Login
How good is your DC simulation model?
In an effort to boost performance, distribution centers everywhere are in a constant cycle of upgrades, adding new technology and material handling equipment. The process is necessary to meet the demands brought on by rising e-commerce volumes, escalating customer expectations, soaring product return rates, and labor shortages.
But it can be difficult to predict how each new upgrade will affect the operation of the system as a whole. So many companies use computer simulations to model the impact of each change on the total workflow, allowing them to study changes virtually before making them physically.
Industries such as automotive manufacturing have used this approach for years to avoid expensive downtime, bottlenecks, and shutdowns. And now the e-commerce sector is adopting the same tactics as retailers, brands, and third-party logistics service providers (3PLs) struggle to keep up with consumers’ demands for faster fulfillment.
But as logistics practitioners embark on these DC simulation exercises, many are finding that they need to apply new strategies in building their simulation models to account for the pace of technological change in material handling automation, says Abhineet Mittal, research science manager for simulation and digital twins with the e-commerce and logistics giant Amazon’s worldwide design engineering team.
“The DC space is changing faster than people can keep up,” Mittal says. “For example, there used to be a rule of thumb that this length of conveyor should work for this volume of throughput. But now, systems are not as simple; robotics and automation go in, so you have a lot of data that goes with the system. You have to do simulations first to be sure.”
In designing its own facilities, Amazon addresses that challenge by creating models composed of multiple independent parts, treating them like the separate ingredients of a cake instead of a blended batter mix.
Mittal calls that approach “modular discrete-event simulation” and says the strategy works because automation doesn’t happen all at once throughout an entire fulfillment center, but step by step for certain tasks. “People used to build monolithic [simulation] models, covering everything from input to output. But now you need modularity within the model, so you can switch out certain modules to see where the problems are and then adjust them,” he says.
That strategy also allows DC designers to identify exactly where problems are likely to occur and swap in a different part virtually, before having to make expensive, last-minute changes on the warehouse floor. “Bottlenecks don’t typically happen within a single automated process, but where two modules meet. Think of it like Lego pieces; using the simulation, you can exchange one and see if it affects operations in a positive or negative way,” Mittal says.
Building modular simulations has also helped Amazon avert potential equipment compatibility problems in an age when most DCs contain automated devices provided by multiple vendors or systems integrators. Those diverse devices often feature different computer interfaces that might not exchange data smoothly unless they are carefully modeled first.BUILD YOUR OWN MODEL
Building a complex model of your distribution center typically requires an expert IT (information technology) staff or a hired consulting firm. That is particularly true for organizations seeking to build a digital twin of their operation, a type of simulation that includes real-time data inputs from the automated systems in the model, provided by links to the programmable logic controller (PLC) chips in the machines themselves.
But at a simpler level, Mittal says companies can build their own basic simulation models using commercial software applications, such as FlexSim or Simulink. “The [developers of these applications are] trying to make modeling as simple as possible. You can use drag-and-drop features and graphical user interfaces (GUIs) and connect it all in a process flow. Anyone can create a simple model in 30 minutes,” he says.
According to Mittal, the key to success is using modular architecture, which allows users to create faster builds and make updates more easily. In that regard, he offers the following tips:
- Organize models as interchangeable, parametric modules with well-defined interfaces;
- Reuse validated components to reduce rebuild effort and accelerate new scenarios;
- Swap or update modules to adapt quickly to design changes without touching the core;
- Scale complex systems by composing from simpler submodels and support hybrid paradigms;
- Enable collaboration via concurrent module development and isolated testing.
Those principles work not just for logistics-focused applications but also to answer questions for a wide range of business sectors. And that can improve warehouse models as well, he says, because they help provide lessons on building an effective model—for example, by guiding users through the process of thinking through such questions as how you’ll get people out of the building in case of a fire, how do you get a large crowd into a stadium, and what’s the best way to stock shelves in retail stores.
“So there’s potential for everyone to learn more. The more you know about [any process], the more you can predict what’s going to happen and how to improve it,” Mittal says. As an example, he cites the 1999 Hollywood movie “The Matrix,” which describes a dystopian future where humanity has been tricked into living in a digital model of reality. “There’s a saying, ‘The Matrix isn’t real, but simulation runs the world,’” he says. “And that’s becoming more and more true every day.”
