Multi-sensor fusion based industrial action recognition method under the environment of intelligent manufacturing

In the context of intelligent manufacturing and Industry 4.0, the manufacturing industry is rapidly transitioning toward mass personalization production. Despite this trend, the assembly industry still relies on manual operations performed by workers, considering their cognitive ability and flexibility. Thereinto, studying operator action perception and recognition methods is a vital filed and of great significance for improving the production efficiency and ensuring product quality. In this paper, a multi-sensor fusion-based data acquisition system is constructed to address the challenge of achieving comprehensive and accurate perception of the assembly process with a single sensor. Then, an action recognition model architecture based on ResNet + LSTM + D-S evidence theory is proposed and established. By fully considering the characteristics of different data, the multi-sensor data values are maximized, data complementarity is achieved, and the recognition accuracy exceeds 97 %. This research is expected to provide guidance for increasing the degree of workshop automation and improving the efficiency and quality of the production process.