Design and Implementation Accurate Three Distance Sensors Device Using Neural Network

Abstract

This paper suggested and created a proposed device for sensing an accurate distance from 5 to 400 cm with three sensors using the proposed on-line Feedforward Backpropagation Neural Network (FFBPN). Firstly, preparing a dataset (distances) with two distances utilizing Arduino and measured distance calculated from three sensors (2 ultrasonic and 1 IR). Secondly, target distances are calculated from the ground manually using a distance tape tool. Thirdly, interfacing Arduino and MATLAB using USB easily saved datasets from Arduino directly to MATLAB, then training, and testing data.  The proposed device divides reading distances into three distances, getting from three sensors to get accurate distances. The first reading was 5-10 cm from down ultrasonic sensor 3.  The second reading was 10-80 cm from middle IR sensor 2. The third reading was 80-400 cm from up ultrasonic sensor 1. The dataset was created and improved using the suggested FFBPN to create a very accurate distance gadget with three sensors. A suggested FFBPN consisted of 3 layers: The first input measured distance layer (from 3 sensors), the second hidden layer with ten neurons, and the third output layer distance. The numerical results showed that the proposed distance device was significantly accurate due to regression result R=1 using the proposed Neural Network (NN), which meanings that the device had 100% fitting. It had the best validation performance in 464 epochs, i.e., is 0.0036122. Furthermore, the proposed on-line FFBPN was fast because the training time equals three seconds.  This distance device was implemented for use in robotic and radar applications to detect objects accurately because this device successfully detects near and far objects.