AGENT-BASED MODELING AND SIMULATION IN CONSTRUCTION

Agent-based Modeling and Simulation (ABMS) is a relatively new development that has found extensive use in areas such as social sciences, economics, biology, ecology etc. Can ABMS be effectively used in finding answers to complex construction systems? The focus of this paper is to provide some answers to this question. Initial experimentation is conducted to understand the advantages of using ABMS either in isolation or in combination with traditional simulation methodologies. The paper provides a summary of this experimentation, conclusions and sets the agenda for future research in this area. 1 CONSTRUCTION SIMULATION Construction researchers and practitioners have used various techniques for studying complex construction systems. These techniques include the basic networking techniques like CPM and PERT, queuing models, productivity models like method productivity delay model (Adrian 1976), operations research tools like linear programming, game theory, simulation, and industrial planning techniques like line of balance method. The advent of simulation methods in construction, occurred in the form of introduction of simple networking concepts, as a modeling framework for studying construction operations. The earliest of these methods was the so-called "link node" model adapted by Teicholz (1963). After that Halpin (1973) developed the CYCLONE format at the University of Illinois. CYCLONE has become the basis for a number of construction simulation systems. CYCLONE simplified the simulation modeling process and made it accessible to construction practitioners with limited simulation background. The application of the construction process simulation ranges from productivity measurement and risk analysis to resource allocation and site planning. A microcomputer version of CYCLONE was developed by Luch and Halpin (1981) at Georgia Tech. This version is called MicroCYCLONE. Paulson (1987) developed the INSIGHT system which is based on the CYCLONE methodology and has a more interactive interface. Touran (1981) focused on automated real time data acquisition and its integration with INSIGHT. Work at the University of Michigan under Carr led to the development of RESQUE (Chang and Carr, 1987) which is also CYCLONE based with advanced resource handling capabilities. Ioannou (1989) developed UM-CYCLONE for advanced construction process modeling. More recently advanced construction simulation initiatives have been launched. Simphony developed at the University of Alberta provides an advanced simulation environment specially tailored for construction researchers and practitioners (AbouRizk et al. 1999, Hajjar and AbouRizk 1999). STROBOSCOPE developed at the University of Michigan and now housed at Virginia Tech is another notable development that has mustered a strong following amongst construction researchers (Martinez 1996, Martinez and Ioannou 1999). Numerous other smaller scale initiatives can also been found in literature. Tommelein et al. (1998) and Halpin and Kueckmann (2002) recently expanded the use of construction simulation into areas pertaining to workflow variability and other lean concepts. Tommelein et al. (1998) and Walsh et al. (2002) have now combined the study of conSawhney, Bashford, Walsh, and Mulky struction supply chain management and construction simulation. Sawhney and Deshpande (2000) developed constructs for Java-based simulation for simulating construction operations over the web. The above described works generally map the history of construction simulation. The common trend has been the heavy use of discrete event approach in construction simulation. Construction simulation has primarily been conducted using discrete event simulation tools (Walsh et al. 2002). This work makes a departure from this well traveled path in that it explores the application of ABMS in construction—as a standalone tool or in combination with discrete event simulation. 2 AGENT-BASED MODELING AND SIMULATION Agent-based modeling and simulation is a methodology in which a simulation experiment is constructed around a set of autonomous “agents” that interact with each other and their underlying environment to mimic the real-world scenario that they replicate (Sanchez and Lucas 2002). ABMS tends to closely resemble how physical, biological, and social systems work in their natural form (Sawhney 2002, Walsh et al. 2003). Some consider this technique a new development; while others simply deem it as a natural extension of existing paradigms such as parallel and distributed discrete-event simulation, and object-oriented simulation (Davidsson 2000). ABMS has been used in a variety of fields including social sciences, ecology, economics, political science and marketing and sales (Bonabeau 2002). In this approach each system is modeled as a collection of autonomous decision-making entities (Bonabeau 2002, Axelrod 1998, Axtell 1999, Sanchez and Lucas 2002). These agents sense and stochastically respond to conditions in their local environments, mimicking complex large-scale system behavior (Sanchez and Lucas 2002). Each agent individually assesses its situation and makes decisions based on a set of rules (Bonabeau 2002). Extremely complex behaviors can arise from repetitive, competitive interactions between agents enabled by the computational power of computers. Researchers can thus explore dynamics out of the reach of pure mathematical methods at the system level, and discover the fundamental rules driving system behavior (Axelrod 1998, Bonabeau 2002). 3 ABMS IN CONSTRUCTION Construction discipline is deeply entrenched in tradition and history. Researchers and practitioners alike are of the view point that there is “central control” behind every construction project; therefore once a plan is created it is assumed that the project will evolve as per this plan and that interaction of construction “entities” will have a minimal impact on this evolution. Few have challenged this approach. Howell (1999) suggested that the happenings within the construction discipline could be better explained based on the agent-based concept. At a micro-level, onsite activities seem to show more “organic” control as compared to the much subscribed central and coordinated control (Walsh et al. 2003, Howell 1999). Systems that show these kinds of behaviors are amenable to the use of agent-based modeling and simulation based inquiry. However, much research needs to be conducted in this field. The construction project is routinely described as a setting in which constant change is a rule rather than an exception (Kim and Paulson 2003), and much of the construction management literature is dedicated to change management. Changes to the project plan occur due to design changes, unexpected delays or interruptions in the supply chain, or field conditions that differ from expectations, among others. Two examples—one from the commercial sector and one from the residential sector—are given here to describe the potential benefits of agent-based modeling and simulation for the construction industry. The first example pertains to construction site safety. Consider the case of a crew of five skilled workers working on the 20th floor of a high-rise building that is under construction. The first thing that leaps in a constructors mind is the safety climate at the work face, since the construction industry is notorious for its poor safety record when compared with other industries (Mohamed 2002). Current approaches to creating a safe construction climate focus upon use of lagging indicators i.e. past accident statistics for similar circumstances (Flinn et al. 2000, Mohamed 2002). No attempt is made to incorporate and study leading indicators relating to organizational, managerial, and human factors (Mohamed 2002). Much of this can be attributed to the lack of availability of modeling and analysis tools to the construction industry. It seems impossible to construct a computer model of the construction environment under study and experiment with safety factors such as trust and support within a group of workers, safety rules and procedures, available safety devices, use of these devices, amount of time available to plan and carry out the work etc. Construction worker safety is further problematic in that some workers are more risk-tolerant than others, and some situations are more risky than others. Nonetheless, accidents happen to both the risk-tolerant and the riskaverse, although in different proportions. Situations can emerge to become more unsafe based on actions taken by workers and situation can become more complex based on interactions amongst and between the workers and the working environment. An agent-based modeling and simulation testbed on the other hand, can be used to mimic the construction environment in which the workers are performing their work along with a heterogeneous set of Sawhney, Bashford, Walsh, and Mulky agents representing these workers to study various aspects of the safety climate. Various aspects of the construction environment can be adjusted along with “fitness and safety” factors of the instantiated “worker” agents to determine the most effective safety plan. Such a proactive approach, unimaginable otherwise will certainly enhance the ability of researchers and practitioners alike to study construction site safety more pragmatically. Simple “whatif” scenarios would allow one to consider the impact of different safety management philosophies on workers of different risk-tolerance, and allow the work to progress in a realistically variable environment, to identify those management practices most directed to zero accidents. Second example pertains to the case of the production homebuilding industry. Bashford et al. (2003) reviewed Securities and Exchange Commission (SEC) filings for 23 publicly traded companies whose core business is the construction of single-family dwellings. The filings were the annual operating statements for the