PhD Proposal by Johnie Sublett
[Advisor: Prof. Mavris]
will propose a doctoral thesis entitled,
A Framework for Anomaly Detection and Fault Management in Human Exploration Systems
Friday, August 19 at 8:00 a.m.
Conference Room #304
Weber Space Science and Technology Building (SST II)
One tap mobile:
997 7434 9126
Dangerous environments, physically and cognitively demanding activities and procedures, and highly specialized support systems make human spaceflight an incredibly difficult operational envelope. Deep-space exploration and planetary environments will be even more challenging, and push the capabilities of current space suits, exploration habitats, and their management systems.
Surface suits will require greater flexibility for the wearer, higher durability in harsh environments, and lower energy expenditure during use. Current space suit technologies rely on gas-retaining liners to maintain their breathing atmosphere and proper body pressurization. Planetary environments will be much more demanding, with surface suits requiring greater flexibility for the wearer, higher durability in harsh environments, and lower energy expenditure during use. Although some solutions have been proposed to improve the effectiveness of contemporary suits for long-term surface operations, there remains a risk of asphyxiation due to accidental suit puncture, especially along the limbs. Much of this risk can be mitigated through the implementation of two systems: a sensor array that is able to detect the location of a suit puncture along limb sections and a series of emergency inflatable cuffs capable of producing air-tight seals at strategic points along the body.
To this end, a hardware-in-the-loop pressure-suit testbed was developed, fabricated, and integrated with an embedded pressure sensor array to demonstrate suit recovery in the event of a puncture, with fault detection provided by embedded IoT devices conducting real-time machine-learning inference and passing resulting anomaly data to a notional habitat management framework.
Today’s spaceflight operations rely on a sophisticated orchestration among the on-orbit crewmembers, hundreds of habitat interfaces and systems, and dozens of terrestrial flight controllers and support staff. For missions beyond cis-lunar space, communications round-trip delays and blackouts will necessitate greater onboard automated systems management, fault-detection systems, and independent incident response, as prognostication and assistance from the Earth could be inherently delayed by twenty minutes or more. Using an increased focus on crew resource management and autonomous systems, these concerns can be managed. The hundreds of interfaces, data management systems, and ad-hoc experiment racks of the ISS can be enclosed within a standardized data reporting and instrument management framework, with reconfigurable controller GUIs and graphical alerts provided by containerized services, reducing crew cognitive impacts and standardize interfaces across the habitat. Wireless monitoring is enabled through the liberal use of IoT low-power devices, providing continual monitoring of critical systems. Autonomous monitoring can be enabled through Deep Neural Network anomaly detection methods, with the wealth of real-time data utilized for semi-supervised training. This large, centrally managed habitat-wide anomaly detection technique can then be distilled using device- and subsystem-specific student-teacher knowledge distillation. The result is a network of neural networks that enables distributed edge-device anomaly detection, providing rapid and redundant anomaly management on embedded devices. Detected anomalies can then be matched to previously predicted fault trees for root-cause estimation or impacted system assessments, providing real-time crisis support to onboard crewmembers.
Contributions are directed in four key areas: the physical-system testing of a notional active pressure sealing cuff on a space-suit like testbed; the deployed hardware-in-the-loop testing of anomaly detection methods on an embedded pressure-sensor array on said testbed; the definition, deployment, and testing of a habitat-focused orchestration, data-management, and fault-management system; and the testing of hybrid (partially distributed computing) anomaly and sensor-management methods on a notional habitat system and sensor swarm.
- Prof. Dimitri Mavris – School of Aerospace Engineering
- Prof. Spencer Bryngelson – School of Computational Science & Engineering
- Prof. Rich Vuduc – School of Computational Science & Engineering
- Dr. Alicia Sudol – School of Aerospace Engineering
- Dr. Carrie Olsen – NASA