Production Simulation Alternatives

A fundamental goal of production simulation is to make predictions about how production capacity and flow behavior influence factory performance.

The most common tool to achieve this task today is discrete event simulation, which is available in a variety of software packages. Popular brand names include AnyLogic, Arena, FlexSim, ProModel, Simio, Tecnomatix, and many others.

Discrete-Event Simulation

In any of these software tools, simulating the behavior and dynamics of complex production systems involves Monte Carlo methods. Monte Carlo means that the simulation randomly plays through possible sequences of events hundreds of times (sampling) to statistically approximate the overall system performance. Using sampling, problems can be solved numerically that are too complex to be solved analytically, as had long been the case for production system dynamics. Popular approaches include discrete event simulation (DES), continuous simulations that use small time steps to loop through non-discrete changes (also known as discrete-time simulation or DTS), and agent-based models which may be based on either DES or DTS. Users and researchers have discussed the pros and cons of the different techniques: DES requiring more time and suffering from inaccuracy when utilizations get high; DTS having lower accuracy in general, especially with low utilizations, and the challenges of deciding on the right time step. Both DES and DTS require capacity, like machine count and kanban count, as inputs.

"Simulation of complicated systems has become quite popular. One of the main reasons for this is embodied in that word “complicated.” If the system of interest were actually simple enough to be validly represented by an exact analytical model, simulation wouldn’t be needed, and indeed shouldn’t be used. Instead, exact analytical methods like queueing theory, probability, or simple algebra or calculus could do the job. Simulating a simple system for which we can find an exact analytical solution only adds uncertainty to the results, making them less precise."
Chapter 1.4 of: Simio and Simulation (Jeffrey S. Smith and David T. Sturrock)

To date, discrete event simulation is the most common way to model production operations and various software packages are available, some of which have detailed 3D animations to show the physical movements of goods and people across the shop floor. Often, DES software is used in conjunction with other tools to help generate different scenarios and sets of inputs in order to plan and optimize new production systems. A big limitation of Monte Carlo methods in general is that each run (hundreds of iterations) just produces a single point answer, i.e. the performance of a previously defined system: there is no inherent information about cost drivers and no optimization other than through trial and error. DES-specific optimization tools exist, but they are very limited (more on that below).

Analytical: Queueing Theory

The analytical methods known as queueing theory, or queueing models, have been around for much longer than simulation methods, and are, by contrast, a mathematical way to model dynamic part flow behavior. Traditional queueing models have been the "gold standard" and thus the reference point for accuracy validation of simulation methods, but have traditionally been limited in the complexity they can capture. There were no models to capture reentrant flows, as would be required for rework, same with batch processing. That's one of the reasons queueing models have not found much adoption in industry. Another reason is the difficulty of choosing and then implementing the correct mathematical expression for each configuration. The scientific community has developed a few methods to simplify the mathematical handling, such as the decomposition method or the queueing network analyzer method, but these did not overcome the limitations regarding system complexity. In terms of required inputs, both methods have the same input requirements and yield the same outputs as discrete-event simulation: resource counts are required inputs to test if a system is capable of making rate. However, they are more precise and repeatable, as the same inputs will always produce the same result. Today, queueing models are not commonly used outside academia, at least there are no off-the-shelf tools available that would facilitate broad adoption, even for the rather narrow use cases they can capture. But certain queueing models, e.g. the M/M/1 queue (a simple system with 1 machine), remain an important tool for validating the accuracy of other methods.

LineLab

LineLab is an alternative to simulation, originally based on analytical approaches to modeling production. LineLab is a software tool that uses an analytical approach to capture factory dynamics. Like previous analytical approaches, it's highly accurate. How? LineLab is partly based on queueing models, but the framework was extended by research conducted at MIT. In LineLab, a single solve can capture the entire production system, including anything flowing in and out, in a single system.

To create LineLab, queueing models were not just rewritten, but extended quite significantly, so LineLab can handle aspects of part flow that are common in industry:

Re-entry flow (e.g. rework, split routing)
Feeder lines (determines the "critical path", which is generally stochastic due to effects of variation)
Shared systems with different products and different routing, that are sharing some workstations (and thus queues) and/or part carriers, co-produced variants
Batching
Pulsed and flow lines
Nested cells, e.g. station has one loading device and multiple chambers; and parallel processing within a cell

One key difference to previous approaches is that LineLab internally uses an optimization solver, which can directly handle its analytical models of factory flow. This is a key differentiator, that often turns weeks of going back between different simulation models and spreadsheets, into a single solve with LineLab.

For one, the system configuration does not need to be an input. If the desired throughput is known, LineLab can optimize capacity and kanban count as well as many other values to find the best way to achieve throughput. Verified with 1000+ processes and 500+ equipment types in a single model LineLab, can co-optimize workstation counts and other parameters throughout, in a single solve. Whereas there are optimizer packages to work alongside discrete event simulation that can optimize 5-15 variables with varying levels or robustness, LineLab can simultaneously optimize thousands of decision variables and, in contrast to DES tools, does not need starting values or bounds. Moreover, any input can be parametric. This is particularly useful in early stage design or in model based systems engineering, when physics-based models need to be considered. Also, the optimization allows for adding arbitrary constraints, like a maximum flow time between any two stations - unprecedented compared to any simulation method. And finally, thanks to the optimization piece, LineLab produces a complete sensitivity analysis that quantifies the role of every single input, making it the most transparent way to model production and operations systems.

Conclusion

There are different methods to model and simulate the performance of planned production systems. Many of the well-known software tools are based on Monte Carlo methods, as this has long been the only way to solve such complex systems. With their 3D animations and daily schedule examples, the well-known software tools will continue to find important applications. LineLab is a comparatively new product based on analytical methods. For the optimization of various variables, such as capacity, in early system planning, LineLab is indispensable.