Neural Network-based road friction estimation using road weather information
| dc.contributor.author | Minges, Florian | |
| dc.contributor.department | Chalmers tekniska högskola / Institutionen för mekanik och maritima vetenskaper | sv |
| dc.contributor.examiner | Selpi, Selpi | |
| dc.contributor.supervisor | Selpi, Selpi | |
| dc.contributor.supervisor | Muppirisetty, Srikar | |
| dc.contributor.supervisor | Lindbohm, Erik | |
| dc.date.accessioned | 2020-04-15T17:31:36Z | |
| dc.date.available | 2020-04-15T17:31:36Z | |
| dc.date.issued | 2020 | sv |
| dc.date.submitted | 2019 | |
| dc.description.abstract | The tire/road friction is an important factor for the overall vehicle performance and stability, and is thus an area of interest for both vehicle and tire manufacturers. Furthermore, accurate road friction estimation (RFE) algorithms could become even more important in the future when autonomous driving systems become more common. The most common limitation of the current RFE models is that they often require very specific testing conditions and tend to struggle when these are not met, which is an issue when trying to generalise the performance for the every-day vehicle usage. One of the most important factors in tire/road friction is the road surface conditions, which in turn is heavily in influenced by the current and past weather. This thesis studies what added value knowledge of the road weather can have on a road friction estimation algorithm, using a simple neural network. We do this by setting up an experiment where we train two models; one that is trained only with data from the in-vehicle sensors, and one that is trained with in-vehicle sensor data and weather data combined. Ultimately, the added weather data is able to improve the best performance of our model by 3.76 percentage points from 45.27% to 49.03% (8.3% improvement in terms of the number of correctly classified samples) when modelled as a 6-class classification problem. When modelled as a binary classification task ([0 - 0.5) and [0.5+]) our model's performance improves by 21.5 percentage points from 54.47% to 75.97% (39.5% improvement in terms of number of correctly classified samples). Finally, we conclude that the added road weather data has a positive in influence in distinguishing high from low friction values, while struggling with distinguishing the low friction nuances. | sv |
| dc.identifier.coursecode | MMSX30 | sv |
| dc.identifier.uri | https://hdl.handle.net/20.500.12380/300762 | |
| dc.language.iso | eng | sv |
| dc.relation.ispartofseries | 2020:05 | sv |
| dc.setspec.uppsok | Technology | |
| dc.subject | Road Friction Estimation | sv |
| dc.subject | RFE | sv |
| dc.subject | Machine Learning | sv |
| dc.subject | Sensor Fusion | sv |
| dc.subject | Road Weather | sv |
| dc.title | Neural Network-based road friction estimation using road weather information | sv |
| dc.type.degree | Examensarbete för masterexamen | sv |
| dc.type.uppsok | H |