Analysing wear behaviour of Al–CaCO3 composites using ANN and Sugeno-type fuzzy inference systems

• Published in: Neural computing & applications Vol. 32; no. 17; pp. 13453 - 13464
• Main Authors: Sosimi, A. A., Gbenebor, O. P., Oyerinde, O., Bakare, O. O., Adeosun, S. O., Olaleye, S. A.
• Format: Journal Article
• Language: English
• Published: London Springer London 01.09.2020; Springer Nature B.V
• Subjects: Algorithms; Aluminum; Artificial Intelligence; Artificial neural networks; Back propagation networks; Calcium carbonate; Composite materials; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Design of experiments; Fuzzy systems; Image Processing and Computer Vision; Inference; Mathematical models; Mechanical properties; Neural networks; Original Article; Probability and Statistics in Computer Science; Process controls; Process parameters; Systems design; Ultrasonic testing; Wear rate; Wear tests; ANN; Stir casting; Sugeno-type FIS; composite; Wear; Al–CaCO
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-020-04753-6

Design of experiment for the development of stir cast calcium carbonate-reinforced aluminium composite is a search for optimum combination of material and process control parameters for best physical and mechanical properties. A soft-computing model can accurately learn the complex interactions between process parameters to provide great insights in the development of this composite. This paper demonstrates and analyses the potential of artificial neural network (ANN) and Sugeno-type fuzzy inference systems (FIS) for wear behaviour prediction of calcium carbonate-reinforced aluminium composites. The models were trained with data collected from the experiment. The data consist of filler particle size of 150 μm with weights fractions varied from 0 to 25 wt%, in step of 5. Wear test data at different time of contacts (30, 60, 90, 120 and 150 s) and variable loads of 2.27 N, 4.54 N and 6.80 N were collected, resulting to 120 length vectors. Comparing the experimental results of wear test with those predicted using the ANN and Sugeno-type FIS, the integration of calcium carbonate particulate enhanced the wear characteristics of Al matrix up to 200%. On the use of back-propagation neural network with 4–3–1 architecture for wear prediction, the Levenberg–Marquardt training algorithm performs better. For Sugeno-type FIS, the Gaussian membership function resulted to the best prediction of wear rate. When ANN and Sugeno-type FIS performance on the test set were analysed based on some statistical parameters, the later returned an R 2 value of 0.9775 as against ANN’s value of 0.3684. The predicted wear rate using ANFIS with Gaussian membership functions was in good agreement with the experimental values.