Step Change Point Estimation in the Mean of Multiple Linear Profiles by Probabilistic Neural Network
Volume 10, Issue 1, Spring 2020, Pages 34-48
https://doi.org/10.48313/jqem.2020.115126
Negin Forouzandeh, Mona Ayoubi, Masoomeh Zeinalnezhad
Abstract The change point is a useful concept in the control of statistical process that assists quality engineers in finding assignable causes and improving the quality of a product or process. It also reduces the time and cost of detecting assignable causes. In this paper, an artificial neural network approach is used to estimate the step change point in phase II monitoring of the mean of multiple linear profiles. the performance of the probabilistic neural network is evaluated in order to estimate the change point utilizing Monte Carlo Simulation. The results of simulations show the fact that the network recommended in estimating the change point has entirely better performance than maximum likelihood estimator in small shifts, considering mean square error criteria, but maximum likelihood estimator method owns better performance in medium to large shifts. In general, in all shift types, maximum likelihood estimator has a better performance in terms of precision, and the proposed probabilistic neural network performs better in terms of accuracy. In addition, another advantage of the proposed approach is the fact that contrary to the maximum likelihood estimator approach, it does not require any knowledge about the change type and can appropriately estimate any kind of change point as well.
Detection of Out-of-Control Parameters in Polynomial Profiles Using an Artificial Neural Network
Volume 2, Issue 1, Spring 2012, Pages 21-26
Reza baradaran kazem zade, Mona Ayoubi
Abstract Profile monitoring is one of the emerging research areas in the field of statistical process control. A profile describes the relationship between a response variable and one or more independent variables. This relationship, which is typically modeled using a regression equation, may be simple linear, multiple linear, polynomial, or in some cases nonlinear. In order to monitor the quality of a process or product whose quality characteristic is expressed as a profile, multivariate control charts must be used due to the correlation among the regression model parameters. Various types of multivariate control charts can be applied to detect small and large shifts in the process.
A major limitation of multivariate control charts is their inability to identify the specific parameter that goes out of control once a change is detected. In this study, after a multivariate MCUSUM control chart in the MCUSUM–Chi-square method signals a shift in a polynomial profile, a multilayer perceptron neural network is employed to identify the out-of-control parameter. The designed network is trained and then tested, and the results demonstrate its strong performance in identifying the parameter that has shifted out of control.