Cutting glass using AI

A true story about how KMP Digitata used AI to reduce waste for our client Home Décor.

Here is the challenge – you have a some massive pieces of glass, you have to cut a bunch of different shapes and sizes out of that glass and ensure that there is minimal waste. Add in to that mix you have an ERP system that handles all of your stock of materials and that system has to be the point of truth for what you hold in stock – so if you cut a piece of glass and there is leftover glass that can be used to cut other pieces of glass then it needs to seamlessly go back into stock.

19th century factory workers cut glass by hand Efficient glass cutting has been a challenge since the first Industrial revolution, which I'm sure they didn't refer to as Industry 1.0

The scale of this problem is such it has it’s own Wikipedia page. Where it gets interesting for Home Decor is that there are multiple choices of glass sheet sizes to choose from so the selection of glass sheet and the cuts to that glass sheet are all potential levers that we can pull to minimise waste when cutting the glass. Our lead developer described this as one of the most challenging pieces of code he’s ever written – this is the story of how it developed and the end result.

Where do you start with a problem like this?

As with all technical challenges we are faced with, we start with the people. The outcome here is clear – reduce material waste, and in order to do this we are assisting the manufacturing team at Home Décor in firstly selecting the most appropriate glass sheet to cut and then work out how best to cut the sheet. The team at Home Decor are already doing this and more – we discovered there was also the potential that offcuts of a certain size would be saved and used later at the discretion of the team – they know that they can get further cuts from these offcuts and this is a really great way of Reducing waste. The team also have to take the time to measure and figure out the best cuts to make – this is delivered with varying success depending on the person figuring out the cuts and is something which takes time, slowing down the manufacturing process. By taking the time to understand the process and how things currently worked we understood the levers we could pull to affect the outcome we wanted and then we could effectively augment that process with our little bit of AI (which I’m now referring to as Augmented Intelligence).

The outcomes of us releasing this would be a significant reduction in material waste and to increase the speed of the glass cutting element of the production process.With that understanding we went away and did some really, really hard maths. We then delivered an output that would make sense to the team cutting the glass (a cut sheet) so that we could compare the solutions our code was generating against that of the team currently in place – the goal being it had to be as good as or better than what existed right now in terms of material waste.

Iteration 1 was not hugely successful – we didn’t even release this version to the client, we failed fast and tried a few things out. We did test this with the team and the feedback looked a bit like this: “The planner has advised that manufacturing should consume 1.5 sheets of 4296-37LGB (1070x2440mm) glass to complete this order, whereas the configurator advises 2 sheets of 4260-30LGB (762x2220mm). These are to be used in the preparation of 4 panels of glass @ 656x725mm. This would only consume 1⅓ sheets”.

There were rotations to consider, specific rules around certain glass sheets being reserved for certain kinds of orders, minimum spacings for the cuts, years of knowledge that we had to get the system up to speed on.

Iteration 2 was much better, we had a working cut sheet that we felt could stack up against the existing solution so we released this to the Home Décor team to take a look at and work with in a test environment – we got some key feedback that allowed us to tweak the code to deliver some more effective cuts and the team asked if we could extend the functionality to show how the doors should be put together in a visual format that reflected what the customer had seen.

Iteration 3 was an all singing, and all dancing multiple page cut sheet that included the glass panel cutting. We launched this in parallel to the old way of working to ensure that the solution was robust and that the team who would be working with this could produce the doors more efficiently with the new cut sheets.

It worked! The glass cutting produced less waste and sped up the production process by eliminating the need for someone to have to work out what glass to use and how to cut it. Ironically some of the late additions to the system were deemed to be less useful and our final released iteration simplified things a little.

We released our 4th iteration to live in 2017 and Home Décor have been using this little bit of AI ever since, by applying our methodology of starting with the people, iterating quickly and being prepared to experiment we’ve made a real difference to the production process and saved the planet just a little in the process as there are savings on the material, transport and energy. We’ve also made some of the inputs content manageable too – so that not every change to the size of a glass sheet or door panel requires a developer to re-write code. It may have been a tough challenge to solve, but the project has made a significant positive difference to our client's business.

Side note: We didn’t really think of this as AI until someone else pointed it out to us, this was just a creative, technology driven solution to a human problem and that is what we are best at. We don’t get too hung up on labels in that way.

If you are wrangling with Industry 4.0, need a simple solution to a complex problem or simply have a challenge you’d like a fresh perspective on then we’d love to hear from you.