Hi everyone!
I’m looking for a small number of collaborators for a research project on AI-based early detection of drift and sensor degradation in space environments. The project involves areas such as sensor modeling, ML/AI, space systems, or data analysis — depending on your background.
Because the project will be developed with support from Imperial College for testing, we’re keeping the team compact and focused.
More about the project:
AI Integrated Sensor Drift Prediction
We aim to teach sensors to tell the truth even when space tries to break them.
Space sensors drift due to radiation, temperature swings, vibration, and aging.
Even tiny drift threatens navigation and mission safety.
What is Sensor Drift?
A sensor should read 25°C**,** but begins reporting 28°C after months in orbit.
Small deviation → large mission risk.
Our Research Focus
AI to detect early drift, predict failure, and auto-correct sensor readings in real time or send alerts about the risk.
If you’re interested, you can either:
Reply directly to this thread with a short description of your background/skills, or
Hi, I would be interested if there was any way I could contribute. My background is more based in person centred quality and risk / incident prediction and management however. What I find fascinating in your project is how / if we can reliably tell early sensor drift or early failure warning from a genuine but unusual truth. I guess that that diagnostic boundary is more where my background fits.”
Hi. I,m happy to catch up if you would like.
I would be particularly interested in how one could separate the four key cases — genuine environmental truth, slow drift, emerging degradation, and the short-term causal anomalies you get from things like radiation hits or power transients, ensuring the AI preserves truth rather than producing stable but misleading outputs.
I’m based in Australia, so a wee bit geographically remote, but remote collaboration is straightforward for me.”
Great to meet you today! Interesting idea on your project, while it is somewhat outside of biology, it still may be interesting to pursue this between the @HardwareAWG and @AIMLawg ? I would contact @vaishnavi.nagesh to get tied into the AIMLawg meeting, happening Jan 13
To answer your question:
Potential data sources
Environmental Data App: https://visualization.osdr.nasa.gov/eda/ This is part of OSDR, and has real ISS telemetry, tied to specific experiments. Example: Compare [Rodent Research] RR-1 vs RR-3 temperature readings to identify [potential] drift patterns across missions?
Telemanom (I think via NASA JPL): https://github.com/khundman/telemanom - A search helped me find this one. Example: SMAP/MSL telemetry with labeled anomalies. LSTM + nonparametric thresholding detects drift without predefined limits.
Standards & Tools
NTRS: https://ntrs.nasa.gov/search This is NASA’s main ‘NASA Technical Reports Server’. A quick search helped me find this example: “Impact of Sensor Degradation on MODIS NDVI Time Series” which documents calibration drift correction methods (figure thats quite relevant?).
ProgPy (Prognostics Python Packages): https://nasa.github.io/progpy/ Also a search helped me find this (i hadnt heard of this before). NASA’s open-source library for prognostics/health management. Includes state estimation, prediction, and uncertainty quantification, it also could support your drift prediction pipeline.
Maybe
NASA Technical Standards: https://standards.nasa.gov/all-standards I think this is the main place to finded needed sensor hardware specs, workmanship requirements, or radiation hardness assurance documentation (in checking it out, NASA-STD-8739 appears to have five entries which covers electrical parts including sensor assemblies).
PHM Society Mirror: https://data.phmsociety.org/nasa/ Interesting, there seems to be a ‘Backup’ site to mirror as a download location for the Prognostics Data Repository datasets (I’m betting because sometimes NASA’s direct downloads are unavailable?). This is via a prognostics and health management society (PHM).
Questions to help you scope a bit what youre looking for:
Which sensor modalities (thermal, radiation, optical) are you targeting?
What ML approaches are you considering? LSTMs, autoencoders, or simpler baselines like KNN/isolation forests? Do you have AI/ML algorithmic expertise? If not, you (or someone you know) may also want to consider taking this AI/ML for space biology open-access on demand course: https://www.nasa.gov/using-ai-ml-for-space-biology-research/
Real-time edge inference or ground-based analysis? Ie., are you doing this more for hardware preparation for flight on the ground, or to develop real-time in situ analysis tool?
Context, I am NOT an engineer, but have tried to stay aware of their domain for a decade so I remember and learn some things along the way
Such useful resources Ryan! @Rinsdew great to meet you. It’s an interesting idea you have.
AI/ML AWG hpsts a monthly meeting on second tuesday every month. The next one would be On Jan13th. It would be great to have you attend that and see if any of the current topics interest you or if you want to propose a new topic.
I concur with Ryan on tying AIML interests with Hardware AWG to further your project. Connecting you to @drravichangle as well to advance both your interests.
Cheers,
Vaishnavi
Let’s connect @Rinsdew to explore the area of interests, I have been working for Federated Machine Learning, Data Center Optimizations and creating sustainability solutions leveraging Green AI practices.
Employing AIML for Space and ISS would be next level utility. Can also explore Multi Agent AI System for OSDR Data Exploration and Analytics.
Thank you @rtscott2001 for your nice words and help! Here’s our approach:
We’re primarily targeting Inertial Measurement Units (IMUs) - specifically gyroscopes and accelerometers used for spacecraft navigation and attitude control. These are the most mission-critical sensors where drift directly causes orientation errors, navigation failures, and mission loss.
Our initial focus is on IMU drift caused by radiation-induced degradation and thermal cycling, which are the primary failure modes in space. We’re building the foundational AI twin architecture with IMUs first, then expanding to other modalities (thermal, optical, pressure) once we validate the approach.
Thanks for the NASA AI/ML course link! We’ll definitely review it - particularly interested in their approaches to limited training data scenarios.
At this stage, SOHAO is primarily a ground-based analysis and validation tool intended for hardware preparation, characterization, and pre-flight testing.
Hi@drravichangle! Thanks for reaching out - would love to connect! Your background in Federated Learning and Green AI is really relevant to what we’re building.
Quick context on our project: We’re developing AI twins for spacecraft sensors (starting with IMUs) that predict drift and failures before they happen. Think predictive maintenance meets edge AI for space hardware.
Where your expertise could be valuable:
1. Federated Learning for Multi-Mission Data
We’re collecting sensor data from multiple spacecraft/missions.
Federated approach could let us train across missions without sharing raw data.
2. Green AI / Edge Optimization
Power budgets on spacecraft are BRUTAL. Model efficiency is critical
I would love to participate in the aforementioned research project.
I’m an embedded systems developer with over 35 years of experience in programming.
My experience with AI is in the form of long-term interest (and some development) in reinforcement learning algorithms and Expert Systems. My adventure with Machine Learning started in the late 90’s when I developed a discrete neuron: the VCL component library, which allowed it to be used (with multiple instances of the same components) as the simplest basic element to build neural networks in any RAD software compatible with the VCL technology.
Thank you for reaching out and for sharing your background. I really appreciate the depth and longevity of your experience, especially your work in embedded systems, early neural network development, and sensor research in space environments.
The project is currently very early-stage and moving quickly, with many aspects still being defined in parallel. That said, the core focus is on sensor reliability, drift prediction, and intelligent monitoring for space systems, so your background is highly relevant.
At this stage, rather than having rigidly defined roles, I’m looking to have conversations with people who are interested in contributing their expertise where it can have the most impact as the project takes shape.
If you’re open to it, I’d be happy to talk with you to walk through the current vision, constraints, and roadmap, and to explore where your skills could best fit within the project.
I built the navigation system with 63% improvement on real NASA datasets.
We are now developing a new filtering approach grounded in quantum mechanics, explicitly modeling multiple noise sources directly within the equations.
The project is moving into the onboard validation stage, where we will conduct tests on UAVs and onboard platforms to verify real-world performance and collect additional flight data.
At this stage, we are looking for guidance, collaboration, and access—whether through introductions to investors, domain experts, professors, laboratories, or anyone working deeply in spacecraft navigation, sensing, or autonomy. Even a short conversation or insight would be incredibly valuable.
I’m building this project from Turkey, where access to space-focused infrastructure and active missions is extremely limited. That constraint is exactly why I’m seeking international collaboration.
elssuna123@gmail.com
so I remember and learn some things along the way
