Estimating the health of turbine engine based on the relationship
between torque margin and density altitude
Kosei Ozeki
1
, Takahiko Masuzaki
2
, Takeru Shiraga
3
, Koji Wakimoto
4
and Nakamura Takaaki
5
1,2,3,4,5
Mitsubishi Electric Corporation, Kamakura, Kanagawa, 2478501, Japan
ozeki.kosei@aj.mitsubishielectric.co.jp
masuzaki.takahiko@dc.mitsubishielectric.co.jp
shiraga.takeru@ea.mitsubishielectric.co.jp
wakimoto.koji@df.mitsubishielectric.co.jp
nakamura.takaaki@dy.mitsubishielectric.co.jp
ABSTRACT
We present an anomaly detection method developed for the
PHM North America 2024 Conference Data Challenge. This
competition is aimed at estimating the health of helicopter
turbine engines (PHM Society, 2024). The task includes the
estimation of the torque margin (regression) and the health
state (binary classification) of turbine engines. We developed
an estimation model using a hybrid algorithm that combines
data-based machine learning and domain knowledge-based
processing. Our method achieved scores over 0.99 for both
the testing and validation datasets. based on the calculation
rules provided by PHM Society. These results were ranked
first among all the participating teams.
1. INTRODUCTION
The Data Challenge was held as part of PHM 2024 (PHM
Society, 2024). This competition is an anomaly detection for
the health of helicopter turbine engines, and the authors
participated as challengers. Turbine engines deteriorate over
time and require regular maintenance (BHT-407-FM-3,
2018). However, engine data is rarely collected
automatically, making time-consuming trend analysis
difficult. Therefore, early detection of deterioration may not
be possible, and the timing of maintenance may be missed
(Bechhoefer, and Kessler, 2022). To address this issue,
machine learning-based methods such as SVM (Cao, Xu,
Huang, and Yang, 2022), (Chakraborty, Sarkar, Ray, and
Phoha, 2010) and Deep Learning (Huber, Palmé, and Chao,
2023), (Luo, and Zhong, 2017), as well as mathematical
approaches (Bechhoefer, and Hajimohammadali, 2023),
(Zhou, Zhou, Li, and Ca, 2023), (Tolani, Yasar, Chin, and
Ray, 2005), are being tried to monitor the performance and
health of the engine. We have been working on the
development of anomaly detection technology and have
proposed various methods. (Nakamura, Imamura, Mercer,
and Keogh, 2020), (Nakamura, Mercer, Imamura, and
Keogh, 2023). At PHMAP 2023 (PHM Society, 2023), The
Data Challenge was held to detect anomalies in spacecraft
propulsion systems. Our team proposed a time series
classification method using the k-NN algorithm and achieved
a score of 99.05%. This score was ranked third among all
participating teams. This method primarily relies on machine
learning algorithms, and we consider that using domain
knowledge is effective for further improving accuracy (Kato,
Kato, and Tanaka, 2023). In this Data Challenge, we
proposed an anomaly detection approach that actively
leverages domain knowledge.
2. PROBLEM DESCRIPTION
In the Data Challenge, the task is to evaluate the health of
helicopter turbine engines. This chapter presents the tasks and
datasets of the Data Challenge.
2.1. Task
This task involves two problems: torque margin regression
and the health state classification. The torque margin is an
indicator of engine health. The health state is classified as
healthy or faulty. Both regression and classification are
evaluated separately, with the final score being the average
of these two. Participants are required to estimate the engine
health using the testing and validation datasets and submit
their model's results. They are also required to provide a
confidence level (class_conf) for their classification results.
A severe penalty is imposed for high-confidence false
negatives (instances where the engine is predicted to be
healthy but is in fact faulty) because overlooking an engine
failure could lead to expensive repair costs.
