Signal Management

Signal Management

Load, generate, inspect, and manage vibration signals in the predictive-maintenance-mcp in-memory repository. Supports CSV, TXT, NPY, WAV, MAT (MATLAB), and Parquet formats.

Prerequisite: The predictive-maintenance-mcp MCP server must be connected.

Core Operations

List Available Signals

On disk: Call list_signals() to browse files in the data/ directory.

In memory: Call list_stored_signals() to see signals currently loaded in the cache with their metadata (sampling rate, duration, sample count).

Load a Signal

Call load_signal(file_path=..., signal_id=..., sampling_rate=...).

• file_path: path to the signal file (CSV, TXT, NPY, WAV, MAT, Parquet)
• signal_id: a short identifier for referencing the signal later (e.g., "bearing_normal_01")
• sampling_rate: required for formats that don't embed it (CSV, TXT, NPY). WAV files include the sampling rate automatically.

If the user provides a file path without specifying a signal_id, derive one from the filename (e.g., "bearing_data.csv" -> "bearing_data").

Inspect Signal Metadata

Call get_signal_info(signal_id=...) to get metadata without loading the full array: sampling rate, duration, sample count, min/max values.

Generate Test Signals

Call generate_test_signal(...) to create synthetic signals for testing.

Parameters:

• fault_type: "normal", "outer_race", "inner_race", "ball_defect", "cage_fault", "gear_mesh", "unbalance", "misalignment"
• duration: signal duration in seconds (default 1.0)
• sampling_rate: Hz (default 10000)
• shaft_rpm: rotational speed (default 1800)
• snr_db: signal-to-noise ratio in dB (default 20)
• signal_id: identifier for the generated signal

This is useful for:

• Demonstrating the diagnostic workflow without real sensor data
• Testing skills and reports with known fault signatures
• Training users on vibration analysis

Clear Signals

• clear_signal(signal_id=...) — remove a specific signal from cache
• clear_all_signals() — clear all cached signals

Visualize a Signal

Call plot_signal(signal_id=...) to generate an HTML time-domain plot of the raw waveform.

Workflow for New Users

1. Call list_signals() to see what data files are available
2. Load one with load_signal(file_path=..., signal_id=...)
3. Call get_signal_info(signal_id=...) to verify it loaded correctly
4. Call plot_signal(signal_id=...) for a quick visual check
5. Proceed with analysis (quick-screening, bearing-diagnosis, etc.)

If no data files are available, generate a test signal:

1. Call generate_test_signal(fault_type="outer_race", signal_id="test_outer")
2. Proceed with analysis on "test_outer"

Important Notes

• Signals are cached in memory; they persist for the session but not across server restarts
• Large signals (>100 MB) may take a few seconds to load
• CSV files must have a single numeric column or the first numeric column is used
• Always confirm sampling_rate for formats that don't embed it
• All signal data stays local — nothing is transmitted externally