Virtual AI Worklaod Orchestration For Heterogeneous Computing In Industry

Abstract

Industry 4.0 smart manufacturing uses Artificial Intelligence (AI) for quality inspection, predictive maintenance, anomaly detection, and real-time automation. These applications run on heterogeneous hardware such as CPUs, GPUs, FPGAs, and TPUs, but efficient workload scheduling is challenging due to differences in performance, energy usage, and latency. Poor task allocation increases execution delay, power consumption, and reduces resource utilization. This paper proposes a Virtual AI Workload Orchestrator (VAWO), a simulation-based framework for optimizing AI task scheduling across heterogeneous devices without real hardware. It integrates workload generation, scheduling algorithms, device modeling, monitoring, and visualization dashboards. Experimental results show that reinforcement learning-based scheduling improves throughput, reduces latency, and lowers energy consumption compared to static methods. The framework also supports sustainability goals by enabling energy-efficient industrial computing.