Browsing by Author "Tuntas, Remzi"

Now showing 1 - 8 of 8

An Anfis Model To Prediction of Corrosion Resistance of Coated Implant Materials
(Springer, 2017) Tuntas, Remzi; Dikici, Burak
In the present study, an adaptive neuro-fuzzy inference system (ANFIS) model has been used for predicting the corrosion resistance of AA6061-T4 alloy coated with micro-/nano-hydroxyapatite (HA) powders by sol-gel technique. The input parameters of the model consist of the HA powder size (micro-/nanoscale, 35 mu m/20 nm), coating thickness (30, 60 and 85 mu m) and potential values, while the output parameter is corrosion current density. The performance of proposed ANFIS model was tested on the potentiodynamic polarization scanning (PDS) curves by comparing experimental and the theoretical results of the coatings. The results showed that the generated PDS curves of the coatings are in definitely acceptable levels with obtained results in our experimental reference study. Then, the combined effect of arbitrary selected coating thickness and HA powder size on corrosion behaviour of the coatings was also predicted by trained ANFIS model without using any experimental data. And finally, the predicted results for the arbitrary selected coating thicknesses were compared by validation tests. The results showed that the ANFIS has potential to be used in industrial applications of biomedical implant materials coated with HA without performing any experiments after detailed systematic studies in the near future.
An Artificial Neural Network (Ann) Solution To the Prediction of Age-Hardening and Corrosion Behavior of an Al/Tic Functional Gradient Material (Fgm)
(Sage Publications Ltd, 2021) Dikici, Burak; Tuntas, Remzi
In this theoretical study, the prediction of the corrosion resistance and age-hardening behavior of an Al/TiC functional gradient material (FGM) has been investigated by using the artificial neural network (ANN). The input parameters have been selected as TiC volume fraction of the composite layers, aging periods of the composite, environmental conditions, and applied potential during the corrosion tests. Current and microhardness were used as the one output in the proposed network. Also, a new three-layered composite has been imaginarily designed to demonstrate the predictive capability and flexibilities of the ANN model as a case study. Artificially aging (T6) process and potentiodynamic scanning (PDS) tests were used for heat-treating and corrosion response of the FGS, respectively. The results showed that the generated PDS curves of the FGM and calculated corrosion parameters of the case study are quite near and in acceptable limits for similar composites obtained values in experimental studies. Besides, this study has been a great success in predicting peak-aging times and its corresponding hardness values more precisely.
An Investigation on the Aging Responses and Corrosion Behaviour of A356/Sic Composites by Neural Network: the Effect of Cold Working Ratio
(Sage Publications Ltd, 2016) Tuntas, Remzi; Dikici, Burak
In the present study, an artificial neural network model has been used for predicting the corrosion behaviour, aging and hardness responses of aluminium-based metal matrix composites reinforced with silicon carbide particle. Hyperbolic tangent sigmoid and linear activation functions are employed as the most appropriate activation function for hidden and output layers, respectively. The developed artificial neural network model is used to predict the corrosion current density, peak aging time and peak hardness of the composites. Feed forward back propagation neural network has been trained by Levenberg Marquardt algorithm. The regression correlation coefficients (R-2) between the predicted and the experimental values of the corrosion current densities are found as 0.99986, 0.99629 and 0.99671 for the training, testing and validation datasets, respectively. Also, some case studies have been predicted by artificial neural network model. Test results indicate that the proposed network can be used efficiently for the prediction of the polarization response, peak aging time and peak hardness of the composites for different SiC volume fractions and deformation ratio without using any experimental data.
A Neural Network Based Controller Design for Temperature Control in Heat Exchanger
(Natl inst Science Communication-niscair, 2019) Tuntas, Remzi
The temperature of outlet fluid of the heat exchanger system is controlled with Artificial Neural Network Based Model Reference (ANNBMR) control method. The ANNBMR controller designed arranges the temperature of the outlet fluid to reach a desired reference value in the shortest possible time, despite the flow and temperature variation of input fluid. The simulation of ANNBMR controller designed with the use of the plant model of the heat exchanger system is carried out in MATLAB Simulink and simulation results are obtained. The simulations results of the proposed ANNBMR controller is compared with the simulations results of the PID conventional controller. The proposed controller shows better performance and reduces both settling time and maximum overshoot and by shortening the error signal between the input and output in a shorter period of time. The obtained simulation results prove that the proposed control method is quite successful.
A New Approach for the Analysis of the Transient and Steady State of Piecewise-Linear Circuits by Neural Networks
(World Scientific Publ Co Pte Ltd, 2008) Tuntas, Remzi; Demir, Yakup; Koksal, Muhammet
This paper presents a new approach based on Feed-Forward multilayered Neural Networks (FNNs) for the transient and steady-state analysis of piecewise-linear circuits. Nonlinear circuits are changed by linear circuits containing switches together with some state matrices and control inequalities by using piecewise linearization approach. One of the two problems arising in the analysis of these circuits is that control inequalities belonging to piecewise linearized components and control times for internally or externally controlled components is needed to determine switching times. Another is that the analysis time is very long. The proposed approach is considered as the solution to the problems. FNNs are used for modeling the piecewise linear circuits. By using the obtained model networks, the switching sequence and switching times from one state to another for transient and steady states are determined. The transient and steady-state solutions are fast accomplished through this knowledge. As an example, a nonlinear circuit is used for demonstrating the utility of the proposed approach, and the results are compared with that of the model constructed at Matlab/Simulink. Example circuit is analyzed in the time of 1 h 42min by using the proposed approach but the time of 2 h 27 min by Matlab/Simulink. Moreover, simulation results demonstrate that the proposed approach yields more accurate approximation for the switched nonlinear circuits.
A New Intelligent Hardware Implementation Based on Field Programmable Gate Array for Chaotic Systems
(Elsevier Science Bv, 2015) Tuntas, Remzi
In the present study, a new intelligent hardware implementation was developed for chaotic systems by using field programmable gate array (FPGA). The success and superior properties of this new intelligent hardware implementation was shown by applying the Modified Van der Pol-Duffing Oscillator Circuit (MVPDOC). The validation of intelligent system model was tested with both software and hardware. For this purpose, initially the intelligent system model of MVPDOC was obtained by using the wavelet decompositions and Artificial Neural Network (ANN). Then, the intelligent system model obtained has been written in Very High Speed Integrated Circuit Hardware Description Language (VHDL). In the next step, these configurations have been simulated and tested under ModelSim Xilinx software. And finally the best configuration has been implemented under the Xilinx Virtex-II Pro FPGA (XC2V1000). Furthermore, the High Personal Simulation Program with Integrated Circuit Emphasis (HSPICE) simulation of MVPDOC has been carried out under ModelSim Xilinx software for comparison with proposed intelligent system. The results obtained show that the proposed intelligent system simulation has much higher speed in comparison with HSPICE simulation. (C) 2015 Elsevier B.V. All rights reserved.
Prediction of Corrosion Susceptibilities of Al-Based Metal Matrix Composites Reinforced With Sic Particles Using Artificial Neural Network
(Sage Publications Ltd, 2015) Tuntas, Remzi; Dikici, Burak
In this theoretical study, the prediction of the corrosion resistance of Al-Si-Mg-based metal matrix composites reinforced with SiC particles has been studied, using an artificial neural network. Four input vectors were used in the construction of the proposed network; namely, volume fraction of SiC reinforcement, aging time of the composites, environmental conditions, and potential. Current was used as the one output in the proposed network. Test results indicate that the proposed network can be used efficiently for the prediction of the corrosion resistance of Al-Si-Mg-based metal matrix composites reinforced with SiC particles, and the methodology is suitable for engineers to study the corrosion of metal matrix composites. In addition, a few forecasts regarding the polarization response for different SiC volume fractions and aging conditions have also been generated without using any experimental data.
Yapay Sinir Ağları Yardımı İle Çağrı Merkezi Uygulamalarında Öngörü Modellemesi
(2019) Ortakaya, Sefa; Tuntas, Remzi
Yapay zekâ teknolojilerinden biri olan Yapay Sinir Ağları (YSA) tahmin, modelleme,sınıflandırma ve bunun gibi birçok sosyal ve mühendislik alanlarında yaygın olarakkullanılmaktadır. YSA'nın paralel yapısı, gerçek zamanlı uygulamalar için önemli bir özellikolup en önemli avantajları modelin esnek ve uyumlu doğasıdır. YSA'lar bir defa eğitimdengeçirildikten sonra yeniden programlamaya gerek kalmadan herhangi bir uygulama içinsorunsuz bir şekilde kullanılabilirler. YSA'lar, uygun öğrenme yöntemini kullanarak girdi veçıktı kalıpları arasındaki doğrusal ve doğrusal olmayan ilişkileri belirlerler. Başka birdeyişle, karmaşık ve doğrusal olmayan sistemlerin giriş ve çıkışları arasındaki korelasyonukullanarak sistemleri modelleme yeteneklerine sahiptirler. Bu çalışmada YSA yöntemi ileçağrı merkezi verilerine yönelik gelmesi beklenen çağrı sayıları ile yapılması beklenengörüşme süreleri tahmin edilmiştir. Çağrı merkezleri kurumlarınmüşterilerinden/vatandaşlarından gelen talep, görüş, öneri, memnuniyetsizlik, şikâyet vb.konularda hizmet verdiği yüksek öneme sahip iletişim birimidir. Çağrı merkezi yöneticilerierken karar almada tahmin yapmak durumuyla karşı karşıyadırlar. Bu nedenle çağrı merkezisistemlerinde yer alan raporlamalarda günlük, haftalık ve aylık periyotlarda gelen çağrısayıları ile karşılanan çağrıların görüşme sürelerinin tahmini önem arz etmektedir. Yapılanbu çalışmada, eğitim verileri olarak daha önceki aylara ait görüşme sayıları ve görüşmesüreleri kullanılmıştır. Öngörü modellemesi için girişten çıkışa doğru ileri beslemeli YSAmodeli elde edilmiş ve Levenberg-Marquardt algoritması ile ağ modeli eğitilmiştir. Giriş,gizli ve çıkış katmanından oluşan bu üç katmanlı YSA'nın gizli ve çıktı katmanları için lineeraktivasyon fonksiyonları kullanılmıştır. Bu çalışmada kullanılan ileri beslemeli ve geriyayılımlı YSA modeli ile gelen çağrı sayıları ve bu çağrıların görüşme süreleri tahmin edilmişve elde edilen bu YSA modelinin öngörü performansı ortaya konarak bu modelin güvenilirve tutarlı olduğu gözlemlenmiştir.