SP-CTNG: Supporting Data and Reproducibility Materials for Stability-Guided Cluster-Number Estimation and Partition Selection for Fault-State Discovery in Industrial Time Series

This record provides supporting data and reproducibility materials for the manuscript “Stability-Guided Cluster-Number Estimation and Partition Selection for Fault-State Discovery in Industrial Time Series.” The archive contains six fixed PCA32 clustering inputs derived from the Tennessee Eastman Process (TEP) and Case Western Reserve University (CWRU) datasets, 486 formal CTNG candidate partitions, six selected SP-CTNG plateaus, 180 seed-level final partitions comprising 134,760 sample-label rows, task and source mappings, supplementary analyses, MATLAB implementation, reconstruction scripts, environment specifications, and validation records. SP-CTNG applies a stability-guided partition-selection layer to ordered candidate paths generated by CTNG. Cluster-count persistence and minimum pairwise partition agreement are evaluated without reference labels, and consensus aggregation is applied only after the selected plateau has been fixed. Independent fixed-feature validation reproduced all 486 formal candidate partitions, all six selected plateaus, and all 180 final partitions. The archive also contains search-range and threshold-subset sensitivity records, controlled same-cluster-count membership-drift records, graph-generated synthetic cases, and Figure 5 reconstruction support. Original TEP/CWRU records and third-party baseline source code are not redistributed. Source links, mappings, version information, and execution guidance are provided in the archive.