Research on power modeling of the industrial robot based on ResNet

With the development of intelligent manufacturing and the popularization of industrial robots (IR), the energy consumption (EC) of IR become increasingly significant and attract attention from academia and industry. Accurate EC modeling and prediction is the premise and basis for energy efficiency optimization of IR. In recent years, intelligent algorithms like neural networks have been introduced into the EC modeling of IR, and high-precision models can be constructed without complex modeling and parameter identification process. In contrast with the direct modeling method based on mechanism model, data-driven modeling has significant efficiency and application advantages. In this paper, we propose a ResNet-based IR power modeling method. The deep learning algorithm is used to establish the precise mapping between the operating parameters and power of the industrial robot. Then, the model is applied to another industrial robot by transfer learning, which realizes the rapid construction and deployment of power models for different IR. The experimental results show that the proposed IR power modeling method can accurately construct the power model and achieve 21.83% less error. After transfer learning, the power model can be accurately constructed even when the amount of data is reduced by 80%, which facilitates the large-scale practical application of IR and their energy efficiency improvement.