iso26262-agent

AI-powered ISO 26262 Functional Safety Agent for Automotive

An MCP (Model Context Protocol) server that provides AI-assisted functional safety engineering tools for automotive ECU development. Generates HARA, FMEA, safety requirements, and compliance checks using RAG + LLM — grounded in ISO 26262 standards.

Built for engineers working on safety-critical systems: inverters, motor controllers, TCUs, ADAS, autonomous drive platforms, and battery management systems.

---

🎯 What It Does

Tool	Input	Output
HARA Generator	System/function description	Hazard events + ASIL classification (A/B/C/D)
FMEA Generator	Component description	Full FMEA table (Failure Mode, S, O, D, RPN, Actions)
Safety Requirements	Function + ASIL level	ISO 26262-compliant FSR/TSR/SWSR requirements
Compliance Checker	Your existing requirements	Gap analysis vs ISO 26262 with clause references

---

🏗️ Architecture

┌─────────────────────────────────────────────┐
│              MCP Client (IDE/CLI)            │
└──────────────────┬──────────────────────────┘
                   │ MCP Protocol
┌──────────────────▼──────────────────────────┐
│           iso26262-agent (MCP Server)        │
│                                              │
│  ┌─────────┐ ┌─────────┐ ┌──────────────┐  │
│  │  HARA   │ │  FMEA   │ │ Requirements │  │
│  │Generator│ │Generator│ │  Generator   │  │
│  └────┬────┘ └────┬────┘ └──────┬───────┘  │
│       │            │             │           │
│  ┌────▼────────────▼─────────────▼────────┐ │
│  │         RAG Pipeline                    │ │
│  │  Query → Embed → Search → Context      │ │
│  └────────────────┬───────────────────────┘ │
│                   │                          │
│  ┌────────────────▼───────────────────────┐ │
│  │         LLM (Gemini API)               │ │
│  │  Context + Prompt → Structured Output  │ │
│  └────────────────────────────────────────┘ │
└──────────────────────────────────────────────┘
                   │
┌──────────────────▼──────────────────────────┐
│          Knowledge Base (194 chunks)         │
│                                              │
│  • ISO 26262 Parts 1-12 concepts            │
│  • ASIL determination tables (S/E/C → ASIL) │
│  • Automotive failure mode patterns          │
│  • Safety requirement templates              │
│  • HARA methodology & examples              │
└──────────────────────────────────────────────┘

---

🛠️ Tech Stack

• Python 3.11+
• MCP Protocol — Model Context Protocol server (standardized AI tool interface)
• Gemini API — LLM for reasoning and generation
• RAG Pipeline — Semantic embeddings + vector search (with TF-IDF fallback)
• Pydantic — Structured data models and validation
• Rich — Formatted CLI output

---

🚀 Quick Start

1. Clone & Install

git clone https://github.com/Abinash009/iso26262-agent.git
cd iso26262-agent
pip install -r requirements.txt

2. Set API Key

export GOOGLE_API_KEY=your_gemini_api_key_here

Or create a .env file:

GOOGLE_API_KEY=your_key_here

3. Run Demo

python3 demo.py

This runs all 3 tools with example automotive inputs and shows formatted output.

4. Run as MCP Server

python3 src/server.py

---

📋 Usage Examples

HARA Generator

Input: "Torque control function in traction inverter for electric truck"

Output:
┌──────────────────────────────────────────────────────────────┬──────┐
│ Hazard                                                       │ ASIL │
├──────────────────────────────────────────────────────────────┼──────┤
│ Unintended forward over-torque while vehicle is stationary   │ D    │
│ Unintended excessive negative torque (regen braking)         │ C    │
│ Sudden total loss of drive torque during overtaking           │ B    │
│ Unintended reverse torque while forward gear engaged         │ D    │
│ Unintended forward motion when reverse gear engaged          │ A    │
└──────────────────────────────────────────────────────────────┴──────┘

FMEA Generator

Input: "DC-DC converter 800V to 12V for auxiliary systems"

Output:
┌───────────────────────────────────────┬───┬───┬───┬─────┬──────────────────────────┐
│ Failure Mode                          │ S │ O │ D │ RPN │ Recommended Action       │
├───────────────────────────────────────┼───┼───┼───┼─────┼──────────────────────────┤
│ HV to LV Isolation Breakdown          │10 │ 2 │ 4 │ 80  │ Reinforced galvanic iso  │
│ Loss of Output (Zero Voltage)         │ 9 │ 4 │ 3 │ 108 │ Dual 12V output paths    │
│ Uncontrolled Output Overvoltage       │ 8 │ 3 │ 2 │ 48  │ Redundant OV protection  │
│ Shoot-Through (High/Low Side)         │ 8 │ 3 │ 2 │ 48  │ HW-enforced dead-time    │
│ Excessive Junction Temperature        │ 6 │ 4 │ 2 │ 48  │ Dual NTC sensors         │
└───────────────────────────────────────┴───┴───┴───┴─────┴──────────────────────────┘

Safety Requirements Generator

Input: function="Emergency braking torque delivery", ASIL=D

Output:
[FSR-001] [FSR] The system shall prevent unintended emergency braking torque delivery
[FSR-002] [FSR] The system shall deliver requested torque within specified time
[FSR-003] [FSR] The system shall maintain symmetrical braking torque delivery
[TSR-001] [TSR] Dual-channel cross-validation of torque demand signals
[TSR-002] [TSR] Dual independent power supply for hydraulic actuation
[TSR-003] [TSR] Redundant pressure sensors for left/right monitoring

---

📁 Project Structure

iso26262-agent/
├── README.md                    # This file
├── demo.py                      # Quick demo script (run this!)
├── requirements.txt             # Python dependencies
├── src/
│   ├── __init__.py              # Package initialization
│   ├── server.py                # MCP protocol server + CLI
│   ├── tools.py                 # 4 safety tools + SafetyAgent class
│   ├── rag.py                   # RAG pipeline (embeddings + search)
│   └── models.py                # Pydantic data models
├── knowledge_base/
│   ├── asil_determination.md    # ASIL classification tables
│   ├── failure_modes_automotive.md  # ECU failure mode patterns
│   ├── hara_methodology.md      # HARA process & examples
│   ├── iso26262_overview.md     # ISO 26262 Parts 1-12 summary
│   └── safety_requirements_patterns.md  # Requirement templates
├── tests/
│   └── test_tools.py            # Unit tests
└── docs/

---

🔑 Key Design Decisions

Decision	Rationale
MCP Protocol	Standardized AI tool interface — works with any MCP client
RAG over fine-tuning	Updatable knowledge base, no retraining needed, source traceability
TF-IDF fallback	Works without API key for demos and testing
Pydantic models	Type-safe, validated, serializable structured output
Automotive-specific KB	Not generic FMEA — tailored to ECU failure patterns
Separated concerns	RAG, Tools, Models, Server — each independently testable

---

🎯 Supported Automotive Domains

• Power Electronics — Inverters, DC-DC converters, motor controllers
• Connectivity — TCU, V2X, telematics
• ADAS / Autonomous Drive — Camera, radar, sensor fusion, path planning
• Battery Management — BMS, cell balancing, thermal management
• Braking / Steering — Brake-by-wire, EPS, stability control
• Body Electronics — Lighting, door control, climate

---

⚠️ Disclaimer

This tool assists engineers in safety analysis — it does not replace human judgement for safety-critical decisions. All outputs should be reviewed by qualified functional safety engineers before use in production.

---

👤 Author

Abinash Kumar — AI & Automotive Connectivity Engineer

• LinkedIn: abinash-kumar-26261657
• GitHub: Abinash009

Related Projects

• sync-spec-agent — RAG + MCP agent for telecom standards
• tcu-test-agent — Automotive ECU test automation (Robot Framework)

---

📄 License

MIT License