Jiawei Xiong’s research while affiliated with Nanjing University of Science and Technology and other places

In practical engineering, there are many challenges when solving multi-objective optimization problems, such as the black box characteristics and time-consuming evaluation. The traditional evolutionary optimization method is usually limited due to the expensive cost and the difficulty in obtaining solutions. To modify the deficiency, a data-driven Bayesian SVR adaptive modeling technique and a constrained multi-objective surrogate-based optimization method is proposed in the context of the small sample. The Bayesian SVR model is first utilized to replace the complex computer model, thus reducing the expensive cost of every call to the actual performance function. Then, the new design by maximizing the aggregation strategy of the constrained expected improvement matrix is chosen. Next, the sample information and the data-driven Bayesian SVR model is adaptively updated, and the optimization is fulfilled step by step. The superior characteristic of the Bayesian SVR model, that is, the powerful ability to explore the boundary and the measurement of the prediction uncertainty, ensures the prediction accuracy and provides an improvement direction for selecting the new sample. In addition, the proposed Chebyshev distance and Manhattan distance aggregation strategy has the advantages of low computational complexity and good applicability for multivariable optimization problems. Test functions and engineering examples show that: 1) The proposed method can effectively reduce expensive simulation costs and improve optimization efficiency for expensive constrained multi-objective problems in the context of the small sample; 2) The Pareto frontier of surrogate-based multi-objective optimization has a certain degree of superiority in convergence, diversity, and space dispersion.

Limited availability of representative time-to-failure (TTF) trajectories either limits the performance of deep learning (DL)-based approaches on remaining useful life (RUL) prediction in practice or even precludes their application. Generating synthetic data that is physically plausible is a promising way to tackle this challenge. In this study, a novel hybrid framework combining the controlled physics-informed data generation approach with a deep learning-based prediction model for prognostics is proposed. In the proposed framework, a new controlled physics-informed generative adversarial network (CPI-GAN) is developed to generate synthetic degradation trajectories that are physically interpretable and diverse. Five basic physics constraints are proposed as the controllable settings in the generator. A physics-informed loss function with penalty is designed as the regularization term, which ensures that the changing trend of system health state recorded in the synthetic data is consistent with the underlying physical laws. Then, the generated synthetic data is used as input of the DL-based prediction model to obtain the RUL estimations. The proposed framework is evaluated based on new Commercial Modular Aero-Propulsion System Simulation (N-CMAPSS), a turbofan engine prognostics dataset where a limited availability of TTF trajectories is assumed. The experimental results demonstrate that the proposed framework is able to generate synthetic TTF trajectories that are consistent with underlying degradation trends. The generated trajectories enable to significantly improve the accuracy of RUL predictions.

Recent advances in multivariate data fusion technology have promoted the applications of neural network-based models for remaining useful life (RUL) prediction. However, the interpretability of these models is usually poor since they are developed in a black-box manner. It is difficult to use them in engineering systems with multiple failure modes (FMs) under various operation conditions (OCs). This work proposes an adaptive deep learning-based RUL prediction framework with FM recognition. First, a FM recognizer fusing physics-informed FM classifier with deep convolutional neural networks (DCNN) is developed, which improves the interpretability and the accuracy of the recognition model. Then, a framework which can adaptively train models and select them for RUL prediction according to FM recognition results is presented. An OC-based smoothing technique is proposed to improve the RUL prediction accuracy and robustness. Extensive experiments based on turbofan datasets are conducted to validate the effectiveness of the proposed framework. The results show that the RUL prediction accuracy is improved by 7% under the proposed framework when compared with other methods. It proves the performance gains of the proposed framework by incorporating prior FM recognition with RUL prediction. It also provides insights for RUL prognostics subject to distinct FMs and OCs.