Development and tuning of automatized methods for food quality and process control in agro-food industries

In this thesis work, techniques traditionally used in Material Science, as the image- based analysis and the NIR spectroscopy, coupled to multivariate data analysis, were investigated in order to develop innovative approaches to evaluate quality parameters in agro-food products and for food process monitoring. The application of innovative approaches aims to overcome some of the disadvantages of the reference analysis, such as the time needed to prepare the samples, to carry out the analysis and to achieve the results, the specialized skills required to the operator, the invasive nature and the lack of environmental sustainability, since often solvents have to be used. Moreover, the reference analytical methods are not suitable for real-time applications and do not allow a prompt intervention of the automatic control in the case of out-of-range parameters. This thesis work is presented in four main parts. The first part (Chapter 1) gives an overview of the agro-food industry value, emphasizing the crucial role of the Material Science to improve the current analytical controls along the entire food supply chain and to develop new technologies. Furthermore, this part provides the conceptual elements, basics of control theory and some examples of the Process Analytical Technology (PAT) application in the agro-food industry. The second part (Chapters from 2 to 4) describes the fundamentals of the instrumental techniques explored in this thesis, while the third part (Chapter 5) presents the chemometric methods applied for data processing. The fourth part is the experimental section and includes three practical case studies, which make up the core of the thesis. The first case study (Chapter 6) focuses on developing an automatized method for hazelnut quality sorting, based on imaging techniques (RGB and multispectral) and the multivariate image analysis (MIA). The presented setup is an off-line method able to detect more than 90% of ‘moldy' hazelnuts (defect type 1), while the discrimination between sound and ‘pest-affected' (defect type 2) hazelnuts was performed correctly at an 80% degree. Enhanced imaging and spectroscopic techniques will be considered in future to detect hazelnut defects directly on the entire hazelnut kernel. In the second case study (Chapter 7), lab-scale drying tests were performed on in-shell hazelnuts in order to define a model of general applicability for hazelnut drying, since different cultivars and drying air parameters (Temperature and Relative Humidity) were considered. The hazelnut moisture content at the equilibrium (Me) and the drying rate (k) were calculated for each drying test by modelling the experimental data with a 2nd order kinetics. Subsequently, the contribution of the drying air parameters and the cultivar on affecting Me and k was evaluated by means of the ANOVA and the MLR analysis. In the third case study (Chapter 8), NIR spectroscopy associated with chemometrics was investigated as a fast and non-invasive method to predict the Arabica/Robusta ratio and the colour of ground roasted coffee blends. The PLS2 algorithm was applied to elaborate a model for the simultaneous determination of these two coffee parameters. A good predictive ability was achieved, with errors equal to 1.28 A.U. and 4.34% w/w for the roasting degree and for the Arabica content, respectively. The final two chapters (9 and 10) summarize the results and perspectives of the research work, highlighting that all the approaches developed and discussed show the potential to be applied for real time determination of food quality parameters and process control at the industrial scale, according to the PAT guidelines. Overall, this thesis has shown that the combination of disciplines such as imaging, spectroscopy and chemometrics can provide practical insights and control possibilities that reference analytical methodologies cannot offer, and this can be also extended to the industrial process as a whole