SAIP2026

Machine Learning-Based Prediction of Tool Wear and Remaining Useful Life Using Multisensor Data Fusion

7 Jul 2026, 17:20

1h 20m

Great Hall (University of the Western Cape)

Poster Presentation Track F - Applied Physics Poster Session 1

Speaker

SAM MKHWEBANE

Description

Abstract
In the era of Industry 4.0, the transition toward fully autonomous man
ufacturing requires robust Tool Condition Monitoring (TCM) to prevent
catastrophic tool failure and workpiece damage. Precise prediction of cut
ting tool wear and remaining useful life (RUL) is necessary for reducing
downtime and optimizing tool usage in automated machining. However,
manual inspection requires frequent machine stoppage, while traditional
monitoring methods are intrusive and struggle to capture the non-linear
nature of wear progression in lathe operations. This study aims to develop
and validate a machine learning model for tool wear and RUL prediction
using multisensor data fusion. Random Forest, Support Vector Regres
sion, and XGBoost models will be trained on available lathe data from
previous studies, then validated on experimental lathe turning data to
be collected under a Taguchi L9 design using AISI 1045 steel workpieces
and uncoated carbide inserts. Multisensor data (force, vibration, acoustic
emission) will be processed to extract 84 time-domain features per cut,
reduced via feature selection. When validated on the experimental lathe
data, all three models are expected to achieve an R2 exceeding 0.85 and
an RMSE below 40 µm, with XGBoost leading in accuracy, demonstrat
ing cross-dataset generalizability. The findings will support tool condition
monitoring systems for smart manufacturing applications, contributing to
reduced downtime, optimized tool usage, and improved production effi
ciency.
Keywords: Tool Wear, Machine Learning, RUL, Random Forest, XGBoost,
Turning, SVR, Taguchi Method