1
Remaining Useful Life Prognostics of Rolling Element Bearings
Based on State Estimation Techniques
Zhen Li
1,2
, Konstantinos Gryllias
1,2
1
KU Leuven, Department of Mechanical Engineering, Celestijnenlaan 300, Heverlee, B-3001, Belgium
2
Flanders Make@KU Leuven, Belgium
zhen.li@kuleuven.be
konstantinos.gryllias@kuleuven.be
A
BSTRACT
Rolling element bearings (REBs) are key components in
rotating machines. 40% of the failures in electrical motors
occur due to bearing faults (Sharma, Abed, Sutton, and
Subudhi, 2015). Consequently, monitoring the health stage
and estimating the remaining useful life (RUL) of the REBs
is essential. Additionally, maintenance of rotating machines
can be scheduled based on the RUL estimation, which will
mitigate potential time wasting, economic losses and hazards
(Wen, Rahman, Xu, and Tseng, 2022).
Research on methodologies for estimating the remaining
useful life of REBs primarily falls into three categories: (a)
methods driven by artificial intelligence (Ma, Yan, Wang,
and Liao, 2023), (b) statistical approaches (Lim & Mba,
2015), and (c) physics model-based methodologies
(Gabrielli, Battarra, Mucchi, and Dalpiaz, 2024). The
authors’ research will focus on the integration of statistical
methods, such as the Kalman Filter (KF) and its variants, with
physics model-based approaches, e.g., lumped mass models,
to enhance the interpretability of RUL estimations under
steady and varying operating conditions. However, current
research may overlook several considerations in practical
applications, including:
(1) Health indicators (HIs) directly affect the accuracy
of RUL estimation. Therefore, they should reflect
the degradation trend. There are the following issues
regarding the HIs: (a) Current research mainly
focuses on vibration signal analysis based on the
cyclostationarity of REBs. The cyclostationarity of
REBs is successfully applied for fault detection and
fault diagnosis of REBs, obtaining exciting
effectiveness based on the comparison between
healthy cases and fault cases (Antoni, Xin, and
Hamzaoui, 2017). However, prognostics
necessitates indicators that are linked with the size
of the defect which theoretically increases with the
passage of time; (b) Few researchers pay attention
to the physics models for the prognostics of the RUL
because the crack inside the REBs cannot be directly
observed. On the one hand, the signal acquired from
sensors can deliver part degradation information,
e.g., the vibration amplitude will increase as the
defect extends intuitively. On the other hand,
fluctuation can appear because of the operation
condition changes, the complexity of the REBs’
structure and the stochasticity of the degradation. So,
how to connect the observable vibration data and the
unobservable crack size, which will improve the
accuracy and interpretability of prognostics, is a
potentially interesting topic.
(2) In the degradation process of REBs, anomaly
thresholds and failure thresholds are required to be
determined which are key parameters for RUL
estimation. The Anomaly thresholds determine
when the RUL estimation should start, i.e., the end
of the health stage (design life of components). The
failure thresholds are required for the RUL
calculation. They are points at which a component
is not allowed to continue to serve. However, these
two thresholds are difficult to be determined in real
cases.
(3) Due to the nonlinearity of the degradation, the
accuracy and stability of KF and its variants are
problematic. For instance, the Extended Kalman
Filter (EKF) uses the first derivatives to linearize the
state and measurement functions to approximate the
posterior probability. Consequently, a poor
approximation may be obtained.
(4) Online RUL estimation should be highlighted
because of the definition of prognostics. From the
practicability perspective, methodologies should be
able to estimate the RUL at the current time based
on historical data. If a methodology requests the
whole data to estimate the RUL of REBs, then it
would be difficult to implement it in the industry.
Zhen Li et al. This is an open-access article distributed under the terms o
the Creative Commons Attribution 3.0 United States License, which
ermits unrestricted use, distribution, and reproduction in any medium,
rovided the original author and source are credited.
