An Advanced Diagnostic Model for Gearbox Degradation Prediction
Under Various Operating Conditions and Degradation Levels
Hanqi Su
1
*
, Jay Lee
2
1,2
Center for Industrial Artificial Intelligence, Department of Mechanical Engineering,
University of Maryland, College Park, MD, 20742, USA
hanqisu@umd.edu, leejay@umd.edu
ABSTRACT
This study introduces a novel three-stage diagnostic method-
ology aimed at enhancing the prediction and classification
of gearbox degradation under various operating conditions
and multiple degradation levels, addressing the complex-
ities encountered in real-world industrial settings. Lever-
aging the latest advancements in data-driven approaches,
from similarity-based methods to residual-based deep con-
volutional neural networks (CNNs) and pseudo-labeling
techniques, our approach systematically classifies data into
known, unknown, and undetermined categories, predicts
known degradation levels, and refines classification mod-
els with augmented pseudo-label data. The efficacy of our
methodology is demonstrated through its remarkable perfor-
mance using the data from the PHM North America 2023
Conference Data Challenge. It achieves scores of 600 / 800
on the testing data and 574 / 813 on the validation data, sig-
nificantly surpassing the first-place scores of 463.5 and 472
in the competition, respectively, setting a new benchmark in
the field of gear fault diagnosis.
1. INTRODUCTION
Over the years, the gear fault diagnosis domain has witnessed
the evolution of numerous data-driven approaches aimed at
identifying and diagnosing gear faults and degradation to
ensure the reliability and efficiency of mechanical systems.
Existing research has shown variant machine learning (ML)
and artificial intelligence approaches for gear fault diagno-
sis (Kumar, Gandhi, Zhou, Kumar, & Xiang, 2020; Zhu et
al., 2023; Su & Lee, 2024), including convolutional neural
networks (CNN) (Zhao, Kang, Tang, & Pecht, 2017; Kreuzer
& Kellermann, 2023), recurrent neural networks (RNN) (Tao,
Wang, S
´
anchez, Yang, & Bai, 2019; Durbhaka et al., 2021),
autoencoders (AE) (Saufi, Ahmad, Leong, & Lim, 2020;
Hanqi Su et al. This is an open-access article distributed under the terms of
the Creative Commons Attribution 3.0 United States License, which permits
unrestricted use, distribution, and reproduction in any medium, provided the
original author and source are credited.
∗
Corresponding author: Hanqi Su (hanqisu@umd.edu)
Z. He et al., 2020), deep belief networks (DBN) (X. Wang,
Qin, & Zhang, 2018; Li, Li, He, & Qu, 2019), etc. How-
ever, in real-world industrial settings, analyzing and diagnos-
ing gearbox degradation may become more complex. Gear-
boxes may operate under a variety of working conditions, and
sometimes, some states of gear health are not known in ad-
vance. This uncertainty adds a significant layer of difficulty
to gear fault diagnosis. Meanwhile, the small size of the data
sets also presents challenges for ML model deployment. Con-
sequently, there is a growing need to explore and develop
innovative solutions that reduce reliance on specialized ex-
pertise and enable the creation of more versatile, automated
systems for gear fault diagnosis. These advancements hold
the promise of making gear fault diagnosis more accessible
and efficient, paving the way for broader applications and en-
hanced operational reliability.
To address these limitations, we propose a novel three-stage
diagnostic approach for predicting gearbox degradation. The
initial stage introduces a similarity-based model designed to
classify data into known, unknown, and undetermined cate-
gories. Subsequently, the second stage employs a residual-
based CNN regression model, focused on the prediction of
known degradation labels. In the final stage, we transition the
regression model to a classification model. We incorporate
pseudo-labeling techniques to assign pseudo-labels to testing
data. The data, now augmented with pseudo labels, is then
used to refine the classification model further. This innova-
tive approach enhances the model’s robustness and its gen-
eralization capabilities, offering a comprehensive solution to
the challenges of gearbox degradation prediction.
The rest of this article is organized as follows: Section 2 in-
troduces the competition and dataset, reviews the relevant ap-
proaches, and data preprocessing module. Section 3 provides
a comprehensive analysis of three-stage ML model construc-
tion. Section 4 reports and discusses the performances of ML
models and summarizes the limitations of this study. Section
5 concludes this paper by highlighting its findings and contri-
butions.
1
