An implementation of a radial basis function network states observer

Knowledge on the internal states of an engineering system is vital for diagnosing faults and controlling the system. However, these internal states are often not directly measurable, therefore they are instead estimated based on the input and output data. This form of estimation is known as a states observer. All model-based states observers will encounter differential equations. In control applications, the majority of the differential equations encountered are ordinary differential equations (ODE). Many ODE solvers utilise high order interpolation and recursive bound checking to guarantee smoothness and accuracy of the solution, which complicate the algorithm and lengthen the computation time. A further drawback of using recursive algorithm is that the computation time cannot be determined beforehand, resulting in an inefficient use of computational bandwidth. An improved implementation of an ANN states observer using radial basis function (RBF) network is proposed in this paper. The proposed RBF network observer has virtually similar performance as a known ODE solver in linear systems, while being much faster and cheaper to implement. Hence, it is more attractive compared to any other methods for areas such as sensors and embedded systems.