Strain Analysis And Data-Driven Decoupling Of Six-Axis Force/Torque Sensors

The six-axis force/torque sensor,which features the ability to sense complete force/torque information in three-dimensional space,plays an initial role in the process of industrial automation and intelligence.There are still some critical issues associated with the design and inter-dimension decoupling of the two typical types of six-axis force/torque sensors(Y-shaped elastic beam and Stewart platform).The prerequisite of designing a six-axis force/torque sensor with elastic beam is to know the strain distribution in its elastic structure.However,the traditional Finite Element Analysis(FEA)based on geometric modeling is time-consuming and inefficient.An efficient and accurate strain analysis is therefore proposed for the six-axis force/torque sensor with Y-shaped elastic crossbeam.The deformation characteristics of the Y-shaped beam under various axial force/torque loads are analyzed,and a simplified mechanical model is established using Timoshenko beam theory.The explicit expressions of various strains are obtained,and the corresponding predictions are compared with the finite element simulations.The results show that the analytical solution agrees quite well with the FEA results,which validates the analytical model.The study therefore provides an effective and accurate tool for the design of six-axis force/torque sensors with elastic Y-shaped crossbeam.In the development of six-dimensional force/torque sensor products,software decoupling algorithm are an important way to improve their decoupling performance.Due to the structural design and manufacturing errors,the traditional decoupling algorithm implemented by calculating the pseudo-inverse matrix of the calibration data usually encounters problems such as incomplete decoupling and ill-condition of the decoupling matrix.With the development of statistical learning approaches,data-driven artificial neural network algorithms have demonstrated an excellent generalization ability.A single hidden layer neural network model is therefore established for the two typical six-axis force/torque sensors.Instead of the conventional uniaxial calibration dataset which is limited and incomplete,a large-scaled dataset is automatically generated by the finite element simulation,which is conducted with ABAQUS by developing codes based on Python.The decoupling performance of the traditional decoupling algorithm and that of the artificial neural network algorithm for the two types of sensors are investigated and compared.It is shown that the latter is more robust than the former,with the decoupling error reduced effectively.