How we teach the computer to read floor plans
With automated floor plan recognition, we use state-of-the-art technology based on artificial intelligence and self-learning systems. But what does that actually mean exactly? We spoke to our machine learning engineer Fidelius about the development of the new function. He is a mathematician and joined our software team last year. Since then, he has been mainly concerned with the development and optimization of floor plan recognition.
What is machine learning or artificial intelligence?
Artificial intelligence (AI) is the property of an IT system to exhibit intelligent, "human-like" behavior. Machine learning is a specific form of artificial intelligence that can be used to automate processes. It involves training a computer to perform certain repetitive tasks. Through training and correction, the results get better and better - just like in real life. In other words, the computer learns.
Why are you interested in machine learning?
I see incredible possibilities in the process in the future. The topic offers a lot of learning potential for me personally and a lot of application potential for Noocoon.
Machine learning is a multifaceted tool. There is hardly a problem that can't be solved with it. The technology that makes this possible is now being developed for it, and I'm looking forward to working on it.
How does the new floor plan recognition differ from the rest of the Noocoon software?
For automatic electrical planning we use rule-based processes: For example, the customer wants 3 lamps in the room. Based on the customer's request, the software plans all the required components down to the last terminal, because we specify to it with the rules what all is needed to install a lamp.
Floor plan recognition has a different task. It is not intended to plan, but to capture information. Specifically, it is to recognize information in a floor plan and extract it for processing, e.g., to create an installation plan.So we teach the computer to read floor plans. Now floor plans are always very similar and drawn to certain specifications, but it is often the details that make the difference here. Inflexible rules, as in the case of installation planning, do not help here. The computer must learn to recognize certain patterns.
That's why we use the machine learning method, which means we show the computer as many examples as possible and instructions on how it should learn. The software then learns independently and can also adapt the results. Our task is to keep correcting it in the process. In this way, the software gradually learns what a room is, how big it is, where a door or window is, etc.
How is such an AI trained - where are the difficulties?
In order to train such an artificial intelligence, a lot of data and computing capacity are required above all. Each training session lasts several hours and you have to train very often. In the process, we keep changing the conditions of the training until the final result is satisfactory. This requires a lot of research to optimize the training. The software can already recognize spaces, but it will soon be able to do even more. It is more or less like an apprentice who can already do some things on his own, but who cannot yet go to the construction site on his own.
Will electricians then soon be replaced by an artificial intelligence?
No, of course not. Such technology should be understood as a tool. In the past, there were screw terminals, for example. Today, plug-in terminals, such as the wagon terminals, make work on the construction site much easier. So with AI, the electrician makes a certain work process easier and shorter.
