Static Friction Breakage Predictor

Objective and Goals

The objective of this project is to develop a machine-learning model capable of predicting the static friction breakage between two materials using physical property ratings and existing friction datasets.
The goal is to create a more accessible, scalable, and cost-efficient alternative to traditional friction experiments and to evaluate whether ML can outperform or match LLM-based predictions.

Requirements

Hardware / Tools

  • Laptop or server

  • Force scale (for optional experimental validations)

Frameworks / Libraries

  • Python

  • TensorFlow / Keras

  • NumPy / Pandas

  • Scikit-Learn

  • API access for LLM property rating generation

  • SQLite (optional future database integration)

Planning

Milestones and Timeline

  • Collect publicly available static friction datasets

  • Build algorithm to extract material names and generate material list

  • Use LLMs to assign 12 physical property ratings to each material

  • Integrate and preprocess datasets (cleaning, combining, feature engineering)

  • Train main ML model to predict friction coefficients

  • Test model accuracy, analyze errors and safety margins

  • Compare ML predictions vs LLM predictions

  • Document limitations and finalize research findings

Roles and Responsibilities

(Draft — roles can expand based on member skills)

Dogac Agirtici — Project Management / Machine Learning Development

  • Dataset collection and integration

  • Creation of material property-rating pipelines

  • Development of hybrid ML model architecture

  • Feature engineering and evaluation

Andranik Khachikyan — Research / Physics Analysis

  • Background research on friction theory

  • Identifying core physical properties influencing friction

  • Validating LLM-generated property scales

  • Supporting experimental design

Johnny Haro — Software & Data Engineering

  • Data preprocessing automation

  • Algorithm implementation and debugging

  • Testing model accuracy and verifying output integrity

Execution

Development Process

  • Weekly development sessions (2 hrs recommended per person)

  • GitHub for collaboration and version control

  • Python used for all model development

  • Iterative approach: prototype → test → refine → evaluate

  • Continuous comparison of ML outputs with real-world friction data

Weekly Meetings

  • Held online (Discord)

  • 30–60 minutes per week

  • Scheduling based on team availability

  • Meeting notes and deliverables stored in shared repository