Fault Detection Methods Based On Optimal Sensor Location

Recently the theoretical study and practical application for the model-based fault detection (FD) technique has made remarkable development. Some research findings have been obtained about knowledge-based optimal sensor location. The sensor locations were often defaulted as known, but the approaches to improving robust fault detection precision based on optimal sensor location have not yet received much attention. For linear time invariant system, this paper mainly concerns methods based on hybrid H_/H_∞ multi-objective optimization design for fault detection system, and proposes a new concept and method for fault detection by combining with optimal sensor location design, which the numerical simulation and experiments indicate the effectiveness and feasibility of these algorithms. The major work and the result of research are represented as follows: For linear time invariant dynamic system, an optimization design method for hybrid H_/H_∞ observer-based fault detection system is proposed. Using the direct Linear Matrix Inequality (LMI) technique, existence conditions of hybrid H_/H_∞ fault detection observer gain and its solutions are obtained, and an iterative algorithm for obtaining its optimal solution is given, so that an optimal fault detection system is formed. The system designed by this method can be achieved with an optimal trade-off between the more sensitivity to faults and the more robustness against the unknown inputs on the premise of guarantying system stable. Simulation example shows the effectiveness of the design method and algorithm. For linear time invariant dynamic system, a non-convex multi-objective optimization design method for H_/H_∞ fault detection filter is proposed. Residual generator constituted of output observer and post-filter is described as non-convexBilinear Matrix Inequalities (BMIs) form. By using the upper bound of non- positive definite terms that can perfect squared in the bilinear matrix inequalities to replace the original non-positive definite terms and transforming BMIs into a set of linear matrix inequalities, solvable conditions of optimization problem and its solution for observer gain and post-filter matrix are obtained, and the convergence of this algorithm is proved. The fault detection filter design also is highly sensitive to faults, and is strongly robust to unknown input. Compared with observer-based fault detection filter without post-filter, this method can give designer more freedom in design. Simulation example and bearing experiment shows the effectiveness of the design method and algorithm.For linear time invariant dynamic system, an approach to hybrid HJHn fault detection observer design based on optimal sensor location is proposed. The system use the offered k sensors to form N measuring points (to measure variables), and by optimally selecting m(m