How to find the variables causing outliers in a multivariate autocorrelated process control: a study in simulation and an extruder machine

• Published in: International journal of advanced manufacturing technology 2022-09, Vol.122 (3-4), p.1497-1511
• Main Authors: Ueda, Renan Mitsuo, Souza, Adriano Mendonça
• Format: Article
• Language: English
• Subjects: Autocorrelation; Autoregressive models; CAE) and Design; Computer-Aided Engineering (CAD; Control charts; Decomposition; Engineering; Extrusion; Industrial and Production Engineering; Mechanical Engineering; Media Management; Multivariate analysis; Original Article; Outliers (statistics); Process controls; Time series; Variables
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-022-10000-0

Conventional multivariate control charts (MCCs) have difficulties finding the responsible variable for an outlier in the manufacturing system in which the variables have autocorrelation. This study suggests a promising solution to solve this problem by combining two different techniques: the vector autoregressive (VAR) model and the T 2 decomposition method; we also compared this proposal with two other traditional MCCs. This study was guided by the design research methodology, and in order to analyze the feasibility of this proposal, simulations were performed, and a test of the method was made using real data of the operation of an extruder machine. Three variables were monitored in the extrusion process, totaling 769 measurements for each time series. At the end of the study, this proposed method was indicated to find the time series causing the signal outliers: (i) VAR residuals helped better explain the causality between the time series; (ii) the T 2 chart effectively indicated the signal outliers; and (iii) the time series causing this variability was found using the T 2 decomposition technique.