By 2060, the estimated amount of worldwide plastic waste will triple and half of it will be sent to landfill. Yet, waste sorting plants that treat this waste face an unsolved recurrent main problem that is the occurrence of jams in conveyor belts. These jams limit the quantity of waste sorted, whereas waste sorting plants have a constraint to sort a certain amount of waste per week. The main causes of these jams are the complexity and variability of the composition of the waste flow (dirt, humidity, etc.). Therefore, a method based on an autoencoder artificial neural network is used to detect potential jams on conveyor belts in waste sorting plants. The method and results on real industrial data are explained and allow concluding on the feasibility of using an autoencoder model to detect potential jams.