v0.3.0 — LLM Edge Predictive Maintenance

What's Changed

Project renamed from Claude Edge Predictive Maintenance to LLM Edge Predictive Maintenance to clarify this is an independent open-source project with no Anthropic or STMicroelectronics affiliation.

Bug Fixes

• ISO 10816 g→mm/s conversion — added frequency-domain integration (_accel_g_to_velocity_rms_mms) so \diagnose_vibration\ reports correct ISO 10816 velocity in mm/s RMS instead of raw g values
• Skill guardrails — all three skills now enforce data-integrity rules: no guessing RPM from spectral peaks, no bearing hypotheses without evidence, mandatory LLM-knowledge disclosure

Docs & Meta

• Disclaimer added to README, CITATION.cff, and landing page footer
• Landing page and BibTeX updated to reflect new project name
• CITATION.cff version bumped to 0.3.0

Full Changelog: v0.2.1...v0.3.0

v0.2.1: Landing Page, Citation & Contrast Fixes

What's New

Landing Page

• GitHub Pages site with Tailwind CSS, dark theme, glass morphism
• System overview, architecture diagrams, conversation examples
• Full responsive design with reveal animations

Citation Update

• Title: "Claude Edge Predictive Maintenance" (was "Claude Edge Diagnostics")
• "IoT Sensors" (was "MEMS Sensors")
• DOI 10.5281/zenodo.18808856 added to BibTeX
• CITATION.cff bumped to v0.2.0

UI/Contrast

• All text white for maximum contrast on dark background
• Font sizes bumped across all sections
• Glass cards more opaque (12% bg, 20% border)
• Stronger CTA button with gradient and glow
• Hero gradient spans full "Claude Edge Predictive Maintenance"

Full Changelog

v0.2.0...v0.2.1

v0.2.0 — DATALOG2 Live Acquisition & Skills Update

v0.2.0: DATALOG2 live acquisition, SDK integration, skills update

New Features:
- DATALOG2 USB-HID live acquisition with duration_s (server-side timer)
- Native .dat file parsing via STDATALOG-PYSDK (sensor + analysis venvs)
- diagnose_vibration with optional RPM (never hallucinate RPM)
- load_signal accepts DATALOG2 acquisition folders (auto-detect ODR)

Skills:
- All 3 skills updated with correct MCP tool names
- DATALOG2 workflow as primary acquisition path
- Proper frontmatter with metadata and trigger phrases
- Pre-built .zip files in skills-zips/ for Claude Desktop install

Docs:
- README aligned with current project state
- SDK install documented for both server venvs
- Claude Desktop config updated for SDK mode

v0.1.0 — Initial Release

Claude Edge Diagnostics v0.1.0

First public release — AI-powered predictive maintenance bridging STEVAL-STWINBX1 edge sensors and Claude via MCP.

Components

• stwinbox-sensor-mcp — 12 MCP tools for STWIN.box USB/serial acquisition
• vibration-analysis-mcp — 11 MCP tools: FFT, PSD, envelope analysis, bearing faults, ISO 10816
• machine-vibration-monitoring skill — sensor setup & baseline workflow
• vibration-fault-diagnosis skill — multi-step fault diagnosis
• operator-diagnostic-report skill — maintenance report generation

Highlights

• Full DSP pipeline: FFT, PSD, spectrogram, Hilbert envelope
• Bearing fault frequencies (BPFO/BPFI/BSF/FTF) with built-in database
• Automated fault classification (unbalance, misalignment, looseness, bearing defects)
• ISO 10816 vibration severity assessment
• CITATION.cff for Zenodo DOI integration

⚠️ This is an early proof-of-concept release. See README for disclaimer.