Search

Support vector data description (SVDD) procedure fits a spherically shaped boundary around the normal data by minimizing the volume of the description. However, the SVDD may not find an efficient boundary if the normal data consist of multiple classes. In addition to the multi-class normal data, some anomaly observations can be available. We propose a generalized SVDD procedure which finds multiple spheres around the multi-class data by incorporating the anomaly observations into the training procedure. Thus, descriptions for each class include as many as their corresponding class observations by keeping the other class and anomaly observations as far as possible. Moreover, we introduce a generalized Bayesian framework which utilizes the relationships among the classes by not only considering the prior information from normal classes but also the anomaly class. Experiments with various simulation studies and real-life applications demonstrate that the proposed approach can effectively identify the anomalies in multi-class data.

Simultaneous monitoring of multi-dimensional processes becomes much more challenging as the dimension increases, especially when there are only a few or moderate number of process variables that are responsible for the process change, and when the size of change is particularly small. In this paper, we develop an efficient statistical process monitoring methodology in high-dimensional processes based on the Bayesian approach. The key idea of this paper is to sequentially update a posterior distribution of the process parameter of interest through the Bayesian rule. In particular, a sparsity promoting prior distribution of the parameter is applied properly under sparsity, and is sequentially updated in online processing. A Bayesian hierarchical model with a data-driven way of determining the hyperparameters enables the monitoring scheme to be effective to the detection of process shifts and to be efficient to the computational complexity in the high-dimensional processes. Comparison with recently proposed methods for monitoring high-dimensional processes demonstrates the superiority of the proposed method in detecting small shifts. In addition, graphical presentations in tracking the process parameter provide the information about decisions regarding whether a process needs to be adjusted before it triggers alarm.

In most manufacturing processes, identifying the faulty process variables that may lead to process changes is crucial for quality engineers and practitioners. There are several parametric procedures for identifying faulty variables with the assumption that they follow multivariate normal distributions. However, in practice, the normality assumption restricts the applicability of such procedures in identifying the faulty variables. In addition, conventional procedures for fault identification are often computationally expensive, especially in high-dimensional processes. Therefore, this article proposes a data-driven Bayesian approach for fault identification that addresses the limitations posed by the normality assumption. The proposed approach is computationally efficient for high-dimensional data compared with existing approaches. Experimental results with various simulation studies and real-life data sets demonstrate the effectiveness of the proposed procedure. - 2018, - 2018 American Society for Quality. This article was made possible by the support of NPRP 5-364-2-142 and NPRP 7-1040-2-393 grants from Qatar National Research Fund (QNRF) and NRF-2015R1C1A1A01051487 from the National Research Foundation of Korea. Scopus

Identifying the faulty variables of the out-of-control signal in high-dimensional process is an important problem for quality control areas. Even though there have been several procedures for fault variable identifications, most of the existing approaches assume the multivariate normal distribution of observations and are sensitive to the correlations between variables. Therefore, in this paper, we propose a new fault variable identification method that does not assume any specific distribution of observations. The proposed procedure based on one class classification method identifies the changed variables by identifying unchanged variables at each step using the information obtained from the previous steps. This strategy can reduce computational times when a few variables are changed in a high-dimensional process. In addition, the proposed procedure is robust to the correlations between variables, resulting in stable performance regardless of the number of changed variables. The experiment results with diverse dataset demonstrate superiority of the proposed distribution-free procedure. Copyright © 2016 John Wiley & Sons, Ltd.

Information-based optimal experimental designs are effective offline quality improvement tools that provide insights into the information under complex engineering situations. In the literature, considerable attention has been focused on the regular design region-based experiments to generate design points for both qualitative and quantitative factors. However, there are several situations while some design points are infeasible due to the cost and resource-related restrictions. In such situations, an appropriate design should be selected to obtain feasible experimental design points. Therefore, this paper is three-fold. One, a D-optimal design is selected over other designs. Two, this paper is to develop models that interconnect experimental design as an information-gathering process in the early design phase with operations research in the optimization phase. To the best of our knowledge, there is not an optimization model for identifying optimum factor level settings by linking the D-optimal design concept to optimization. Thus, a 0–1 mixed-integer nonlinear programming model is proposed to obtain an optimal operating condition for both qualitative and quantitative factors. Relaxation and constraint enforcement concepts are also presented to solve the proposed optimization model. Besides, comparison studies of the proposed optimization model and counterparts are also conducted. Finally, the proposed methodology may have a potential impact to enhance complex engineering situations for both qualitative and quantitative factors in a linearly restricted experimental design region.

Traditional anomaly detection procedures assume that normal observations are obtained from a single distribution. However, due to the complexities of modern industrial processes, the observations may belong to multiple operating modes with different distributions. In such cases, traditional anomaly detection procedures may trigger false alarms while the process is indeed in another normally operating mode. We propose a generalized support vector-based anomaly detection procedure called generalized support vector data description which can be used to determine the anomalies in multimodal processes. The proposed procedure constructs hyperspheres for each class in order to include as many observations as possible and keep other class observations as far apart as possible. In addition, we introduce a generalized Bayesian framework which does not only consider the prior information from each mode, but also highlights the relationships among the modes. The effectiveness of the proposed procedure is demonstrated through various simulation studies and real-life applications.

Being able to effectively measure similarity between patents in a complex patent citation network is a crucial task in understanding patent relatedness. In the past, techniques such as text mining and keyword analysis have been applied for patent similarity calculation. The drawback of these approaches is that they depend on word choice and writing style of authors. Most existing graph-based approaches use common neighbor-based measures, which only consider direct adjacency. In this work we propose new similarity measures for patents in a patent citation network using only the patent citation network structure. The proposed similarity measures leverage direct and indirect co-citation links between patents. A challenge is when some patents receive a large number of citations, thus are considered more similar to many other patents in the patent citation network. To overcome this challenge, we propose a normalization technique to account for the case where some pairs are ranked very similar to each other because they both are cited by many other patents. We validate our proposed similarity measures using US class codes for US patents and the well-known Jaccard similarity index. Experiments show that the proposed methods perform well when compared to the Jaccard similarity index.