An investigation of the suitability of smartphone devices for road condition assessment

The measurement of road roughness is important for asset management decision making. Not only is road roughness an indicator of road condition and thereby a means of determining road maintenance needs, but it is also used to determine vehicle operating costs (i.e. fuel consumption and vehicle maintenance). Road agencies with large road networks, because of resource issues, are unable to record the condition of the entire network on a sufficiently frequent basis to adequately determine road condition and therefore identify proactive road maintenance requirements. This research investigates whether a smartphone based system may be a suitable means for measuring road roughness at sufficient accuracy and if data from such a system could be used to inform asset management decision making and provide the road user with information about vehicle operating cost of using different routes. This research by means of an in depth review of the literature and the use of a vehicle dynamics package, identified the factors which can most influence the accurate measurement of road roughness by smartphone based systems and quantified the relative importance of these factors. The investigation found that measured vehicle body acceleration, speed, vehicle type and smartphone type are very influential inaccurately determining road roughness from a smartphone type approach. Thereafter, a variety of computational methods were trialled on a multi-variable dataset that had been built using a vehicle dynamic package, to determine if the algorithms could be used to infer road roughness from a dataset which might be available from a smartphone based system. As a result of this analysis, the random forest machine learning algorithm was identified as the most suitable for the task at hand. It was found that the developed algorithm could be used to determine precise measures of road roughness if data concerning vehicle speed and type, sprung mass, smartphone type and vehicle body acceleration were available. The same algorithm could also be used to classify road condition if only vehicle speed, vehicle type and measured vehicle vertical body acceleration were available in the dataset.