Evaluation of Food Fineness by the Bionic Tongue Distributed Mechanical Testing Device

In this study, to obtain a texture perception that is closer to the human sense, we designed eight bionic tongue indenters based on the law of the physiology of mandibular movements and tongue movements features, set up a bionic tongue distributed mechanical testing device, performed in vitro simulations to obtain the distributed mechanical information over the tongue surface, and preliminarily constructed a food fineness perception evaluation model. By capturing a large number of tongue movements during chewing, we analyzed and simulated four representative tongue movement states including the tiled state, sunken state, raised state, and overturned state of the tongue. By analyzing curvature parameters and the Gauss curvature of the tongue surface, we selected the regional circle of interest. With that, eight bionic tongue indenters with different curvatures over the tongue surface were designed. Together with an arrayed film pressure sensor, we set up a bionic tongue distributed mechanical testing device, which was used to do contact pressure experiments on three kinds of cookies—WZ Cookie, ZL Cookie and JSL Cookie—with different fineness texture characteristics. Based on the distributed mechanical information perceived by the surface of the bionic tongue indenter, we established a food fineness perception evaluation model by defining three indicators, including gradient, stress change rate and areal density. The correlation between the sensory assessment and model result was analyzed. The results showed that the average values of correlation coefficients among the three kinds of food with the eight bionic tongue indenters reached 0.887, 0.865, and 0.870, respectively, that is, a significant correlation was achieved. The results illustrate that the food fineness perception evaluation model is effective, and the bionic tongue distributed mechanical testing device has a good practical significance for obtaining food texture mouthfeel information.


Experimental Material
According to the classification methods of mechanical properties about food texture, we referred to GB/T 29604-2013 Sensory Analysis-General Guidelines for Establishing a Reference for Sensory Characteristics [1] and selected three kinds of solid foods with different texture characteristics such as apples, bananas and cookies as experimental materials. Wash and peel the apple, then use a knife to cut the sheet with a thickness of 10mm along the longitudinal axis.
The apple sample was prepared using a 10 mm diameter sample preparation device to form a cylinder with a diameter of 10 mm and having 10 mm height. Wash and peel the banana, then use a knife to cut the sheet with a thickness of 10mm. The banana sample was also made by using a 10 mm diameter sample preparation device to form a cylinder with a diameter of 10 mm and having 10 mm height. The cookie was cut with a knife into a cube with the side length of 10 mm, and the sample without cracks was taken. After each test, the bionic tongue indenter and the stage were cleaned to ensure the consistency of measurement conditions.

Sensory Assessment
The food sensory assessment is a kind of statistical-based psychological activity, which is a quantitative and qualitative description of the food by consumers through sensory perception.
According to the national standard GB/T 10220-2012 Sensory analysis-Methodology-General guidance [2]. 20 evaluators (10 men and 10 women) were selected and trained to constitute in an evaluation group to conduct sensory assessment on sample texture of apple, banana and cookie. Under the precondition of guaranteeing sample texture features and basic sampling principles, the samples were numbered and the samples are numbered and submitted to the evaluator for independent evaluation according to a random coding. What's more, we should guarantee the evaluator to wear an eye mask to avoid subjective psychological cues due to its appearance and color. The taste detergent was used to remove residual taste in the oral cavity between two samples or two evaluation activities. Moreover, we used a linear scale detection method to mark the intensity of the sensory perception on a linear scale of 10 cm. The two endpoints of the 10cm line correspond to the lowest score and highest score of the sensory assessment. Each sample was evaluated for three times and it was summarized by using the mean value method. In the end, the linear scale was converted into the corresponding value in proportion for statistical analysis.

Contact Pressure Distribution
In this experiment, there were three kinds of food including apple, banana and cookie.
Each food had 3 parallel samples. The device in the Figure 10 was used for contact pressure test. According to the above experimental method, each parallel sample was tested under 8 bionic tongue indenters with different curvature differences. In the data collection interface of the analysis software, the image of pressure distribution at each moment in the sensing region of the arrayed film pressure sensor in constant speed pushing process was displayed in real time. The image was featured by the annular distribution with the center of peak stress. The image of pressure distribution and the matrix of the pressure values corresponding image at each moment was stored in the data frame. We set to acquire one data frame every 0.02s in this experiment.
The data frame covered from contacting different foods to leave food pieces completely was different. According to the characteristics of the experimental data among the three foods, the maximum range in the data frame was selected as the subsequent calculated sample size, that was, the 17-80 data frames were defined as valid data frame. Within the valid data frame, the maximum matrix of pressure value was 14×12. Thus, the middle sensing area 5×5cm 2 of the arrayed film pressure sensor was defined as the valid pressure collection area.
Considering that the curvature distribution of 8 bionic tongue indenters' surface has a large difference, which results that mechanical information perceived by different bionic tongue indenters are different. The images corresponding to the maximum total pressure in the apple sample perceived by 8 bionic tongue indenters were shown in Figure 1. From figure 1, it can be seen that the bionic tongue distributed mechanical force testing device can obtain a large amount of experimental data. The pressure matrix over the contact surface in each data frame during the pressing contact process can lay the foundation for the study of food fineness.

Establishment of Fineness Perception Evaluation Model
In this study, we used the distributed pressure data from the valid data frame of the contact pressure distribution map to establish a food fineness perception evaluation model.
According to the pressure distribution change data, we selected 3 major indicators to establish the food fineness perception evaluation model. Three indicators have been introduced in the manuscript.
Three indicators including gradient, stress change rate and areal density are the same as the index to describe the fineness, which are weighted to get the comprehensive fineness index I: Where α, β, γ are the weight coefficient of gradient value, stress change rate, areal density.
In order to avoid the influence of subjective randomness, this paper uses entropy weight (2) Standardized formula is represented as: The standardization matrix = ( ) is gained by calculating, expressed as: (4)

2) Index Information Entropy Acquisition
The information entropy Ev of the v-th indicator can be expressed as: where, = ∑ =1 ⁄ , when =0, then the following equation is established as: If the entropy value Ev of a certain index is smaller, illustrating that the degree of variation of the index value is greater, the more amount of information provided, the greater the role of the index in comprehensive evaluation, and the greater the weight should be.

3) Weight of Each Indicator Acquisition
The weight expression of the vth indicator is as follows: The result of the calculation was expressed as: w1=0.339; w2=0.412; w3=0.249, that was, the weight coefficients of three indicators including gradient, stress change rate, areal density were    Figure 3 shows the changing process of the fineness index value, which was following the gradually increasing pressure on apples, bananas and cookies by using bionic tongue indenter S2. According to the defined three indicators including gradient, stress change rate and areal density, the smaller the gradient index value is, the finer the food pieces in the frame is; the smaller the stress change rate, the finer the stress sensing point over tongue surface perceived; The smaller the areal density, the larger the contact area, indicating that the food pieces are finer. Therefore, the smaller the fineness index value is, which indicates that food pieces are finer. From the trend of the three curves, it could be seen that as the stress on the food block increased, the fineness index gradually decreased and tended to be stable, indicating that following the increasing pressure on the food block based on bionic tongue indenter device, the tongue surface could perceive that food pieces were more and more fine, and the fineness index tended to a stable value eventually. In the trend of three foods' fineness changing, it could be seen that the fineness of bananas is the best and the fineness of cookies is the worst, which is in line with the actual perception of human tongue.

Sensory Assessment
The apples, bananas, and cookies were evaluated by 20 evaluators for sensory assessment. The result was recorded on a straight line of 10cm. We measured it and made data statistics, among them, the two endpoints of the 10 cm line were respectively corresponding to the lowest score and the highest score of the sensory assessment, that is, the higher the score, the finer the perception. The fineness perception score results of three kinds of foods are as in Figure 4.  shows that the true fineness perception of humanity is that the fineness of bananas is the best, followed by the apples, and the fineness of cookies is the worst. The conclusion of the sensory assessment and the results from above-mentioned experiment are the same.

Correlation Analysis between Sensory Assessment and Fineness Evaluation Model Results
In order to explore the correlation between the food fineness perception evaluation model results established based on the mechanical information perceived by the bionic tongue indenter and the sensory assessment. Pearson correlation analysis was used to calculate the correlation coefficient and to obtain the correlation coefficient matrix. The correlation coefficient between the two variables X, Y is defined as: where X is the sensory score of food fineness. Y is the fineness index calculated by the evaluation model. ( , ) is the covariance of variable X and Y. EX is the mean of X. EY is the mean of Y. Var(X) is variance of X. Var(Y) is the variance of Y.
The sample correlation coefficient is expressed as: where is a random sample of capacity n taken from variable X. is a random sample of capacity n taken from variable Y. Note：①* indicates that correlation is significant at the 0.05 level. ②** indicates that correlation is extremely significant at the 0.01 level. Table 1 shows that we gained the correlation coefficients of three kinds of foods between This shows that the food fineness perception evaluation model is effective, and the bionic tongue distributed mechanical testing device has a good practical significance for simulating to the food texture.