Development of PAT tools using guided microwave spectroscopy and chemometrics for meat and dairy processing applications Ming Zhao,¹ Bhavya Panikuttira,¹.

Slides:



Advertisements
Similar presentations
11-1 Empirical Models Many problems in engineering and science involve exploring the relationships between two or more variables. Regression analysis.
Advertisements

PCA for analysis of complex multivariate data. Interpretation of large data tables by PCA In industry, research and finance the amount of data is often.
Regression analysis Relating two data matrices/tables to each other Purpose: prediction and interpretation Y-data X-data.
Analysis of Variance Outlines: Designing Engineering Experiments
Quantifying soil carbon and nitrogen under different types of vegetation cover using near infrared-spectroscopy: a case study from India J. Dinakaran*and.
Introduction The use of qNMR for purity measurement has been steadily growing in recent years. The assessment of the purity of calibration materials and.
1 Two methods for modelling the propagation of terahertz radiation in a layered structure. GILLIAN C. WALKER 1*, ELIZABETH BERRY 1, STEPHEN W. SMYE 2,
A NEW PERSPECTIVE TO VISIBLE NEAR INFRARED REFLECTANCE SPECTROSCOPY: A WAVELET APPROACH Yufeng Ge, Cristine L.S. Morgan, J. Alex Thomasson and Travis Waiser.
Characterization of Soil Shrink-Swell Potential Using the Texas VNIR Diffuse Reflectance Spectroscopy Library Katrina M. Hutchison, Cristine L.S. Morgan,
Zhaohua Wu and N. E. Huang:
Elaine Martin Centre for Process Analytics and Control Technology University of Newcastle, England The Conjunction of Process and.
A Simulation Model for Predicting the Potential Growth of Salmonella as a Function of Time, Temperature and Type of Chicken Thomas P. Oscar, Agricultural.
,. Sugar measurements in soybeans using Near Infrared Spectroscopy Introduction  Soluble carbohydrates are the third compound of soybeans by weight (11%),
Model-based Classification in Food Authenticity Studies D. Toher 1,2, G. Downey 1 and T.B. Murphy 2 Presented by: Deirdre Toher 1 Ashtown Food Research.
SPECTRAL AND HYPERSPECTRAL INSPECTION OF BEEF AGEING STATE FERENC FIRTHA, ANITA JASPER, LÁSZLÓ FRIEDRICH Corvinus University of Budapest, Faculty of Food.
Multipurpose analysis: soil, plant tissue, wood, fruits, oils. Benchtop, portable Validation in-built, ISO compliant Little or no sample preparation. Rapid.
AE 469/569 TERM PROJECT DEVELOPING CORRECTION FUNCTION FOR TEMPERATURE EFFECTS ON NIR SOYBEAN ANALYSIS Jacob Bolson Maureen Suryaatmadja Agricultural Engineering.
Walloon Agricultural Research Center Walloon Agricultural Research Center, Quality Department Chaussée de Namur, 24 – 5030 GEMBLOUX - Tél :++ 32 (0) 81.
Permeation is the passage of contaminants through porous and non-metallic materials. Permeation phenomenon is a concern for buried waterlines where the.
CEMTREX, INC. MODCON SYSTEMS LTD NIR Systems July 2010.
Expression profiling of peripheral blood cells for early detection of breast cancer Introduction Early detection of breast cancer is a key to successful.
Effects of Inulin on Rheological Attributes of Processed Cheese Effects of Inulin on Rheological Attributes of Processed Cheese Rahul Patel*, Hans Zoerb*,
Blue: Histogram of normalised deviation from “true” value; Red: Gaussian fit to histogram Presented at ESA Hyperspectral Workshop 2010, March 16-19, Frascati,
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. 1 Part 4 Curve Fitting.
Analytical approaches for 1 H NMR wide line spectra of Soil Organic Matter Alexander Jäger 1, Marko Bertmer 1, Gabriele E. Schaumann 2 References Introduction.
BIO-PROCESS ENGINEERING GROUP, Dept. Agricultural & Bioresource Engineering, U of S 2008 CSBE International Meeting Microwave drying characteristics of.
Food Quality Evaluation Techniques Beyond the Visible Spectrum Murat Balaban Professor, and Chair of Food Process Engineering Chemical and Materials Engineering.
Chapter 4 Linear Regression 1. Introduction Managerial decisions are often based on the relationship between two or more variables. For example, after.
Use of spectral preprocessing to obtain a common basis for robust regression 5 spectral preprocessing combinations gave significantly higher RPDs (α =
1 11 Simple Linear Regression and Correlation 11-1 Empirical Models 11-2 Simple Linear Regression 11-3 Properties of the Least Squares Estimators 11-4.
36th ICAR Session and Interbull Meeting Niagara Falls, June 2008 Potential Estimation of Minerals Content in Cow Milk Using Mid- Infrared Spectrometry.
STATISTIC MODELING OF RESULTS IN CIVIL ENGINEERING Częstochowa, 2004.
REVIEW ON PROJECT WORK Measurement… Prof. Andras Fekete Department of Physics and Control Corvinus University of Budapest.
Rapid, On-site Identification of Oil Contaminated Soils Using Visible Near-Infrared Diffuse Reflectance Spectroscopy Chakraborty, S. 1, D. Weindorf 1,
Measuring Soil Properties in situ using Diffuse Reflectance Spectroscopy Travis H. Waiser, Cristine L. Morgan Texas A&M University, College Station, Texas.
Data collection  Triticale samples from 2002 to 2005 (Iowa, USA).  Foss Infratec™ 1241 (transmittance instrument).  Crude protein analysis by AACC Method.
Standardization of NIR Instruments: How Useful Are the Existing Techniques? Benoit Igne Glen R. Rippke Charles.
A WIRELESS PASSIVE SENSOR FOR TEMPERATURE COMPENSATED REMOTE PH MONITORING IEEE SENSORS JOURNAL VOLUME 13, NO.6, JUNE 2013 WEN-TSAI SUNG, YAO-CHI HSU Ching-Hong.
Table of Contents Chapter 1 Introduction to Physical Science Chapter Preview 1.1 What Is Physical Science? 1.2 Scientific Inquiry 1.3 Measurement 1.4 Mathematics.
COMPARATIVE STUDY BETWEEN NEAR- INFRARED(NIR) SPECTROMETERS IN THE MEASUREMENT OF SUCROSE CONCENTRATION.
APPLICATION OF CLUSTER ANALYSIS AND AUTOREGRESSIVE NEURAL NETWORKS FOR THE NOISE DIAGNOSTICS OF THE IBR-2M REACTOR Yu. N. Pepelyshev, Ts. Tsogtsaikhan,
Potential of Hyperspectral Imaging to Monitor Cheese Ripening
Stats Methods at IC Lecture 3: Regression.
Random signals Honza Černocký, ÚPGM.
Chapter 4 Basic Estimation Techniques
Novel desalting method for protein recovery from fat rendering waste water. Functional properties of recovered proteins. Carlos Álvarez1, Liana Drummond1,
M.L. Amodio, F. Piazzolla, F. Colantuono, G.Colelli
Term project for the coursework AE 569
,Branko Vranic1, Nada Tarek2 ,V. Frost 3, G. Betz1
Predicting extractive content of Eucalyptus bosistoana heartwood by near infrared spectroscopy Yanjie Li and Clemens.
Matteo Reggente Giulia Ruggeri Adele Kuzmiakova Satoshi Takahama
ELECTROMYOGRAPHICAL COMPRESSION SHORTS TO PREDICT LACTATE THRESHOLD
RAPID DETERMINATION OF PROTEIN CONTENT IN PROTEIN POWDER FINISHED PRODUCT USING NEARINFRARED (NIR) Roney Christiana, Yanjun Zhangb, Kan Heb, Piyush Purohita,
Authors: Asa Gholizadeh, Nimrod Carmon, Aleš Klement, Luboš Borůvka, Eyal Ben-Dor, Sabine Chabrillat Contact: Effects of Measurement.
D. Varga, A. Szabó, L. Locsmándi, Cs. Hancz, R. Romvári
Louise Fortunato, Sulaf Assi, Paul Kneller and David Osselton
Interval selection complexity
E.V. Lukina, K.W. Freeman,K.J. Wynn, W.E. Thomason, G.V. Johnson,
Statistical Methods For Engineers
Introduction to Instrumentation Engineering
Agricultural Sustainability Through Cover Crops Andres Tapia, Dr
Deyuan Kong and Sara Mcmillen, Chevron Energy Technology Company USA
Relationship between mean yield, coefficient of variation, mean square error and plot size in wheat field experiments Coefficient of variation: Relative.
Estimating the nutritional quality of milk fat in cow milk
J/   analysis: results for ICHEP
Volume 5, Issue 4, Pages e4 (October 2017)
Presentation of the main new features in
© The Author(s) Published by Science and Education Publishing.
14 Design of Experiments with Several Factors CHAPTER OUTLINE
Volume 109, Issue 7, Pages (October 2015)
Presentation transcript:

Development of PAT tools using guided microwave spectroscopy and chemometrics for meat and dairy processing applications Ming Zhao,¹ Bhavya Panikuttira,¹ and Colm O’Donnell¹ ¹ UCD School of Biosystems and Food Engineering & UCD Institute of Food & Health, University College Dublin, Dublin 4, Ireland Introduction The adoption of a process analytical technology (PAT) approach in food processing would facilitate real-time process monitoring. In this study the potential of guided microwave spectroscopy (GMS) in combination with chemometric methods was investigated for the development of PAT tools for meat and dairy processing applications. The current study on GMS was carried out in a static manner to mimic the relative static measurement moments during in-line monitoring. Specifically the potential of GMS technology to predict the fat and moisture composition of ground beef and milk samples was investigated using microwave spectra recorded from 100 to 2200 MHz at 4 MHz increments. Partial least squares regression (PLSR) models were developed using a modified ensemble Monte Carlo uninformative variable elimination (EMCUVE) procedure PLS to predict both fat and moisture content of ground beef; PLSR models based on all the spectral variables and variable importance on projection (VIP) selected spectral variables were also developed to predict the coagulum strength of cheese milk by storage modulus (G′ (Pa)). Figure 1. The ε scan in-line process analyser (Thermo Fisher Scientific Inc., Takkebijsters Breda, The Netherlands). Figure 2. The general work flow. Methods & Materials Study I. Forty five ground beef samples (2.2 kg per sample) were measured using the ε scan in-line process analyser (Thermo Fisher Scientific Inc., Takkebijsters Breda, The Netherlands) under the static condition at constant density. Each sample was compressed into a ‘U’ shaped stainless steel mould designed to produce a sample size fitting to the measurement field (7.5×10×22.6 cm³) in the chamber of the analyser to form a brick shape with density of 1497 kg/m³. Guided microwaves were generated over 100-1800 MHz by the electro-magnetic field and trespassed the distance between the transmitter and the acceptor (Figure 3). Study II. Cheese milk samples were prepared in three different coagulation conditions by adding rennet [0.20 g , 0.28 g and 0.36 g] into 1000 ml of low fat milk respectively. For each sample, GMS measurements were taken at 5 th , 15 th, 25 th, 35 th, 45 th and 55 th minutes after the addition of rennet with constant temperature control (~ 32 °C). Guided microwave spectral data over 100-2200 MHz were collected (Figure 4). All the electronic signals were collected by the acceptor and saved as data form in the embedded computer system. For calibration of the system, 12 normalised spectra were collected for each sample and saved as .csv files. These data were imported into Matlab 2014a (The Mathworks, Natick, MA, USA) for chemometric approaches. For ground beef samples, Partial least squares regression (PLSR) models were developed on the full range of spectral data (X) and chemical reference data (Y) for the predictions of both fat and moisture content in ground beef. In-house produced samples (n=36) were used as the calibration data set while the commercial samples (n=9) formed a validation set. A modified ensemble Monte Carlo uninformative variable elimination (EMCUVE) was also employed to improve the performance of all PLSR models and to increase the consistency of prediction results using the most informative spectral variables (Esquerre et al., 2011). For cheese milk, PLSR models were developed using leave-one-out cross validation of the whole sample set for the prediction of G’ (Pa) of cheese milk coagulation conditions; the most informative spectral variables were decided using variable importance on projection (VIP). Figure 3 a) a ‘U’ shaped stainless steel mould designed to adapt the size of the measurement field (7.5×10×22.6 cm3) of the ε scan analyser chamber and the moulded ground beef samples; b) Bench measurement using the ε scan analyser: 1-measurement chamber, 2- transmitter, 3-receiver, 4-temperature probe. Figure 4 Bench measurement of the ɛ scan analyser on cheese milk: 1- measurement chamber, 2-transmitter, 3- receiver, 4-temperature probe. Results Study I. For ground beef samples, PLSR models were developed using full spectral variables (n= 226) over the microwave frequency range of 100-1800 MHz (Table 1). PLSR models were also developed using the least number of spectral variables (i.e. n=16, 17 or 22 and n= 72, 73 or 74 for fat and moisture contents respectively (Table 2 ) which were decided by EMCUVE with a threshold value around 1.6-1.8. For both fat and moisture content prediction, the retained informative spectral variables were almost fixed. Generally, less than 4 or 6 PLS loadings were involved to develop prediction models. The values of R²CV and R²P were stabilized at 0.93 and 0.90 respectively for calibration and prediction of fat content, while their corresponding values of RMSECV and RMSEP were around 2.13-2.18 %w/w and 1.72-1.83 %w/w. For the moisture content prediction, the values of R²CV and R²P were fixed at 0.89 and 0.82 with a RMSECV of 1.93%w/w and a RMSEP of 1.77 %w/w. The best performed PLSR model were highlighted in red colour in Table 2 and shown in Figure 5. Figure 6. a) GMS spectra of low fat milk and milk samples with rennet additions measured at 5th min of coagulation; mean spectrum at each measurement time of b) Trial 1 (0.28 g rennet in 1000 ml low fat milk), c) Trial 2 (0.36 g rennet in 1000 ml low fat milk), d) Trial 3 (0.20 g rennet in 1000 ml low fat milk). Figure 5. GMS mean spectrum of ground beef with spectral variable selection using EMCUVE for a) fat content prediction; d) water content prediction; regression coefficient plots of the seclected variables for b) fat content; e) moisture content; PLSR plots based on microwave spec tral variables using EMCUVE - predicted value vs. reference value: c) fat content ; f) moisture content. Table 1. Summary of PLSR model performances (ɛ scan – 100-1800 MHz) for fat and moisture content in ground beef. Figure 7. Rheometer data (G’ (Pa)) over milk coagulation time (min). Table 3. Summary of PLSR model performances for the prediction of storage modulus (G’ (Pa)) in cheese milk.   ɛ scan (0-1800 MHz) cal. Set n=36 val. Set n=9 (spectral variables, v=226) PLS loadings R²CV RMSECV R²P RMSEP Bias fat normalisation 6 0.91 2.63 1.56 -0.35 normalisation+ SNV 3 0.87 3.12 0.89 1.77 0.56 normalisation+ MSC 1 0.82 3.59 0.73 2.78 -0.99 moisture 4 0.86 2.21 0.85 2.34 0.02 normalisation+SNV 0.84 2.18 2.36 1.55 -0.58 ɛ scan   cal. set n=15 Data pre-treatment Spectral variables retained PLS loadings R²C RMSEC R²CV RMSECV Bias normalisation (norm.) 460 5 0.93 2.63 0.74 5.4 -0.293 8 4 0.94 2.49 0.81 4.54 -0.139 norm. + SNV 0.97 1.59 0.92 3.01 -0.101 6 0.96 1.87 2.98 -0.038 norm. + MSC 0.98 1.36 2.64 -0.169 2.04 0.91 2.92 0.003 Study II. Spectral plots of cheese milk samples of 3 trials ( i.e. 0.28 g (T1) , 0.36 g (T2) and 0.20 g (T1) of rennet in 1000 ml of milk ) obtained at coagulation time points were shown in Figure 6. The measurements of rheological parameter (G’) were shown in Figure 7. These results were used as reference values (Y) of PLSR modelling. In the summary of PLSR results (Table 3), the best performed PLSR prediction model was developed using normalization and multiplicative scatter correction (MSC) pre-treated spectral data over 100-2200 MHz. The value of R²CV was at 0.95 with a RMSECV of 2.28 G’(Pa) (Figure 8a). The model developed using four spectral variables at 592, 940, 1340, 1984 MHz selected using VIP algorithm also shows satisfactory performance (R²CV-0.91, RMSECV- 2.92) of G’ (Pa) prediction (Figure 8b). a) b) Table 2. Summary of PLSR model performances (ɛ scan spectral variables retained using EMCUVE) for fat and moisture content in ground beef.   ɛ scan cal. Set n=36 val. Set n=9 Data pre-treatment with EMCUVE Spectral variable retained PLS loadings R²CV RMSECV R²P RMSEP Bias fat normalised (v=22) 22 6 0.93 2.18 0.90 1.79 -0.46 normalised (v=17) 17 2.13 0.89 1.83 -0.51 normalised (v=16) 16 4 1.72 -0.33 moisture normalised(v=72) 72 1.93 0.82 1.77 -0.05 normalised (v=74) 74 normalised (v=73) 73 Figure 8. PLSR plots for G’ (Pa) prediction based on normalization and MSC pre-treated a) 460 spectral variables and b) 4 spectral variables . References Conclusion Esquerre, C., Gowen, A. A., Downey, G., & O’Donnell, P. C. (2011). Selection of variables based on normalised partial least squares regression coefficients in an ensemble Monte Carlo procedure. Journal of Near Infrared Spectroscopy. 19, 343-350. This study demonstrated the potential of GMS for PAT applications in meat and dairy processing. The results also showed that spectral variable selection method can efficiently eliminate non-informative variables in order to improve consistency of model prediction and can also be implemented to a low-cost GMS system design. Acknowledgement The authors acknowledge funding for this work from the Irish Department of Agriculture, Food and the Marine under the Food Institutional Research Measure (FIRM) programme over 2013-2017.