Evaluation of machine learning algorithms for detection of road induced shocks buried in vehicle vibration signals. (March 2019)

• Record Type: Journal Article
• Authors: Lepine, Julien; Rouillard, Vincent; Sek, Michael
• Is Part Of: Proceedings of the Institution of Mechanical Engineers. Volume 233:Number 4(2019:Apr.)
• Page Start: 935
• Page End: 947
• Publication Date: 2019-03
• Subjects: Heavy vehicle systems -- in-vehicle data recorders -- machine learning -- protective packaging -- shocks detection -- vehicle noise/vibration; Mechanical engineering -- Congresses; Transportation engineering -- Congresses; 629.2
• DOI: 10.1177/0954407018756201
• Languages: English
• ISSNs: 0954-4070

Abstract

Road surface imperfections and aberrations generate shocks causing vehicles to sustain structural fatigue and functional defects, driver and passenger discomfort, injuries, and damage to freight. The harmful effect of shocks can be mitigated at different levels, for example, by improving road surfaces, vehicle suspension and protective packaging of freight. The efficiency of these methods partly depends on the identification and characterisation of the shocks. An assessment of four machine learning algorithms (Classifiers) that can be used to identify shocks produced on different roads and test tracks is presented in this paper. The algorithms were trained using synthetic signals. These were created from a model made from acceleration measurements on a test vehicle. The trained Classifiers were assessed on different measurement signals made on the same vehicle. The results show that the Support Vector Machine detection algorithm used in conjunction with a Gaussian Kernel Transform can accurately detect shocks generated on the test track with an area under the curve (AUC) of 0.89 and a Pseudo Energy Ratio Fall-Out (PERFO) of 8%.