Research on a China 6b heavy-duty diesel vehicle real-world engine out NOx 1 emission deterioration and NOx correction for humidity and temperature using 2 on-board sensors and big data approach

7 China VI standard proposed higher requirements for durability of heavy-duty diesel vehicles 8 emissions. Previous research which took advantages of both on-board sensors and big data 9 approach to get the NOx deterioration factor was rather scarce. This paper used big data approach 10 to study the deterioration of engine out NOx emission based on 254,622 km operation data getting 11 from the on-board sensors or ECUs. Meanwhile, a formula for on-board NOx correction for ambient 12 humidity and temperature had been fitted. The analyses revealed that engine out NOx would not 13 be deteriorated during the useful life or even longer (deterioration factor was 1.014 @700,000km). 14 For a steady working condition, the engine out NOx mass flow (g/h) is negatively linearly correlated 15 with absolute humidity (R 2 = 0.997). If Ha was lower than 12g/kg, Ha almost had no effect on engine 16 out NOx concentration (ppm). Otherwise, there was also a negatively linear relationship between 17 them (R 2 = 0.978). It is hoped that the methods and conclusions of this paper could provide some 18 enlightenment for future NOx emission deterioration research.


Introduction
NOx sensors location and aftertreatment configuration (schematic plot). 84 uploads the data packets to the cloud platform with 1Hz frequency based on https protocol through 93 its built-in wireless communication module. After receiving the data packet, the cloud platform 94 analyzes, stores and visualizes it. Meanwhile, the cloud platform provides functions such as data 95 retrieval, download, statistical analysis, API (Application Programming Interface) and so on. 96 The Parameters collected in the durability test including ambient parameters, vehicle 97 parameters, engine parameters and aftertreatment system parameters were shown in Table 2   The "Raw Dataset" used in this paper consisted of 11,250,364 lines (observed value) and 37 118 Figure 5(a) shows the ambient temperature in the "Raw Dataset" with abnormal points. Figure  126 5(b) shows the modified ambient temperature. As we know, generally, ambient temperature would 127 not suddenly severe change in a short time, so the pre-value or post-value both can be used as an 128

alternative. 129
In this paper, the previous value fill method ("ffill" method in python) is used to deal with the 130 modifiable outliers. In the same way, this method was also fit for the outliers of other signals which 131 would not suddenly severe change in a short time, such as atmospheric pressure, longitude/latitude, 132 fuel/urea level, odometer, etc. 133  As shown in Figure 6(a), the engine out NOx concentration in the "Raw Dataset" has many 136 abnormal points which were less than zero. Since the engine out NOx concentration was related to 137 the engine working condition, and for in-use vehicles, the engines usually operate in transient 138 conditions, it is impossible to replace the abnormal points with the pre-value or post-value. 139 Generally, for the high-thin datasets, the unmodifiable outliers can be directly deleted.  Due to the concentration was measured on a dry basis for the on-board NOx sensor, the 148 measured concentration shall be converted to a wet basis according to the following Equations (1-149 T is temperature of the intake air,℃; a R is relative humidity of the intake air,%; a P is saturation vapour pressure of the intake air, kPa; B P is total barometric pressure, kPa.

 NOx mass flow 151
The calculation of the instantaneous NOx mass flow shall be according to the following 152

Feature selection 169
In this paper, we mainly focused on engine out NOx. However, the vehicle location parameters, 170 aftertreatment system parameters, and some indicative parameters (e.g. fuel level) didn't directly 171 affect the engine out NOx emission. These features can be dropped during data preprocessing. 172 Moreover, the derived parameters obtained in section 3.1.2 can be incorporated into the dataset 173 for engine out NOx analysis. 174 After the above data preprocessing, we can get the "Clean Dataset", and the parameters of 175 "Clean Dataset" were shown in Table 3

Results and Discussion
Based on the data getting from 254,622 km durability test of the tested vehicle, we carried out 179 the research for engine out NOx emission deterioration, at the same time, completed the NOx 180 correction for ambient air temperature and humidity. 181

182
In this paper, the "Clean Dataset", "Basic Dataset" and "Analysis Dataset" were obtained 183 according to the methods shown in Section 3.1, the datasets scale，specific methods and constraint 184 conditions were shown in the Table 4. 185 The "Raw Dateset" in Table 4 was the data of 254,622 km durability test. The "Clean Dataset" 186 was got by data preprocessing described in Section 3.2 in this paper. 187   should ensure the dataset used for the research was a "steady" one. 203 On the "Basic Dataset", although the engine speed and engine torque were almost the same, 204 the dataset contained many transition operating points, some outliers, and shot noise, etc. Therefore, 205 in order to obtain the dataset of a certain steady working condition as accurate as possible, it is 206 necessary to filter the "Basic Dataset" to eliminate the influence of transition working conditions 207 and noise points on the final results. 208 For a given diesel engine and a given working condition, engine out NOx is mainly affected 209 by intake air parameters and fuel injection parameters. The intake air parameters are main affected 210 by ambient air parameters. So, we could use these parameters to filter the "Basic Dataset", then we 211 would get the "Analysis Dataset". 212  Intake air parameter filtering 213 Figure 8(a) shows the intake air mass flow versus odometer on "Basic Dataset". As it can be 214 seen, there were many outliers and shot noise points, which were basically generated by transition 215 Figure 8(b) shows the frequency distribution of the intake air mass flow. As it can be seen, the 217 intake air mass flow distribution was relatively concentrated, so interquartile range (IQR) can be 218 used for filtering. If the value was less than the minimum or greater than the maximum showed in 219 Figure 9, it would be dropped. 220  We got "Dataset_D" after filtering injection timing by IQR. 237 can be seen, for a given operation condition of the tested vehicle, the larger the injection timing is, 239 the higher the fuel consumption is. As we know, the larger the injection timing is, the higher the 240 combustion temperature is, the higher combustion temperature would lead to higher engine out NOx 241 emission. So, we had taken both fuel consumption rate and injection timing as the constraint 242 conditions to ensure the consistency of the injection parameters. 243 to -0.587 (on "Dataset_D"). 261 As it was shown in Figure 12(a) Step2: Calculating the instantaneous engine out NOx mass flow at the observed Ha by using k 269 and b in step 1 (Equation 12 Step4 : Setting a range [0.95, 1.05] for ratio, filtering by the ratio, then getting a new dataset. 272 Step5: Repeating step1-4 until the difference of two successive iterations k is less than 1%. 273 "Analysis Dataset" respectively, as it can be seen, after the abovementioned filtering, the engine 279 power on "Analysis Dataset" was all almost the same in 254,622 km durability test. So "Analysis 280 Dataset" could be considered as a "steady" dataset.

Analysis of engine out NOx emission deterioration 284
On "Analysis Dataset"，the Pearson correlation coefficient between engine out NOx mass flow 285 and Ha was -0.947 (Table 5), without considering the deterioration , engine out NOx mass flow was 286 most related to Ha. 287 Therefore, further taking Ha = 15±0.5 g/kg which could cover a wider range of mileage or 290 seasons (Figure 11(b)) as a constraint condition, we can got a new dataset named "Dataset_E" on 291 which we could analyse the deterioration of engine out NOx emission. 292 The deterioration function and deterioration factor were shown in Table 6. As it can be seen, 297 within 254,622 km, for the maximum weight working condition (including ambient condition) of 298 the tested vehicle, the deterioration factors of engine out NOx mass flow and engine out NOx brake 299 specific emission were 1.001 and 1.001, respectively 300  see Figure 2) of the tested vehicle from September 2019 to November 2020 on "Clean Dataset", the 309 ratio of relative humidity greater than 50% was 93.5%, and that greater than 60% was 84.5%. Figure  310 15(b) shows the distribution of relative humidity on "Analysis Dataset", the ratio of relative 311 humidity greater than 50% is 97.8%, and that greater than 60% is 93.4%. Figure 15( Figure 16(b) showed the distribution of ambient temperature on "Clean 317 Dataset" and "Analysis Dataset" respectively. As it can be seen, the ambient temperature is mainly 318 affected by the season (Figure 16(c)), and the ambient temperature is almost above 0℃ in Southeast 319 China all the year round.
Where: RHa is relative humidity, %;Ta is ambient temperature,℃. 330 If the relative humidity remained the same or varied in a small range, Ha had an exponential 331 relationship with the ambient temperature. 332 When the ambient temperature was at a low level, the variety of relative humidity has little 333 effect on Ha, in other words, the lower ambient temperature would directly lead to lower Ha in 334 27 winter. 335 The larger the Ha is, the larger the specific heat capacity of the intake air is. Larger Ha would 336 lead to lower combustion temperature and lower engine out NOx emission. So, Ha has a significant 337 effect on engine out NOx emission. 338 339 Figure 18 The relationship between Ha and ambient temperature/relative humidity For a given steady working condition (e.g. "Analysis Dataset"), according to ideal gas law PV= 368 nRT, the intake air mass flow will be affected by intercooler outlet temperature. 369 To sum up, for a given steady working condition, whether ambient temperature or Ha, they had 390 different effect on engine out NOx concentration (ppm) at different sections; the exhaust mass flow 391 rate was mainly affected by the ambient temperature; engine out NOx mass flow (g/h) was mainly 392 affected by Ha. So, we can take the engine out NOx mass flow (g/h) as the correction target, Ha as 393 the independent variable to study a formula for NOx correction for ambient air temperature and 394 Step2: Setting Ha = 10.71 g/kg as the reference, that is to say Kh,d = 1 @ Ha = 10.71 g/kg . 400 Step3: Calculating the average value of engine out NOx mass flow @ Ha = 10.71±0.5 g/kg by 401 Step6: Using k and b obtained in step 5 to correct engine out NOx mass flow. 406 "Analysis Dataset". As it can be seen, after ambient temperature and humidity correction, engine 415 out NOx mass flow was on the same level in different seasons, so as the engine out NOx brake 416 specific emission (Figure 25(b)). 417 Table7 shows the deterioration function (linear) of the corrected engine out NOx mass flow 420 (g/h) and engine out NOx brake specific emission (g/kW.h). 421 Compared to Table 6, the deterioration factor (@254,622 km) of engine out NOx was 0.004 422 correction for ambient air temperature and humidity were suitable for on-board sensors and could 427 be used for the deterioration analysis for the in-use vehicle engine out NOx emission. 428

430
This study had analyzed the deterioration of the engine out NOx by using 254,622km durability 431 test data and big data approach for the tested vehicle. Then, within the odometer that no engine out 432 NOx deterioration had happened, we had completed the NOx correction for ambient temperature 433 and humidity for NOx sensor. The main conclusions of our present study can be summarized to the 434 following: