The QuenchMiner™ Expert System for Quenching and

Distortion ControlAparna S. Varde, Mohammed Maniruzzaman,

Elke Rundensteiner and Richard D. Sisson Jr.

Worcester Polytechnic Institute,

Worcester, MA, USA

Introduction

• QuenchMiner™ : Started as Web-based tool for analyzing quenching experimental data

• Data Mining features added: discovering interesting patterns in data sets to guide decisions

• Enhanced into Expert System for Decision Support in Heat Treating

Logo summarizing QuenchMiner™ functions

What is an Expert System

• Computer program with knowledge of specialist

• Does representation and reasoning using knowledge

• Solves problems or gives advice to users

Structure of an Expert System

QuenchMiner™ for Decision Support• Tool to support or assist the users decisions in selecting

quenchants, parts, conditions to achieve a desired output for quenching in the industry. Analyses user cases.

• Goal: Estimate parameters of interest computationally– Desired Suspension of Part– Average Cooling Rate– Cooling Uniformity– Average Heat Transfer Coefficient– Residual Stress– Hardness– Distortion Tendency– Cracking Potential

• Given: Quenching experimental input– Quenchant Information– Part Conditions– Manufacturing Details

System Architecture• User interaction

through Web

• Data Mining to get knowledge from quenching data, build rules

• Decision Making using Artificial Intelligence techniques e.g. Rule Interpreters

Architecture of QuenchMiner™ Expert System

QuenchMiner™ Demo

• Demo available at,

http://mpis.wpi.edu:9006/database http://mpis.wpi.edu:9006/database

• Screendumps of demo shown here:

– Menu Screen

– Case Input Screen

– Case Output Screen

Methodology for Analysis• Identify causes (variables) of effects (parameters of interest)

• Represent causal relationship by Fishbone (Ishikawa) diagrams

Fishbone diagram for Cooling Rate

Decision Making

Application to Distortion Control

• Distortion: One of the biggest problems in quenching. Part gets deformed in shape and / or size while being quenched.

• QuenchMiner™ – estimates tendency for distortion in part given input conditions

(computationally, without performing experiment)

– identifies causes of distortion in given case by analysis

– Thus assists users decisions to select conditions to minimize distortion

• Example: Shown in Screen-dump

Conclusions• QuenchMiner™ developed as an Expert System for

Decision Support in Heat Treating

• Analyzes cases in Quenching and Distortion Control

• Uses Data Mining and Artificial Intelligence techniques

• Only supports or assists users decisions

• Demo available, System Ready for use

• Ongoing research and user feedback will provide further improvements to QuenchMiner™.