Riassunto analitico
Nowadays, one of the major challenges for vehicle manufacturers is the reduction of pollutant and greenhouse gas emissions, as these two factors accelerate climate change and deterioration of the air quality, affecting both the environment and human health. The role of automotive engineers is to reduce the contribution of the sector to these events. European and international regulations have become progressively stringent over the years, imposing rigorous limits on emissions of chemicals such as CO2, NOx, HC, CO. Future vehicles must be more efficient and less polluting, complying with regulations but remaining competitive in the tough automotive market.
This thesis activity, carried out at AVL in Italy s.r.l., in the gasoline engine calibration department, explores innovative solutions in the calibration and testing field, extensively using data analysis and AI, to improve the efficiency of the team’s activities.
The main part of the work concerns the realization of a Python tool that performs statistical analysis on data obtained from both road tests and chassis-dyno tests. The program manages big datasets, allowing the user to achieve maximal insight into the data, focus on specific areas of the ECU software, discover anomalies and facilitate the comparison of different calibrations. Once the improvements in the efficiency in terms of performance and emissions have been stated, a secondary part of the activity using a different approach has been developed.
The trending and extensive use of AI in data-driven insights and efficiency improvement combined with the great availability of data, has led to its usage for the second approach. This second tool exploits Machine Learning techniques, as clustering and anomaly detection, to overcome a limit of the first tool: the comparison and validation of tests with different boundary conditions. It can also detect possible sensor failures and individualise the working range of the sensors.
The first tool, together with the second, has been used on different calibrations and the case study of a defective catalyst, to give a better understanding of the benefits of working on real cases.
Since testing and calibration are expensive, it is essential to minimize these activities, that can be achieved either through an optimized data analysis, but seeking to develop tools that use AI could give an additional support to the calibration phase and improve the productivity and efficiency.
|
