Rapid Non-Invasive Capacitive Assessment of Extra Virgin Olive Oil Authenticity

Abstract

Economically motivated adulteration (EMA) and/or cross-contamination are the two major factors resulting in the substandard quality of premium edible oil like extra virgin olive oil (EVOO) produced in food and beverage (F&B) fast-moving consumer goods (FMCG) industries. Current quality assurance methods (e.g., spectroscopy and chromatography) in FMCG involve intrusive sample extraction and ex situ analysis in a laboratory using expensive bulky instrumentation, which is neither integrable inline nor scalable to match the production throughput. Such techniques do not meet the industrial requirements of in situ testing, non-intrusive analysis, and high throughput inspection (100% product verification) leading to food loss and package waste from unwanted batch rejects. Herein, a low-cost electrical approach based on capacitance is proposed to show the proof of concept for screening EVOO-filled containers non-invasively for adulteration without any sample extraction by capturing the differences in the dielectric properties of mixed oils. The sensor system displayed a fast response (100 ms) and low detection limits for different adulterants (olive oil (32.8%), canola oil (19.4%), soy oil (10.3%) and castor oil (1.7%)), which is suitable for high-throughput (>60 sample/min) screening. Furthermore, a low-cost automated system prototype was realized to showcase the possibility of translating the proof of concept for possible scaling up and inline integration.

1. Introduction

Quality assurance (QA) in the food and beverage (F&B) industries is of paramount importance to brandish brand integrity and product quality. QA is the foremost boundary between consumers and false foods. It ensures the delivery of high-quality products for consumers and forestalls producers from falsifying claims about product quality. A notable industry that thrives on high-speed mass production like no other is the F&B fast-moving consumer goods (FMCG) industry. Yet, amidst the industry’s battle to keep up with a rapidly evolving and expanding consumer base, quality is still at its core. Food adulteration involves lowering food quality by either replacing its major constituents with inferior alternatives or removing major ingredients from the food item. Adulterants are any materials that make the food sub-standard or misbranded or contain extraneous matter [1]. Besides intentional adulteration, storage, mixing and filling conditions in production affect the food quality and ingredient composition. Implementation of quality control tests at high speeds in FMCG supply chains is a major challenge. This has garnered monumental interest in cost-effective yet cutting-edge inspection technologies by both producers and food inspection agencies alike to check for the authenticity of food products at high speeds.

Extra virgin olive oil (EVOO) is a prominent example of a high-value FMCG product. Having an impressively low free acidity of not more than 0.8% (m/m), EVOO is markedly the most nutritious and healthiest oil [2], thereby creating a remarkably huge demand and generating market revenue of 1410.2 million USD in 2020, which is projected to further rise to 1465.6 million USD by 2026 [3]. Such in-demand products incite the producers to adulterate EVOO, consequently making EVOO one of the leading types of food that experiences adulteration [4]. When intentional, it is usually done to illicit monetary gain and thus, is known as economically motivated adulteration (EMA) [5]. In the case of EVOO, common adulterants include olive oil (free acidity of less than 1% (m/m)), soy oil, castor oil, and canola oil. Adulteration cost the European Union about 4 million Euros per year [6]. Besides the economic loss, adulteration oils could lead to poor nutritional quality and have major health implications due to the presence of allergic, genotoxic and/or carcinogenic compounds. The other form of loss in edible oil authenticity arises from cross-contamination in processing lines. Oils of similar nature are subjected to processing in the same feedlines after cleaning to save on the cost of constructing new lines and maintain sustainability by adopting a smaller machinery footprint. Additionally, passage through multiple lines for storage, filling, treatment, and dispensing could affect the oil authenticity and introduce unwanted foreign matter. Therefore, the detection of edible oil adulteration is crucial for maintaining food quality and safety.

In general, olive oil is mainly composed of oil fraction and non-oily fraction. The oil fraction is the major portion constituting 98–99% of the total oil. It consists of triglycerides containing monounsaturated fatty acids (MUFA, 55–85%) such as oleic acid [7]. It also contains palmitoleic acid; saturated fatty acids (12.7%), such as palmitic, stearic, and arachidic acids; and polyunsaturated fatty acids (10.1%), including linoleic and linolenic fatty acids. The non-oil fraction contains smaller amounts of minor compounds (1–2%) such as fatty alcohols [8], sterols [9,10], waxes [11], tocopherols [12,13], phenolic compounds [14,15], carotenes and chlorophylls [16,17], n-alkanes and n-alkenes [18] and volatiles [19], which also determines the organoleptic characteristics of olive oil. The antioxidant properties of olive oil are linked to the radical scavenging actions of phenolic compounds such as hydroxytryrosol and oleuropein, which reduce the formation of superoxide anions, neutrophil respiratory burst, and hypochlorous acid thereby lowering the risk of coronary heart disease and cancer associated with the Mediterranean diet [20]. The high levels of unsaturated fatty acids in olive oil improve the lipid profile by reducing the low-density lipid (LDL)/high-density lipid (HDL) ratio in humans. The resistance against the oxidation of LDL by phenolic compounds helps in preventing ischemic heart disease besides improving endothelial function needed for improving the hemostatic system in hypercholesterolemic individuals [21]. Monounsaturated fatty acids like oleic acid induce a positive effect on natural killing cell activity by boosting the immune system [22]. Olive oil increases the concentration of gastrointestinal hormones, which in turn act on the synthesis or secretion of enzymes and other constituents of the digestive juices, thereby promoting faster absorption and metabolism [23]. High oleic acid content has a beneficial effect in curing fatty liver diseases by reducing the accumulation of triacylglycerols in the liver, improving postprandial triacylglycerols glucose, and glucagon (like peptide-1) responses in insulin-resistant subjects, and upregulating glucose transporter-2 expression in the liver [24].

To avail of the maximum nutritional benefits of olive oil, the olive oil quality needs to be of the highest order. This is possible only by controlling both pre-harvest and post-harvest factors. Several pre-harvest factors such as cultivar, edaphoclimatic conditions, environmental conditions, pedalogic conditions, tree age, treatment, fruit ripening and harvest time influence the olive oil quality. For example, the highest content of campesterol, cholesterol, β-sitosterol, and total sterol was detected in Picholine cultivar [9]. Picholine Marocaine cultivar done at 600–900 m altitude showed that the chlorophyll and carotenoids content gradually increases, which is nutritionally preferable [25]. In general, olive flowering requires spring temperatures, growth and development need high temperatures and olive fruit ripening needs heavy rain [26]. Pedologic (sandy, clay, stony, limestone, gypsum and brown soil) conditions influence the α-tocopherol, the sterols composition and the volatiles of the olive oils [27]. Total phenol content is influenced by the age of the tree, and 30–50 years-old olive trees showed the highest phenol content compared with 10–30 years old trees [25]. The phenols and o-diphenol contents of the oils decrease during fertilization treatments [28]. Boron alters the biochemical functions of phenylalanine ammonia-lyase, peroxidase and polyphenol oxidase activities [29]. Linoleic acid increases during ripening due to oleate desaturase that changes oleic acid to linoleic acid in the olive [30]. Some of the post-harvest factors including fruit storage, leaf removal and fruit washing, fruit crushing, oil extraction systems and oil storage also influence the olive oil quality. Storage of more than 24 h is not recommended between harvesting and processing of olive fruits because longer storage time may hydrolyze the triglycerides to free fatty acids with the action of lipases, in the presence of moisture [31]. Leaf discards used in olive fruit harvesting raise the degree of green color, which gives an organoleptic sensation (color, aroma and taste) [32]. The recommended olive crushing time is 20–30 min because crushing the fruit within this time window produces less peroxide and prolonged crushing could degrade the quality of the olive oil [33]. During storage in plastic containers, reactions occur between oxygen and the unsaturated fatty acids in olive oil occur thereby degrading its quality over time [34]. Pressure, centrifugation and percolation are used for olive oil extractions in olive oil mills from a paste form. Olive oil extraction using natural decanting is time-consuming, laborious and may cause contamination whereas centrifugation reduces time, labor and contamination [33].

Among the several grades of olive oil defined by the European Commission [35] and the International Olive Council (IOC) [36], EVOO is considered the best quality olive oil since is obtained directly from the olive fruit by mechanical or other physical processes without altering oil properties. EVOO has a free acidity, expressed as oleic acid, of not more than 0.80 g per 100 g and the other physico–chemical and organoleptic characteristics correspond to those fixed for this category in this standard [37]. Olive oil obtained by blending refined olive oil and virgin olive oil other than lampante olive oil, having a free acidity content, expressed as oleic acid, of not more than 1 g per 100 g, forms regular olive oil [37]. Due to the premium quality, the price of EVOO is the highest among different grades and is vulnerable to adulteration. Conventional methods such as Fourier transform-infrared (FTIR) spectroscopy [38], near-infrared (NIR) reflectance spectroscopy [39], Raman scattering spectroscopy [40], nuclear magnetic resonance (NMR) spectroscopy [41], high-performance liquid chromatography (HPLC) [42] and gas chromatography (GC) [43] are widely used for edible oil authenticity assessment. These techniques rely on sample extraction from packaged products and offline analysis in a laboratory to identify the chemical composition, fatty acid content, etc. Such offline methods are expensive because of the sophisticated instrumentation that requires a high level of expertise for data analysis and interpretation. This is further compounded by the high implementation cost for large-scale adoption and low inspection throughput due to random sampling, which is not attractive for inline monitoring. In addition, unsealing the packages/containers for sample extraction leads to food loss and package waste generation affecting environmental sustainability [44]. The absence of a reliable in situ oil quality/authenticity monitoring system in conjunction with the drawbacks of the current methods are the motivating factors to seek other rapid low-cost non-destructive evaluation systems to assess edible oil quality. The exponential progression in high-speed electronics and sensor technologies opens avenues to seek alternative non-destructive technologies in the spectrum of food quality checking. The desirable characteristics of such a system are as follows: (i) cost-effectiveness, (ii) high-speed sensing, (iii) in situ placement, (iv) non-intrusiveness, and (v) reduced equipment footprint for seamless inline integration.

Recently, dielectric spectroscopy, a technique which provides dielectric information of a medium as a function of frequency, as a screening tool for authenticity check of edible oils has garnered interest because different fatty acid compositions manifest distinguishable behavior across the electromagnetic spectrum [45]. Dielectric spectroscopy is a closely related technique used to study the response of a sample via the interaction of an external field with its electric dipole moment within a broad frequency range. It is used to estimate the dielectric permittivity of a material as a function of frequency. Unlike the infrared spectrum used in Raman (0.3 THz–120 THz) [46], FTIR (12–120 THz) [47], and NIR spectroscopy (118–384 THz) [48], dielectric spectroscopic characterization covers a relatively lower frequency range (few MHz to GHz); which is economically viable for low-throughput offline inspection when compared with the other spectroscopic methods [49]. Despite its attractive features, dielectric spectroscopy faces several challenges for deployment in an inline high-throughput inspection environment. Being a wide-range frequency response method capable of measuring the dielectric permittivity directly, the operational frequency range should be high (MHz-GHz) to induce polarization. Such high-frequency analysis requires an expensive microwave probe station and/or bulky benchtop vector network analyzer (VNA) equipment, which limits in situ inspection and large-scale implementation in a high-throughput inspection setting such as FMCG. Moreover, to estimate the dielectric properties accurately, the electric field needs to be concentrated near the material under test. Consequently, the dielectric probe is in proximity or contact with the sample to reduce the influence of package/container dielectric effects. The current research on portable dielectric spectroscopy-based sensors is limited to intrusive offline testing using concentrated high-frequency waves for a small sample volume (few uL) and/or contact analysis from a dedicated ex situ inspection region [50,51] and no reports on in situ and non-intrusive testing of finished packaged goods are available. However, the promising aspects of dielectric spectroscopy give the incentive to pursue an electrical response method, which can overcome the current drawbacks of dielectric spectroscopy.

Capacitance (C = ε_rε_oA/d), is one quantity, which is a function of the material property (dielectric constant or relative permittivity of the material, εr and εo is the permittivity of free space = 8.85 × 10–12 Fm⁻¹), the material dimensions (A, electrode area traversing the field), and the distance (d) from the material under test, that offers an electrical readout capable of overcoming the aforementioned drawbacks of traditional methods and dielectric spectroscopy [52,53,54]. Capacitive sensing distinguishes material types and physical changes to the material under test, allows contactless operation, provides adjustable sensing depth, is low-cost, fast and requires low power [55,56,57]. The close correlation between capacitance and dielectric property could be extrapolated to both quantitatively and qualitatively assess the deviation in the authenticity of edible oils via blending with low-grade alternatives. Such oil mixtures possess a different dielectric characteristic compared with authentic oils, which is readily distinguishable via indirect capacitance profiling at a single frequency without requiring expensive probes and wide-range frequency characterization to estimate the actual dielectric constant. Hitherto, both capacitive sensing and dielectric spectroscopy, have been explored for food/oil quality characterization [58,59]. Food spoilage detection [60], water content estimation in edible oils [61], moisture measurement in food [62] and adulteration in olive oil [63,64] are some of the applications reported using offline capacitive/dielectric analysis on extracted samples. Albeit in its infancy, capacitance-based non-destructive testing of food container/package quality has shown great potential for further exploration [65,66,67,68]. This provides the incentive to extrapolate the concept of capacitance for adulteration detection. To the best of our knowledge, the proposed capacitive approach herein is novel for direct through-container screening of EVOO for authenticity check without requiring oil extraction from a sealed container. In general, the proposed detection method offers an economical pathway for high-throughput non-invasive edible oil inspection warranting further research for adulteration detection. Furthermore, capacitive sensing offers bespoke electrode designs, sensing topologies and sensor placement configurations with controllable field penetration depth and strength to accommodate end-product aesthetics and overcome any mechanical constraints during customized inline integration. The through-container testing scheme does not require any oil sample to be extracted for offline analysis, which prevents unwanted waste generation (food and package) from the disposal of tampered containers, which is contrary to the existing offline industrial solutions.

In this work, a portable low-cost capacitance measurement setup comprising two parallelly placed rectangular electrodes demonstrates the feasibility of non-invasive adulteration detection in packaged containers without any form of sample extraction. EVOO-filled containers adulterated to varying degrees with different adulterants served as the dielectric medium between the electrodes. The electrostatic field between the electrodes generates capacitance signals of varying strengths depending on the degree of adulteration and the type of adulterant. The proof of concept was further translated into a small-footprint automated preliminary prototype targeted towards the development of a standalone system with inline integration potential.

2. Materials and Methods

2.1. Preparation of Blended Oils

Several commercially available edible oils were purchased from retail stores including extra virgin olive oil (EVOO, Naturel^®, Andalusia, Spain), olive oil (OO, Naturel^®, Andalusia, Spain), soybean oil (SOY, FairPrice, Singapore), sunflower oil (SF, FairPrice, Singapore), vegetable oil (VEG, blend of 50% palm oil and 50% soybean oil, FairPrice, Singapore), canola oil (CAN, FairPrice, Singapore) and castor oil (CAS, Ambika, India). EVOO and OO are products of Spain with an expiry date in 2025. All other oils were products of Singapore. No information on the harvest and/or geographical origin was provided unless specified otherwise. All oils were purchased and tested in 2022. Commercial vegetable oils are blended mixtures of palm oil with other edible oils for stability since palm oil is solid below 40 °C. EVOO was spiked using 4 common adulterants (OO, SOY, CAN and CAS) to make binary mixtures with 6 degrees of adulteration (2.5%, 5%, 10%, 20%, 40% and 60% (v/v)). Plastic containers made of polyethylene terephthalate (PET) with a wall thickness of ~1.2 mm were used for the sensing experiments. 15 mL oil volume was used throughout the study unless stated otherwise.

2.2. Sensing Setup

Figure 1 shows the experimental setup, sensing configuration and sensor electrode design used in this study. Rectangular-shaped sensor electrodes were designed in-house using printed circuit board (PCB) design software (EasyEDA, JLCPCB, Shenzhen, China) and outsourced for fabrication on FR4 PCB substrates (JLCPCB, China). Two electrodes spaced 27 mm apart were connected to two excitation terminals (INA and INB) of a capacitance sensing module (FDC2214EVM, Texas Instruments, Dallas, TX, USA). The FDC214 capacitance board uses a sinusoidal excitation signal with a default frequency of 5 MHz and a peak-to-peak voltage (Vp-p) of 1.8 V and is operated in a differential sensor topology (refer to Section 3.4), where two terminals are excited by the same signal to produce a strong sensing field necessary to cover the entire diameter (sensing depth) of the container with a non-polar liquid like oil [69].

2.3. Capacitance Measurements

Real-time capacitance data was logged using Sensing Solutions EVM GUI (Texas Instruments, Dallas, TX, USA). All measurements were done in a laboratory ambience (room temperature (RT) ~25 °C and relative humidity (RH) ~60%) unless stated otherwise. Statistical analysis, data plotting, linear fitting and hypothesis testing were performed using OriginPro software (v2019b, Northampton, MA, USA). To demonstrate measurement reproducibility and repeatability, at least 6 containers were tested per container/oil sample with 3 repeats for a total of ~12 s (4 s per sample), thereby yielding ~300 data points (100 ms sampling rate) per oil type unless stated otherwise. For sensor characterization, capacitance responses to all oils were recorded before adulteration experiments. The responses were obtained using the relative capacitance (ΔC/C_O = (C − C₀)/C₀), with air as the baseline (blank), where C₀ and C are capacitances of the sensor in air and with container (filled and unfilled), respectively.

To obtain the sensor calibration curve for adulteration tests, four commonly used adulterant oils, namely, OO, SOY, CAN, and CAS were mixed with EVOO in varying proportions and evaluated. The relative capacitance (ΔC_EVOO/C_EVOO = (C_AO − C_EVOO)/C_EVOO) of the calibration curve was calculated using EVOO as the reference (blank), where C_AO and C_EVOO are capacitances of EVOO oil (blank) and adulterated EVOO (blending with OO, SOY, CAN, or CAS), respectively. The slope of the sensor calibration curve ((ΔC_EVOO/C_EVOO) vs. the degree of adulteration (%)) curve gives the sensor sensitivity (s). The limit of detection (LOD) is given by 3σ_b/s, where σ_b is the standard deviation of the blank EVOO measurements [65,66,70,71]. A linear fit (y = a + bx, where b and a represent the slope s and the y-axis intercept, respectively) to the calibration curve gives the correlation coefficient of determination (R²).

3. Results and Discussion

3.1. Sensor Characterization

In general, the polarizable nature of the liquid under test dictates the signal strength. For example, liquids with high water content (diploes) are expected to have very high capacitive signals due to high dielectric constant values (ε_{r ~}80) (e.g., milk, water, juice, etc.) while insulating liquids (e.g., oil, honey, etc.) devoid of water molecules/dipoles are not easily polarizable and have very low dielectric constant values (ε_r < 5) [72,73], which makes sensing mixed oils difficult, and more so when carried out non-intrusively because of the thick insulating package material acting as a barrier and reducing the denser electric field lines (near the electrode surface) from accessing the actual material under test (oil). Sensor characterization was performed to assess the sensor stability in air before actual adulteration testing to ensure that the unblended oil responses are separable, to begin with. The mean capacitance (µ) ± standard deviation (σ) of the capacitive sensor in air obtained from a recording of 1.5 h was ~55.675 ± 0.003 pF (air). The relative standard error expressed in percentage (σ_rse = σ/µ√n), n = 5664) was estimated to be ~0.006%. (Figure 2a). The low error percentage confirms the stability of the recordings at such a high data acquisition speed, which is promising for implementation in high-throughput quality check scenarios in FMCG lines. Figure 2b shows the real-time capacitance of a sensor in air, with an empty container and unblended oil-filled containers. The average capacitance (µ ± σ) of the sensor increased from 55.696 ± 0.0004 pF to 55.735 ± 0.0004 pF upon insertion of an empty container, and to 55.983 ± 0.064 for unblended oil-filled containers in the capacitive field. The relative standard error (σ_rse) estimated for air, unfilled, EVOO, OO, SOY, SF, VEG, CAN, and CAS, was 0.007%, 0.007%, 0.004%, 0.004%, 0.001%, 0.001%, 0.002%, 0.001% and 0.002%, respectively. Figure 2c shows the relative capacitance (ΔC/C₀) expressed in % for unfilled and unblended oil-filled containers with air as the baseline. The sensor displayed a positive capacitance change upon placing both empty and filled containers in the inter-electrode gap in the following order air < unfilled < EVOO < OO < SOY < SF ≈ VEG ≈ CAN < CAS. ΔC/C₀ of unblended oil-filled containers is at least a 6-fold (~0.52 ± 0.12%) greater compared to unfilled containers (~0.06%), which signifies the higher dielectric contribution of oil towards the overall capacitance compared to the container (refer to Section 3.4 and 3.5) since the container material has a very lower dielectric constant (~2–3 for PET [67,68]) and container wall thickness (~1.2 mm).

ΔC/C₀ for all unblended oils are substantially greater (ranging from 487 to 974-fold) compared to their relative standard errors (ranging from 0.001% to 0.007%), and 87 to 620-fold greater than the relative standard error of blank (air). Furthermore, ΔC/C₀ of OO, SOY, SF, VEG, CAN, and CAS, with respect to EVOO, increased by ~0.012%, ~0.197%, ~0.212%, ~0.209%, ~0.215% and ~0.355%, respectively. Even after replacing the baseline reference as EVOO, the observed responses for all oils remained significantly greater (3 to 230-fold) than their above-stated individual relative standard errors, and at least 3 to 86-fold greater than the relative standard error of EVOO (blank). Hence, for actual adulteration calibration (Section 3.5), EVOO replaced air as the blank measurement (baseline reference). Additionally, statistical hypothesis testing performed using two t-test confirms that the results are statistically significant with p < 0.01 for oil identification and classification (refer to Figure S1 in Supporting Information (SI)). Overall, the sensor characterization underscores the ample capacitive signal strength and the good separability between EVOO and adulterant oils, which is crucial for oil-type classification and adulteration detection.

3.2. Sensor Validation

The capacitance testing performed using FDC2214EVM was calibrated against a standard LCR system (Keysight, Impedance Analyzer, E4990A, 20 Hz to 30 MHz) to validate the obtained observed responses and trend. E4990A gives the absolute capacitance value of capacitive sensors whereas FDC2214 is intended for monitoring capacitance changes in applications. The absolute capacitance values of the sensor connected to FDC2214EVM and E4990A are expected to be different due to inherent system baseline differences from internal circuit configurations. A quick test showed that the change in capacitance obtained using FDC2214 evaluated using a standard 50 pF ceramic capacitor was found to be close to the absolute capacitance measured using E4990A (refer to Figure S2 in SI). Hence, the relative capacitance (ΔC_EVOO/C_EVOO) was used for comparison and validation of the sensing methodology and system. The cumulative capacitance responses from E4990A for all unblended oils with EVOO as blank showed good response and separability as seen in Figure 3, which is consistent with the earlier observed trend. Unlike the E4990A, which can perform the cable length compensation correction during calibration to accommodate the 1 m probe cable length, the FDC2214EVM is affected by the stray parasitic capacitance of the wirings to the terminals leading to slightly lower response levels. Nevertheless, the closeness of the observed responses using both systems validate the sensing methodology.

3.3. Sensing Methodology Optimization

The optimization of the capacitive sensing protocol is important to achieve good capacitive signal strength with observable separability between EVOO and adulterant oils before its extension to authenticity checking using binary solution blends. Such optimization is attained through experimental testing to find the right concoction of sensor configuration, sensing area and operational frequency.

FDC2214 allows multi-channel measurements (up to four) with each channel having two terminals (INA and INB) for excitation, thereby allowing two modes of operation for any given channel: (i) single-ended mode, where INA/INB is connected to one electrode and another electrode is connected to FDC2214 ground terminal; and (ii) differential configuration, where each sensor electrode is connected to one excitation terminal (INA and INB) of the corresponding channel. The single-ended configuration offers a greater sensing range but with reduced sensitivity compared to the differential configuration for a given total sensor plate area, thereby making it useful for displacement-based sensing applications [69]. On the other hand, the differential configuration produces a stronger sensing field but with a reduced sensing range, which is attractive for material property analysis (e.g., liquid level sensing) [69]. This was verified experimentally in Figure 4a, which shows that ΔC_EVOO/C_EVOO of adulterant oils are much higher and separable from EVOO for the differential mode vis-à-vis single-ended mode, which affirms that a stronger field encompassing the oil is responsible for the observed behaviour. Hence, a differential configuration capable of generating a strong electrostatic field was adopted in this work to detect low dielectric changes from oil mixing. The electrode area determines the strength of the electric field, coverage of the material under test and the penetration depth of the sensing field. Experiments were conducted to study the influence of electrode area and downselect the optimal electrode area to elicit the maximum response as shown in Figure 4b using four different electrode areas (A) (A1—100 mm², A2—168 mm², A3—468 mm² and A4—900 mm²). Reducing the electrode area decreases the electric field strength and penetration depth rendering it difficult to distinguish the oils. On the other hand, increasing the electrode area beyond a threshold lowers the overall response since most of the electric field dissipates onto the container’s empty overhead space and ambience (air has a very low dielectric constant), resulting in poor coverage of the actual contents. To develop low-cost yet high-speed detection systems for screening non-polar liquids, the selection of the correct excitation frequency range is critical. Most of the commercially available capacitance development boards offer an excitation frequency in the range of 10–100 kHz [65,66]. However, such low frequencies are insufficient to generate a strong electric field for assessing non-polar liquids. To demonstrate that a higher frequency is desirable for testing non-polar liquids, a comparison of the raw capacitances of the two FDC boards is shown in Figure 4c. FDC2214EVM with higher noise immunity and a higher frequency of operation (5 MHz) showed clear separability between the unblended oils while FDC1004EVM operating at a lower frequency (25 kHz) failed to show any distinction between the oils, which is consistent with the literature [64].

3.4. Sensing Mechanism

As mentioned earlier, capacitance (C) is related to the relative permittivity or dielectric constant (ε_r = ε/ε_o) of the material under test, where ε_o is the permittivity of the free space and ε is the permittivity of the material. Figure 5 shows the schematic of the capacitive sensing working principle for oil adulteration detection. ε_rB, ε_rO, ε_rA and ε_rAO denote the permittivity of air, good oil plus container, adulterant oil plus container, and blended oil plus container, respectively, and C_B, C_O, C_A and C_AO represent their corresponding capacitances. Since C ∝ ε_r, the presence of oil plus container as dielectric (ε_rO, ε_rA, ε_rAO > ε_rB) in the sensing area elevates the output capacitance (C_O, C_A, C_AO > C_B), where ε_rB ~1 (air has the lowest dielectric constant). The permittivity of the adulterant oil (ε_rA) dictates the sign of capacitance change (C_AO). Adulteration using higher dielectric oil (ε_rA > ε_rO) produces a positive change in capacitance (C_AO > C_O) and vice versa (C_AO < C_O) for a lower dielectric oil (ε_rA < ε_rO).

3.5. Sensor Response Calibration and Performance

There exists a close correlation between the density and fatty acid unsaturation with the dielectric constant.

Table 1. Dielectric constant, fatty acid content and density of all oils used in this study. Extra virgin olive oil−EVOO, olive oil−OO, soybean oil−SOY, sunflower oil−SF, vegetable oil−VEG, canola oil−CAN and castor oil−CAS.

Property	EVOO	OO	SOY	SF	VEG	CAN	CAS
Dominant unsaturated fatty acids (% m/m)	Oleic acid (55–83%) Linoleic acid (3–21%) [35,36,37,74,75]	Oleic acid (55–83%) and Linoleic acid (3–21%) [35,36,37,74,75]	Linoleic (48–59%) and Oleic acid (17–308%) [37,76,77]	Linoleic acid (52–65%) and oleic acid (26–36%) [37,76,77]	Oleic acid (20–29%) Linoleic (27–31%) [37,78,79]	Oleic acid (57–62%) and linoleic (15–22%) [37,76,77]	Ricinoleic (87–90) [80]
Density (kg/L at 20 °C)	0.909 [81]	0.909 [75]	0.931 [82]	0.932 [82]	0.925 [82]	0.917 [82]	0.959 [80]
Dielectric constant (εr)	2.81 [83,84]	3.005 [83]	3.020 [85]	3.020 [84]	3.01 [85]	3.015 [85]	4.5 [86]

summarizes the dielectric constant, density and dominant fatty acid composition present in the oils used in this study. The density of fatty acids increases in the following order: oleic acid < linoleic acid < ricinoleic acid (Figure 6). In common fatty acids found in oils (oleic acid and linoleic acid), the density increases with the degree of unsaturated bonds, i.e., cis-alkene functional groups (C=C bonds). The density of a substance in the liquid state is largely dependent on the geometry of the constituent molecules and its subsequent effect on the intermolecular forces. The relationship between the asymmetrical structure of the substituent groups across the bond and the overall dipole moment of the fatty acid dictates the degree of polarizability and density. As the number of cis-alkene bonds increases, so does the polarity of the molecule, allowing stronger dipole-dipole, resulting in the molecules packing closer together. Ricinoleic acid has extensive hydrogen bonding between its molecules lending it its high density. In general, the more mass per unit volume of oil, the more molecules there are to be polarized within the sensing field, leading to a higher dielectric constant. It is to be noted that CAN, VEG, SF and SOY shown in Figure 2c have very close capacitance values even though SOY and SF have higher dielectric constants and densities compared to VEG and CAN. This is due to the presence of both oleic and linoleic acid in significant amounts as opposed to the other oils with a clear dominant fatty acid. In general, the observed capacitance of oils is still in agreement with the density/dielectric property trend shown in Table 1. Since SF and VEG have similar fatty acid composition, dielectric constant, and capacitance response as SOY and CAN, respectively, they were omitted from the adulteration experiments.

Figure 7a–c shows the sensor calibration plots of binary mixtures generated by spiking EVOO with OO, CAN, SOY and CAS oils to various degrees (2.5–60%). The sensor response (ΔC/C) showed a positive relationship with the degree of adulteration. This is attributable to a combination of increased oil density and dielectric contribution from the adulterant towards the overall capacitance.

Table 2. Summary of sensor performance parameters for EVOO adulteration detection. Extra virgin olive oil-EVOO, Olive oil−OO, soybean oil−SOY, canola oil−CAN and castor oil−CAS.

Sample	Sensitivity (s)	Limit of Detection (LOD) (%)	Correlation Coefficient (R2)
EVOO + OO	9.14 × 10−5	32.8	0.918
EVOO + CAN	1.56 × 10−4	19.4	0.837
EVOO + SOY	2.91 × 10−4	10.3	0.926
EVOO + CAS	1.80 × 10−3	1.7	0.911

summarizes the sensor performance parameters (s, LOD and R²) extracted from the linear fitting of the sensor calibration plots in Figure 7 for the different adulterants. The increased composition of denser fatty acids like ricinoleic acid (in CAS) and linolenic acid (in SY) compared to oleic acid (in CL) contributes to the greater oil densification from mixing, which is responsible for the differences in the sensor sensitivity for the same degree of adulteration. Olive oil with its closeness to EVOO in chemical composition showed the lowest sensitivity and the poorest detection limit making it the hardest adulterant to detect using capacitance.

In general, to minimize variability during batch production in F&B FMCG and ensure the reliability of offline product quality tests, the ambient conditions are kept relatively constant. The testing ambience boundaries were set as air-conditioned (A/C, 25 °C and RH~60%) and non-air-conditioned (non-A/C, 31 °C, RH~90%) ambience, respectively, to closely resemble the offline laboratory and F&B production environment for tropical countries (such as Singapore). The effect of humidity and temperature on the sensor response was assessed separately. From Figure 8a, at constant humidity (RH~60%), the bare sensor capacitance drift constitutes only a 0.01% change for every 1 °C rise in temperature. Moreover, this change in response occurs in 1.5 h, thereby constituting an almost insignificant change of 1.85 × 10⁻⁵%s⁻¹. A rise in humidity at a constant temperature (25 °C) produces a steeper positive slope in the response from the additional moisture content (dielectric contribution) as shown in Figure 8b. The humidity-induced response albeit higher compared to the temperature-induce response during the same time frame of 1.5 h, is still insignificant since the baseline capacitance drift causes a 0.006% change for every 1% change in RH (3.27 × 10⁻⁵%s⁻¹). The change in total bare sensor capacitance caused by the drift between the two boundary conditions is ~6.48 × 10⁻⁵%s⁻¹. Overall, both humidity and temperature effects on the sensor are slow and have a very small magnitude, which is insignificant compared to oil-induced responses occurring in less than a second from the calibration curves in Figure 7. Hence, the sensor is expected to function without much loss in sensitivity for RH in the range of 60–90% and temperature between 25–30 °C. To further validate the above claims, 60% adulterated oils were tested at different humidity levels. Figure 8c shows the closeness of the ΔC_EVOO/C_EVOO values for different RH, which affirms the low impact of external ambience on the sensor response. soybean oil (SOY, FairPrice, Singapore).

The type of container (material and size) could affect the sensitivity of the sensor. Hence, trials were conducted directly on commercial EVOO containers available in the market to check the efficacy of the method for industrial implementation and further validate the sensing methodology. The commercial EVOO comes in two types of packaged containers, glass (1 L) and PET (2 L) with a container wall thickness of 5 mm and 0.6 mm, respectively. To account for the bigger size of glass (70 mm width), and PET bottle (110 mm width) compared to the small containers (15 mL, 27 mm width) used earlier, two rectangular electrodes with an area of 416 mm² area were used in a differential configuration. EVOO and 20%-SOY blended EVOO samples were tested on both containers and compared against the earlier results. Since the sensor area used was different, normalized (by area) relative capacitance (Air/EVOO as blank) was used for comparing the sensor response to container material/blending.

Figure 9a shows the two types of commercial EVOO (Naturel^®, Andalusia, Spain), sensor electrode placement configuration and sensor electrode design used for these trials. Figure 9b shows the normalized (by area) relative capacitance comparison (air as blank) of the three types of containers tested in this study. Clearly, the response of the glass bottle increased by 10 to 20-fold compared to the two PET bottles, which is attributable to the combination of greater wall thickness (5 mm) and higher dielectric constant (3–4) of glass compared to PET bottles (0.6 mm and 1.2 mm) with a dielectric constant of 2–3. Furthermore, the 2 L bottle response was 50% lower compared to the 15 mL vials due to the reduction in container wall thickness by half. The thicker glass container and higher dielectric constant of glass is expected to lower the sensor response to adulteration detection since the electric field strength is decaying with distance and the effective electric field reaching the contents inside is diminished. Similarly, the reduced thickness of the 2 L PET bottle is expected to increase the response to blending as more field is penetrating the oil contents. Figure 9c,d show the increase in sensor response upon blending as discussed under Section 3.4 and Section 3.5. Figure 9d confirms the above-mentioned postulations, wherein the sensor response to 20%-SOY blending was lowest in glass bottles and highest in the 2 L-PET bottles when compared to the 15 mL PET containers. This reduction in sensor response for glass bottles could be easily circumvented by increasing the electrode length (vertically) to enhance the penetration depth of the sensing field.

3.6. Automated EVOO Adulteration Detection System

The graphical user interface (GUI) of FDC2214EVM is limited by its re-programming capability, which prevents its implementation for developing fully automated systems for inline implementation and remote monitoring. Hence, for further development, the USB to I2C serial communication for data acquisition was performed using Arduino instead of the FDC2214 Sensing solutions GUI and the FDC2214 sensing module was re-designed in-house and fabricated in a foundry (JLCPCB, Shenzhen, China).

Figure 10a shows the Arduino-based sensing setup for automated detection of adulteration in EVOO. The in-house designed FDC2214 was interfaced with Arduino IDE via Arduino Pro Mini (3.3 V, 8 MHz, Sparkfun, Niwot, CO, USA). The Arduino Pro Mini was connected to a laptop via an FTDI basic (3.3 V, 8 MHz, Sparkfun, Niwot, CO, USA) breakout chip for USB to serial IC data transmission. The baud rate was set to 51,200 and the data acquisition rate was 100 ms. RealTerm serial terminal open-source software was used for the data logging process. A relative capacitance (ΔC_EVOO/C_EVOO) threshold (>0.015%) was set in the Arduino code with a LED output showing two states for EVOO (OFF) and adulterated EVOO (ON), respectively. Kst real-time open-source dataset plotting tool was used for live data visualization. The response of the in-house developed Arduino-based system is comparable to the FDC2214EVM as shown in Figure 10b, which confirms the consistency of the earlier observed trend and demonstrates the successful translation of the feasibility study to a preliminary prototype. This Arduino-based system offers several advantages in terms of cost, footprint and multiplexing summarized in

Table 3. Comparison of FDC2214EVM and in-house designed FDC2214 with Arduino.

Criteria	FDC2214EVM	Arduino + FDC2214
Cost (USD)	510	54
Footprint (cm2)	25	17
Multiplexing FDC (channels)	Not possible (4)	Yes (32)
Automated threshold detection	No	Yes

. Figure 10c shows the photographs of the FDC2214EVM and the components of the in-house designed FDC2214 system. The newly developed system is 89% lower in cost and 32% smaller in footprint, compared to the FDC2214EVM-based system, respectively. Although the FDC2214 used in both systems have four channels, the Arduino-based system offers greater flexibility in terms of multiplexing for multi-channel monitoring, which could significantly boost the inspection throughput. Such an Arduino-based system further allows multi-channel cum multi-product monitoring as shown in Figure 10d. Four FDCs (FDC1, FDC2, FDC3 and FDC4) were multiplexed using one in-house designed TCA9548A, 1-to-8 I²C multiplexer (Adafruit, New York, NY, USA) to generate 16 active channels. Channel 1 (CH1) of FDC1 and FDC2 were connected to two different (sensor area and inter-electrode gap) in-house designed sensor electrodes to monitor two separate non-polar liquids (honey and oil) packaged in different containers (size, shape and wall thickness) simultaneously. The different baselines and capacitance changes seen in the live plotting tool (Figure 10e) for the two liquids affirm the detection of different packaged liquid dielectrics. It is possible to operate up to 8 FDCs yielding 32 active channels using a single multiplexer, which is not possible in FDC2214EVM due to the limitations of the dedicated GUI. Herein, the components are designed separately and connected via wires to evaluate the proof of concept. Hence, further shrinking of the footprint and reduction in system cost is possible by integrating all individual components on a single PCB board for batch fabrication.

3.7. Comparison with Other Industrial Adulteration Detection Methods

Spectroscopy and chromatographic techniques are widely used industry standards for the authenticity check of edible oils [87].

Table 4. A benchmark table of the proposed capacitive detection against the standard industrial edible oil authenticity testing methods.

Criteria	Spectroscopy	Chromatography	This Work
Response time	5 s–1.5 h	15 min–40 min	<1 s
Sample extraction	Yes	Yes	No
Throughput (samples/min)	<10	0	>60
Cost/system (USD)	1200–150,000	3000–15,000	50–500
Detection limit (%)	<1% [88,89,90]	<1% [91,92,93]	>10% (except for castor oil > 1%)
Footprint	Large	Very Large	Small
Deployability	Offline	Offline	Inline
Waste generation	Yes	Yes	No

shows a comparison of the proposed capacitive system against the current industrial standards employed for oil adulteration detection. As expected, the detection limits of our proposed method are lower compared to traditional methods due to the presence of the container/packager barrier between the oil and the sensing surface. On the other hand, the existing industrial techniques, despite showing low detection limits, require sample extraction, specific sample volume, sample treatment/preparation, dedicated cuvettes and skilled operators to perform offline analysis ex situ, which significantly increases the screening time. Consequently, the number of samples inspected is reduced, which results in unnecessary batch rejection of samples based on the output of random sampling. The proposed system outperforms existing industrial methods in terms of cost, footprint and detection throughput while offering a novel non-destructive tool to aid quick screening for packaged products.

4. Conclusions and Future Work

In summary, the proposed non-destructive capacitive detection scheme successfully classified good and adulterated oils non-invasively without requiring any sample extraction via through-package/container testing. The proposed method indirectly relies on the differences in dielectric properties of the oil from blending to make qualitative and quantitative assessment of oil capacitance deviations due to blending without complex requiring complex dielectric calculations and sophisticated instrumentation. The adulteration detection limit and sensitivity of the capacitance sensing system depend on the nature (density, fatty acid content and dielectric constant) of the adulterant. The fast detection time and sensitivity achieved were achieved using simple electronic circuitry making it cheap and scalable for large-volume and high-speed production in FMCG lines. The non-invasive screening allows 100% inline package/container inspection for oil quality check, without the need to tamper with the container/package sealing for sample extraction thereby reducing wastage from batch re-jects. The major advantage of this quick screening method would be in situ quality control check, without overwhelming the few available laboratories with unnecessary requests. In other words, if the capacitive-based analysis gives a ‘positive’ result (i.e., adulterant might be present), then further specific analyses could be requested from dedicated laboratories for finding chemical composition; on the other hand, if the result is negative, then there is no need to carry out additional expensive and time-consuming analyses. This tool is not aimed at quantitative estimation of the chemical composition of EVOO fatty acids due to adulteration, or dielectric permittivity evaluation, instead, it focuses on providing a platform for rapid in situ qualitative screening, which is absent in current food industries.

Some of the limitations of the proposed system include its inability to perform authenticity checks through metallic containers (very few edible oils are packaged in metal jars) as the metallic body acts as the ground and drains the capacitive field. A possible solution to overcome this is to couple the capacitance electrodes with inductance to measure the impedance of the container with oil under test. Additionally, the applicability of the proposed method to edible oils in general needs evaluation on a case-by-case basis considering the vast range of adulterants and variations in oils due to pre- and post-harvest factors.

The current work involves in-house PCB design of a fully integrated capacitance board with all the components (electrodes, Arduino, FDC2214 and multiplexer) assembled on board to further reduce the cost and footprint. The future work would involve the software and hardware design for automation of a standalone capacitance-based adulteration testing prototype system with a pass/fail criterion suitable for analysing moving packaged containers with different edible liquids (e.g., honey, milk, etc.).