Potential of Piezoelectric MEMS Resonators for Grape Must Fermentation Monitoring ^†

Abstract

In this study grape must fermentation is monitored using a self-actuating/self-sensing piezoelectric micro-electromechanical system (MEMS) resonator. The sensor element is excited in an advanced roof tile-shaped vibration mode, which ensures high Q-factors in liquids (i.e., Q ~100 in isopropanol), precise resonance frequency analysis, and a fast measurement procedure. Two sets of artificial model solutions are prepared, representing an ordinary and a stuck/sluggish wine fermentation process. The precision and reusability of the sensor are shown using repetitive measurements (10 times), resulting in standard deviations of the measured resonance frequencies of ~0.1%, Q-factor of ~11%, and an electrical conductance peak height of ~12%, respectively. With the applied evaluation procedure, moderate standard deviations of ~1.1% with respect to density values are achieved. Based on these results, the presented sensor concept is capable to distinguish between ordinary and stuck wine fermentation, where the evolution of the wine density associated with the decrease in sugar and the increase in ethanol concentrations during fermentation processes causes a steady increase in the resonance frequency for an ordinary fermentation. Finally, the first test measurements in real grape must are presented, showing a similar trend in the resonance frequency compared to the results of an artificial solutions, thus proving that the presented sensor concept is a reliable and reusable platform for grape must fermentation monitoring.

1. Introduction

The fermentation of grape must into wine involves the interaction between yeasts, bacteria, fungi, and viruses. This complex biochemical process has been recognized and studied since the pioneering investigations of Louis Pasteur in the 1860s. During such a fermentation process, yeasts utilize sugars and other constituents of the grape must for their growth, converting these to alcohol (ethanol), carbon dioxide, and other metabolic end products [1]. A serious problem in winemaking occurs when the yeast growth and the alcoholic fermentation stops prematurely, which results in a wine with residual, unfermented sugar and a concentration of ethanol less than expected. This is indicated in minor deviations of the physical properties, such as density and viscosity, when compared to an ordinary fermentation, and the wine is commonly referred to as being stuck, or sluggish [2,3].

There are several approaches to monitor such a fermentation of wine according to different parameters related to the process. El Haloui et al. [4], Nerantzis et al. [5], and Koukolitschek [6] reported on density determination based on differential pressure measurements or flexural oscillators, respectively. By monitoring the CO₂ release, the fermentation process can also be monitored, as reported in [4]. Another approach is to determine the yeast cell population evolution by means of impedance [7] or turbidity measurements [8]. The propagation velocity in grape must, determined by ultrasound measurements, can also be used for monitoring the fermentation process [9,10]. Lastly, a reflective technique based on fiber optics was reported in [11,12]. Basically, all of these different approaches have their specific drawbacks; some lack enough accuracy, others have only been tested in discrete must samples and, in some cases, the sensor output performance deteriorates dramatically due to an increasing deposition of tartaric acid crystals on the active surface of the sensors [12]. In recent years, cantilever-like micro-electromechanical system (MEMS) resonators have become a reliable platform for various sensing applications [13]. One significant advantage of such resonators is that they can usually be fabricated using silicon micromachining technology and are, therefore, smaller and much cheaper to produce, compared to traditional quartz crystal resonators. So-called “fluidic channel resonators”, in which the fluid passes through a fluidic channel in the moving part of the resonator, is beneficial due to their high-precision capabilities, especially for molecular detection [14], blood coagulation [15], or to quantify other physical properties of fluids, such as mass density or viscosity [16,17]. With such a system only a small amount of liquid (5 μL [18]) is needed and masses can be detected down to the attogram regime [19]. For actuation and sensing, lasers are often used to drive such micro- [20] or nano- [21] mechanical resonators. This approach is considered to be a very accurate and flexible, with the drawback of making the measurement setup bulky, expensive, and mobile integration impracticable. A disadvantage of fluidic channel resonators is their low reusability, which is caused by potential clogging of the fluidic channel. Therefore, such resonators are often designed as one-time use devices, which increases the cost of the overall system. In contrast, solid resonators are often easier to clean, which increases reusability, but generally lacks in precision compared to fluidic channel resonators. The recent improvements in the field of piezoelectric solid MEMS resonators for liquid monitoring purposes predestines such a system for sensing applications, where decent accuracy is required [22,23,24]. The piezoelectric actuation and readout mechanism keeps the sensor device reasonably small and does not require any further laser-based measurement equipment. In this particular field, the question arises whether a micro-machined solid resonator is capable to detect the minor deviations in the physical properties of the grape must during the fermentation process, and whether it offers a promising alternative to the presented measurement approaches. Most recently, Toledo, et al. [25] introduced a phase locked loop-based oscillator circuit in combination with a commercial lock-in amplifier to track the oscillation frequency of the solid resonators [26]. Using this measurement setup, in combination with an evaluation procedure presented in [27], it could be shown that the monitoring of grape must fermentation using solid MEMS resonators is, in principal, possible.

In this study we will focus on the reliability and the reusability of such a solid resonator and what impact potential contaminations caused have on the sensor characteristics despite the harsh liquid environment. Based on these investigations, an estimation of the precision of the sensor concept is given. Furthermore, different measurement and evaluation procedures are used to validate the data presented in [25]. Finally, test measurements in a real grape must are presented and compared to those of the artificial solutions, thus proving the suitability of the presented concept for grape must fermentation monitoring.

2. Experimental Details

2.1. Sensor Specification

The piezoelectric MEMS resonator is fabricated on four-inch SOI-wafers and features a length of L = 2524 μm, a width of W = 1274 μm, and a thickness of T = 20 μm, passivated with a SiO₂/Si₃N₄ bi-layer of t_iso = 250/80 nm. For actuating and sensing, an aluminium nitride (AlN) thin film with a thickness of t_AlN = 1 μm is sputter-deposited onto the plate surface and is sandwiched between two chromium/gold thin film electrodes with equal thickness for the bottom and top electrode of t_be = t_te = 50/450 nm. In Figure 1, a typical sensor chip after packaging and wire bonding with a released single side-clamped resonator, and its top view as an inset, are shown.

The corresponding mode shape is illustrated by finite-element method (FEM) eigenmode analyses of the fourth-order (15-mode) of the roof tile-shaped mode in Figure 2a in side and, in Figure 2b, in top view. In Figure 3 a schematic cross-sectional view is presented, showing the electrode design optimized for the fourth-order of the roof tile-shaped mode and four electrodes (w_o = 335 μm, w_i = 272 μm) with alternating anti-parallel ($\downarrow \uparrow \downarrow \uparrow$) electric excitation to ensure in addition a collection of almost all generated polarization charges without cancellation. The result of this tailored electrode design is an increased deflection in z-direction and, in further consequence, an increased strain related conductance peak as reported in [28], which is one of the major reasons why this oscillation mode is used in this study. Furthermore, it is worth mentioning that the resonance frequencies in liquids do not exceed 600 kHz, which simplifies the electrical measurement procedure and minimized any pressure-induced energy losses by sound waves, which requests compressive fluid properties as reported in [29]. Secondly, high Q-factors above 100 are achieved with this type of mode, which facilitates a high sensor sensitivity. To overcome the obstacle of parasitic current caused by the conductivity of the model solutions, the entire sensor element, including bond wires and ceramic package, is passivated with an amorphous silicon dioxide thin film with a thickness of ~4 μm.

2.2. Artificial Model Solutions and Sensing Principle

Two sets of model solutions are prepared, using a R200D microscale from Sartorius (Goettingen, Germany), representing an ordinary (N_1–9) and a stuck/sluggish (S_1–9) wine fermentation process, consisting of fructose, glucose, glycerol, and ethanol [30]. The specific compositions are listed in Table 2 and Table 3 and show significant differences in the fructose, glucose, and ethanol concentrations, when comparing ordinary (Table 2) and stuck/sluggish (Table 3) wine fermentation processes. The resonator is immersed in these model solutions and actuated in the fourth-order of the roof tile-shaped mode with an Agilent 4294 A impedance analyzer (excitation voltage V_exc = 500 mV AC, Keysight, Böblingen, Germany). The actuation in the resonance leads to an increased average deflection and to increased strain on the sensor surface, respectively. Due to the piezoelectric effect, polarization charges are generated, which are detected by the impedance analyzer as an increased conductance peak ΔG. A typical peak characteristic of the piezoelectric MEMS resonator is shown in Figure 4 when immersed in the model solution N₁. This measurement procedure lasts only ~10 s. Subsequently, the sensor is cleaned with dish soap and isopropanol, followed by a 20 min drying process at room temperature to avoid tartaric acid crystal deposition, as reported in [12].

From these output characteristics, the resonance frequency f_res and Q-factor Q are determined using the Butterworth-van Dyke equivalent circuit, in combination with a Levenberg-Marquardt algorithm [31]. Once f_res and Q are obtained for all solutions (N₁–N₉ and S₁–S₉), an evaluation procedure taken from [32] is used to determine the actual density ρ_f using:

$${\rho }_{\mathrm{f}}=\frac{\mathrm{Im}(Z_{\mathrm{m}})-{\omega }_{\mathrm{res}}{L}_{\mathrm{m}}+\frac{1}{{\omega }_{\mathrm{res}}{C}_{\mathrm{m}}}-\mathrm{Re}(Z_{\mathrm{m}})+R_{\mathrm{m}}}{{\omega }_{\mathrm{res}}{d}_1}$$

(1)

Here, Z_m describes the impedance of the equivalent resonance circuit including a resistance R_m, an inductance L_m and a capacity C_m at a certain angular frequency ω_res in liquid. The constant parameter d₁ is determined in a calibration liquid with known viscosity and density, using:

$$d_1=\frac{\mathrm{Im}(Z_{\mathrm{m}})-{\omega }_{\mathrm{res}}{L}_{\mathrm{m}\text{\_}\mathrm{air}}+\frac{1}{{\omega }_{\mathrm{res}}{C}_{\mathrm{m}\text{\_}\mathrm{air}}}-d_2\sqrt{{\mu }_{\mathrm{f}}{\rho }_{\mathrm{f}}{\omega }_{\mathrm{res}}}}{{\omega }_{\mathrm{res}}{\rho }_{\mathrm{f}}},$$

(2)

with:

$$d_2=\frac{\mathrm{Re}(Z_{\mathrm{m}})-R_{\mathrm{m}\text{\_}\mathrm{air}}}{\sqrt{{\mu }_{\mathrm{f}}{\rho }_{\mathrm{f}}{\omega }_{\mathrm{res}}}}$$

(3)

where R_{m_air}, L_{m_air}, and C_{m_air} are the corresponding unperturbed resonator values in air and are given in

Table 1. Characterization of the resonator in air. R_{m_air}, L_{m_air}, and C_{m_air} represent the unperturbed resonator in air.

fres	Q	Rm_air	Lm_air	Cm_air
[kHz]	[1]	[kΩ]	[mH]	[fF]
1014.1	329.6	1.365	70.662	348.77

. For this initial calibration process values supplied by the manufacturer for μ_f and ρ_f of N₄ and S₄ are used to calculate d₁ and d₂, as shown in Equations (2) and (3), respectively. Once these two constants are obtained, all further values of ρ_f can be determined using Equation (1).

2.3. Real Grape Must Fermentation

For the investigation of a real fermentation process, Airen grapes are processed into grape must with a subsequent filtering process to avoid damaging the sensor by large grape skin particles. After this procedure, the grape must is inoculated with 0.2 g/mL of Saccharomyces cerevisiae strain (UCLM S325, Fould-Springer, Ciudad Real, Spain) previously rehydrated, as described in the supplier guidelines. A cylindrical fermenter (mini-Bioreactor Applikon Buitechnology B.V., Delft, The Netherlands) is filled with 3 L of inoculated must, where the temperature was controlled at 28 °C. The fermentation monitoring is carried out by means of the extraction and analysis of 7 mL-samples approximately every 5 h during the following six days. After extraction, the samples were centrifuged for one minute at 1000 rpm (Universal 32R Hettich, Tuttlingen, Germany) and kept refrigerated until analysis.

3. Results

In Figure 5, the results from the reproducibility measurements are depicted for N₃ and N₇ of an ordinary fermented grape must in Figure 5a, and for S₇ of a stuck fermented grape must in Figure 5b. The measurement procedure, including the cleaning process presented in the previous part of this paper, is repeated for 10 times. Thereby, standard deviations in the resonance frequency f_res of 0.102% for N₃, 0.085% for S₇ and 0.094% for N₇ are obtained, thus showing a high potential for the targeted application. Higher standard deviations are obtained for the Q-factor (~11%) and the conductance peak (12%), which, however, have minor impacts on the final calculation of the density values, indicated by a standard deviation of the calculated density values of 1.2% for N₃, 1.1% for N₇, and 1.0% for S₇.

The compositions of the artificial wine solutions for an ordinary (N_1–9) wine fermentation process, listed in

Table 2. Composition of the model solutions representing a desired wine fermentation process from N₁ (raw grape must) up to N₉ (ordinary fermented model solution) [30]. The measured values are obtained using a R200D microscale from Sartorius.

Solution	Fructose		Glucose		Glycerol		Ethanol
-	[g/L]		[g/L]		[g/L]		[% v/v]
-	Nom.	Meas.	Nom.	Meas.	Nom.	Meas.	Nom.	Meas.
N1	110	109.90	100	99.90	0	0	0	-
N2	90	89.90	80	80.00	0	0	1	-
N3	70	69.95	30	30.03	5	4.99	6	-
N4	60	59.97	20	19.99	5	5.00	8	-
N5	40	39.90	10	9.95	6	5.99	9	-
N6	20	19.90	2	2.00	7	6.97	12	-
N7	8	8.02	2	2.00	7	6.99	13	-
N8	5	5.00	2	2.00	7	6.99	13	-
N9	2	2.00	1	0.99	9	8.99	14	-

, shows a decrease in fructose and glucose concentration from 110 and 100 g/L (N₁) to 2 and 1 g/L (N₉), respectively. Likewise, the glycerol and ethanol concentration are increasing from zero to 9 g/L and 14% v/v. In the case of a stuck fermented wine process, as it is listed in

Table 3. Composition of the model solutions representing a stuck wine fermentation process from S₁ (raw grape must) up to S₉ (stuck fermented model solution) [30]. The measured values are obtained using a R200D microscale from Sartorius.

Solution	Fructose		Glucose		Glycerol		Ethanol
-	[g/L]		[g/L]		[g/L]		[% v/v]
-	Nom.	Meas.	Nom.	Meas.	Nom.	Meas.	Nom.	Meas.
S1	110	109.90	100	99.98	0	0	0	-
S2	90	90.00	80	79.98	0	0	1	-
S3	80	79.98	60	60.01	3	3.00	3	-
S4	70	69.98	30	29.98	5	4.99	6	-
S5	66	65.99	27	27.03	5	4.99	7	-
S6	64	64.00	26	26.06	5	5.01	7	-
S7	62	61.99	24	24.00	6	5.99	7	-
S8	58	57.99	22	22.00	6	5.99	7	-
S9	57	56.99	21	21.00	7	7.01	8	-

, the fructose and glucose concentrations decrease to 50 and 20 g/L, respectively, and remain constant at these values. In parallel, the increase in the ethanol concentration stops as well, and does not exceed a value of 8% (v/v). These significant changes in the composition of the model solutions affect the resonance response of the MEMS sensor as shown for both an ordinary fermentation in Figure 6 and a stuck fermentation in Figure 7. In the first case, the resonance frequency increases monotonically for the ordinary fermentation (see Figure 6). In contrast, the increase of the resonance frequency during the investigation of the stuck fermentation plateaus at ~575 kHz.

In Figure 8, the results of the resonance response analysis are compared to the ethanol concentration σ_Ethanol for both sets of model solutions. Significant differences in the resonance frequencies f_res are evident with increased σ_Ethanol values when comparing ordinary and stuck fermentation, as well as a good correlation between σ_Ethanol and f_res. The combination of unfermented sugars and lower concentration of ethanol in a stuck fermentation (see Table 3) leads to a flat resonance characteristics, allowing to detect stuck fermentation. The results from the repeatability measurements of N₃, N₇, and S₇ from Figure 5 are used to estimate the precision of the MEMS sensor. For a further analysis of this particular progress a second resonator with the same geometry is excited in the first roof tile-shaped mode (12-mode). The resonance frequencies in both sets of model solutions amount around 1/4 of the resonance frequency of the 15-mode (~160 kHz) and are depicted in Figure 9, showing a similar characteristics in the frequency response. Compared to the results presented in Figure 7 the absolute change in resonance frequency is also ca. four times lower, leading to a lower sensitivity for the 12-mode. Therefore, all further evaluations are performed using the results from the resonator excited in the 15-mode. In Figure 10 the resonance frequencies for both parameter sets (N_1–9 and S_1–9) as a function of the corresponding density values are depicted, showing a linear increase in f_res from 567 kHz (N₁) to 582 kHz (N₉), and again a premature stop at ~575 kHz for the stuck fermentation.

As a next step, the density values of the model solutions are evaluated with a Stabinger SVM3000 viscometer (Anton Paar, Graz, Austria). The results are listed in

Table 4. Determined density values (Res.) for ordinary and stuck/sluggish model solutions and their comparison to reference values (Stabinger) obtained with a Stabinger SVM3000 viscometer.

Ordinary				Stuck/Sluggish
Sol.	Density ρ [g/cm3]			Sol.	Density ρ [g/cm3]
-	Stabinger	Res.	Dev.	-	Stabinger	Res.	Dev.
N1	1.081	1.082	−0.001	S1	1.075	1.075	-
N2	1.065	1.063	0.002	S2	1.063	1.064	−0.001
N3	1.032	1.033	−0.001	S3	1.050	1.050	-
N4	1.020	1.023	−0.003	S4	1.031	1.031	-
N5	1.007	1.010	−0.003	S5	1.028	1.023	0.005
N6	0.994	0.994	-	S6	1.025	1.024	0.001
N7	0.988	0.989	−0.001	S7	1.020	1.029	−0.009
N8	0.985	0.978	0.007	S8	1.023	1.020	0.003
N9	0.996	0.995	0.001	S9	1.021	1.020	0.001

and depicted in Figure 10, showing an almost linear decrease for N_1–9 from 1.081 (N₁) to 0.996 g/cm³ (N₉) and a premature stop in the density decrease for S_1–9 with a minimal value of 1.020 g/cm³ (S₇). Furthermore, the results of the density determination are compared to those from the Stabinger viscometer in Figure 11, showing low deviations and the possibility to distinguish between ordinary and stuck fermentations at an early stage of the fermentation process.

Finally, the results of the frequency response analysis for real grape must fermentation are presented in Figure 12, showing a similar trend in the change of f_res, compared to those of the artificial solutions presented in Figure 8, which indicates the high potential of the presented sensor system to withstand the harsh conditions during the fermentation process even in real grape must.

4. Conclusions

In this paper, an approach to monitor the fermentation processes in winemaking by analyzing the frequency response of a piezoelectric MEMS resonator is presented. The sensor is excited at the fourth-order of the roof tile-shaped mode in several artificial grape must model solutions, representing an ordinary and a stuck wine fermentation process, respectively. During the artificial fermentation, both an increasing ethanol and a decreasing total sugar concentration reduce the density of the grape must despite the counteracting effect of the increasing glycerin concentration. This decrease in density leads to higher resonance frequencies. Due to the high Q-factor of the MEMS sensor in liquid media (~100 in isopropanol), even minor shifts in resonance frequency can be detected with high precision, which enables the possibility to monitor these changes of the physical properties during the fermentation processes. Reproducibility and reusability measurement were performed, showing low standard deviations in the resonance frequencies of ~0.1% and moderate deviations in the Q-factor (~11%) and conductance peak (~12%). These results show that the resonance frequency is hardly effected by potential surface contaminations. Due to the fast measurement cycle of 10 s and easy cleaning procedure, harsh contaminations, such as the deposition of tartaric acid crystals, could be avoided. Nevertheless, minor surface contaminations reduce the precision of the calculated density values leading to a standard deviation of ~1.1%. Finally, test measurements in real grape must were performed resulting in similar sensor characteristics compared to the results of the model solutions, demonstrating the high potential of the presented sensor concept for grape must fermentation monitoring.