FASTENER CLUSTER (K-MEANS)¶
Groups fasteners into k clusters using k-means with k-means++ initialization on selected numeric features (default: peak_axial_kN, peak_shear_kN, fos_min). Returns per-fastener cluster labels + per-cluster summary (count, centroid in real units, sample IDs). Useful for identifying ‘families’ of similarly-loaded joints in large datasets (e.g. cluster the BiW bolts that all see similar crash load profiles, separate from the few outliers driving failures). Standalone in-PHP implementation — no external ML dependencies. Pipe master_table from lsdyna_fastener_failure_analysis or lsdyna_joint_map_batch_analysis directly into this worker.
When to use¶
Tagged: cluster, fastener, fos, grouping, k-means, kmeans, ml, outlier.
Inputs¶
| Label | ID | Type | Default | Required | Description |
|---|---|---|---|---|---|
| Master Table | master_table | dataset | — | Per-fastener row dataset from a previous fastener-analysis worker. | |
| Number of Clusters (k) | num_clusters | integer | 4 | Target cluster count (2-20 typical). | |
| Features (CSV) | features | string | peak_axial_kN,peak_shear_kN,fos_min | Comma-separated numeric column names from master_table to cluster on. | |
| Max Iterations | max_iterations | integer | 50 | k-means iteration cap (converges early on most datasets). | |
| Random Seed | random_seed | integer | 42 | Seed for reproducible k-means++ initialization. |
Outputs¶
| Label | ID | Type | Description |
|---|---|---|---|
| Cluster Labels | labels | dataset | Per-fastener: id + cluster integer (0 .. k-1). |
| Cluster Summary | cluster_summary | dataset | Per-cluster: count, real-unit centroid for each feature, z-score centroid, up to 10 sample IDs. |
| Meta | meta | dataset | Run metadata — k, features used, iterations, points used/skipped, feature mean/std stats. |
Disciplines¶
- engineering.fastener.analysis
- ml.unsupervised.clustering
Auto-generated from platform schema. Worker id: fastener_cluster. Schema hash: 39417225968e. Hand-curated docs in workerexamples/ override this page when present.