Clustering techniques are tailored to find internally homogeneous groups of observations. In industrial processes that rely on batches, grouping batches with similar profiles provides valuable information about process control and monitoring. This paper proposes a variable selection approach based on the kernel function and Principal Component Analysis (PCA). The clustering quality is assessed through the Silhouette Index (SI). When applied to three industrial processes, the proposed approach retained an average of 5.16% of the original variables, yielding on average a 235.4% more precise batch grouping. We also performed a simulation experiment.
Clustering analysis; Variable selection; Kernel; Batch processes
