Hardware design of neural network system state observer

Modern development of software has allowed us to solve complex and non-linear control systems. Software such as MATLAB, MAPLE, etc allows control engineers to design their systems easily through graphical programming. However, implementation of a software-based control design on hardware such as FPGA or DSP are met with a difficult challenge as hardware constraints tend to limit the performance or affect the output of the control design. An example would be a state observer which is comprised of differential equations. To process the differential equations on embedded systems using standard algorithms is not feasible and with increasing complexity of control designs, state observers are restricted to general purpose processors. Based on previous research, an alternative approach to developing state observers is by using neural networks to solve the differential equations. This paper presents a methodology on implementation of a neural network based state observer onto FPGA. The study of this research showed that the state observer design can be implemented on hardware without the need to sacrifice performance or increase the cost of implementation. The prospects of this research can be applied to fields such as fault tolerant or self-diagnostic systems.