Architecture of an Onboard Neural Network Computer Based on FPGA Crystal with Dynamic Reconfiguration

This paper studies the development of principles for constructing the hardware architecture of neural network computers (NNCs) used as accelerators and coprocessors for the implementation of expert and intelligent systems in objects with a high degree of criticality, including the onboard equipment of aircraft. A model of a NNC built on the basis of an FPGA crystal is developed, which allows dynamic deployment of artificial neural networks on its basis without the need to reprogram the FPGA crystal itself. The principles of deployment and training of a neural network based on the presented NNC are described. An example of its use in an expert system that analyzes the state of an onboard information and computing network of an aircraft for the occurrence of abnormal situations in real time and performs parrying of arising failures by the method of dynamic reconfiguration is given.