Machine Learning in Aerospace{0}


By Louis M.

The aerospace industry is a complex and heavily data-reliant field which requires a great deal of research, design, and production for proper execution of its products and services. Machine learning has played an active role in the development of technology in aerospace to aid in this process, providing valuable information that would otherwise be difficult to obtain or unobtainable using traditional methods. The development of autonomous systems of control and execution for vehicles and aircraft has been a field of particular interest and help in this industry as well. As the development of this field progresses, it will continue to play a key role in the development of the industry as a whole.

The development of aerospace technologies is a long, costly, and arduous process with high risks in case of any product or individual part failures. The ability to use prior data from such failures under similar conditions is a promising field which may allow for predictive data analysis of ongoing and future projects to ensure clarity in the potential risks and rewards for a given project. In “Using Big Data and predictive machine learning in aerospace test environments,”(Armes, Refern 2013), this is elaborated. RMA, or “Returned Materials Authorization” is a system which processes the failures and necessary returns or scrapping of failed parts and tests which is involved in any projects’ ongoing development process. The potential for machine learning in this system and integration with the “R&D” (Research and Development) department is huge. The case in point of this application is as follows: “By posing the question: ‘These certain products have not failed yet, but are at risk of failing’ and having a solid justification for that probability of failure, the engineer/analyst can affect real change and give executives the information necessary to prevent disastrous product failures in the field.” Armes and Refern point out that the ability of machine learning applied in this manner, to organize and present data effectively in this way would prove a huge boon to all future aerospace development projects and increase the efficiency of these projects.

In addition to analysis, the aerospace industry is heavily involved in applying machine learning to autonomous systems. Since 2003, for example, JPL has been operating the “Autonomous Sciencecraft Experiment” onboard an orbiting satellite. It uses the “‘Continuous Activity Scheduling Planning Execution and Replanning’ software”, or CASPER, to aid in the autonomous study and analysis of “science events” which it can detect from orbit. It is able to “study … short-lived science events(such as volcanic eruptions, dust storms, etc.)” and “reduces instrument setup time by using autonomy software to take advantage of execution information to streamline operations”(jpl.nasa.gov). The ability of CASPER to more indpendently detect these “science events” from orbit and relatively independently create and execute plans of action rapidly for analysis purposes means that such events can be far more efficiently catalogued and studied by scientists for a variety of purposes. Private enterprise has been entering the field of autonomous systems in aerospace as well, with the company Near Earth Autonomy now working on a grant from NASA to develop autonomous systems for takeoff and landing vehicles and aircraft. The key point of focus for their research is the ability of such a vehicle, with the inclusion of a “Safe50 software module,” which will “accurately guide the [Unmanned Aircraft System] from take-off to landing in a fully autonomous manner, outside of the operator’s visual line of sight, without a direct link with a base station, and with intermittent GPS reception,” (Geiver 2016). This is a crucial field of development for the entire aerospace industry, and with ongoing advances in this sort of technology the pain and dangers of launch and landing operations has the potential to be reduced significantly, and without the heavy emphasis on simultaneous and complex outside controls (via controller/gps/nav) which is currently a necessity.

In short, the aerospace industry is one in which the applications of machine learning are crucial to most future developments of technology and system capability. Design and Production analysis for various technologies and projects, along with the creation of autonomous systems to aid in the function of these technologies are moving forward with haste and will continue to do so. The integration of these systems and technologies within aerospace research will allow for previously impossible scientific analysis of data, and previously impossible levels of control and safety for vehicles and aircraft. The cost-effectiveness and safety guarantees promised by machine learning integration is clear; industries within aerospace will undoubtedly continue efforts in this area as long as they prove to be fruitful, and at the moment the promise of this seems to be enormous.

References:

Armes, Tom, & Refern, Mark. (2013, September 19). Using Big Data and predictive machine learning in aerospace test environments. IEEE Xplore, Retrieved from
http://ieeexplore.ieee.org.proxy.library.cpp.edu/document/6645085/authors

(Author N/A, Publication Date N/A) Autonomous Sciencecraft Experiment. Retrieved from
http://www-aig.jpl.nasa.gov/public/projects/ase/

Geiver, Luke. (2016, June 09). Near Earth, NASA making first, last 50ft of UAS flight safer. UAS Magazine. Retrieved from
http://www.uasmagazine.com/articles/1496/near-earth-nasa-making-first-last-50-ft-of-uas-flight-safer