Neural network-based ultrasonic level measurement of fluids

This thesis describes the practical implementation of neural network-based ultrasonic level measurement of fluids (in the range of cm to meters) and will be designed a neural network-based ultrasonic range finder sensor for this goal. By a neural network algorithm, this sensor can be advanced including temperature, humidity, and other parameters (such as pressure, CO2 in medium). A Multi Layer Perceptron (MLP) neural network (NN), and Back-Propagation Algorithm which is used as common for it?s learning is used in this thesis is . MLP NN is designed by Matlab Neural Network Toolbox. Necessary input-output data set for training and test of MLP NN is obtained using the approximate formula published in JASA [1993] by Owen Cramer Hardware interface uses an Atmel Atmega32 8-bit AVR microcontroller to facilitate the generation of 40 kHz signal burst which is used in the transmitter circuit, and also to process the received signal for measuring the time of flight of reflected waves and exact distance of the obstruction. Level of fluids is momentarily calculated by flight time and speed of reference sound. Temperature and Humidity Sensor (SHT11) provides temperature and humidity data to microcontroller. These data are the inputs of MLP. The outputs of MLP are values of corrected level of fluids. In this project, The program for this device is developed in MCS Electronic / BASCOM-AVR IDE.