ABSTRACT
ABSTRACT Unfortunately, high density human crowds or pedestrian flow has previously led to tragedies and or fatalities. In addition, crowd behavior as a reaction to an incident aggravates the complexity and disruption of human flow. Results of such situations include trampling and crushing. Therefore, it is important to monitor such crowd motion for danger warning and prevention. In this study, we are proposing a novel approach to provide a continuous crowd load prediction on pedestrian bridges, with particular focus on high density of crowds. The approach is based on continuous monitoring methods known as structural health monitoring (SHM). An important novelty in the approach is the choice of Fiber Bragg Gratings (FBG) Fiber Optic Sensors (FOS) over traditional monitoring instrumentations due to its numerous advantages. The approach also employs machine learning techniques to generate prediction models from training data gathered from FOS. The machine learning techniques applied were Support Vector Machines (SVMs) and exponential Gaussian Process (GP) regression applied on vibration data collected from FOS sensors mounted on a pedestrian bridge. The approach was validated using laboratory experiments on a real scaled bridge model. Different loading scenarios were investigated to predict the weight of individuals, groups as well as continuous flow of people activities. Current results demonstrated promising results capable of predicting the total load weight with an mean error of 10 kg and the activity type as fast or slow with an accuracy above 90%.
