Journal of Chemometrics最新文献

To address the challenges of extracting features from complex industrial process data, the reliance of numerous fault detection methodologies on presupposed data distribution types, and the limited generalization capacity of fault detection, this manuscript introduces a sophisticated algorithm for industrial process fault detection. This algorithm harnesses the information gain adaptive (IGA) technique for feature selection and a synergistic model decision mechanism. Initially, the process involves the computation of information gain via decision trees, coupled with the determination of the $��k�$ value through cross-validation. This strategy enables the adaptive selection of features, thereby facilitating data dimensionality reduction and effective feature extraction. The subsequent phase introduces a ternary statistical measure monitoring group for the detection of linear faults, while autoencoders and one-class SVM methodologies are applied for the monitoring of nonlinear faults. The culmination of this approach is the development of an innovative weighted decision mechanism, designed to amalgamate the findings from both linear and nonlinear detection avenues, yielding more dependable detection results. The validation of this algorithm employs datasets from the water chillers process and Tennessee Eastman (TE) process, demonstrating the IGA-combined model's superior performance over isolated linear or nonlinear detection algorithms in terms of detection accuracy and robustness. Notably, the efficacy of this method is not contingent upon specific assumptions regarding data distribution, rendering it a versatile and efficacious tool for the fault detection in industrial processes.

Given that an effective process monitoring implementation should take both the spatial and temporal variations into account, a novel online process monitoring scheme based on a newly formulated algorithm titled as spatial–temporal deviation analysis (STDA) is proposed. Different from the mainstream process monitoring methods that focus on characterizing the spatial and/or temporal variation in the historical normal samples, the proposed STDA algorithm is designed to adaptively and timely train a pair of projecting vectors to uncover potential deviation in the spatial–temporal variation of online monitored samples, so as to guarantee consistently enhanced monitoring performance. Instead of utilizing a fixed projecting framework trained offline, the STDA algorithm is repeatedly executed once a newly measured sample become available for online monitoring. Therefore, the proposed STDA-based method could consistently ensure its effectiveness for online fault detection, because a projecting framework targeted to revealing deviation in spatial–temporal variation is dynamically determined for different online monitoring samples in a timely manner. Finally, the salient monitoring performance achieved by the proposed STDA-based approach is evaluated through comparisons with other counterparts.

Multivariate calibration techniques and machine learning algorithms are inextricably linked within the realm of chemometrics and data analysis. Classical least squares (CLS) modeling, a fundamental multivariate regression approach, has traditionally been utilized for quantitative analysis tasks, establishing relationships between predictor variables (e.g., spectroscopic data) and response variables (e.g., chemical concentrations). However, a unique feature of CLS is its ability to handle scenarios with partial knowledge of the independent variable matrix, making it an intriguing candidate for qualitative pattern recognition and discriminant analysis applications. This study proposes a novel approach, Classical Least Squares Discriminant Analysis (CLS-DA), which combines the principles of CLS modeling with discriminant analysis objectives. The performance of CLS-DA is comprehensively evaluated using two real-world datasets: chemical analysis of three wine cultivars and mid-infrared spectroscopy of minced meat samples (pork, chicken, and turkey). The results are compared against the well-established Partial Least Squares Discriminant Analysis (PLS-DA) method, a widely adopted technique for classification tasks in chemometrics. For both sets of experimental data, CLS-DA and PLS-DA showed comparable efficiency. For the classification of three types of wine, the accuracy of the proposed method was 94.3%, while the accuracy of the reference method was 98.1%. For the classification of minced meat samples, the accuracies of CLS-DA and PLS-DA were 97.2% and 94%, respectively for all three groups. The findings demonstrate the potential of CLS-DA as a straightforward and interpretable supervised pattern recognition technique, exhibiting comparable classification performance to PLS-DA. The study highlights the advantages of CLS-DA, including its ability to operate within the original data space and its flexibility in accommodating partial knowledge scenarios. The proposed CLS-DA approach presents a promising alternative for discriminant analysis, offering new perspectives on the applications of classical least squares modeling in chemometrics.

Currently, the majority of methods to monitor cancer treatment through the analysis of body fluids are based on a highly selective detection of single molecules or cells. In this study, we are considering the analysis of the aqueous medium of liquid samples, that is, water, itself, using aquaphotomics and near-infrared spectroscopy (NIR) for spectral data acquisition and processing, within cancer research. Water, as a molecular system, is a rich source of information about the current state of a patient, which can be extracted from near-infrared spectra of liquid samples via simple algorithms based on multivariate data analysis. The reported results, obtained ex vivo of body fluids, demonstrate the potential of aquaphotomics in cancer research.

Here, we demonstrate mid-field ¹H NMR spectroscopy combined with chemometrics to be powerful in the classification and authentication of motor oils (MOs). The ¹H NMR data were processed with a new algorithm for simultaneous phase and baseline correction, which, for crowded spectra such as those of the refinery products, allowed for more accurate estimation of phase parameters than other literature approaches tested. A principal component analysis (PCA) model based on the unbinned CH₃ fingerprint region (0.6–1.0 ppm) enabled the differentiation of hydrocracked and poly-α-olefin-based MOs and was effective in resolving mixtures of these base stocks with conventional base oils. PCA analysis of the 1.0- to 1.14-ppm region enabled the detection of poly (isobutylene) additive and was useful for differentiating between single-grade and multigrade MOs. Non-equidistantly binned ¹H NMR data were used to detect the addition of esters and to establish discriminant models for classifying MOs by viscosity grade and by major categories of synthetic, semisynthetic, and mineral oils. The performances of four classifiers (linear discriminant analysis [LDA], quadratic discriminant analysis [QDA], naïve Bayes classifier [NBC], and support vector machine [SVM]) with and without PCA dimensionality reduction were compared. In both tasks, SVM showed the best efficiency, with average error rates of ~2.3% and 8.15% for predicting major MO categories and viscosity grades, respectively. The potential to merge spectra collected from different NMR instruments is discussed for models based on spectral binning. It is also shown that small errors in phase parameters are not detrimental to binning-based PCA models.