Remaining useful life prediction combining temporal convolutional network with nonlinear target function

Maintenance is a critical aspect of complex products through entire life cycle, often requiring coordination of production planning and available resources, while previous studies appear to have rarely addressed. With this in mind, this paper presents a prescriptive maintenance framework based on digital twins for reducing operational risk and maintenance costs of complex equipment clusters. Virtual entities are firstly constructed for each single asset in multiple dimensions, which use real-time or historical sensing data collected from the physical entities to predict the corresponding remaining useful life (RUL). Then such RUL information is incorporated into a stochastic programming model with chance constraints to enable dynamic decision making. In particular, a risk-based optimization model is formulated to take full account of the physical distances between facilities and production gaps. Further, a dual-sense pyramidal transformer model is proposed to sense important details of data in both time and space while capturing temporal dependencies at different scales. We have demonstrated through a maintenance case of turbofan engines that the proposed scheme significantly lowers total maintenance costs while reducing frequent visits from maintenance personnel.

Section: Deep Learning For Remaining Useful Life (Rul) Predictionmentioning

confidence: 99%

Prescriptive maintenance for complex products with digital twin considering production planning and resource constraints

Mao

Liu

Qiu

et al. 2023

“…For decades, as one application of signal processing techniques, fault vibration signal analyses have been studied in numerous techniques, for instance, spectral kurtosis (SK) [1,2], empirical wavelet transform [3], empirical mode decomposition [4,5] and deep learning [6,7]. These studies have made great processes in improving accuracy of bearing fault diagnosis.…”

Section: Introductionmentioning

confidence: 99%

Sparsity-assisted signal decomposition via nonseparable and nonconvex penalty for bearing fault diagnosis

Liao,

Huang,

Qiu

et al. 2024

Monitoring vibration signals from a fault rotatory bearing is a commonly used technique for bearing fault diagnosis. Owing to harsh working conditions, observed signals are generally contaminated by strong background noise, which is a great challenge in extracting fault bearing signal. Sparsity-assisted signal decomposition offers an effective solution by transforming measured signals into sparse coefficients within specified domains, and reconstructing fault signals by multiplying these coefficients and overcomplete dictionaries representing the abovementioned domains. During the process, observed vibration signals tend to be decomposed, and fault components are extracted while noise is diminished. In this paper, a nonseparable and nonconvex log (NSNCL) penalty is proposed as a regularizer for sparse-decomposition model in bearing fault diagnosis. A convexity guarantee to the sparse model is presented, so globally optimal solutions can be calculated. During the process, tunable Q-factor wavelet transform with easily setting parameters, is applie din signifying multi-objective signals with a sparse manner.Numerical examples demonstrate advantages of the proposed method over other competitors.

“…Data-driven RUL prediction methods, which aim to establish nonlinear mapping relationships between the RUL and collected monitoring data, do not require specialized knowledge and have a better generalization performance [12]. It has swept the RUL prediction field and mainly includes traditional machine learning-based methods [13] and deep learning-based methods [14]. Compared with machine learning-based methods, deep learning-based methods do not need to rely on complex feature engineering and can achieve end-to-end automatic extraction of feature information from raw monitoring data with a better prediction performance [15,16].…”

Section: Introductionmentioning

confidence: 99%

Multiscale global and local self-attention-based network for remaining useful life prediction

Zhang,

Song,

et al. 2023

Remaining useful life (RUL) prediction plays an important role in Prognostics Health Management (PHM) and can significantly improve equipment safety and reliability. Recently, while deep learning based methods have swept the RUL prediction field, most existing methods still have difficulties in simultaneously extracting multi-scale global and local degradation feature information from raw multi-sensors monitoring data. To address these issues, we propose a novel Multi-scale Global and Local Self-attention based Network (MGLSN) for RUL prediction. MGLSN consists of the global and local feature extraction sub-networks based on self-attention, which work in parallel to simultaneously extract the global and local degradation features of equipment and can adaptively focus on more important parts. While the global network captures long-term dependencies between time steps, the local network focuses on modeling local temporal dynamics. The design of parallel feature extraction can avoid mutual influence of information from global and local aspects. Moreover, MGLSN adopts the multi-scale feature extraction design (multi-scale self-attention and convolution) to capture the global and local degradation patterns at different scales respectively, which can be combined to better reflect the degradation trend. Experiments on widely used CMAPSS, N-CMAPSS and PHM08 challenge datasets provided by National Aeronautics and Space Administration (NASA) show that MGLSN significantly outperforms the state-of-the-art RUL prediction methods.