A Probabilistic Copula-Based Fault Detection Method With TrAdaBoost Strategy for Industrial IoT

With the development of information and technology, Industrial Internet of Things (IIoT) are widely spread. In the face of heterogeneous network access technologies, IIoT data presents characteristics, such as massiveness, heterogeneity, and dynamics. In this work, an inductive transfer modeling fault detection method based on the vine copula-based dependence description (VCDD), named TrAdaBoost VCDD (TAVCDD), is proposed for fusing the target and source data. Especially, there is too little data in target domain operation condition to establish a good monitoring model. In the modeling process, a generalized local probability (GLP) index is used to set the selection probabilities of samples in the target and source domains. In addition, two adaptive indexes are used to determine the number of training samples in one iteration and the number of iterations. To avoid unbalanced samples between the target and source data sets, an adaptive sample selection strategy called active VCDD (AVCDD) is introduced to explore the underlying distribution of the target sample space and to select the most “informative” samples from the source data set. The proposed TAVCDD method is compared with six state-of-the-art methods via a numerical example, the Tennessee Eastman process, and the ethylene cracking furnace process.