An experimental-based artificial neural network performance study of common rail direct injection engine run on plastic pyrolysis oil

ABSTRACT

Artificial neural network model was constructed to analyse and evaluate the engine performance. The experiments were conducted on a diesel engine with the blend of plastic pyrolysis oil with diesel and ethanol. Three input layer with two hidden layers and five output layers were used in artificial neural network modelling. The learning algorithm called feed-forward back-propagation was applied for the hidden layer. To train the neural network, 70% of the complete data from the experimentation was selected and 30% in predicting from the neural network. The model developed for prediction has excellent agreement as observed from the correlation coefficient (R) within the range of 0.964–0.9816. Statistical analysis shows that the ANN predicted and experimental results are in close agreement with each other. Overall, it could be concluded that it is a mean to predict the virtual sensing in studying the real time with established artificial neural network architecture. In addition, common rail direct injection engine operation could give complete freedom from diesel and thereby provides energy security and sustainable of a nation.

KEYWORDS:

• Common rail direct injection (CRDI)
• plastic pyrolysis oil (PPO)
• ethanol
• engine performance
• exhaust gas recirculation (EGR)
• artificial neural network (ANN)

1. Introduction

For the past few decades, there is significant growth in the use of vehicles worldwide. At the same time, the pollution given out by these vehicles has raised the global warming issue. With the depletion of conventional fuel and an increase in the pollution level has led many researchers to work on alternative fuels that yield lower emissions without compromising in the power. The quality of air is deteriorating due to the emission especially nitrogen oxide (NOx) from the diesel engine powered with diesel, which is a key factor to be considered for research work as it affects human health. Some of the countries like Brazil and China have implanted heavy taxes on these vehicles keeping in mind the elimination of such vehicles. Even the Indian government in some of the cities has banned the vehicle that runs on diesel with 200 CC (National Green Tribunal). In order to use the diesel engine, alternative fuels are the promising sources of energy but fuels must yield lower emissions. The number of experimental work on diesel engine has been carried out under the influence of mixture formation and injection strategies (Theinnoi, Suksomponga, and Temwutthikuna 2017; Banapurmath et al. 2018). The main challenge for emission control can be done by earlier and after treatment control devices which are involved in converting NOx/particulate matter (PM) to comparative benign compounds [Johnson TV 2006]. Various methods to reduce the NOx emission from a diesel engine are EGR, variation in injection strategies, water injection and reducing ignition delay (ID). Among these EGR and injection strategies were proved to be effective methods currently being applied to reduce the NOx (Robert et al. 2002; S.V. Khandal et al. 2017a). In EGR, the fraction of exhaust gases is recirculated into the intake manifold which mixes up with air being inducted into the combustion chamber (Fang et al. 1999). It has also observed that with different injection strategies, the NOx was reduced when the engine operated with pyrolysis oil (Sudershan et al. 2017). Research studies (McGeehan 2005; Johnson 2008; Zhao 2010; Badami, Nuccio, and Trucco 1999; Suh et al. 2011) has establishes a clear evolution of pollution norms that challenged the design and development of diesel engine with modern technology. This paradigm shifts in the technologies of a diesel engine which satisfy the specified emission norms along with best fuel economy. Among the numerous latest technologies, common rail direct injection (CRDI) finds a renewed interest (Balusamy 2010; Shimazaki 2003). CRDI technology has proven to be low emission with reduced fuel operating consumption due to homogeneous mixture formation (Pickett 2004; Minato, Tanaka, and Nishimura 2005; Desantes et al. 2004; Pierpont et al. 1995; Payri 2006) than the conventional diesel engine but NOx emission was higher (Hountalas, Mavropoulos, and Binder 2008). However, when pyrolysis oil (PPO) blended with diesel and ethanol in the percentage improvement in the emission were observed without much compromising in its performance (Sudershan et al. 2019). EGR has been employed widely because of simple operation with higher efficiency along with cost-effectiveness (Maiboom 2008). One of the main drawbacks of experimental work is that it consumes more time in the analysis of the engine performance and it is expensive. To overcome such problem, modelling tools were developed as an alternative. One such technique is modelling with ANN. This technique was used to estimate the required output parameters by utilising the experimental data available. Hence, ANN model of physical phenomenon in complex systems without using mathematical relations could be advantageous in predicting the responses. Researchers used ANN in predications of engine performance and different thermal systems (Kalogirou 2000; Pacheco-Vega 2001; Betchtler 2001; Prieto 2001; Chouai, Laugeier, and Richon 2002; Sozen, Arcaklioglu, and Ozalp 2003; Ertunc and Hosoz 2006; Hosoz and Ertunc 2006). The application of the ANN model was applied to a diesel engine in predicting its performance and emission character. The performance and emissions of a diesel engine under the effect of injection pressure using ANN were reported (Canakci, Erdil, and Arcaklioglu 2006; Arcaklioglu 2004; Arcaklioglu and Celikten 2005). IT, IP and EGR are the important parameters in the performance analysis of the CRDI engine. The experimental work on CRDI engine fuelled with rapeseed oil (RSO) and its blends showed lower NOx emissions, where the fuel IT varied from 3° BTDC to 9° BTDC. The diesel equivalent levels of soot emission and 22% reduction in NOx emissions were reported at 30% RSO and IT of 3° BTDC and IP of 1200 bar (Labecki et al. 2012). When EGR of 20% was employed with retarded single injection of the fuel at 9° BTDC, NOx reduced further at the expense of higher smoke, HC and CO emissions and lower PP and HRR were also reported as compared to neat diesel operation. However, when IP increased beyond 800 bar, NOx emission was higher than the CI operation, but soot emission lowered (Labecki et al. 2012; Song et al. 2012). The SOME indicated faster ignition, lower pre-mixed spike and lower PP when the start of energising was between 2° and 6° BTDC. Also, the volume of particulate increased significantly with speed (Yoon, Hyung, and Woo 2008). The UOME gave better performance at IT of 10° BTDC and IP of 900 bar, when a CI engine is operated at a rated speed of 1500 rpm at 80% and 100% loads (Basavarajappa et al. 2015). ANN model was developed using experimental data to predict performance and emission characteristics. The developed ANN model was capable of mapping the PM–NOx–BSFC trade-off potential of the CRDI engine operation (Roy S et al. 2014). Another work reported that the developed ANN model has commendable accuracy in predicting the performance characteristics as the error was in the range of 0.045–1.66% and correlation coefficients were in the range of 0.99833–0.99999 (Amit Ray et al. 2015, Chakraborty, Roy, and Banerjee 2016). The ANN architecture wherein the MOPSO algorithm was employed to improve the emission-performance trade-off characteristics. The results indicated a 10.2%, 30.6%, 25.4% and 9.4% improvement in the performance-emission trade-off footprint as compared to experimental results at 20%, 40%, 60% and 80% full load, respectively (Banerjee et al. 2017).

1.1. Motivation for the present study and objectives

The present study investigates the use of alternative fuel that replaces diesel which is depleting due to its high consumption rate. In this regard, various combinations of blends with different engine operating parameters were used to check the feasibility of the alternative fuel. Further to reduce the number of experiments to be carried out, ANN application was utilised. This will help to carry out minimum experimental work.

After a detailed literature review, it was observed that scarce literature is available in predicting the engine performance of common rail direct injection (CRDI) engine using ANN model with error analysis. In this context, experimental work was carried out to know the combined effect of EGR and injection strategies on the performance of diesel engine modified to operate in CRDI mode using ANN prediction models. Owing to facts like complex behaviour besides nonlinearity of CI engine operations, ANN is an appropriate modelling technique to study CRDI engine performance. Therefore, an effort is made to study the CRDI engine performance using ANN with statistical errors. In addition, it will substitute a virtual platform for diagnosing the system in real time. The results obtained by ANN for the engine performance studies were impressive, i.e. the ability to predict BTE, HC, NOx, PP and HRR was commendable. Statistical error analysis has been also carried out to compare the ANN predicted results with experimental results.

2. Conversion method of waste to oil

Figure 1. Experimental set up for plastic pyrolysis oil

Figure 1 represents the pyrolysis oil obtained from waste plastic. The waste solid plastics were utilised for the production of plastic pyrolysis oil (PPO). Wastes collected are sorted, cleaned, cut into pieces and fed into fixed bed reactor. Inert gas was inducted from the chamber to make it free from oxygen. The reactor is subjected to heating to about 300–400^oC for 3–4 hours. On reaching 370^oC, gases are produced which are condensed in the condenser. The condensate is a pyrolysis oil. The properties of pyrolysis oil are shown in Table 1. The blend used in the current work was 70:30 (PPO 70% and Diesel 30% on volume basis). The fuel properties of the blends are shown in Table 2.

Table 1. Fuel properties

Property	PPO	Diesel
Density (kg/m^3)	835.5	840
Kinematic viscosity (cSt)	2.52	2
Flash point (^oC)	40	50
Fire point (^oC)	43	56
Calorific value (kJ/kg)	40248	44500
Cloud point(^oC)	-	2
Pour point (^oC)	-	−16
Cetane number	46	55

Table 2. Properties of blended fuels

Blends	PPO: Diesel 70:30
Properties
Density (kg/m^3)	830
Kinematic viscosity (cSt)	2.4
Flash point (°C)	45
Fire point (°C)	52
Calorificvalue (kJ/kg)	40,671
Cetane number	47

3. Experimental setup

Rated capacity of the engine is 5.2 kW with 1500 rpm and specification of the same are presented in Table 3. Figure 2 represents the schematic diagram of the CRDI diesel engine, which is a single-cylinder, 4-stroke and water-cooled engine.

Figure 2. Schematic diagram of the CRDI experimental test rig

Table 3. Specifications of CI engine

Make and model	Kirloskar, TV1
Engine type	Single cylinder four stroke direct injection diesel engine
Orientation	Vertical
Ignition system	Compression Ignition
Bore x Stroke	87.5 mm x 110 mm
Displacement volume	660 cc
Compression ratio	17.5: 1
Arrangement of valves	Overhead
Combustion chamber	Open Chamber (Direct injection)
Rated power	5.2 KW @1500 RPM
Cooling Medium	Water cooled
Air measuring manometer
Made	MX 201
Type	U- Type
Range	100 – 0 – 100 mm
Eddy current dynamometer
Model	AG – 10
Type	Eddy current
Maximum	7.5 (kW at 1500–3000 RPM)
Flow	Water must flow through Dynamometer during the use
Dynamometer arm length	0.180 m
Uel measuring unit range	0–50 ml

3.1. Uncertainty analysis

The uncertainties for the present investigation are given in Table 4. Four readings were taken to overcome the measurement errors. The result of averaged values presented in the analysis. Table 5 provides the specification of exhaust gas analyser and Table 6 provides the specification of smoke meter used in the current work.

Table 4. Uncertainties for the present investigation

Measured variable	Accuracy (±)
Load (N)	0.1
Engine speed (rpm)	1
Temperature, (^oC)	1
Fuel consumption (g)	0.1
Measured variable	Uncertainty (%)
HC	±1.2
CO	±2.5
NOx	±2.3
Smoke	±2.0
Calculated parameters	Uncertainty (%)
BTE (%)	±1.2
HRR (J/^oCA)	±1.3

Table 5. Specification of Exhaust Gas analyser

Type	DELTA 1600 S
Object of measurement	Carbon Monoxide (CO), Carbon dioxide(CO2) and Hydrocarbon (HC)
Range of measurement	HC = 0 to 2000ppm as C2H4 CO = 0 to 10% CO2 = 0 to 16 % O2 = 0 to 21% NOX = 0 to 5000 ppm (as Nitric Oxide)
Accuracy	HC = ± 30 ppm HC CO = ± 0.2% CO CO2 = ± 1% CO2 O2 = ± 0.2% O2 NOX = ± 10% ppm NO
Resolution	HC = 1 ppm CO = 0.01% Vol CO2 = 0.1% Vol O2 = 0.01% Vol NOX = 1 ppm
Warm up time	10 min (self-controlled) at 20^o C
Speed of Response Time	Within 15 sec. for 90% response
Sampling	Directly sampled from tail pipe
Power source	100 to 240 V AC/50 HZ
Weight	800 g
Size	800 mm x 210 mm x 50 mm

Table 6. Specifications of Smoke Meter

Type	HARTRIDGE SMOKEMETER-4
Object of measurement	Smoke
Measuring range opacity	0–100%
Accuracy	± 2% relative
Resolution	0.1%
Smoke length	0.43 m
Ambient Temperature Range	−5 ^o C to +45^o C
Warm up time	10 min (self-controlled) at 20^o C
Speed of Response Time	Within 15 sec. for 90% response
Sampling	Directly sampled from tail pipe
Power supply	100 to 240 V AC/50 HZ
Size	100 mm x 210 mm x 50 mm

3.2. EGR set-up

EGR setup was used in investigating the feasibility of pyrolysis oil in reducing the NOx emission. The exhaust gases were cooled before it flows the intake manifold. To regulate the exhaust gases, a control valve is provided. EGR percentages were varied in steps of 5% from 15% to 25%.

3.3. ANN approach for the present study

ANN approach is not new in the field of the diesel engine to predict its performance. The present study focuses on ANN approach in predicting the performance of CRDI engine with EGR and different injection strategies.

4. Results and discussions

In this section experimental and ANN model results were presented and analysed.

4.1. Experimental results

In the present work, the experiments were carried out at 80% load with the influence of injection strategies such as injection timing (IT), Injection pressure (IP) along with exhaust gas circulation (EGR). Fuel IT was varied from 15^obTDCto 5^obTDC, IP from 800 bar to 1000 bar and EGR with 15% to 25% in steps of 5%. With three inputs at three levels, 27 combinations are prepared using design of experiments (DOE) and are given in Table 7. Table 4 provides the details of experimental data used for ANN modelling. The current results and discussions highlights on construction of ANN prediction model with different CI engine operating parameters. This paper provides the findings with PPO and diesel blend (70:30 ratio by volume).

Table 7. Experimental data for modelling of ANN

Trial no.	Parameter settings			Characteristics
	IT (t) (^obTDC)	IOP (p) (Bar)	EGR (e) (%)	BTE (%)	Smoke (HSU)	HC (ppm)	NOx (ppm)	Pmax (bar)	HRR (J/^0CA)
1	15	800	15	28	53	51	600	69	66
2	15	800	20	27.5	56	54	502	67.5	64
3	15	800	25	26.92	63	58	403	65	60
4	15	900	15	28.5	50	49	620	72	68
5	15	900	20	28	54	52	520	70	66
6	15	900	25	27.4	60	56	421	67	63
7	15	1000	15	28.3	52	50	608	70	67
8	15	1000	20	27.8	55	53	510	68	65
9	15	1000	25	27.1	62	57	411	66	62
10	10	800	15	28.5	50	46	622	72	68
11	10	800	20	28	53	49	515	70	66
12	10	800	25	27.4	58	54	410	67	62
13	10	900	15	29.2	47	42	650	74	70
14	10	900	20	28.6	50	46	540	72	68
15	10	900	25	28	55	51	428	69	65
16	10	1000	15	28.8	49	45	630	73	69
17	10	1000	20	28.3	52	48	526	71	67
18	10	1000	25	27.7	56	53	414	69	64
19	5	800	15	28.1	52	50	605	70	67
20	5	800	20	27.6	55	53	507	68	65
21	5	800	25	27.0	62	57	408	66	61
22	5	900	15	28.6	49	48	625	73	69
23	5	900	20	28.1	53	51	524	71	67
24	5	900	25	27.5	59	55	425	67.5	64
25	5	1000	15	28.14	51	49	612	71	68
26	5	1000	20	27.91	53	52	515	69	66
27	5	1000	25	27.1	60	56	416	67	63

4.2. Construction of ANN model using MATLAB

In the present study, series of experiments are performed on the CRDI engine with three different input parameters at certain step intervals. As per the studies of (Najafi et al. 2009; Sozen, Arcaklioglu, and Ozalp 2003; Roy et al., 2014b), it can be concluded that different ratios are to be assumed for test and training data. Input parameters are t (IT), P (Injection pressure) and e (EGR). Outputs for the experiments are measured and recorded in Table 7. Network parameters developed for MATLAB are shown in Table 8. ANN was repeated for five times to find out ideal inputs for the maximum and minimum values of the given output parameters with higher accuracy than readings. Totally, 1331 possible combinations of the 11 reading of three input parameters were made and matrix was created. Artificial Neural networking’s regression tool of the MATLAB was utilised to obtain the solution. The default settings for the training function are used, input and target data has been provided as the recorded data from the experiments in the module. Layer 1 neurons are selected and 18 as arbitrary value. After creating this network, the next step is to provide data for the training of ANN network.

Table 8. Network developed for MATLAB

MATLAB
Topology	3 inputs 5 outputs and 2 hidden layers with 9 hidden neurons in each layer
Data	Training Subject: 70% randomly selected observation data
	Test Subject: 30% randomly selected observation data
Activation	Log-sigmoid
Training algorithm	Feed Forward Back-Propagation
Loss function	Minimum MSE
Stopping criteria	Stop the network training when the variation error starts increasing

The developed model has been tested against statistical errors viz., root-mean-square error (RMSE) and mean absolute percentage error (MAPE). These are calculated using relations given in the literature (Chakraborty, Roy, and Banerjee 2016) and are provided in Table 9.

Table 9. Fraction of variance [R²], RMSE and MAPE values

Parameters	R^2	RMSE	MAPE (%)
BTE	0.99938	1.024	0.8
HC	0.99567	0.899	1.1
NOx	0.99862	0.725	0.45
PP	0.99999	0.135	0.61
HRR	0.99999	0.239	0.72

4.3. Prediction of ANN for the experimental data

From the establishment of the experimental work, an ANN model was developed to predict BTE, HC, NOx, PP and HRR as shown in Figure 3. The results obtained by ANN for the engine performance studies were impressive, i.e. the ability to predict BTE, HC, NOx, PP and HRR was commendable. Figure 3 indicates the steady and admirable concurrency of the network over the entire range of operation. Further, it indicates the innate sensitivity and toughness of the network in its ability to map the performance, combustion and emission values simultaneously with excellent accuracy.

Figure 3. Architecture of the proposed ANN

4.3.1. Brake thermal efficiency

Figure 4(a,b) represents predicted BTE values of CRDI engine. The ANN predicted value of BTE exhibits low mean square with experimental data. The predication of ANN on BTE is in persistent with the experiment results in all cases. ANN predicted BTE results when compared with experimental BTE results have R², RMSE and MAPE values of 0.99938, 1.024 and 0.8, respectively. The predicted maximum BTE was observed to be 29.1669% at IT of 9^obTDC with IP 860 bar and EGR of 15% while minimum BTE was predicted as 27.03033% at IT of 15^obTDC with IP of 820 bar and EGR of 25%.

Figure 4. (a) Variation of BTE with IT and EGR (b) Correlation coefficient for BTE

4.3.2. Unburnt hydrocarbon

Figure 5(a,b) shows the HC emissions of CRDI engine with ANN prediction. The ANN predicted value with the observed data was 0.97762. Minimum HC was predicted as 43 at IT of 11^obTDC with IP 920 bar and EGR of 15% while predicted maximum HC was observed to be 59 ppm at IT of 13^obTDC with IP of 1000 bar and EGR of 25%. The predicted values are in good agreement with the experimental results. ANN predicted HC results when compared with experimental HC results have R², RMSE and MAPE values of 0.99567, 0.899 and 1.1, respectively.

Figure 5. (a) Variation of HC with IT and EGR (b) Correlation coefficient for HC

4.3.3. Oxides of nitrogen (NOx)

Figure 6(a,b) represents the prediction of NOx emission of the CRDI engine. ANN predicted value was observed to be 0.96466. To reduce NOx emission IT of 13^obTDC with IP 960 bar and EGR of 25% could be used. Minimum NOx was observed to be 311 ppm at IT of 13^obTDC with IP 960 bar and EGR of 25% while maximum NOx was 677 ppm with IT 12^obTDC with IP of 920 bar and EGR of 15%. ANN predicted results when compared with experimental results have R², RMSE and MAPE values of 0.99862, 0.725 and 0.45, respectively.

Figure 6. (a) Variation of NOx with IT and EGR (b) Correlation coefficient for NO_x

Table 10 provides the comparison of current results with current emission regulation. It clearly shows that the results obtained are similar or very close to the current emission norms.

Table 10. Comparison of results with current emission regulation

Emissions	CO gms/Kwhr	HC gms/Kwhr	NOx gms/Kwhr	PM gms/Kwhr
BS-IV	1.5	0.96	3.5	0.02
Expt. Results	1.4	0.977	3.6	0.02

4.3.4. Peak pressure and heat release rate

Predication for PP and HRR was depicted in Figures 7(a) to 8(b). NN tool exhibits an excellent agreement with experimental results and correlation coefficient (R) for PP and HRR were 0.98168 and 0.96702, respectively. ANN predicted PP results when compared with experimental PP results have R², RMSE and MAPE values of 0.99999, 0.135 and 0.61, respectively. ANN predicted HRR results when compared with experimental HRR results have R², RMSE and MAPE values of 0.99938, 0.239 and 0.72, respectively.

Figure 7. (a) Variation of PP with IT and EGR (b) Correlation coefficient for PP

Figure 8. (a) Variation of HRR with IT and EGR. (b) Correlation coefficient for HRR

5. Conclusions

The present study reveals the results of the CRDI engine run on PPO with different IT, IP and EGR using ANN models developed. The following conclusions are drawn within the range of process parameters selected in the study:

• CRDI engine operation was smooth with PPO injected at maximum IP of 1000 bar.

• ANN predicted results were in good agreement with the CRDI engine experimental results.

• The results predicted by ANN matches with experimentation data with overall high accuracy with R in the range of 0.964 to 0.9816

• It is an effective forecasting tool employed to replace the experiment work for the desired output.

• It provides a complete platform in predicting the virtual sensing in optimising the real time with established ANN architecture.

• The predicted maximum BTE was 29.1669% at IT of 9^obTDC, IP of 860 bar and EGR of 15%.

• Statistical analysis shows that the ANN predicted and experimental results are in close agreement with each other.

On the whole, it can be concluded that to reduce the NOx emission IT of 13^obTDC with IP 960 bar and EGR of 25% have to be used for CRDI engine. This reduced the NOx emission by 38% as compared to CI mode, without much compromise in BTE. In addition, CRDI engine operation could give complete freedom from diesel and thereby provides energy security and sustainable of a nation.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the Deanship of Scientific Research, King Khalid University [Grant Number R.G.P.1/86/40.].

Notes on contributors

S. V. Khandal

Dr. S. V. Khandal is currently working as Associate Professor in the Department of Mechanical Engineering, Sanjay Ghodawat University, Kolhapur-416118, Maharashtra State, India. He has experience of 12 years in teaching and 07 years in research and published 40 plus research articles in peer reviewed journals and 1 book chapter. His research interests are in the areas of alternative fuels, thermodynamics, combustion and renewable energy sources.

Sudershan B. Gadwal

Dr. Sudershan B. Gadwal is currently working as Associate professor and Head of the Department in Mechanical Engineering at A.G. Patil Institute of Technology, Solapur, India. He completed his PhD. in Mechanical Engineering and M.Tech in Thermal power Engineering. He is having 16+ years of teaching experience and 05 years in research and published more than 10 research articles in Peer revived journals. His research area includes Alternative fuels, sustainable energy, Refrigeration and Air conditioning. He has published papers in International journal and National conferences.

Venkatesh A. Raikar

Dr. Venkatesh A. Raikar is currently Professor in the Department of Mechanical Engineering and vice chancellor, Sanjay Ghodawat University, Kolhapur-416118, Maharashtra State, India. He has experience of 35 years in teaching and 15 years in research and published 25 plus research articles in peer reviewed journals. His research interests are in the areas of robotics, industry 4.0 and renewable energy sources.

T.M. Yunus Khan

Dr. T. M. Yunus Khan received his Ph.D. degree from University of Malaya, Malaysia. He is currently working as Assistant Professor in Department of Mechanical Engineering, King Khalid University, Saudi Arabia. His research focuses are Heat transfer, fluid flow and biodiesel production and IC engines.

Irfan Anjum Badruddin

Dr. Irfan Anjum Badruddin received his Ph.D. from University Sains Malaysia, Malaysia. He is currently working as Professor in Department of Mechanical Engineering, King Khalid University, Saudi Arabia. His interest includes, heat transfer in porous medium, blood flow in human body, biodiesel production, Renewable energy.

Appendix 1

Table

[x y z] = ndgrid(sets{:}); cartProd = [x(:) y(:) z(:)]; sets = {[15 14 13 12 11 10 9 8 7 6 5], [800 820 840 860 880 900 920 940 960 980 1000], [15 16 17 18 19 20 21 22 23 24 25]}; [x y z] = ndgrid(sets{:}); cartProd = [x(:) y(:) z(:)];