In this project we propose a method for detecting sensor anomalies in time-series data based on the difference between the predicted and actual output of a sensor. We use a linear regression model to predict the output of a sensor based on the outputs of sensors on other channels. We additionally validate our model on fault-free data to determine how well our model correlates with actual sensor output. Then, we can run our model in real time and can detect sensor anomalies when the correlation between the actual and expected sensor data is far from the expected correlation.
Read the full report at https://github.com/devYaoYH/hackillinois_2020/blob/master/presentation/hackmath.pdf
We learned a lot about statistical techniques, and learned even more when trying to implement ones we already knew. It's a lot easier to just write these things down on paper then to try to calculate them efficiently!
We mostly built in python and additionally wrote up a report in LaTeX!
Log in or sign up for Devpost to join the conversation.