Before the wash, AI sorts what came back

Hotels sometimes return linen bundled inside a tightly wrapped sheet. At Nor Tekstil’s Ålesund laundry in Norway, that creates a job before articles can be recognised: open the bundle. JENSEN describes a dedicated bundle breaker that prepares these arrivals for robotic separation.[1]

It is a useful introduction to industrial laundry AI. The incoming material has to become a sequence of identifiable, inspectable articles, then batches the washing line can use. Cameras, radio tags and X-rays answer different questions along the way. Their value comes from connecting those answers to the machinery handling the fabric.[4]

Three questions before an item joins a batch

A tag can identify a garment without revealing what was left inside it. An ordinary photograph describes the visible textile; an X-ray offers a view of its contents. Knowing where a shirt belongs and deciding whether it can continue through the process are separate tasks.[4]

Buttons and zips belong in the training set

Inwatec software engineer Martin Jakobsgaard describes neural networks trained with human-labelled imagery. ODIN learns from examples of components that should pass, such as buttons and zips, and objects that should cause rejection, such as knives and scalpels. The classification problem is therefore more specific than detecting metal: many legitimate parts of a garment contain it.[2]

HEIMDAL has a different training task. It recognises article types using visible characteristics such as colour, pattern, texture and size, with training adapted to the laundry’s product range. That makes the customer’s actual mix of linen and workwear part of the installation work. A useful category is one the receiving laundry needs to distinguish.[2]

In a separate technical account, Inwatec engineer Tudor Morar explains where X-ray interpretation becomes harder. Thin plastic and paper can resemble the textile in the image; polyester itself is a plastic. A previously unseen object is more likely to be caught when it resembles examples the model has learned. These are material and training limits of the described system, rather than a universal detection guarantee.[3]

Sorting becomes a physical production system

JENSEN describes Ålesund’s four THOR robots as having a combined capacity of at least 4,000 pieces per hour, serving 32 main categories and ten lower-volume categories. This is capacity specified for the four together, rather than measured shift output.[1]

A second operating example, La Nuova in New Zealand, shows where the sorted material goes. Inwatec’s customer account describes robot separation, camera and RFID identification, and ODIN inspection feeding a bag system. Sorted batches can then move into the tunnel washer. The machinery turns classification into a prepared load for the next process.[5]

La Nuova managing director Brad Craig told the supplier that the business wanted to remove unpopular physical tasks without using the investment simply to dismiss people. He reported positive staff feedback. That is a named customer’s account of the change in work; the published case gives no measured percentage reduction in labour or increase in output.[5]

A buffer can be as useful as a faster robot

Ålesund’s overhead buffer can hold up to 30 tonnes of sorted linen, according to JENSEN. The automated system can continue preparing material for the next day’s washing.[1]

Capacity and storage solve different production problems. A fast sorter can still leave a washer waiting for suitable linen. A buffer gives the laundry some freedom to prepare batches earlier and release them when needed.[1][5]

Report an error or suggest a correction