Papers by Author: Jin Wu Xu

Abstract: In order to solve the time delay problem between the process parameters and the quality indicators in the modeling processes, a method of time delay estimation on COREX parameters is proposed based on Dynamic Time Warping (DTW) algorithm. The method solves the problem existing in the conventional methods which demand the number of calculating sample to be same. Taking the real field data from Baosteel COREX-3000 as the research object, the DTW distances between the process parameters and the quality indicators are calculated, and then the delay time is estimated. The real field data are used for verification, the results show that the proposed method can estimate the time daley effectively, and the prediction accuracy of model which used time delay estimation becomes higher. It provides an effective measure for model preprocessing.

Abstract: Hot strip tail flick is an abnormal production phenomenon, which brings many damages. To recognize the tail flick signals from all throwing steel strip signals, a feature extraction method based on morphological pattern spectrum is proposed in this paper. The area between signal curves after multiscale opening operation and the horizontal axis is computed as the pattern spectrum value and it reflects the geometric information differences. Then, support vector machine is used as the classifier. Experimental results show that the total correct rate based on pattern spectrum feature reached 96.5%. Compared with wavelet packet energy feature, the total correct rate is 92.1%. So, the feasibility and availability of this new feature extraction method are verified.

Abstract: Aimed at the problem of low resolution and cross term interference of the traditional time-frequency analysis methods, a new time-frequency filtering method based on generalized S transform is proposed. The method is extended under the premise of the linearity, lossless invertibility, high time-frequency resolution of S transform. On the basis, a coefficient which is direct to the signal energy distribution is introduced. In this way, the resolution of the S transform can be adjust adaptively. Eventually, this method is applied to the time-frequency filtering. The results of simulation and faulty bearing show that the proposed methodology can achieve good effect of noise reduction, and be more suitable for the non-stationary characteristics of vibration signals.

Abstract: In COREX processes, the cold gas is produced in melter gasifier, after being cooled and dust controlled, blown into the blast furnace and used in the reduction reaction. The cold gas content plays a key role in the reaction of lump ore and pellets reduction. A prediction model of COREX cold gas content of carbon dioxide is proposed based on modified orthogonal signal correction partial least squares algorithm (MOSC-PLS). Firstly, the input and output variables of the model are selected according to the COREX processes principle. Secondly, MOSC algorithm is used to preprocess the data, in order to remove the irrelevant information between the input and output variables of the model. Finally, prediction model is built based on PLS. The real field data of cold gas content of carbon dioxide from Baosteel COREX are used for verification. The results show that MOSC-PLS has an advantage over the orthogonal signal correction partial least squares (OSC-PLS) in prediction accuracy. Thus the necessary decision supports and analysis tools for the cold gas content control are provided.

Abstract: Technology and metallographic of three kinds of representative gears were analyzed in this paper according to characteristics of carburizing and quenching, and technology problems met with during the carburization process of large low-alloy gears in Citic Heavy Industries Co., Ltd were also taken into account. Three conclusions are demonstrated through experimental results. Firstly, although carburization technology is normal, grinding cracks can be caused by hard particles included in matrix. Secondly, if carbon concentration of carburization layer is too high or too steep in carburizing technology, carbide will aggregate in reticular or structural stress will not be average during the subsequent quenching technology. And therefore grinding cracks are caused. Moreover, retained austenite of surface layer and subsurface layer will be over standard and thus grinding cracks will be caused if quenching temperature is too high during quenching technology after carburizing.

Abstract: The application of AE measurement for condition monitoring of the slow-speed and heavy-load equipment is gaining ground. However, different AE features are sensitive to fault in varying degrees when they are employed in the trend analysis. Therefore, in this paper the focus is on the selection of the appropriate AE features for the trend analysis. First the AE features are ranked by Laplacian Score method according to their importance, and then a new feature index is obtained by the fusion of the features with their ranking scores, which serve as weight coefficient in this condition. The degradation data of the bearings in the belt conveyor are used to prove that the proposed method is effective.

Abstract: In order to predict product quality and optimize production process, the product quality model needs to be built. According to the fact that the common methods always cost long training time and can not realize real-time update, an online product quality model based on the online support vector regression is here proposed. The real field data of zinc coating weights from strip hot-dip galvanizing are used for validation. The results show that the models based on the online support vector regression have a higher prediction precision and shorter training time than traditional support vector regression, which is convenient to complete the real-time update. The zinc coating weights forecasting model based on the online support vector regression for multi-group data has an average of the relative prediction error of 4.35%, thus for the model will be used as an analysis tools for the quality control.

Abstract: In order to monitor nonlinear production process effectively, multivariate statistical process control based on kernel principal component analysis is applied to process monitoring and diagnosis. Squared prediction error (SPE) statistic of the kernel principal component analysis (KPCA) model is used for process monitoring, and the fault causes of the production process could be tracked by the methods of data reconstruction and the optimal neighbor selection strategy. Simulation data and Tennessee Eastman process data are used for model validation, as a result the proposed method has better performance on abnormality detecting, compared with multivariate statistical process control based on linear principal component analysis. What is more, the causes of the faults are tracked effectively, thus the production process can be adjusted to prevent substandard products.