Forecasting the Dielectric Properties of the Insulation Material using an Embedded ANN Microcontroller

This paper describes a laboratory application for the forecasting the dielectric properties, using an artificial neural network embedded in a microprocessor. The dielectric permittivity and loss factor of the insulation material were determined as a function of temperature and frequency inputs by an implemented ANN model. For this purpose, a three-layer feed-forward back-propagation ANN model with two inputs and two outputs was created. The Levenberg-Marquardt back-propagation algorithm was selected as a training function, the gradient descent method with momentum weighting and bias learning function as an adaptive learning function, and mean squared normalized error as a performance function. The first layer of the network consists of three neurons, the second layer of six neurons with a tangent sigmoid activation function, and the outputlayer of two outputs with a linear transfer function. The initial values of weights and biases were selected randomly. Input and layer of two outputs with a linear transfer function. The initial values of weights and biases were selected randomly. Input and output data were rescaled to have the minimum possible determination errors. The microprocessor-based method is a low-cost determination system independent of computers and can be used in both laboratory and industrial applications.