Liquid Sensing : Development and Characterisation of an Electronic Tongue Based on Electrochemical Methods

Sammanfattning: A new sensor technology for liquid sensing is reported, i.e. the electronic tongue based on electrochemical methods. Such a system involves the combination of non-selective sensors (metal electrodes) and a signal processing part. It determines attributes (quality, status of a process etc) rather than single parameters. Transient current responses are obtained as a result of large potential pulses consecutively applied to each electrode. Unique response patterns for liquid samples are interpreted with mathematical and statistical methods, for example principal component analysis (PCA). In this thesis, focus was put on general aspects of development and characterisation of the sensor system, since the long-time objective is to apply it to industrial processes. This was accomplished by studying sensitivity, selectivity, long-term stability and signal processing.The proper choice of sensors and measurement techniques is important for successful discrimination of liquid samples. For example, liquid washing detergents were discriminated with PCA by merging data from several voltammetric electrodes like copper, glassy carbon, gold, iridium, silver and platinum. In another study, fermented milk samples were successfully discriminated as sensors of different measurement technologies were combined, i.e. conductometry, potentiometry and voltammetry. Sensors of each technique alone were not able to separate all samples.Adsorption and/ or porous layer formation lead to drift in electrode responses. The former can inhibit or enhance an electron rate transfer and the latter increases the electrode area and thus, the observed current. Drift counteraction was studied in two different ways. First, mathematical algorithms like component correction (CC) and additive correction (AC) were applied to different data sets. Both methods require the use of reference samples. It was found that CC was better to remove drift in current responses; CC was not as sensitive as AC to dissimilar changes between reference samples and other liquids. Physical treatments like polishing and electrochemical cleaning were also studied. In electrochemical cleaning, a high and a low potential was applied to each electrode prior to analysis. In addition, settings like potential and time in the electrochemical procedure were studied for each electrode with multivariate experimental design. Unfortunately, true optima were not found.Signal processing to reduce large data sets (about 8000 variables) was also employed. Three methods were compared, a chemical/physical method (three parameters were taken from the current transient obtained for each potential step), hierarchical PCA (HPCA) and wavelet transformation (WT). It was found that HPCA compressed the data sets to the largest extent, but that discrimination performance improved for the chemical/physical method compared to the other techniques.