hello people! sooo I’m working on a research project , the project is about finding anomalies in a different way , standard methods include a threshold based monitoring system which is simple to implement yet it has its own problems like it misses sub-threshold faults for example a 5% solar degradation that are critical but below the 15% alarm line and they also cause an alarm fatigue through cascading alerts what i mean by this is that it can trigger 50 different false alarms from just one fault and the scientist or the operator often gets confused and the satellite dies, then a team of scientists at NASA invented LSTM but it lacks explainability and its a black box ML meaning that it takes weeks or even months of training and could hallucinate just like any other existing AI model which means its unstable so to solve this problem im using physics and i basically made a framework with a causal inference engine
the way this works is that instead of just looking at raw data points and guessing what’s wrong, the engine uses the actual laws of physics to understand the “why” behind the telemetry. because it’s a causal model, it knows that if Variable A changes, Variable B must react in a specific way based on physical constraints. this lets the system distinguish between a real structural failure and just noisy sensor data, which stops the cascading false alarms.
it solves the “black box” problem because you can actually see the logic path the engine took to find the fault so it’s not just a probability, it’s a physical derivation. this makes it way more stable than an LSTM because physics doesn’t change, so there’s no “hallucinating” or needing months of training data to understand how a satellite should behave in orbit. anyway, i’ve reached a certain limit of doing things alone and i cant keep up with the development anymore, so its open for contributions! 
Repository: GitHub - rudywasfound/aethelix: Causal Inference for Multi-Fault Satellite Failures · GitHub
DOI: Aethelix: A Causal Inference for multi fault scenarios on a satellite.