Condition Base Maintenance of Naval Propulsion Plants

This project titled Condition Based Maintenance of Naval Propulsion Plants deals with experiments were conducted using a numerical simulator of a naval vessel (Frigate) characterized by a Gas Turbine (GT) propulsion plant with the aim of determining the performance decay or degradation of the vessel. Condition Base Maintenance is also known as predictive maintenance. Three measures of the performance decay of the vessels considered were: Ship speed (linear function of the lever position lp), Compressor degradation coefficient kMc and Turbine degradation coefficient kMt.

It makes good sense to develop a model capable of predicting the performance decay of any given vessel, as a predictive maintenance which is a form of preventive maintenance is usually more cost-effective than corrective maintenance. See

In this project, a number of regression models were considered to see which model will be the best for predicting the performance decay of naval propulsion plants.

Data Source: Machine Learning Repository

Data Features:

1. Gas Turbine shaft torque (GTT) [kN m]
2. Gas Turbine rate of revolutions (GTn) [rpm]
3. Gas Generator rate of revolutions (GGn) [rpm]
4. Starboard Propeller Torque (Ts) [kN]
5. Port Propeller Torque (Tp) [kN]
6. HP Turbine exit temperature (T48) [C]
7. GT Compressor inlet air temperature (T1) [C]
8. GT Compressor outlet air temperature (T2) [C]
9. HP Turbine exit pressure (P48) [bar]
10. GT Compressor inlet air pressure (P1) [bar]
11. GT Compressor outlet air pressure (P2) [bar]
12. Gas Turbine exhaust gas pressure (Pexh) [bar]
13. Turbine Injecton Control (TIC) [%]
14. Fuel flow (mf) [kg/s]

Selected (Utilized) Features:

1. Gas Turbine shaft torque (GTT) [kN m]
2. Gas Generator rate of revolutions (GGn) [rpm]

Data Targets:

1. Ship speed (v) [knots] (a linear function of Lever position (lp) )
2. GT Compressor decay state coefficient.
3. GT Turbine decay state coefficient

Feature Selection Method Used:

The Variance Inflation Factor (VIF) was used to check for the presence of multicollinearity between the features. It was found that when considering all the features, they had very high VIF's which were greater than traditional benchmark of a VIF of 5 or below and above the permissible benchmark of a VIF 10 or below (suggest by some scholars). When considering three of more features, they had high VIF's still. Hence, only two features were selected because their VIF's were less than 5.

Data Visualization:

The relationship between Gas Generator rate of revolutions and ship speed is a curve, which means that a linear model is not suitable.

The relationship between Gase Turbine shaft torgue and ship speed is a curve, which means that a linear model is not suitable.

Regression Techniques used:

1. Linear Regression
2. K-Nearest Neighbor (KNN)
3. Support Vector Machine (SVM)
4. Decision Tree (DT)

Evaluation Metrics:

1. Coefficient of Determination
2. Mean Squared Error

The best Model:

When the using the metric Mean Squared Error, the model with the lowest mean squared model is the best among all models under consideration. However, while using the Coefficient of Determination, the model with the highest Coefficient of Variation is preferable. Using both metrics, the Decision Tree came up as the best model

View Code on Kaggle

Use API