1 “The Last Planner”: Shielding production through weekly work plans a Technique
N. Optimization of construction operations through simulation and genetic algorithms
a. Technique
Another tool available to construction managers to accelerate project completion is the use of simulation. Simulation optimization is defined as the process of maximizing information retrieval from simulation analysis without carrying out the analysis for all the combinations of input variables (Carson and Maria 1997, Marzouk and Moselhi 2004). Computer-based simulation is one technique that has been lately used to model uncertainties associated with construction operations, particularly earthmoving operations, with the objective of optimizing construction processes to reduce construction costs and delivery time.
Discrete-event simulation has been used to analyze and design construction operations for over three decades (Martinez and Ioannou 1999). General-purpose simulation tools and languages have been developed with the intention of targeting a very broad domain and to be used with almost any type of operation. Conversely, special-purpose simulators are tools that have been designed for specific construction operations, targeting thus a narrower domain. In addition, frameworks such as modeling paradigm or simulation strategy and others have also been created to guide users in the development of simulation models (Hooper 1986, Balci 1988, Martinez and Ioannou 1999).
The most important characteristics of simulation tools are their simulation strategies and their level of flexibility. Most simulation systems use two strategies: process interaction (PI) and activity scanning (AS). However, other models have also been implemented combining event scheduling (ES) with PI or AS. Flexibility refers to the capability of the simulation tool to model complex situations and to adapt to different application requirements. Thus, simulation systems can be very flexible and programmable, and allow for modeling complex and detailed
construction operations; other systems can be very simple and non-programmable tools with limited modeling capabilities which therefore allow only for modeling simple construction operations. However, simple simulation systems still generate efficient results and are easier to learn.
The difference between PI and AS strategies is given in the viewpoint from which they are written. PI models are written from the point of view of the entities or transactions that flow through the system. This strategy is intended for modeling operations where the moving entities have several attributes and the resources (including machines) that serve these entities have few attributes or interactions (Hooper 1986, Martinez and Ioannou 1999). AS strategies, on the other hand, are written from the point of view of the various activities being performed. AS models focus on identifying these activities and the conditions under which they take place. Whether a simulation tool is PI or AS-based has an important impact on the way the system is modeled and the way it is presented to the computer (Evans 1989, Martinez and Ioannou 1999). Some
researchers argue that one strategy is superior to the other; nonetheless most available research agrees that both strategies have equal power and its usefulness depends rather in the system they are intended for. IP systems are more suitable for manufacturing purposes in which materials undergo a fixed pattern after they arrive to the system, and only leave as final products. In contrast, AS are more appropriate for modeling construction operations which involve many interacting resources that can perform in different states and where different conditions are required to carry out activities (Martinez and Ioannou 1999). Most models are also represented
using activity cycle diagrams (ACDs), which are networks that naturally describe three-phase AS models.
Several general-purpose simulation systems have been developed specifically for modeling construction operations based on some form of ACDs and AS or three-phased AS strategies. Some of these include CYCLONE (Halpin and Rigs 1992), REQUE (Chang 1986), COOPS (Liu 1991), CIPROS (Odeh 1992), STEPS (McCahill and Bernold 1993), and STROBOSCOPE (Martinez 1996).
Marzouk and Moselhi propose a framework called SimEarth that specifically allows the optimization of earthmoving operations through the use of computer simulation and genetic algorithms (Marzouk and Moselhi 2004).
The optimization process uses computer simulation and genetic algorithms to search for a near- optimum fleet configuration, taking into account their availability to contractors. The genetic algorithm considers a series of qualitative (i.e. type of resources and their combinations) and quantitative (i.e. quantity of each resource used) variables that determine the production of earthmoving operations. The simulation tool allows for estimating the time and cost of these operations which enable efficient planning of earthmoving operations (Marzouk and Moselhi 2004). By optimizing earthmoving operations, the time and cost of performing these can be minimized, reducing overall project schedule and costs. In addition, the framework allows for time-cost tradeoff analysis and the performance of what if scenarios with respect to fleet configurations.
b. Implementation
The SimEarth framework was developed and implemented in Microsoft environment, and consists of the following components: Earth Moving Simulation Program (EMSP), Equipment Cost Application (ECA), Equipment Database Application (EDA), Hauler’s Travel Time Application (HTTA), Earth Moving Genetic Algorithm (EM_GA), and Output Reporting Module (ORM) (Marzouk and Moselhi 2004).
ECA is a spreadsheet application developed to provide the user with the total hourly owning and operating costs and their respective breakdown. The application was design to be fully
compatible with the Caterpillar performance handbook to enhance its applicability in the industry. EDA is a database that contains essential equipment characteristics such as hauler’s allowable speeds, and it supplies the entire system with the information to be used in the simulation process. HTTA is a fuzzy clustering model that estimates hauler’s travel time (Marzouk and Moselhi 2004).
Earth Moving Simulation Program is the simulation tool that performs replications of
earthmoving operations based on a predefined set of resources and entities. The program utilizes discrete event simulation and object-oriented modeling which facilitates the modeling of
construction operations. EMSP contains the main activities of earthmoving operation which include loading, hauling, dumping, and returning. These are also classified into two types: bound-to-happen and conditional activities.
The simulation program uses a three phase simulation approach by tracking activities in 3 phases. Thus, in phase one, the first activity is removed and the simulation time is advanced to
the next time. In phase two, all due bound-to-happen activities are carried out, and, in phase three, all possible conditional activities are performed.
Earth Moving Simulation Program receives its input from the framework in two different ways, through parameters that are passed through its main function and by reading from external files. EMSP’s main function passes different types of parameters including simulations performed either in a test manner or an analysis manner, interactions among equipment in the fleet under consideration, selected fleet scenario for simulation analysis, presence or use of second hauler, selected set of activities involved in the simulation process, number of simulation runs, and conditions for simulation analysis termination. The external files, on the other hand, feed the program with external information related to soil type, scope of work, equipment characteristics, and possible durations of the involved activities represented in the form of probability density functions (Marzouk and Moselhi 2004).
To start the simulation analysis, the construction manager or the applicable user is required to specify the type of secondary activities involved in the project and the associated fleet scenario to be tested. Secondary activities can include spreading, compacting, and such. The main activities of loading, hauling, dumping and returning are selected by default. Then, the user has to specify the type of equipment available in the project to perform all main activities and the equipment available for each identified secondary activity and its corresponding model. After the equipment for both primary and secondary activities has being selected, the user introduces into the program all the relevant physical characteristics of the system such as characteristics of the material to be hauled, and any other that may impact any activity. Once all physical entities of the system have been established, these are mapped by their representative classes. For example, for the main activity of hauling, the class that represents this object contains the characteristics of the hauler unit including unit type, model, payload, and hourly owning and operating costs. These characteristics are the data member variables of that particular class. The framework’s optimization module uses a genetic algorithm called Earth Moving Genetic Algorithm and Pareto optimality. The genetic algorithm contains two measures of fitness that allow the calculation of project duration and project total cost. These measures are obtained based on the pilot runs carried out by the simulation engine. Thus, based on the information contained in the database and inserted by the user, the simulation program runs a series of replications of the earth operation processes under different pre-established scenarios, and the generic algorithm calculates the time required by each piece of equipment to complete its task. The total cost is also calculated based on direct and indirect costs. The direct costs are estimated based on the time equipment is assigned to the project, and its associated owning and operating costs. The indirect costs, on the other hand, can be of two types, time related and time
independent. The user is thereby able to define the types of indirect costs that are to be applied according to project characteristics.
The following example illustrates the use of the program in selecting the most appropriate fleet configuration to optimize earthmoving operations. Three different fleet scenarios are considered as shown in table 3 to find the near-optimum fleet configuration through simulation techniques. The example involves moving a specific amount of earth from a certain distance. It is necessary to know specific characteristics of the soil such as loose and bank densities. All these
parameters, along with the characteristics of each fleet scenario are introduced into the
simulation system. The user is also required to introduce the probability distributions associated with the duration of the main and secondary activities which in the example are spreading and compacting.
Equipment characteristics Scenario 1 Scenario 2 Scenario 3
Loaders Range (1-10) for scenarios
Type CAT 992G CAT 990SII CAT 988F
Bucket capacity (m3) 12.3 9.2 6.9
No. of passes 4 3 3
Hourly owning and operating cost
(dollars/h) 300 250 175
Haulers Range (15-20) for scenarios
Type CAT 777D CAT 773D CAT 769C
Payload (ton) 81.7 45.8 33.46
Hourly owning and operating cost
(dollars/h) 215 160 130
Dozers Range (1-10) for scenarios
Type CAT D8R
Cycle production (m3) 27
Hourly owning and operating cost
(dollars/h) 150
Soil Compactors Range (1-10) for scenarios
Type CAT CS-583C
Cycle production (m3) 19.1
Hourly owning and operating cost
(dollars/h) 90
Table 3. Characteristics of fleet scenarios (taken from Marzouk and Moselhi 2004, pp. 111)
Based on the inputs made by the user, the simulation engine carries out a series of pilot runs, and the generic algorithm calculates the required measures of fitness of project total duration and project total cost. The system then returns the calculated total project duration based on entered information related to the total quantity of earth to be moved, the daily production, and the scheduled hours per day; and it calculates project total cost from the scheduled hours per day, the equipment hourly cost, the number of equipment associated to each scenario, the total
quantity of earth to be moved, daily production, scheduled working days per month, time-related indirect cost, and time-independent indirect cost (Marzouk and Moselhi 2004).
Ultimately, the system provides the construction manager or the applicable user with two solutions. The first one contains the minimum calculated cost and its associated duration, and the second solution provides the minimum calculated duration and its associated cost. Thus, the construction manager can perform time-cost tradeoff analysis and select the fleet configuration that best minimizes operations duration.
c.
Advantages
Computer simulation and genetic algorithms have been extendedly applied within the
construction industry because they are an efficient tool to optimize construction operations. By optimizing construction operations, the time and cost of carrying them out is minimized. Optimization of construction operations also enhances the efficiency of construction processes. The framework presented allows the identification of a near-optimum fleet for earthmoving operations while it presents a series of useful features. It develops efficient optimizations as it takes into account both qualitative variables such as the type of resources and combinations
used, etc., and quantitative variables such as the quantity of each resource used, etc. In addition the model considers and accounts for actual availability of equipment. Finally, it allows users to consider and evaluate what if scenarios related to fleet configurations and time-cost tradeoff analysis. All these features allow the construction manager to efficiently reduce the uncertainty associated with construction operations and therefore improve construction planning.
d. Key elements to ensure a high degree of success
The implementation of technologies involving simulation and the use of genetic algorithms require knowledge and expertise. Therefore, training is substantial for a proper implementation of these techniques.
During experimentations with simulation technologies, the construction manager or engineer changes the different parameters in the model or the logic of the operations. Therefore, it is also important that the construction manager or the engineer learns how to properly enter the needed inputs and how to control the different permitted options that the program allows in order to obtain adequate results.
Despite of its powerful features, the simulation system is only a tool that allows the engineer to find optimality searches related to construction operations, or the near-optimum fleet in the case of the framework presented. The search for this optimality is modeled by the simulation program but it has to be guided by the knowledge and experience of the engineering and construction manager. Therefore, knowledge and understanding of the operations under analysis is also vital for the users to be able to efficiently employ simulation and optimize construction operations. The development of complex simulation models can also be substantially enhanced when combined with 3D models. Simulation modeling and 3D visualizations can be very helpful in designing complex construction operations and making optimal decisions. A visualized
simulated representation of the construction operation is a more realistic tool which provides the user with comprehendible feedback, indispensable for adequate analysis. In addition, the 3D visualization tool can provide valuable insight into the details of construction operations that are usually hard to perceive and therefore disregarded (Kamat and Martinez 2001).
e.
Disadvantages
Typically, the effort and knowledge involved in the development and implementation of
simulation models tend to limit the use of simulation in construction (Mohamed and AbouRizk, 2005). Building and utilizing simulation applications require experience, technical knowledge of the construction system and of the simulation technology, and substantial investments in time and money, and given the relatively short duration of a project’s construction process, the potential achievable benefits may not always be perceived as to be worth the associated costs of implementation. Furthermore, without the proper expertise and tools, modeling the simulation system can become a time consuming and ultimately worthless investment (Mohamed and AbouRizk, 2005).
Once the model has being built, its utilization is not a trivial process. Carrying out simulation runs and experimentations typically involve modifications in the topology of the model which again requires more effort, knowledge and expertise (Mohamed and AbouRizk, 2005).
f.
Applicability and use
There is an increasing number of simulation and genetic algorithm applications within the construction industry that have been developed aiming at optimizing operations during
construction with the intention of minimizing the total duration and costs of these. Because of the potential benefits and advantages that simulation tools offer in the analysis and planning of construction, its adoption has been slowly gaining acceptance within the industry.
However, there are also a series of important factors that have hindered a broader implementation of simulation technologies. First, as most technologies, implementing
simulation models require high investments substantially increasing construction costs. Then, effort, time, knowledge and experience are also prerequisites not only to build the simulation model but to run it as well. Finally, because of the uniqueness of construction projects, the development of the simulation representation is a time-consuming task compared to overall construction duration. All of these issues have contributed to a limited applicability of simulation and such technologies (Mohamed and AbouRizk, 2005).