Multi-class and Multi-label Classification of an Assembly Task in Manufacturing

Human action monitoring is a tool that could help improve performance and efficiency in industrial assembly. Monitoring actions is a very complicated task to solve due to the complexity of the tasks to be classified added to the lack of data within the sector. Human action monitoring in the industrial environment is a complex problem to perform due to the difficulty of differentiating very complex tasks. Current methods are able to solve the problem of simple tasks while having difficulties during task transitions. Our approach aims to solve the problem of action classification in the industrial domain using deep learning models. By creating a multi-label classification model, we obtain a multi-label accuracy of 94.48% on a set of 12 tasks in the assembly of an industrial tool. The lessons learned in this work can serve as a basis for the construction of deep learning models for classifying complex actions in real time of industrial assembling tasks.