Dynamic simulation as decision support tool

So far we have mentioned how simulation presents itself as an alternative tool to the traditional problem solving methods used by decision makers within the various SCM processes.

Let us now go into the detail of this application to evaluate the main aspects inherent to the application of this methodology in the SCM context: how is a simulation model developed? What is needed for its operation? How is it actually used?

In the next paragraphs we will answer these and other questions, trying to provide a complete discussion of all the aspects inherent to the implementation of a dynamic simulation model for the governance of decisions made within SCM processes.

Which type of simulation?

When talking about simulation, we are considering a vast range of methodologies and applications that are used in very different fields. For example, FEM analysis (Finite Elements analysis), which is often carried out during the design and prototyping of a new product to test its mechanical properties in advance, represents a particular use of simulation.

In cases of application of simulation to the supply chain, we can summarize three different approaches existing in the literature:

 

1. System Dynamics

System dynamics is a modeling approach that captures the dynamic interrelationships within complex systems over time. It employs feedback loops, stocks, and flows to represent the system’s structure and behavior, allowing for the simulation of various scenarios. Developed by Jay W. Forrester of MIT, this methodology is widely used in areas such as business, economics, and environmental studies. System Dynamics aids in understanding and predicting system behavior, facilitating strategic decision-making by illustrating how changes in one part of the system can reverberate throughout the entire interconnected system. A simple example of most used System Dynamics models are SIR (Susceptible, Infectious, Recovered) epidemiologic models.

2. Agent-Based modelling

Agent Based Modeling: considers groups (clusters) of agents and how they influence each other depending on the relationships that bind them. A cluster is representQative of the (appropriately simplified) behavior of a certain object or individual, therefore the level of detail of this approach can vary widely depending on how much of the real behavior of individuals needs to be represented. Traditional applications of this methodology concern epidemiological models for estimating the market share and/or level of adoption of new products within a closed (market) system.

3. Discrete Event

Discrete Event Modeling: it is the simulation method that allows you to reach the maximum level of detail, whereby each object in reality is represented as an entity , depicted by its own typical parameters and its own operating and interaction logics. It is often used to solve complex problems relating to the branch of queuing theory, whereby the analyzed process is broken down into its main elements in order to determine the expected performance when the stochastic behavior of the parameters and inputs to the system varies (e.g. queue of customers at a post office).

To select the correct simulation method, a first critical decisional level is the level of abstraction of the methods. In facts, each method serves a specific range of abstraction levels, and a modeler should be deciding upon which method to use based on the problem’s characteristics.

Simulation methods

System Dynamics (SD):

System Dynamics operates at a high level of abstraction, focusing on the overall structure and behavior of a system. It represents key variables and their interconnections using causal loop diagrams (CLDs) and Stock and Flows Diagrams (SFDs), emphasizing feedback loops and accumulations. SD provides a holistic view of system dynamics but may simplify individual components, behavior, and intricate interactions.

In facts, the system behavior is defined by simple parameters interactions and mathematical relationships of parameters. For example, in an SD environmental model, the model can estimate the resource cost over time (e.g. oil) based on a mathematical the interaction of other parameters (e.g. oil extraction rate, oil consumption), but does not allow to detail the consumption in the whole world (e.g. considering tariffs) nor define different consumer behaviors (e.g. drivers in the US tend to make more miles by car than people in the Netherlands). Agent Based Modelling can overcome the SD limitations.

Agent-Based Modeling (ABM):

Agent-Based Modeling allows for a more granular level of abstraction by simulating individual agents with unique characteristics and behaviors. ABM captures the heterogeneity and autonomy of entities within a system, offering a detailed perspective on how individual components interact. This method provides a nuanced understanding of emergent patterns and behaviors, making it suitable for studying complex systems with diverse actors. Bringing back the example from SD, in this case a modeler would be able to define oil consumers from the US and from the Netherlands. Eventually, both of them could be clustered in sub-clusters based on their wealth or location (e.g. people living in big cities run less miles). This is configurable, up to each single entity behavioral simulation. In this case, oil consumption can be estimated with more precision compared to SD, and the model can predict the future consumption based on the people elasticity to price changes.

Discrete Event Modeling (DEM):

Discrete Event Modeling focuses on specific occurrences or events within a system, providing a detailed level of abstraction. It is best used to represents processes as sequences of discrete events, capturing the timing and order of activities. DE specifically focuses on process flows, making it the best method to address Supply Chain problems, which can be broken down in many sub-processes (e.g. logistics, storage, manufacturing…). While also Discrete Events allow the creation of agents, it is important to clarify the difference between Discrete Event and ABM agents.

In DE, there are two types of agents: passive and active agents. In ABM, agents are only active due to the fact that each agent has its own defined behavior and interact with each other. 

To give an example of “passive” agent in DE, we can think of a production order or a lot in a manufacturing simulation model. This agent is made of configurable parameters: size, weight, color, just to provide some, but is not characterized by a pre-defined behavior as it is inanimate. Instead, other active agents (e.g. machines and human resources) containing the process flow will define the routing of the order from the beginning to the end of their life.

The tailored approach

The first fundamental element to understand about the creation of a simulation model is that it is developed starting from scratch, without any “preset” models to rely on. This is due precisely to the need to grasp the typical nuances of the business, company and supply chain : if on the one hand there may be an agreed framework  on how to develop the data useful for the decision-making process, on the other hand the modeled reality is peculiar and therefore requires ad hoc modeling.

In this sense we can say that creating a simulation model for a specific company can be compared to the construction of a house: the client will rely on an architect and provides him indications about his wishes, both in terms of expected output and constraints (economic and not),. The architect will create a project capable of accommodating all requests. This does not mean that there is an absolute best project: just as happens with a house, by relying on different designers you obtain different projects that are all valid.

The critical issue therefore lies in finding the model that correctly represents the right trade-off between precision of the result provided, reliability, scalability and onerousness of implementation and maintenance (measured in cost and time).

 

 

Confidence level of detail
Complexity vs accuracy

It is clear that, once the type of simulation model has been defined, the advantages of having an ad hoc model far outweigh the costs of its creation. Among them we include:

  • Obtaining a local optimum solution, valid only for the analysis context. This allows to overcome the traditional limits of analytical methods: they provide global optimal solutions but with a reduction in the realism of the analyzed problem (the system must necessarily be translated into constraints), which can actually jeopardize the applicability of the solution provided.
  • The possibility of “manually” managing levers, parameters and thresholds to dynamically adapt the results provided by the model to changes in the business and therefore in company objectives. This type of interaction with the model, made possible through appropriate parameter setting interfaces, allows decision makers to experiment freely and without consequences with different decision-making alternatives.
  • Recurrent use of the model, to carry out ex-post or ex-ante analyses. The model also represents a useful benchmark tool for performing root-cause analysis of problems that affected the supply chain in a specific period or for verifying the causes of unexpected results.

The scalability of the model also allows the conformation of the modeled system to be modified in a short time following changes in the structure of the analyzed supply chain.

From this point of view, the addition of a warehouse, a factory or new products/components within the analyzed supply chain is much less costly, in terms of implementation times, compared to modifying/adding constraints of a analytical method (algorithm).

Mastering complexity

When we talk about complexity, it is often difficult to quantify it in a unique and easily understandable way. If on the one hand for the evaluation of traditional methods (analytical algorithms) complexity is “measurable” according to the theory of computational complexity, which analyzes the time and computational capacity required to solve a given problem, on the other hand the same methodology cannot be applied to simulation models as they “lack” a problem to solve.

Therefore, leaving aside the mathematical aspect linked to the measurement of complexity, we can limit ourselves to carrying out a qualitative analysis based on the main elements that contribute to identifying the computational complexity of a model (whether analytical or simulation).

01

Problem solving in supply chain processes

Problem by problem: Analytical methods and Dynamic Simulation compared

02

Dynamic simulation as decision support

How to choose the right simulation methodology for a tailor made approach to reliable data-driven decision

03

The simulation dictionary

Model, Scenarios, Simulations, Digital Twin: what is what?

04

Skills and technologies

Learn about the skills and technologies involved in this cutting-edge technology

05

Simulation modelling roadmap

From beginner to level expert: the road ahead

06

Applying Simulation to the S&OP process

MRP, Scheduling, and in between Simulation Modelling: discover how