Hydraulic shovel digging phase simulation and force prediction using machine-learning techniques

The efficient and effective utilization of hydraulic excavators relies on, among other factors, a better understanding of the formation resistance during the excavation process. Studies on hydraulic shovel excavators have focused on optimizing the digging process by considering the reaction of the b...

Full description

Saved in:
Bibliographic Details
Published inMining engineering Vol. 74; no. 1; pp. 31 - 33
Main Authors Azure, Jessica W A, Ayawah, Prosper E A, Kaba, Azupuri G A, Kadingdi, syth A, Frimpong, Samuel
Format Journal Article
LanguageEnglish
Published Littleton Society for Mining, Metallurgy, and Exploration, Inc 01.01.2022
Subjects
Algorithms
Angle of repose
Artificial intelligence
Automation
Bulk density
Contact angle
Contact force
Dredging
Excavation
Excavators
Forces
Hydraulics
Joints (timber)
Learning algorithms
Machine learning
Mining engineering
Performance evaluation
Profits
Root-mean-square errors
Simulation
Size distribution
Support vector machines
Online AccessGet full text

Cover

More Information
Summary:The efficient and effective utilization of hydraulic excavators relies on, among other factors, a better understanding of the formation resistance during the excavation process. Studies on hydraulic shovel excavators have focused on optimizing the digging process by considering the reaction of the boom, stick and bucket, and analyzing the resulting moments at the shovel's front-end joints. These studies failed to understand resistive forces from a statistical viewpoint. The purpose of this paper is to develop machine-learning models capable of predicting the formation of resistive forces during shovel excavation. The shovel's excavation was simulated in the PFC 5.0 environment using a typical field-size hydraulic shovel bucket to generate the shovel contact forces required to overcome formation resistance. Bulk density, angle of repose, rock fragment size distribution and formation height were investigated. The recorded formation reaction forces were analyzed using six machinelearning algorithms. The results indicate that the support vector machine was the best model, with coefficient of determination (R2) of 0.76, root-mean-square error (RMSE) of 0.077 and mean absolute error (MAE) of 0.057. The results obtained show that machine-learning techniques are useful and reliable means of predicting formation resistive forces. This research is a preliminary step toward generating models for resistance forces that potentially enhance shovel automation.
ISSN:0026-5187