Open Platform Communications Assignment Sample

• Subject Name : Communication

Results and Discussion: Open Platform Communications

The following sections present the performance evaluation of the low-cost OPCs prior to the correction and the subsequent results post their machine learning field calibration. Firstly, the performance of the low-cost OPCs is compared against the reference instruments. The assessment for the same is discussed as per the criteria mentioned in the CFR Parts 58- Ambient Air Quality Surveillance (Subchapter C) of EPA. Thereafter, the results of (1.) and (2.) of the field calibration method are discussed. Finally, the cumulative results of the proposed calibration method are compared with the reference concentrations in the test dataset to determine if the low-cost OPCs meet the EPA’s criteria after the field calibration.

OPC stands for Open Platform Communications. It is referred to a collection of specifications and standards for the purpose of carrying out industrial telecommunications. The original standard was developed in 1996 by an industrial automation task force. This was developed under the name of OLE for Process Control. OPC helps to establish a strong relationship between control devices manufacturers by different organizations specializing in manufacturing. Once OPC Foundation was initially released in 1996, it was created in order to maintain a certain standard. The name was changed to Open Platform Communications in the year of 2011, as it has been adopted beyond the areas and specializations of process control. The area covered by OPC is increasing day by day with its applications in discrete manufacturing, building automation, process control and more. Also, it has grown beyond its original field of Object Linking and Embedding. It has a wide application including implementation in Microsoft’s .NET Framework, XML, and the TCP format.

The specification of OPC was founded on COM, OLE, and technologies of DCOM that are developed by Microsoft. These technologies are developed for the family of Microsoft Windows operating system. A standard series of objects and interfaces are defined by the specification. These are useful for the purpose of writing as well as reading real-time data and information.

There is a rise in organizations nowadays that have become engaged in commercializing sensors of low cost, also known as low-cost sensors (LCS). These devices are capable of monitoring and analyzing outdoor air pollution. The primary benefit here of using low cost sensors is the increased range of special coverage while carrying out air qyality analysis and their monitoring within cities and other remote locations. At the present time there are hundreds and thousands of low-cost sensors available commercially in the market ranging from several hundred to several thousand dollars. These price ranges of low cost sensors have been set considering different individual needs and requirements so that the best device can be chosen from the wide range of options available in the market in order to meet specific individual or organizational needs. At the same time scientific studies are being carried out in order to make them better in terms of efficiency and reliability on those devices. Independent evaluation has been reported according to scientific literature regarding the independent performance analysis and evaluation of LCS against a given set of measurements taken as reference for approximately 110 low cost sensors.

Moreover, according to the reports it has been found that low cost sensors can sometimes create hindrances as they can be unstable sometimes. These devices can get affected by atmospheric conditions, such as acquiring cross sensitivities from compounds that are interfering in nature. This may change the performance of the low cost sensors depending on the location of the site. The performance of these large numbers of sensors can be reached by carrying out effective quantitative analysis. The initial steps include carrying out surveys involving questionnaires that will help to gather individual views and data regarding these low cost sensors. As personal experience will be involved here, the chance to get biased and altered responses will be a lot lower. Other methods such as in-depth interviews can be conducted as well with the experts to grasp the ideas of low cost sensors and their practical application in day to day activities. Published reports and relevant testing laboratories can be used for this purpose along with peer-reviewed journals that consists of testing of various types of low cost sensors in the given research papers.

While carrying out the comparison of the LCS systems against the given reference standards relevant metrics of the topic highlighted the low cost sensors that are most cost effective and can be utilized to analyze and monitor the air quality and the pollutants present in the air in a given region or area. These readings should consist of a good level of agreement that is represented by determination coefficient of R2>0.75, having a slope near to 1.0. The possibility to possess a versatile and capable low cost sensor is highlighted in the review, having the ability to operate and function with more than one pollutant, and preferably with the help of low cost sensor data treatment. This will help to unlock the untapped potential of these devices which will help to make individual lives a lot easier and better in terms of efficiency, reliability, and managing time. Various fields of science and technology will be unlocked due to this leading to revolutionary advancements and discoveries in the given field.

On the other hand indoor air quality can be determined with the help of low cost sensors as well. Indoor air quality describes the quality of air and the level of pollutants present in a building or an enclosed interior space. Microbial contaminations play a significant role in affecting the indoor air quality (IAQ) of an area. Moreover, these microbial contaminators depend on several factors such as the humidity, temperature, and the gaseous pollutants present in the building or room. These gaseous pollutants include carbon dioxide, carbon monoxide, organic compounds that are volatile in nature, and more. Adverse health effects are common among the occupants in those areas. This in turn, can be minimized with the aid of an air quality monitoring system. A low cost sensor might be helpful here combined with other apparatuses and parts to determine the intensity if pollution within the building. Once this is identified, other measures such as purification of air and removing the pollutants from the air can be thought of.

The OPC-N2 calculations of particle mass assume a negligible contribution from particles below approximately 0.38 μm.”31 However, ignoring their contribution can lead to a severe underestimation in the PM concentrations. To prove this hypothesis, the contribution of sub-0.38 µm particles to the PV for the whole data set, measured using the reference instruments, was calculated. Figure 9 presents the box-plot of this contribution. The average percentage contribution of sub-0.38 µm particles was found to be ~40% for PV2.5 and ~27% for PV10. Under no circumstance can this contribution be assumed as negligible. Additionally, the majority of the low-cost optical-based PM sensors tend to neglect this contribution because of their inability to detect particles in this range. Therefore, it becomes vital for all such sensors to perform equivalent corrections, as proposed, before forecasting size-integrated concentrations.

Performance Evaluation of OPC-N2

Fig.2 portrays the comparison of low-cost OPCs’ firmware and calculated PV2.5 and PV10 relative to the reference concentrations for the entire data collection campaign. In this duration, when the ReNEWW house was occupied, the OPCs’ raw calculated PV2.5 mean for a 30-minute time-averaging window was 0.52 µg/m3 (OPC 1) and 0.54 µg/m3 (OPC 2). For the same duration and time-averaging window, the mean PV2.5 measured by the reference instrument was 2.65 µg/m3. This suggests a major underestimation of PV2.5 calculated from both the OPCs. Similar findings were observed for OPCs’ calculated PV10 as well (Mean concentrations - OPC 1: 2.84 µg/m3, OPC 2: 2.76 µg/m3 and Reference: 7.45 µg/m3; 30-minute time-averaging window). However, the firmware outputs of the OPCs showed a little improvement in measurements compared to the raw calculated concentrations (Mean PV2.5 - OPC 1: 1.91 µg/m3, OPC 2: 2.02 µg/m3; Mean PV10 - OPC 1: 3.87 µg/m3, OPC 2: 4.37 µg/m3; 30-minute time-averaging window). This is because the selected OPC performs a proprietary self-calibration for its firmware outputs [17]. Nevertheless, despite the improvements, the firmware outputs also underestimated the particle concentrations compared to the reference values.

To quantify the difference in OPCs’ and reference instrument’s PV measurements, the accuracy metrics discussed in the previous section were estimated. Each of the accuracy metric is calculated for PV concentrations averaged over 30-minute window, and for the period when ReNEWW house was occupied. The MAPE for comparison of OPCs’ calculated PV2.5 relative to the reference PV2.5 was observed to be 79.81 % (OPC 1) and 78.97 % (OPC2). For calculated PV10, the MAPE was found to be 67.67 % (OPC 1) and 67.59 % (OPC 2).

The ReNEWW house remained unoccupied between 15th December 2018 and 5th January 2019. For this duration, substantial lower PV concentrations were recorded by both the OPCs (Mean PV2.5 - OPC 1: 0.15 µg/m3 (Calculated) and 0.53 µg/m3 (Firmware), OPC 2: 0.14 µg/m3 and 0.49 µg/m3 (Firmware); Mean PV10 - OPC 1: 0.36 µg/m3 and 1.57 µg/m3 (Firmware), OPC 2: 0.34 µg/m3 and 1.67 µg/m3 (Firmware); 30-minute time-averaging window). The reference instruments, being costly in nature, were not operated during this period.

While measurements provided by OPCs (calculated and firmware) differed substantially from the reference concentrations, the OPCs were observed to behave similarly with each other throughout the measurement campaign (Fig.2). To confirm this hypothesis, the mean CV between PV concentrations (averaged over 30-minute window) measured by both OPCs (calculated and firmware) was calculated for the entire data collection campaign. The mean CV values for calculated and firmware PV2.5 were 4.82 % and 4.07 %, respectively. For PV10, CV values observed were 4.94 % and 8.73 % for calculated and firmware concentrations, respectively. These reported CV values were within the limits recommended by EPA (in the CFR Parts 58- Ambient Air Quality Surveillance (Subchapter C)). This indicates that both the OPCs were consistent with respect to each other in terms of output concentrations. In a laboratory analysis, Sousan et al. [17] reported a similar range of mean CV values for the selected OPC’s PM concentrations (4.2% to 9.5%).

Although a disparity between the absolute measurements given by the OPCs and reference instruments can be observed in Figure 2, they appear to follow a similar pattern. In order to better understand this, the average within-day variation of the PV concentrations from all the instruments were plotted and shown in Figure 3. The values shown represent the PV concentrations given by the instruments averaged at each hour of the day. A general trend can be seen here in the plots where low PV concentrations are observed from 12 AM to 6 AM, and the concentration gradually increases as the day progresses. Such a trend is observed as there are limited human activities during the night hours, which increases during day and evening time. Also, both the calculated and firmware PV values given by both the OPCs seem to agree very well with respect to each other. This corroborates the low CV values obtained for the OPC-N2. In addition, both the OPCs were successful in capturing the peaks corresponding to the reference concentrations. However, in terms of absolute value, the OPCs seem to under- estimate. To quantify the difference in measurements, the accuracy metrics discussed in the previous section were estimated.

During other times, when all the sensors and reference instruments were active, the r, R2, slope, MAPE and p-values of t-test for the OPCs’ firmware and calculated PV values against the reference concentrations were calculated. Table 4 summarizes the results.

The slope and r values estimated for both calculated and firmware concentrations, given by both the OPCs, did not meet the EPA’s performance criteria. Also, lower values of R2 and higher MAPE values are observed throughout. Finally, the p-values for the t-tests are <0.05 for all the measurements. Therefore, at 95% confidence, it is statistically concluded that the measurements given by both the OPCs differed from the reference concentrations.

The difference, however, is more in calculated PV concentrations compared to Firmware PV. 13 However, despite this calibration, the error values remain as high as ~40-50%, and there still exists a statistical disagreement between the firmware values and the reference concentrations. This correction is, therefore, deemed incompetent. However, despite the errors in the absolute concentrations, the OPC-N2s were found to have good precision and are capable of detecting changes in concentration with time. Similar findings were given by Sousan et al.13 while evaluating Alphasense OPC-N2 in laboratory conditions and Crilley et al.11 in ambient conditions.

Correction for OPC-N2

Counting Efficiency Correction Figure 4 shows the comparison of counting efficiencies for both the OPCs. For each bin, its counting efficiency is represented at its mean diameter in the plot. Before correction, the particles < 0.8 µm exhibited counting efficiency values < 1. This indicates that the particles < 0.8 µm were under-estimated. The value gradually increased and reached a maximum value of 1.88 at 1.195 µm. For 0.8 – 1.6 µm diameter range, the OPCs seem to overestimate the particle count. After that, the OPCs started to underestimate again. However, its performance improved with increase in particle diameter. This inconsistency in the counting efficiency is one of the key reasons for OPC-N2’s poor performance in its detection range. The explanation for this variation is the poor performance of its laser. This shape of the raw counting efficiency curve, increasing values at the beginning, reaching a maxima, and falling after that to reach a minima, was similar to what Sousan et al.13 observed for OPC-N2 in laboratory conditions.

However, after the proposed counting efficiency correction was applied, a significant improvement in its value was seen in case of both the OPCs. The counting efficiencies were close to 1 for all the bins, post correction. Therefore, the proposed correction was successful in improving the performance of the OPCs significantly in its detection range. The correction for the missing mass for sub-0.38µm particles in the OPCs is discussed next.

Correction for Sub-0.38 μm Particles

After obtaining the corrected volume concentrations (0.38-10 µm) for both the OPCs, they were used in the second GPR function to obtain the volume distributions below 0.38 µm. The training of this GPR function yielded an adjusted R2 of 0.72. This indicates that the features included to build this function could successfully explain ~72% of the variations in the data. The application of this correction on one sample volume distribution from OPC 1 is shown in Figure 5. This data was collected on 2nd of December at 18:00 hours. The sampling interval was 30 minutes. The obtained full range volume distribution is compared against the reference volume concentrations in the figure 5.

The curve represented in red line is the predicted curve below 0.38 µm, which seems to agree well with the reference concentration values. The green line represents the fitted curve after the volume concentrations were corrected. Therefore, cascading both the corrections, the size distribution, ranging from 0.1 to 10 µm, was obtained. It can be seen in this example that the distribution obtained after corrections in this OPC is consistent with the reference concentrations. Thereafter, the curve prediction was carried out for all the corrected volume concentrations in test dataset of OPC 1 and OPC 2. The cascaded curve of predicted distribution and the corrected distribution for both the OPCs are compared against the reference distributions in Figure 6.

After combining both the corrections, the OPCs’ volume distributions seem to closely match the reference distributions. Once the full range volume concentrations were obtained (0.1-10 µm), the size-integrated PV concentrations (PV2.5 and PV10) for both the OPCs were computed. The comparison of these concentrations against the reference values is presented in Figure 7. It is evident that, post both the corrections, the agreement between the OPCs’ PV2.5 and PV10 with reference concentrations has substantially improved. To quantify this improvement, the accuracy metrics, r, R2, slope, MAPE and p-value, were computed for calculated, firmware and corrected PV concentrations, against the reference values. The results are summarized in table 5.

Before correction, high MAPE and low R2 values were observed for both the OPCs’ calculated PV values. In addition, the r and the slope values didn’t meet the EPA’s criteria. Also, the p-values concluded a statistical difference between the OPCs’ calculated PV concentrations and the reference concentrations. The OPCs’ firmware calibrated PV concentrations, on the other hand, showed an improvement in OPCs’ performance compared to calculated PV values, however, the values still failed to meet the EPA’s recommended values for r and slope. Also, statistical disagreement among firmware PV and the reference PV is observed.

However, the corrected PV concentrations for both the OPCs showed a significant reduction in MAPE compared to both calculated and firmware PV concentrations. Also, high R2 values were obtained (> 0.97). In addition, the r and the slope values, for the corrected PV, met the EPA’s criteria. Finally, the p-values for all the t-tests between the corrected PV concentration and the reference concentrations are > 0.05. This indicates that after the corrections were applied, there were no statistical evidence of disparity between them.

To better understand this improvement, the correlation plots for the calculated, firmware and the corrected PVs (PV2.5 and PV10) against the reference PVs (PV2.5 and PV10) are shown in figure 8. Ideally, the scatter points on these plots should lie on the line y=x; however, under practical conditions, any point close to this line can be considered as an acceptable measurement. However, the calculated and firmware PV2.5 and PV10¬ for both the OPCs are far off the line y=x. The firmware PV values are heteroscedastic, and the calculated PV values consistently underestimate the PV concentrations. This indicates unsatisfactory measurements. On the other hand, the corrected PV values are homoscedastic and closer to the line y=x, thus bolstering the results given in table 5. Therefore, with all these findings, it can be concluded that both the proposed corrections were successful in addressing OPC-N2's measurement limitations and improved its performance substantially.

There are several advantages of using this correction methodology for low-cost optical sensors. Firstly, the correction functions used in this methodology is non-parametric, hence is expected to perform better than parametric regression models. For example, Holstius et al.26 used a linear regression model for a low-cost sensor and could achieve a maximum R2 value of 0.72. In another study, Magi et al.17 used a multiple linear regression model and achieved a R2 of 0.60. On the other hand, in this study using GPR models, the R2 achieved is in the range of 0.97-0.98. This signifies that the developed correction model can explain up to 97-98% of variation in the data. The explanation for better results here is that the correction methodology is size-distribution based, and then conversion into size-integrated concentrations. Contrarily, models develop ed by Magi et al.17 and Holstius et al.26 corrected directly for size-integrated concentrations, thereby losing fundamental information on particle composition.

Another major advantage of using this correction methodology is that it enables the user to obtain a wide range size-distribution (0.1-10 µm) using a low-cost sensor. None of the correction methodology available in the literature serves this purpose. It is extremely important to know the complete size distribution before calculating the size-integrated concentrations. This is because any proportion of the size distribution curve that is missing or neglected in the analysis can cause errors in the final output of the sensor. For example, OPC-N2’s manual states, “The OPC-N2 calculations of particle mass assume a negligible contribution from particles below approximately 0.38 μm.”31 However, ignoring their contribution can lead to a severe underestimation in the PM concentrations. To prove this hypothesis, the contribution of sub-0.38 µm particles to the PV for the whole data set, measured using the reference instruments, was calculated. Figure 9 presents the box-plot of this contribution. The average percentage contribution of sub-0.38 µm particles was found to be ~40% for PV2.5 and ~27% for PV10. Under no circumstance can this contribution be assumed as negligible. Additionally, the majority of the low-cost optical-based PM sensors tend to neglect this contribution because of their inability to detect particles in this range. Therefore, it becomes vital for all such sensors to perform equivalent corrections, as proposed, before forecasting size-integrated concentrations.

Despite these advantages of the proposed correction methodology, there are certain limitations that needs to be addressed here for future research in this area. Firstly, all the results in this study are expressed in terms of size-integrated Particle Volume concentration. Although Particle Volume is analogous to Particle Mass, the latter is popularly used. Therefore, to obtain the PM concentration, it would require incorporating the size-resolved aerosol densities in these correction models.

Another shortcoming of this study is that the proposed framework is limited for indoor aerosols only. For ambient aerosols, temperature and relative humidity should be included in the models. Finally, the proposed correction model is data-driven; hence, with change in conditions, it would be necessary to re-train the models. Researchers can explore the use of physical correction models in the future.

Remember, at the center of any academic work, lies clarity and evidence. Should you need further assistance, do look up to our Business Communications Assignment Help