An Artificial Neural Network Architecture to Classify Workers’ Operations in Manual Production Processes

The recent Industry 4.0 paradigm is disruptively changing the manufacturing landscape. Where fully automated settings are not feasible or economically viable, Industrial Internet of Things sensors are gaining traction due to their flexibility and affordable costs. In such a scenario emerges crescent attention to digitizing the human factor. Based on this, this manuscript proposes an integrated digital architecture in which a radio-frequency-based indoor positioning system is adopted to anonymously tag human operators. The highly unbalanced spatio-temporal dataset is fed into a recurrent neural network architecture that classifies without overfitting the manual picking/deposit activities in products’ stocking areas of a real and low standardized manufacturing job shop with a performance of 0.93.