Application of Random Forest Method for Estimating Rejected Product in Industrial Conveyor Belt

Abstract

In most companies, the unloading system still poses a major problem which constitutes repetitive stops of the chain because of cracking at the level of the carriage and the intensive loss of the large quantities of the raw material. This problem can cause a time waste, a high cost and even stopping the production line. This paper presents an enhanced procedure able to achieve relevant classification of weight product in conveyor belt in order to supply quantitative estimation of rejected or not rejected (R/nR) cases. The studied database contains the different weight of both rejected and not-rejected product. The results show that the use of a random forest classifier is an effective way to improve estimation and classification for fast and truthful industrial diagnostic. Compared to other machine learning methods, the proposed method provided a significant performance reaching more than 90% of accuracy.