NPP leverages TOMRA deep learning technology to purify production of rPET

Nord Pal Plast is a company specializing in the production of recycled PET flakes derived from post-consumer plastic bottles. The subsidiary of the European group Dentis was established in 2003 and is based in Lesquin, France. The company was the first to test a new deep learning application developed by TOMRA Recycling Sorting that enables the separation of clear and light blue transparent bottles from hard-to-detect white opaque bottles to create a clean PET bottle stream.

Regulatory change drives innovation at sorting facilities

Under the Circular Economy Law, France has set an ambitious target of recycling 100% of plastics by 2025. To achieve this, consumers nationwide received instructions to sort household packaging waste into separate paper/carton, metal, and plastic bins. To meet the demands of these regulatory changes and increase the volume of recycled plastic content available to the market, Nord Pal Plast (NPP) initiated a project with TOMRA Recycling Sorting in 2022.
Amed Tuwi, Application Developer - Deep Learning at TOMRA Recycling Sorting, and Alexandre Cliche, Quality Manager at NPP
Amed Tuwi, Application Developer - Deep Learning at TOMRA, and Alexandre Cliche, Quality Manager at NPP
Even with conventional sorting technologies, creating a pure recyclable PET fraction was challenging due to hard-to-detect plastic contaminants, such as multilayer packaging. Milk bottles in France, for example, are often composed of multiple layers of different colored plastics, with the outermost white layer containing titanium dioxide that provides UV protection. While producer responsibility organizations like the French Eco-emballages have called for reducing opacifying colorants in PET (2) these materials are abundant in waste and need to be separated from the target fraction using the most advanced sorting technologies.

​Working in collaboration with the research and development engineers at TOMRA Recycling Sorting, a new deep learning-based application for purifying PET was tested on the sorting lines at NPP. TOMRA software engineers trained the AUTOSORT® with GAIN™ deep learning technology to detect opaque PET objects, foils, textiles and films that are considered contaminants when producing rPET. The machine’s powerful combination of deep learning software and cutting-edge sensors now makes it possible to create a mono fraction of transparent PET with outstanding purity levels.

“With our latest GAIN™ application, we have targeted the separation of white opaque PET from clear PET, which was previously difficult to solve with conventional technology”, explains Amed Tuwi, Application Developer - Deep Learning at TOMRA Recycling Sorting. As project leader for the new GAIN™ application, Tuwi commented: “NPP is a highly recognized and future-forward player in the industry –they were a great collaborative partner to test our new PET cleaning application on an industrial scale.”

Building a competitive edge with technology

In operation since 2003, NPP is a PET bottle recycling company with a processing capacity of 40,000 metric tons of PET bottles per year. Specializing in clear and colored PET bottle recycling, NPP has used TOMRA sorting technology for over 12 years. The company’s sorting plant in Lesquin features six AUTOSORT® machines, two of which have been equipped with GAIN deep learning technology.
“Not only has deep learning technology significantly improved our purity levels of clear PET, but it also allows us to expand our operations and create new revenue streams. With this new application for PET, we can now recycle white opaque PET bottles with UV barriers and easy-to-separate body sleeves – items found in supermarkets nationwide”, explains Alexandre Cliche, Quality Manager at NPP.

The plastic sorting facility uses one AUTOSORT® with GAIN™ technology to purify the PET bottle input stream from contaminants. The second machine with deep learning technology then recovers recyclable PET bottles from the first sorting step. This ensures that NPP recovers as much material as possible for recycling, maximizing yield.

"The new deep learning application for PET halves the final residual contamination, and we are now achieving 100 ppm (parts per million) contamination. What's more, this is achieved with a minimum number of machines – an investment that really pays off when it comes to operating costs”, comments Frédéric Durand, Managing Director of TOMRA France.

Mixed PET fraction prior to sorting process.
Mixed PET fraction prior to sorting process
After successfully implementing TOMRA’s deep learning technology to achieve higher purity levels of secondary raw materials, NPP has been selected by Citeo in its national tender as a trusted industrial partner to operate the recycling of PET bottles.


​1.TOMRA Recycling Sorting designs and manufactures sensor-based sorting technologies for the global recycling and waste management industry to transform resource recovery and create value in waste.

2. Eco-emballages. (2017, February). PET opaque: le programme d’actions d’Eco-Emballages pour 2017. Retrieved from Eco-emballages: