Dynamic Modeling Method of Six-axis Force/Torque Sensors Based on IGWO-GRU*

The strain-based six-axis force/torque sensor often faces challenges in accurately measuring complex and refined dynamic phenomena, leading to significant dynamic errors. To overcome this issue, it is essential to establish a high-precision dynamic model. A dynamic modeling method for six-axis force/torque sensors based on deep learning is presented in this paper. The proposed method utilizes the gated recurrent unit (GRU) neural network to construct an accurate dynamic model. The GRU neural network is capable of capturing long-term dependencies and handling sequential data effectively, making it suitable for modeling the interdimensional coupling between the dynamic force/torque and the sensor’s dynamic response. To address the subjective judgment involved in selecting hyperparameters for deep learning, the improved grey wolf optimization (IGWO) algorithm is employed. In this study, the proposed IGWO-GRU method is applied to dynamically model six-axis force/torque sensors under negative step excitation and impact excitation. The evaluation of the method yields an average mean squared error (MSE) of 0.056417 and 0.05590 for the six channels in the respective cases. Through comparison with other algorithms, the results demonstrate the accurate modeling effect and the generalizability of the proposed method. Moreover, the findings provide new insights for the dynamic modeling of other types of sensors.