Combination of LF-NMR and BP-ANN to monitor water states of typical fruits and vegetables during microwave vacuum drying

Highlights

• A rapid real-time nondestructive detection method of combination of LF-NMR and BP-ANN was set up.

• The information of A_total and T₂₃ can be used to analyze the drying behavior.

• The information of A₂₀, A₂₁ and A₂₂ can be used as the fingerprint characteristics of material discrimination.

• The BP-ANN model of mixed species had excellent predictive performances with R² 0.9969 and RMSE 0.0184.

Abstract

To set up a rapid real-time nondestructive detection of moisture content, this paper reported the results of a combination of LF-NMR and BP-ANN to monitor the relationship between drying parameters and state of water under different microwave vacuum drying conditions. Three kinds of materials, carrot (fruit), banana (vegetable) and pleurotus eryngii (edible fungus), were tested in the experiment of applicability. The resulted showed that the information of A_total and T₂₃ can be used to analyze the drying behavior and the information of A₂₀, A₂₁ and A₂₂ can be used as the fingerprint characteristics of material discrimination. Three classic models (PLS, SVM and BP-ANN) were compared to study the prediction ability of moisture content with the inputs of A₂₀, A₂₁, A₂₂, A₂₃ and A_total. The performance of BP-ANN model was the best. Although the BP-ANN model of mixed species was not as good as the BP-ANN model of single fruit or vegetable, it still had excellent predictive performances with R² 0.9969 and RMSE 0.0184 to meet the needs of current industry and production.

Introduction

As one of the most important drying technologies (Zhang, Tang, Mujumdar, & Wang, 2006), microwave vacuum drying has been widely used in many fields, especially in the drying of heat sensitive products (Lagnika et al., 2018; Lv, Zhang, Wang, & Adhikari, 2018; Sun, Zhang, & Mujumdar, 2018). However, in the face of severe global energy and labor cost challenges, new drying technologies such as high-efficiency, energy-saving, intelligent and precise have become a major demand for the transformation and upgrading of the global drying industry. The difficulty of online detection and low level of intelligence restricts the development of microwave vacuum technology in fruit and vegetable food drying (Sun, Zhang, Mujumdar, & Yang, 2019; Zhang et al., 2006). For example, the lack of real-time online detection technology for moisture content of materials in the drying process makes it difficult to know the drying state of materials and judge the drying endpoint, resulting in poor product quality and high energy consumption (Duan et al., 2013). What's more, too few drying equipment functions, mainly rely on artificial judgment and control feedback, automatic intelligence level to be improved. At present, the integrated control technology of whole-process drying based on artificial intelligence has become a research hotspot.

It is very important to know the water state of materials for the control of drying process. Low field nuclear magnetic resonance technology (LF-NMR) is a nondestructive rapid detection technology developed according to the proton energy level transition (Tylewicz et al., 2016; Yan et al., 2016). With the advantages of high sensitivity, strong pertinence and short detection time (Hu, Lan, Zhang, & Xie, 2017), LF-NMR has been widely used in the field of food science (Fan & Zhang, 2018), such as quality discrimination and classification (Corsaro et al., 2016; Ribeiro et al., 2014; Santos, Pereira, & Colnago, 2016), composition analysis (Carneiro et al., 2016; Cheng, Zhang, Adhikari, & Islam, 2014; Tao & Ngadi, 2017), quality detection (Fundo et al., 2016; M. Y.; Li et al., 2014; Soares et al., 2017; Xin, Zhang, & Adhikari, 2013), especially online monitoring of moisture content during materials drying process (Gudjonsdottir, Arason, & Rustad, 2011; Lv et al., 2018; Sun et al., 2019). At the point of fruit and vegetable drying, Li found that the signal intensity of free water peak (A₂₃) had good a correlation with dielectric properties in Chinese yam (L. Li, Zhang, & Yang, 2019) and moisture content in apple (L. Li, Zhang, Bhandari, & Zhou, 2018). Using LF-NMR, Lv (Lv, Zhang, Bhandari, Li, & Wang, 2017) tried to design a smart device for online measurement of moisture state and six kinds of vegetables were tested with high accuracy (P > 0.950). Combined with magnetic resonance imaging (MRI), LF-NMR was also used to investigate the distribution of moisture during different drying methods (Lv et al., 2018; Lv, Zhang, Bhandari, Yang, & Wang, 2016).

The application of artificial neural network (ANN) was comprehensively introduced in food drying from aspects of kinetic model, physicochemical properties and quality analysis, intelligent control design (Sun et al., 2018). Generally speaking, ANN is used to predict product indicators (moisture content, color, nutrition et al.) with the inputs of drying parameters (power, time, temperature et al.)(Guiné, Cruz, & Mendes, 2014; Mahjoorian et al., 2017). With four inputs of mass, temperature, thickness and drying time, Husna (Husna & Purqon, 2016) used a BP-ANN model to predict the moisture content of durian slice. In Mortaza's study (Aghbashlo, Mobli, Rafiee, & Madadlou, 2013), the spraying air flow rate, aspiration rate, peristaltic pump rate, drying air temperature, and outlet air temperature were used to predict the performance indices of capsules' residual moisture content, particle size, bulk density, encapsulation efficiency, and peroxide value. The topology structure of ANN was also studied to improve the accuracy of prediction. Sun discussed the influence of topology parameters of transfer function, training function, and the number of neurons on ANN performance (Sun et al., 2019). In Khawas's study (Khawas, Dash, Das, & Deka, 2015), the BP-ANN optimized by genetic algorithm was found more accurate in prediction of various banana parameters during drying. Some researchers made attempts to apply ANN to the automation and intelligent control of industry. Nadian (Nadian, Abbaspour-Fard, Martynenko, & Golzarian, 2017) designed an intelligent integrated control of hybrid hot air-infrared dryer based on ANN combined with fuzzy logic technology. Evaluated for kiwifruit drying, the system exhibited excellent performance between product quality and energy consumption.

Through LF-NMR and ANN were widely used in food drying, the combination technology based on LF-NMR and ANN is rarely used in microwave vacuum drying for fruits and vegetables. The main objectives of this research were to establish a fast real-time nondestructive technique based on combined LF-NMR and ANN to provide technical support for automation and intelligence of drying equipment. Three kinds of typical fruits and vegetables (carrot, banana and pleurotus eryngii) were tested, and the performances of different type of BP-ANN models such as single fruit and vegetable type model and mixed type model were studied.