Saint-Gobain Construction Products Rus (SGSP RUS) (Nord Clan: ML Sense)

2024: Izover mineral wool quality control at the Saint-Gobain plant

The production of mineral wool is a continuous conveyor line. A feature of the technology is the periodic ingress of foreign substances into the product, or into the so-called mineral wool carpet. As a result, defects form on the minvat. To prevent defective products from going on sale, it was decided to introduce a system based on machine vision and neural networks ML Sense from Nord Clan.

Before the Isover manufacturer applied the ML Sense solution, the control specialist had to visually track defects on the surface of the mineral carpet.

𝓲

Nord Clan

Mineral wool moves along three conveyor strips at a speed of up to 16 m/min in parallel and continuously. The surface is embossed, so it is difficult to closely monitor defects for a long time, literally begins to ripple in the eyes.

Stages of ML Sense implementation in mineral wool production

1. Training of the ML Sense system based on neural networks and machine vision.

We collected a datacet from photographs where each type of defect is marked and classified. At the same time, it had to be borne in mind that the defects are very similar in appearance, but have a different nature of origin, not obvious to a non-specialist. To train neural networks, the ML Sense team figured out the intricacies of production, essentially becoming specialists in the mineral wool department.

𝓲

Nord Clan

We ultimately set the platform's recognition accuracy to 99%. We connected the analyst, the notification system. Adapted the interface for customer tasks.

𝓲

Nord Clan

2. Design and installation of the software and hardware complex, installation of equipment in production.

We selected video cameras, lighting fixtures for more accurate recognition of defects. They were installed on the mast mounts. A feature of this attachment was that the metal shells for the equipment were designed according to their own patented scheme of the Nord Clan. It is important that video cameras are protected from industrial dust and external influences.

𝓲

Nord Clan

3. Development of a unique marking device with a control system.

As soon as machine vision detects the defect, it sends a signal to the control unit of the device. At this moment, pneumatic detectors are activated, which supply air pressure to the sprays. Depending on the location of the defect on the carpet, this or that marker is activated.

𝓲

Nord Clan

4. Start-up of the system and training of personnel.

While working on the project, a team of Nord Clan engineers went to production several times to test the operation of the system. And only after both parties were convinced that the system works stably and without failures, we handed over all the equipment to the customer for operation, trained personnel, signed acts of acceptance - transfer.

Results

What has changed in the plant:

1. Replaced visual inspection with machine vision. Management no longer needs to hope only for the attentiveness and good vision of the quality department operator.
2. Introduced domestic software, which means they solved the issue of import substitution at an industrial enterprise. ML Sense is the result of the work of a Russian company. Included in the register of domestic software.
3. Increased the economic effect. If earlier the plant suffered financial losses due to the return of a poor-quality minvata, now these costs have been reduced to zero.

According to our estimate, based on the experience of implementing ML Sense at other enterprises, now the plant saves from 15 million rubles a year. This economic effect was achieved due to significant reductions in financial losses due to product complaints, as well as by reducing the salaries of quality control specialists.