Modeling of the polluting emissions from a cement production plant by partial least-squares, principal component regression, and artificial neural networks

A Portland cement process was taken into consideration and monitored for one month with respect to polluting emissions, fuel and raw material physical-chemical properties, and operative conditions. Soft models, based on linear (partial least-squares, PLS, and principal component regression, PCR) and nonlinear (artificial neural networks, ANNs) approaches, were employed to predict the polluting emissions. The predictive ability of the three regression methods was evaluated by means of the partition of the dataset by Kohonen self-associative maps into both a training and a test set. Then, a "leave-more-out" approach, based on the use of a training set, a test set, and a production set, was adopted. The training set was used to build the models, the test set was used to select the number of latent variables or the neural network training endpoint, and the production set was used to produce genuine predictions. ANNs proved to be much more effective in prediction with respect to PLS and PCR and, at least in the case of SO2 and dust, provided a predictive ability comparable with the experimental estimated uncertainty of the response. This showed that it is possible to satisfactorily predict the two responses. Such a prediction will result in the prevention of environmental and legal problems connected to the polluting emissions.     

Prediction of consumer liking from trained sensory panel information: Evaluation of neural networks

This study was set out to establish artificial neural networks (ANN) as an alternative to regression methods (multiple linear, principal component and partial least squares regression) to predict consumer liking from trained sensory panel data. The sensory profile and acceptability of 10 market samples of beef bouillon products were measured. The products were distinct as evaluated by the trained sensory panel. A total of 100 regular beef bouillon product users from Manila measured overall liking, flavour, aftertaste and mouthfeel of the products. Curve fitting method was applied to identify sensory drivers of consumer liking. The sensory drivers of consumer liking were used as explanatory variables in artificial neural networks and regression methods. To overcome the limitations of regression methods we have used artificial neural network techniques to model consumer liking score as a function of trained sensory panel scores and achieved quite encouraging results. Our simulation experiments show that though the regression methods such as multiple linear regression (MLR), principal component regression (PCR) and partial least square (PLS) give an accurate prediction of consumer liking scores, this approach is not robust enough to handle the variations normally encountered in trained sensory panel data. ANNs were trained using the sensory panel raw data and transformed data. The networks trained with sensory panel raw data achieved 98% correct learning, the testing was in a range of 28–35%. Suitable transformation method was applied to reduce the variations in trained sensory panel raw data. The networks trained with transformed sensory panel data achieved about 80–90% correct learning and 80–95% correct testing. It is shown that due to its excellent noise tolerance property and ability to predict more than one type of consumer liking using a single model, the ANN approach promises to be an effective modelling tool.     

Application of Artificial Neural Networks (ANNs) in Wine Technology

In recent years, neural networks have turned out as a powerful method for numerous practical applications in a wide variety of disciplines. In more practical terms neural networks are one of nonlinear statistical data modeling tools. They can be used to model complex relationships between inputs and outputs or to find patterns in data. In food technology artificial neural networks (ANNs) are useful for food safety and quality analyses, predicting chemical, functional and sensory properties of various food products during processing and distribution. In wine technology, ANNs have been used for classification and for predicting wine process conditions. This review discusses the basic ANNs technology and its possible applications in wine technology.     

An artificial neural network model for predicting flavour intensity in blackcurrant concentrates

Artificial neural networks (ANNs)—machine learning acquiring knowledge in training and using deduced relationships to predict responses—were studied to rationalise concentrate use in fruit drinks production. Sets of ANNs were developed for predicting flavour intensity in blackcurrant concentrates from gas chromatographic data on flavour components (37) in 133 sorbent extracts from blackcurrant concentrates varying in season, geographical origin and processing technology. Sensory data was collected using ratio scaling on flavour intensities in drinks from concentrates. Relationships between chromatographic and sensory data for concentrates of three seasons (1989, 1990 and 1992) were modelled by ANNs with back propagation using principal component regression scores as input. Predictions were compared with a global model from random concentrates from all three seasons. In predicting overall flavour intensity, ANN models were better fitted than partial least square regression. Ability of artificial neural networks to simulate non-linear relationships observed in human perceptions could explain such improvements.     

Consumer acceptability and purchase intent of probiotic yoghurt with added glucose oxidase using sensometrics,artificial neural networks and logistic regression

This research aimed to identify the drivers of acceptance and purchase intent of a probiotic (Bifidobacterium longum BL05) nonflavoured yoghurt supplemented with glucose oxidase, and to model the consumers’ acceptability using sensometrics and artificial neural networks (ANN). Consumers (n = 100) evaluated the degree of liking of yoghurt assays in respect of appearance, aroma, taste, texture and overall linking. Sensometric techniques – multiple linear regression (MLR), partial least squares regression (PLS), principal component regression (PCR) – and ANN were used to model the overall liking. Sensory drivers of global acceptance and purchase intent were also determined using logistic regression (LR). Hierarchical cluster analysis (HCA) identified three consumer segments that presented differences in all sensory attributes evaluated (P < 0.05). The ANN model showed the best performance to predict overall liking, followed by the MLR, PLS and PCR, indicating that taste and texture were the most significant attributes impacting the yoghurts overall liking. In accordance with the logistic models, overall acceptance and purchase intent could be predicted with 81.94 and 85.49% accuracy, respectively. The logistic regression indicated that taste was the attribute that contributed significantly (P < 0.0001) to higher scores for purchase intent and was considered the driver of acceptance.     

Comparison between linear and nonlinear PLS methods to explain overall liking from sensory characteristics

The influence of sensory characteristics on overall liking can be statistically studied with Partial Least Squares (PLS) regression methods. To correctly model nonlinear dependence relationships, some nonlinear PLS extensions are useful. The purpose of the present paper is to compare performances and results of three PLS methods, using a real data set: regular PLS with sensory attributes as explanatory variables; PLS with attributes and their respective squares; and a new nonlinear PLS extension, called ASPLS. In case of a nonlinear dependence relationship between sensory characteristics and hedonic responses, this last method is shown to be worth considering.     

A hybrid Bayesian-neural network approach for probabilistic modeling of bacterial growth/no-growth interface

A hybrid probabilistic modeling approach that integrates artificial neural networks (ANNs) with statistical Bayesian conditional probability estimation is proposed. The suggested approach benefits from the power of ANNs as highly flexible nonlinear mapping paradigms, and the Bayes' theorem for computing probabilities of bacterial growth with the aid of Parzen's probability distribution function estimators derived for growth and no-growth (G/NG) states. The proposed modeling approach produces models that can predict the probability of growth of targeted microorganism as affected by a set of parameters pertaining to extrinsic factors and operating conditions. The models also can be used to define the probabilistic boundary (interface) between growth and no-growth, and as such can define and predict the values of critical parameters required to keep a desired pre-specified bacterial growth risk in check. A modular system incorporating the various computational modules was constructed to illustrate the application of the hybrid approach to the probabilistic modeling of growth of pathogenic Escherichia coli strain as affected by temperature and water activity. The proposed approach was compared to other techniques including the traditional linear and nonlinear logistic regression. Results indicated that the hybrid approach outperforms the other approaches in its accuracy as well as flexibility to extract the implicit interrelationships between the various parameters. Advantages and limitations of the approach were also discussed and compared to those of other techniques.     

Nondestructive Prediction of Moisture and Sodium Chloride in Cold Smoked Atlantic Salmon (Salmo salar)

ABSTRACT: Salt and moisture contents in cold-smoked salmon were determined using short-wavelength near-infrared (SW-NIR) reflectance spectroscopy (600 to 1100 nm). Partial least square (PLS) regression models yielded the best results among 3 linear regression methods tested. Back-propagation neural networks (BPNN) exhibited a somewhat better capability to model salt and moisture concentrations (Salt: R²= 0.824, RMS = 0.55; Moisture: R²= 0.946, RMS = 2.44) than PLS (Salt: R²= 0.775, RMS = 0.63; Moisture: R²= 0.936, RMS = 2.65). Selection of samples from different axial locations on a fish did not affect the prediction error for salt or WPS but affected the prediction error for moisture.     

Real-time process modeling of particleboard manufacture using variable selection and regression methods ensemble

This study focuses on the real-time prediction of mechanical properties such as internal bond strength (IB) and modulus of rupture (MOR) for a wood composite panels manufacturing process. As wood composite panel plants periodically test their products, a real time data fusion application was developed to align laboratory mechanical test results and their corresponding process data. Fused data were employed to build regression models that yield real-time predicted mechanical property values when new process data become available. The modeling algorithm core uses genetic algorithm to preselect a meaningful subset of process variables. Calibration models are then built using several regression methods: multiple linear regression, ridge regression, neural networks, and partial least squares regression (PLS). Four different predicted response values were generated for each new record of real time process variables. On-line validation results showed good performance of the ridge regression method with a 0.89 correlation coefficient between actual and predicted MOR values, a root mean square error (RMSEP) of 1.05 MPa and a mean normalized error of 9 %. IB was best predicted by PLS with a 0.81 correlation coefficient between actual IB and PLS predicted IB values, a RMSEP of 75.1 kPa, and a mean normalized error of 15 %.     

Estimation of Dielectric Properties of Cakes Based on Porosity, Moisture Content, and Formulations Using Statistical Methods and Artificial Neural Networks

Dielectric constant (DC) and dielectric loss factor (DLF) are the two principal parameters that determine the coupling and distribution of electromagnetic energy during radiofrequency and microwave processing. In this study, chemometric methods [classical least square (CLS), principle component regression (PCR), partial least square (PLS), and artificial neural networks (ANN)] were investigated for estimation of DC and DLF values of cakes by using porosity, moisture content and main formulation components, fat content, emulsifier type (Purawave™, Lecigran™), and fat replacer type (maltodextrin, Simplesse). Chemometric methods were calibrated firstly using training data set, and then they were tested using test data set to determine estimation capability of the method. Although statistical methods (CLS, PCR and PLS) were not successful for estimation of DC and DLF values, ANN estimated the dielectric properties accurately (R ², 0.940 for DC and 0.953 for DLF). The variation of DC and DLF of the cakes when the porosity value, moisture content, and formulation components were changed were also visualized using the data predicted by trained network.     

Detection of irradiated beef by nuclear magnetic resonance lipid profiling combined with chemometric techniques

The combination of ¹H NMR lipid profiling with multivariate analysis was applied to differentiate irradiated and non-irradiated beef. Two pattern recognition chemometric procedures, stepwise linear discriminant analysis (sLDA) and artificial neural networks (ANNs), provided a successful discrimination between the groups investigated. sLDA allowed the classification of 100% of the samples into irradiated or non-irradiated beef groups; the same result was obtained by ANNs using the 1 kGy irradiation dose as discriminant value suggested by the network. Furthermore, sLDA allowed the classification of 81.9% of the beef samples according to the irradiation dose (0, 2.5, 4.5 and 8 kGy). ¹H NMR lipid profiling, coupled with multivariate analysis may be considered a suitable and promising screening tool for the rapid detection of irradiated meat in official control of food.     

Quantification of aspartame in commercial sweeteners by FT-Raman spectroscopy

Efficient methods are proposed herein for the quantification of aspartame in commercial sweeteners. These methods are based on a treatment of Raman data with partial least squares (PLS), principal component regression (PCR) and counter-propagation artificial neural networks (CP-ANN) methods. For the three chemometric techniques used, the relative standard errors of prediction (RSEP) calculated for calibration and validation data sets were on the order of 1.8–2.2%. Four commercial preparations containing between 17% and 36% of aspartame by weight were evaluated by applying the developed models. Concentrations found from the Raman data analysis agree perfectly with the results of the UV–Vis reference analysis, with the recoveries in the 98.7–100.8%, 98.6–101.1% and 97.8–102.2% ranges for the PLS, PCR and CP-ANN models, respectively. The proposed procedures can be used for routine quality control during the production of commercial aspartame sweeteners.     

Application of artificial neural network (ANN) for the prediction of thermal resistance of knitted fabrics at different moisture content*

Thermal resistance of the fabrics is one of the decisive parameters in terms of comfort; however it can change due to wetting. Therefore, thermal resistance of wetted fabric is important for comfort performance of garments. In recent years, artificial neural networks (ANN) have been used in the textile field for classification, identification, prediction of properties and optimization problems. ANNs can predict the fabric thermal properties by considering the influence of all fabric parameters at the same time. In this study, ANNs were used to predict thermal resistance of wetted fabrics. For this aim, two different architectures were experienced and high regression coefficient (R²) between the predicted (training and testing) and observed thermal resistance values were obtained from both models. The obtained regression coefficient values were over 90% for both models. Then it can be said that ANNs could be used for predicting thermal resistance of wetted fabrics successfully.     

Nondestructive Determination of Moisture and Sodium Chloride in Cured Atlantic Salmon (Salmo salar) (Teijin) Using Short-wavelength Near-infrared Spectroscopy (SW-NIR)

ABSTRACT: Moisture (49.70 to 74.20% w/w) and salt (0.13 to 12.30% w/w) concentrations in cured Atlantic salmon ( Salmo salar ) or teijin were determined by short-wavelength near-infrared (SW-NIR) reflectance spectroscopy (600 to 1100 nm) using partial least square regression (PLS) and artificial neural networks (ANN) calibration methods. ANN and PLS yielded similar results (Salt: ANN RMS = 1.43% w/w, PLS RMS = 1.37% w/w; Water, ANN RMS = 2.08% w/w, PLS RMS = 2.04% w/w). Sampling the dorsal or ventral portion of the fish did not appear to affect the prediction error of the salt or moisture models.     

Intelligent Estimation of the Canola Oil Stability Using Artificial Neural Networks

In the present study, a multi-layer perceptron neural network and radial basis function (RBF) network were used to estimate the oxidative stability of canola oil during storage. Artificial neural networks (ANNs) were used to model oxidative stability of canola oil during storage, and comparison was also made with the results obtained from a regression analysis. The oxidative stability of canola oils was considered as dependent variable, and independent variables were selected as time (in week), variety, C14:0, C16:0, C18:0, C20:0, C18:1, C18:2, C18:3, and C22:1 fatty acid content. The results were compared with experimental data and it was found that the estimated oxidative stability by RBF neural network is more accurate than multi-layer perceptron network and regression model. It was also found that the oxidative stability of canola oil decreased with increase in storage time and C18:3 fatty acid content.     

Application of near infrared spectroscopy for the rapid determination of antioxidant activity of bamboo leaf extract

This study was carried out to evaluate the feasibility of using near infrared (NIR) spectroscopy for determining three antioxidant activity indices of the extract of bamboo leaves (EBL), specifically 2,2-diphenyl-1-picrylhydrazyl (DPPH), ferric reducing/antioxidant power (FRAP), and 2,2′-azinobis-(3-ethylbenz-thiazoline-6-sulfonic acid) (ABTS). Four different linear and nonlinear regressions tools (i.e. partial least squares (PLS), multiple linear regression (MLR), back-propagation artificial neural network (BP-ANN), and least squares support vector machine (LS-SVM)) were systemically studied and compared in developing the model. Variable selection was first time considered in applying the NIR spectroscopic technique for the determination of antioxidant activity of food or agricultural products. On the basis of these selected optimum wavelengths, the established MLR calibration models provided the coefficients of correlation with a prediction (r_pre) of 0.863, 0.910, and 0.966 for DPPH, FARP, and ABTS determinations, respectively. The overall results of this study revealed the potential for use of NIR spectroscopy as an objective and non-destructive method to inspect the antioxidant activity of EBL.     

Predicting acetic acid content in the final beer using neural networks and support vector machine

Most acetic acid found in beer is produced by yeast during fermentation. It contributes significantly to beer taste, especially when its content is higher than the taste threshold in beer. Therefore, the control of its content is very important to maintain consistent beer quality. In this study, artificial neural networks and support vector machine (SVM) were applied to predict acetic acid content at the end of a commercial‐scale beer fermentation. Relationships between beer fermentation process parameters and the acetic acid level in the fermented wort (beer) were modelled by partial least squares (PLS) regression, back‐propagation neural network (BP‐NN), radial basis function neural network (RBF‐NN) and least squares‐support vector machine (LS‐SVM). The data used in this study were collected from 146 production batches of the same beer brand. For predicting acetic acid content, LS‐SVM and RBF‐NN were found to be better than BP‐NN and PLS. For the comparison of RBF‐NN and LS‐SVM, RBF‐NN had a better reliability of model, but lower reliability of prediction. SVM had better generalization, but lower reliability of model. In summary, LS‐SVM was better than RBF‐NN modelling for the prediction of acetic acid content during the commercial beer fermentation in this study. Copyright © 2013 The Institute of Brewing & Distilling     

PREDICTION OF MECHANICAL PROPERTIES OF CUMIN SEED USING ARTIFICIAL NEURAL NETWORKS

In this paper, two artificial neural networks (ANNs) are applied to acquire the relationship between the mechanical properties and moisture content of cumin seed, using the data of quasi-static loading test. In establishing these relationship, the moisture content, seed size, loading rate and seed orientation were taken as the inputs of both models. The force and energy required for fracturing of cumin seed, under quasi-static loading were taken as the outputs of two models. The activation function in the output layer of models obeyed a linear output, whereas the activation function in the hidden layers were in the form of a sigmoid function. Adjusting ANN parameters such as learning rate and number of neurons and hidden layers affected the accuracy of force and energy prediction. Comparison of the predicted and experimented data showed that the ANN models used to predict the relationships of mechanical properties of cumin seed have a good learning precision and good generalization, because the root mean square errors of the predicated data by ANNs were rather low (4.6 and 7.7% for the force and energy, respectively).

PRACTICAL APPLICATIONS

Cumin seed is generally used as a food additive in the form of powder for imparting flavor to different food preparations and for a variety of medicinal properties. Physical properties of cumin seeds are essential for the design of equipment for handling, harvesting, aeration, drying, storing, grinding and processing. For powder preparation especially the fracture behavior of the seeds are essential. These properties are affected by numerous factors such as size, form and moisture content of the grain and deformation speed. A neural network model was developed that can be used to predict the relationships of mechanical properties. Artificial neural network models are powerful empirical models approach, which can be compared with mathematical models.     

Comparison of PLS, PCR and MLR for the quantitative determination of foreign oils and fats in butter fats of several European countries by their triglyceride composition

Three methods for evaluation of gas Chromatographic data have been compared. Multiple linear regression (MLR) as developed by Precht, principal component regression (PCR) and partial leastsquares regression (PLS) have been applied to the detection of foreign fat added to pure butter fat samples obtained from several European countries. The emphasis was put on building a calibration model for the general detection of foreign fat and therefore only one was built for the quantitation of various vegetable oils and lard MLR, as developed for German butter fat, was found to be appropriate also for the detection of the addition of about 3–5% foreign fat depending on the formula used. PCR calibration leads to a model with 11 factors indicating a detection limit of about 3 % foreign fat added. PLS seems to offer the lowest detection-limits (of about 2%) of the methods compared.     

Scanner Digital Images Combined with Color Parameters: A Case Study to Detect Adulterations in Liquid Cow’s Milk

This study presents a proposition of a low-cost screening method for detection and quantification of adulterations in liquid cow’s milk samples. The studied adulterations were made with water and NaOH. Digital images from the adulterated samples were obtained using a flatbed scanner, and the means of ten color parameters were used to evaluate the information from images: red, green, blue, hue, saturation, value, relative colors (r, g, and b), and intensity. Regression models for water quantification were proposed using multiple linear regression (MLR), principal components regression (PCR), and partial least squares (PLS). The best models were obtained using PCR and PLS, with root mean square error for prediction smaller than 7%. These results were compared with near-infrared (NIR), and the prediction capability was very similar. In the case of adulterations with NaOH, the colors B, S, g, and b presented the highest differences between fresh and adulterated milk samples.