Everything we have learned from our production by creating an Artificial Intelligence

Data, data, data. Any artificial intelligence bases its effectiveness on handling a huge amount of information. So do the algorithms we are training in the Teaming.Ai project. And that has had a singular effect for Industrias Alegre. Because, by working for the machines, we have deepened our own knowledge. And this is a valuable breakthrough for our company that puts us a little bit further on the road to automation.


As we explained a few months ago in this article, our company is involved in the European Teaming.AI project, a research developed together with fifteen partners from eight different European countries. The proposal is aimed at achieving an artificial intelligence capable of predicting the deviations that usually occur in plastic injection molding machinery and thus avoid wasting time and resources.

A paradigm shift by measuring our own data

In order for this artificial intelligence to be able to make its predictions reliably, we have had to first tell it how production is being carried out and then point out to it which parts have been OK and which have been labeled as incorrect. It is also given a startup protocol. And the system, by analyzing the parameters with which those parts were injected (far more than a human being could consider), draws its own conclusions and, over time, is able to predict possible errors and warn the production plant before they happen.

"This approach has meant a change of mentality for the entire Industrias Alegre team," says Alejandro Espert, head of Process Engineering and coordinator of the project in our company. And he continues: "Does this mean that we didn't measure before? Not at all. It's just that our quality methodology involves making empirical measurements of what is happening. And when we find that deviations are occurring, we make the appropriate corrections because we know that the wrong parts are going to be produced. But it's always empirical. And, although a lot of measurements are made, we don't have real-time data because no one is constantly measuring at the machine and that's why, occasionally, the errors occur."

Now, when working on the project, it has been necessary to study the entire production process in depth and analyze many parameters, those that we used to measure and those that were not considered. Espert gives us an example: "Until now, we would open a mold and it was enough to take a look at the part to know that the process was going well. But in order to provide data to the Artificial Intelligence (which is our fundamental task in the consortium) it has been necessary to install a thermographic camera that measures the temperature of the part as it comes out of the mold. What temperature is it, 285 degrees? Well, we know that 285 is the right temperature. So, we have worked for the machine, but now we also know it ourselves.

From intuitive knowledge to scientific knowledge

And in the measurement, we had the second big challenge: determining how to collect the data.

"It was a very unique task because we had to measure temperatures, pressures, and so on. And that has meant asking a lot of questions: Do we need to know what temperature the polymer is at inside the chamber? If so, we must put a sensor inside the machine, but: what kind of sensor should it be, where is it going to be housed, how does it work, how does it connect to send data?"

What to measure, how to do it, how to communicate it... In other words, we have gone from having a handful of data and relying on the great intuitive knowledge of the expert personnel working with the machines to consciously accumulating significant knowledge of our entire production.

"It is knowledge that we had largely hidden, and we have been able to extract it by participating in this project," says Espert once again. And Amparo Vázquez, our R&D&I manager, confirms this: "This process now allows us to have scientific data and to move more solidly towards automation. We know that in the end it is the human being who decides in the factory, but now we have more objective data to control the whole process".

The Teaming-AI project is still moving forward, now in the hands of those who are creating the software. But for us it has already had the enormous value of getting to know us a little better, nurturing a know-how with many decades of experience and allowing us to move forward in the field of quality and innovation.

You may also be interested in