Response surface methodology-based model for prediction of NO and NO_{2} concentrations in nonthermal plasma-treated diesel exhaust
- 88 Downloads
Abstract
Nonthermal plasma (NTP) technique for diesel exhaust emission control has been in the interest of the researchers from the last two decades, and it almost became a laboratory-proven technique for its efficiency over conventional techniques. A prior prediction of effectiveness of the process may lead to overcome the constraints in bringing this technique to real-time applications of diesel exhaust pollution control. In this present study, an attempt is made to find out the most dominating parameters and to predict the sum of NO and NO_{2} concentrations in NTP-treated diesel exhaust with respect to variations in operating parameter values using response surface methodology (RSM). Experiments have been conducted following 3^{N} full factorial design and collected the data by varying voltage, flow rate, temperature, discharge gap and initial (NO + NO_{2}) concentrations. The regression coefficients of the RSM-based mathematical model have been obtained by training it using these experimental data. The root-mean-square error (RMSE) is 4.7 ppm during the training. When this model is tested for a data other than that given during training (test data), the RMSE is 5.6 ppm. Further, the results are also compared with the model derived using the only available method in the literature, i.e., dimensional analysis, and found to be performing better.
Keywords
Air quality Diesel exhaust Nonthermal plasma NO_{X} removal Prediction of NO_{X} Response surface methodology1 Introduction
Diesel exhaust emission control has become a challenge for the researchers, as the emission standards are becoming more stringent throughout the world. On the other hand, demand for diesel engines is increasing continuously. Thus, there is a need to look for a technology to substitute the existing conventional catalytic-based filtering systems.
Even though NTP treatment for diesel exhaust pollution control has been proven to be an efficient technique, the studies are still at the laboratory level. To bring it into real-time applications, prior prediction of pollutant concentrations with the treatment would be helpful. If the NO_{X} concentrations are accurately predicted through modeling, it facilitates to estimate the pollutant removal efficiencies for those parametric variations, which cannot be performed in the laboratory due to experimental limitations [1]. Experimental studies were conducted by the researchers throughout the world to know the effects of various operating parameters such as applied voltage [2, 3], flow rate [4], temperature [5], residence time [6], reactor configurations [7] and electrode configurations [8]. While there is sufficient literature available regarding the chemical kinetics of the treatment, there is no much work done to quantify the effects of different parameters on the NO_{X} removal from the exhaust. Further, this modeling also helps the researchers to plan for the real-time applications through providing the knowledge of the relation between different parameter values and NO_{X} removal.
The multi-criteria decision-making method was used to evaluate diesel exhaust emission characteristics by Hoseinpour et al. [9], while gasoline fumigation was added to reduce pollutant emissions under different operating conditions. A plant-specific multi-year and multi-parameter coal power stack emission model has been proposed by Walvekar and Gurjar [10] using the emission factor-based approach. However, very few studies have been carried out to predict the NO_{X} (NO + NO_{2}) concentrations in the diesel exhaust after the NTP treatment based on the parameter values.
In most of the studies [1, 11, 12], dimensional analysis has been used for this purpose by relating the NO_{X} concentration with two or three other operating parameters of the NTP process. The disadvantage of the models derived using this method is that the root-mean-square error (RMSE) between experimental and predicted values of NO_{X} concentrations would be more for an experimental data other than that by which the model is trained. The reason for this is the less experimental data requirement for the training of the model. It can be observed that the RMSE for the test data is 9.58 ppm in a study by the authors of this present study (Allamsetty and Mohapatro [13, 14]).
Deriving a mathematical model and training it with more number of experimental data with a wide range of parameter values can be helpful in accurately predicting the pollutant concentrations for the test data also. Thus, in this present study, an effort has been made to predict the sum of NO and NO_{2} concentrations in diesel exhaust with NTP treatment using RSM.
RSM has been used to model various types of parameters such as surface roughness and temperature of turning operation [15, 16], kerf width of laser machining [17], bending strength of aluminum alloys [18]. It has also been used in [19, 20, 21, 22] to predict the pollutant concentrations in the emissions from a diesel engine based on various control parameters such as engine speed, compression ratio, brake power and injection parameters.
In this present study, the sum of NO and NO_{2} concentrations (NO + NO_{2}) has been predicted using RSM-based predictive model, with respect to the changes in parameters voltage (V), flow rate (Fr), temperature (T), discharge gap (Dg), initial sum of NO and NO_{2} concentrations (NO + NO_{2})_{i}. The effect of each parameter on the response has been analyzed numerically to know the dominant parameters. Values of the regression coefficients of the model have been derived, and analysis of variance has been performed on the model. Then, the model has been used to predict the values of (NO + NO_{2}) and compared them with their corresponding experimental values for the test data. The model has also been compared with the model derived using dimensional analysis with respect to its performance during the testing.
2 Experimental details
2.1 Experimental setup
Exhaust composition before treatment
(NO + NO_{2})_{i} (ppm) | NO (ppm) | NO_{2} (ppm) | CO (ppm) | O_{2} (%) | CO_{2} (%) |
---|---|---|---|---|---|
200 | 294 | 6 | 240 | 14.5 | 6.3 |
250 | 236 | 14 | 344 | 13.9 | 6.8 |
300 | 280 | 20 | 448 | 13.2 | 7.5 |
The high voltage applied to the plasma reactor during the NTP treatment has been generated and varied using a high-voltage AC test set (0–30 kV, 50 Hz, Make: Rectifiers & Electronics). This high voltage has been measured using a voltage divider (2000:1 ± 5%, Make: IWATSU, Model: HV-P60A, DC to 50 MHz, within − 3 dB) and a digital storage oscilloscope (Make: RIGOL, Model: DS 1074: 70 MHz).
The inner and outer diameters of the reactor are 15 mm and 17 mm, respectively. A layer of aluminum foil is wrapped to form the ground electrode over the reactor, between inlet and outlets, for a length of 280 mm. This can be described as the effective discharge length. The reactor is made up of borosilicate glass, which acts as the dielectric barrier when the high voltage is applied. The parameter Dg has been varied during the experiments by changing the diameter of the high-voltage electrode. The inner diameter of the reactor is maintained constant at 15 mm. Thus, when the electrode diameters are changed among 3 mm, 4 mm and 5 mm, they formed discharge gaps of 6 mm, 5.5 mm and 5 mm, respectively.
2.2 Design of experiments
The design of experiments (DoE) is a crucial aspect of RSM, which was formerly developed for the model fitting of physical experiments. Later, these strategies were also implemented for numerical experiments with an objective of selecting the input data points where the response needs to be experimentally found out.
Operating parameters and their levels
Operating parameters | Levels | ||
---|---|---|---|
− 1 | 0 | + 1 | |
Voltage (kV) | 16 | 21 | 26 |
Flow rate (lpm) | 4 | 10 | 16 |
Temperature (°C) | 25 | 50 | 75 |
Discharge gap (mm) | 5 | 5.5 | 6 |
Initial sum of NO and NO_{2} concentration (ppm) | 200 | 250 | 300 |
The NTP treatment primarily causes NO to NO_{2} conversion reactions leading to a decrement in NO concentration and an increment in NO_{2} concentration. If the applied high voltage keeps on increasing, the electric field gets intensified leading to the decrement in NO_{2} also. The possible chemical reactions those can take place in the reactor were given in a study by Saavedra et al. [24] with their rate constants. Major reaction pathways have been mentioned in a previous study of the authors [25]. The operating parameters have been varied in such a way to cover all the possible combinations with all the considered levels of each parameter, and corresponding (NO + NO_{2}) concentrations are noted down.
2.3 Experimental results
3 Model derivation
Estimated coefficients of quadratic RSM model
Term | Coefficient | SE coefficient | t-Stat |
---|---|---|---|
C | 382.8295 | 87.7362 | 4.3634 |
V | − 10.6173 | 1.5534 | − 6.8349 |
Fr | − 9.6690 | 1.0411 | − 9.2873 |
T | 0.7192 | 0.2554 | 2.8158 |
Dg | − 17.7552 | 28.8792 | − 0.6148 |
(NO + NO_{2}) | − 0.6918 | 0.1684 | − 4.1087 |
VFr | 0.5571 | 0.0151 | 36.9309 |
VT | − 0.0280 | 0.0036 | − 7.7217 |
VDg | 0.9144 | 0.1810 | 5.0517 |
V(NO + NO_{2}) | 0.0066 | 0.0018 | 3.6583 |
FrT | 0.0094 | 0.0030 | 3.1018 |
FrDg | − 0.0981 | 0.1508 | − 0.6506 |
Fr(NO + NO_{2}) | 0.0038 | 0.0015 | 2.5473 |
TDg | 0.0199 | 0.0362 | 0.5504 |
T(NO + NO_{2}) | − 0.0028 | 0.0004 | − 7.6256 |
Dg(NO + NO_{2}) | 0.0159 | 0.0181 | 0.8757 |
V ^{2} | − 0.1813 | 0.0256 | − 7.0805 |
Fr^{2} | 0.0516 | 0.0178 | 2.9022 |
T ^{2} | 0.0002 | 0.0010 | 0.1649 |
Dg^{2} | 0.6444 | 2.5600 | 0.2517 |
(NO + NO_{2})^{2} | 0.0030 | 0.0003 | 11.8548 |
The SE coefficient is an estimate of the standard deviation of the sampling distribution of the corresponding parameter. In other words, it is the ratio of standard deviation and the square root of the sample size. A lower value of SE coefficient indicates a more precise estimation. The t-stat is the ratio of the coefficient to its standard error. It is used to test the null hypothesis that the corresponding coefficient is zero against the alternative that it has some value other than zero, given the other predictors in the model. From the t-stat values obtained, it can be said that the coefficient values are well estimated. The values of both R^{2} and adjusted R^{2} of the model have been found to be 0.993, which indicate that this model is closely fitted with the experimental data. The RMSE is found to be 4.7 ppm. Further, the F value for the model has been found to be 1660 with respect to the constant model, which indicates a significant regression relationship between the response parameter and the operating parameters. The p value is 1.76e^{−229}, which indicates a strong significance of the model. According to these results, it can be said that the model is well derived and suitable to be used for prediction of the response parameter, i.e., (NO + NO_{2}).
4 Results and discussion
4.1 Surface plots
The trend followed by the response parameter, i.e., (NO + NO_{2}), with respect to the increase or decrease in the operating parameters can be observed in these plots. The apparent decrement in the (NO + NO_{2}) with the variation in V from 16 to 26 kV can be noticed in every subfigure of Fig. 3. It can be seen from Fig. 3a that the effect of V is less when the Fr is 16 lpm. The effect of the second parameter on the response can be observed in the remaining plots also and can be compared with each other. For example, from Fig. 3a, b, it can be noticed that the effect of Dg, when it is varied from its lower level to upper level, is lesser compared to that of Fr. Similarly, a decrement in (NO + NO_{2}) with an increase in T and with a decrease in Dg and (NO + NO_{2})_{i} can be observed from Fig. 3b, c and d, respectively.
4.2 Analysis of variance
Analysis of variance for the quadratic RSM model
Source | SS | DF | MS | F value | P value |
---|---|---|---|---|---|
V | 220,101.4 | 1.0 | 220,101.4 | 9951.1 | 2.2E−186 |
Fr | 91,027.0 | 1.0 | 91,027.0 | 4115.5 | 2.8E−145 |
T | 11,567.2 | 1.0 | 11,567.2 | 523.0 | 2.8E−60 |
Dg | 6343.1 | 1.0 | 6343.1 | 286.8 | 7.4E−42 |
(NO + NO_{2}) | 367,234.0 | 1.0 | 367,234.0 | 16,603.3 | 1.2E−210 |
VFr | 30,166.9 | 1.0 | 30,166.9 | 1363.9 | 9.4E−97 |
VT | 1318.8 | 1.0 | 1318.8 | 59.6 | 3.9E−13 |
VDg | 564.4 | 1.0 | 564.4 | 25.5 | 9.1E−07 |
V(NO + NO_{2}) | 296.0 | 1.0 | 296.0 | 13.4 | 3.2E−04 |
FrT | 212.8 | 1.0 | 212.8 | 9.6 | 2.2E−03 |
FrDg | 9.4 | 1.0 | 9.4 | 0.4 | 5.2E−01 |
Fr(NO + NO_{2}) | 143.5 | 1.0 | 143.5 | 6.5 | 1.2E−02 |
TDg | 6.7 | 1.0 | 6.7 | 0.3 | 5.8E−01 |
T(NO + NO_{2}) | 1286.2 | 1.0 | 1286.2 | 58.1 | 7.0E−13 |
Dg(NO + NO_{2}) | 17.0 | 1.0 | 17.0 | 0.8 | 3.8E−01 |
V ^{2} | 1108.9 | 1.0 | 1108.9 | 50.1 | 1.9E−11 |
Fr^{2} | 186.3 | 1.0 | 186.3 | 8.4 | 4.1E−03 |
T ^{2} | 0.6 | 1.0 | 0.6 | 0.0 | 8.7E−01 |
Dg^{2} | 1.4 | 1.0 | 1.4 | 0.1 | 8.0E−01 |
(NO + NO_{2})^{2} | 3108.4 | 1.0 | 3108.4 | 140.5 | 1.9E−25 |
e | 4910.2 | 222.0 | 22.1 | 1.0 | 5.0E−01 |
The calculated F values mentioned in the table for most of the terms of the model are greater in magnitude than the critical value of F, i.e., 3.88. If the F value of any term is lesser than the critical value of F of the model, then the P value increases and indicates the insignificance of that particular term. This has happened for few of the interaction and square terms of the model. The P values are comparatively higher for the interaction terms involved with the parameter Dg, i.e., FrDg, TDg and (NO + NO_{2})_{i}Dg, and square terms, i.e., T^{2} and Dg^{2}. However, the P values of linear terms along with the term associated with Dg are much lesser than 0.05, indicating that those terms are statistically significant at the 95% confidence level. In other words, it can be said that the changes in response values are associated with changes in parameter values and their inclusion in the model is meaningful.
4.3 Model validation
The second part of the validation process is testing the quadratic RSM model for an entirely new set of data which has not been used during training or while finding out the regression coefficients. Experiments have been conducted to obtain the test data of twelve sets with a random variation in the operating parameter values.
The operating parameter details of this test data set of twelve experiments are given in Table 4 along with the predicted values of NO + NO_{2} concentrations. In this table, experimental values of NO + NO_{2} concentrations are mentioned as (NO + NO_{2})_{e} and predicted values of NO + NO_{2} obtained using RSM model are mentioned as (NO + NO_{2})_{p_RSM}.
Test data with predicted (NO + NO_{2}) using quadratic RSM and dimensional analysis-based models and their errors
Experimental order | V (kV) | Fr (lpm) | T (°C) | Dg (mm) | (NO + NO_{2})_{i} (ppm) | (NO + NO_{2})_{e} (ppm) | (NO + NO_{2})_{p_RSM} (ppm) | E_{RSM} (ppm) | (NO + NO_{2})_{p_Dim} (ppm) | E_{Dim} (ppm) |
---|---|---|---|---|---|---|---|---|---|---|
1 | 25 | 8 | 25 | 5 | 250 | 146.7 | 152.5 | 5.8 | 154.8 | 8.1 |
2 | 18 | 4 | 50 | 5.5 | 300 | 260.7 | 254.7 | − 6 | 224.1 | − 36.6 |
3 | 24 | 12 | 75 | 6.5 | 200 | 141.4 | 148.2 | 6.8 | 107.1 | − 34.3 |
4 | 16 | 12 | 25 | 5 | 250 | 232.6 | 229.0 | − 3.6 | 264.5 | 31.9 |
5 | 20 | 8 | 75 | 6 | 300 | 251.1 | 243.3 | − 7.8 | 243.7 | − 7.4 |
6 | 21 | 6 | 50 | 5 | 200 | 138.7 | 132.9 | − 5.8 | 92.4 | − 46.3 |
7 | 17 | 10 | 100 | 5.5 | 250 | 210 | 208.7 | − 1.3 | 189.5 | − 20.5 |
8 | 23 | 14 | 50 | 6 | 300 | 262.1 | 262.8 | 0.7 | 275.9 | 13.8 |
9 | 19 | 4 | 25 | 6.5 | 200 | 177.9 | 169.4 | − 8.5 | 116.9 | − 61.0 |
10 | 23 | 16 | 50 | 6 | 250 | 209.7 | 216.8 | 7.1 | 198.7 | − 11.0 |
11 | 15 | 4 | 25 | 5 | 300 | 286.4 | 284.4 | − 2 | 288.8 | 4.4 |
12 | 20 | 18 | 75 | 6 | 200 | 187.2 | 192.5 | 5.3 | 137.8 | − 49.4 |
4.4 Comparison with dimensional analysis-based model
The predicted values of NO + NO_{2} concentrations obtained using the dimensional analysis-based model are mentioned as (NO + NO_{2})_{p_Dim} in Table 5. The error in these predicted values is mentioned as E_{Dim}. It can be observed from this column of Table 5 that for the experiments with order numbers 1, 5, 8, 10 and 12, the error is lesser compared to that of the remaining experiments. From this, it can be understood that the error is more when there is simultaneous variation in two or more operating parameters with respect to the data for which the model is trained.
5 Conclusion
A prior knowledge about the outcome of any experimental process would always be helpful in bringing the theory toward practical applications. In this present study, an attempt is made to predict the sum of NO and NO_{2} concentrations in diesel exhaust during the NTP treatment, with respect to changes in operating parameters, using RSM. The 3^{5} full factorial design has been adopted for conducting the experiments with three-level variations in the five operating parameters: Those are V, Fr, T, Dg and (NO + NO_{2})_{i}. From the main effect plots, it is noticed that the change in (NO + NO_{2})_{i} and V affects the (NO + NO_{2}) more than the other parameters.
Regression coefficients of the quadratic RSM model have been obtained by feeding it with the experimental data. The results of the chosen model have been analyzed using the surface plots, ANOVA and model validation. From the F values and P values obtained with ANOVA, the model can be described as significant at the 95% confidence level.
The predicted values are observed to be in good agreement with the corresponding experimental values. The RMSEs are found to be 4.7 ppm and 5.6 ppm for training data and test data, respectively. From all these results, this quadratic RSM model can be described as well suited for the prediction of (NO + NO_{2}) concentrations in diesel exhaust during NTP treatment.
Notes
Compliance with ethical standards
Conflict of interest
The authors declare that they have no conflict of interest.
References
- 1.Jasmin K, Bhattacharyya A, Rajanikanth BS (2015) Prediction of NO_{X} removal efficiency in plasma treated exhaust: a dimensional analysis approach. In: 3rd international symposium on new plasma and electrical discharge application and on dielectric materials (ISNPEDADM), new electrical technologies for environment, Reunion island (Indian Ocean)Google Scholar
- 2.Song C-L, Bin F, Tao Z-M et al (2009) Simultaneous removals of NO_{x}, HC and PM from diesel exhaust emissions by dielectric barrier discharges. J Hazard Mater 166:523–530. https://doi.org/10.1016/j.jhazmat.2008.11.068 CrossRefGoogle Scholar
- 3.Morgan N, Ibrahim D, Samirm A (2017) NO_{X} emission reduction by non thermal plasma technique. J Energy Environ Chem Eng 2:25–31. https://doi.org/10.11648/j.jeece.20170202.12 CrossRefGoogle Scholar
- 4.Namihira T, Tsukamoto S, Wang D et al (2001) Influence of gas flow rate and reactor length on NO removal using pulsed power. IEEE Trans Plasma Sci 29:592–598. https://doi.org/10.1109/27.940952 CrossRefGoogle Scholar
- 5.Rajanikanth BS, Srinivasan AD, Subhankar das (2005) Enhanced performance of discharge plasma in raw engine exhaust treatment-operation under different temperatures and loads. Plasma Sci Technol 7:2943–2946. https://doi.org/10.1088/1009-0630/7/4/015 CrossRefGoogle Scholar
- 6.Talebizadeh P, Rahimzadeh H, Babaie M et al (2015) Evaluation of residence time on nitrogen oxides removal in non-thermal plasma reactor. PLoS ONE 10:1–16. https://doi.org/10.1371/journal.pone.0140897 CrossRefGoogle Scholar
- 7.Osawa N, Suetomi T, Hafuka Y et al (2012) Investigation on reactor configuration of non thermal plasma catalytic hybrid method for NO_{X} removal of diesel engine exhaust. Int J Plasma Environ Sci Technol 6:119–124Google Scholar
- 8.Mohapatro S, Allamsetty S, Madhukar A, Sharma NK (2017) Nanosecond pulse discharge based nitrogen oxides treatment using different electrode configurations. High Volt 2:60–68. https://doi.org/10.1049/hve.2017.0011 CrossRefGoogle Scholar
- 9.Hoseinpour M, Sadrnia H, Ghobadian B, Tabasizadeh M (2018) Exhaust emission characteristics of a diesel engine on gasoline fumigation: an experimental investigation and evaluation using the MCDM method. Int J Environ Sci Technol. https://doi.org/10.1007/s13762-018-1667-1 CrossRefGoogle Scholar
- 10.Walvekar PP, Gurjar BR (2013) Formulation, application and evaluation of a stack emission model for coal-based power stations. Int J Environ Sci Technol 10:1235–1244. https://doi.org/10.1007/s13762-012-0131-x CrossRefGoogle Scholar
- 11.Roslan NA, Buntat Z, Sidik MAB (2013) Application of dimensional analysis for prediction of NO_{X} removal. In: IEEE 7th international power engineering and optimization conference, Langkawi, Malaysia, pp 218–222Google Scholar
- 12.Mukherjee DS, Rajanikanth BS (2016) DBD plasma based ozone injection in a biodiesel exhaust and estimation of NO_{X} reduction by dimensional analysis approach. IEEE Trans Dielectr Electr Insul 23:3267–3274. https://doi.org/10.1109/TDEI.2016.005919 CrossRefGoogle Scholar
- 13.Allamsetty S, Mohapatro S (2018) Prediction of NO and NO_{2} concentrations in ozone injected diesel exhaust after NTP treatment using dimensional analysis. In: 10th international conference on applied energy (ICAE), Hong KongGoogle Scholar
- 14.Allamsetty S, Mohapatro S (2018) Prediction of NO_{X} concentration in nonthermal plasma-treated diesel exhaust using dimensional analysis. IEEE Trans Plasma Sci 46:2034–2041. https://doi.org/10.1109/TPS.2018.2827400 CrossRefGoogle Scholar
- 15.Kahraman F (2009) The use of response surface methodology for prediction and analysis of surface roughness of AISI 4140 steel. Mater Technol 43:267–270Google Scholar
- 16.Khidhir BA, Oqaiel WA, Kareem P (2015) Prediction models by response surface methodology for turning operation. Am J Model Optim 3:1–6. https://doi.org/10.12691/ajmo-3-1-1 CrossRefGoogle Scholar
- 17.Sivaraos Milkey KR, Samsudin AR et al (2014) Comparison between Taguchi method and response surface methodology (RSM) in modelling CO_{2} laser machining. Jordan J Mech Ind Eng 8:35–42Google Scholar
- 18.Jaya HT, Tunggal D (2011) Prediction of the bending strength using response surface methodology. 4^{th} Int. Conf. Modeling Simul. Appl. Optim., Malaysia, 2–5Google Scholar
- 19.Sivaramakrishnan K, Ravikumar P (2014) Optimization of operational parameters on performance and emissions of a diesel engine using biodiesel. Int J Environ Sci Technol 11:949–958. https://doi.org/10.1007/s13762-013-0273-5 CrossRefGoogle Scholar
- 20.Wahono B, Ogai H (2012) Construction of response surface model for diesel engine using stepwise method. In: 2012 joint 6th international conference on soft computing and intelligent systems (SCIS) and 13th international symposium on advanced intelligent systems (ISIS), Kobe, pp 989–994Google Scholar
- 21.Yilmaz N, Ileri E, Atmanli A et al (2016) Predicting the engine performance and exhaust emissions of a diesel engine fueled with hazelnut oil methyl ester: the performance comparison of response surface methodology and LSSVM. J Energy Resour Technol 138:052206. https://doi.org/10.1115/1.4032941 CrossRefGoogle Scholar
- 22.Anderson A, Devarajan Y, Nagappan B (2018) Effect of injection parameters on the reduction of NO_{X} emission in neat bio-diesel fuelled diesel engine. Energy Sources Part A Recover Util Environ Eff 40:186–192. https://doi.org/10.1080/15567036.2017.1407844 CrossRefGoogle Scholar
- 23.Byrne D, Taguchi S (1987) Taguchi approach to parameter design. Qual Prog 20:19–26Google Scholar
- 24.Saavedra HM, Pacheco MP, Pacheco-Sotelo JO et al (2007) Modeling and experimental study on nitric oxide treatment using dielectric barrier discharge. IEEE Trans Plasma Sci 35:1533–1540. https://doi.org/10.1109/TPS.2007.905951 CrossRefGoogle Scholar
- 25.Mohapatro S, Allamsetty S (2017) NO X abatement from filtered diesel engine exhaust using battery-powered high-voltage pulse power supply. High Volt 2:69–77. https://doi.org/10.1049/hve.2016.0084 CrossRefGoogle Scholar