A Novel Multiscale Deep Health Indicator with Bidirectional LSTM Network for Bearing Performance Degradation Trend Prognosis

Abstract

As rolling bearings are the key components in rotating machinery, bearing performance degradation directly affects machine running status. A tendency prognosis for bearing performance degradation is thus required to ensure the stability of operation. This paper proposes a novel strategy for bearing performance degradation trend prognosis, including health indicator construction techniques and a performance degradation trend prediction method. To more accurately represent the degradation trend, the multiscale deep bottleneck health indicator is proposed as a new synthesized health indicator to remove high-frequency detail signals from features, which can reduce possible fluctuations in conventional synthetic health indicators. A suitable method for selecting the statistical characteristics required for fusion is also presented to solve the problem of information redundancy that affects trend representation. In addition, a stacked autoencoder network is used for deep feature extraction of selected statistical features. A bidirectional long short-term memory network prediction model is also proposed for the prediction of degradation trend, which can make full use of historical and future information to improve prediction accuracy. Finally, experiments are carried out to verify the effectiveness of the proposed method.

Fusing sensitive degradation features with uncertainty analysis for RUL prediction of rotating machines

“…In this paper, a neural network model based on ResNet-GRU is built. The output of the model is the fusion features index called ResNet-GRU-HI, referred to as ResG-HI for short [29,30]. Figure 4 shows its flow structure.…”

A new time–space attention mechanism driven multi-feature fusion method for tool wear monitoring

“…Then, two fully-connected layers are used to generate the attention weights. It is computed as, 𝑴 = 𝜎(𝑾 1 (𝑾 0 * 𝑭 𝑎𝑣𝑔 + 𝒃 0 ) + 𝒃 1 ), (9) where 𝜎 is the sigmoid function, 𝑊 0 , 𝑊 1 , 𝑏 0 , 𝑏 1 are the weight and bias of the first and second fully-connected layers respectively, 𝐹 𝑎𝑣𝑔 represents the squeezed state maps from GAP. Finally, the weighted attention weight and the output come from 1D-CNN is used as the last 1D-CNN layer.…”

“…It has the advantages of convenience and low cost. After years of research, numerous studies have shown that a single monitoring signal cannot exactly describe the state of machine work for monitoring in advanced manufacture [8][9][10]. With the rapid development of sensing and computing technologies, extracting different kinds features from different sensor signals to represent the tool wear status that makes RUL prediction more reliable [11] [12].…”

RUL prediction method for rolling bearing using convolutional denoising autoencoder and bidirectional LSTM

“…Chen et al [19] divided the LSTM into two stages before and after degradation points, obtaining different degradation features and using them as LSTM inputs to achieve higher prediction accuracy. Wang et al [20] proposed a method that decomposes fused features on multiple scales and removes high-frequency components to reduce volatility, then combines stacked self-encoders with Bi-LSTM models to achieve high-quality feature extraction and satisfied precision. Wang et al [21] proposed a prediction model of LSTM under an attention mechanism, which makes full use of past information under long sequences to improve the prediction ability.…”