Road classification using built-in self-scaling method of Bayesian regression

Abstract This paper proposes the use of the built-in self-scaling (BS) method for ISO classification of road roughness. The technique employs the transfer function between the vehicle body acceleration as input and the suspension travel as output. This transfer function has a nonzero dc gain, which is important for application of the BS method. Frequency response magnitude patterns corresponding to this transfer function are estimated via Bayesian regression, capitalizing on the inherent properties of the BS method where the prior dc gain is incorporated into the formulation. This strategy leads to high classification accuracy. The proposed approach requires only low-cost sensors. It possesses a short detection time of 0.5s and a short training time of 5s for each road class. The method is model-free and does not require recalibration when the load carried by the vehicle changes. Additionally, it is capable of handling varying vehicle velocity and is effective for both passive and active suspensions. A laboratory-scale experiment shows that the proposed technique increases the percentage of correct classification by an average of 34% in the case of constant road profiles, compared with a state-of-the-art method using augmented Kalman filtering. A corresponding value of 24% is achieved for a varying road profile. The significant improvement in the accuracy of road classification is impactful as it will enable controller design for suspension systems to be enhanced resulting in more comfortable ride and higher vehicle stability.