chosen to describe the dynamic behavior is the external cone angle. This parameter has been detected by using an experimental injection chamber, a multi-hole (8 holes) injector for GDI applications and recorded by a high-speed C-Mos camera. The images have been elaborated by a fuzzy logic and neural network algorithm and are processed by using a chaos deterministic theory. This procedure carries out a map distribution of the working point of the spray and determines the stable (signature of the spray) and instable behavior.