XTRACTIS FOR PREDICTIVE MAINTENANCE
PREDICTING THE FUNCTIONAL DEGRADATION OF A NAVAL PROPULSION UNIT
BENCHMARK: XTRACTIS VS. RANDOM FORESTS, BOOSTED TREES & NEURAL NETWORKS
The aim of this study is to predict the functional degradation of a Naval Propulsion Unit compressor, from measurements on a gas turbine in a stable state obtained by digital simulation of frigate. It illustrates the ability of the xtractis Trustworthy AI to automatically induce knowledge in the form of predictive and intelligible mathematical relationships in order to model a very complex phenomenon.
In the end, xtractis confirms this great complexity, generates a regression model composed of 306 decision rules without chaining, using 12 predictors out of 14, and predicts a decay state for the infinity of points in the decision space with proven reliability.
- Help field experts and maintenance managers to understand the causal relationships between certain turbine parameters and its upcoming state of degradation.
- Carry out predictive maintenance actions specific to each turbine, thanks to quick and systematic diagnosis, thus avoiding critical damage.
- Modeling Type & Reference Data
- Xtractis Automatic Induction Process
- Top-IVE: Best Predictive & Intelligible Model
- Performance of the Top-IVE
- Intelligibility of the Model & Explainability of the Decision
- Example: Prediction for Simulation #6945
- Resources of the xtractis Process (Induction + Validation + Deduction)
- Benchmark xtractis versus Random Forest, Boosted Trees & Neural Networks
- Conclusions & Advantages of the xtractis Trustworthy AI