딥러닝 기반의 컨베이어 벨트 수명 예측 모델 개발

Abstract

This thesis proposes an RUL estimation model which predicts the remaining useful life (RUL) of a head pulley, a component used in a belt conveyor system to drive the belt by pulling and pushing the belt. Belt conveyor systems are widely used in a myriad number of production plants and factories due to relatively low costs and their versatility \cite{conveyors}. While the manufacturers that implemented the belt conveyor systems have benefited from the convenience of safely transporting various materials without excessive human labor and resultant time consumption, the production sites had to also consider the damages, degradation, and malfunction of the belt conveyor systems as time passed by. To keep the system running in good conditions, RUL estimation is essential to run the system smoothly.

Recently, neural networks are used to create models that effectively predict the RUL of the machinery or the component due to the neural networks' ability to automatically extract useful features and to find non-linear patterns. To do so, a large amount of data is required to train and test models due to the nature of deep learning. However, there are not many datasets that specialize in RUL prediction, and existing RUL datasets are mostly limited to bearings. This thesis uses a dataset specially collected to predict the RUL of a head pulley, which is another commonly used component but not widely researched. The dataset is collected using a custom sensor on a small belt conveyor system installed near the building of POSTECH Institute of Artificial Intelligence (PIAI).

The proposed model uses sensor data attached to a small belt conveyor to predict the RUL of the head pulley of the conveyor system. The model classifies the input data into 6 classes (thickness) and then computes the predicted RUL using a soft-voting classifier after using boosting to create an ensemble of models.