Tomra uses deep learning for wood recycling

Tomra Recycling is using deep learning, a subset of artificial intelligence, in wood recycling applications. 

The company has combined its AUTOSORT technology with its deep learning-based sorting add-on, GAIN, to create a solution that can distinguish between and sort different types of wood-based materials.  

The primary application for Tomra Recycling’s new solution is sorting Wood A – non-processed wood – from Wood B – processed wood products such as MDF (medium-density fiberboard), HDF (high-density fiberboard), oriented strand board (OSB) and chipboard.  

Tomra Recycling’s X-TRACT solution quickly became popular with chipboard manufacturers to produce a clean recycled woodchip fraction by sorting and separating out the inert material (glass, stones, ceramics, etc.) and metals. Once the X-TRACT unit has removed these impurities, the recovered woodchip is of sufficiently high quality to be used in the production of standard chipboards.  

Higher purity 

In recent years, however, Tomra Recycling has been approached by an increasing number of customers who are looking to use recycled wood of a much higher purity level in their production processes. To achieve these specific purity requirements, in addition to removing the inert material and metals in the infeed stream, other impurities including engineered wood composites as well as polymers, would have to be removed.  

As these materials are not distinguishable using x-ray technology, the X-TRACT unit was unsuited to this sorting task. Tomra Recycling’s deep learning experts developed an application that combines Tomra’s AUTOSORT unit with its deep learning-based sorting add-on, GAIN.  

Tomra’s Wood A vs Wood B application uses deep learning technology to sort and extract impurities that couldn’t previously be detected, making it possible for the first time to detect, analyze and sort every different wood type, therefore cleaning up the real wood fraction.   

Tomra is the first company in the world to use deep learning technology to detect and separate different wood types, targeting Wood B (processed wood composites) as impurities to leave a clean Wood A fraction (non-processed wood), or, depending on customers’ requirement, producing individual high purity engineered wood composite fractions out of the infeed stream.