Simulation dictionary: Models, Scenarios, Simulations and Digital Twins
1. Models
With the term “model” we entail the representation of an object, entity or process in an artificial (virtual) environment, capable of reproducing its functioning and performance according to a certain degree of reliability .
This definition therefore allows us to state that a model supporting a generic SCM process, whether based on simulation or not, must replicate the input data flows (sales plan, production plan, financial objectives, etc.), process them correctly and provide the decision maker with a vision of the expected results.
The definition of model also introduces some important concepts which we summarize below:
- A model constitutes a representation of something observable/perceptible in reality: just as the same object can be represented in various ways depending on the painter who creates the painting, the way the modeler decides to represent the observed phenomenon is often subjective and it lends itself to the interpretation of the person who creates the model and to his perception of reality. Therefore it is important that in the phase of defining the model requirements there is always an expert figure about the reality studied, whether linked to the business or technical field.
- The phenomenon represented by the model can be of any type: from a physical object (e.g. the functioning of a warehouse or a factory) to an abstract process (e.g. the phases of the management process of a tender). The description of the phenomenon to be modeled must be as framed as possible, to avoid having to include factors, variables and interactions in the model that are not necessary for its analysis.
- The environment in which the model is conceived, developed and simulated is purely artificial, or better yet, virtual: while for a model of a building there are different implementation environments (from CAD software to 3D printing of the model), the simulation model exists only virtually and must therefore be developed using specific software. If on the one hand the virtuality of the simulation model does not allow easy recognition of the effort and complexity necessary to create it, on the other hand it enables a key advantage of this type of applications: the possibility of creating copies of the model on which is possible to make experiments with variations, additions, changes, etc. at zero cost and without having any type of repercussion on the reality analyzed. The difficulty and burden of creating a precise simulation model is therefore compensated by the fact that it can be modified freely and infinitely without incurring any danger or additional cost. This advantage leads the decision maker to adopt a new type of decision-making approach based on experimentation (trial and error approach) at zero cost and time. This approach allows to test a multitude of alternative ideas and solutions and to direct the decision maker’s mind towards non-standard solution paths. These way of reasoning is in any case different from the traditional decision making mindset, with evident benefits regarding effectiveness and efficiency of the implemented solutions.
- The model must reproduce the functioning of the observed phenomenon: this presupposes in-depth knowledge of how the phenomenon operates. However, although the functioning of processes, companies, etc. is often known, the results often tend to deviate from those expected. This is mainly due to the presence of stochastic and irrational factors within any type of phenomenon. In the simulation model, unlike other applications, it is also possible to take these biases into consideration and study their effect on the results, to establish effective containment or abatement strategies.
- The creation of a simulation model requires establishing a priori the appropriate level of detail to analyze the phenomenon. This is a key aspect as it affects on the one hand the quality of the results provided by the model and on the other the time and design effort for its implementation. Therefore, even for simulation, as for any project, there is a trade-off between quality and time (cost) of implementation: a clear definition of how the model must work for determining the results is fundamental for the correct estimate of the development time.
Having briefly clarified the fundamental concepts underlying the definition of model, we then move on to analyze how many and which models may be necessary depending on the analysis needs.
We can distinguish two main cases: the one inherent to the creation of a single model and the one which instead requires the creation of multiple models.
1.1 Unique model
In the case of single development, it is necessary to create a model because we want to know the state of the art and the performance of a phenomenon (usually process) that may exist or still be created.
An application example may regards a situation where management needs to obtain information regarding company processes that are not able to be monitored correctly through the existing information systems and technologies.
Let’s think for example about monitoring and estimating CO2 equivalent emissions in relation to a supply chain: it is unthinkable to analytically measure emissions at each step of the supply chain and for each process. However, in order to have an estimate of the environmental impact and what to do to mitigate it, the creation of a simulation model can represent the right application for the calculation and tracking of all emissions, both direct (scope 1) and indirect (scope 2 and 3).
In others situations in which the model concerns a process that does not yet exist, it is clear that the simulation in this case not only performs an analysis function but also a design function of the characteristics of the new process. A classic example of this use concerns the design and subsequent set up of a supply chain starting from scratch (for example for the marketing of a new product or for the entrance into a new market). In this case the model, built according to the strategic objectives that the company wants to pursue, allows to verify structural choices that would otherwise be made in conditions of limited visibility (as often happens, there is always clear evidence of the investments while the returns are much more uncertain and estimated on the basis of previous experiences). From this point of view, the effort inherent to the creation of the simulation model is amply rewarded by the efficiency that it is able to guarantee to the structure of the new supply chain.
1.2 Multiple models
When it is necessary to create multiple models, we can deal with two situations: complementary models and alternative models.
The creation of complementary models is necessary when the scope of analysis is so broad and varied as to require a subdivision of the project scope into several parts, each of which will have its own underlying model. This is the case, for example, of the study of complex corporate realities: having to create a model capable of simulating the entire functioning of a supply chain, from the analysis of demand to the scheduling of production orders, it is convenient to create multiple modules (models) depending on the purpose and level of detail they need to consider. The scheduling model will therefore be completely different from that for the management of distribution logistics and so on.
The case of alternative models instead arises when management is called upon to evaluate possible evolutions of company processes with respect to the actual way of operating. In this case, a model is created to represent the starting situation (As Is model) and then copies are created.
Making appropriate modifications on models’ copies (To Be models) they become capable of simulating the functioning of the process under substantial changes introduced by management.
Let’s take for example the case of a company that distributes its products locally through a simple distribution network (a central depot with couriers) that at a certain point decides to review its network to improve the level of service and at the same time optimize costs. As possible decisions, the company wants to evaluate the alternative of a two-level distribution network (1 central depot, N peripheral depots) or to use alternative means for deliveries (e.g. micro-mobility vehicles, which would allow to significantly increase the number of vehicles in circulation). It is clear that choices of this magnitude, which substantially modify the current supply chain, require the modification of the starting model and the creation of alternative models (To Be) in order to correctly evaluate the impact of this type of strategic decisions on company objectives and performance.
2. Scenarios
The concept of scenario could easily be confused with that of model. We said that the model is what represents a phenomenon in a virtual environment and which replicates its functioning under specific conditions. The scenario is instead identified by the input dataset of a specific model, and therefore identifies what happens to the model when it operates under specific conditions.
According to this definition, a model can therefore have different scenarios, that is, it can be simulated several times by feeding it with different data sources. On the contrary, the same scenario, i.e. the same starting dataset, can be simulated on multiple alternative models (in case of multiple models).
Scenarios are often used to test different situations of what can happen to the analyzed model and therefore represent events that are out of the control of the decions maker (i.e. the company). It is therefore clear that the generation and testing of the model with different scenarios is particularly useful when operating in conditions of high uncertainty or when it is necessary to develop a risk analysis and related containment plans (where each scenario corresponds to the occurrence of certain risks).
Returning to an example mentioned previously regarding the design from scratch of a supply chain (for which the creation of only one model is therefore sufficient), examples of possible scenarios that management might be interested in testing could regard:
– changes in demand expectation: substantial growth or decrease in demand (which may reflect situations of unexpected appreciation of the product by the market or lack of market maturity);
– changes in the demand presentation pattern (same overall volume but distributed according to different demand curves than expected, for example with strong seasonal peaks);
– contraction or expansion of demand (in terms of market segments) following the entrance/exit of competitors in the reference market;
– changes in the network: unexpected changes in production/storage/transport capacity of network partners (e.g. bankruptcy of a transport company, fire in a warehouse, etc.); unexpected changes inthe distribution structure (e.g. closure of roads or trade routes); etc.
– exogenous events: such as staff strikes; lack of raw materials or components following global disruptions (such as economic crises or wars); temporary/permanent closures of facilities (for example following lock-downs to contain epidemics); etc.
All these examples constitute valid test scenarios for different supply chain models. Obviously the choice of scenarios to simulate depends on the object of the simulation and the risk propensity of the company’s management.
Nonetheless, in the scenario definition phase it is often a good idea to involve figures who are experts in risk management or who have gained experience in the field of study, in order to select and prioritize the risk/opportunity scenarios that are most likely to occur or that can lead to greater harm/benefit.
A further reason that may push the decision maker to have to develop multiple scenarios may instead concern pure design choices of the model variables. In fact, it may happen that it is necessary to test different combinations (sets) of variables depending on the decision-making alternatives available.
For example, let’s think of a model for analyzing the productivity of a department made up of N production lines and M work teams (one team per line): in this case it could be useful to create a series of scenarios where the possible combinations of teams and lines are evaluated and possibly optimized. In this case the number of scenarios to be tested is potentially equal to the Cartesian product of the two sets (NxM) and can be therefore high: for this reason it is always good to perform a prior qualitative screening of the scenarios, eliminating from the analysis the ones that cannot be implemented due to technical constraints.
3. Simulations
When we talk about simulation we mean the so-called run of the model, i.e. its execution in order to visualize its results. Each simulation therefore corresponds to a specific run of the model and will be characterized by a set of specific results.
The need to carry out more than one simulation lies in the nature of the model: if there are no stochastic variables in it but only deterministic ones, then all the runs (simulations) performed for the same scenario will lead to the exact same results. In this case we can limit ourselves to carrying out a single simulation, giving the model the role of “calculator” of the variables considered.
Obviously we will run as many simulations as scenarios considered.
However, if the model presents one or more stochastic variables, in this case the repeated run of each scenario is necessary to determine statistically valid values (exploiting the central limit theorem) for the results obtained. In fact, by running multiple simulations of the same scenario with stochastic variables, we will see that the results always vary (obviously the dispersion of the results depends on the degree of influence that the random variables have on them). Therefore, to get a valid idea of where these results can be positioned according to a certain reliability interval, it is necessary to carry out a certain number of simulations (heuristically greater than 30) and evaluate the average, assuming the validity of the central limit theorem.
In this case, therefore, the model takes on the actual function of a simulator, understood as a generator of sets of potentially infinite results due to the random nature of the variables that make up the model. Obviously, as the number and importance of stochastic variables increase, it is a good idea to considerably increase the number of simulation runs carried out for each scenario, in order to consider the widest possible spectrum of cases (and derive important information on the best case and worst case results).
In conclusion, we can summarize what has been stated until this paragraph with the following schematization.
4. Digital twin
In the ever-evolving landscape of digital technologies, the concept of digital twins has emerged as a groundbreaking innovation with profound implications across various industries. From manufacturing and healthcare to urban planning and beyond, digital twins are revolutionizing how we conceptualize, analyze, and optimize complex systems. However, within the realm of digital twins, there exist two primary paradigms: monitoring and simulative. Let’s delve into the intricacies of each, explore their applications, and understand the added value they bring to different types of analyses.
Monitoring Digital Twins:
At its core, a monitoring digital twin is a virtual representation of a physical asset, process, or system. It continuously collects real-time data from sensors, IoT devices, or other sources embedded within the physical counterpart. This real-time data is then synchronized with the digital twin, providing stakeholders with a comprehensive view of the asset’s performance, condition, and behavior.
Applications:
- Predictive Maintenance: By analyzing real-time data streams, monitoring digital twins can predict when equipment or machinery is likely to fail, enabling proactive maintenance activities. This minimizes downtime, reduces repair costs, and prolongs the lifespan of assets.
- Performance Optimization: Monitoring digital twins enable organizations to monitor key performance indicators (KPIs) in real-time and identify opportunities for optimization. Whether it’s improving energy efficiency in a manufacturing plant or enhancing resource utilization in a smart building, real-time monitoring facilitates data-driven decision-making.
- Quality Control: In manufacturing environments, monitoring digital twins can track and analyze production parameters in real-time, allowing for early detection of defects or deviations from quality standards. This ensures that products meet specifications and enhances overall product quality.
Simulative Digital Twins:
In contrast to monitoring digital twins, simulative digital twins focus on creating a virtual replica of a physical asset, process, or system, incorporating sophisticated models and algorithms to simulate its behavior under various conditions. These digital twins enable what-if scenario analysis, predictive modeling, and optimization through simulation.
Applications:
- Design and Development: Simulative digital twins play a crucial role in the design and development phase of products and systems. Engineers can simulate different design configurations, assess their performance, and optimize parameters before committing to physical prototypes. This accelerates the product development cycle, reduces costs, and enhances innovation.
- Operational Planning: In complex systems such as transportation networks or urban environments, simulative digital twins facilitate operational planning by simulating various scenarios and predicting their outcomes. For example, urban planners can simulate traffic flow, pedestrian movement, and environmental impact to optimize city infrastructure and urban mobility.
- Risk Management: Simulative digital twins enable organizations to assess and mitigate risks by simulating potential scenarios and evaluating their consequences. Whether it’s simulating natural disasters, cyber-attacks, or supply chain disruptions, digital twins provide valuable insights into risk exposure and resilience strategies.
Added Value through Analysis:
Both monitoring and simulative digital twins offer distinct advantages in terms of analysis and decision support.
- Real-time Insights: Monitoring digital twins provide real-time insights into the performance and condition of physical assets, enabling proactive decision-making and rapid response to changes.
- Predictive Capabilities: By leveraging historical data and predictive analytics, both monitoring and simulative digital twins can anticipate future trends, identify patterns, and forecast potential outcomes.
- Optimization Opportunities: Whether it’s optimizing operational processes, resource allocation, or system performance, digital twins enable organizations to identify inefficiencies, bottlenecks, and opportunities for improvement.
- Risk Mitigation: By simulating various scenarios and assessing their impact, digital twins help organizations mitigate risks, enhance resilience, and make informed decisions in complex and uncertain environments.
In conclusion, digital twins represent a paradigm shift in how we conceptualize, analyze, and manage complex systems. Whether it’s through real-time monitoring or sophisticated simulation, digital twins offer unparalleled insights, predictive capabilities, and optimization opportunities across diverse industries and applications. As organizations continue to embrace digital twins as a strategic asset, the potential for innovation and transformation is limitless.
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
04
Skills and technologies
Learn about the skills and technologies involved in this cutting-edge technology
06
Applying Simulation to the S&OP process
MRP, Scheduling, and in between Simulation Modelling: discover how