Papers by Keyword: Data-Driven

Abstract: Inverse identification of material parameters from experimental data is a long-standing challenge, especially when calibrating complex constitutive models characterized by a large number of parameters. Heterogeneous mechanical tests combined with full-field measurements provide a large amount of information for material parameter identification but lead to high computational costs when used within a Finite Element Model Updating (FEMU) framework. This work presents an exploratory study on the use of surrogate-assisted Bayesian optimization to assess its potential for reducing the number of simulations required for FEMU-based calibration using data from a single notched tensile test. FEMU cost function is applied based on the discrepancy between experimental and numerical strain fields. A Gaussian Process surrogate model is iteratively constructed, and new sets of material parameters are selected using an Expected Improvement criterion. The results are discussed in terms of convergence behaviour and optimization efficiency, providing insight into the suitability of Bayesian optimization for solving inverse identification problems.

Abstract: Machine learning models are effective tools for predicting and reducing noise levels in industrial gear systems. In this study, we compare different machine learning methods to investigate the effects of different gear modification parameters on noise levels. Four different predictive models was used. Random Forest Regressor, XGBoost, Gradient Boosting Machines and neural network. The study concluded that Random Forest and Gradient Boosting Machines models were the most effective. Both models achieved low mean squared error values 6.10 and 6.67. Further tests with synthetic data confirmed the stability of these models. Current sustainability trends show that the integration of machine learning into industrial applications fits well with manufacturers' objectives. However, it is currently challenging to determine which machine learning methods are most effective in optimizing noise reduction. This paper seeks to address this gap by comparing the accuracy and reliability of these models. Based on the results, the use of machine learning models is recommended to reduce noise levels in geared systems.

Abstract: This paper discusses the state-of-art of adaptive control approaches for nonlinear systems to date and presents a new classification framework, in which the existing adaptive control approaches can be broadly classified into two categories: model-driven methods and data-driven methods. The principle, main research progress, and inherent problems of these methods are reviewed. Finally, some practical considerations and future directions are also briefly explored and discussed.

Abstract: Aiming at the features of nonlinearity, time delay, high dimension, big noise and others in the electric furnace smelting process, and the smelting process parameters some control process needs cannot be obtained by sensors, so the online parameter prediction method is of great significance for dynamic optimization control of system. In this paper, by using chaos theory and phase space reconstruction method, based on the idea of data-driven modeling, the temperature prediction model of crucible in the smelting process of ferroalloy electric furnace is established based on multivariate time series. Experiments show that the established data-driven prediction model can well predict the temperature change in the electric furnace smelting process, the root mean square errors of prediction are respectively less than 0.0715,and this has good guidance for practical production.

Abstract: With the rapid development of information science and technology, data-driven approaches are already being the research tide in many fields. BP neural network (BPNN), support vector machine (SVM) and least squares support vector machine (LS-SVM) are introduced and adopted to simulate fluctuating time-series wind speeds in this paper. The regression-prediction models developed by implementing machine interpolation learning are established respectively. And the original speeds used as learning and forecast samples for the simulation of the data-driven approaches are obtained through AR numerical modeling. Based on the comparison of evaluation index, the results show that the simulated fluctuating wind speeds through SVM and LS-SVM are more accurate than the simulated speeds through BPNN, but the simulation time of LS-SVM and BPNN are shorter than the SVM.

Abstract: To resolve the problem of a large amount of automated test scripts and test data files, through the test tool QTP, data-driven and keyword-driven testing mechanism, a test automation framework based on three layer data-driven mechanism is designed, including the design of the TestSet managing test case files, the design of the TestCase storing test cases and the design of the TestData storing test data.Through controlling the test scale and applying the test data pool, reconfigurable and optimization of test scripts are designed. The methods above can decouple the test design and the script development, make test cases and data show a more humane design, make test scripts and test data on the business level optimized and reusable, and make the number of script files and the test data files reache a minimum, which reduces the occupied space.

Abstract: Complexity control theory based on the model depict the complexity of the controlled object more accurately through research model and data-driven control theory of Internet of things is more urgent to be explored through the data to find out the system of internal control mechanism. Look from the appearance, the complexity of data including a lot of sample, high dimension, strong time-varying factors. The representation of implicit information is huge amounts of data redundancy, high-dimensional data clustering features, the time-frequency characteristics of time-varying data, etc. At present, the data-driven control theory research is just beginning [1-3], the scientific problem is the lack of unified definition and theoretical framework. This paper will analyze the sewage treatment system of some sewage treatment plant, making the internal scientific questions are more specifically.

Abstract: As one of the most widely used energy storage systems, lithium-ion batteries are attracting more and more attention, and the estimation of lithium-ion batteries remaining useful life (RUL) becoming a critical problem. Generally, RUL can be predicted in two ways: physics of failure (PoF) method and data driven method. Due to the internal electro-chemical reactions are either inaccessible to sensors or hard to measure; the data-driven method is adopted because it does not require specific knowledge of material properties. In this paper, three data-driven algorithms, i.e., Support Vector Machine (SVM), Autoregressive Moving Average (ARMA), and Particle Filtering (PF) are presented for RUL prediction. The lithium-ion battery aging experiment data set has been trained to implement simulation. Based on the RUL prediction result, we can conclude that: (1) ARMA model achieved better result than SVM, however, the result shows a linear trend, which fail to properly reflect the degradation trend of the battery; (2) SVM often suffers from over fitting problem and is more suitable for single-step prediction; and (3) PF approach achieved a better prediction and reflected the trends of degradation of the battery owing to its combined with specific model.

Abstract: P2P technology has become one of the most effective approaches to internet video broadcast for its simple deployment, good scalability and high cost-effectiveness ratio. This article proposed technology based P2P-based online video teaching system,P2P technology will be compared with traditional technologies, described the demand characteristics of online video, online video systems network architecture and design method, discussed the media data distribution algorithm from the node management, buffer management, data-driven,Video source coding.Finally get research conclusions.

Abstract: How to monitor and predict icing load of power transmission lines are important problems for the reliability of power grid. A model based on data-driven is presented here to predict the icing load of transmission line, which is available to forecast the icing disaster of it. The fitfulness, which influencing the prediction results of icing load, is analyzed and discussed in this paper. According to the results of simulation, this model has a good accuracy of prediction if the training data and prediction data are in the same icing process. If the icing process is not same but contiguous, it also can predict the degree of icing load qualitatively.