Prediction of density, porosity and hardness in aluminum–copper-based composite materials using artificial neural network
Introduction
The term “composite” broadly refers to a material system which is composed of a discrete constituents (the reinforcements) distributed in a continuous phase (the matrix) which derives its distinguishing characteristics from the properties of its constituents, from the geometry and architecture of the constituents, and from the properties of the boundaries (interfaces) between different constituents. Composite materials are usually classified on the basis of the physical or chemical nature of the matrix phase, e.g., polymer matrix, metal–matrix and ceramic matrix composites.
Aluminum matrix composites (AMCs) refer to the class of light-weight high performance aluminum centric material systems. The reinforcement in AMCs could be in the form of continuous/discontinuous fibers, whisker or particulates, in volume fractions ranging from a few percent to 70% (Surappa, 2003). Candan and Bilgic (2004) produced Al–60 vol.%SiCP composites by pressure infiltration technique and investigated their corrosion behavior, in general they found that Al-based composites are weaker in 3.5% NaCl solution compared to Al–4 wt.%Mg alloy Al-based composites are usually reinforced by Al2O3, SiC, and C. In addition SiO2, B, BN, B4C may also be considered. Properties of AMCs can be tailored to the demands of different industrial applications by suitable combinations of matrix, reinforcement and processing route. In the last few years, AMCs have been utilized in high-tech structural and functional applications including aerospace, defense, automotive, and thermal management areas, as well as in sports and recreation.
The major advantages of AMCs compared to unreinforced materials are mentioned in most of articles dealt with aluminium-based composite materials (e.g. Szezepanik and Sleboda, 1996, Smith and Hashemi, 2006, Chaudhurky et al., 2004) which can be summarized as follows: greater strength, improved stiffness, reduced density, good corrosion resistance, improved high temperature properties, controlled thermal expansion coefficient, thermal/heat management, enhanced and tailored electrical performance, improved wear resistance and improved damping capabilities.
Among the various methods to fabricate metal matrix composites, stir casting method has drawn keen attraction among the researchers due to its industrial feasibility. The major limitation in many cases in fabricating metal matrix composites by liquid phase route resides upon the incompatibility of the reinforcement and the matrix (Candan and Bilgic, 2004, Hassan et al., 2007). This problem in case of Al-based metal matrix composite is due to the formation of stable tenacious oxide film, resulting in poor wettability with the ceramic particle. One of the common practices to improve wettability of an Al melt is through addition of small amount of reactive metals like magnesium and titanium prior to the incorporation of ceramic particle. In the present work, 4 wt.%Mg was added to Al to improve wettability as recommended by Candan and Bilgic (2004); Hassan et al. (2007) and argon flux was used during melting and pouring tasks to reduce oxidation effect.
The use of artificial neural networks (ANNs) represents a new methodology in many different applications of composite materials including prediction of mechanical properties (e.g. Altinkok and Koker, 2006, Durmus et al., 2006, Lee et al., 1999, Mukherjee et al., 1995, Zhang et al., 2002, Zhang et al., 2003). It is a promising field of research in predicting experimental trends and has become increasingly popular in the last few years as they can often solve problems much faster compared to other approaches with the additional ability to learn from small experimental data. ANN was used to predict wear loss and surface roughness of AA 6351 aluminum alloy by Durmus et al. (2006). In their study, experimental and ANNs results have been compared and they showed coincidence to a large extent. The use of ANN for prediction of physical properties and tensile strengths in particle reinforced aluminum matrix composites showed satisfactory results if ANN used as prediction technique (Altinkok and Koker, 2005, Altinkok and Koker, 2006). Vassilopoulos et al. (2007) used ANN in spectrum fatigue life prediction of composite materials. In their study they found that the main benefit of this tool is that only a small portion, in the range of 40–50%, of the experimental data is needed for the whole analysis. Thus, expensive and time consuming tests required by the conventional way for the establishment of S–N curves could be significantly reduced without significant loss of accuracy.
Section snippets
Materials
The test materials studied in this work were a mixture of aluminium (commercial grade Al, ∼99% purity) and copper (∼97% purity and 40 mesh sizes) as a matrix and silicon carbide as reinforcements. About 1000 g of commercial grade Al and different weight percentages of copper powder (0, 1, 2, 3, 4, and 5 wt.%) were taken to prepare the composite by compocasting. Magnesium (∼99% purity ingots) added in small quantities (fixed weight percentage 4 wt.%) to promote wettability between metal matrix and
Modeling with neural networks
Artificial neural networks are considered as artificial intelligence modeling techniques. They have a highly interconnected structure similar to brain cells of human neural networks and consist of the large number of simple processing elements called neurons, which are arranged in different layers in the network. Each network consists of an input layer, an output layer and one or more hidden layers. One of the well-known advantages of ANN is that the ANN has the ability to learn from the sample
Training and validating of ANN
The ANN was trained and implemented using fully developed feed forward backpropagation neural network. For the training problem at hand the following parameters were found to give good performance and rapid convergence: two input nodes; namely Cu (wt.%) and SiC (vol.%), one hidden layer with 10 neurons and 3 output neurons which are hardness, density and porosity. Sigmoid activation function was selected to be the transfer function between all layers. The ANN architecture is shown in Fig. 1.
A
Conclusions
The use of ANN in prediction hardness and some physical properties for aluminium–copper-based composite materials has been studied in this investigation. The ANN gives satisfactory results when compared to the experimental measurements. The maximum absolute relative error for predicted values does not exceed 5.99%. Therefore, by using ANN values, satisfactory results can be estimated rather than measured and hence reduce testing time and cost.
Among other training parameters, it was found that
Acknowledgements
The authors gratefully acknowledge the assistance of the committee of scientific research/Jordan University of Science and Technology for its support of this research (grant No. 29/2007). The authors would like also to gratefully acknowledge the use of Machine shop and the laboratory facilities at Jordan University of Science and Technology, Irbid, Jordan.
References (22)
- et al.
Modeling of the prediction of tensile and density properties in particle reinforced metal matrix composites by using neural networks
Mater. Des.
(2006) - et al.
Corrosion behavior of Al–60 vol.%SiCP composites in NaCl
Mater. Lett.
(2004) - et al.
The use of neural networks for the prediction of wear loss and surface roughness of AA 6351 aluminum alloy
Mater. Des.
(2006) - et al.
Artificial neural networks for the prediction of mechanical behavior of metal matrix composites
Acta Met. Mater.
(1995) - et al.
Aging response of aluminum alloy 2024/silicon carbide particles (SiCP) composites
J. Mater. Sci. Eng. A
(2004) - et al.
The use of neural networks for the prediction of fatigue lives of composite materials
Compos.: Part A
(1999) - et al.
Prediction on tribological properties of short fiber composites using artificial neural networks
Wear
(2002) - et al.
Artificial neural network predictions on erosive wear of polymers
Wear
(2003) - et al.
Artificial neural networks in spectrum fatigue life prediction of composite materials
Int. J. Fatigue
(2007) Application of artificial neural networks to predict the carbon content and the grain size for carbon steel
Egypt J. Solids
(2002)
Effect of casting temperature on the microstructure and wear resistance of Compocast A356/SiCp composites: a comparison between SS and SL routes
J. Mater. Process. Technol.
Cited by (155)
Smart-Cast: An AI-Based System for Semisolid Casting Process Control
2024, Procedia Computer SciencePrediction of elastic wave propagation in composite bars using deep learning techniques
2023, Materials LettersThe prediction of MOE of bamboo-wood composites by ANN models based on the non-destructive vibration testing
2022, Journal of Building EngineeringElectrochemical and mechanical characteristics of Al-<inf>(x)</inf>SiC composite fabricated by high energy compaction
2022, Journal of the Indian Chemical SocietyNeural networks and correlation analysis to improve the corrosion prediction of SiO<inf>2</inf>-nanostructured patinated bronze in marine atmospheres
2022, Journal of Electroanalytical ChemistryEffect of interfacial bridging atoms on the strength of Al/CNT composites: machine-learning-based prediction and experimental validation
2022, Journal of Materials Research and Technology