Machine Learning for Launch Assurance

According to the 30th Space Launch Wing Director of Operations, “If we don’t change something before 2022, we won’t be able to keep up with our planned launch schedule, it’s 3x what it is today. It’ll break us”.     The 30th & 45th Space Wings provide launch assurance for national launch missions. As the cost of launch continues to decrease, the frequency of launches increases with it.  However, the process, procedures, and technologies used for mission assurance has not kept pace. This proposal would provide the United States Space Force (USSF) with a much-needed solution to automate the pattern of life launch data for day of launch events.  The USSF needs a way to automatically assess if a launch sensor is within normal limits. Without automation,  it is impossible to keep up with the exponentially increasing number of sensors and the increasingly complex pattern of life behaviors.  The complexity of the task increases even further when one accounts for the growing number of launch providers.    The Department of Defense has accelerated continuous delivery through its DevSecOps journey, however, Artificial Intelligence and Machine Learning (AI/ML)  solutions are significantly lagging. Adoption of AI/ML capabilities is hindered by it’s lack of integration with common DoD DevSecOps reference architectures.  To incorporate AI/ML capabilities, the DoD must extend and mature the benefits of DevSecOps to include AI/ML tools and pipelines.  Platform One has built a DevSecOps compliant cyber security stack known as  “Big Bang”. Big Bang provides an Infrastructure as Code and Configuration as Code (IaC/CaC) platform that serves as the foundation for over 40+ weapon systems in the DoD. This foundation implements a scalable infrastructure platform that is ripe for expansion and maturation of capabilities. Machine Learning Operations (MLOps) is one such capability.  Machine learning operations (MLOps) have, thus far, been mostly “demoware” and challenging to adopt across the DoD.  There are a few notable exceptions to this, but a significant barrier to adoption lies within the ability to accredit and certify a platform that enables the development of algorithms that can deploy to production environments.  The challenge is not in developing ML algorithms or finding/determining the right commercial tools to leverage. The problem is and has been, developing a secure and accredible capability that is both infrastructure agnostic and incorporates the same cyber requirements as other DevSecOps systems.