At MOSIMTEC, we often say our job is to answer the questions our clients are not asking. For example, a distribution client came to us confident their planned goods-to-person (GTP) system could meet expected throughput, but they were concerned about storage capacity. Because they thought storage capacity would be a problem, they really wanted to understand the right mix of SKUs to house in the ASRS. By asking the right questions and building a valid simulation model, we were able to find the throughput was going to be a problem, despite what initial static spreadsheets predicted. The customer was able to play with operational strategies in the simulation model to overcome the equipment design shortcomings before the system went live.
An introductory view of simulation modeling may say the client asks questions, tells us what to model, tells us what simplifying assumptions we can make, tells us what reports they want; and we program. The reality is that your simulation consultant should be completely in tune with your goals and objectives, so they can:
- Ask you the right questions
- Lead you to questions you didn’t even know you had
- Advise you on the best dashboards to understand system behavior
- Provide an in depth understanding of the system being modeled
- Define why some areas are likely to benefit from simplifying assumptions and other areas need more detailed modeling
- Advise you on novel ways to solve the problem at hand
One reason simulation often gets skipped is because you already have really smart people working on really hard problems. Your people may be working on robotics or vision systems, which are inherently very technical in and of themselves. It is often easy to forget that the overall system performance is a different problem that may need a different mindset to analyze.
A Simple Example
If I asked a brilliant mechanical engineer or software engineer how many pieces of equipment is needed to process 250 batches per hour, when 1 piece of equipment can do 25 an hour; many often confidently respond that the answer is 10.
If you did, you are not alone, Chat GPT answered 10 when I asked this question. This is the wrong answer a lot of the time, but it is easy to be confident in this wrong answer.
Why is this wrong most of the time? Most systems have variability in processing times, arrival rates, equipment reliability, blocking and starving from upstream and downstream resources. 
In a simple system, like the 250 batches per hour example, someone with system design expertise may realize they need to use queuing theory and add some additional downtime factors to correctly size the equipment. Is the mechanical engineer or software engineer a bad engineer if they forgot to add extra capacity to account for variability? Of course not. They were busy focusing on their area of expertise and providing value in a very effective way in that area. If you ask ChatGPT several rounds of very specific questions on a simple queuing system, you will eventually get a more valuable answer.
The right question was not “how many pieces of equipment do I need?” The right question was “How many pieces of equipment do I need to meet my desired system performance?” Which then brings up the questions like:
- What is your desired throughput and maximum wait times?
- Do you have limits on your queue space in front of the operation?
If the system is simple enough with limited variability, you may still be able to answer the question in a spreadsheet or with some basic queuing theory formulas. However, what happens when our system is slightly more complex?
A Slightly More Complex Example
Let’s say I have 2 positions in a robotic workstation. I want to process 225 units per hour. Each part may require some variable time due to customer requirements, but the averages in the 2 positions 1 and 2 are 1.2 minutes and 0.9 minutes respectively. The stations are expected to fail occasionally, causing a pause in processing.
The quick math that many smart people may do often looks like this:
Math is straightforward and makes sense, so the facility or equipment design continues with these numbers. The design team is also extra confident, because they are fairly sure of the average processing times at each station. Because you cannot have a partial station, the system has some built-in excess capacity at each station, providing even greater confidence that this system will work.
However, if we built this machine, we could have an infeasible system. The queue of parts in front of the first station could grow infinitely over time.
You can play with this system yourself with a simple AnyLogic Cloud model, which allows you to adjust several system conditions, such as average processing times and percentage of time failed. You can also adjust the capacities of the 2 processing stations and experiment with adding a buffer between the 2. Try the model yourself below. After hitting the play button, you can adjust capacity and other settings via sliders and checkboxes.
Interactive Simulation Model
By playing with this simple simulation model, we find the system can stabilize with an additional unit of capacity at each resource.
A buffer between the 2 stations also helps the system by reducing the times the first station is blocked by the 2nd station. The buffer between the 2 stations also spreads the items in queue to 2 locations, instead of 1 mega queue in front of the first station.
This alternative design is feasible, but significantly different from the original, spreadsheet-based plan. Many system designers may be surprised to learn that they need over 30% excess capacity just to account for variability.
In the cloud model, we are looking at an extremely simple system. What if this was a system with:
- Significantly more stations
- Routing between stations that varied by order type
- Required shared resources, like operators, to setup batches
Even the smartest engineers would struggle to understand if the system could meet required throughput and order turn-around-times without a simulation model. At MOSIMTEC, using simulation to design systems with complex interactions and variability is what we do. Let your people focus on what they do best and rely on the experts in simulation modeling. If you know your really smart people want to do simulation modeling, MOSIMTEC can also provide training and mentoring to get them to be world class modelers as quickly as possible.
About the Author
Amy Brown Greer (amy.greer@mosimtec.com) is a Principal Simulation Engineer at MOSIMTEC with over 21 years of experience as a simulation and process improvement consultant. Amy has delivered solutions for the transportation, healthcare, distribution, manufacturing, and service industries. These solutions and recommendations are based on simulation model results, statistical analysis, mathematical modeling, and other traditional Industrial Engineering tools. Amy believes simulation and modeling technology should contribute to improving an organization’s overall performance and a good consultant must take the time to understand the customers’ systems and objectives before diving into model development.
Prior to joining MOSIMTEC, Amy led the simulation efforts of the American Red Cross Biomedical Services. She was also a simulation consultant for 6 years at TranSystems, which included being the organization’s simulation practice area leader. Amy holds a B.S. in Industrial Engineering from Tennessee Tech, and a M.S. in Industrial Engineering from Virginia Tech. She is a registered professional engineer, has served on the PE exam development committee, and is a Fellow of the Institute of Industrial and Systems Engineers (IISE).