Sorting: Avoiding Waste by Detecting Foreign Material

Advanced detection of foreign material is of paramount importance for potato product manufacturers, to guarantee food safety and to preserve brand equity. But a second, equally important role a well-tuned sorting process accomplishes is to improve yield by minimizing wasted good product that is removed in error.
Product presentation makes a big difference to a sorter’s efficiency. Selecting the ideal infeed with added functionalities such as dewatering, de-oiling and aligning, in addition to spreading, singulating and monolayering, enables the product to be prepared and presented to the sorter’s inspection zone in a manner that maximizes sort performance.
Improving Good-to-Bad Ratio
Karel Van Velthoven, advanced inspection systems product marketing manager at Key Technology, explains that there is a natural trade-off between a sorter’s defect removal rate and good-to-bad ratio – the objective is to improve both simultaneously. A more aggressive sort results in more false rejects, and a less aggressive one leads to more FM and defects passing through the sorter’s ‘accept’ stream. Given this inherent trade-off, removing FM and major defects always gets priority over losing good product to the sorter’s ‘reject’ stream.
Another way to reduce waste, according to Marco Colombo, global category director potatoes at TOMRA Food, is using the sorting data to optimize upstream processes.
“The data generated by our sorting solutions can also be used for an upstream adjustment of the potato peeler, proactively managing raw material infeed and steam times according to the expected product specifications. In other words, we can depict the ideal potato profile going to the peeler, while at the same time optimizing its steam time.”
The very definition of ‘good’ and ‘bad’ with respect to quality is a hurdle, as it varies over time, explains Rasmus Andersen, head of software, Newtec Engineering.
“With highly customized sorting solutions that have many outlet configurations, we focus on improving the precision of the quality rather than the accuracy, to achieve consistency in all outlets. To address this, we have improved the optics in our cameras, which allows us to enhance the identification of the different defect types.”
As reported by all industry players interviewed for this article, the way forward in improving sorting precision lies with AI and machine learning developments.
“Even when using the most advanced methods (hyperspectral cameras combined with RGB cameras) there are cases where perfect classification becomes complicated. The target is to find the best compromise between safest possible rejection and lowest possible false reject (safety vs. yield) in the classification. Gathering more information per object and smarter analysis of the gathered information will be the way to achieve this,” representatives of InSort point out.
You can read the rest of this article in the Summer Issue of Potato Business Digital magazine, which you can access by clicking here.