Materials Science and Engineering A363 (2003) 203–210 Modeling of tribological properties of alumina fiber reinforced zinc–aluminum composites using artificial neural network K. Genel a,∗ , S.C. Kurnaz b , M. Durman b a Mechanical Engineering Department, Engineering Faculty, Sakarya University, Esentepe, Sakarya 54187, Turkey b Metallurgical and Materials Engineering Department, Engineering Faculty, Sakarya University, Esentepe, Sakarya 54187, Turkey Received 10 March 2003; received in revised form 7 July 2003 Abstract It is known that the strength of a metal, as well as wear resistance can be successfully improved by fiber reinforcement. In this study, multiple-layer feed-forward artificial neural network (ANN) modeling for tribological behavior of short alumina fiber reinforced zinc–aluminum composites has been established. The specific wear rate and coefficient of friction obtained from a series of the wear tests were used in the formation of training sets of ANN. Samples of composite material with 10, 15, 20 and 30 vol.% fiber contents were prepared by the pressure die-casting method. Wear tests with pin-on-disc arrangement had performed at a constant sliding speed of 1 m/s under four different loads (5, 10, 20 and 40 N). The results of experimental tests showed that wear behavior and friction coefficient of the composites were significantly affected by the fiber volume fraction. The specific wear rate decreased with increasing fiber volume fraction and increased with increasing load. The coefficients of friction of the composites were higher than that of the unreinforced matrix alloy. The modeling results confirm the feasibility of the ANN and its good correlation with the experimental results. The degrees of accuracy of the prediction were 94.2 and 99.4% for specific wear rate and friction coefficient, respectively. It is concluded that ANN is an excellent analytical tool if it is well trained. This means considerable cost and time saving. Finally, using ANN modeling data and experimental data, 3D plots and empirical expressions for specific wear rate and friction coefficient related to load and fiber volume fraction were established. © 2003 Elsevier B.V. All rights reserved. Keywords: Die-casting; Composite; Short fiber; Wear; Artificial neural network 1. Introduction A new family of zinc–aluminium alloys (ZA8, ZA12 and ZA27) has been showing a successful trend in replacing cast Fe, Al and Cu alloys in various technological applications due to their superior mechanical and tribological properties and better castability with a wide range of casting processes [1]. The ZA27 alloy is known for its high strength/density ratio and good wear strength because of the higher aluminum content. Moreover, strength, modulus of elasticity, creep and wear resistance is enhanced by fiber reinforcement to form composite materials [2]. Many studies showed that the addi- tion of short saffil fiber (-Al 2 O 3 ) to zinc–aluminum alloys, especially ZA27, increased the wear resistance and coeffi- cient of friction [3–5]. It is known that wear resistance of composite material is mainly related to fiber volume frac- ∗ Corresponding author. Tel.: +90-264-346-0353x319; fax: +90-264-346-0351. E-mail address: kgenel@sakarya.edu.tr (K. Genel). tion, morphology, strength of interface and fiber orientation. Moreover, load, speed of sliding and humidity also alter wear performance of material. Accordingly interpretation of wear behavior requires a lot of experimental data obtained from different series of test, which consumes extensive time and money. In this study, two artificial neural networks are pro- posed as the framework for developing a model that pre- dicts the tribological properties of alumina fiber reinforced zinc–aluminum alloy matrix composites based on the load, fiber volume fraction and fiber orientation. Artificial neu- ral network (ANN) modeling is inspired by the biological nerve system and is being used to solve a wide variety of complex scientific and engineering problems. This math- ematical technique is especially useful for simulations of any correlation that is difficult to describe with physical models because of the ability to learn by example and to recognize patterns in a series of input and output values from example cases. The interest in ANN modeling in the fields of materials science and physical metallurgy has 0921-5093/$ – see front matter © 2003 Elsevier B.V. All rights reserved. doi:10.1016/S0921-5093(03)00623-3