INCREASING ION SELECTIVE ELECTRODES PERFORMANCE USING NEURAL NETWORKS

O Postolache,Helena Ramos,M Pereira,P Girao

doi:10.47839/ijc.2.1.158

Abstract

This paper reports the implementation of a neural processing structure as a component of an intelligent measuring system that uses ion selective electrodes (ISEs) as sensing elements of heavy metal ions (Pb+2, Cd+2) concentration. The neural network (NN), designed and implemented to reduce errors due to ion interference and to pH and temperature variations, is of the multiple-input multiple-output Multilayer Percepton (MLP-NN) type. The NN is a component of a virtual instrument that includes a PC laptop, a PCMCI data acquisition board with associated conditioning circuits and the specific ISE sensors. A practical approach concerning the optimal neural processing solution (number of NN structures, number of neurons, neuron transfer functions) to increase the performance of low cost ISEs is presented. Results are presented to evaluate the performance of the NN intelligent ISE system and to discuss the possibility of transferring the acquisition and processing task to a low cost acquisition and control unit such as a microcontroller.

Highlights

Ions of heavy metals, such as lead and cadmium, are an undesired presence in river waters
Weights and biases of neural network (NN)-T, NN pH sensor-processing block (NN-pH) and NN-ion selective electrodes (ISEs) stored in files are directly accessed according to the acquired voltage on the T, pH, ISE_Cd and ISE_Pb channels and used as factors and terms involved in matrices multiplication that lead to CCd+2 and CPb+2 values
In order to evaluate the advantage of the NN utilization in the presented application, errors with NN processing (NN T, NN-pH and NN-ISE) are compared with errors without correction

Summary

INTRODUCTION

Ions of heavy metals, such as lead and cadmium, are an undesired presence in river waters. The monitoring of its concentration requires the utilization of measuring systems often based on ion selective sensors (ISE). Glass electrodes ISEs are characterized by fragility, temperature dependent sensitivity and limited selectivity [1] All these limitations make their use unsuitable for permanent field measurements required for continuous river water quality monitoring. Other groups of ISEs are solid-state crystal membrane sensors, such as NASICON, and PVC membranes, such as ELIT, which detect the ions (i.e. Pb+2,Cd+2, Hg+2) in aqueous solutions [2][3]. These types of ISEs are robust enough to permit its field utilization. Solutions in the multivariable modeling area use both the fuzzy system [4] and neural networks [5][6] to perform the modeling of sensor with non-linear characteristics and multiple influence factors

ISE SYSTEM

NN temperature sensor processing block

NN pH sensor processing block

RESULTS AND DISCUSSION

CONCLUSION