Teaching machine learning by implementing automatic detection of pavement in bicycles rides using smartphone accelerometer data

Machine learning is becoming a very important topic in modern society and it is also mandatory that engineering undergraduate students have some background on the field. It can be applied in mechanical engineering in various forms such as machinery design, fluid mechanics analysis and vibrations monitoring. During bicycle riding, the cyclist is subject to mechanical vibrations due to the pavement irregularities, and a prolonged or frequent exposition to vibrations might become harmful and bring problems to hand and arms. In this work we propose a vibration measurement procedure using a bicycle and a smartphone accelerometer with data acquisition performed by free applications. The time data is processed externally, and four machine learning algorithms are trained with real data obtained from different road pavement conditions: asphalt pavement, dirt road and stone pavement. The mechanical vibrations of two positions in bicycle structure (handlebar and frame) were also investigated. To process the acquired data, eight different statistical features were considered. In general, the Gradient Boosting Machine classifier presented the best performance according to its scores on each testing condition, with 84.99% and 96.38% accuracies in classifying vibration signals with 3 and 2 classes, respectively. The proposed teaching methodology was approved by different groups of students. For most of them, it was a first contact with the field, and helped stimulate interest in the subject and for future applications.