lgdimaggio/predictive-maintenance-mcp/report-generation

Report Generation

Orchestrate predictive-maintenance-mcp tools to produce professional diagnostic reports. Supports time-domain plots, FFT spectrum, envelope analysis, ISO 20816-3 evaluation, feature comparison, PCA visualization, and full DOCX documents.

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

Available Report Types

Workflow

Step 1 — Determine Report Type

Ask the user or infer from context which report(s) are needed. For a "full" or "comprehensive" request, generate multiple reports in sequence.

Step 2 — Verify Signal Availability

Call list_signals(scope="memory") to confirm the signals are loaded. Load with load_signal(filepath="<file>", signal_unit="g") if needed (raw binary .bin/.raw/.dat files additionally need sample_format and sampling_rate — see the signal-management skill). All report tools take signal_id — the handle returned by load_signal.

Step 3 — Gather Parameters and Generate

Time-domain plot: plot_signal(signal_id="<id>", time_range=[0.0, 2.0], show_statistics=True)

FFT report: generate_fft_report(signal_id="<id>", max_freq=5000, num_peaks=15, rpm=1797) (rpm is optional — when given, harmonic markers are drawn at 1x/2x/3x RPM)

Envelope report: generate_envelope_report(signal_id="<id>", filter_low=500, filter_high=5000, max_freq=500, bearing_freqs={"BPFO": 107.4, "BPFI": 162.2}, rpm=1797) (bearing_freqs overlays expected fault frequency markers; pass rpm only as the measured shaft speed: it overrides the speed the signal declares, and without either the shaft-order and sideband tests do not run)

ISO 20816-3 report: generate_iso_report(signal_id="<id>", machine_group=2, support_type="rigid")

• machine_group is 1 (large, >300 kW) or 2 (medium, 15-300 kW)
• The signal must have a DECLARED unit; if the assessment is refused, re-load

with load_signal(filepath="<file>", signal_unit="mm/s", overwrite=True) after confirming the unit with the user.

Feature comparison (signal_groups maps a label to a list of signal_ids): generate_feature_comparison_report(signal_groups={"baseline": ["baseline_1"], "faulty": ["OuterRaceFault_1"]}, features_to_plot=["rms", "kurtosis"])

PCA visualization (requires a model trained via train_anomaly_model): generate_pca_visualization_report(model_name="anomaly_model", test_signal_ids=["<id>"])

Full DOCX (sections maps section titles to content): generate_diagnostic_report_docx(signal_id="<id>", title="Pump P-101 diagnostic", sections={"Summary": "...", "Findings": "..."})

Step 4 — Report to User

After generation:

Report generated successfully:
  Type: {report_type}
  File: {returned file path}
  Signal: {signal_id}
 
Key findings:
  {2-3 bullet summary}
 
Open the HTML file in a browser for the full interactive report.

Report filenames are timestamped — consecutive runs never overwrite each other. Call list_html_reports() to list all generated reports, or list_html_reports(file_name="<report>.html") for one report's details.

Step 5 — Composite Reports (Optional)

For "full" or "comprehensive" requests, generate in sequence:

1. FFT spectrum report
2. Envelope analysis report
3. ISO 20816-3 report
4. Full DOCX report (synthesizes everything)

Provide a combined summary referencing all report paths.

Important Notes

• NEVER display raw HTML content in chat — always provide the file path
• Signal unit declaration is CRITICAL for ISO reports (no unit = refusal)
• Reports are self-contained HTML with embedded Plotly charts
• DOCX reports include embedded images and can be shared externally
• Reports document the evidence for an engineer's decision — they do not

replace engineering judgment

• All processing is local — reports contain only processed results, not raw data