1
A Multidisciplinary Framework for Vibration-Based Gear Fault
Diagnosis Using Experiments, Modeling, and Machine Learning
Lior Bachar
1
, Jacob Bortman
1
1
Departmant of Mechanical Engineering, Ben-Gurion University of the Negev, Beer-Sheva, 8410501, Israel
liorbac@post.bgu.ac.il
jacbort@bgu.ac.il
ABSTRACT
Vibration-based gear diagnosis is crucial for ensuring the
reliability of rotating machinery, making the monitoring of
gear health essential for preventing costly downtime and
optimizing performance. This study proposes a
multidisciplinary framework to enhance fault diagnosis, that
aligns with digital twin principles by integrating experiments,
dynamic modeling, physical preprocessing, and machine
learning. Within this framework, we focus on three core
procedures: domain adaptation to reduce discrepancies
between measured and simulated data; physical
preprocessing, grounded in in-depth investigations dictating
signal processing and feature engineering techniques; and
learning algorithms, encompassing the process of training
AI-based models. The framework is benchmarked through a
comprehensive case study of localized tooth fault diagnosis,
using controlled-degradation tests and realistic simulations.
First, we detect faults using unsupervised learning
algorithms; then, we use zero-shot-learning for classifying
between localized and distributed faults; finally, we adopt a
few-shot-learning strategy for severity estimation. Above all,
this hybrid framework aligns with the accelerating field of
physics-informed machine learning, by combining physical
knowledge and advanced algorithmics with machine
learning. This contributes to the PHM community by offering
valuable insights into integrating different aspects of
research, thereby enhancing performance in diagnosis tasks.
1. INTRODUCTION
Vibration-based gear diagnosis has made significant strides
over the years, resulting in a concise general framework that
typically encompasses data collection, signal processing,
feature extraction, and health indicator construction (Kumar,
Gandhi, Zhou, Kumar, and Xiang, 2020; Kundu, Darpe, and
Kulkarni, 2021). However, challenges persist, particularly in
the accelerating field of digital twins and physics-informed
machine learning (DENG et al., 2023). The availability of
labeled faulty measured data remains limited, a shortcoming
that must be acknowledged in any data-driven diagnosis
strategy. Furthermore, each fault type manifests differently,
necessitating tailored methods to overcome these differences
and recognize their unique characteristics in the signature in
early stages. Condition-based maintenance, as illustrated in
Figure 1, typically encompasses diagnosis, including fault
detection, classification, and severity estimation; and
prognosis, which involves remaining useful life estimation
(Kumar et al., 2020). This work contributes a comprehensive
framework for gear fault diagnosis that aligns with the
growing area of digital twins. Dynamic models can be
instrumental, both for bridging theoretical insights with
practical applications (Mohammed & Rantatalo, 2020) and
for generating synthetic training data (Bachar et al., 2023).
While other approaches do not necessarily rely on dynamic
modeling, we choose to incorporate them as indispensable
assets (Dadon, Koren, Klein, and Bortman ,2018). By
combining physical preprocessing with synthetic data, the
proposed framework aims to improve generalizability of AI-
based algorithms for fault diagnosis. Section 2 outlines the
proposed framework. Section 3 summarizes extensive
controlled-degradation tests in gears. Section 4 introduces the
general flow of incorporating dynamic models. Section 5
presents a case study on localized tooth fault diagnosis,
benchmarking the effectiveness of the proposed framework.
Figure 1. General stages of PHM.
2. THE MULTIDISCIPLINARY FRAMEWORK
Figure 2 presents a block diagram of the proposed
framework, which assumes that users have abundant labeled
healthy data, limited labeled faulty data (if any), and a rich
database of both healthy and faulty simulated data. The first
step involves using domain adaptation to enhance the
