Intelligent load management for on-board aircraft generator systems

Amarantidis-Koronaios, Georgios Alexandros (2021). Intelligent load management for on-board aircraft generator systems. University of Birmingham. Ph.D.

Preview

AmarantidisKoronaios2021PhD.pdf
Text - Accepted Version
Available under License All rights reserved.
Download (6MB) | Preview

Abstract

This thesis describes research that has been undertaken to develop an Intelligent Load Management system that assists in overheat protection of on-board aircraft generators; both under normal operation and under fault conditions.

There is an ongoing demand to optimise military aircraft performance by decreasing weight, operating costs and gas emissions, whilst increasing overall reliability. Recently, a move towards a more electric aircraft has become increasingly popular to address these demands. Flight critical systems such as: cabin pressure, flight control, surface actuation, landing gear, breaking, etc. which were conventionally controlled by pneumatic or hydraulic systems, are now included in a wider electrical network. This conversion to electrical systems increases the need for constant and uninterrupted provision of power. Generators are limited by the amount of power they can provide. An excess demand of current for a prolonged period of time can lead to overheating, which in turn, can lead catastrophic failure due to insulation degradation. The state of the art overheat protection method is using a thermo-mechanical fuse. In case of overheat the generator trips offline, with only essential systems remaining operable, in order to prevent further damage.

The proposed alternative is to produce an “intelligent fuse” where models, knowledge of the mission profile, and temperature measurements are combined to predict future temperatures and manage the loading of the generators. A lab-based AC generator system was used as the main plant for this research. Based on that generator, a lumped parameter thermo-electric model was derived. It was further used as a simulation tool for faults and as a surrogate for a generator when multiple generators exist in the same system. This approach provided a high fit (<97%) in scenarios that had been advised by BAE Systems for all relevant temperatures.

A large part of the predictive methods used, revolved around using linear models. Both white box and black box approaches were assessed; with autoregressive exogenous models (ARX) providing the best performance for estimating the temperatures of sensitive parts of the generator using the information of measured currents.

In order to accommodate for faults during the mission, adaptive models were created. They considered variations of measured currents and measured temperatures to more accurately estimate future temperatures. These models either took the form of ARX or neural networks. Each provided their distinct advantages; with neural networks achieving more accurate prediction with high prediction horizons, whilst ARX being more robust throughout. This made ARX the preferred candidate for this application.

To ensure appropriate load management, both open loop and closed loop techniques were explored. Both neural networks open loop and closed control loops provided suitable solutions, with the control loops providing the least amount of performance compromise when no load management was necessary.

These ideas were extended to multi-generator systems. They were tested in simulation scenarios and in a hardware in the loop scenario, where the relative efficiency gain of having ILM in the system was subsequently assessed. ILM is also validated and tested on a BAE System aircraft generator in the Brough test facilities. Tests were run both under normal conditions and after a fault was introduced to the machine. The adaptive models provided an overall fit of at least 90% for the relevant temperatures under all conditions.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):