JAXA Develops Fault Detection Algorithms for the Health Management of Space Propulsion Systems
MATLAB Allows Engineers to Develop Machine Learning Algorithms Without Extensive Data Science Experience
“The machine learning app in MATLAB was absolutely instrumental in building the propulsion system fault diagnostic model. The application was able to simply rank the features and utilize them to build and evaluate a classification learning model. We were able to evaluate the computational cost, versatility, and accuracy of the models through trial and error with a variety of models. It is an excellent toolbox for those who do not have in-depth knowledge of machine learning to introduce machine learning to their domain.”
Key Outcomes
- Deploying MATLAB for predictive maintenance on space propulsion systems reduced development cycle time and costs
- MATLAB enabled end-to-end development of fault diagnosis and predictive maintenance applications
- An on-site standalone app enabled immediate evaluation and characterization of measurement data
The Japan Aerospace Exploration Agency (JAXA) is developing prognostics and health management (PHM) technology for spacecraft propulsion systems to improve the safety and reliability of future missions to the moon, Mars, and deep space. The target is thrust anomalies caused by filter clogging or valve fault in the supply system. Recently, the requirements for propulsion systems have become more demanding, and there is a need to identify these faults quickly and accurately.
To address this challenge, JAXA is developing a pressure-sensing method using a noninvasive fiber Bragg grating (FBG) sensor to expand the amount of information available in the propulsion system. Using MATLAB® and Signal Processing Toolbox™, the team preprocessed the time-series data, calculated the FFT, and applied a high-pass filter. The surge data is synchronized with the peak detection function.
In addition, the team utilizes Predictive Maintenance Toolbox™ to interactively explore and rank features used to train and compare machine learning models. This approach allows JAXA engineers to develop, evaluate, and deploy PHM algorithms with a high degree of accuracy.
Using these tools, JAXA is effectively addressing the challenges associated with propulsion system health monitoring in space applications, driving improvements to make spacecraft operations safer and more reliable even when engineers do not have extensive data science experience.
