Skip to content

The objective of the project is to classify steel plates fault into 7 different types. The end goal is to train several machine Learning Algorithms for automatic pattern recognition.

Notifications You must be signed in to change notification settings

Anurag-Singh-creator/Steel-Plate-Fault-Detection

Repository files navigation

Steel-Plate-Fault-Detection

View the Presentation

Introduction

Project Title: Applied Machine Learning and Data Engineering in Business Context Project Focus: Steel Plate Fault Detection Objective: Develop an end-to-end cloud solution for fault detection in steel plates for XYZ Products. Team Members: Alisa Ilina, Anurag Singh, Eirik Egge, Henry Stoll, Magnus Eliassen, Olivia Lundholm

Executive Summary

Identifying the Problem:

The most common production faults like Bumps, K-Scratches, and Z-Scratches are primarily influenced by factors such as conveyor length, steel plate thickness, and the steel type.

Automating Fault Detection:

Implementation of the XGBoost model which correctly classifies over 80% of all faults. This allows for the transition from manual fault classification to an automated detection system.

Moving to the Cloud:

The proposed cloud architecture leverages existing data sources, enabling efficient fault detection and providing a continuous flow of data-driven insights into the production process.

Scaling for the Future:

The vision extends beyond just analysis. The aim is to scale automated fault diagnosis across the production line, paving the way for future high-value use cases in the domain.

Value Proposition:

Delivering substantial value across the production line through cloud-based fault detection mechanisms.

Background & Challenges

Company Profile:

XYZ Products, a renowned steel manufacturer located in Northern Sweden.

Objective:

Achieve a reduction in costs associated with production faults.

Challenges:

Inefficient manual quality control mechanisms. Significant variable costs due to faulty steel plates. Plates detected as faulty post-sale, resulting in customer dissatisfaction. Manual data collection process for fault labeling. Decision-making often rooted in human assumptions instead of data-driven insights.

Model Selection & Performance

XGBoost:

Achieved an accuracy of over 80%, marking it as the best performing model for this application.

Comparative Analysis:

Other models like MLP, Decision Tree, and LogReg were also evaluated, but XGBoost outperformed them all.

Process Delta: Transitioning to Automation

Current System: Manual fault detection and classification lead to increased costs, reduced customer satisfaction, and a lack of actionable insights. Proposed System: An automated fault detection and prevention system powered by machine learning and cloud infrastructure. This includes: Automatic fault detection replacing manual labeling. AI-powered optical recognition of steel plates. Dynamic feedback loops providing actionable insights into the production process, facilitating preventive maintenance and data-driven decision-making.

About

The objective of the project is to classify steel plates fault into 7 different types. The end goal is to train several machine Learning Algorithms for automatic pattern recognition.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published