Machine learning methods help accurate estimation of the hydrogen solubility in biomaterials

This study investigates the hydrogen solubility in five industrially relevant biochemicals, i.e., eugenol, furfural, furan, allyl alcohol, and furfuryl alcohol. Gene expression programming, three different artificial neural networks, and least-squares support vector regression (LS-SVR) are checked in this regard. The ranking analysis employing seven well-known statistical criteria confirmed that the LS-SVR is the most accurate model for the given purpose. The developed LS-SVR is superior to the Peng-Robinson, Soave-Redlich-Kwong, and perturbed-chain statistical associating fluid theory thermodynamic-based correlations proposed in the literature. The LS-SVR model predicts hydrogen solubility in the considered biochemicals with the relative absolute deviation of 1.91%, mean squared error of 6.4 × 10^?7, and regression coefficient of 0.99924. Both experimental and modeling observations approved that furan has the maximum tendency to absorb hydrogen molecules. A pure simulation analysis indicates that the maximum hydrogen solubility of 0.11 (mole fraction) can be absorbed by furan at pressure = 14.93 MPa and temperature = 402 K.     

Simultaneous fault diagnosis of proton exchange membrane fuel cell systems based on an Incremental Multi-label Classification Network

Fault diagnosis plays an important role in the operation of proton exchange membrane fuel cell (PEMFC) systems. In some certain working conditions, multiple faults can occur simultaneously. And, to the best of our knowledge, very few studies have yet to design an algorithm specifically for simultaneous fault diagnosis in PEMFC systems. Therefore, a novel simultaneous fault diagnosis algorithm, based on multi-label classifier chain named Incremental Multi-label Classification Network (IMCN), is proposed. To develop and optimize IMCN, a PEMFC model is constructed based on the commercial software AVL CURISE M to simulate data when simultaneous multiple faults occur. To further verify the generalization performance and practical effect of IMCN, a bench experiment using a high-power PEMFC system is conducted, which has similar boundary conditions as the boundary conditions embedded in simulation model. And, whether in experiment or simulation, corresponding verification methods are adopted to verify the success of simultaneous multiple faults embedding. Experimental data testing shows that, the subset accuracy, Hamming loss, Jaccard index, precision and recall of IMCN reaches 0.973, 0.029, 0.921, 0.961 and 0.956 respectively (better than Multi-Label MLP classifier, Label powerset MLP classifier, etc.), and the proposed simultaneous fault diagnosis method has achieved excellent results.     

A data-driven digital-twin model and control of high temperature proton exchange membrane electrolyzer cells

The high temperature proton exchange membrane electrolyzer cells (HT-PEMEC) are promising for hydrogen generation from fluctuating and intermittent renewable energy. In this study, a data-driven method is developed to study the dynamic behavior of HT-PEMEC. This method combines multiphysics simulation and nonlinear system identification, avoiding expensive experimental costs and time-consuming full multiphysics calculations. Dynamic models for predicting the power consumption, hydrogen production and temperature are identified, and the verified fit is 96.31%, 97.87%, 87.73%, respectively, which demonstrated the accuracy of the identification model. Subsequently, the identification model was used to predict the dynamic behavior of HT-PEMEC and design control strategies. Fuzzy control strategy and neural network predictive control strategy are implemented to alleviate overshoot and suppress fluctuations so as to improve the durability of the electrolyzer. Moreover, compared with the fuzzy control strategy, the neural network predictive control strategy reduces the power overshoot by approximately 92%. This data-drive digital-twin model can not only guide dynamic experimental research, but also can be extended to study the dynamic behavior of various fuel cells and electrolyzer cells.     

A unified data-driven design framework of optimality-based generalized iterative learning control

This paper proposes a unified design framework for data-driven optimality-based generalized iterative learning control (DDOGILC), including data-driven optimal ILC (DDOILC), data-driven optimal point-to-point ILC (DDOPTPILC), and data-driven optimal terminal ILC (DDTILC). First, a dynamical linearization in the iteration domain is developed. Then three specific DDOGILC approaches are proposed. Both design and analysis of the controller only require the measured I/O data without relying on any explicit model information. The optimal learning gain can be updated iteratively, which makes the proposed DDOGILC more adaptable to the changes in the plant. Furthermore, the proposed DDOPTPILC and DDOTILC only depend on the tracking error at specific points, and thus they can deal with the scenario when the system outputs are measured only at some time instants. Moreover, the proposed DDOPTPILC and DDOTILC approaches do not need to track the unnecessary output reference points so that the convergence performance is improved.     

Frame-wise detection of relocated I-frames in double compressed H.264 videos based on convolutional neural network

Relocated I-frames are a key type of abnormal inter-coded frame in double compressed videos with shifted GOP structures. In this work, a frame-wise detection method of relocated I-frame is proposed based on convolutional neural network (CNN). The proposed detection framework contains a novel network architecture, which initializes with a preprocessing layer and is followed by a well-designed CNN. In the preprocessing layer, the high-frequency component extraction operation is applied to eliminate the influence of diverse video contents. To mitigate overfitting, several advanced structures, such as 1 × 1 convolutional filter and the global average-pooling layer, are carefully introduced in the design of the CNN architecture. Public available YUV sequences are collected to construct a dataset of double compressed videos with different coding parameters. According to the experiments, the proposed framework can achieve a more promising performance of relocated I-frame detection than a well-known CNN structure (AlexNet) and the method based on average prediction residual.     

Multi-objective decision analysis for data-driven based estimation of battery states: A case study of remaining useful life estimation

Data-driven methods, which can explore the relationship among battery external parameters and battery states automatically without establishing complicated battery model, have been intensively applied to estimate state of charge (SOC), state of health (SOH) and remaining useful life (RUL) etc. Nevertheless, relatively few researches have been done on the selection of data-driven model parameters and the determination of model with the optimal comprehensive performance. To address these questions, this paper presents a multi-objective decision method for data-driven based estimation of battery states. This method adopts the combination of the analytic hierarchy process and the entropy weight method together with integrating subjective and objective weights. The mean absolute error and root squared mean error of training-set, validation-set and test-set are used as accuracy indexes, and modeling time is seen as computation burden index. These seven indexes are applied as objective criteria for the multi-objective evaluation method, successfully evaluating the comprehensive performance of estimation model. Moreover, with three cases for RUL estimation, the specific application process of selecting the model with the optimal comprehensive performance by the proposed method is presented in detail.     

Predictive modelling for contact angle of liquid metals and oxide ceramics by comparing Gaussian process regression with other machine learning methods

Wettability has a major effect on the performance of the corrosion of ceramic refractory under normal operating conditions. Contact angle measurement is available to characterize the wettability of liquid metals and oxide ceramics. Therefore, it is necessary to develop a contact angle prediction model with generalizability. This work emphasizes on developing a model for predicting the contact angle of a liquid metal with a solid oxide and analyzes the influence of factors affecting the contact angle when contact angle is predicted. In this paper, six contact angle prediction models are developed based on machine learning methods and contact angle data from the previous literature. The comparison between six contact angle prediction models evidences that the gaussian process regression (GPR) model has the best prediction accuracy and reaching 96%. Furthermore, the comparative results indicate that when surface energy of metal, surface energy of oxide, formation free energy of oxide, and bandgap energy of oxide are ignored respectively, the prediction accuracy of the model decreases by 4%, 3%, 1% and 1% respectively.     

Performance degradation analysis and fault prognostics of solid oxide fuel cells using the data-driven method

Widespread commercial implementation of the solid oxide fuel cell (SOFC) systems is hindered by their high cost, insufficient durability and poor reliability. Fault prognostics of these systems are extremely difficult due to the complicated interactions of their constituting components. This paper proposes a data-driven method of fault prognostics of SOFC systems based on the voltage signal. The voltage signal is first decomposed into a trend component and several fluctuation components with the empirical mode decomposition (EMD). The minimal-redundancy-maximal-relevance criterion (mRMR) is then applied to determine the most relevant fluctuation component. A Gauss mixture model with different humps is obtained from the distribution of the trend component and the fluctuation component in different periods. Finally, the similarity of different humps is calculated and adopted as the health indicator (HI). A fault warning is successfully issued approximately 70 h in advance. Meanwhile, the validity of the proposed method is confirmed by the measured microstructure and element distribution at different degradation stages using the scanning electron microscopy (SEM) and energy dispersive X-ray detector (EDX). These results demonstrate that the proposed method can predict the fault occurrence during the SOFC operation.     

数据驱动下交互网络群智感知任务分配仿真

采用当前方法在交互网络中分配群智感知任务时,分配任务所用的时间较长,存在用户完成任务的概率较低,分配效率低的问题,提出一种数据驱动下交互网络群智感知任务分配方法。分析群智感知环境,将能耗最小化和时间最小化作为感知任务分配的约束条件,通过信誉模型计算用户在交互网络中的信誉值,将感知任务优先分配给信誉值高的用户,在交互网络中设置因子,控制用户完成任务花费的代价,通过用户处理感知任务对应的效用值对用户信誉值进行更新,在交互网络中竞争下一个感知任务,直到分配完交互网络中存在的感知任务为止,实现交互网络群智感知任务的分配。仿真结果表明,所提方法的分配效率高、任务完成率高。