Record Details

Comparison of Some Neural Network and Multivariate Regression for Predicting Mechanical Properties of Investment Casting

DSpace at IIT Bombay

View Archive Info
 
 
Field Value
 
Title Comparison of Some Neural Network and Multivariate Regression for Predicting Mechanical Properties of Investment Casting
 
Creator SATA, A
RAVI, B
 
Subject artificial neural network
investment casting
mechanical properties
multivariate regression
prediction penalty index
COOLING RATE
A356 ALLOY
IRON
MICROSTRUCTURE
CLASSIFICATION
BEHAVIOR
QUALITY
MODELS
 
Description Investment casting enables producing complex shapes with good accuracy and surface finish. A key goal for investment castings used in automobile, aerospace, chemical, biomedical and other critical applications is to be free of internal defects and to possess mechanical properties within the desired range. At present, casting quality is ascertained by destructive testing at the end of production cycle, leading to the possibility of scrapping the entire batch. In this work, the mechanical properties of investment castings have been predicted based on process parameters and chemical composition, by employing artificial neural network (ANN) and multivariate regression (MVR). The data of related process parameters (wax making, shell making, dewaxing, melting etc.), chemical composition of the alloy, and the resulting mechanical properties (ultimate tensile strength, yield strength, and percentage elongation) for 800 heats were collected in an industrial investment casting foundry. Three different ANN models: back propagation, momentum and adaptive, and Levenberg-Marquardt, with varying number of neurons in the hidden layer (from 20 to 45 in steps of 5) were trained using a portion of the data and tested with remaining data. A prediction penalty index (PPI) was developed to compare the relative predictive capability of various neural network and MVR models. It is observed that both ANN and MVR could predict the mechanical properties well, though MVR gave slightly better results. For the ANN model, better results were produced when the number of neurons in the hidden layer was equal or slightly higher than the number of input parameters.
 
Publisher SPRINGER
 
Date 2014-12-28T13:10:01Z
2014-12-28T13:10:01Z
2014
 
Type Article
 
Identifier JOURNAL OF MATERIALS ENGINEERING AND PERFORMANCE, 23(8)2953-2964
1059-9495
1544-1024
http://dx.doi.org/10.1007/s11665-014-1029-1
http://dspace.library.iitb.ac.in/jspui/handle/100/16676
 
Language English