Using Neural Networks to Identify Wired Peripherals Connected to Integrated Modular Avionics Hardware

This paper presents an approach to detect electronic components attached to Integrated Modular Avionics modules without any smart or additional interface based on their electrical properties using neural networks. Different LEDs and electric motors are classified, neural networks with different hyperparameters trained and accuracies compared. The neural network approach is compared to a simpler least-square approach. The best performing network is exported into C code and successfully tested on an IMA module. Based on the detected peripherals, the IMA module executes the corresponding application. The approach is validated using a fictive aircraft lighting system.