Goal
In this project, I was given a simulated faulty hydraulic pressure sensor model and accompanying data. I was tasked with creating a new, more accurate model that generalizes better. The original sensor data is plotted in the top image on the right. Clearly, there is significant deviation between the desired output and the original model.
Method
In order to solve this problem, I tested various filters and regressors. I built and compared linear, quadratic, cubic, trigonometric, and logarithmic models. I also tried various filters to reduce noise in the data. Eventually, I arrived on a 2nd order polynomial regression model that utilized a median filter. All of this project was done using Matlab.
Results
- Improved R² from 0.057 to 0.96
- Reduced RMSE from 0.02 to 0.005
- Tested the robustness of the model with bootstrapping of 100 model fits and found high generalizability
- Computed the covariance matrix and Mahalanobis distance (0.019) to show model generalizes well to unseen data
- Bottom image illustrates the improved fit compared to the desired output