International Journal of Industrial Engineering & Production Research

---

In certain statistical process control applications, quality of a process or product can be characterized by a function commonly referred to as profile. Some of the potential applications of profile monitoring are cases where quality characteristic of interest is modelled using binary,multinomial or ordinal variables. In this paper, profiles with multinomial response are studied. For this purpose, multinomial logit regression (MLR) is considered as the basis.Then, the MLR is converted to Poisson GLM with log link. Two methods including Multivariate exponentially weighted moving average (MEWMA) statistics, and Likelihood ratio test (LRT) statistics are proposed to monitor MLR profiles in phase II. Performances of these three methods are evaluated by average run length criterion (ARL). A case study from alloy fasteners manufacturing process is used to illustrate the implementation of the proposed approach. Results indicate satisfactory performance for the proposed method.

---

Studying the social networks plays a significant role in everyone’s life. Recent studies show the importance and increasing interests in the subject by modeling and monitoring the communications between the network members as longitudinal data. Typically, the tendency for modeling the social networks with considering the dependency of an outcome variable on the covariates is growing recently. However, these studies fail in considering the possible correlation between the responses in the modeling of social networks. Our study use generalized linear mixed models (GLMMs) (also referred to as random effects models) to model the social network according to the attributes of nodes in which the nodes take a role of random effect or hidden effect in the modeling. The likelihood ratio test (LRT) statistics is implemented to detect change points in the simulated network streams. Also, in the simulation studies, we applied root mean square Error (RMSE) and standard deviation criteria for choosing an appropriate distribution for simulation data. Also, our simulation studies demonstrates an improvement in the average run length (ARL) index in comparison to the previous studies.
 

---

The number of nonconforming items in a sample is monitored using the fraction defective known as the np-chart. The performance of the np-chart in Phase II depends on the accuracy of the estimated parameter in Phase I. Although taking large sample sizes ensures the accuracy of the estimated parameter, it can be impractical for attributes in some cases. Recently, the traditional c-chart and the np-chart with some adjustments have been studied to guarantee the in-control performance. Due to technology progresses, researchers have faced high-quality processes with a very low rate of nonconformity, for which traditional control charts are inadequate. To ameliorate such inaccuracy, this study develops a new method for designing the np-chart, such that the in-control performance is guaranteed with a pre-defined probability. The proposed method uses Cornish-Fisher expansions and the bootstrap method to guarantee the desired conditional in-control average run length. Through a simulation study, this study shows that the proposed adjustments improve the np-charts’ in-control performance.

---

In the last few decades, profile monitoring in univariate and multivariate environment has drawn a considerable attention in the area of statistical process control. In multivariate profile monitoring, it is required to relate more than one response variable to one or more explanatory variables. In this paper, the multivariate multiple linear profile monitoring problem is addressed under the assumption of existing autocorrelation among observations. Multivariate linear mixed model (MLMM) is proposed to account for the autocorrelation between profiles. Then two control charts in addition to a combined method are applied to monitor the profiles in phase II. Finally, the performance of the presented method is assessed in terms of average run length (ARL). The simulation results demonstrate that the proposed control charts have appropriate performance in signaling out-of-control conditions.
 

---

Statistical monitoring of dynamic networks is a major topic of interest in complex social systems. Many researches have been conducted on modeling and monitoring dynamic social networks. This article proposes a new methodology for modeling and monitoring dynamic social networks for quick detection of temporal anomalies in network structures using latent variables. The key idea behind our proposed methodology is to determine the importance of latent variables in creating edges between nodes as well as observed covariates. First, latent space model (LSM) is used to model dynamic networks. Vector of parameters in LSM model are monitored through multivariate control charts in order to detect changes in different network sizes. Experiments on simulated social network monitoring demonstrate that our surveillance monitoring strategy can effectively detect abrupt changes between actors in dynamic networks using latent variables.