Font Size: a A A

Modeling And Control For Sandwich Hysteresis Systems

Posted on:2008-05-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:L W MaFull Text:PDF
GTID:1118360242476004Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
With the development of society and scientific technology, theoretical subjects were improved steadily. The technology revolution would make the society and scientific technology progress because the new theories have been applied to practice. In this situation, the research topics scientists have to face will often be changeable. Nonlinearities often exist in control systems. It is a challenge to the analysis and controller design for the control systems. The existence of nonlinearity is one of the significant factors which constrain the performance of the system. Recently, a class of nonlinear system, i.e. so-called sandwich systems have attract the control engineers'attention. Nonlinear sandwich system is the system in which a nonlinear subsystem is embedded in between two dynamic linear subsystems. Gang Tao and Avinash Taware began to research sandwich dead-zone systems and sandwich backlash systems several years ago. It is known that dead-zone is static and backlash can be considered as a single-loop hysteresis. In contrast to them, hysteresis is a type of dynamic and undifferentiable nonlinearity. It is clear that hysteresis is more complex than dead-zone and backlash. The study on sandwich hysteresis systems has seldom been found in literature so far. Therefore, sandwich system with hysteresis is the main topic studied in this paper. It is known that neural networks have many advantages, such as: self-learning, associative memory, high speed sought optimization solution, and so on. It has been proved that neural network can approximate arbitrary smooth nonlinear functions. However,both theory and practice have proven that the traditional modeling schemes including neural networks can only approximate functions with one-to-one or multiple-to-one mapping.In this dissertation ,a continuous transformation is used to construct an elementary hysteresis operator (EHO). Then based on the EHO, an expanded input space is constructed. It can be proved that the relation between the input space and the output space of the neural network is a one-to-one mapping. Based on the constructed expanded input space, a neural network is employed to approximate the hysteresis.Though the EHO-based NN model is promising to approximate hysteresis, the modeling residual is still obvious in experimental results. A more precise method,i.e. the elementary hysteretic model (EHM) is proposed. The precision is not guaranteed if the coefficient is equal to 1 in the modeling procedure. Therefore, it is necessary to propose a new method to calculate coefficient in order to improve the EHM-based NN model. Based on the new EHO, the improved EHO-based NN hysteresis model is constructed.The existent compensation methods for hysteresis usually depend upon the inverses model based technique. Continuous transformation is used to construct an elementary inverse hysteresis operator (EIHO). Then based on the EIHO, an expanded input space is constructed. It can be proved that the relation between the input space and the output space of neural network is a one-to-one mapping. Based on the constructed expanded input space, a neural network is employed to approximate the inverse hysteresis. A improved EIHO-based NN inverse hysteresis model is also constructed.At last, an inverse hysteresis model is trained by indirect training. The inverse hysteresis model is used to cancel the influence of the hysteresis sandwiched between two linear blocks. The effects of the linear blocks are canceled by using state feedback control strategy. In this way, the inverse hysteresis model can directly cascade with the hysteresis to compensate for the effect of the hysteresis. The result of the simulation shows that the output of the controlled system can track the desired output and the inverse hysteresis model can almost cancel the disturbance of the hysteresis.
Keywords/Search Tags:sandwich hysteresis system, hysteresis, inverse hysteresis model, elementary hysteresis operator, elementary inverse hysteresis operator, neural network, expanded-input space method
PDF Full Text Request
Related items