Chapter 3 Methodology
3.6 Simulation model
3.6.1 Simulation modelling tool
Generally, a simulation model can be developed based on three main categories o f simulation modelling tools (Robinson, 2004):
i) Spreadsheets
ii) Programming languages iii) Specialist simulation software
Pidd (2004) reports that writing the simulation using programming languages, such as FORTRAN was the only option available in the late 1950s. However, these general purpose languages are still useful nowadays since they give the developer the flexibility to design a model with minimum restriction on output format, even though it requires a longer developing time (Shannon, 1975 and Robinson, 2004).
Furthermore, Brooks et al. (2001) and Seila (1995) point out that low modelling cost is another advantage o f using this approach.
The earliest specialist simulation software like SIMULA, GPSS and SIMCRIPT, also known as simulation language (Brooks et al., 2001), or special purpose language (Shannon, 1975) was able to simplify the modelling process and reduce the modelling time by having extra features, such as experimental support, a well-suited syntax, automatic data generation, collection and reporting o f statistics as well as animation facilities (Pidd, 2004; Brooks et al., 2001; and Shannon, 1975). Then, a new generation o f specialist simulation software started to appear with the existence o f powerful computers that have the capability to develop, execute and analyse the models visually and interactively. Such software can be classified as general purpose packages such as Arena, Extend, SIMUL8, and AweSim, or application-oriented simulation packages such as ProModel, ServiceModel, MedModel, MODSIM 111, AutoMod, WITNESS and SIMPROCESS which are suitable for modelling a specific application (Law and Kelton, 2000). A detailed summary o f simulation software including the typical applications, cost o f software, and the features available for model building is provided in Swain (2007).
The detailed history and development o f simulation software can be obtained from Robinson (2005) and Pidd and Carvalho (2006). Among such software, ProModelPC, WITNESS and Simul8 are the softwares most commonly used by academia and industry to develop simulation models according to a survey conducted by Hlupic (2000).
Although these simulation specialist packages offer advantages in terms o f being able to develop simulation models easily in a graphical manner through the available menu functions, the user requires extra time to explore all features provided in the software as well as time to obtain the skills needed to develop the simulation models. On the other hand, spreadsheet simulation is convenient for developing simulation models since the user is generally familiar with spreadsheets like Microsoft Excel. Ragsdale (2004) stated that the electronic spreadsheet has been reported to be one o f the most effective and useful approaches for developing computer models by millions o f business people. Coles and Rowley (1996) also report the increasing use o f spreadsheets by managers as decision support systems. In addition, Evans (2000) and Pecherska and Merkuryev (2005) have shown that the spreadsheet is a powerful tool for teaching both static and dynamic simulation models. With spreadsheets, the user is able to integrate graphics to visualise the results that are updated dynamically with the change o f model input. A spreadsheet also provides statistical tools and functions that allow the user to perform an analysis o f the result directly (Evans, 2000). Seila (2005) indicated that a large number o f functions are available in the spreadsheet and an automation programming language such as Visual Basic Application (VBA) is among spreadsheet features which are capable o f being a platform to conduct a simulation.
Moreover, Seila (2005) reported that spreadsheet simulation is appropriate in developing stochastic models and undertaking sensitivity analysis o f the models with variation o f the unknown parameters, but has a limitation with regards to modelling the complex algorithm and large amounts o f data.
Table 3.4 presents a comparison o f the main categories o f simulation modelling tools based on several features, such as range o f application and the time required to obtain the skills and build the model. Generally, programming languages provide a high range o f application, flexibility and small execution time. However, the specialist simulation software is better than programming language in terms o f duration o f
model building, ease o f use and ease o f model validation. As has been previously discussed, users only require a short time to obtain the software skills necessary to use the spreadsheet simulation that is also affordable in terms o f price. The long execution time for spreadsheet simulation might be overcome by using powerful computer specifications or performing the simulation on multiple computers using a parallel and distributed simulation approach (Brooks et al., 2001).
Table 3.4: A comparison of main categories of simulation modelling tools (Source: Robinson, 2004)
Features Spreadsheet Programming
language
Specialist
simulation software
Range o f application Low High Medium
Modelling flexibility Low High Medium
Duration o f model building Medium Long Short
Ease o f use Medium Low High
Ease o f model validation Medium Low High
Run-speed Low High Medium
Time to obtain software skills Short (medium Long Medium for macro use)
Price Low Low High
Apart from the cost and learning time factors, Sezen and Kitapci (2007) indicate that spreadsheet simulation and commercial simulation software can also be distinguished
‘based on the appropriateness to the specific need’. According to Robinson (2004) and Pidd (2004), the majority o f specialist simulation software are Visual Interactive Modelling Systems (VIMS). As the result, the simulation model can be developed interactively by select the model’s objects via the available menus in the software. For example the “machine” object can be used to model part o f a manufacturing process whereas a “counter and people” object can be used to model the service model or queuing model. In addition, the logic and the flow o f entities for the simulation model can be defined through an existing menu. The simulation software also provides a visualisation o f the model to display the animation when the model has been executed and user is able to interact with the model at any particular time to obtain the results
or modify the model. Thus, the simulation software is more appropriate to develop the event-scheduling approach that is able to define the flow o f entities, event logic and visualize the movement o f entities to get more understanding o f the model. On the other hand, Robinson (2004) stated that the ‘spreadsheet is a relatively straightforward approach to develop a simple time-slice model’ but it is ‘difficult to develop a model animation using spreadsheet’. Nevertheless, several researchers have preferred to use spreadsheet simulation to model a queuing system since the time to develop the queuing model with other applications requires extra time to gain knowledge o f programming languages or special purpose simulation languages (Seal, 1995)
An analysis o f the outcome generated by the spreadsheet and the commercial simulation software can be used to examine the performance and impact o f the simulation model using different simulation modelling tools. Interestingly, the deterministic supply chain analysis performed by Chw if et al. (2002) showed similarity o f the total cost margin for a one year simulation period between spreadsheet and supply chain guru software, with less than 1% o f deviation. It is probably because the rounding o f the input data and truncating o f the processes during the simulation model development process. However, Chwif et al. (2002) claimed that the spreadsheet result might be misleading if the model has variation in demand.
The analysis o f inventory management problems by Zabawa and Mielczarec (2007) using Monte Carlo proved that the spreadsheet is capable o f producing a similar outcome with Extend software for variable demand and lead time.
To choose an appropriate simulation software tool for this study, an analysis o f a simple deterministic IRP model using Microsoft Excel Spreadsheet and Pro Model software is conducted in order to test the capability o f these simulation tools in terms o f the accuracy o f results, modelling time and execution time. A detailed explanation o f the IRP model notations and objective function is presented in Chapter Five. For simplicity, the simulation model is analysed using spreadsheet and Pro Model software for 100 periods o f simulation time using similar demand data. Figure 3.5 shows the screen shoots o f the simulation model for spreadsheet and Pro Model software.
m - n X,
Figure 3.2: Simulation model screen shots (a) Pro Model (b) Spreadsheet
As predicted, the results o f six different inputs generated by spreadsheet and Pro Model software as presented in Table 3.5 showed similarity for all measurement costs.
Table 3.5: Results of comparison analysis between Spreadsheet and Pro Model software
However, throughout the modelling and execution phase o f the simulation model, the spreadsheet model outperformed the Pro Model software. The total modelling time for the simulation model using the Pro Model software was longer than the modelling time for the spreadsheet model. Extra time was required to learn and build an advanced simulation model, even though the author had been used to this software for basic simulation modelling previously. Building this model not only required determining general elements o f the model like locations, entities, path networks and resources but also required using advanced elements, including attributes, variables and external file features. Further, the operation logic statement to specify the activity
of entities at a defined location based on specific conditions needed advanced programming logic skill. Further information regarding the modelling elements and other features in the Pro Model software can be found in Harrell and Price (2002) and Benson (1997).
Surprisingly, in this analysis we found that the execution time o f a 100 simulation period for the Pro Model software at full speed was longer than the execution time for the spreadsheet model at each change o f input parameters. Therefore in this thesis we adopt spreadsheet simulation as a simulation tool to develop the model. This decision supported by Sezen and Kitapci (2007) who also contended that ‘spreadsheets can be effectively used in modelling and simulation o f supply chain inventory problems’.