ELECTRONIC TONGUE / ELECTRONIC NOSEPrepared by Patrycja Ciosek ELECTRONIC TONGUE / ELECTRONIC NOSE (ETongue, ENose) are systems for automatic analysis and recognition (classification) of liquids or gases, including arrays of non-specific sensors, data collectors and data analysis tools. Electronic tongues are used for liquid samples analysis, whereas electronic noses - for gases. The result of Etongue/Enose can be the identification of the sample, an estimation of its concentration or its characteristic properties. This new technology has many advantages. Problems associated with human senses, like individual variability, impossibility of on-line monitoring, subjectivity, adaptation, infections, harmful exposure to hazardous compounds, mental state, are no concern of it. Synonyms of an electronic tongue: artificial tongue, taste sensor Synonyms of an electronic nose: artificial nose, olfactory system APPLICATIONS OF E-TONGUES/E-NOSES:Foodstuffs Industry
Medicine
Safety
Environmental pollution monitoring
Quality control of air in buildings, closed accommodation (i.e. space station, control of ventilation systems) Chemical Industry
Legal protection of inventions - digital "fingerprints" of taste and odours SENSING METHODS APPLIED
PATTERN RECOGNITIONThe electronic tongue or nose system performance is dependent on the quality of functioning of its pattern recognition block. Various techniques and methods can be used separately or together to perform the recognition of the samples. After measurement procedure the signals are transformed by a preprocessing block. The results obtained are inputs for Principal Components Analysis, Cluster Analysis or Artificial Neural Network. Measurement Sensors arrays' outputs are arranged in data matrix (Fig. 1).
Each sample is characterized by unique and typical set of data, forming "fingerprint" of an analyte in m-dimensional pattern space. Preprocessing Preprocessing is the phase in which linear transformation on the data matrix is performed (without changing the dimensionality of the problem) in order to enhance qualitative information. Typical techniques include manipulation of sensor baseline, normalization, standarization and scaling of response for all the sensors in an array. Principal Component and Cluster Analysis A multi-sensor system produces data of high dimensionality - hard to handle and visualize. Principal Component Analysis (PCA) and Cluster Analysis (CA) are multivariate pattern analysis techniques reducing dimensionality of the problem and reducing high degree of redundancy. PCA is a linear feature-extraction technique finding most influential, new directions in the pattern space, explaining as much of the variance in the data set as possible. This new directions - called principal components - are the base for a new data matrix. Usually 2 or 3 of them are sufficient to transfer more than 90% of the variation of the samples. The base principle of Cluster Analysis is the assumption of close position of similar samples in multidimensional pattern space. Similarity between each 2 samples is calculated as a function of the distance between them - usually in Euclidean sense - and displayed on a dendrogram (Fig. 2).
Artificial Neural Networks (ANN) Neural Networks are information processing structures imitating behavior of human brain. Their main advantages, such as: adaptive structure, complex interaction between input and output data, ability to generalize, parallel data processing and handling incomplete or high noise level data make them useful pattern recognition tools. There are many possible architectures and algorithms available in the literature, but the most common in measurement applications is feed-forward network (multilayer perceptron MLP) and back-propagation learning algorithm. The base units of artificial neural networks are neurons and synapses. Neurons are organized in layers and connected by synapses. Their task is to sum up their inputs and non-linear transfer of the result, which is then transmitted via synapsis with modification by means of the synapsis weights - this signal, in turn, is the input for the next layer of the network (Fig. 3).
The use of ANN involves 3 phases:
REFERENCES:
LINKS:Commercially available e-noses/e-tongues:
Chemometrics: |
|
