Study on Hybrid XG Boost–LSTM Predictor with a Digital-Twin Monte Carlo Virtual Check for Risk-Aware Scheduling in MSME CNC Manufacturing cluster

Abstract

Smart scheduler proposed in this research is using Risk aware scheduling methods for micro Small and medium enterprises involved in precision components manufacturing processes using Computerised Numerical Control (CNC) machines for a group of operators in an industrial cluster is challenging because of its non-regular availability and also based on different parametric requirements. The normal rule-based scheduling approaches fails in solving such issues for machine condition, tool wear, queue congestion, and deadline-related uncertainty. The study shows most hybrid machine-learning models cannot predict process risk, and do not reflect in ever evolving real-time state of the physical manufacturing system. To address this limitation, this study proposes a two-level decision-fusion framework that integrates a hybrid XG Boost–LSTM risk predictor with a Digital Twin-based Monte Carlo virtual verification mechanism. The hybrid model generates a predictive delay-risk estimate, while the Digital Twin simulates candidate job execution under stochastic tool-wear, machine-availability, and queue-congestion conditions to obtain an independent risk estimate. These estimates are combined using an interpretable weighted fusion model, which is incorporated into a priority-aware machine-selection and scheduling strategy. The proposed framework was evaluated using a synthetic six-machine MSME-style precision-CNC manufacturing environment through independent replications, multi-scenario experiments, alternative fusion models, and Monte Carlo convergence analysis. The results show that fused scheduling with improved pooled on-time delivery to 74.0%, compared with 72.3% for the hybrid predictor alone and 70.3% for availability-based scheduling. The improvement over predictor-only scheduling was statistically significant compared to previous research achievements. Furthermore, the predictive and Digital Twin risk estimates exhibited near-zero correlation indicates complementary information. Linear fusion also matched or outperformed Bayesian and logistic-regression fusion alternatives while maintaining low computational latency, with worst-case scheduling latency below 50 ms for six candidate machines. The findings demonstrate the feasibility of combining predictive intelligence with Digital Twin-based virtual verification for real-time, risk-aware scheduling in MSME manufacturing Key words: Digital twin, cyber-physical systems, hybrid machine learning, decision-level fusion, risk-aware scheduling, Monte Carlo simulation, Industry 5.0, small and medium enterprises.