Research On Digital Filtering For The Six-Axis Force Sensor

The multi-axis force sensor is not only the most important sensor of the intellectualized robot, but also the groundwork of engineering mechanical testing and measure. The six-axis force sensor is the most widely used in many types of multi-axis force sensor. It can detect the size of six directions force in the three-dimensional space. However, there is a large number of noise mixing in the output signal, such as themal noise of resistance strain gauges, the noise of electromagnetic device in amplifier circuit and the interference of high temperature generated by the elastic body creep. Consequently, how to effectively eliminate the noise has a great influence on real-time tracking of the system’s state vector. The main aim of this master thesis is to build the space-state model of six-axis sensor and to filter colored noise about output signal of the sensor by using suitable digital filtering algorithms.(1) State-space modelBased on time-series analysis, the whitening process of linear colored interference is achieved by the equivalent of Gauss-Markov model. According to the main trend of the strain modes, the thin linear state-space model is established. By introducing the ideas of mutual coupling filter, nonlinear colored noise is considered as white noise, high-term colored noise and the whole process interference.Then a series of colored noise model is built by combining nonlinear colored noise with state transition matrix.(2) Adaptive Kalman filteringTo solve the problem that the standard Kalman filter can’t gain the optimal estimation because of the state-space model error of the sensor. The paper discards the idea of mutual coupling filtering and proposed four kinds of adaptive EKF algorithms. By simplifying higher-order function state-model, the model error can transfers the state transition matrix to the system interference matrix and system control matrix. The multiple adaptive factors are constructed through the model of three sections function and fuzzy adaptive control algorithm. By combining with the chaos weeds optimization algorithm, the model error is repaired and optimal estimation is acquired by the technology of dynamically adjusting the weight of state prediction in the filter estimation.(3) Improved particles filteringTo enhance the robustness performance of Adaptive EKF filtering, the paper use the particle filter algorithm for nonlinear system filtering problems and propose UPF, LOPF, GSO-PF to optimize the resampling process of the traditional particle filter and keep the diversity of particles. Meanwhile, based on the framework of the traditional discrete sliding mode controller, the improved controller which has the replanning switching function can amend the simplified model error by repairing the feedback sine wave signal.(4) Improved wavelet threshold transformTo eliminate the noise signal based on non-Bayesian framework, the paper proposes two kinds of improved wavelet threshold method to redefine the relationship between wavelet coefficients and wavelet threshold by the gray theory and the concepts of relative entropy. The paper uses the ability of the focus of wavelet transform to combine each signal with another section and research the model of three attached circular plate. Then the performance judgement of wavelet transform is achieved by identifying the damage index.(5) SVR adaptive filteringIn order to verify the correctness of the six-axis force sensor model, the paper presents adaptive filter of improved support vector regression (SVR) which can identify the state-model of E-type membrane. The real difference of SVR algorithm between theoretical value and the measured value can eliminate by noise canceler and become the input of noise-free state model. Finally, the feedback signal error of the sensor output signal is repaired by sliding mode controller(6) Interactive multi-excitation filteringTo filter the noise of more dynamic coupling excitation, the paper put the information of multi-directional excitation force into the system control matrix to establish multi-target state space model by transferring combined strain to combined force. By redefining the weight ratio of excitation force through the model probability of interactive particle algorithm, the system state can be tracked by using the joint estimation of three strategies.