Application of Neural Network in Simple Tool Wear Monitoring and Indentification System in MDF Milling

Marcin Zbieć

doi:10.5552/drind.2011.1020

Abstract

This paper deals with simple neural network-based diagnostic system, applied to tool wear prediction in MDF milling. Ten tools were used for the test, and each one was consequently worn in the process of MDF milling. During the wearing process, the key process parameters were measured, such as cutting and thrust forces, temperature and power consumption. The neural network-based system was used for tool wear prediction of all the tools except the fi rst one, based on data collected during the previous attempts. The test has shown that the proposed system has good prediction accuracy and that it could be a useful tool in the optimization of the woodworking process.

Highlights

Real-time tool wear and breakage monitoring are based on various measurable factors related to the cutting process, such as cutting and normal forces, machined surface temperature, work power demand and feed mechanisms and acoustic emission related to cutting and material fracture
Tools used for MDF milling were marked with M1-M10 symbols
Accuracy ranging from 3 to 10% means that the diagnostic system is capable of identifying the condition of tool wear with astounding precision, in the order of several micrometers

Summary

Introduction

Real-time diagnostic of tool wear should allow more control of the whole machining process by eliminating the production spoilage caused by worn or catastrophically crashed tools. Real-time tool wear and breakage monitoring are based on various measurable factors related to the cutting process, such as cutting and normal forces, machined surface temperature, work power demand and feed mechanisms and acoustic emission related to cutting and material fracture. AI system based on the neural network seems to be the simplest working solution for tool wear prediction. Artificial neural networks allow rejection of regular, time consuming statistical and mathematical analysis. Trend prediction and generalization properties can replace an experienced machine operator and complex analysis made with traditional methods (Gawlik, 1997)

Objectives

Methods

Results

Conclusion