Search

In this paper, an analytic condition is given for determining delayed positive feedback controller for stabilizing an oscillatory system. The $\tau $ -decomposition and $D$ -decomposition methods are employed in deriving this condition. The obtained results are then used to stabilize second order delay systems by a proportional controller. Under-damped and over-damped systems are treated separately, where the Smith-predictor structure is used in the over-damped case to obtain the stability conditions. Illustrative examples are given to show the effectiveness of the proposed approach.

There are quite tight timing requirements in deterministic low latency network. The IEEE 802.1 Time-Sensitive Networking (TSN) task group has proposed several traffic shapers to satisfy real-time communications requirements. Traditionally, the performance of TSN is analyzed by simulations, whereas these methods cannot cover all corner cases. This paper firstly presented formal models of the TSN’s time-aware and peristaltic shapers using UPPAAL, tactically solving the problem mentioned above. Afterward, we verified some properties of the shapers models, of which the results could evaluate whether these shapers are able to satisfy strict timing requirements or not. Based on the models, we can discuss about the performance of time-critical traffic combining the preemption mechanism. More-over, we can also analyze resource utilization and transmission latency. Under the time properties analysis and verification of TSN traffic shapers, we can provide engineers with an accessible reference that may assist them in developing the TSN.

In the iron and steel industry, vacuum refining is the previous process of continuous casting, and the VD furnace is one of the key equipment for vacuum refining. The precise control of the end temperature is an important guarantee for the quality of subsequent cast steel products. In this paper, based on the XGBoost algorithm and the LightGBM algorithm, the VD furnace molten steel terminal temperature prediction model will be established respectively, and their performance will be compared. First, perform data preprocessing on the original industrial data, fill in missing values, use Tukey’s test method to detect and delete outliers, and use box-cox data transformation to normalize the data. Then, the random forest feature selection algorithm was used to select 8 variables as the input of the model. Finally, the temperature prediction models based on the XGBoost and LightGBM algorithms were established. The experimental results show that the models based on the XGBoost algorithm and the LightGBM algorithm both have good prediction effects, and the prediction accuracy within ±10℃ reached 89.86%, and the overall performance of the LightGBM model is better than XGBoost model.

Transfer entropy is a data-driven model-free method, which can obtain the causal relationship between variables, and has been applied in the fields of complex industrial process modeling, monitoring and fault diagnosis. Transfer entropy can detect the causal relationship between variables without assuming any basic model. Because the signal will appear non-stationary time series in which both mean and variance change, so the method of partial symbol transfer entropy is proposed to detect the causality of non-stationary time series, and then the causal network diagram is obtained. The effectiveness of the method is verified by simulation and actual industrial case (Tennessee-Eastman process).

Artificial intelligence and its deep learning have broad application areas, including medical diagnosis, industrial area, data analysis, identification, system control, self-driving, and games. Deep learning methods show the advantage of using neural networks to solve complex tasks. The network training speed is highly dependent on the hardware environments and its training methodology. Training a large dataset on a Convolutional Neural Network (CNN) can last for weeks with a limited number of GPUs. Although cloud training is a good method to train large networks within a short time, it is usually costly. What is more, since a good network needs a lot of parameter tunings to get an optimal result, a network needs to train multiple times with different parameter settings. The training speed and the cost for large networks have become a constraint to verify algorithm results when computing resources are limited. Thus, fast training a large network with a limited number of computing resources is needed and challenging, especially for large datasets. In this paper, we propose new fast training methods to accelerate the CNNs training time from the data sampling level. We introduce a flat reduced random sampling strategy and a bottleneck reduced random sampling strategy for deep learning model training. Furthermore, the theoretical analysis for the proposed fast training methods are derived by four theorems and two corollaries to show the properties and benefits of the proposed new methods. The experiments of these fast training methods are run on three image classification datasets: CIFAR-10, CIFAR-100 and ImageNet. The experimental results show that our proposed new training methods can achieve significant training time percentage reduction with a very small accuracy cost correspondingly, i.e., to well provide the fast training method for deep learning CNN models as our theoretical analysis shows and predicts.

The purpose of this research is to improve the analysis of grade evaluation of soil corrosion rate for pipeline. With the characteristics of the existing mathematical models and the factors of pipeline soil corrosion, a combined mathematical model of the rough set, the decision tree and support vector machine (RS-DT-SVM) is proposed to evaluate the grade of soil corrosion rate for pipeline. This method utilizes the rough set (RS) and decision tree (DT) to manage by standardization analysis of the data of pipeline soil corrosion in the early stage. According to the characteristics of solving the problems of over learning, local minima, nonlinearity and dimension disaster, support vector machine (SVM) is used to build the intelligent evaluation of rate grade of soil corrosion. With algorithm characteristics of machine learning, it can be built without determining the specific function expression. Taking the actual and objective corrosion data of loam as an example, the model has a relative superiority in the evaluation grade analysis of multi-factor control system. It also provides a method guidance of evaluation grade of soil corrosion rate for pipeline.

Ladle Furnace (LF) is a kind of refining equipment outside the furnace, and the end temperature of its molten steel is an important factor affecting the quality of later finished steel. Aiming at the characteristic that it is difficult to continuously measure the temperature of molten steel in LF, based on the field data of Baosteel's LF furnace, a two-layer structure Stacking integrated model framework is designed, the first layer contains four base learners: ridge regression algorithm, support vector regression, random forest, and XGBoost, the second layer uses the XGBoost model. Based on five modeling methods of ridge regression algorithm, support vector regression, random forest, XGBoost and Stacking designed, after analyzing and selecting the main factors affecting the LF temperature, the prediction models of the LF furnace molten steel endpoint temperature were established. The experimental results show that compared with a single model, the prediction effect and stability of the Stacking model combining ridge regression algorithm, support vector regression, random forest, and XGBoost have been significantly improved. The prediction model can not only reflect the effect of different factors on LF, but also, it can predict the end temperature of LF molten steel relatively accurately. The prediction accuracy of the end temperature of molten steel at ± 10°C can reach more than 80%.

In this paper, the issue of stabilization for uncertain rectangular descriptor fractional order systems (FOS) with 0

This paper deals with the functional observer (FO)-based fault detection and isolation problem for time-delayed linear systems. First, a new method to design the low-order FO estimating the partial states for linear systems with time-varying delay is presented. Conditions for the existence of the proposed observer and the stability of resulting error dynamics are derived. Based on the proposed observer, a novel residual generator is constructed to detect actuator faults. Furthermore, a bank of residual generators for fault isolation is also constructed for time-delayed systems. The simulation results of the numerical example demonstrate that the proposed observer-based fault diagnosis scheme works effectively.

The tree-based machine learning functions on the divide-and-conquer principle and is known to perform well in certain applications. In this paper, we first give a new data partitioning rule using the mean of the data columns to grow the tree till the child nodes are small in size. Then, the local regression is applied to leave nodes to enhance the resolution of the node outputs. Randomization is introduced at tree growth and forest creation. The local prediction accuracies on the leaves are used to select a subset of the test data for actual predictions. The case study on the diagnosis of autistic spectrum disorder shows that the proposed method achieves the prediction accuracy of the ensemble at above 96% with reduced variance, which is much better than those reported in the literature.

This paper firstly focuses on yield estimation problem on post-layout-simulation of high dimensional SRAM arrays. Post-layout-simulation is much more credible than pre-simulation. However, it introduces strong relationship among SRAM columns. The Multi-Fidelity Gaussian Process model between the small and the large SRAM arrays near Optimal Shift Vector (OSV) is built. An iterative strategy is proposed and Multi-Modal method is applied to obtain more prior knowledge of the small SRAM arrays and further accelerate convergence. Experimental results show that the proposed method can gain 5-7x speedup with less relative errors than the state-of-the-art method for 384D cases.

In order to avoid the spread of new coronavirus, it is necessary to keep a distance, wear masks and wash hands, not close contact. When the academic research or market survey needs to fill in the questionnaire, it is a safe and necessary choice to use the network electronic form. In order to prevent the spread of virus, it is necessary to keep a safe distance between people. The demand for electronic forms is more diverse and vigorous. To prevent the spread of virus through interviews and questionnaires, we can change the way of filling in electronic forms to achieve the purpose of maintaining a safe distance. As a result, the demand for electronic forms has increased, and the electronic forms can be more convenient. At the same time, the anti-counterfeiting function of electronic signature is used. The undeniable characteristics of personal signature can be customized and diversified. At present, the system development and technology of web forms are various and mature. In order to increase the employability of college students, the implementation course of online spreadsheet is added to the professional courses of University, which has gradually become a popular option of software engineering for students' graduation design. From the system analysis and system design, JavaScript, HTML5, CSS3 and other technologies are selected. Through the programming framework on the server, through the browser, it is convenient to use the web spreadsheet. The system functions are successfully implemented, and the graduation design of university professional courses is completed, which is a reference for successful cases of course practice. This model is a successful example of system development. Combined with the theory of university courses and the practice of project design, the system development, design and system testing are completed successfully within the effective time limit. From the system analysis and system design, JavaScript, HTML5, CSS3 and other technologies are selected. Through the programming framework on the server, through the browser, it is convenient to use the web spreadsheet. The system functions are successfully implemented, and the graduation design of university professional courses is completed, which is a reference for successful cases of course practice. This model is a successful example of system development. Combined with the theory of university courses and the practice of project design, the system development, design and system testing are completed successfully within the effective time limit.

In the binocular vision system, the parallax image is different when the real face of the three-dimensional structure is viewed from different viewing angles, while the face in photos and videos is flat or the planar structure is the same when viewed from different angles, so it is proposed A new method that uses a manual face binocular disparity map made by a synchronized binocular visible light camera combined with a convolutional neural network (convolutional neural network, CNN) to achieve two classifications. First, the Harr cascade classifier is used to extract faces in binocular images at the same time, and the extracted partial faces are made into a data set; then, the disparity map form that meets the experimental hypothesis is proposed and analyzed; finally, the ResNet18 network is used to compare different methods Binocular live body detection solution for human face. The accuracy rate of 99.12% is achieved on the self-made data set in the laboratory, and it has a good resistance to facial attacks on photos and videos. The results show that the method has a small amount of data, low hardware cost and good robustness to human faces due to changes in light and angle.

The endpoint temperature of the molten steel in the VD (Vacuum Degassing) furnace is an important parameter determining the quality of the finished steel products. Based on the in-depth analysis of the vacuum refining process of the VD furnace, combined with the field data, the effective preprocessing of the data was completed. Then the NNG (Non-Negative Garrote) variable selection algorithm is used to determine the input variables. An integrated ELM (Extreme Learning Machine) molten steel endpoint temperature modeling method based on AdaBoost.RT is proposed. Experimental simulation results show that the AdaBoost.RT-ELM prediction model has significantly improved prediction accuracy than a single ELM.

The Granger causality is a very effective method for the root cause diagnosis. However, the synergy and redundancy between variables will affect the accuracy of the test result. This article introduces an unnormalized Granger causality method to detected the synergy and redundancy between variables and defined a redundancy index. Based on the index, a variable stratification method is proposed. The layered result obtained by this method can directly reflect the order in which the variables are affected by the fault. In addition, further Granger causality tests based on the layered results can make fault propagation paths more accurate. Two typical faults of the TE process were used to verify the effectiveness of the method.

Plant wide oscillations are very common in industrial processes. When a control unit oscillates in the process, the oscillation will propagate through the connectivity between units, which will cause problems such as poor product quality and higher energy consumption. Therefore, it is very important to diagnose the root cause of plant wide oscillations [1]. In this dissertation, the kernel Granger causality is proposed for root cause diagnosis, which effectively solves the problem that the linear Granger causality cannot deal with nonlinear data. A small nonlinear numerical example shows the validity of kernel Granger causality. In the root cause diagnosis of industrial cases, the method also successfully detected the correct root point (LC2).

Granger causality analysis is one of the most widely used methods in root cause diagnosis. This method can get effective results in many cases, but there are still some problems and underutilization of data is one of them. Granger causality analysis only used the fault relate data segment. This paper proposes a novel root cause diagnosis framework based on Granger causality analysis, and attempts to combine the normal and fault data to make the result more accurate. The main ideal is to test the change of causality intensity before and after the fault to optimize the result of the fault propagation paths. Tennessee Eastman(TE) process data and TE data was used to verify the effectiveness of the method.

Transfer entropy (TE) is a data-driven, model-free method that can obtain causal relationships between variables and is used in the modeling, monitoring, and fault diagnosis of complex industrial processes. Transfer entropy can detect the causal relationship between variables without the assumption of any basic model, but it is complicated and takes a long time to calculate. In order to solve this limitation, this paper proposes a method of causal network analysis and root cause detection based on parameter variable sequence transfer entropy, which is more robust than traditional transfer entropy, faster in calculation speed, and strong in anti-interference ability, thereby improving the causal path accuracy. The causal network diagram can be obtained by calculating the transfer entropy between sequences, and the root cause of the fault can be found. The effectiveness and accuracy of the method are verified by simulations and actual industrial cases (Tennessee-Eastman process).

Plant-wide oscillations are very common in industrial processes. When a control unit oscillates during the process, the oscillations will propagate through the connectivity between the units, which will cause poor product quality and higher energy consumption. It is important to diagnose the root cause of plant-wide oscillations. Generally, methods for estimating Granger causality use linear models such as autoregressive models. This paper proposes using Granger causality analysis based on the neural network for root cause diagnosis, which effectively solves the problem that Granger causality analysis based on linear models cannot handle non-linear data. The Granger causality detection model based on neural network is successfully applied to the plant-wide oscillation root location of industrial process, and the correct root cause is detected, which proves the feasibility and effectiveness of the method.