Data Pre-processing and Sensor-Fusion for Multivariate Statistical Process Control of an Extrusion Process

Abstract

In most manufacturing processes, data related to a product are collected across several process steps. Ensuring good data quality is essential for subsequent process modeling, monitoring, and control. Although data for a given process might already be available in digitized form in the process control systems or industrial databases, it is in most cases not so that the data can directly be used in its original form for process modeling. Pre-processing is often needed before modeling, which may include operations such as time alignment by handling different sampling frequencies and lag time, handling of missing values, and detection of sample outliers. Specific considerations must be made for processes with both continuous and batch process steps due to different data structures. This paper describes an industrial use case for extrusion monitoring starting from structured raw data and ending up with real-time multivariate statistical process control (MSPC) applying a sensor-fusion approach and feature extraction. The MSPC also enables in-depth analysis for identifying process variables in the case of samples lying outside of the normal operating conditions (NOC).