The analysis of remaining useful life (RUL) of systems
and equipment
enables the prevention of failures, so that effective maintenance
can be performed in time to correct failures that are close to happening.
The degradation signal of a variable can be used as a basis for estimating
the RUL of a given system since this signal is modeled mathematically
correctly. In this paper, the RUL of cyclic processes is analyzed
with the determination of the number of remaining cycles (NRC) in
order to maximize production, guaranteeing operational safety. Two
approaches will be considered: Bayesian methodology and time series.
The Bayesian methodology is based on Bayesian inference to update
the stochastic parameter, providing better representativeness in the
estimation of the NRC. The deterministic parameters and the hyperparameters
in the prior distribution of the stochastic parameter are estimated
through the maximum likelihood estimation method, while the stochastic
parameter in the degradation model of a system can be updated every
time a new degradation data is obtained. On the other hand, the time
series is based on training sets to be able to fit a model that is
similar to the set used for validation. In the estimation of NRC,
a stationary model (simple exponential smoothing), a nonstationary
model (double exponential smoothing), and a model that considers the
component of seasonality (triple exponential smoothing) are considered.
A case study of a temperature swing adsorption unit for natural gas
dehydration will be used to evaluate these two approaches in predicting
NRC in cyclic processes. In this case study, we propose a novel cycle-packaging
methodology that creates a new dimension, allowing the application
of NRC forecasting methodologies, which is the main contribution of
this article. The results suggest that the Bayesian methodology is
the most indicated in the NRC estimation, while the time series are
adequate to identify the cyclic pattern of the process.