Methods for interpreting the out‐of‐control signal of multivariate control charts: A comparison study

Abstract

Multivariate control charts have proved to be a useful tool for identifying an out‐of‐control process. However, one of the main drawbacks of these charts is that they do not indicate which measured variables have been shifted. To overcome this issue, several alternative approaches that aim to diagnose faults the responsible variable(s) for the out‐of‐control signal and help identify aberrant variables may be found in the literature. This paper reviews several techniques that are used to diagnose the responsible variable(s) for the out‐of‐control signal and attempt to make a comparative study among them. In particular, we evaluate the performance of each method under different simulation scenarios in terms of successful identification of the out‐of‐control variables. Special attention has also been given to the computational approaches and especially in the ability of artificial neural networks to identify out‐of‐control signals. The obtained results show that: there is no particular method that can be considered as panacea; the artificial neural networks' performance depends heavily on the training data set.

Using a Time Delay Neural Network Approach to Diagnose the Out-of-Control Signals for a Multivariate Normal Process with Variance Shifts

“…However, one difficulty in the recognition phase that will be encountered is that the number of categories of classifiers' output nodes increases when more categories are involved with an attribute multivariate process. Other soft computing classifiers, such as the extreme learning machine, rough set, random forest and hybrid modeling techniques [37,38], may be worth investigating to decrease the number of output categories in the future.…”

Misleading signals in joint schemes for the mean vector and covariance matrix

“…As expertly put by Zhang et al, it is important to develop procedures that can be employed after a signal for diagnostic purposes so that we are able to pinpoint which parameter has actually shifted, the mean vector or the covariance matrix. Proposing such procedures is what we believe the most important future task, and they may comprise the use of additional control charts or other plots, as proposed by Lou and Reynolds in the univariate case, or to adapt the techniques that authors such as Maravelakis et al, Das and Prakash, and Bersimis et al used to diagnose faults in out‐of‐control conditions while monitoring the mean vector.…”

Exponential random graph modeling for diagnosis of out‐of‐control signals in social network surveillance: Signal interpretation in social networks with ERG modeling

“…Then all pairs of variables with significant term are removed. The higher‐order terms are computed in this way until no variable is left in the reduced set 26 . Using this way all variables and their combination that are responsible for out‐of‐control signals are detected.…”

“…A complimentary comparison and review studies were performed by Puig 25 that compared more methods for multivariate statistical quality control and also some methods for latent‐based multivariate statistical process control. Bersimis et al 26 extended Das and Parkash 20 research, which is based on artificial neural networks and seems to be one of the most promising techniques for fault diagnosis.…”

“…So when any changes in global behavior pattern or the behavior of the network components happened that leads to a fundamental change in the network, the control chart signals and diagnosis approach will interpret this out‐of‐control situation. For this purpose, in this study, MTY, 25,26,28 which is one of the most promising methods for diagnosing out‐of‐control signals, is employed.…”