“Marine” Diesel Rate of Heat Release

89 In order to estimate the general applicability of the ANN and phenomenological modeling approach, the heavy-duty engine trained network and calibrated model are applied to the marine diesel operating conditions in a blind try, i.e. without recalibra- tion. While the blind try ANN simulation cannot predict any of the combustion characteristics, the phenomenological model simulation, except for the short com- bustion durations, accurately predicts trends and variations in ϕ 10 , ϕ 50 and ϕ 90 values c.f. Figure 8.7 and Section 4.4.3.

8.3 Nitrogen Oxide Soot Emissions

In addition to the ROHR characteristics investigation above, the ANN approach is used to simulate both emissions of nitrogen oxide and soot for all three engines. Common network architectures as described in Section 3.4.2 are used, with the seven key operating condition parameters c m , BMEP, p Inj , Δ t Inj , ϕ SOI , x EGR , and λ global as inputs, and the specific nitrogen oxide and soot emissions m NO and m soot as outputs.

8.3.1 “Heavy-Duty” Diesel

Both measured and simulated specific nitrogen oxide and soot emissions for the standard heavy-duty diesel operating conditions Table A.2 are given in Figure 8.8. Whereas in the training sets there are only minimal variations noted between the measured and simulated emissions, distinct deviations are observed for several veri- fication operating conditions. While the deviations in specific nitrogen oxide emis- sions are caused only by operating condition settings outside the range of the trained settings e.g. 2324, 293032, deviations in specific soot emissions appear for operating conditions even within the range of the trained settings e.g. 37. a b Fig. 8.7 Comparison of Blind Try Marine Diesel Engine ROHR Simulations: a Artificial Neuronal Network, b Phenomenological Modeling N o rm a li z e d T im e 1 5 ϕ 9 = 1 [ a .u .] -700 -600 -500 -400 -300 -200 -100 100 200 Marine Operating Conditions [ -] 5 10 15 20 25 ϕ SOC Sim Meas ϕ 10 Sim Meas ϕ 50 Sim Meas ϕ 90 Sim Meas N o rm a li z e d T im e 1 5 ϕ 9 = 1 [ a .u .] -20 20 40 60 80 100 120 140 Marine Operating Conditions [ -] 5 10 15 20 25 ϕ SOC Sim Meas ϕ 10 Sim Meas ϕ 50 Sim Meas ϕ 90 Sim Meas 90 Besides the negative specific soot mass predicted for operating condition 31, a large discrepancy between the measured and simulated soot emissions for operating condition 37 is noted. The operating condition settings and measured, ANN, and phenomenological model simulated specific soot emissions for the operating condi- tions 13, 18, and 37 given in Table 8.4 indicate, that an increase in specific soot emissions with increasing EGR rate is expected. Whereas this is true for the mea- sured and the simulated values for the operating conditions 13 and 18, only the model prediction for the operating condition 37 shows this behavior. While the measured emission for 37 EGR = 17 is more than 300 of the value mea- sured for 18 EGR = 27 , the ANN predicted emission for 37 is even 50 lower than that for 13 no EGR, implying that the ANN does not capture the influence of EGR. Although the experimental soot measurements of operating con- dition 37 are suspected to be statistical outliers, the decrease in soot emissions for an increase in EGR from 0 to 17 , as predicted by the ANN, is even less plausible.

8.3.2 “Marine” Diesel

Similar to the heavy-duty results given in Figure 8.8 a, the ANN nitrogen oxide emission simulations for the two-stroke marine diesel engine show an almost perfect agreement between measured and simulated values for training operating conditions and a qualitatively correct prediction of the measured values for all verification oper- ating conditions Figure 8.9 a and b. a b Fig. 8.8 Heavy-Duty Diesel Emissions ANN Training Verification: a Nitrogen Oxide Emissions, b Soot Emissions