Data-Driven Estimation of Coastdown Road Load

Emissions and fuel economy certification testing for vehicles is carried out on a chassis dynamometer using standard test procedures. The vehicle coastdown method (SAE J2263) used to experimentally measure the road load of a vehicle for certification testing is a time-consuming procedure considering the high number of distinct variants of a vehicle family produced by an automaker today. Moreover, test-to-test repeatability is compromised by environmental conditions: wind, pressure, temperature, track surface condition, etc., while vehicle shape, driveline type, transmission type, etc. are some factors that lead to vehicle-to-vehicle variation. Controlled lab tests are employed to determine individual road load components: tire rolling resistance (SAE J2452), aerodynamic drag (wind tunnels), and driveline parasitic loss (dynamometer in a driveline friction measurement lab). These individual components are added to obtain a road load model to be applied on a chassis dynamometer. However, lab-tested quantities may not account for environmental noise factors and qualitative vehicle characteristics leading to a significant residual road load between the track-tested and lab-tested road loads. Regression modeling techniques are explored for estimating this residual road load and the challenges are discussed. Additionally, a technique is developed to choose feature selection metrics using simulation of multivariate non-gaussian continuous and discrete data having similar statistical properties as the data obtained from automotive road tests. Using the selected features, two regularized regression techniques are experimented with. The first technique models the residual road load power, while the second technique models a polynomial relationship between vehicle speed and residual road load.