.png){kind=icon}
.png){kind=icon}
Return
AI-Driven Circular Manufacturing Framework for Predictive Material Flow Optimisation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.iswa.2026.200651.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The move towards circular manufacturing needs the implementation of intelligent systems that can minimize material waste and optimize manufacturing efficiency using predictive and adaptive control. In this paper, an AIbased Digital Twin approach for predictive material flow optimization in sheet-cutting operations is presented, providing a basis for future prescriptive decision support. A Discrete Event Simulation (DES) model was created in MATLAB SimEvents to simulate a sheet-cutting cell, producing process-level data for forty simulation runs (100 parts per simulation run) with part areas between 1.20 and 9.41 m2 (average = 4.22 m2, std dev = 2.27 m2). Analysis of material flow behaviour showed an average material utilization of 81.15% and corresponding average scrap amount of 18.85%, with 41.0% of parts needing rotation to fit. Three machine learning models, namely XGBoost, Random Forest, and Neural Network, were trained on the DES data using five-fold time-series cross-validation for predicting scrap amount. XGBoost showed better prediction accuracy (mean R2 = 0.73, RMSE = 2.35, MAE = 1.74) than Random Forest (R2 = 0.65, RMSE = 2.70, MAE = 1.98) and Neural Network (R2 = 0.44, RMSE = 3.27, MAE = 2.76). SHAP analysis showed that the real-time sheet-level variables used_on_sheet and remaining_on_sheet are the most influential in scrap prediction, and dependence analysis highlighted nonlinear relationships such as the inverted-U shape for cumulative used area. Feature interaction analysis confirmed the most important interaction between util_percent and used_on_sheet (0.1423), thus proving that the risk of scrap is a function of the joint operational intensity and material state. The approach shows a strong alignment with IATF 16949 and ISO 9001 quality management system standards.
Keywords:
Digital Twin
Discrete Event Simulation (DES)
XGBoost
SHAP
material utilisation
circular manufacturing
predictive optimisation
Predictive Quality Assurance (PQA)
Journal
I
IF:
4.3
Papers:
90
Citations:
0