Ž . Applied Surface Science 161 2000 131–138 www.elsevier.nlrlocaterapsusc Neural networks in studies on oxidation behavior of laser surface engineered composite boride coatings Anuradha Godavarty a , Arvind Agarwal b , Narendra B. Dahotre b, ) a Department of Chemical Engineering, UniÕersity of Tennessee Space Institute, Tullahoma, TN 37388, USA b Department of Materials Science and Engineering, Center for Laser Applications, UniÕersity of Tennessee Space Institute, Tullahoma, TN 37388, USA Received 29 December 1999; accepted 19 February 2000 Abstract Neural computing has been used to determine the kinetics and mechanism of oxidation of composite titanium diboride Ž . TiB coating. The use of neural network software reduces the required extensive experimental study by learning the trend 2 Ž . of the mass gain curves and predicting the effect of variables temperature and time within its region of learning. The ‘‘Professional IIrPLUS version 5.23’’ neural networks software was trained from the initial experimental results obtained Ž . from thermogravimetric analysis TGA . The oxidation rate was found to be parabolic. There was no change in the oxidation mechanism of TiB coating in the temperature range of 6008C to 10008C in comparison to earlier minimal experimental 2 study. The activation energy for oxidation of composite TiB coating was 210 kJrmol. Application of neural networks in 2 Ž these oxidation studies can further be expanded to study the effect of other parameters such as the gaseous environment air . or pure oxygen and their flow rates. q 2000 Elsevier Science B.V. All rights reserved. Ž . Keywords: Neural networks; Oxidation; Composite TiB coating; Thermogravimetric analysis TGA 2 1. Introduction Neural computing provides an approach that is closer to human perception and recognition than conventional computing. The special characteristics of neural computing such as learning by example, distributed associative memory, fault tolerance, pat- tern recognition and synthesis make them useful in a w x variety of applications 1–3 . Neural computing has found its application in various fields like digital ) Corresponding author. Tel.: q 1-931-393-7495; fax: q 1-931- 454-2271. Ž . E-mail address: ndahotre@utsi.edu N.B. Dahotre . communication, environmental pollution analysis, sales forecasting, biomedical systems such as diag- nosing coronary artery disease, jet engine diagnostic system and detection in synthetic aperture radar im- w x ages 1–3 . However, neural networks have their own limita- tions. One of the limitations is that good results can be obtained only if the data to be tested is within the range of the trained data. Extrapolation of the test results beyond the trained data is likely to give errors. Secondly, there are always end effects in the trained and tested software, which cause increased wx error in the output data at the end points 1. To avoid the end effects in training the data, theoretical data points should be added to the input parameters. 0169-4332r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved. Ž . PII: S0169-4332 00 00279-8