Application d’un réseau de neurones artificiels auto-encodeur pour la détection de bourrages sur tapis convoyeurs en centre de tri de déchets
Résumé
In France, household waste is collected and sorted in waste sorting centers for selective waste collection. From 2005 to 2017, the amount of recyclable waste increased by 35%. By 2026, the Organization for Economic Co-operation and Development (OECD) has estimated that the amount of plastic waste worldwide will have tripled. And, only less than a fifth will be recycled and half will go to landfill. Sorting centers encounter a major and persistent problem in the sorting of household waste that has not been solved, conveyor belts jamming. The incoming waste flow is of variable composition and consistency (material, shapes, humidity, entanglement of waste, etc.) resulting in significant jams. Detecting potential jams is therefore necessary. An autoencoder artificial neural network is tested to detect potential jams on conveyor belts in waste sorting centers. The method and the results on real data are detailed.