Multi-sensor data fusion and deep learning-based prediction of excavator bucket fill rates

Abstract

Accurately predicting the bucket fill rate of excavators is a challenging task due to factors such as material flowability and the complex coupling interactions between the material and the bucket. To address this challenge, this paper proposes a bucket fill rate prediction method based on multi-sensor data fusion and deep learning. The ITCBAM model was developed by integrating a Convolutional Block Attention Module (CBAM) into the InceptionTime framework, leveraging multi-source sensor data such as cylinder displacement and stereo vision to enable precise predictions of fill rates. Results show that the ITCBAM model achieves prediction errors of 9.48% and 10.65% on the familiar and unfamiliar test sets, respectively. Compared to physical models and other deep learning models, it demonstrates higher prediction accuracy and stronger generalization capability. This method facilitates excavation decision-making, enhances construction efficiency, and provides valuable insights for further research on the automation and real-time prediction of construction machinery.