Trains OneClassSVM and LOF models to detect anomalies in vibration signals using predictive-maintenance-mcp server. Supports PCA visualization, feature extraction, and prediction workflows.
How this skill is triggered — by the user, by Claude, or both
Slash command
/predictive-maintenance:anomaly-detectionThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Train and use machine learning models to detect anomalous vibration patterns.
Train and use machine learning models to detect anomalous vibration patterns. Supports OneClassSVM and Local Outlier Factor (LOF) algorithms with PCA visualization for clustering analysis.
Prerequisite: The predictive-maintenance-mcp MCP server must be connected.
Gather a set of signals representing NORMAL machine operation. Call
list_stored_signals() to check available signals, or load signals with
load_signal(...).
Ask the user:
For each baseline signal, call extract_features_from_signal(signal_id=...) to
verify data quality. Signals with very high kurtosis or ISO Zone C/D should be
excluded from the normal baseline.
Call train_anomaly_model(signal_ids=[...], model_type="oneclass_svm").
Parameters:
"oneclass_svm" or "lof" (Local Outlier Factor)The model is saved to the models/ directory for future use.
Run predict_anomalies(signal_ids=[...]) on the training signals themselves.
Most should be classified as normal (anomaly_score near 0). If too many are
flagged, increase contamination or review the baseline data.
Load the signals to analyze with load_signal(...).
Call predict_anomalies(signal_ids=[...]).
The tool returns for each signal:
| Anomaly Score | Classification | Action |
|---|---|---|
| Low (near 0) | Normal | No action needed |
| Medium | Borderline | Monitor more frequently |
| High | Anomaly | Investigate with bearing-diagnosis or quick-screening |
Call generate_pca_visualization_report(signal_ids=[...]) to project all
signals into 2D space. Normal signals cluster together; anomalies appear as
outliers.
npx claudepluginhub lgdimaggio/predictive-maintenance-mcp --plugin predictive-maintenancePerforms prognostic assessment on vibration signals: trend analysis, degradation onset detection, and RUL estimation using predictive-maintenance-mcp server.
Detects anomalies and outliers in datasets using ML like Isolation Forest, One-Class SVM, LOF, autoencoders. For analyzing unusual patterns or deviations.
Provides guidance on classical ML with scikit-learn: classification, regression, clustering, pipelines, preprocessing, model evaluation, and hyperparameter tuning.