Physics-Informed Deep Neural Network for Bearing Prognosis with Multisensory Signals

Abstract

Prognosis of bearing is critical to improve the safety, reliability and availability of machinery systems, which provides the health condition assessment and determines how long the machine would work before failure occurs by predicting the remaining useful life (RUL). In order to overcome the drawback of pure data-driven methods and predict RUL accurately, a novel physics-informed deep neural network, named degradation consistency recurrent neural network, is proposed for RUL prediction by integrating the natural degradation knowledge of mechanical components. The degradation is monotonic over the whole-life of bearings, which is characterized by temperature signals. To incorporate this knowledge of monotonic degradation, a positive increment recurrence relationship is introduced to keep the monotonicity. Thus, the proposed model is relatively well-understood and capable to keep the learning process consistent with physical degradation. The effectiveness and merit of the RUL prediction using the proposed method are demonstrated through vibration signals collected from a set of run-to-failure tests.

Challenges on prognostics and health management for wind turbine components

“…Generally, interpretable ML model training is critical for improving the transparency and trustworthiness of ML models [49]. One way to improve this interpretability is to include physical knowledge into the model [4,70], converting them into hybrid models [44]. The parallel fusion structure, integrating physics models and ML modules, provides the advantage of leveraging the strengths of both approaches simultaneously, playing the role of compensators in improving accuracy, robustness, and interpretability while enabling a comprehensive understanding of systems' degradation behavior [69].…”

“…In pursuit of this objective, an emerging field known as physics informed neural networks (PiNNs) is raising. PiNNs can assimilate features aligned with scientific principles, thereby progressing towards the development of interpretable and more accurate deep neural models [70]. For instance, well-known physical fatigue damage models can be combined with data-driven layers to model degradation processes in bearings, finally merging into a cumulative damage model for RUL prediction [71].…”

A Novel Deep Model with Meta-learning for Rolling Bearing Few-shot Fault Diagnosis

“…With the advancement in technology, the structure of modern machinery and equipment is becoming increasingly complex, meanwhile, the high requirements of reliability and increased precision must be met. In recent years, deep learning-based intelligent fault diagnosis techniques have attracted a lot of attention due to their merits, such as robust feature extraction capability, effective processing models, cost-effective in calculation and analysis [1]- [5]. However, the excellent performance of these deep models is based on massive amounts of labeled data [6], [7].…”

Clustering Federated Learning for Bearing Fault Diagnosis in Aerospace Applications with a Self-Attention Mechanism

“…Various research on essential diagnosis issues, such as deep learning methods [3,4], knowledge transfer [5][6][7][8][9], fault decoupling and detection [10][11][12], imbalance data augmentation, and model generalization [13][14][15][16], have been carried out. For example, Syed Muhammad Tayyab et al [17] used machine learning through optimal feature extraction and selection for intelligent fault diagnosis of machine elements.…”