1
A Comprehensive Approach to Fault Classification of Helicopter
Engines with Adaboost Ensemble Model
Peeyush Pankaj
1
, Sammit Jain
2
, and Shyam Joshi
3
1,2
MathWorks India, Trillium Building, Blocks I & J, Embassy Tech Village, Bangalore, India 560103
ppankaj@mathworks.com
sammitj@mathworks.com
3
MathWorks USA, 5810 Tennyson Parkway Suite 425, Plano, TX, USA 75024
shyamj@mathworks.com
ABSTRACT
This work is based on the PHM North America 2024
Conference Data Challenge’s datasets of Helicopter turbine
engine performance measurements. These datasets were
large and moderately imbalanced. This submission produces
compelling results using MATLAB for all the necessary
visualizations, feature engineering, model exploration,
explainability, and confidence margin estimation. All these
tools will be generally applicable to data-driven AI/ML
modeling and predictions.
The MathWorks team score on Testing Data was 0.9686 at
the close of competition. This was further improved to
0.9867. The Validation Data submission scores were also
improved. Our approach demonstrates the effectiveness of
combining strategic data processing, feature engineering, and
model optimization. High prediction metrics and
explainability were demonstrated.
1. INTRODUCTION
The health monitoring of helicopter turbine engines is a
critical component in ensuring operational safety and
efficiency. With the increasing availability of large-scale
operational datasets, there is a growing opportunity to
leverage advanced data analytics and machine learning
techniques to enhance predictive maintenance strategies.
Such predictive maintenance strategies traditionally rely on
prognostics data trends over time and comparing these
against performance characteristic curves informed from
domain expertise (Bechhhoefer & Hajimohammadali, 2023).
However, this study addresses the PHM North America 2024
Conference Data Challenge, which involves predicting the
health of helicopter turbine engines by estimating torque
margins and classifying engine health status, where none of
the time-dependent information in the dataset was available.
The approach we used is described in Figure 1 below.
Figure 1. Block diagram representing workflow adopted for
failure prediction modeling on the helicopter engine dataset.
