Study On Soft Sensor Modeling Methods And Applications

Soft sensor technology is one of the most important research directions in area of process control. In this dissertation, several issues and the corresponding solutions about soft sensor technology are discussed based on the real industrial process and the main contributions are described as follows.1. Many soft sensor methods are introduced in literatures of science and show good performance in their special applications, but in the more wide range of practice, drawbacks of these methods appeared. In order to study on the scope of application, we investigate frequently-used soft sensing methods based on several industry applications and get some useful conclusions.2. A multi-model soft sensing method based on Affinity Propagation, Gaussian process and Bayesian committee machine is presented. It uses Affinity Propagation clustering arithmetic to cluster training samples according to their work modes. Then, the sub models are estimated by Gaussian process regression. Finally, in order to get a global probabilistic prediction, Bayesian committee machine is adopted to combine the outputs of the sub estimators. The proposed method have been applied to predict H₂S and SO₂ concentrations of sulfur recovery unit. Practical applications indicate it is useful for the online prediction of quality specifications in industry processes.3. An adaptive observer is a recursive algorithm for joint state-parameter estimation of parameterized state space systems. Previous works on globally convergent adaptive observers consider unknown parameters either in state equations or in output equations, but not in both of them. In this paper, a new adaptive observer is designed for linear time varying systems with unknown parameters in both state and output equations. Its global convergence for simultaneous estimation of states and parameters is formally established under appropriate assumptions. A numerical example is presented to illustrate the performance of this adaptive observer.4. Based on the techniques of high gain observer and adaptive estimation theory, an adaptive observer is proposed for state fault and sensor fault estimation in a class of uniformly observable nonlinear systems. It is first assumed that a high gain observer exists for the fault-free system. With a parametric model of sensor fault, a high gain adaptive observer is designed for fault estimation. In order to establish the global convergence of the adaptive observer, in addition to the usual conditions for high gain observer convergence, a persistent excitation condition is also required, like in most recursive parameter estimation problems.5. A full vehicle active suspension system is considered with the dynamics of the four actuators. The aging coefficients of suspension system componets are modeling as unknown time varying parameters. An adaptive observer is designed to estimate the aging coefficients. Simulation result shows the aging coefficients could be estimated rapidly.