Audio-Based Bayesian Model for Productivity Estimation of Cyclic Construction Activities

Construction heavy machines often perform their routine tasks in the form of cyclic operations (e.g., cycles of digging, swinging, and dumping for a hydraulic excavator), and forecasting those cycle times is an important step toward scheduling, cost estimation, and productivity analysis of construction projects. The current state of research for automated cycle time prediction for construction operations is based on processing kinematic data (e.g., acceleration) or implementing computer vision algorithms. Both methods have certain limitations (e.g., it is necessary to directly attach kinematic sensors to machines, and computer vision algorithms are very sensitive to lighting conditions and occlusions). In addition, current methods predict cycle time values once (usually at the beginning of projects) and do not include an auto-update component capable of adjusting results over time and due to variations, such as changes in jobsite conditions or impacts of learning curve, for example. To address these two important issues, the authors propose an audio-based Bayesian system for estimating cycle times of cyclic construction activities. The sounds generated during routine operations of construction equipment and machines are treated as the main source of data for this project, and efficient signal processing and machine learning algorithms are implemented to process recorded audio files at construction jobsites and extract useful information. A robust denoising algorithm was developed to improve the quality of audio files, and Bayesian statistical models are utilized to include historical data for cycle time estimation enhancement. Case studies illustrate that implementing robust audio signal processing techniques, along with a Markov chain-based filter, enable the system to accurately forecast cycle times of construction activities for multiple days of operation.