Approaching a simulation-based problem

To tackle a numerical simulation process that returns accurate results, it is crucial to rigorously follow a series of procedures that partly precede and partly follow the actual modeling of the system. We can separate the simulation process workflow into the following individual steps:

1. Problem analysis
2. Data collection
3. Setting up the simulation model
4. Simulation software selection
5. Verification of the software solution
6. Validation of the simulation model
7. Simulation and analysis of results

   To fully understand the whole simulation process, it is essential to analyze the various phases that characterize a study based on simulation in depth.

   Problem analysis

   In this initial step, the goal is to understand the problem by trying to identify the aims of the study and the essential components, as well as the performance measures that interest them. Simulation is not simply an optimization technique and therefore there is no parameter that needs to be maximized or minimized. However, there is a series of performance indices whose dependence on the input variables must be verified. If an operational version of the system is already available, the work is simplified as it is enough to observe this system to deduce its fundamental characteristics.

   Data collection

   This represents a crucial step in the whole process since the quality of the simulation model depends on the quality of the input data. This step is closely related to the previous one. In fact, once the objective of the study has been identified, data is collected and subsequently processed. Processing the collected data is necessary to transform it into a format that can be used by the model. The origin of the data can be different: sometimes, the data is retrieved from company databases, but more often than not, direct measurements in the field must be made through a series of sensors that, in recent years, have become increasingly smart. These operations weigh down the entire study process, thus lengthening their execution times.

   Setting up the simulation model

   This is the crucial step of the whole simulation process; therefore, it is necessary to pay close attention to it. To set up a simulation model, it is necessary to know the probability distributions of the variables of interest. In fact, to generate various representative scenarios of how a system works, it is essential that a simulation generates random observations from these distributions.

   For example, when managing stocks, the distribution of the product being requested and the distribution of time between an order and the receipt of the goods is necessary. On the other hand, when managing production systems with machines that can occasionally fail, it will be necessary to know the distribution of time until a machine fails and the distribution of repair times.

   If the system is not already available, it is only possible to estimate these distributions by deriving them, for example, from the observation of similar, already existing systems. If, from the analysis of the data, it is seen that this form of distribution approximates a standard type distribution, the standard theoretical distribution can be used by carrying out a statistical test to verify whether the data can be well represented by that probability distribution. If there are no similar systems from which observable data can be obtained, other sources of information must be used: machine specifications, instruction manuals for the machines, experimental studies, and so on.

   As we've already mentioned, constructing a simulation model is a complex procedure. Referring to simulating discrete events, constructing a model involves the following steps:

8. Defining the state variables
9. Identifying the values that can be taken by the state variables
10. Identifying the possible events that change the state of the system
11. Realizing a simulated time measurement, that is, a simulation clock, that records the flow of simulated time
12. Implementing a method for randomly generating events
13. Identifying the state transitions generated by events

    After following these steps, we will have the simulation model ready for use. At this point, it will be necessary to implement this model in a dedicated software platform; let's see how.

    Simulation software selection

    The choice of the software platform that you will perform the numerical simulation with is fundamental for the success of the project. In this regard, we have several solutions that we can adopt. This choice will be made based on our knowledge of programming. Let's see what solutions are available:

    • Simulators: These are application-oriented packages for simulation. There are numerous interactive software packages for simulation, such as MATLAB, COMSOL Multiphysics, Ansys, SolidWorks, Simulink, Arena, AnyLogic, and SimScale. These pieces of software represent excellent simulation platforms whose performance differs based on the application solutions provided. These simulators allow us to elaborate on a simulation environment using graphic menus without the need to program. They are easy to use but many of them have excellent modeling capabilities, even if you just use their standard features. Some of them provide animations that show the simulation in action, which allows you to easily illustrate the simulation to non-experts. The limitations presented by this software solution are the high costs of the licenses, which can only be faced by large companies, and the difficulty in modeling solutions that have not been foreseen by the standards.
    • Simulation languages: A more versatile solution is offered by the different simulation languages available. There are solutions that facilitate the task of the programmer who, with these languages, can develop entire models or sub-models with a few lines of code that would otherwise require much longer drafting times, with a consequent increase in the probability of error. An example of a simulation language is the general-purpose simulation system (GPSS). This is a generic programming language that was developed by IBM in 1965. In it, a simulation clock advances in discrete steps, modeling a system as transactions enter the system and are passed from one service to another. It is mainly used as a process flow-oriented simulation language and is particularly suitable for application problems. Another example of a simulation language is SimScript, which was developed in 1963 as an extension of Fortran. SimScript is an event-based scripting language, so different parts of the script are triggered by different events.
    • GPSS: General-purpose programming languages are designed to be able to create software in numerous areas of application. They are particularly suitable for the development of system software such as drivers, kernels, and anything that communicates directly with the hardware of a computer. Since these languages are not specifically dedicated to a simulation activity, they require the programmer to work harder to implement all the mechanisms and data structures necessary in a simulator. On the other hand, by offering all the potential of a high-level programming language, they offer the programmer a more versatile programming environment. In this way, you can develop a numerical simulation model perfectly suited to the needs of the researcher. In this book, we will use this solution by devoting ourselves to programming with Python. This software platform offers a series of tools that have been created by researchers from all over the world that make the elaboration of a numerical modeling system particularly easy. In addition, the open source nature of the projects written in Python makes this solution particularly inexpensive.

Now that we've made the choice of the software platform we're going to use and have elaborated on the numerical model, we need to verify the software solution.

Verification of the software solution

In this phase, a check is carried out on the numerical code. This is known as debugging, which consists of ensuring that the code correctly follows the desired logical flow, without unexpected blocks or interruptions. The verification must be provided in real time during the creation phase because correcting any concept or syntax errors becomes more difficult as the complexity of the model increases.

Although verification is simple in theory, debugging large-scale simulation code is a difficult task due to virtual competition. The correctness or otherwise of executions depends on time, as well as on the large number of potential logical paths. When developing a simulation model, you should divide the code into modules or subroutines in order to facilitate debugging. It is also advisable to have more than one person review the code, as a single programmer may not be a good critic. In addition, it can be helpful to perform the simulation when considering a large variety of input parameters and checking that the output is reasonable.

Important Note

One of the best techniques that can be used to verify a discrete-event simulation program is one based on tracking. The status of the system, the content of the list of events, the simulated time, the status variables, and the statistical counters are shown after the occurrence of each event and then compared with handmade calculations to check the operation of the code.

A track often produces a large volume of output that needs to be checked event by event for errors. Possible problems may arise, including the following:

• There may be information that hasn't been requested by the analyst.
• Other useful information may be missing, or a certain type of error may not be detectable during a limited debugging run.

After the verification process, it is necessary to validate the simulation model.

Validation of the simulation model

In this step, it is necessary to check whether the model that has been created provides valid results for the system in question. We must check whether the performance measurements of the real system are well approximated by the measurements generated by the simulation model. A simulation model of a complex system can only approximate it. A simulation model is always developed for a set of objectives. A model that's valid for one purpose may not be valid for another.

Important Note

Validation is a where the level of accuracy between the model and the system is respected. It is necessary to establish whether the model adequately represents the behavior of the system. The value of a model can only be defined in relation to its use. Therefore, validation is a process that aims to determine whether a simulation model accurately represents the system for the set objectives.

In this step, the ability of the model to reproduce the real functionality the system is ascertained; that is, it is ensured that the calibrated parameters, relative to the calibration scenario, can be used to correctly simulate other system situations. Once the validation phase is over, the model can be considered transferable and therefore usable for the simulation of any new control strategies and new intervention alternatives. As widely discussed in the literature on this subject, it is important to validate the model parameters that were previously calibrated on the basis of data other than that used to calibrate the model, always with reference to the phenomenon specific to the scenario being analyzed.

Simulation and analysis of results

A simulation is a process that evolves during its realization and where the initial results help lead the simulation toward more complex configurations. Attention should be paid to some details. For example, it is necessary to determine the transient length of the system before reaching stationary conditions if you want performance measures of the system at full capacity. It is also necessary to determine the duration of the simulation after the system has reached equilibrium. In fact, it must always be kept in mind that a simulation does not produce the exact values of the performance measures of a system since each simulation is a statistical experiment that generates statistical observations regarding the performance of the system. These observations are then used to produce estimates of performance measures. Increasing the duration of the simulation can increase the accuracy of these estimates.

The simulation results return statistical estimates of a system's performance measures. A fundamental point is that each measurement is accompanied by the confidence interval, within which it can vary. These results could immediately highlight a better system configuration than the others, but more often, more than one candidate configuration will be identified. In this case, further investigations may be needed to compare these configurations.