A Novel Deep Learning Approach for Machinery Prognostics Based on Time Windows

Abstract

Remaining useful life (RUL) prediction is a challenging research task in prognostics and receives extensive attention from academia to industry. This paper proposes a novel deep convolutional neural network (CNN) for RUL prediction. Unlike health indicator-based methods which require the long-term tracking of sensor data from the initial stage, the proposed network aims to utilize data from consecutive time samples at any time interval for RUL prediction. Additionally, a new kernel module for prognostics is designed where the kernels are selected automatically, which can further enhance the feature extraction ability of the network. The effectiveness of the proposed network is validated using the C-MAPSS dataset for aircraft engines provided by NASA. Compared with the state-of-the-art results on the same dataset, the prediction results demonstrate the superiority of the proposed network.

Evaluating Image Classification Deep Convolutional Neural Network Architectures for Remaining Useful Life Estimation of Turbofan Engines

“…This indicates that a very deep model is not required for RUL estimation, so the benefits that the residual-and inceptionbased architectures offer are not realized. The applicability of CNNs on the N-CMAPSS data set, demonstrated in this work, and their past success on the original CMAPSS data set (Sateesh Babu, Zhao, & Li, 2016;Li et al, 2018;H. Yang et al, 2019), show the broad applicability that they have in RUL estimation.…”

“…This data set contains a large amount of run-to-failure data, so this work takes a data-based approach. Past research on the original CMAPSS data set demonstrated the applicability of convolutional neural networks (CNNs) (Li, Ding, & Sun, 2018;H. Yang, Zhao, Jiang, Sun, & Mei, 2019), long short-term memory (LSTM) (da Costa, Akeay, Zhang, & Kaymak, 2019;Zheng, Ristovski, Farahat, & Gupta, 2017;Wu et al, 2020), and hybrid methods, which merge CNN and LSTM (Zhao, Huang, Li, & Iqbal, 2020) for RUL estimation.…”

Fusing physics-based and deep learning models for prognostics

“…Deep CNN architectures have been proven to provide excellent performance on time series for predictive maintenance in recent works [6,35,[38][39][40]. Concretely, a solution based on 2D convolutions with 1D filters and no pooling layers was proposed in [6] for the prognostics benchmark problem [11].…”

Aircraft Engine Run-to-Failure Dataset under Real Flight Conditions for Prognostics and Diagnostics

“…Since its release as PHM Challenge [10] in 2008, the CMAPSS dataset [11] has been one of the most widely used prognostics datasets. Some recent examples that are also among the best performing prognostics models applied to the CMAPSS dataset are deep learning based methods such as convolutional neural network (CNN) [12,13], long shortterm memory networks (LSTM) [14][15][16][17][18][19] or hybrid networks combining CNN and LSTM layers [20,21]. The CMAPSS dataset provides simulated run-to-failure trajectories of a fleet comprising large turbofan engines.…”