Skip to content

Latest commit

 

History

9 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

RoadRakshak

Predictive Pavement Maintenance SystemRoad Rakshak is a data-driven Machine Learning project designed to predict road roughness—specifically measured by the Mean Roughness Index (MRI)—before major damage or potholes occur.Instead of waiting for roads to deteriorate and relying on expensive, reactive repairs, this project builds a predictive system that simulates how traffic loads, weather conditions, and structural design interact to break down roads over time.

💡 The Core Idea & The FASTag Connection To train a highly reliable machine learning model, decades of precise structural history are required. For this project, the model was trained using the massive, real-world Long-Term Pavement Performance (LTPP) database managed by the Federal Highway Administration (FHWA). How it works in the real world: While historical data was used to teach the model the laws of road physics, the system is designed to be deployed using live FASTag traffic volumes collected at national toll plazas. The cumulative vehicle traffic counted by electronic tolling scales serves as the primary live input to predict future highway deterioration dynamically.

📊 The Dataset & Cleaning (The Cartesian Purge) The raw database contained over 290,000 messy laser-scan entries taken across various timelines. To make the data mathematically sound, a rigorous cleaning process was executed:Sensory Noise Removal: Multiple laser scans often reported slightly conflicting roughness scores for the exact same road conditions. These conflicting entries were mathematically averaged to create a stable ground truth.The Cartesian Purge: Duplicate records caused by timeline misalignment across separate databases were tracked down and eliminated. The Clean Output: The raw data was successfully refined down to 42,311 pure, unique physical road states containing structural, traffic, and climate variables.

🛠️ Feature Engineering (Injecting Pavement Physics) Initial models built using only basic traffic and climate data hit a strict accuracy ceiling of around 54%. Roads break down differently based on how thick they are and what they are made of. To fix this, several physical attributes were engineered into the dataset:Pavement Age: Calculated by finding the exact time difference between the road's construction year and the year the inspection took place.Total Structural Thickness: Combined the measurements of all structural layers of the road to capture its literal load-bearing capacity.Initial Roughness ($IRI_0$): Sourced the very first roughness rating of the road when it was brand new to establish a starting baseline.Material Category Encoding (One-Hot Encoding): Categorized road types (like Asphalt vs. Concrete) into independent binary columns. This prevents the machine learning models from misinterpreting arbitrary text labels as numbers, allowing the math to adapt to different physical material behaviors.

Datasets & Data Requirements

This project utilizes specific datasets for training and road structural distress prediction. Due to their size, these files are not included in this repository.

Filename Description Purpose
COLLECTED_RAW_DATA.csv Initial VAHAN/FASTag logs Raw ingestion
X_scaled_final.csv Pre-processed feature set Model training
bucket.mdb Historical pavement health database from LTTP requested data Ground truth mapping
enhanced_raw_master.csv Combined telematics master file Predictive analysis

About

Predicting Pavement Structural Distress using Traffic Volumes derived from FASTag International Roughness Index Modeling For Jointed Plain Concrete Pavement

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages