AI Control for Preventing Engine Failures
Budget: €250 – €750 EUR
Problem Statement: Find control ranges for parameters to prevent engine failure
scenarios through the implementation of ML models.Objective:
The objective of this study is to develop an efficient and reliable ML model for PdM
of aircraft engines.
The primary focus is doing a comprehensive study of existing PdM techniques and
ML models used in the aviation industry as base models.
This study also aims to identify key indicators of engine health and optimise
the control parameters to remain within tested ranges obtained through the
model and methodology.
The impact of this study would be significant, potentially reducing maintenance
costs, minimizing downtime, and improving overall operational efficiency in the
aviation industry.
1. Oversampling, and Undersampling before modelling to make sure to have
balanced data.
2. Focus on each area of failure and non-failure state after modelling and see the
accuracy
3. Use the best-chosen model which predicts the Failure state properly, to be fed
into PDP and SHAP to see the ranges of control parameter framework for user
alerting.
4. Genetic algorithm, Random forest, Support Vector Machine
Not just predicting Remaining Useful Life (RUL) of engines like most C-MAPSS-based studies.
Instead, focusing on:
Controlling — Not Just Predicting
want to find the optimal sensor/control parameter ranges (e.g., pressure, temperature, vibration) that keep the engine in a safe/non-failure state — and use this to trigger real-time alerts before failure happens.
"Previous Work "
Main Goal: Predict RUL or failure time using ML models
Tool Used: Mostly Python, TensorFlow, Keras
Optimization Approach: Train DL models like LSTM or GRU for RUL
Real-Time Capability: Often missing or post-analysis only
Interpretability: Sometimes used SHAP for model explainability
System Goal: Estimate how long the engine will last
Work to do :
Main Goal:Prevent failure by optimizing control parameters to stay in safe range
Tool Used:KNIME (industry tool with low-code + integration + visual workflows)
Optimization Approach:Genetic Algorithms to find dynamic ranges that reduce failure risk
Real-Time Capability:Real-time integration and alerting using KNIME
Interpretability:Directly extract control ranges using SHAP + PDP for maintenance planning
System Goal:Build an early-warning + optimization system that tells: “If your sensor_11 exceeds this range, act now”
scenarios through the implementation of ML models.Objective:
The objective of this study is to develop an efficient and reliable ML model for PdM
of aircraft engines.
The primary focus is doing a comprehensive study of existing PdM techniques and
ML models used in the aviation industry as base models.
This study also aims to identify key indicators of engine health and optimise
the control parameters to remain within tested ranges obtained through the
model and methodology.
The impact of this study would be significant, potentially reducing maintenance
costs, minimizing downtime, and improving overall operational efficiency in the
aviation industry.
1. Oversampling, and Undersampling before modelling to make sure to have
balanced data.
2. Focus on each area of failure and non-failure state after modelling and see the
accuracy
3. Use the best-chosen model which predicts the Failure state properly, to be fed
into PDP and SHAP to see the ranges of control parameter framework for user
alerting.
4. Genetic algorithm, Random forest, Support Vector Machine
Not just predicting Remaining Useful Life (RUL) of engines like most C-MAPSS-based studies.
Instead, focusing on:
Controlling — Not Just Predicting
want to find the optimal sensor/control parameter ranges (e.g., pressure, temperature, vibration) that keep the engine in a safe/non-failure state — and use this to trigger real-time alerts before failure happens.
"Previous Work "
Main Goal: Predict RUL or failure time using ML models
Tool Used: Mostly Python, TensorFlow, Keras
Optimization Approach: Train DL models like LSTM or GRU for RUL
Real-Time Capability: Often missing or post-analysis only
Interpretability: Sometimes used SHAP for model explainability
System Goal: Estimate how long the engine will last
Work to do :
Main Goal:Prevent failure by optimizing control parameters to stay in safe range
Tool Used:KNIME (industry tool with low-code + integration + visual workflows)
Optimization Approach:Genetic Algorithms to find dynamic ranges that reduce failure risk
Real-Time Capability:Real-time integration and alerting using KNIME
Interpretability:Directly extract control ranges using SHAP + PDP for maintenance planning
System Goal:Build an early-warning + optimization system that tells: “If your sensor_11 exceeds this range, act now”