# A Few More Turns on the Changeover Wheel – Part 2: Improving the Sequence

In my last post I looked at how to create a changeover sequence. However, this was only the first draft of such a sequence. For a truly good sequence, you need to spend some more time optimizing the sequence. Try to get a better sequence, even though it is impossible to find the perfect solution even for a moderate number of products. I also give a suggestion on how to visualize a changeover wheel in Excel.

## Introduction

Just a brief recap, in my last post I created a changeover sequence for an ice cream machine, where you could change the waffle (cone, cup, sandwich), the flavor (vanilla, raspberry, chocolate), and the size (small, medium, large), for a total of 27 products. I sequenced the products starting with the most cumbersome part of the changeover (the waffle), followed by the flavor (light to dark), and finally the easy-peasy change of the size.

## Optimize the Changeover Sequence

Sequencing one variable after another sounds straightforward, but the devil is in the details. In reality there are often more complications. What if the slowest changeover in the second largest variable takes longer than the fastest in the largest variable? What if, in our example, changing from chocolate to vanilla takes longer than changing from a cone to a cup? What if a smaller variable influences the time it takes for a longer variable to change?

Also important: If you change two variables at once (e.g., changing both the waffle and the flavor), is this in parallel or consecutively? In other words, is the changeover duration the sum of the two individual changeovers, or is it the longest of the two individual changeovers? Do you change the flavor while also changing the waffles, or do you first change the waffle and then the flavor?

Even if one variable change is independent from another variable change, there is probably still quite a bit of potential to improve. Below on the left is the basic, non-optimized changeover wheel from my last post. Now compare it with the changeover wheel on the right. It still has all the 27 product types, it still has all the same sequences for each variable, but it has much fewer changeovers. For example, in the first wheel above we changed flavors a total of nine times. In the wheel below, we have only six changeovers for the flavor. The number of changeovers has been reduced by one-third, saving time, effort, and wasted ice cream for cleaning during changeover!

Similarly for the size of the ice cream. In the first wheel, we changed the ice cream size a total of 27 times. In the second, optimized changeover wheel, we changed the size only 18 times. While the size of the product was our least significant variable, it does save some time. This could get even more complicated if one variable affects the effort for another variable. For example, what if changing the waffle type would take longer if it is a large ice cream portion?

Again, your goal is to get the sequence that takes the least time for the changeover effort. To be more precise, it should take the least cost for the changeover effort, including not only time but also energy, people involved, wasted material, and any other cost that goes into the changeover effort. There is an optimal solution out there. If you have some variables where the sequence does not matter or where the changeover time is zero, then you probably have multiple optimal solutions. The challenge is finding it.

The first challenge is to have all the data… which you probably don’t. For optimizing the changeover sequence, you are probably relying on the gut feeling of the people who do this (and please, DO involve the people that do the changeover into the sequencing of the changeover!). This is, for most companies, probably the most sensible approach. You may not get the optimal solution, but hopefully you get something close with a reasonable effort.

Even if you have all the data, it is not easy to find the optimal solution. The number of possible sequences (in mathematical language: the permutations) depends on the number of products you have. For n products, the number of sequences is n!, where the “!” is not an exclamation mark but a mathematical operator for factorial. For example 3! = 1*2*3 = 6. And this number of permutations goes up at an eye watering speed! For 5 product variants, it is still a manageable 120 possible sequences. But for our example with 27 different product types, we could make 27! = 10 888 869 450 418 400 000 000 000 000 combinations. Even a computer that could test one million sequences per second would still take magnitudes longer than the existence of the universe to check them all. And that is for sequencing a measly 27 products.

If you have 100 different product types (which is by no means rare), you have 100! = 93 326 215 443 944 200 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 000 possible changeover sequence combinations. For a measly 172 products, my Excel can no longer even calculate the number of possible sequences. The point I am getting at: You cannot do an exhaustive comparison of all sequencing combinations. No matter if you are human(s) or a computer, you have to take shortcuts to determine the sequence. Don’t aim for perfection, but for something good enough. If you find the perfect solution long after our sun has burned out and the universe has ended, then your changeover sequence is no longer necessary.

## A Note on Visualizing the Changeover Wheel

Creating the changeover wheel is not easy. Visualizing it is not easy, either. Creating multiple overlapping circular graphs is a bit tricky in Excel. There is one chart option called “sunburst,” which looks like it is perfectly suited to create changeover wheels. The sunburst below is a display of a calendar year including quarters, months, weeks, and days.

Unfortunately, by default the Excel sunburst sorts the entries by size, where the largest entries come first and shorter ones afterward. The sunburst below starts in December, followed by October (more days than November). The shortest month February ends the year. In short, the sequence is completely messed up, and I have not found any way to stop Excel from automatically doing this. (Damn you, Microsoft. So tantalizing close to the perfect solution, and yet utterly useless for my purposes…)

To my knowledge, the best way to create a changeover wheel in Excel is to use a doughnut or pie chart, with multiple such charts on top of each other to represent the different variables. It is a bit of work, but doable. There are also custom solutions out there, but I did not look into those.

In any case, don’t get stuck on the wheel. A simple excel table is also perfectly adequate, as long as you remember to start at the top again when you get to the bottom. In this series of posts I simply use a wheel because it is a better visualization. In my next post, I will show you how to properly use a changeover wheel, as there are some pitfalls too. Now, go out, optimize your changeover sequence (but don’t take eons to do so), and organize your industry!

PS: This blog post was inspired by a master thesis by my student Milena Oberle: “Entwicklung eines Steuerungskonzepts zur Stabilisierung volatiler Auftragsfolgen einer variantenreichen Mischfertigung unter Berücksichtigung der zukunftsweisenden Erfolgsfaktoren einer Smart Factory“, Hochschule Karlsruhe, 2022.