Real time operating system based electronic nose was developed with an array of commercial metal oxide semiconductor (MOS) sensors and homemade zinc oxide based sensor, for real time quality analysis of raw milk. The sensors were empanelled and calibrated towards various concentrations of volatile organic compounds (VOCs) such as ethanol, hexanal, acetic acid, acetaldehyde, etc., which are responsible for off-flavors in milk produced due to microbial contamination, chemical reactions and genetic effects in cow. In addition to MOS sensors, an electrochemical based carbon dioxide (CO2) sensor was incorporated in the sensor array and calibrated towards various concentrations of CO2 to detect the spoilage of milk due to mastic disease in cows. The selectivity and discrimination efficiency of developed electronic nose was confirmed with principal component analysis.
Natural odours are complex mixtures of small organic molecules with relative molecular masses in the range of 18 to 300 microns [1] that are carried by air along with many other chemical compounds. The odour properties, including perceived intensity, of a molecule depend on the shape, the size and characteristics of the polar group, so that a low volatility species can be sensed intensely, while the other present in high concentration can have only a low intensity smell. Olfaction may recognize a harmful substance, but more frequently it will be sensitive to natural and harmless compounds that result from biological processes. Electronic nose, a mimic of human olfaction, plays a decisive role in food quality discrimination, where human sense of smell fails due to fatigue, narrowness in detection limit and exposure to toxic organic compounds. Since quality control of starting milk is one of the most important tasks in the production of dairy products [2] and chemical analysis in the milk remains a great challenge to the sensory evaluation panel for many decades due to the heterogeneous nature of milk [3], a need for rapid and cost effective discrimination of milk and dairy products becomes substantial. At this juncture, electronic nose, an array of gas sensors with pattern recognition system came to the scenario in order to replace the highly sophisticated analytical instruments (for example, mass chromatograph, FTIR and UV-Visible spectrometer) whose usage is being limited mainly because of their higher cost and lack of portability, despite their accuracy. Analysis of milk aroma is the best promising tool to obtain information about the milk and one of the best methods to isolate the flavor compound is headspace analysis [3]. Headspace of milk from a healthy cow typically contains complex mixtures of VOCs (acetone, 2-pentanone, 2-butanone, toluene, limonene, heptanal, etc.) at various concentrations [3-5]. The concentration of these VOCs varies due to the factors such as bacterial metabolism, ageing, photo-oxidation and presence of pro-oxidant metals such as copper, iron and nickel.
Milk from a healthy cow contains only few bacteria which, may multiply and the rate of multiplication will increase, as time passes from the time of milking to the time of processing; it also depends on the standard of milking, handling practices and storage. These bacterial growths, lead to the spoilage of milk with the production of off-flavor. As the microbial spoilage of starting milk severely affects the industrial quality, due to undesirable aroma, physical defects and metabolic toxicity, identification of spoilage in milk before production of products becomes mandatory. Likewise, spoilage of milk due to various other factors should also be identified at the earliest stage to avoid complications and complaints in the final product. This can be achieved by early detection of VOCs at headspace [5] produced during spoilage. Fox example, milk spoilage due to microbial/bacterial contamination can be identified from the presence of acetaldehyde, 3-methyl-1-butanol, acetic acid and ethanol [4, 6] in headspace of the milk, which are absent in raw milk before spoilage. Likewise hexanal, pentanal and dimethyl disulfide [4] are good indicators of fluorescent light damaged products. Whereas pentanal, hexanal, heptanal and isopentanal will indicate copper induced oxidation in milk [7] ageing of milk can be significantly identified through the presence of dimethyl sulfide, ethyl acetate, 2-heptanone, pentanal, etc. [8]. Ampuero et.al, [4] observed that the headspace of milk from the cow bearing genetic defect contains trimethyl amine. And Erikssona et.al, [9] observed that the milk from the cow affected by mastic disease contains higher concentration of CO2 than the healthy reference milk.
From the literature it is noted that, so far to discriminate the quality of milk and their derivatives, sensors of various kind were exposed to the headspace of milk samples of various ages or stored under various condition or from various farms after treatments and most widely used, pattern recognition methods like principle component analysis (PCA) [2, 10] or artificial neural network (ANN) [4, 11] has been carried out. Though PCA is a traditional method of analysis and reduce the dimensionality factor, during real time analysis it is not user friendly. These require tedious training processes and hectic sample collections.
Hence in order to make sensor systems portable, cheaper or usable for monitoring food quality, sensor elements and system as a whole: should require low power for operation; should have no need for external carrier or reactant gases; should be made compatible with microelectronic technologies; should have reversible and reproducible responses; should have adequate sensitivity and should have response times lower than one minute and long term stability if used in continuous operation or response times of a few seconds. With this motivation in mind a Real Time Operating System (RTOS) based embedded electronic nose system that can detect the volatile organic compounds (VOCs) as given in Table 1 which are good marker for milk spoilage due to various reasons has been developed.
Since the electrical properties of the metal oxide semiconductors vary [12-14] according to the environment to which they are exposed through the process of adsorption / desorption, the metal oxide semiconductor (MOS) based sensors are found to be suitable candidates for detecting VOCs / gases at various concentrations in the range of parts per million (ppm). Hence commercial MOS and homemade sensors were used to form the sensor array in embedded electronic nose. Further to avoid complications in the final embedded model static headspace analysis was carried out.
II. MATERIALS AND METHODS
As TGS 2620, 822, 813 (FIGARO, Japan) which are made of SnO2 (n-type material) [15] and homemade nanostructured zinc oxide (ZnO) thin film based sensors are sensitive towards aromatic compounds of our interest, they were empanelled to form the array to construct an electronic nose. The nanostructured ZnO n-type material [14] thin film was coated on an ultrasonically cleaned glass substrate using spray pyrolysis technique by dissolving 0.05 M of zinc acetate dihydrate (99.9% purity, Merck, India) in 50 mL of deionized water at an optimized condition. The Structural characterization of the film was carried out using X-ray diffractometer with Cu Ka radiation of wavelength 1.5418 Å (D8 Focus, Bruker, Germany) at the scanning rate of 0.02°/min and generator setting of 40 kV. Surface morphology was obtained from field emission SEM (FE SEM, 6701F, JEOL, Japan). The X-ray diffraction pattern of as-deposited film shown in Fig. 1 (a) indicates polycrystalline nature of film with hexagonal wurtzite structure. The peak positions are well in agreement with JCPDS (File no 36-1451). The SEM micrograph shows spherical shaped well connected nanograins and is shown in Fig.1 (b).
Electrical characterization of these sensors was performed with the home made testing chamber. The testing chamber consists of 6500cc capacity glass jar with a septum provision at one side of the jar to inject desirable concentration of VOCs/gases and another side of the jar had a provision to connect the chamber to vacuum pump. The testing chamber has been pumped down to low pressure, to enhance the response of sensors towards specific VOC.
Each sensor element has been individually calibrated towards various concentrations of VOCs. The calibration procedure initiated with the preheat treatment [15-19] of sensor element in order to remove the ambient moisture. The sensors were allowed to reach equilibrium condition by placing it in dry environment. Equilibrium state was determined by evaluating the stability of the sensor response with respect to time. If the measured resistances appear to be stable then a steady baseline is recorded and known concentrations of VOCs namely acetaldehyde, acetic acid, dimethyl sulfide, hexanal, ethanol (Merck, Germany: purity - 99.8%), trimethyl amine (Alfa Aesar, USA: purity - 99.8%), were injected into the glass chamber through the calibrated syringe via septum provision individually. The consequent response of the sensors in that environment was observed under optimized condition. After each exposure, dry air was allowed into the testing chamber to remove the VOC from the sensor environment and reestablish the same baseline for the next chemical exposure. Each measurement was repeated three times in order to exactly estimate and calibrate the response of sensors towards specific VOC and to make sure the reproducibility of the sensor response.
Since the presence of higher concentration of CO2 is a good marker for milk affected by mastic disease, carbon dioxide sensor, TGS 4161 (FIGARO, Japan) [20] was inducted as a fourth sensor in the electronic nose. The block diagram of the setup used to calibrate CO2 gas sensor is shown in Fig.2. The preheated sensor was placed inside a high vacuum gas-testing chamber (HINDHIVAC- 12A4D Model). Desired concentrations of CO2 was admitted inside the chamber along with nitrogen (N2) as carrier gas [21,22] through mass flow controllers (AALBORG-USA) and the response of sensor towards various concentrations of CO2 was acquired using (Keithley- 6517A) electrometer, which is integrated with the personal computer.
From the observed sensor response, calibration table has been formed to construct the look-up table, which is being embedded in the microcontroller AT89C51rd2. In the lookup table formation, different threshold levels were fixed as per the response of individual sensor in array towards selected VOCs along with it response and recovery time.
For real time quality analysis of milk and dairy products, sensor signals were conditioned using op. amp IC TL279 and converted into their respective digital data using IC AD0808. Response of all the four sensors was collected. And the sensor, which showed steep variation to that particular real time environment in the sensor array, was identified by comparing the response of each sensor to their respective zero ppm level. Further, the response of the identified sensor was compared with the corresponding calibrated data in the embedded database for decision-making.
In the present embedded approach, final result was fixed not only based on the steep initial response of a particular sensor but also depends on the response of the other sensors in the array which showed significant gradual variation with an increase in concentrations of that particular VOC/gas. According to the resultant data and the corresponding threshold limit for decision-making, final result was displayed in the LCD module and the same was stored for further processing. This was effectively carried out using Keil Compiler and RTOS software RTX-51. In the RTOS based embedded coding, round robin task scheduling was provided to read the sensor signal at a particular time interval (based on the response and recovery time) and hence selection of particular sensor output from an array was achieved. In addition, the voltage output from each sensor was compared at different time intervals after the exposure to headspace.
The efficiency of developed model in real time quality analysis was studied by collecting raw milk from local farms without any preservatives and heat treatments. The sensor array was exposed directly to the headspace of milk samples at 3 cm distance and the response of each sensor was observed. Further to reconfirm the efficiency of embedded model in discriminating milk quality most widely used traditional PCA was performed using MATLAB with the data corresponding to various concentrations of VOCs and response of milk during real time analysis. Further, the efficiency of the model was crosschecked through gas chromatography (GC) (PerkinElmer Clarus500) with Elite-5ms capillary column (5%Phenyl 95% dimethylpolysiloxane) of length 30m and id 250µm. He @ 1ml/min was used as carrier gas with an oven and needle temperature of 308 K and 363 K respectively.
III. RESULTS AND DISCUSSION
In the current investigation, VOCs which are predominantly responsible for the undesirable aroma / off-flavor during milk spoilage were presented to the sensor array and response of the sensor array was observed. In this section methodology carried out to form the look up table and the database for decision making during real time analysis of milk quality is discussed.
A. Response of sensors towards VOCs
Various concentrations of hexanaldehyde were presented to the sensor array and the response of each sensor was observed. From the response curves of sensors it was observed that as the concentration of hexanaldehyde increases, voltage drop across load resistance (VRL) in the sensor's measurement (15-18) circuit increases. This is because, when MOS exposed to the environment having VOCs, oxygen (O2) get adsorbed on the surface of MOS by removing electrons from MOS as shown in equation (1). Further the O2- on the surface of sensing element react with VOC and give up electrons to the sensing material as expressed in equation (2).
O2 + 2 e-(MOS) O2- (1)
(X+) + O2- (Y) + e- (2)
Where, X and Y represent the analyte and resultant product respectively. This results in an increase in charge carriers of the sensing element, which in turn increases the conductivity of MOS in the presence of VOCs [14]. Since the load resistance was connected in series with the sensor element, the voltage drop across. Therefore change in the value of VRL at each concentration was observed.
Similarly various concentrations of DMS, acetaldehyde, acetic acid, ethanol and TMA were presented to the sensor array and the response of each sensor was observed and is shown in Fig. 3. From these response curves of sensors it was observed that in the case of hexanaldehyde, as the concentration of hexanaldehyde increases, TGS 822 sensor shows gradual variation. Likewise for acetaldehyde and acetic acid, TGS 2620 and 813 exhibited gradual variation respectively. Hence response of TGS 2620 was taken in to consideration for decision making in the case of presence of acetaldehyde and 813 for acetic acid.
By adopting the same procedure, the response of the sensors array towards ethanol, DMS and TMA were observed and the sensor TGS 813 found to respond well for ethanol. In the case of DMS all the sensors exhibit higher sensitivity. Hence response of all the sensors in the array was taken in to consideration for the detection of DMS.
Compared to other sensors in the array, response of ZnO thin film towards TMA was found to be appreciable with better sensitivity and is shown in Fig.3 (f). It was observed that the electrical resistance of film decreases as the concentration of TMA increases, due to the formation of product with an electron as given by following mechanism [19].
(CH3)3N + O2‑ NO2 + CO2 + H2O + e- (3)
B. Response of sensors towards CO2
Among all the sensors in the array, response of TGS 4161 sensor towards various concentrations of CO2 is found to be appreciable and is shown in Fig. 3 (g). It has been observed from the response that as the concentration of CO2 increases VRL of sensor's measurement circuit has also been increased. This nature of response is due to the solid electrolytic type of TGS 4161 which functions similar to a battery i.e., as concentration of CO2 increases VRL of the measurement circuit increases. The sensing element of TGS 4161 consists of cation (Na+) solid electrolyte formed between two electrodes together with a printed heater (RuO2) substrate. The cathode (sensing element) consists of lithium carbonate and gold, while the anode (counter electrode) is made of gold. When the sensor is exposed to CO2 gas, the following electrochemical reaction occurs [20].
Li2CO3 + 2Na+ Na2O + 2Li+ + CO2 (4)
The calibrated values of TGS 4161 sensor is also integrated in the embedded database as standard reference to differentiate the healthy milk from mastic milk by detecting higher concentration of CO2 during real time analysis.
IV. PERFORMANCE VERIFICATION OF DEVELOPED ELECTRONIC NOSE DURING REAL TIME ANALYSIS
To verify the performance of the developed electronic nose, the sensor array has been exposed to headspace of milk samples and the response of the sensor array has been observed. From the response, it was noted that as time increases, response of TGS 2620 and TGS 822 sensors was predominant. Milk which was tested in the present work is from a single healthy cow without any genetic defects and mastic disease, hence bacterial contamination in milk started only after 7 hours from milking. Based on the response of the sensors and decision making commands in the look up table, the developed electronic nose displayed the spoilage level through LCD panel, which indicate the presence of ethanol.
A. Confirmation through principle component analysis and gas chromatography
Principle component analysis (PCA) was performed with sensors in sensor array and response of the sensor array towards various concentrations of VOCs as well real time analysis as first and second principle component respectively, to confirm the efficiency and effectiveness of developed electronic nose towards raw milk quality discrimination. From the PCA shown in Fig. 4, it is observed that milk after 7th hour falls in the cluster labeled ( ). This cluster contains VOCs responsible for off flavor in milk due to various spoilage factors. The resultant response from PCA confirms the spoilage of milk due to bacterial contamination, which is in well agreement with the response of developed embedded based electronic nose as shown in Fig 5. Where, TGS 813 response confirms the presence of ethanol, which is a good marker for microbial/bacterial contamination in raw milk.
The same milk sample was taken in a 5ml vial and gas chromatography was carried out. The results confirm the presence of ethanol which is shown in Fig. 6. Hence the developed electronic nose with calibrated sensors array, conditioning circuit and microcontroller system is found to be efficient in real time quality discrimination of milk.
V. VALIDATION OF DEVELOPED PROTOTYPE MODEL
Validation of the developed prototype model has been carried out in two phases. In first phase, validation was done by injecting very low concentrations (4-10µL of VOC in 30 mL of milk) of VOCs (ethanol, acetaldehyde, acetic acid, TMA, DMS, hexanal, etc.,) directly in to milk, which are responsible for off-flavors during milk spoilage and by detecting the same. During this phase, milk samples were collected from local farms and experiments were carried out over a period of 25 days repeatedly with the sample from same farm to confirm the reproducibility.
From these studies, it is observed that the texture of the milk hasn't changed even at higher concentrations (> 50ppm) levels of, ethanol, TMA, DMS and hexanal. On the other hand at higher concentrations (> 50ppm) of acetic acid and acetaldehyde the texture changed. In the case of acetic acid, the milk curdled and in the presence of acetaldehyde the colour turned to pale pink as shown in Fig.7.
From this observation, it is clearly understood that the presence of lower concentrations (<50 ppm) of VOCs like acetaldehyde, acetic acid, ethanol, TMA, DMS and hexanal can not determined from the change in colour or texture.
In phase two, in order to confirm the efficiency of the developed prototype model in detecting the off-flavors generated at initial stage of milk spoilage due to bacterial contamination, ageing, mastic disease, genetic defect and light damage were studied by collecting various milk samples from local dairy where nearly 25 lakh liters of milk from the near-by 40 centers were collected daily and processed further before it is being packed for sale at user end. In the dairy, the test performed on a routine basis are to find components, added water, inhibitors, plate loop count, sediment analysis and somatic cell count. Analysis of milk samples were carried out by trained people through standard methodologies like, chemical test or by using litmus paper or by heating the sample. All these methods are found to be time consuming. Also the accuracy depends on the trained people. Therefore the developed prototype model is employed to determine the quality of milk. The experiments were carried out for a period of 2 months repeatedly to confirm the efficiency and reproducibility of the model with the samples from various farms.
The milk quality detection efficiency of the developed prototype model is found to be appreciable compared to the regular tests carried out in the dairy by the experts and trained people.
VI. CONCLUSION
An RTOS based embedded approach has been carried out with the empanelled sensors TGS 822, 2620, 813, ZnO based chemical sensor and 4161, which are calibrated towards various concentrations of VOCs (hexanal, acetic acid, acetaldehyde, dimethyl sulfide, ethanol, trimethyl amine) and CO2 for real time quality discrimination. Since the VOCs/gases are major contributor for off-flavors in the milk and their derivatives, headspace analysis is found be a desirable method for quality discrimination. Therefore, quality discrimination of raw milk was carried out on the basis of database formed with the calibrated data of sensors and fixed threshold limit. The resultant embedded electronic nose was found to be cost effective, portable and user friendly with significant sensitivity and appreciable selectivity. PCA was carried out to confirm the efficiency of developed embedded based electronic nose and the results were found to be in well agreement with the developed embedded system. Hence the developed embedded based model can be effectively used for real time quality discrimination of milk with substantial efficiency.
