Multibeam Optical System and Neural Processing for Turbidity Measurement

Abstract

This paper presents a turbidity measuring system based on a modulated four infrared (IR) light beam architecture with advanced data processing. The turbidity sensing component consists of a pair of IR light-emitting diodes (LEDs) connected to a current drive controlled through the pulsewidth modulated (PWM) outputs of a multifunction input/output board. The scattered and transmitted IR light in the media under test is detected by a two-channel IR photodiode module that includes a set of transimpedance and programmable gain amplifier. The voltages proportional to the detectors' output currents, are acquired using a 12-bit ADC included in a microcontroller and RS232 transmitted to a laptop personal computer (PC) that works as an advanced control and processing unit. Using optimal neural network processing architectures, an accurate extraction of the turbidity information is performed. A practical approach concerning the neural network architectures [multilayer perceptron single-input-single-output (SISO), multiple-input-single-output (MISO)] including neural network training and testing is discussed in the paper. The multi-input architectures prove to be a robust and general solution for the proposed application. Results from a turbidity measuring system that was designed for automated standalone remote operation with sensing channel autocalibration capabilities are presented