电熔镁熔炼过程中的异常工况(如半熔化工况)直接影响产品质量、威胁人员和生产安全,有必要及时诊断.但与异常直接相关的超高温熔池温度(>2850℃)难以利用温度传感器检测,目前现场主要依靠工人在定期巡检时人眼观察炉壁来诊断,工作强...电熔镁熔炼过程中的异常工况(如半熔化工况)直接影响产品质量、威胁人员和生产安全,有必要及时诊断.但与异常直接相关的超高温熔池温度(>2850℃)难以利用温度传感器检测,目前现场主要依靠工人在定期巡检时人眼观察炉壁来诊断,工作强度大、安全度低、诊断不及时.针对上述问题,本文提出一种炉体动态图像驱动的电熔镁炉异常工况实时诊断方法.结合电熔镁炉熔炼各区域温度分布的空间特征、正常工况下熔炼温度变化和水雾扰动引入的图像时序特征、以及异常工况下温度异常区域持续发亮扩大的特征,在对炉体动态图像进行空间多级划分的基础上,提出了一种多级动态主元分析(Multi-level dynamic principal component analysis,MLDPCA)动态图像分块建模方法.在此基础上,提出基于MLDPCA的逐级诊断方法与基于贡献图的异常定位方法.最后,采用某电熔镁生产现场的实际图像进行方法验证,结果表明了所提方法的有效性.展开更多
针对基于动态主元分析的故障检测方法存在的主元个数较多以及计算效率低等问题,本文提出基于混合动态主元分析(Hybrid Dynamic Principal Component Analysis,HDP-CA)的复杂过程故障检测方法。该方法采用分步策略消除数据之间的自相关...针对基于动态主元分析的故障检测方法存在的主元个数较多以及计算效率低等问题,本文提出基于混合动态主元分析(Hybrid Dynamic Principal Component Analysis,HDP-CA)的复杂过程故障检测方法。该方法采用分步策略消除数据之间的自相关和互相关性,提高了故障检测的精度和效率。对TE过程典型故障和热连轧过程中断带故障检测结果表明:HDPCA方法提取的主元个数少于DPCA方法提取的主元个数。并且,基于HDPCA的T2和SPE统计量的检测性能和检测精度都由于基于DPCA的统计量。因此,本文提出的方法可以准确有效地检测出故障。展开更多
为了提高故障检测准确率,提出了基于动态受控主元分析(dynamic controlled principal component analysis,DCPCA)模型的故障检测方法。首先,利用DCPCA提取动态受控主元(dynamic controlled principal component,DCPC),所得DCPC包含过程...为了提高故障检测准确率,提出了基于动态受控主元分析(dynamic controlled principal component analysis,DCPCA)模型的故障检测方法。首先,利用DCPCA提取动态受控主元(dynamic controlled principal component,DCPC),所得DCPC包含过程的自回归特性和与控制输入之间的动态因果关系,使得构建的DCPCA模型更精确。然后,针对传统方法只对过程变量进行静态空间结构的故障检测,忽略了动态特性的问题,基于DCPCA模型适时应用检测综合指标,对系统进行静态重构误差和动态模型误差的双重检测,使得检测结果更全面。最后,基于田纳西-伊斯曼(Tennessee-Eastman,TE)过程的仿真结果验证了所提方法的可行性和有效性。展开更多
Dynamic principal component analysis(DPCA) is an extension of conventional principal component analysis(PCA) for dealing with multivariate dynamic data serially correlated in time.Based on the fact that the measured v...Dynamic principal component analysis(DPCA) is an extension of conventional principal component analysis(PCA) for dealing with multivariate dynamic data serially correlated in time.Based on the fact that the measured variables in relation to chunk monitoring of the industrial fluidized-bed reactor are highly cross-correlated and auto-correlated, this paper presents a practical strategy for chunk monitoring by adopting DPCA in order to overcome the shortcomings of the conventional method.After introducing the basic principle of DPCA, both how to determine the time lagged length of data matrix and how to calculate the nonparametric control limits when the dynamic data are not subject to the assumption of independently identically distribution(IID) were discussed.An appropriate DPCA model based on the real data from a industrial fluidized-bed reactor was built, with parallel analysis and empirical reference distribution(ERD)method to select time lagged length and control limits, respectively.During data pretreatment, data smoothing was used to reduce noise and the serial correlations to some degree.The simulation test results showed the effectiveness of the DPCA based method.展开更多
文摘电熔镁熔炼过程中的异常工况(如半熔化工况)直接影响产品质量、威胁人员和生产安全,有必要及时诊断.但与异常直接相关的超高温熔池温度(>2850℃)难以利用温度传感器检测,目前现场主要依靠工人在定期巡检时人眼观察炉壁来诊断,工作强度大、安全度低、诊断不及时.针对上述问题,本文提出一种炉体动态图像驱动的电熔镁炉异常工况实时诊断方法.结合电熔镁炉熔炼各区域温度分布的空间特征、正常工况下熔炼温度变化和水雾扰动引入的图像时序特征、以及异常工况下温度异常区域持续发亮扩大的特征,在对炉体动态图像进行空间多级划分的基础上,提出了一种多级动态主元分析(Multi-level dynamic principal component analysis,MLDPCA)动态图像分块建模方法.在此基础上,提出基于MLDPCA的逐级诊断方法与基于贡献图的异常定位方法.最后,采用某电熔镁生产现场的实际图像进行方法验证,结果表明了所提方法的有效性.
文摘针对基于动态主元分析的故障检测方法存在的主元个数较多以及计算效率低等问题,本文提出基于混合动态主元分析(Hybrid Dynamic Principal Component Analysis,HDP-CA)的复杂过程故障检测方法。该方法采用分步策略消除数据之间的自相关和互相关性,提高了故障检测的精度和效率。对TE过程典型故障和热连轧过程中断带故障检测结果表明:HDPCA方法提取的主元个数少于DPCA方法提取的主元个数。并且,基于HDPCA的T2和SPE统计量的检测性能和检测精度都由于基于DPCA的统计量。因此,本文提出的方法可以准确有效地检测出故障。
文摘为了提高故障检测准确率,提出了基于动态受控主元分析(dynamic controlled principal component analysis,DCPCA)模型的故障检测方法。首先,利用DCPCA提取动态受控主元(dynamic controlled principal component,DCPC),所得DCPC包含过程的自回归特性和与控制输入之间的动态因果关系,使得构建的DCPCA模型更精确。然后,针对传统方法只对过程变量进行静态空间结构的故障检测,忽略了动态特性的问题,基于DCPCA模型适时应用检测综合指标,对系统进行静态重构误差和动态模型误差的双重检测,使得检测结果更全面。最后,基于田纳西-伊斯曼(Tennessee-Eastman,TE)过程的仿真结果验证了所提方法的可行性和有效性。
文摘Dynamic principal component analysis(DPCA) is an extension of conventional principal component analysis(PCA) for dealing with multivariate dynamic data serially correlated in time.Based on the fact that the measured variables in relation to chunk monitoring of the industrial fluidized-bed reactor are highly cross-correlated and auto-correlated, this paper presents a practical strategy for chunk monitoring by adopting DPCA in order to overcome the shortcomings of the conventional method.After introducing the basic principle of DPCA, both how to determine the time lagged length of data matrix and how to calculate the nonparametric control limits when the dynamic data are not subject to the assumption of independently identically distribution(IID) were discussed.An appropriate DPCA model based on the real data from a industrial fluidized-bed reactor was built, with parallel analysis and empirical reference distribution(ERD)method to select time lagged length and control limits, respectively.During data pretreatment, data smoothing was used to reduce noise and the serial correlations to some degree.The simulation test results showed the effectiveness of the DPCA based method.
