A comparison study of pattern recognition algorithms implemented on a microcontroller for use in an electronic tongue for monitoring drinking waters

Abstract

A portable electronic tongue has been developed using an array of eighteen thick-film electrodes of different materials forming a multi-electrode array. A microcontroller is used to implement the pattern recognition. The classification of drinking waters is carried out by a Microchip PIC18F4550 micro-controller and is based on neural networks algorithms. These algorithm are initially trained with the multi-electrode array on a Personal Computer (PC) using several samples of waters (still, sparkling and tap) to obtain the optimum architecture of the networks. Once it is trained, the computed data are programmed into the microcontroller, which then gives the water classification directly for new unknown water samples. A comparative study between a Fuzzy ARTMAP, a Multi-Layer Feed-Forward network (MLFF) and a Linear Discriminant Analysis (LDA) has been done in order to obtain the best implementation on a microcontroller.

Introduction

Sensors based on electrochemical techniques are used to determine the concentration of specific chemical compounds, or the accurate measurement of physiochemical parameters. But generally they have an important drawback; namely that of susceptibility to interference from other species that mask the species of interest. However this drawback can be converted to advantage if, instead of looking for that type of accurate measurement, another kind 7 measurement of a rather more qualitative nature is employed, such as the discrimination or classification of samples of complex chemical nature. Under this concept, electronic tongue systems that employ different sets of non-specific electrodes were developed some years ago [1]. Each of the electrodes provides a signal that is proportional to the set of species in the system under analysis. As electronic tongue systems tend to produce a qualitative result, multivariate analysis techniques are generally required in order to process the data obtained from the measurements.

Various electrochemical techniques have been used in electronic tongues, such as potentiometry [2], voltammetry [3] or impedance spectroscopy [4]. These have been used in several applications, including waste water control [5] and food analysis [6].

Potentiometric techniques have as their main desirable feature simplicity of measurement method and electronic equipment. Various different types of electrodes have been used in potentiometry, such as membranes [7] or metal surfaces [8]. In this latter electrode, a voltage is obtained that is proportional to the concentrations of all species present in solution and hence its quantification is difficult to determine whenever the aqueous medium is complex [9].

A method for obtaining a multi-electrode of easy construction and simple operation is to employ inks from thick-film hybrid circuit technology [10] because there are many different types of inks and each has a key chemical element that can become the active element of the sensor.

Most systems of electronic tongues remain in the laboratory version, which requires the presence of a computer and, specially above all, two separate processes, one for taking measurements and another for data processing. If it is desired for these systems to have industrial application however, it is necessary to unify these two phases into a single system. The best method for achieving a single system is the use of microcontrollers in systems which, in addition to the measurement of potential, are able to perform the analysis of relevant data using a software program implemented in the microcontroller memory. Thus portable electronic tongues are becoming popular as they offer simplicity, reliability and use in field [11]. Some systems using microprocessors have been presented as electronic tongues [12] but the system presented in this communication has as its main novelty the development and comparison of three types of pattern recognition algorithms. Pattern recognition algorithms have become a critical component in the implementation of electronic tongues and noses and have been used successfully in these applications [13]. For implementation in portable equipment the algorithm must be transferable to a microcontroller which has a limited amount of memory. Thus the perfect pattern recognition algorithms will require high accuracy, to work fast to work in real-time and have low memory requirements in order to be implemented in a microcontroller. Not all pattern recognition algorithms are able to reach each of these requirements. In this communication three pattern recognition algorithms have been used, Fuzzy ARTMAP, Multi-Layer Feed-Forward (MLFF) and Linear Discriminant Analysis (LDA).

MLFF is the most popular type of artificial neural network (ANN); basically it is formed basically by three layers of neurons (input, hidden and output). They require a training stage, where the weights of each neuron are set, and another validation stage [14]. The Fuzzy ARTMAP network uses the so-called adaptive resonance method and is based on the use of prior actions to predict subsequent steps [15]. For LDA the method is a probabilistic parametric classification technique and maximizes the variance between categories and minimizes the variance within categories, by means of a data projection from a high dimensional space to a low dimensional space. In this way, a number of orthogonal linear discriminant functions equal to the number of categories minus one are obtained [16], [17]. Such algorithms have been used in electronic noses [18] and electronic tongue systems [19] giving important benefits such as: simplicity of implementation of computer algorithm, speed of calculation and the attainment of good and reliable results with a small number of measures.

The aim of this paper is to present a potentiometric electronic tongue system that uses an electrode assembly constructed in thick film technology whose data analysis system consists of a pattern recognition algorithm implemented on a microprocessor system. As an example application of this system, an analysis has been made of various types of drinking water that have different concentrations and types of salts. The implemented pattern recognition algorithm is able to perform a classification of these water samples using the data obtained from potentiometric measurements. This example can be extended to other industrial applications such as quality control of water purification, wastewater discharges, quality control of drinks and, in general, in cases where it is appropriate to conduct qualitative measures quickly, easily, economically, and not necessarily carried out by specialized personnel.