Coupling human activity recognition and wearable sensors for data-driven construction simulation

Construction discrete event simulation (DES) models consist of a complex system of interrelated activities. An essential step in DES design is the process of input modeling which entails the estimation of the attributes (e.g. duration, precedence logic) of simulated activities. The quality of the simulation output is directly proportional to the quality of the input modeling. Traditional simulation models are commonly built upon engineering assumptions and subject matter expert opinions of simulation parameters. In this paper, a machine learning-based framework is designed and implemented to extract durations of activities performed by construction workers from wearable sensors. This framework uses accelerometer and gyroscope sensors embedded in smartphones that are worn by field workers. Data analysis and processing is applied to the collected data to train machine learning algorithms capable of detecting and classifying construction workers’ activities. Once the activities are identified, their durations are calculated using the time stamps of the collected data. Results indicate that smartphones can be used as cost-effective, ubiquitous, and computationally powerful means of enabling data-driven DES models with enhanced reliability over traditional simulation models.