AutoAssist 🚗

AI-Powered OBD-II Vehicle Health Monitoring & Anomaly Detection Platform

AutoAssist is an end-to-end vehicle telemetry analytics platform that uses OBD-II sensor data, machine learning, and interactive visualizations to detect abnormal vehicle behavior, assess vehicle health, and provide actionable insights through a modern web dashboard.

The project combines data engineering, anomaly detection, health scoring, backend APIs, and frontend analytics into a complete vehicle monitoring system.

---

Dashboard Preview

Dashboard

Anomaly Center

Telemetry Explorer

Reports

---

Features

Vehicle Health Monitoring

• Vehicle health score generation
• Risk level classification
• Session-level vehicle assessment
• Health trend analysis

Anomaly Detection

• Isolation Forest based anomaly detection
• Sensor behavior analysis
• Per-record anomaly scoring
• Binary anomaly classification
• Session anomaly summaries

Driving State Classification

Automatically classifies telemetry into:

• Idle
• Traffic
• Cruising
• Acceleration
• Deceleration

Interactive Analytics Dashboard

• Health overview
• Telemetry visualization
• Sensor exploration
• Anomaly investigation
• Session reports
• Correlation analysis

CSV-Based Workflow

• Upload vehicle telemetry datasets
• Automatic preprocessing
• Feature engineering
• Anomaly analysis
• Health assessment generation

---

System Architecture

CSV Upload
     │
     ▼
Preprocessing Pipeline
     │
     ▼
State Classification
     │
     ▼
Feature Engineering
     │
     ▼
Isolation Forest Model
     │
     ▼
Anomaly Detection
     │
     ▼
Health Score Engine
     │
     ▼
FastAPI Backend
     │
     ▼
React Dashboard

---

Machine Learning Pipeline

Feature Engineering

The platform generates engineered features from raw OBD-II telemetry:

• RPM Delta
• Speed Delta
• MAF Delta
• MAP Delta
• Throttle Delta

Input Features

rpm
speed
maf
map
throttle_pos
rpm_delta
speed_delta
maf_delta
map_delta
throttle_delta

---

Model

Algorithm:

IsolationForest(
    n_estimators=200,
    contamination="auto",
    random_state=42
)

Model Artifacts

models/
├── isolation_forest.pkl
├── scaler.pkl
└── feature_config.json

---

Validation Results

Validation performed on:

2,693,087 telemetry records

Overall Performance

Metric	Value
Total Records	2,693,087
Total Anomalies	82,823
Overall Anomaly Rate	3.08%

Driving State Analysis

Driving State	Anomaly Rate
Idle	2.86%
Traffic	2.47%
Cruising	2.70%
Acceleration	17.81%
Deceleration	12.28%

Key Findings

• Acceleration and deceleration states exhibit significantly higher anomaly rates due to transient vehicle dynamics.
• Steady-state driving (traffic, cruising, idle) maintains consistent low anomaly rates.
• The model learned a compact normal driving region with minimal anomaly inflation.

---

Technology Stack

Backend

• Python 3.11+
• FastAPI
• Pandas
• NumPy
• Scikit-Learn
• Joblib
• Pydantic

Frontend

• React
• TypeScript
• TanStack Start
• TanStack Router
• React Query
• Tailwind CSS
• Recharts

Machine Learning

• Isolation Forest
• StandardScaler
• Custom Feature Engineering Pipeline

---

Project Structure

OBDAnomalyDetector/
│
├── client/
│   ├── routes/
│   ├── components/
│   ├── hooks/
│   └── api/
│
├── server/
│   ├── preprocessing/
│   ├── analytics/
│   ├── ml/
│   └── api/
│
├── models/
│
├── data/
│
├── reports/
│
├── docs/
│
└── README.md

---

Running Locally

Clone Repository

git clone https://github.com/AadilSandeep/OBDAnomalyDetector.git

cd OBDAnomalyDetector

---

Backend Setup

Install dependencies:

pip install -r requirements.txt

Run FastAPI:

uvicorn server.api.main:app --reload

Backend URL:

http://localhost:8000

Swagger Documentation:

http://localhost:8000/docs

---

Frontend Setup

cd client

npm install

npm run dev

Frontend URL:

http://localhost:3000

---

Supported Dataset Requirements

The current version expects telemetry containing the following core signals:

RPM
Vehicle Speed
Mass Air Flow (MAF)
Manifold Absolute Pressure (MAP)
Throttle Position

Datasets using different column names may require updates to the preprocessing mapping configuration.

---

Current Limitations

• Requires a supported OBD-II telemetry schema
• Designed for offline CSV analysis
• No direct OBD-II hardware integration yet
• No fault-code (DTC) interpretation
• No real-time telemetry streaming

---

Future Scope

Dataset Flexibility

• Automatic schema detection
• Alias-based column mapping
• Support for multiple OBD-II dataset formats

Real-Time OBD-II Integration

Planned support for:

• ELM327 Bluetooth Dongles
• ELM327 Wi-Fi Dongles
• Live vehicle telemetry streaming

Future Architecture

Vehicle
    │
    ▼
OBD-II Dongle
    │
    ▼
Live Telemetry Stream
    │
    ▼
AutoAssist
    │
    ▼
Real-Time Health Monitoring

Real-Time Health Monitoring

Future versions aim to provide:

• Live anomaly detection
• Streaming health scores
• Driver alerts
• Continuous health tracking

Diagnostic Intelligence

Future research direction:

Anomaly Detection
        │
        ▼
Diagnostic Engine
        │
        ▼
Root Cause Suggestions

Potential capabilities:

• Airflow anomaly detection
• Sensor drift identification
• Intake system monitoring
• Early fault prediction

---

Project Status

Version

v1.0 MVP

Completed

• Data preprocessing pipeline
• Driving state classification
• Feature engineering
• Model training
• Anomaly detection
• Health score engine
• Validation pipeline
• FastAPI backend
• React frontend
• End-to-end CSV analysis workflow

---

License

This project was developed for academic, research, and educational purposes.

---

Author

Aadil Sandeep

Computer Science Engineering

Vehicle Analytics • Machine Learning • Automotive AI