How to bring predictability to logistics operations through simulation techniques

Certainly, we've all been faced with decisions to be taken without having much data to support them, whether in personal or professional life. However, these “choices in the dark”, if they are not correct, can incur great financial losses in the corporate environment and, depending on the foreseen changes in the company's operation, can bring with them a great inertia that ends up making it difficult to reverse them in the short term. term. Therefore, it is essential that, even in an uncertain scenario, it is possible to predict the performance of the system to be implemented or modified, with the estimation of key indicators of the operation and identification of its main bottlenecks.

Going to the field of logistics, imagine that you have just received the mission to scale the resources related to all the operations involved in boarding and disembarking passengers at a new airport. The project includes, for example, the dimensioning of the check-in, search/security, passport verification, boarding waiting hall, immigration control, customs and gates. It is reasonable to predict that the investments required for this infrastructure work are high, so it is imperative that its dimensioning is as accurate as possible. Otherwise, the system may operate improperly, with bottlenecks that may impair the service offered to passengers.

Do you agree that considering only the average number of times spent in each of the steps involved in each flow (loading and landing) and the productivity of employed resources can lead to errors in sizing? And the reason for this is quite simple: practically every operation has an associated uncertainty, that is, the rate of arrival of passengers, the time spent to perform a task, as well as the productivity of the resources used, have a variability that can be translated through of parameters such as standard deviation, variability and coefficient of variation. Thus, as important as knowing the position measures that characterize the operation, such as the average, it is necessary to have dispersion measures, such as the standard deviation, in hand, so that a system is well dimensioned and so that an adequate estimate can be made. of its performance indicators.

In this sense, simulation is a very powerful tool available on the market through various software such as Simul8®, Arena®, Simio® and AnyLogic®. Next, a video with a simulation of the problem we just discussed, an airport terminal, made by Vancouver Airport Services using Simio® software.

Let us now detail, in a simplified way, the main steps for modeling a system using discrete event simulation:

Figure 1: Main steps for building a simulation model

• First, it is necessary to collect a data sample large enough to extract the statistical parameters of interest to the problem, such as mean, trimmed mean, mode, median, quartiles and standard deviation. Based on these parameters, it is important to “clean” discrepant data, called outliers, which can “pollute” the sample if not removed. This removal of outliers must be judicious and always duly justified.
• With the data already treated, the next step is to find out which probability distribution best fits the collected data. Among the most well-known distributions, we have the normal, uniform, exponential and triangular distributions. Each of these distributions has a set of parameters that characterize it. For example, the normal is defined by the mean and standard deviation, while the exponential is characterized only by the mean. There are software available on the market specialized in ranking the distributions that best reflect the collected sample, among which we can mention StatFit® (Simul8 ® complement) and Input Analyzer (linked to Arena®).

Figures 2: Examples of continuous probability distributions and associated parameters [2]

• Before starting the computational modeling of the problem, it is important to structure the conceptual modeling. In it, the main blocks of the process flowchart used for modeling the problem will be defined, with activity sequencing, queuing forecast and decisions to be taken during the simulation.
• Next, the conceptual model must be translated into a computational model using some software available on the market. Simpler, smaller simulations with few resources involved can even be run in Excel®, but as the size of the problem grows, the use of specialized software becomes necessary. In a simplified way, the computational models are constituted by:

1. CDD-KYC Entities – objects flowing in the system, such as passengers in an airport terminal;
2. Attributes – characteristics associated with an entity, such as preference in serving a passenger;
3. Processes – activities to be performed in the model, which may or may not be associated with some resource, such as service at the check-in counter;
4. Resources – elements that make it possible to assist entities in a process, such as totems or airline employees checking passengers in;
5. Rows – Formed when all resources available to perform a process are busy when a new entity arrives to be attended to, such as the queues that form at the check-in counter when all employees or totems are busy.

In general, the software allows simulating particularities of the operation, such as random resource failures, the maximum time that entities are willing to wait in line before giving up service, scheduled arrivals in time windows, synchronization of activities and drawing based on a rule of thumb. some attribute to some entity. It is also at this stage that we must insert the inputs related to each probability distribution of each activity (such as the rate of arrival of entities to the system and service times), the costs associated with each resource (if you want to generate a cost report at the end simulation) and the total time you want to run the simulation.

Figures 3: Example of computational modeling in Simul8® of the flow of a passenger, including departure at the origin airport, flight and disembarkation at the destination airport [3]

• Still in the computational modeling, we must define which performance indicators we want to know about the system in order to format the results report generated by the software. Among the indicators that are usually monitored, we have the average queue time and the average queue formed in each process (to identify bottlenecks), the utilization rate of each resource employed in each activity (to identify possible idleness and/or teams oversized), average total time entities remain in the system and number of entities that gave up service due to excessive waiting time in the queue.
• After these steps, the simulation must be started, which is usually accompanied by a graphical interface in which it is possible to monitor the arrival of entities to the system, the execution of activities and the formation of queues. This graphical interface can be quite simple or more detailed and realistic, as shown in the previous Vancouver Airport Services video. To avoid that some very discrepant random data generated during the simulation interfere with the results obtained, it may be interesting to replicate the simulation as many times as necessary to reach a pre-established confidence interval for the results.
• Finally, the system performance indicators report generated by the software should be analyzed. With this data in hand, it is possible to assess whether the resources are adequately sized for the service that is planned to be offered to customers, whether there is any bottleneck in the system and how the activities interact with each other.

The application of simulation is not restricted to the dimensioning of transport systems, such as airports, ports, docks of a distribution center, toll plazas or subway stations. It is possible, among many other activities, to evaluate the performance of a production line or even the operation of a warehouse.

Whether to test a new operating model, or to scale or plan a system “from scratch”, simulation is an important tool to assist logistics professionals in decision-making in environments with associated uncertainty.

References:

[1] https://www.youtube.com/watch?v=JuXwEbAvk2Q

[2] https://www.slideserve.com/brenna-hardin/continuous-probability-distributions

[3] https://blog.simul8.com/simul8-tip-label-based-batching-and-collecting/

[4] https://www.anylogic.com/

[5] https://www.simul8.com/

[6] https://www.arenasimulation.com/

[7] https://www.simio.com/index.php

[8] Botter, RC Lecture notes for the course PNV5005 – Modeling and analysis of intermodal transport systems using simulation techniques. Master's Program in Logistics Systems Engineering Poli-USP. 2019.