Methods of Robust Virtual Sensor Design

The problem of increasing the efficiency of technical systems by methods of functional diagnostics using virtual sensors is described in cases where the available physical sensors are not enough to reduce the complexity of diagnostic tools or to localize defects with the required depth. The use of additional physical sensors to achieve the desired result may require additional costs, in addition, the reliability of such sensors is usually low. The problem of constructing robust virtual sensors for technical systems described by nonlinear models under the action of external disturbances has been posed and solved. Two ways of solving this problem are proposed based on the identification and Jordan canonical forms. Relations are given that allow constructing a sensor of minimal complexity for estimating a given component of the state vector of a technical system. In this case, the constructed sensor will be insensitive or minimally sensitive to perturbations, which is achieved by using the singular value decomposition of matrices describing the perturbations and the original system. The synthesized virtual sensor can become an addition to the existing physical sensors or replace a failed physical sensor. Theoretical provisions are illustrated with a practical example. The results obtained can be used to solve problems of building fault-tolerant systems.