![](https://csdnimg.cn/release/download_crawler_static/16539945/bg4.jpg)
on the proposed method, a hybrid approach integrating EEMD, data analysis and modified GPR, as shown in Fig. 4, is carried
out.
The raw degradation
D
yðtÞ is calculated using Eq. (2):
D
yðtÞ¼yðtÞyð0Þ ð2Þ
where yðtÞ represents the measurement value at time t, yð 0Þ is the initial value, and
D
yðtÞ denotes the degradation variable of
the performance characteristic.
3.2. Time-frequency characteristic extraction and analysis of multiscale signal
In this section, EEMD and characteristic analysis are adopted to analyse the multiscale characteristics of signals. Here, the
performance characteristic of time series is considered as a hybrid signal consisting of multiscale components [3]. The cycle
information, mutable information and degradation tendency can be divided into different frequency scales using the decom-
position method. Characteristic analysis is performed to determine the main characteristics hidden in the divided subsignals.
3.2.1. Data decomposition based on EEMD
In contrast to traditional frequency domain decomposition approaches such as the fast Fourier transform (FFT) and WD,
EEMD is an empirical, direct and self-adaptive signal decomposition method that decomposes a time series based on its
traits. Although EEMD is based on EMD, it can effectively address the mode mixing that appears during EMD [31].
Fig. 3. Simplified structure diagram of the optocoupler.
Fig. 4. Framework of the proposed hybrid EEMD-GPR approach.
4 D. Xu et al. / Mechanical Systems and Signal Processing xxx (2017) xxx–xxx
Please cite this article in press as: D. Xu et al., RUL prediction of electronic controller based on multiscale characteristic analysis, Mech.
Syst. Signal Process. (2017) , https://doi.org/10.1016/j.ymssp.2017.11.036