Research Of Non-liner Decoupling Algorithm For Piezoelectric Six-axis Force/Torque Sensor

As an important pillar industry of modern information technology,sensor technology is regarded universally at home and abroad as the one with the greatest potential.Featuring by rich measuring force and high precision,it is widely used in areas such as robot technology,aircraft landing force detection and transient thrust measurement of Rocket engine.Among those the Piezoelectric Six-Axis Force/Torque Sensor,with favorable dynamic characteristics,high precision,high sensitivity,and high reliability,can be applied to areas requiring high natural frequency,thus being one of the greatest potential.It usually adopts multi-point support measurement method which may be affected by factors in assembly,manufacturing and layout thus leading to coupling in multi-dimensional output.Therefore,researches on decoupling of the piezoelectric six-axis force/torque sensor have been especially important.Under the support of the National Natural Science Foundation of China(51205165),this paper is focused on the research on the decoupling of the Piezoelectric Six-Axis Force/Torque Sensor.The major parts are as follows:First,this paper introduces the structural type,layout of force sensing element and the measuring principles of the sensor,builds up the measurement model,conduct the calibration and loading experiment and analyze the reasons of coupling errors of the sensor according to the results of calibration experiment and comes up with the appraisal indicators of coupling of the sensor.Second,on the premise that the measurement system model is linear,the paper adopts two kinds of static linear decoupling algorithm based on the average calibration matrix and the least squares method to build up two kinds of lineal decoupling algorithm respectively.Then the paper gets the calibration matrix by adopting the training sample data through the two models.After that,the paper makes comparative analysis of the decoupling effects of the two linear decoupling algorithm by adopting non-training sample data.Third,due to the non-linear characteristics of the measuring system of the sensor,there are some limitations to adopt linear decoupling algorithm.Therefore,on the basis of the linear decoupling models,the paper comes up with non-linear decoupling algorithm based on back propagation neural network and radial basis function neural network.Then,principles of thetwo neural networks are introduced to build up the two kinds of non-linear decoupling algorithms.After choosing training samples to train network models,the paper adopts non-training samples to make decoupling effects certification and comparative analysis of the trained models.Forth,to decoupling the sensor through single algorithm will be limited by its defects.However,it's possible to overcome respective defects by mingling different kinds of algorithms and improve the decoupling performance.On such an theoretical basis,this paper comes up with the decoupling algorithms of least squares support vector machine and genetic algorithm improvement neural network.Also,the paper builds up the decoupling models of those two algorithms and certifies their decoupling effects,which shows that the measuring performance of the sensor has been greatly improved after decoupling by hybrid algorithm.The results prove the feasibility and validity of the hybrid decoupling algorithm.