Modeling the effect of vibration on the quality of stirred yogurt during transportation

Abstract

When transporting yogurt, vibrations and sharp movements can damage its quality. This study developed a model to connect the changes in yogurt quality with the transportation distance as simulated by the total number of vibrations. Linear regression analysis showed that there was a significant negative correlation between the water holding capacity and hardness of the yogurt over the same transport distance (p < 0.05). The yogurt vibration model was established by combining principal component analysis with a Back-Propagation Artificial Neural Network model. The number of training iterations was 2669, with a correlation coefficient of 0.96611, indicating that the model was reliable. The optimal transportation distance was determined to be within the range from 20 rpm for 8 h to 100 rpm for 4 h.

Introduction

Yogurt, with its unique sensory characteristics is popular with consumers because of its beneficial nutritional functions and good flavor (Cheng, 2010). With the continuous improvement of the life quality in China and the rapid growth of yogurts increased, the fresh milk bar is emerging in China in recent years, and the transportation radius of cod-chain milk products such as yogurt is undergoing an increasing widen. However, it is challenge to implement testing and control of yogurt quality during long-distance transportation. Texture is one of the most predominant factors that affect the quality and marketability of yoghurt as well as the flavor, which directly exhibits yoghurt mouth feel, flow behavior and appearance (Jaworska et al., 2005; Soukoulis et al., 2007). Yoghurt textural properties of gels can be assessed by texture profile analysis, oscillatory shear, rotational viscometry and various sensory methods. Textural defects include wheying-off, weak body, and lumpiness (Lucey, 2004). Many parameters such as starter culture, milk base composition, homogenization conditions, incubation temperature, heat treatment, stabilizers and postmanufacture handling have been well known to influence texture formation of yoghurt (Hassan et al., 2003; Shaker et al., 2001). Previous research reported that pumping operation caused textural defects during yoghurt fermentation, filling and storage (Lucey, 2004). It was believed to be the resultant vibration due to pumping processing to disturb milk gels. Vibration impact on gel textural properties during yogurt fermentation was systematically analyzed and stirred yogurt exhibited a reduced water-holding capacity and product viscosity with large gel particles resulted from vibration during gel formation (Körzendörfer et al., 2018). However, vibration and shock inevitably occurred and disrupted yogurt textural quality during haulers or trucks transportation (Das et al., 2019). During yogurt transportation along the cold chain from manufacturer to retailer, yogurt can inevitably be damaged by poor handling and sharp movements. This treatment leads to the separation of the yogurt whey, the destruction of the curd texture, and deterioration in the physical and chemical indices and microstructure (Soukoulis et al., 2007). These changes of flavor, texture and taste of the yogurt and so can directly affect consumer acceptance and preference. Many efforts e.g. improving starter culture, optimizing fermentation process and post manufacture treatment have been performing to reduce the impact of these risk factors on the quality of yogurt (Das et al., 2019; Petchwattana and Naknaen, 2016). The pH value of yogurt samples has been found to change little under different storage conditions at different storage times (Salvador and Fiszman, 2004). Similarly, no color changes or color differences were detected in set-type yogurts during storage (Mani-López et al., 2014). Regarding the effect of different storage temperatures on the number of possible lactic acid bacteria in yogurt, and the effect on its quality, Zhang et al. concluded that the best storage temperature was 4 °C. During storage, the pH value and titratable acidity of yogurt was found to fluctuate mildly, with the L. plantarum viable counts remaining stable (Li et al., 2017). Regarding shelf life, Cai (2012) developed a shelf life prediction model for yogurt during storage, and Sofu and Ekinci (2007) used artificial neural networks (ANN) to model the data obtained to predict the shelf life of yogurt. An artificial neural network (ANN) is a computer model based on the principle which a highly interconnected system of simple processing elements can disclose the complex interrelationships between independent and dependent variables (Devillers, 1996). The most popular ANN is the back-propagation ANN (BP-ANN) in numerous industrial and scientific areas. For example, BP-ANN has been proposed to predict the particle size distribution of nano materials (Fu et al., 2005; Khanmohammadi et al., 2010). In this work, BP-ANN was applied to propose the vibration model during yogurt transportation. Up to now, the effect on yogurt quality during transportation of the distance travelled has been little studied, and the contradictory requirements of the demands of consumers and manufacturers are yet to be resolved. The impact of transportation on the quality of yogurt has become a problem for many dairy companies. The present study aims to simulate transport vibration through a laboratory simulation, and to explore the changes in the physical and chemical properties of yogurt after vibration treatment. A yogurt vibration model will then be created using MATLAB software to predict the quality of yogurt.

Materials and methods

Materials

Skim milk powder was provided by New Hope Shuangfeng Dairy Co. Ltd. (Hangzhou, China) Sucrose was purchased from supermarkets and the Yo-Mix 495 starter culture from Danisco Group (Epernon, France). The other reagents used were of analytical grade.

Yogurt preparation

The yogurt was produced as follows: milk powder was reconstituted to the same composition as liquid milk [milk powder, deionized water, sucrose 6% (w/w)], then sterilized (5 min at 95 °C). After cooling the milk to 40 °C, bacteria (0.08% w/w) were added then the mixture was incubated for 8 h at 40 °C. After stirring, the product was put in 100 mL bottles then placed in a refrigerator at 4 °C for 3 h, to eventually become the finished samples for the following experiments.

Simulated vibration conditions

The yogurt samples were placed on a vibration test table produced by Gaotian Experimental Equipment Co. Ltd. (WuHan, HuBei, China) in an incubator at 4 ± 1 °C. The vibration frequency was set at 40, 100, or 160 rpm, for periods of 0, 2, 4, or 6 h. This enabled the effects of different vibration frequencies on the simulation of yogurt during road transportation to be studied.

Testing the physicochemical indicators of the quality of the yogurt samples

Hardness and viscous force

Yogurt samples (50 mL each cup) were used to determine the hardness and viscous force. The specific parameters were set up on Texture analyzer (TX.XT Plus.Stable Microsystems Ltd, Godalming, UK) as described by Wang: a 35 mm diameter pressure plate with a test speed of 1.0 mm/s, a test depth of 30.0 mm, and a rebound speed of 10.0 mm/s. The maximum value of the positive region of the curve from the analyzer represented hardness, and the area of the negative region of the curve, viscous force.

Acidity

The acidity of the yogurt was measured using GB 5009.239-2016 (Chinese Standard, 2016).The acidity of the yogurt samples was determined by titrating a yogurt: deionized water mixture (1:4, v/v) using 0.1 M NaOH solution. Phenolphthalein was used as the indicator for the acidity measurement. The titration acidity (°T) of the yogurt sample was expressed as the volume (mL) of NaOH titrated to 100 mL of the yogurt.

Water holding capacity

Five g of yogurt was weighed accurately (W₁) then centrifuged at 5400×g for 10 min by applying MR 23i high-speed refrigerated centrifuge produced by Shanghai Xintian Biotechnology Co. Ltd. (Shanghai, China). After standing for 1 min, the supernatant was weighed (W₂). The formula for calculating the water holding capacity (WHC) of the yogurt sample was as follows:

$$\text{WHC \%}=\left(W_1-W_2\right)/W_1\times 100$$ 1

where W₁ is the sample weight, and W₂ the weight of the supernatant liquid.

Viscosity

The viscosity of the yogurt was measured using an NDJ-8s digital rotary viscometer (US Solid, Cleveland, OH, USA) applying a No. 62 rotor with the rotating speed of 1.5 rpm, a measurement time of 30 s, and a measurement temperature of 4 °C. Record data after five dynamic tests.

Sensory evaluation

Sensory evaluation was conducted according to the method developed by the International Dairy Federation (Karagül-Yüceer et al., 2006). The samples were rated by ten assessors (5 males and 5 females at the average age of 24) who had undergone the professional training required to fully understand the sensory characteristics of yogurt, and were familiar with the evaluation criteria for yogurt. The panelists needed to judge the following sensory attributes, color (milky white level), aroma (typical yogurt fragrant), taste (Good for sweet and sour), and organizational status (smooth mouth feel for texture) of yogurt. The excellent acceptability of yogurt was set 100 points including color (20 points), aroma (20 points), taste (30 points) and organizational status (30 points). Each sensory attribute was rated to describe hedonic scale where the highest score indicates the highest quality and the lowest score represents the lowest quality. During the evaluation process, the assessors rinsed their mouths between samples with warm water to ensure the accuracy of the experimental data. The four sensory attributes were weighted to provide a total sensory score.

Model building

Back-Propagation Artificial Neural Network (BP-ANN) is a well-known multi-layered technique for building models. The network parameters needing to be set are the number of input and output neurons, the number of layers in the multi-layer network, and the number of neurons in the hidden layer. During training, the initialization of the network also needs to be considered along with the practical problems. Khanmohammadi et al. (2009) found that the ANN model was reliable for modeling protein content in yogurt, while still being rapid and simple, with no sample preparation steps.

Principal component analysis (PCA) recombines multiple indicators which are correlated with each other into a new set of non-correlated comprehensive indicators. When combining the principal components and a BP-ANN to construct the yogurt vibration model, PCA reduces the dimensionality of the original data then the BP -ANN model is constructed to reduce the average relative error (Yang, 2016). At first, six indicators (Five physicochemical indicators and the total sensory score) are set as input layers (Fig. 1), but after processing by PCA, two indicators are obtained and the total sensory score as input layers, because the PCA-BP-ANN can not only achieve faster convergence speed but also reduce the average relative error and improve the accuracy of the evaluation results. Then constructed and connected through the functions of the hidden layer, to finally output the greatest transport distance (in terms of total number of vibrations) while retaining the quality of the yogurt.

Fig. 1.

Neural network structure

Data analysis and processing

The experimental data were mainly subjected to one-way analysis of variance (ANOVA) using SPSS 22 software (IBM Corp., Armonk, NY, USA), with individual sample means being compared (at a significance level of p ≤ 0.05) using the Duncan multiple comparison test. The yogurt vibration model was established using MATLAB software (Version 2014b, The Mathworks, Natick, MA, USA) and the analysis was carried out using Origin 8 software (OriginLab Corp., Northampton, MA, USA).

Results and discussion

Yogurt sensory evaluations

The yogurt samples were vibrated for 6 h at three different frequencies (40, 100, and 160 rpm) then scored by ten assessors. Table 1 showed that as the vibration frequency increased, the overall score gradually decreased. The vibration action significantly decreased the taste of the yogurt and obviously deviated consumers’ expect to yogurt great taste. Sensory evaluation can directly reflect the feelings of consumers, but the individual differences are relatively large, which can well reflect the different tastes of consumers.

Table 1.

Effect of vibration frequency on the sensory attributes score of yogurt

Vibration frequency (rpm)	Sensory attributes
	Color (20 points)	Aroma (20 points)	Taste (30 points)	Organizational status (30 points)	Total score (100 points)
40	18.42 ± 1.55a	17.50 ± 1.45a	22.64 ± 4.09a	25.28 ± 3.20a	83.85 ± 6.38a
100	17.73 ± 2.05a	16.60 ± 1.80b	22.33 ± 3.85a	24.60 ± 3.33a	81.26 ± 7.25b
160	17.80 ± 1.55a	16.50 ± 1.27b	20.50 ± 5.78b	24.90 ± 4.53a	79.70 ± 10.50c

Values in the same row followed by the different superscript letter means the significant difference (p < 0.05) and the same superscript letter does not differ significantly according to the Duncan test

(1) Color: 0 = yellow to 20 = milky white; (2) Aroma: 0 = peculiar smell to 20 = Very fragrant; (3) Taste: 0 = bad taste to 30 = great taste; (4) Organizational status: 0 = extremely grainy to 30 = very smooth

Based on the five physical and chemical indicators of yogurt quality in Table 2, linear regression analysis has shown a significant negative correlation between water holding capacity and hardness and the same transportation distance (an equal number of vibrations) (p < 0.05). This indicated that the gel structure of yogurt was damaged during transportation, which would affect its quality.

Table 2.

Effect of vibration frequency on five physical and chemical indicators of yogurt quality

Vibration frequency (rpm)	Time (h)	Acidity (°T)	WHC (%)	Viscosity (cP)	Hardness (N)	Stickiness (N)
40	0	86.670 ± 1.08	0.266 ± 0.01	1977 ± 1.02	0.276 ± 0.01	0.894 ± 0.01
	2	87.239 ± 2.15	0.251 ± 0.01	1917 ± 2.21	0.253 ± 0.02	0.726 ± 0.02
	4	87.507 ± 0.80	0.247 ± 0.02	2543 ± 1.36	0.235 ± 0.01	0.774 ± 0.01
	6	88.336 ± 2.30	0.239 ± 0.01	2540 ± 0.65	0.232 ± 0.01	0.759 ± 0.02
100	0	73.066 ± 0.94	0.259 ± 0.01	2055 ± 0.79	0.277 ± 0.01	0.702 ± 0.01
	2	73.590 ± 0.22	0.236 ± 0.01	1593 ± 0.58	0.228 ± 0.01	0.867 ± 0.02
	4	74.073 ± 1.62	0.220 ± 0.01	2070 ± 1.69	0.222 ± 0.02	0.691 ± 0.01
	6	75.065 ± 2.43	0.224 ± 0.01	2495 ± 0.54	0.212 ± 0.02	0.686 ± 0.01
160	0	77.603 ± 1.62	0.260 ± 0.01	2115 ± 2.06	0.273 ± 0.02	0.465 ± 0.02
	2	77.484 ± 0.95	0.238 ± 0.00	1380 ± 1.27	0.219 ± 0.01	0.611 ± 0.02
	4	79.339 ± 0.61	0.226 ± 0.02	1650 ± 0.82	0.211 ± 0.01	0.742 ± 0.01
	6	81.995 ± 1.11	0.215 ± 0.02	1139 ± 0.69	0.207 ± 0.01	0.644 ± 0.01

Analysis of the physical and chemical indicators of yogurt quality

Five indicators, for acidity, viscosity, hardness, WHC and stickiness, were measured at different vibration frequencies. The results showed that as the vibration time increased to 6 h under 40 to 100 to 160 rpm, the acidity of the sample increased by 1.92%, 2.74%, and 5.66% respectively compared to without vibration (Table 2). The WHC, hardness and viscosity decreased to different extents. Increasing the vibration frequency and vibration time exacerbated the damage to the yogurt structure through increasing the number of gel microspores, weakening the structural density of the protein network, and reducing the ability to bind water. This made the whey more likely to precipitate, thus decreasing the viscosity and water holding capacity of the yogurt samples (Bringe and Kinsella, 1993; Lucey and Singh, 1997). Therefore, determining the maximum transport distance without degrading the quality of the yogurt sample is critical to the transport distance and plant efficiency.

Principal component analysis (PCA)

Table 3 showed the matrix diagram for the rotating elements and the component graph analysis in the rotating space from PCA. This shows that the five physical and chemical indicators can be classified into two components—PC1 and PC2—where hardness and WHC, respectively, are the main components, with no loss of data. On the basis of this information, the dimensions of the factors can be reduced and thereby improve the quality of the data so water holding capacity and hardness were used as the principal component factors to analyze the problem.

Table 3.

Principal components 1 and 2 with loading factors from principal component analysis of the physical and chemical indicators of yogurt quality

Group	Element
	PC1	PC2
Hardness	0.873
Acidity	0.799
Viscosity	0.663
WHC		0.823
Stickiness		− 0.782

Correlation between water holding capacity, hardness and vibration frequency, distance and sensory evaluation

After vibration at different frequencies for different times (The transport distance in present work refers to the vibration time multiplied by the vibration frequency, expressed in units of r), the physicochemical structure of the yogurt changed. The WHC and hardness were significantly negatively correlated with the distance of vibrations (p < 0.05), with correlation coefficients of − 0.622 and − 0.600, respectively. The number of vibrations was then fitted against the WHC and hardness values. Figure 2(A) shows the Fourier fitting the WHC and distance using MATLAB software, R was 0.9475 and SSE was 0.00009183 and the general model Fourier equation was f(x) = 25.78 − 25.52 cos(0.00000095x) − 1.457 sin(0.00000095x). The root mean square error was extremely small, and the fitting degree was closer to 1, indicating that the model was reliable, it shown that as the number of vibrations increased, the WHC of the yogurt declined, possibly because the secondary bond of the hydrogen bond and other secondary bonds were vibrated. If damaged, gel networks are prone to moisture and binding, and free water precipitates more easily. Figure 2(B) also fit the hardness and distance by Fourier fitting, R was 0.9145, SSE was 0.0003429 and the general model Fourier equation was as followed f(x) = 2956 − 2955 cos(0.0000001461x) − 18.94 sin(0.0000001461x), so the fitting result was great. The hardness of the yogurt decreased as the number of vibrations increased in line with the transport distance, because the network structure of the yogurt had been destroyed, resulting in a looser gel.

Fig. 2.

Effect of number of distance on the WHC, hardness, score of yogurt samples with Fourier fitting

The transportation distance, as indicated by the number of vibrations, was correlated with the sensory evaluation score. However, as the judgment of each assessor varies, the error in the measurement process may lead to inconsistency. Figure 2(C) also fits the score and distance by Fourier fitting, R was 0.9167, SSE was 0.01232 and the general model Fourier equation was f(x) = 8.188 − 0.119 cos(0.0001026x) + 0.152 sin(0.0001026x), it shows that yogurt treated in the range from 20 rpm for 8 h to 100 rpm for 4 h (distance = frequency × time) had an acceptable mouth feel.

BP-ANN model prediction

Use MATLAB software to call the Excel table with data samples. The interface diagram for network training is including input-layer-layer-output, the training error curve in Fig. 3(A) and the target training curve showing in Fig. 3(B). Two indicators obtained by principal component analysis and total sensory scores were used as input layers. The number of implicit neurons is very important because it directly affects the performance of the network. Two hidden layers were used with ten neurons in the first layer, and only one in the second layer. The initial number of iterations for network training was 3000 with a correlation coefficient of 0.96611, indicating that the selected data reflected the content of the original data well. The model is Output = 0.67 * Target + 0.14. After making 2669 iterations, the training error had reached the specified value, the network training was completed, so the training network code could be saved and used for next tests. By combining principal component analysis with a BP–ANN, the sensory evaluation score was also used as an indicator to reflect the quality of the yogurt, and allowed a model of yogurt quality change and transportation distance to be constructed.

Fig. 3.

Training error and target training curve for BP-ANN

Compliance with ethical standards

Conflict of interest

All authors declare that they have no conflict of interest.