Speed independent road classification strategy based on vehicle response: Theory and experimental validation Yechen Qin a , Zhenfeng Wang a , Changle Xiang a , Ehsan Hashemi b , Amir Khajepour b , Yanjun Huang b,⇑ a School of Mechanical Engineering, Beijing Institute of Technology, Beijing, People’s Republic of China b Mechanical and Mechatronics Engineering Department, University of Waterloo, Waterloo, ON, Canada article info Article history: Received 5 January 2018 Received in revised form 9 June 2018 Accepted 21 July 2018 Keywords: Road classification Frequency domain classifier Vehicle system responses Road estimation abstract This paper presents a speed-independent road classification strategy (SIRCS) based on sole measurement of unsprung mass acceleration. The new method provides an easy yet accu- rate classification methodology. To this purpose, a classification framework with two phases named off-line and online is proposed. In the off-line phase, the transfer function from unsprung mass acceleration to the road excitation is firstly formulated, and a random forest-based frequency domain classifier is then generated according to the standard road definition of ISO 8608. In the online phase, unsprung mass acceleration and vehicle velocity are firstly combined to calculate the equivalent road profile in the spatial domain, and then a two-step road classifier attributes the road excitation to a certain level based on the power spectral density (PSD) of the equivalent road profile. Simulations are carried out for different classification intervals, varying velocity, system uncertainties and measure- ment noises. Road experiments are finally performed in a production vehicle to validate the proposed SIRCS. Measurement of only unsprung mass acceleration to identify road classification and less rely on the training data are the major contributions of the proposed strategy. Ó 2018 Elsevier Ltd. All rights reserved. 1. Introduction Road condition identification attracts much attention of vehicle manufacturers and government because of vehicle safety and comfort [1,2]. In America, about a quarter of major urban roads are in poor conditions, resulting in extra vehicle main- tenance cost of nearly four hundred dollars per driver per year [3]. A report in 2013 revealed that annual road maintenance expenditure was about 20,000 million Euros in EU [4], and roads with poor conditions decrease passengers’ feeling and increase travel times [5]. Better road condition information can not only help drivers and advanced vehicle control systems [6–9], but also assist in a better road maintenance scheduling. Generally speaking, reported road estimation algorithms could be divided into three distinct types, i.e. contact measure- ment, non-contact measurement and system response based estimation [10]. The first type requires specially designed pro- filometer, which restricts its commercial application [11]. For the second method, Mono/Stereo cameras and LiDAR are applicable thanks to the vigorous development of autonomous vehicles. Although accurate estimation can be obtained with the non-contact measurement method, high cost of the instruments impedes its application in middle- to low-end vehicles. https://doi.org/10.1016/j.ymssp.2018.07.035 0888-3270/Ó 2018 Elsevier Ltd. All rights reserved. ⇑ Corresponding author. E-mail address: huangyanjun404@gmail.com (Y. Huang). Mechanical Systems and Signal Processing 117 (2019) 653–666 Contents lists available at ScienceDirect Mechanical Systems and Signal Processing journal homepage: www.elsevier.com/locate/ymssp