AI-Based Work Sampling and Productivity Analytics for Sustainable Industrial Operations

Manual work sampling is still one of the most common methods for estimating the proportion of shop-floor time spent on value-added activities, necessary support activities, and waste, but it has the disadvantages of being restricted to occasional human observation, observer bias, and not being able to be used in real time for decision-making. The objective of this study is to design and test an AI-based work sampling and productivity analysis system performing continuous classification of workstation activity using a computer vision activity recognition system, IoT sensor stream data, and automated statistical processing, while converting it to productivity and sustainability indicators. The framework was then trialed for a 12-week period in a medium-sized discrete manufacturing case study facility and compared against parallel manual work-sampling observations. The AI system achieved an overall activity-classification accuracy of 93.4% (macro F1-score 0.91) when compared to manual observation as the reference standard, identified a higher proportion of value-adding time (61.8% vs 54.2%), and enabled operational changes that increased the productivity index by approximately 17% by week 12. Concurrently, energy consumption per unit decreased by 4.6 to 3.9 kWh/unit, and the scrap/waste rate decreased from 6.8% to 4.5%, providing a tangible sustainability co-benefit. The results indicate that using AI-enabled work sampling can significantly improve observation density and objectiveness compared to manual work sampling and directly connect productivity gains with results that drive resource efficiency and waste reduction, areas critical for sustainable manufacturing.