Embedded Files
RUL Estimation of a Particulate Filter
RUL Estimation of a Particulate Filter
The objective of this project is to obtain an estimation of the remaining useful life (RUL) of a particulate filter. The PHM2020 dataset contains flow rate, filter upstream and downstream pressure data for flow across the filter with different sized particulate matter in the water.
Dataset
Dataset
The flow across the filter is dependent on the pressure differential across it, and the RUL of the filter in that instant (degree of clogging) among other factors. Here the pressure across the filter is seen to rise way more sharply in the case of larger particles.
Pressure Differential vs Time for Two Particles
Flow Resistance, a Derived Quantity vs Time
Models
Models
Random Forest Model
Random Forest Model
A Random Forest model for the remaining useful life (RUL) of the filter was trained. It was important to use a causal data for model training and evaluation. The estimated RUL is plotted by simulating the real experiment where data only upto a certain time stamp is fed into the model for RUL prediction. The RUL converges to zero as the filter pressure differential approaches the 20 psi end of life (EoL) limit.
K-Nearest Neighbors Model
K-Nearest Neighbors Model
A KNN model was also attempted, but due to the sparse nature of the dataset it did not perform as well as the RF model. The same 20 psi end of life (EoL) limit is imposed in this case.
All figures are mine.
Google Sites
Report abuse