A Comparison And Application Of Pattern Recognition Techniques In Materials Selection Using Dimensionless Numbers

A comparison and application of pattern recognition techniques in materials selection using dimensionless numbers
Joseph Werner Schmidt-Castañeda, Francisco Javier Sandoval and Dante Jorge Dorantes-Gonzalez Instituto Tecnológico y de Estudios Superiores de Monterrey - Campus Estado de México {schmidt, ddorante, fsandova}@itesm.mx

Abstract -The objective of this paper is to present a tool anda comparison of techniques that can help in the material selection process based on pattern recognition and using dimensionless parameters from Π theorem. It is known that the main disadvantage in materials selection is that a huge database is needed to make the selection and also to make clusters of materials with some characteristics, so we need an easier and automatic way to cluster materialswith respect to their physical and mechanical properties, and particularly for material machinability. This paper also shows a possible application related to material selection using this approach to complement the conceptual design of a part coding system using neural network techniques. * Index Terms - Materials selection, Π theorem, dimensionless numbers, physical and mechanical properties,neural networks, pattern recognition.

that artificial intelligence techniques [9] like pattern recognition can help a lot in problems where there is a missing mathematical model or information about the case of study, so in problems like machinability [10] and materials selection we can use these techniques working with dimensionless numbers to make clusters of materials with the samecharacteristics of machinability to have an enormous database containing materials and their properties [11]. This enables us to generate material working parameters form only a limited selection of physical and mechanical properties. 2. METHODOLOGY Pattern recognition was selected as an intelligent tool for classification and recognition of groups or clusters with similar characteristics of machinability.The algorithms studied for this job were K-means, Fisher´s, lambda classification, expectation maximization and Isodata algorithms. All of them use statistical tools to separate clusters of points from different groups in different ways, one of them will be selected for the present application [12-17]. Materials selection process was the issue which initiated this work [18]. There are many ways ofselecting a material, but the most used is the Ashby map [8], where material clusters are made considering two properties of the materials. The information defining these clusters is held within a large database (image 1). The aim of this thesis is to process this clustering without the need of a database and to include more properties in one representation. As we need to have a many properties inone graph and we want to have dimensionless points in the graph we need to use dimensionless numbers theory that helps when developing experimental techniques. Fluid Mechanics uses dimensional analysis which does not give a complete solution. The success of this analysis depends on the skill to define parameters that would be applied. If one of the variables is omitted, the result will beincomplete and incorrect conclusions will be generated.

I. INTRODUCTION Group technology is a technique and a philosophy that improves the efficiency in production by means of grouping a great variety of parts using information of their shape, dimensions and working process routing [1, 2]. The main requirement of group technology is to have a coding system and a part classification that describes partcharacteristics, as well as their geometrical form, material and working process routing to produce these parts with a code number, gathering this way the parts with similar codes in a specific manufacturing cell or group of machines [3-5]. Parts grouped into families are widely used in manufacturing in order to get profit from their similarities [6]. Grouping parts into families can make easier...