Graphite Flake Research Report

Automatic Classification of Graphite Flake in Cast Iron using Neural Network

Abstract

The traditional method to classify graphite flakes which gives different properties to the cast iron is based on subjective observation of microstructure image which is compared with reference charts from ISO-945. This method is slow, less accurate and involves large computation. To overcome these disadvantages, number of experiments has taken place to automate classification process. These involve extracting features like fractal dimension, roughness, auto-regression and shape of the flakes which involves large computation and take huge operating time.

In this paper, the method is explained to automate the classification of flakes using Artificial Neural

This process is called ‘Blob Extraction’ or ‘Region Detection’.

The labeling algorithm states that assign same label number to pixels which are adjacent to each other where adjacency of pixels is determined based on 4-connectivity or 8-connectivity rule.

Area of a flake is number of pixels with same label. Mathematically, it is described as follows:

Area(i)=∑▒〖p(i)〗

Where i=1 to n, number of flakes in a zone and p (i) is pixel with label i.

Inter particle distance: Inter particle distance is defined as the minimum distance between the two flakes under consideration. It is the minimum among the distance calculated between every boundary pixel of one flake to every boundary pixel of the other.

Inter particle distance(i,j)=min⁡[d_(m,n) (i,j)]

d_(m,n) (i,j)=sqrt( (x_m-x_n )^2+(y_m-y_n )^2)

Where m, n are the number of boundary points of flake i, j respectively, (x_m, y_m), (x_n,y_n) are x and y coordinates of boundary point of i, j flakes respectively.

Chord length: Chord length is the number of pixels between boundary points of the flake which is counted row or column wise. Here, we chose both minimum and maximum chord length of the flake to training the