Deep learning assisted distinguishing of honey seasonal changes using quadruple voltammetric electrodes.

The work presents innovative quadruple disk iridium, platinum, and iridium-platinum voltammetric electrodes with a special design, dedicated to the testing of samples with a complex organic composition. Noble metal wires are tightened in one silver rod, and therefore each of them acts as a single sensor. It was demonstrated that the signals of the iridium-platinum sensor combine the electrode responses constructed from one metal, which increases the possibilities and range of applications of this sensor, and it can be used as an electronic tongue. These single and combined noble metal electrodes were successfully verified to profile the seasonal variability of honey collected from an apiary in Malopolska (voivodeship in Poland). Data obtained by the differential pulse voltammetry, according to the principles of green chemistry, without using any reagents, were interpreted by principal component analysis, preceded by the optimized variable selection procedure. The best results in distinguishing 12 honeys were obtained using a multimetallic electrode. The classification model calculated using deep convolutional neural networks indicated the proper belonging of honeys to the groups with 100% accuracy for the training and validation set. The proposed solution proved that noble metals quadruple disk electrodes are a promising tool supporting voltammetric profiling of samples and this strategy, considering deep learning, can be developed to a large extent.

Pattern recognition in voltammetric signals by ASD trilinear decomposition.

This work proposes a modern approach in separation of voltammetric current components which applies a pattern recognition strategy. Application of the trilinear decomposition algorithm ASD was first time demonstrated for detection and modelling of faradaic and capacitive components, recorded as total signal in DPV experiments. Using calibration set of the model substance i.e. K4Fe(CN)6, sampled with the frequency of 1 kHz, the signal components were extracted, which shape was consistent with the theoretic relations. Fitting in the area of peak expressed by r was 0.9957 for faradaic current and 0.9916 for capacitive component. Also the successive stages of the redox process was identified and interpreted. Essential influence for the quantitative analysis was observed. DP voltammograms utilizing all faradaic current samples were characterized by zero baseline and improved calibration line parameters. A wide application list of this modern strategy was also proposed.

Faradaic and capacitive current estimation by means of Independent Components Analysis and 1kHz sampling.

In this work it was demonstrated for the first time that Independent Components Analysis non-deterministic multivariate strategy enables estimation of the Faradic and double layer charging current components in differential pulse and anodic stripping voltammetry. The signals separation can be realized based on the different shapes of the capacitive and Faradaic current functions. The set of calibration voltammograms measured with the 1kHz sampling frequency is the sufficient data to retrieve signals compatible with the theoretical shapes, with r2 greater than 0.987 for Faradaic component and greater than 0.994 for capacitive component. In the effect, DP voltammograms of Pb(II) and Cd(II) recorded on silver annular band electrode were constructed using only Faradaic component. Quantitative analysis was performed and calibration models were calculated by different waiting time and sampling time, what demonstrated the possibility of significant improvement of sensitivity in considered experiments.

Novel method for standard addition signal analysis in voltammetry.

A new procedure for standard addition analysis in voltammetry based on the slope of the peak at the inflection point (SSAA) is proposed. It is fully automated and does not require any subjective user decisions. It has been tested for both simulated and experimental data. Experimental data included DPASV measurements of Pb(ii) and Cd(ii) on RAgABE, and Tl(i) on BiABE as a working electrode. In each case the procedure offered good accuracy, precision and perfect reproducibility.

Unsupervised pattern recognition methods in ciders profiling based on GCE voltammetric signals.

This work presents a complete methodology of distinguishing between different brands of cider and ageing degrees, based on voltammetric signals, utilizing dedicated data preprocessing procedures and unsupervised multivariate analysis. It was demonstrated that voltammograms recorded on glassy carbon electrode in Britton-Robinson buffer at pH 2 are reproducible for each brand. By application of clustering algorithms and principal component analysis visible homogenous clusters were obtained. Advanced signal processing strategy which included automatic baseline correction, interval scaling and continuous wavelet transform with dedicated mother wavelet, was a key step in the correct recognition of the objects. The results show that voltammetry combined with optimized univariate and multivariate data processing is a sufficient tool to distinguish between ciders from various brands and to evaluate their freshness.

Voltammetric classification of ciders with PLS-DA.

Voltammograms recorded on the glassy carbon electrode (GCE) may be a chemical fingerprints of food samples, enabled distinguishing the origin of the considered products. In this work the objects of the study was 5 Polish ciders of various brands. For each sample 10 scans were recorded by DPV in the potential range between -0.2 and 1.0 V in Britton-Robinson buffer at pH 2.0. The signals preprocessing realized by baseline correction with 4-th degree polynomial and normalization (in 0 to 1 interval), performed to reduce problems with insufficient signal's repeatability associated with mechanical renovation of the electrode surface before each measurement. The PLS-DA classification models were built using the training set and then validated using the samples absent in the learning process. The final multi-class model with optimized complexity enables classification of the ciders with 100% sensitivity and specificity, with the exception of one cider, where specificity was 95% (for validation set).