Schlumberger, October 2018 - Present
Principle Artificial Intelligence Scientist, Software Technology Innovation Center
- Working on prognostics and health management problems.
SGT Inc., NASA Ames Research Center, December 2010 - September 2018
Principal Artificial Intelligence Expert at Schlumberger [October 2018 – present]SGT Inc., NASA Ames Research Center [December 2010 – September 2018]
(Work in collaboration with M. J. Daigle)
- Task Lead for the ``Health Management of Aerospace Systems'' task at SGT Inc.
- Developed an automated framework for the real-time assessment and modeling of the safety of the National Airspace System (NAS).
- Developed methodologies for enhanced decision making abilities to assure resilient operations in the presence of adverse events.
- Developed an approach for diagnostic reasoning using prognostic information for unmanned aerial systems to improve diagnostic accuracy and enable decisions to be made at the present time to deal with events in the future.
- Developed algorithms for diagnosing faults in a Forward Osmosis Water Recycling System (WRS) deployed at NASA Ames Research Center and predicting when the WRS would need maintenance actions, such as changing of the membranes due to fouling.
- Assessed the impact of deploying unrestricted Unmanned Aircraft Systems (UAS) in the National Airspace System (NAS) of the United States through a simulation-based approach.
- Studied the effect of damage progression model choice on prognostics performance. Developed different damage progression models for a centrifugal pump, and investigated how model changes influence prognostics performance.
- Member of the SAE HM-1 Integrated Vehicle Health Management (IVHM) Committee developing an Aerospace Recommended Practices on the ``Guidelines for Writing IVHM Requirements for Aerospace Systems'' (ARP6883). This document is aimed at providing working engineers and program managers clear guidance on the systems engineering aspects of IVHM design in aerospace systems.
- Developed a systematic process for verifying prognostics algorithms as the prognostics algorithm. Demonstrated this process using a battery health management system deployed on an electric unmanned aerial vehicle as well as a planetary Rover.
- Developed an integrated distributed diagnosis and prognosis approach that is formally rooted in structural model decomposition, allowing the system model to be decomposed in order to define local (i.e., subsystem- and component-level) diagnosis sub-problems that can be solved in parallel. This research has been applied several complex real-world systems, for example, multi-tank systems such as used in propellant loading systems for NASA's ground based launch support systems, and a spacecraft power storage and distribution systems.
- Part of the team that built an electro-mechanical actuator (EMA) test stand that allows testing of EMA fault detection and prognosis technologies in flight environment, thus substantially increasing their technology maturity without injecting faults in any real aircrafts. This test stand is christened the FLyable Electro-mechanical Actuator (FLEA) testbed.
- Developed and implemented a prognostic health management system for the FLEA testbed that is aimed at detecting and tracking several fault types. Till date, the FLEA testbed and the algorithm on-board have completed two test flights of the stand on US Air Force C-17 aircraft and several more experiments on UH-60 Blackhawk helicopters.
- Modeled large scale hybrid systems using hybrid bond graphs and derived high-fidelity simulation models for the Advanced Water Recovery System developed at NASA Johnson Space Center.
- Developed a methodology for distributed fault diagnosis of complex physical systems. Implemented algorithms to analyze the diagnosability of continuous systems and design distributed fault diagnosers that communicate a minimal number of measurements amongst themselves to generate globally correct diagnosis results through local analysis, without any central coordinator. These algorithms have been applied to the Advanced Water Recovery System.
- Developed a unified framework for diagnosing abrupt (fast) and incipient (gradual) parametric faults by combining a qualitative diagnosis scheme with a Dynamic Bayesian Network-based diagnosis approach.
- Developed an approach for improving the efficiency of inference of Dynamic Bayesian Networks (DBNs) by factoring the DBN into smaller factors such that each factor is conditionally independent from all other factors given the measurements that are communicated to it.
- Developed a distributed scheme for diagnosing abrupt and incipient faults by combining qualitative diagnosis schemes with inference algorithms applied to factored DBNs.
- Developed models of a spacecraft electrical power distribution system called the Advanced Diagnostic and Prognostic Testbed (ADAPT) at NASA Ames Research Center that was used in empirical human-in-the-loop testing of a fault management interfaces for next generation spacecraft performed by NASA Ames researchers involving real pilots.
- Architected a software tool suite for the modeling and simulation of hybrid systems modeled as hybrid bond graphs, called the Modeling and Transformation of HBS for Simulation (MoTHS) that is being extended to the META project funded by DARPA.
- Developed high-fidelity hybrid models for the Advanced Diagnostics and Prognostics Testbed (ADAPT)
- Improved the efficiency of simulation software for hybrid bond graphs. Developed the VIRTUAL ADAPT simulator which is used by researchers as a portable alternative to ADAPT for ADAPT-related projects. Used by researchers at NASA Ames Research Center for experimental user interaction studies for the Advanced Caution and Warning System Project at NASA Ames Research Center.
- Implemented the Fault Adaptive Control Technology (FACT) software as a test article for ADAPT. Performed diagnosis experiments using FACT on ADAPT.
(Work in collaboration with M. J. Daigle)
- Developed component-based hybrid models for the Advanced Diagnostics and Prognostics Testbed.
- Developed software to construct and execute hybrid bond graphs as MATLAB Simulink models.
- Implemented a Low Complexity MPEG2 AAC Encoder
- Generated worst-case audio streams for comprehensive testing of decoders
- Conducted an in-depth study of various techniques for testing defects in welds.