Hybrid Edge–Cloud Intelligent Assistive Framework for Inclusive Industrial Training and Adaptive Quality Monitoring

Abstract

Accessible vocational training for individuals with disabilities demands intelligent, real-time systems capable of adaptive, personalised support within complex industrial environments. The Hybrid Edge–Cloud Intelligent Assistive Framework (ECIAF) in this work is a specific purpose-built design that combines low-latency edge-based visual monitoring with cloud-hosted deep reinforcement learning for scalable, responsive assistive support for technically disadvantaged learners and workers. The edge tier implemented using a lightweight ResNet-18 convolutional model achieves an average inference latency of 11.90 ms and defect detection accuracy of 97.20% on the MVTec AD benchmark, significantly exceeding real-time responsiveness requirements for assistive purposes. The cloud tier uses a Deep Q-Network (DQN) agent that is trained using a logistic regression environment model adapted from the SECOM semiconductor manufacturing dataset to dynamically calibrate ten process-support parameters representative of scaffolding level, task complexity, and feedback intensity in inclusive education. The adaptive support mechanism in more than 500 training episodes achieved a near-zero error rate, achieving a 99.96% reduction in monitored task errors, surpassing all defined inclusive training objectives. The tri-tiered architecture of the framework—perception, adaptive intelligence and an accessible multimodal interface—is tightly integrated with Universal Design for Learning principles and assistive technology design standards, enabling support for a wide range of cognitive, sensory and physical disabilities. The study highlights the ECIAF as a technologically sound and pedagogically grounded basis for future inclusive industrial education systems.