LOTUS: A Frugal AI-Driven Simulation-Optimization Framework for Process Improvement in Resource-Constrained Indian MSMEs

Abstract

Micro, Small, and Medium Enterprises (MSMEs) contribute nearly 30% of India's gross domestic product and about 45% of its exports, yet most units operate with scarce data, limited computing infrastructure, and tight capital budgets, which places conventional Industry 4.0 digital-twin and simulation-optimization tools beyond their reach. This paper proposes LOTUS (Low-data Optimization with Twin-Unified Surrogates), a frugal, artificial intelligence-driven simulation-optimization framework explicitly designed for resource-constrained Indian MSMEs. LOTUS couples a lightweight open-source discrete-event simulation (DES) twin of the shop floor with a sample-efficient Gaussianprocess surrogate and a Bayesian-optimization search layer, so that near-optimal process and dispatching decisions can be identified using very few expensive simulation runs, entirely on a commodity laptop and without cloud dependence. The framework is validated on a simulated eight-machine auto-components machining job shop representative of a Tier-2/Tier-3 Indian cluster. Against a first-in-first-out heuristic baseline and a genetic-algorithm baseline, LOTUS improved mean daily throughput by 18.3%, reduced makespan by 14.1%, lowered work-in-process by 22.4%, cut energy consumption per part by 15.1%, and reduced the scrap rate from 6.2% to 4.5%, while requiring roughly 90% fewer true simulation evaluations than direct search. All results are averaged over 30 independent replications and verified with 95% confidence intervals and paired t-tests (p < 0.01). Estimated annualized savings approach INR 5.5 lakh for a single unit, offering a deployable, low-cost digital pathway aligned with national MSME digitalization initiatives.