I was very fortunate to be able to work on this project at the Fraunhofer Institute for Organic Electronics, Electron Beam und Plasma Technology in Dresden.  The project was about retrofitting a slitting machine into an existing roll-to-roll coating process line. I also wrote my bachelor thesis about this project. The thesis ended up having over one hundred pages, which is why this is just a very brief summary that leaves out most of the thoughts, boundary conditions and small details. The project is a great example of how I approach complex problems and optimization.

The picture above schematically shows the process line. For reference, the entire length of the line is about 15 meters.


The existing roll-to-roll process at the Institute lacked a possibility to separate the processed webs into several webs of smaller width.  The processed webs had to be cut in longitudinal direction. This had become necessary because of unforeseen issues with different other processes. Furthermore, because of customer requests, the team wished to be able to freely alter the webs width in-line.

I was the only person working on the project, so I had to self-manage everything. The strategy I started out with was to buy components from existing suppliers that sell entire slitting and cutting machines for the industry and implement them myself. It turned out that almost all of these companies are poorly run so I was only able to generate a couple of options ranging from 2-20k euros just in parts with multiple months delivery time. I then developed my own concept that cost around 300 euros and fully met the specifications, which with little doubt most of the other systems would not have achieved. 



My approach to building anything is shaped by our understanding of reality as being made up by quantifiable building blocks (see Personal page). There is no way around doing the math. I do this in order to be less wrong or to have the solutions that I develop be less suboptimal. Below is explained how I successfully solved the “Web Slitter” problem. Since the approach that I choose to use is so universal, all of it can also be applied to almost anything else.

Broadly speaking, every creation or product can be understood as a (very complex) function. There are certain sets of quantitative value (shape, material properties etc.) that can be altered to inflict change in a set of other quantifiable variables (i.e. utility). Only because there is a causal link between these sets, engineering is possible. In the case of building a slitting machine, it is mainly about arranging matter (input) in an optimal way that “scores highest” in the boundary conditions (sets of outcome). The goal of the web slitter is to provide one or more longitudinal slits in the web at the lowest cost. So in order to build the best machine possible, I had to find the optimal arrangement of particles that provide the optimum slit quality while having the lowest cost. Below is an in-depth explanation of how this very abstract non-linear and discrete optimizing can be applied to real-world problems.

In order to come up with the best solution possible I had to either creatively develop a large number of different concepts, take the one with the best cost-to-performance ratio and optimize it further (numerical, Newton-Raphson method), or to reason towards the best solution (analytical method). This approach can be visualized taking a way simpler function like -x2+1

The analytical approach uses logic to asses true statements and the maximum of the function. For -x2+1 could be argued that the function is symmetric and therefore the maximum has to be at zero. An other way to discover the functions maximum value is the numerical or iterative approach. This approach takes one or multiple random points of the function and changes the argument incrementally to increase the functions result value. In the case of the web slitter I had developed numerous concepts and ran them through a model that I had created to test them. The model included mostly data that I had found in various literature. Some concepts could be real-world-tested through prototypes. That way I was able to derive the value and cost of each concept and moved on with the best concept.  For -x2+1 one can creatively pick valid sets of arguments like (-1,0). Since the goal is to achieve the maximum value or score, one would have to optimize the quantitative sets in different directions. Because of -x2+1 being a two-dimensional function there are only two options to alter the argument, either by decreasing or increasing it. In the case of the web slitting machine the amount of variability or dimension is given through the number of atoms or particles that can be arranged times the number of different available particle options for these.  If one were to optimize the argument for the function-x2+1 using the iterative approach, one would have to keep repeating taking two points (higher and lower than current), say (-1.2,-0.44) and (-0.8,0.36) and continue with the point that has the highest value while also decreasing the arguments distances between the steps. 

It is obvious that using logic to safely derive that a functions maxima is the method to choose since it provides the exact maxima. However, often functions are too complex or have other twitches that make it not very economical to reason to the best solution (i.e. wiki TSP). For those solutions an numerical approach is needed. 

A simple simulation was conducted to make sure that the displacement would be tolerable.

Although I do create something entirely new, because the boundary conditions are defined externally, the optimal result is predetermined and just has to be discovered. Because of that, building or optimizing something feels to me very mechanical.  


Although I managed to implement a very good solution, I have to admit that I did a couple of mistakes in the making. The biggest mistake actually involved my approach. I had given little consideration to my intuition and midway through the project my method had become a little wacky.  As explained above, I try to minimize the degree to which I use intuition because it just is error prone. Intuition is the neural network that is our brains cortex that solves functions through a learning process. That means that by definition it is completely wrong at times and almost always wrong to a certain degree. I however came up with solutions that were very non-intuitive. If something is totally non-intuitive, one has to be extra cautious with the found solution. I made the mistake to not mistrust the solutions that I had developed because I valued my intuition too little. I was lucky to be supervised by people that helped me spot and fix my mistakes.  That also taught me a lot on pushing through projects especially when it feels as if the job is almost done.

I think that I made a good job managing the project and especially my tasks. However, because I was very focused on doing whatever had the highest priority at that moment, I often realized when I had wasted my time optimizing something that turned out to be obsolete later, so I definitely learned from that.

I am very thankful for the opportunity that the Fraunhofer Organization and especially my supervisor Dr. Günther have given me. I have to really thank all members of the institute that I worked with for providing the very best support possible.