The Future of Avionics: Machine-Learned and Certified

March 2, 2023

The emerging advanced air mobility market is driving the convergence of airborne electronic systems that are highly integrated with unprecedented functionality. Legacy avionics systems simply can’t meet these new requirements: namely because they lack computing power. Furthermore, the transition to fully automated and autonomous operations necessitates machine learning, enabling capabilities that traditionally are limited to human pilots, such as passively detecting and avoiding airborne traffic and identifying safe landing locations.

Embedded computing professionals still need to gain experience with artificial intelligence (AI). They need to gain knowledge on this topic and may need help building systems that support and incorporate AI. On top of the constraints for low size, weight, and power, making achieving high computational performance difficult, they face two more challenging circumstances when they are looking to incorporate AI into their systems. First, machine learning (ML) and neural network applications are computationally hungry beyond anything the embedded industry has seen before.

And second, no ML application has yet been certified by aviation regulators – and this raises a high bar for developers on both software and hardware levels. For example, processor manufacturers may typically withhold details about how a multi-core processor’s shared cache works (e.g., when cache lines are evicted), despite the common understanding that cache operation significantly impacts application performance. Without understanding the behavior of multi-core processors, system developers can struggle to introduce interference and guarantee mitigation of all potential failure conditions. Did you know that when a 4-core processor is used for avionics, up to three cores have to be disabled for the purpose of certification because the interaction between the cores cannot be sufficiently managed for the certifying authorities?

Intel®, the silicon manufacturer, and Daedalean, the Swiss-based company delivering Situational Intelligence™ for aviation and emerging advanced air mobility, have created a reference architecture that makes it easier for companies to incorporate what they have coined situational intelligence – the ability not only to understand and make sense of the current environment and situation but also anticipate and react to a future situation – in the cockpit. After years of working with the algorithms for this machine-learned aviator and testing them on various equipment in the lab and onboard aircraft in flight, Daedalean came to understand how the application capabilities can be enabled, starting from the silicon.

The reference design is based on the 11th Gen Intel® Core™ i7 and Intel® Agilex™ F-Series FPGA. This architecture satisfies all requirements for future certifiable machine-learned avionics systems requiring high-performance computing at low SWaP. One example application is a vision-based situational awareness system that leverages neural networks with high-resolution, high-throughput camera inputs. We discussed key challenges and constraints limiting silicon selection in the whitepaper and provided a reference architecture based on this.

Intel’s 11th gen Core™ i7 processor offers a unique advantage for aerospace suppliers since Intel has introduced the Airworthiness Evidence Package (Intel® AEP). It provides aircraft-embedded manufacturers with processor artifacts and the analysis and mitigation of non-deterministic and unintended behavior. The AEP supports DO-254 certification up to design assurance level (DAL) A.

By utilizing CPU solutions from Intel, Daedalean created a system design for a sensor computer powerful enough to process AI/ML for use in aviation applications and with the available documentation to support certification.

Recently, Intel published the jointly written whitepaper. This whitepaper addresses the challenges of implementing ML in certifiable aerospace applications. It is the first document ever to present a real-world working example and provide guidance on how to approach these challenges in general: how to ensure that your ML-based system can meet the computational requirements, certification requirements, and the size, weight, and power (SWaP) limitations at the same time. The approach described in the document paves the way for a new generation of airworthy equipment, driving the aviation industry’s need for high-performance embedded computing.

Source