This paper shows how we used two state-of-the- art ARINC 653P4 capable Integrated Modular Avionics (IMA) modules to implement a self-detecting and self-configuring platform application using a machine learning application. To tackle the challenge of out-of-sample peripherals, we show that the application can use an autoencoder to check if the acquired peripheral information behaves similar to the training data. Compared to the previous work, this ability enables a more robust detection of connected simple peripherals - solely based on the electrical properties. The autoencoder improves the ability to correctly detect peripherals for our test samples to 100%. In general, the autoencoder allows us to distinguish among input data that is similar to the training data and out-of-sample data. We acquired out-of-sample data for aged, failed, unknown peripherals, as well as interface open scenarios. Based on these, we created metrices that allow to evaluate which scenario is most likely - including the current maintenance state of a peripheral. This offers more detailed information to the maintenance organizations - both for corrective and preventive maintenance.We demonstrate how two IMA modules use neural networks to detect connected peripherals, agree on the application to execute and jointly execute the correct application without prior knowledge about which peripherals connect to which device.
Using Autoencoders to Identify Aged, Faulty and Unknown Peripherals in the Adaptive IMA System
01.10.2023
1643118 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
A New Approach to Identify Faulty Sections in Distribution System
British Library Online Contents | 2000
|An Intelligent Transport System Methodology to Identify and Controlling of Faulty Vehicles
Trans Tech Publications | 2014
|