1. Introduction
Food texture is an important quality indicator in foods, alongside factors such as appearance, taste, and odor [
1,
2,
3,
4]. Szczesniak defines texture as a sensory property arising from the response of the tactile senses to physical stimuli, and is a multi-parameter attribute [
5]. In order to evaluate food attributes, it is necessary to establish the physical characteristics that occurred during the compression or fracture of given foodstuffs.
In terms of the evaluation of food attributes, texture profile analysis (TPA) has been widely used for various foods. Working within the TPA framework, Szczesniak divided the mechanical characteristics of physical properties into five basic parameters, namely: hardness, cohesiveness, viscosity, springiness, and adhesiveness [
6,
7,
8]. These parameters were determined on the basis of measurement data and force versus time, and consist of two time compression and decompression curves. Fracturability is also determined from measurement data and entails a rapid drop in force. In general, instruments used for TPA measure the force by load cell at low sampling frequencies of up to 100 Hz. Hence, instruments with only load cells are insufficient for textures such as crispness and crunchiness with fracturing. As these qualities are popular in many countries [
9], a novel texture sensor design is needed in order to measure the details of fractures.
Various studies have proposed methods and devices to measure the vibrations that occur during fractures. Vibration data are suitable for the evaluation of detailed textures, especially crispness and crunchiness [
10]. Chen et al. analyzed the crispness of biscuits by means of an acoustic envelope detector [
11]. Meanwhile, Varela et al. revealed that the crispness of almonds had a high correlation with the number of vibrations and height of the vibration peaks [
12]. Taniwaki et al. developed a measurement system equipped with a piezoelectric sensor and analyzed measurement data using the fast Fourier transformation [
13]. Taniwaki et al. also assessed the correlation between mechanical and acoustic characteristics [
14]. In measuring the crispness of commercial potato chips, Salvador et al. employed a load cell and microphone and measured force and sound data, which were then subjected to a principal component analysis [
15]. Akimoto et al. evaluated the wet crisp texture of fruits based on vibration, which they measured using a device with a free-running probe [
16]. In turn, Sakurai et al. developed a measurement system for vertical and horizontal vibrations, which comprised a swingarm device with multiple accelerometer sensors that measured the vibration without the use of electric actuators [
17].
The abovementioned studies revealed the importance of vibration for evaluating food texture. However, the evaluation of texture by means of both force and vibration is not a common practice by food manufacturers. We consider there to be two key problems in this area. One is that there is no standardized evaluation method for food texture. If there is sufficient measurement data, artificial intelligence-utilizing techniques may solve this problem. Neural networks have been used to estimate food texture in terms of crispness and crunchiness based on force and acoustic signals [
18]. The other issue is that there is no standard sensor for simultaneously measuring force and vibration. A standard sensor could constitute an integrated-type device equipped with force and vibration sensors, for instance.
In this study, we propose a novel magnetic food texture sensor that corresponds to the tactile sense of the human tooth. We developed a magnetic food texture sensor that imitated the structure of part of the tooth [
19]. A urethane elastomer within the sensor was adopted as the periodontal ligament beneath the tooth. However, the elastomer had low repetitive durability. Therefore, we redesigned it to have higher durability. The new sensor primarily consists of a probe, linear slider, spring, and circuit board. The probe is cylindrical in shape and includes a permanent magnet. Both sides of the spring are fixed to the probe and circuit board. The linear slider enables the smooth, one-axis motion of the probe during food compression. Two magnetoresistive (MR) elements and one inductor on the circuit board were used to measure the probe’s motion. The usage of the MR element and inductor is based on the report that the periodontal ligament, which is under human and animal teeth, has two types of mechanoreceptors with different response characteristics [
20]. The slowly adapting type has a function to produce sustained responses to static stimulation by force. The rapidly adapting type produces transient responses to the onset and offset of stimulation by vibration. The MR elements and the inductor in the texture sensor have roles of slowly adapting and rapidly adapting, respectively. With respect to sensors using a magnet, various tactile sensors were proposed [
21,
22,
23]. They focused on the softness of robotic skin and used soft materials as a probe or surface layer. One of their advantages was a wireless structure between the deformation part and measurement part by using the magnet. The texture sensor in this study has a common advantage of the wireless structure and has a structural difference using the probe made from hard material, the linear slider, and the spring. In the following section, the structure and calibration method of the sensor are described. Fundamental experiments were performed to evaluate the sensor’s range, resolution, repetitive durability, and frequency response. Moreover, the sensor is used to measure seven types of chicken nuggets with different coatings. The difference between the force and vibration measurement data is then revealed based on the discrimination rate of these.
4. Discussion
In this study, we developed a magnetic food texture sensor to simultaneously measure force and vibration occuring by means of a probe. Using the prototype texture sensor, we conducted a series of fundamental experiments to confirm the range, resolution, repetitive durability of force, and frequency response. As
Figure 5 shows, we confirmed the force range of 10, 50, 80, and 150 N using four types of springs. These ranges correspond to, for instance, sliced apples, raw green beans, raw pears, and raw carrots [
1]. We can employ the texture sensor’s full range by appropriately replacing the spring on the basis of the object’s hardness. The spring of the texture sensor is composed of plated iron and affects the magnetic field generated by the permanent magnet it contains. As the texture sensor was calibrated to include the effect of the spring on the magnetic field, it is considered that stable force measurement was even possible with the magnetic spring.
In the experiment conducted for the resolution, we confirmed the force and its error by means of repetitive pushes of 0.02 mm. The resolutions of the four texture sensors were less than 1% of the force range, and the error was less than the resolution. The force was calculated from the voltages of the two MR elements using Equation (
1). As the noise in the voltage was small, the texture sensor had a small resolution. Even if the texture sensor is equipped with a spring with a high spring constant, the common AD conversion ports in the circuit board ensure that the resolution is almost constant, as the AD conversion ports in the circuit board are common. This means that we can estimate the resolution of the force when replacing the spring.
With respect to repetitive durability, as
Figure 7 displays, the force was almost constant, with virtually no tendency to modulate up and down over 1000 repetitions of pushes. This indicates that each sensor had sufficient durability for at least 1000 uses. However, 1000 tests are not enough for industrial applications, and for this about 10,000 tests will be required. As the guaranteed number of compressions of the spring is 300,000, the texture sensor could continue the measurement for 10,000 repetitions, but the durability of the other components is not guaranteed. For instance, the probe was produced by a 3D printer, and it may break after 1000 repetitions. Furthermore, the shape of the texture sensor’s probe generally depends on the object of measurement. The probe used in this study is cylindrical in shape, but wedge-shaped or wide and flat surfaces could be required. Further confirmation of durability is therefore needed, including probes with shapes other than cylindrical.
The texture sensor features two different elements. As shown in
Figure 8, the MR element responds with high sensitivity to frequencies of 100 Hz or less, whereas the inductor responds to frequencies of 10–1000 Hz. As the instruments used in TPA mainly feature load cells for measuring force, they are limited to measurements in the low-frequency band, such as the MR element in this study. On the other hand, the texture sensor described herein measures the displacement and vibration generated in a probe with different elements. Some researchers have suggested the presence of rapidly adapting mechanoreceptors in the periodontal ligament [
20,
26]. The range of their response frequencies has not been reported, but when referring to the response frequencies of rapidly adapting mechanoreceptors in the skin, Bolanowski et al. reported that the high-frequency range for the perception of vibration was from 40 to 500 Hz [
27]. The inductor of the texture sensor satisfied this range. The frequency response of the inductor was considered suitable for measuring textures with sudden fractures, such as arising from crispness and crunchiness qualities. In order to evaluate fracturing textures, the combination of the MR element and the inductor will be important.
To confirm the effectiveness of the texture sensor for the measurement of food, seven kinds of chicken nuggets with different coatings were evaluated. As is shown in
Figure 9, there was some difference in the waveforms of force and vibration, but only on the basis of differences in the coatings. Although the TPA only determines the physical properties from the force data, it is considered difficult to evaluate textures because each sample has individual differences. In this study, the average data on force and vibration for each type of nugget was determined by means of DBA, and the DTW distance between the average data and measured data was then calculated. As
Figure 10 indicates, the DTW distance between the average data and measured data of the same kind was least at the highest percentage of 75% and the lowest of 20%. The mean percentages were 51.4% and 45.7% for the force and vibration, respectively, which means that about half of the samples could be discriminated when the DTW distance was used as an evaluation value. In
Table 5, the force
scores of C1, C3, and C5 were higher than the vibration, and vice versa for C2. The others differed below 10 in terms of
score of force and vibration. This result indicates that C1, C3, and C5 are better characterized by force data than vibration, and vice versa for C2. In other words, the texture sensor that simultaneously measures force and vibration captured different features of the textures of the coatings. In this experiment, the evaluation was performed on the basis of the DTW distance, as the distance between the two sets of data was calculated after expanding and contracting in the time direction and aligning the peaks. As this is a rough comparison though, we would like to analyze and compare the area of force and the numbers and heights of the vibration spikes in greater detail in future research.
The main limitations of this study are as follows. The durability over 1000 repetitions was investigated, but durability beyond this number and in accordance with aging could not be evaluated. In order to quantify the features of the force and vibration, we employed the DTW distance. Moreover, other features should be used, e.g., physical quantities and the geometrical features of the waveform. The experiment dealt only with chicken nuggets with different coatings. The experimental results also depend on only the cylindrical probe. However, it would be necessary to further evaluate the texture sensor for a wide range of targets that also generate vibration, such as potato chips and crackers.