Artificial Neural Network Modeling of NixMnxOx based Thermistor for Predicative Synthesis and Characterization

Author(s) T.D. Dongale1 , K.G. Kharade2, N.B. Mullani1, G.M. Naik3,  R.K. Kamat4,

1 Computational Electronics and Nanoscience Research Laboratory, School of Nanoscience and Biotechnology, Shivaji University, Kolhapur, 416004 India

2 Department of Computer Science, Shivaji University, Kolhapur, 416004 India

3 Department of Electronics, Goa University, Goa, 403 206 India

4 Department of Electronics, Shivaji University, Kolhapur, 416004 India

Issue Volume 9, Year 2017, Number 3
Dates Received 02 March 2017; revised manuscript received 10 May 2017; published online 30 June 2017
Citation T.D. Dongale, K.G. Kharade, N.B. Mullani, et al., J. Nano- Electron. Phys. 9 No 3, 03042 (2017)
DOI 10.21272/jnep.9(3).03042
PACS Number(s) 07.05.Mh, 84.32.Ff, 07.05.Tp
Key words ANN (51) , Thermistor, Soft computing, Modeling (17) .
Annotation As foremost sensors of ambient conditions, temperature sensors are regarded as the most vital ones in wide-ranging applications touching the societal life. Amongst the temperature sensors, NTC thermistors have captured their unique place due to the favorable metrics such as highest sensitivity, low cost, and ease of deployment. Transition metal oxides especially the NixMnxOx are widely used for thermistor synthesis in spite of the main difficulty of predicting the final sensor characteristics before the actual synthesis. In view of the above, we report an Artificial Neural Network (ANN) technique to accomplish the synthesis with predictable results saving valuable resources. In the said ANN modeling we use hyperbolic tangent sigmoid transfer function for input layer and linear transfer function for the output layer. Levenberg-Marquardt feed-forward algorithm trains the neural net. We measure the performance of the ANN model with regard to mean square error (MSE) and the correlation coefficient between expected output and output provided by the network. Moreover, we uniquely model the resistance-temperature (R-T) characteristics of different thermistor samples using optimized ANN structure. To model such sort of behavior, we provide nickel content, room temperature resistance, and concentration of oxalic acid as an input data to the network and predict the nickel acetate and manganese acetate concentration. The accomplished ANN modeling evidences a lower number of hidden neuron architecture exhibiting optimum performance as regards to prediction accuracy. The lower number of hidden neurons signifies a lesser amount of memory required for prediction of different chemical composition. Thus, we demonstrate exploitation of modeling, simulation and soft computational approaches for predicting the best suitable chemical composition and thus establish the synergy between the materials science and soft computing paradigm.