Opportunistic method for road surface noise labelling : data cleaning

Road surface type and degradation contribute significantly to the rolling noise emission. In recent times, due to the innovation in vehicle propulsion, rolling noise also becomes a main factor in noise emission for lower order roads. Monitoring and labelling these roads, requires considerably more effort than monitoring primary roads and highways due to their large number. Therefore, we propose an opportunistic method where vehicles that are on the roads for other purposes, are used for rolling noise monitoring. The proposed method may also have some additional benefits over the standard CPX regarding the distribution of tires used and the spread of typical driving speeds. However, measurement conditions are not as well known and may influence the results obtained from individual vehicles significantly. The abundance of measurements data from many vehicles will nevertheless allow to eliminate any modifiers and confounders. To that end, a machine learning cleaning algorithm inspired by denoising auto-encoders has been designed and implemented. This cleaning algorithm improves the convergence of measurements, giving the same quality of measurements with a lower number of passages and cars on a road segment.