Simulating the behavior of trade crews in construction using agents and building information modeling Lola Ben-Alon, Rafael Sacks ⁎ Virtual Construction Lab, Technion IIT, Haifa, Israel abstract article info Article history: Received 30 September 2015 Received in revised form 22 October 2016 Accepted 2 November 2016 Available online 16 November 2016 Simulation is particularly useful for testing different production control and information flow methods in con- struction, because field experiments suffer from difficulties with isolating cause and effect. Existing methods such as Discrete Event Simulation are limited in their ability to model the behavior of crews and of individuals who make decisions subject to their perceptions of uncertain conditions. Agent-Based Simulation may offer a bet- ter solution because agents can be applied with behavioral models. The aim of this work was to build an exper- imental tool capable of reflecting the emergent nature of production in construction. This required capturing trade crew behaviors through interviews and encapsulating the behavior in software agents. The system models trades' decision-making and situational awareness while using a Building Information Model to define the phys- ical and the process environment for the simulation. The resulting simulation tool was validated by testing pre- dictable scenarios, which resulted in similar patterns to those found in an actual construction site. It was then applied to explore the emergent outcomes of more complex scenarios. © 2016 Elsevier B.V. All rights reserved. Keywords: Agent-Based Simulation Building information modeling Discrete Event Simulation Emergent construction 1. Introduction Research of production control systems in construction is limited by the capabilities of the available research methods. Among research methods used to date are work studies [1,2], action research [3–6] and simulation [7–9]. Both work-studies and action research are performed ‘in situ’ and thus can only study one control system in one project at a time. They cannot be used to compare or to evaluate the different out- comes that would be obtained if changes were made to the control par- adigm or its parameters on a given project; projects cannot be repeated. Given the inherent variability and uncertainty of parameters that influ- ence the outcomes of construction projects – such as material, labor, equipment and information flows [10, p.3, 11] – these methods also suf- fer from significant drawbacks in terms of isolating cause and effect. It is very difficult to differentiate the effects of any given experimental inter- vention from the influences of parameters that the researchers cannot control, such as design changes, material shortages, weather effects, un- stable subcontractor resource allocations, etc. The Hawthorne effect [12] and the learning curve effect add to the problems of measuring the im- pact of interventions on site. For these and other reasons (such as the limitations of research bud- gets), computer simulations have become the method of choice for comparative research of production systems in construction. Discrete Event Simulation (DES) applications, implemented in languages such as STROBOSCOPE and CYCLONE, have provided general and special pur- pose frameworks for simulating construction operations and construc- tion management processes [13,14]. Examples abound: Tommelein et al. [9] used DES to illustrate the effect of variable production rates on productivity and cycle times in the ‘Parade of Trades’ simulation; Brodetskaia et al. [8] used DES to test the impact of production control policies on throughput (TH),on quantities of work in progress (WIP) and cycle time (CT) in high-rise apartment construction; and Bashford et al. [15] demonstrated the relationship between system loading and cycle times for the case of custom house building. However, due to the nature of DES, these simulations did not model the decision-making behavior of the trade crews nor the effect of move- ment within a geometrically realistic working environment. Their use has been limited to predetermined events of specific construction pro- cesses and general purpose frameworks for developing simulations of construction operations [14,16]. Such research typically uses a “top- down” approach to modeling and understanding the impacts of produc- tion control on labor productivity. In a top-down approach, the se- quence of events is governed by the availability of crew, materials, information and other preconditions at each time step as events are evaluated, but the subjective behavior of trade crews and their human leaders who function within a certain perception of the construction project reality, is not modeled and does not affect the outcomes [17]. Like many economic systems, building construction projects can be considered to be emergent production processes whose outcomes are the results of the actions of the individual economic agents who Automation in Construction 74 (2017) 12–27 ⁎ Corresponding author. E-mail addresses: lola.ben.alon@gmail.com (L. Ben-Alon), cvsacks@technion.ac.il (R. Sacks). http://dx.doi.org/10.1016/j.autcon.2016.11.002 0926-5805/© 2016 Elsevier B.V. All rights reserved. Contents lists available at ScienceDirect Automation in Construction journal homepage: www.elsevier.com/locate/autcon