Using Safety Critical Artificial Neural Networks in Gas Turbine Aero-Engine Control

摘要

'Safety Critical Artificial Neural Networks' (SCANNs) have been previously defined to perform nonlinear function approximation and learning. SCANN exploits safety constraints to ensure identified failure modes are mitigated for highly-dependable roles. It represents both qualitative and quantitative knowledge using fuzzy rules and is described as a 'hybrid' neural network. The 'Safety Lifecycle for Artificial Neural Networks' (SLANN) has also previously defined the appropriate development and safety analysis tasks for these 'hybrid' neural networks. This paper examines the practicalities of using the SCANN and SLANN for Gas Turbine Aero-Engine control. The solution facilitates adaptation to a changing environment such as engine degradation and offers extra cost efficiency over conventional approaches. A walkthrough of the SLANN is presented demonstrating the interrelationship of development and safety processes enabling product-based safety arguments. Results illustrating the benefits and safety of the SCANN in a Gas Turbine Engine Model are provided using the SCANN simulation tool.