Study on Static–Dynamic Characteristics and Decoupling of Agricultural Heavy-Load Six-Dimensional Force Sensor Based on Improved XGBoost Model

Precision operation is one of the crucial means to reduce costs and consumption in the agricultural production process. To enhance the operational efficiency of machinery, combined operation machines are often used for tillage operations. A large unbalanced inertial load is caused during operation, which also exhibit features such as continuity and uncertainty. A six-dimensional force sensor can simultaneously test and output six types of load information, which is an effective means for intelligent detection. However, inter-dimensional coupling of the sensor is unavoidable. Most existing studies on the static and dynamic characteristics as well as decoupling of six-dimensional force sensors are conducted for light-load and low-interference scenarios, which cannot be applied to farmland tillage scenarios characterized by heavy load and high interference. In this paper, focusing on the heavy-load working conditions of the six-dimensional force sensor, static and dynamic calibration tests based on full-scale force and torque loading were carried out to analyze the static and dynamic characteristic indicators of the six-dimensional force sensor. Additionally, research on static and dynamic decoupling of the six-dimensional force sensor based on the improved XGBoost model was conducted, and the decoupling effect of the improved XGBoost model was verified. The test results show the following: (1) The BO-XG Boost model based on Bayesian optimization exhibited superior performance. Compared with the conventional XGBoost model, the improved XGBoost model reduced the tuning time by 35.5% and achieved a lower RMSE. (2) The decoupling performance of the improved XG Boost model was significantly superior to that of Light GBM. When the loading frequency condition was 20 Hz, the R2P values of Fx, Fy, Fz, Mx, My and Mz were increased by 1.77%, 3.12%, 10.67%, 1.36%, 0.16%, and 0.18% respectively. At a loading frequency condition of 40 Hz, the R2P values of the six types of loads were improved by 0.78%, 0.72%, 0.90%, 0.50%, 1.77%, and 0.55% respectively, with corresponding improvements observed at other frequencies. When the loading frequency condition was 160 Hz, the R2P values of Fx, Fy, Fz, Mx, My and Mz all exceed 0.99. The improved XGBoost model could significantly reduce the coupling of electrical signals from other channels to the main channel, making it suitable for the dynamic decoupling of this sensor.