Search Results

Artificial Intelligence is becoming very important and useful in several scientific fields. Machine learning methods, such as neural networks and decision trees, are often proposed in applications for internal combustion engines as virtual sensors, faults diagnosis systems and engine performance optimization. The high pressure of the intake air coupled with the demand of lean conditions, in order to reduce emissions, have often close relationship with the knock events. Fuels autoignition characteristics and flame front speed have a significant impact on knock phenomenon, producing high internal cylinder pressures and engine faults. The limitations in using pressure sensors in the racing field and the challenge to reduce the costs of commercial cars, push the replacement of a hardware redundancy with a software redundancy.

The present work is the continuation of the research activity developed by the same authors in last years about the use of recent technologies (Artificial Neural Networks) for the set up of “software redundancy” modules to be implemented On Board for the use in Diagnostic Systems. In the present work, a system based on Artificial Neural Networks models for automotive engines Fault Diagnosis and Isolation purposes is set-up and analysed. Four sensors/actuators (throttle valve, rotational speed, torque and intake manifold pressure) are considered, and the respective acquired data are used to train and test four ANN modules correlating the different quantities. An FDI scheme is presented which generates fault codes sequences by suitably treating the primary residuals, obtained by comparing experimental data with the calculated ones by the ANN modules. The robust fault isolation capabilities of the proposed FDI system are presented and discussed.

In the last decades, worldwide automotive regulations induced the industry to dramatically increase the application of electronics in the control of the engine and of the pollutant emissions reduction systems. Besides the need of engine control, suitable fault diagnosis tools had also to be developed, in order to fulfil OBD-II and E-OBD requirements. At present, one of the problems in the development of Diesel engines is represented by the achievement of an ever more sharp control on the systems used for the pollutant emission reduction. In particular, as far as NOx gas is concerned, EGR systems are mature and widely used, but an ever higher efficiency in terms of emissions abatement, requires to determine as better as possible the actual oxygen content in the charge at the engine intake manifold, also in dynamic conditions, i.e. in transient engine operation.

In recent years there has been an increasing worldwide effort to limit polluting emissions from road vehicles. The On Board II Diagnostic (OBD II) regulations adopted by California Air Resources Board (CARB) are among the most restrictive rules. They require on-board devices which monitor emission control systems in order to identify deterioration or malfunction of components. For automotive purpose, the high cost of achieving hardware redundancy can be reduced by substituting software redundancy. This approach requires an engine model definition. In this work the application of the Artificial Neural Networks (ANNs) technology, is analyzed and validated by experiments. First model has been tested under varying load conditions with very encouraging results.

In this paper, a further step along a research line concerning the set up of a Fault Diagnosis system for OBD-II purpose is presented. The suitability of Artificial Neural Networks for the use as engine simulation modules in the framework of a software redundancy approach has been analyzed. Experimental tests were performed, by acquiring four main engine operational parameters. Using this knowledge base, the performance of a wide variety of different Net Types was analyzed and discussed. Peculiar aspects of the possible industrial applications of this methodology are also deeply examined.

In this paper the evaluation of the suitability of the Artificial Neural Networks for setting up simulation modules for “analytical redundancy” was further carried out. The performance of the ANN modules was enhanced, by taking into account the engine dynamics for the simulation of fast engine transients and obtaining satisfactory results. Working toward actual on board application in Fault Diagnosis systems, some ANN modules were implemented in an on-line system which acquires signals from an engine mounted on a test bench and compares in real time the experimental values with the estimated ones. In this way, it was possible to perform long duration tests of ANN's behaviour, substantially confirming the results of the conventional off-line analysis.