OK
From 6 to 8 October, the Gran Via venue of Fira de Barcelona hosts Alimentaria FoodTech 2026, the trade show for machinery, technology and ingredients for the food industry. According to El Demócrata, it will bring together more than 330 companies from 22 countries across some 15,000 m², with 38 start-ups and more than 140 specialists in the programme. The announced innovations include AI-based machine vision systems that locate contaminants and defects on the line, sensors to reduce product waste and water consumption, and cold high-pressure processing to extend shelf life.
Its director, Ricardo Márquez, sums it up in Oleo Revista: at the show, he says, visitors will see how the food of the future will be produced. Vinetur adds digital twins and traceability to the topics, and Interempresas highlights AI applied to predictive maintenance and quality inspection, against a backdrop of staff shortages in specialised roles.
What it means for a packaging plant
What is new is not that people are talking about artificial intelligence, but that it now comes built into specific equipment: a camera that rejects a pack, a sensor that flags waste, a model that anticipates a breakdown. That brings it closer to the plant, but it also means judging it like any other machine: by what it does on your line, not on the stand.
In vision, AI makes sense where fixed rules fail: natural products that are never identical, flexible packaging, foreign bodies of unpredictable shape. But it learns from images, and the right ones are images of your own defects, under your lighting and at your line speed. Before buying, be clear about the quality criteria and measure two things at once: the defects that get through and the good packs that are rejected, because a system that throws away too much product is also a source of waste.
For waste and consumption, a sensor is only useful if its data can be linked to the line, the shift, the SKU and the stoppage that caused it. A water meter or a count of lost product without that link gives you a number, but not a cause you can act on.
In predictive maintenance, the model learns how motors, gearboxes and bearings normally behave and raises an alert when something drifts. It needs weeks of clean data and pays off first on the equipment that stops the whole line when it fails. The sensible approach is to start with those few assets, measure mean time between failures and mean time to repair, and extend once the alert has proven that it arrives in time. Our page on applied artificial intelligence explains how each of these fields fits into a plant.
Five questions to take to the show
- What it was trained on: which images or signals the model uses and how many examples from your own line it needs to work.
- False rejects: ask for the figure in real production, not in a demo, and who adjusts the system when conditions change.
- Product changeovers: what happens with a new pack or format; if retraining is needed, how long it takes and who does it.
- Where the data goes: whether it stays in the plant and how it reaches your systems (OPC UA, Modbus TCP, SQL).
- How results are proven: OEE, waste or avoided stoppages, measured before and after on your line.
AI is already at the show; the useful question is which specific problem on your line it solves and how that will be measured. If you are considering a project of this kind, let's talk.
