INVITED SPEAKER

Azadeh is a postdoc researcher at Jiangsu University, China. She has a BS in statistics and an MS in mathematical statistics from the University of Guilan, Rasht, Iran. She graduated with a PhD in statistical inference from Ferdowsi University of Mashhad, Mashhad, Iran, in 2019. 

She has been a visiting professor at Mashhad University since 2020 and at the University of Guilan since 2022. She has taught various courses in statistics for BS students, such as Statistics and Probability for Engineering Students, Mathematical Statistics I, Fundamentals of Probability, and Statistics II.

She has published over 10 publications in international journals and proceedings in international seminars.

Title: Profile control chart based on maximum entropy.

Abstract:

Monitoring a process over time is crucial in manufacturing processes to reduce the waste of money and time. Some charts, such as Shewhart, CUSUM, and EWMA, are popularised to monitor processes with a single intended attribute with various shift ranges. In some cases, the process quality is characterized by different types of proles. The purpose is to monitor profile coefficients instead of a process mean to detect small undesired shifts in the response variable. Small shifts are severe in manufacturing processes having restricted standards like pharmacology, etc. The classical control charts are unable to discover such shifts and end in unqualied productions or capital wastage. In this paper, we tackle the issue with a nonparametric (distribution-free) chart made of the maximum entropy principle for linear profile data. We have an explanatory variable here, but it is easy to extend it to have more independent variables. Two methods are proposed for surveying the intercept and slope of the simple linear profile, one by one and simultaneously. Then, their results are compared here along with the classical method. The first method is based on the maximum entropy principle, and the second is the linear regression. The T²-Hotelling statistics applied to transform two coefficients into a scalar. So, if we add more variables to the model, it is easy to handle them because all the variables are transformed into a scalar. Therefore, it is dimensional-friendly and can be extended to models with more variables and parameters. Finally, a simulation study is applied to see the performance of both methods in terms of the second type of error and average run length. Eventually, two real data examples are presented to demonstrate the applicability of the proposed chart. The first is semiconductors, and the second is pharmaceutical production processes. The performance of methods is relatively similar. The maximum entropy plays an overriding role in correctly identifying differences in the pharmaceutical example, while linear regression misdiagnosed these changes, although it acts better than the classical method. The classical method detects some control points as out-of-control and vice versa.