Ammonia concentration detection method for an internal combustion engine

Abstract

Method for calculating NH3 and NOx concentrations downstream from an SCR system of an internal combustion engine (E) of a vehicle by means of a statistical module, the method comprising the following steps: using temperature measurements acquired upstream (T1) and downstream (T2) of the Selective Reduction Catalyst (SCR) as first input variables and NOx mass flow measurements acquired upstream (N1) and downstream (N2) of the SCR as second input variables; calculating an NH3 concentration downstream from the SCR catalytic converter as a function of the first and second inputs.

Description

[0001] DESCRIPTION

[0002] "AMMONIA CONCENTRATION DETECTION METHOD FOR AN INTERNAL COMBUSTION ENGINE"

[0003] Field of the invention

[0004] The present invention relates to a method for ascertaining a nitrogen oxide (NOx ) concentration and ammonia (NH3 ) concentration downstream from a Selective Reduction Catalyst ( SCR) of an internal combustion engine , preferably a Diesel engine .

[0005] Description of the prior art

[0006] Modern Diesel , future hydrogen, and other lean burning combustion engines use an SCR system to reduce nitrogen oxide (NOx ) . Thereby, a water urea-based solution (Adblue®) is inj ected into the hot exhaust gases upstream of the SCR system, which is trans formed into ammonia (NH3 ) , which in turn reduces the NOx on the catalyst surface into nitrogen (N2 ) and water (H20) . The amount o f inj ected Adblue must be adj usted such that neither NOx nor NH3 occurs in the tailpipe , hence ensuring maximum NOx reduction . During engine operation, both NOx and NH3 concentrations might occur in the tailpipe due to imperfect reduction or Adblue dosing . However, excess ive NOx emiss ions will result from insuf ficient Adblue inj ection, while secondary NH3 emissions will result from excessive Adblue inj ection . Thus , Adblue dosing control is a critical component of SCR systems as NH3 emissions are undesirable .

[0007] To determine the optimal amount of inj ected Adblue , a NOx sensor is installed in the tailpipe downstream of the SCR . Automotive NOx sensors , however , are cross-sensitive to both NOx and NH3 making it impossible to distinguish between them . This is problematic both for the controls to distinguish between Adblue over- and underdosing and for the monitoring of NH3 emissions , the latter being a new requirement of the upcoming Euro 7 legislation .

[0008] NH3 could be monitored using a speci fic NH3 sensor . However, the installation of such an NH3 sensor introduces signi ficant costs and a substantial development ef fort for a technology that could end in the next years .

[0009] KR20230044668A attempts to avoid the implementation of an NH3 sensor by implementing a first NOx sensor installed at an input end of the SCR system and a second NOx sensor arranged at the tail of the SCR system and a neural network that compares the measured values of the sensors to obtain a predicted value of NH3 concentrations . However, such method can return acceptable results only during homologation or other predetermined operation cycles , which are carried out in virtual and predetermined operating conditions . KR20230044668A mentions the possible implementation of further signals relating to the amount of intake fresh air, the amount of urea water inj ection, and the exhaust gas temperature , in addition to the concentration of nitrogen oxides in the SCR system . However, no further detail is given in connection with the above NOx signals comparison .

[0010] US2020224570A1 discloses another solution to avoid the implementation of an NH3 sensor . Here , the estimation is based on a plurality of state variables of the internal combustion engine and on the NH3 storage level of the SCR system . Summary of the invention

[0011] The main scope of the present invention is the implementation of a method for controlling the AdBlue inj ection and reliably monitoring NOx and NH3 emissions , which assures comparable results in comparison with the physical implementation of an NH3 sensor in the tailpipe downstream of an SCR .

[0012] The main principle of the invention is the statistical analysis of the NOx concentration over the SCR system in connection with the exhaust gas mass flow on the basis of the temperature measure at the SCR system .

[0013] The dynamics of the NH3 and NOx concentrations at the tail of the SCR are di f ferent because of di f ferent reaction and transportation pathways of the two species in and through the SCR itsel f . These dynamics depend on the temperature and the exhaust mass flow . Hence , NOx and mass flow measurements are statistically correlated on the basis of the temperature . This correlation assures an af fordable way to predict both the NH3 and NOx concentrations in the tailpipe downstream of the SCR .

[0014] It should be noted that , according to the present invention, neither any signal representative of the AdBlue inj ection, nor any consequent estimation of the NH3 storage level in the SCR is accounted for . This aspect is extremely relevant i f one considers the possible implementation of a series of two or even more SCR with their own AdBlue inj ection means .

[0015] Therefore , it should be clear that an af fordable method of estimating the NOx and NH3 concentrations which is independent of the speci fic layout of the after-treatment system is particularly relevant .

[0016] The occurrence of NH3 tailpipe concentrations is , due to chemical reactions and desorption, highly characteri zed by the temperatures , the mass flow and the NOx sensor signals , which makes these signals inevitable for an accurate distinction of NH3 and NOx concentrations in the tailpipe downstream of an SCR .

[0017] According to the present approach, the information whether NH3 concentrations are present or not is already contained in the evolution of NOx, mass flow and temperature measurements over time . In particular, the temperature is the "pivot variable" , which mainly influences the above dynamics .

[0018] An arti ficial intelligence architecture is trained such that it extracts the physical information about the NH3 concentration from temperature , NOx sensor, and exhaust gas mass flow signals and predicts NOx and NH3 concentration levels in the tailpipe . In other words , the arti ficial intelligence architecture acts as a virtual sensor, which in combination with the cross-sensitive NOx sensor, distinguishes between NOx and NH3 and provides an accurate estimation of the NH3 concentration .

[0019] The present method is simple enough to be implemented in the today' s engine control unit , thus , no speci fic computing hardware is necessary . Hence , the proposed solution is signi ficantly cheaper and more accurate than the real implementation of an NH3 sensor . These and further obj ects are achieved by means of the attached claims , which describe preferred embodiments of the invention, forming an integral part of the present description .

[0020] Brief description of the drawings

[0021] The invention will become fully clear from the following detailed description, given by way of a mere exempli fying and non limiting example , to be read with reference to the attached drawing figures , wherein :

[0022] Fig . 1 shows an internal combustion engine provided with an after-treatment system including an SCR and AdBlue® inj ection means and a control unit programmed to perform the strategy subj ect of the present invention;

[0023] Fig . 2 discloses in detail the after-treatment system according to Fig . 1 ;

[0024] Fig . 3 discloses the overlapping of the ranges of the training and validation data according to the strategy subj ect of the present invention in connection with a neural network implemented in the control unit of Figs . 1 and 2 ;

[0025] Fig . 4 discloses a diagram showing the experimental results achieved by means the present invention . The same reference numerals and letters in the figures designate the same or functionally equivalent parts . According to the present invention, the term " second element" does not imply the presence of a " first element" , first , second, etc . . are used only for improving the clarity of the description and they should not be interpreted in a limiting way .

[0026] Detailed description of the preferred embodiments

[0027] According to the present invention, with the aid of Figs . 1 and 2 , an internal combustion engine E , generally a Diesel type , provided with an intercooler AC to cool the fresh air pumped by the turbo-charger TCU and preferably an EGR cooler EGC to cool possible exhaust gasses to be recirculated .

[0028] The engine includes an intake pipe IP arranged to introduce fresh air in the engine cylinders and an exhaust pipe OP arranged to direct exhaust gas ses through the After- Treatment System (ATS ) , which has the task to reduce pollutants contained in the exhaust gasses .

[0029] The optional presence of the turbo-charger implies that the exhaust gasses , along the exhaust pipe OP, meet the turbine of the turbo-charger in a per se known configuration .

[0030] The ATS , according to Fig . 2 can optionally include several components , such as a Diesel Oxidation Catalyst ( DOC ) , a particulate trap ( DPF) , an NH3 oxidation catalyst , or any known device arranged to cooperate with the SCR to reduce pollutants such as HC and CO, etc . . . Instead of or in combination with the SCR, an SCR coated particulate trap may be integrated into the ATS . Moreover, two or more SCRs in any configuration as described above may be placed as a series of SCRs .

[0031] According to the present invention, what matters is the SCR, the sensors arranged to measure certain quantities upstream and downstream of the SCR, and a control unit ECU configured to calculate the NOx and NH3 concentrations downstream of an SCR .

[0032] The SCR could be also in the form of a particulate trap coated in such a way to develop the functions of an SCR . Usually, the particulate trap coated to combine also the SCR functions are named as "SCR on Filter" or SCRoF .

[0033] The inj ection means INJ are arranged to inj ect AdBlue® upstream of the SCR catalytic converters according to the exhaust gas flow GF direction .

[0034] In the following SCR means "SCR catalytic converter" , "SCR coated particulate trap" , or a combination of both .

[0035] Usually, but not necessarily, the inj ection means , in the form of inj ector, coupled with a pump and an AdBlue reservoir, inj ect the AdBlue in the pipe or a box arranged immediately upstream of the SCR to promote the AdBlue hydrolyzation .

[0036] According to the present invention, a first NOx sensor N1 and a first temperature sensor T1 are arranged upstream of the SCR system, namely upstream of the relating AdBlue inj ection means or upstream of the DOC, and a second NOx sensor N2 and a second temperature sensor T2 are arranged downstream of the SCR .

[0037] According to a preferred embodiment of the invention the DOC is provided of an electric heater in order to render prompt the response of the DOC at cold start .

[0038] The present solution needs to acquire the NOx mass flow; this means that in case the NOx sensors are capable to produce signals representative of the NOx mass flows , it is enough to the present scopes , otherwise , i f the NOx sensors are capable to produce signals representative of the NOx concentrations , then also the measurement or calculation of exhaust gas flow is needed .

[0039] The exhaust gas mass flow can be calculated in di f ferent ways , for example on the basis of pressures / temperatures measurements acquired at the intake pipe IP, at the engine outlet and on the engine speed, according to well-known models , such as the speed-density model .

[0040] An equivalent quantity to the exhaust gas mass flow is the space velocity, which can be calculated on the basis of the exhaust gas mass flow and the geometry of the SCR .

[0041] According to the tests executed, the implementation of the space velocity instead of the exhaust gas mass flow renders more accurate the training process of the statistical model .

[0042] The exhaust gas flow reaching the ATS is given by the di f ference between the total exhaust gas flow and the EGR gas stream .

[0043] EP3633169 teaches to determine the gas flow through a turbine on the basis of pressure and temperature acquired at the engine out and the pres sure acquired immediately downstream of the turbine . Alternatively, turbine models can be implemented to estimate the exhaust gas flow reaching the ATS , such as Guz zella , Onder : " Introduction to Modelling and Control of Internal Combustion Engine Systems" ISBN3-540-22274-X, Springer-Verlag, Berlin 2004 .

[0044] According to the present invention a control unit ECU is operatively connected to the above sensors Nl , N2 , Tl , T2 and to any further sensor needed to calculate the exhaust gas mass flow according to any of the above known techniques .

[0045] In the control unit ECU a neural network is implemented, namely a statistical module which has , as inputs , the signals generated by the above sensors and the exhaust gas mass flow calculated by means any known way to obtain such input .

[0046] It should be clear that the measured or calculated NH3 storage level is not , in any way, an input of the neural network after training .

[0047] This means that the number of SCRs arranged in series with or without corresponding inj ection means is completely irrelevant .

[0048] According to a first preferred implementation of the invention, multiple SCR systems , in series , are implemented . The sensors Nl , T1 are arranged upstream of the first SCR system met by the exhaust gasses and the sensors N2 , T2 are arranged downstream of the last SCR system met by the exhaust gasses .

[0049] According to a second preferred implementation of the invention, the sensors N2 , T2 are arranged immediately downstream of the first SCR met by the exhaust gasses , and further sensors N3 , T3 are arranged downstream of the second SCR system . In this way, the method is applied over each of the SCR systems , means the NH3 and NOx concentrations are determined immediately downstream of each SCR system, separately .

[0050] According to the present approach, the following observations are proposed in order to j usti fy the selection of the above inputs . The obtained NOx and NH3 concentrations downstream of the first SCR system are used as input to the second SCR system . Speci fically, the second SCR system now has a more accurate information about the input NOx concentration, as compared to the uncorrected sensor signal , and additionally an information about the NH3 concentration at the inlet , which may be taken into account in the subsequent SCR system in addition to the second AdBlue© inj ector .

[0051] NH3 adsorbs on the catalyst surface , from where it reacts with NOx . The rate of adsorption depends on the exhaust gas mass flow and on the temperature , both of the gas and the catalyst .

[0052] The rate of reaction between NOx and NH3 depends on the temperature of the catalyst , which in turn depends on the evolution of the temperatures of the exhaust mass flow .

[0053] NH3 , which does not react with NOx, desorbs from the catalyst surface . This process strongly, exponentially, depends on the catalyst temperature .

[0054] The amount of desorbed NH3 leaving the catalyst depends on the amount of stored NH3 on the catalyst surface in the last part of the SCR, which in turn depends on the adsorption, reaction, and desorption of NH3 along the entire SCR channels .

[0055] The signal produced by the NOx sensor in the tailpipe downstream of the SCR is representative of the sum of nonconverted NOx and desorbed NH3 . In view of what above , the concentration of both those quantities depends on the amount of NOx carried into the SCR, which is given by the reading of the NOx concentration through the upstream NOx sensor and the exhaust gas mass flow, the amount and axial distribution of stored NH3 , and the temperature .

[0056] The evolution of the downstream NH3 concentration is driven by the correlation between the upstream and downstream NOx concentrations , the exhaust gas mass flow, function of space velocity, catalyst volume , and the catalyst temperature , function of upstream and downstream catalyst temperatures .

[0057] The concentration of NH3 can be reproduced with the upstream and downstream NOx sensor signals , the upstream and downstream temperatures , and the space velocity .

[0058] The downstream temperature may be reproduced by the upstream temperature and the space velocity and thus can be left out in an alternative representation .

[0059] In view of the present considerations , the present invention bases the prediction of the NOx and NH3 concentrations in the tailpipe downstream of the SCR system on :

[0060] NOx mass flow measurements over the SCR, namely measured upstream and downstream of the SCR, Temperature measurements over the SCR system, namely measure upstream and downstream of the SCR system .

[0061] As disclosed above , when only NOx concentrations can be measured, then also exhaust mass flow passing through the ATS shall be measured .

[0062] About the neural network selection

[0063] Three main dynamic phenomena can be observed in an SCR system

[0064] Fast NOx storage dynamics , due to the weak adsorption, and transport delays ,

[0065] Slow NH3 storage dynamics , due to strong adsorption, Slow thermal energy storage and thus slow temperature dynamics .

[0066] Phenomena with di f ferent dynamics can be managed ef ficiently with Long Short-Term Memory ( LSTM) networks .

[0067] However, other recurrent neural networks (RNN) could be used .

[0068] Alternatively, static neural networks or other static machine learning models , e . g . , random trees , could be used applied on moving window data . However, using static neural networks leads to a substantial use of computer resources , both in terms of memory and calculation power .

[0069] About the training and validation data

[0070] The training data are a selection of data obtained from measurements and optionally from detailed simulation models .

[0071] The measurement data are bound to the behavior of the vehicle / engine on the road or on the bench .

[0072] One key aspect is the selection of the "right" data for training the neural network . Indeed, it is needed to be sure that all critical operating conditions of the system are covered, since a neural network cannot be used for extrapolation .

[0073] Thus , the aim of the "right" data acquisition is uni form distribution of all the values represented by the signals acquired .

[0074] System knowledge is used to remove regions from the training data which are invalid for the neural network to learn, such as , regions where the engine is not running or regions with inactive status bytes .

[0075] Preferably, the training data is cut into segments to prevent overfitting .

[0076] The length of one segment cut is chosen such that it includes the slow dynamics from temperature .

[0077] The segments are shuf fled after every learning iteration such that the initial state value of the neural network, which refers to the initial NH3 storage level of the SCR, does not af fect the prediction .

[0078] The NH3 concentration is learnt from the evolution of the input signals over time .

[0079] It is requested that the training data contain the entire temperature range of operation, preferably in combination with the entire ranges of the other quantities , such as NH3 / N0x ratio , exhaust gas mass flow or space velocity, and NH3 peak levels in the tailpipe downstream of the SCR system .

[0080] The first stage of the method is the acquisition of physical data from a real test bench engine plus ATS . In this stage even an NH3 sensor is implemented to acquire said NH3 peak level and preferably any gradient of the magnitudes of interest listed above .

[0081] Then acquisitions are made in order to assure uni formity as depicted above in terms of temperature range . Then the NH3 sensor is no more used, because the aim of the present solution is to avoid any implementation of the NH3 sensor after the training of the arti ficial intelligence training . Then, being the temperature the most relevant quantity which triggers di f ferent behaviors in the SCR system, it is important to implement training data, which include a uni form distribution of temperature values to trigger all temperature levels . This goes for all of the most relevant quantities . The uni formity of the respective values is determined by means of histograms with a defined number of bins , whose values represent the relative occurrence of the level of the respective quantity such as temperature level . In other words , a matrix is constructed, where each cell includes the number of histogram bins for a certain dimension and a certain range on this particular dimension . The selection of dimensions is chosen such that they enclose all relevant operating points of the ATS .

[0082] Preferably, beyond the above essential quantities for training, certain gradients are added to ensure suf ficient dynamic content in the data to train the dynamic neural networks . In particular, the gradients of the NOx concentrations and the temperatures are taken into account . This becomes crucial to enclose transient operations . In addition, it is preferable to implement also gradients in combination with recurrent neural networks .

[0083] According to a preferred aspect of the invention, the comparison between the training data and the validation data is a methodology to check, i f the training data form a well distributed "hull" around the range of the validation data . For example , i f the training data don' t cover the full range of the validation data or the intended operation range , the set o f training data has to be extended . Thus , the methodology is for the design or selection of the training data .

[0084] Fig . 3 shows an example of a training data and validation data selection . A desired "hull" distribution of the training data around the range of the validation data implies that the area to the training data must enclose the area of the validation data .

[0085] In other words , the neural network requires no extrapolation during prediction on the validation data .

[0086] According to a preferred embodiment of the invention, some training data are generated by per se known model , in order to supplement the training data acquired on-the-road .

[0087] Fig . 4 shows that the prediction of the neural network ( solid line ) is accurate compared to the measurement ( dotted line ) . The NH3 concentration from second 50 to second 150 is low, which is a result of low temperatures and thus limited NH3 desorption activity . The prediction of the network is accurate in all regions indicating that the network correctly predicts the system dynamics .

[0088] This invention can be implemented advantageously in a computer program comprising program code means for performing one or more steps of such method, when such program is run on a computer . For this reason, the patent shall also cover such computer program and the computer- readable medium that comprises a recorded message , such computer-readable medium comprising the program code means for performing one or more steps of such method, when such program is run on a computer .

[0089] Many changes , modi fications , variations and other uses and applications of the subj ect invention will become apparent to those skilled in the art after considering the speci fication and the accompanying drawings which disclose preferred embodiments thereof as described in the appended claims .

[0090] The features disclosed in the prior art background are introduced only in order to better understand the invention and not as a declaration about the existence of known prior art . In addition, said features define the context of the present invention, thus such features shall be considered in common with the detailed description .

[0091] Further implementation details will not be described, as the man skilled in the art is able to carry out the invention starting from the teaching of the above description .

Claims

CLAIMS1 . Method for calculating NH3 and NOx concentrations downstream from an SCR system o f an internal combustion engine (E ) of a vehicle by means of a statistical module , the method comprising the following steps :- using temperature measurements acquired upstream ( Tl ) and downstream ( T2 ) of the Selective Reduction Catalyst ( SCR) as first input variables and NOx concentration measurements acquired upstream (Nl ) and downstream (N2 ) of the SCR as second input variables and measurements or calculations of exhaust gas mass flow through the SCR as third input ; calculating an NH3 and NOx concentration downstream from the SCR catalytic converter as a function of the first , second and third inputs .2 . Method according to claim 1 , wherein the NH3 concentration is calculated evaluating an evolution of NOx concentration, mass flow and temperature measurements over time by means of said statistical model .3 . Method according to claim 1 or 2 , wherein no input from an NH3 sensor is evaluated and / or wherein the statistical module acts as a virtual NH3 sensor, which in combination with the NOx concentration input is capable to distinguish between NOx and NH3 concentrations providing said NH3 concentration calculation .4 . Method according to any one o f the previous claims 1 - 3 , wherein said exhaust gas mass flow is calculated by means of model on the basis of pressure and temperature measurements acquired upstream and downstream of the combustion engine (E ) in combination with a measure of an engine speed revolution .5 . Method according to any one o f previous claims 3 or 4 , wherein said exhaust gas mass flow calculation is in the form of space velocity calculation .6 . Method according to any one o f previous claims 3 or 4 , wherein the internal combustion engine is provided of a turbocharger ( TCU) and wherein said exhaust gas mass flow is calculated by means of model modelling a turbine of the turbocharger .7 . Method according to any one of the previous claims , wherein NH3 storage level is neither measured, nor calculated in order to calculate said NH3 concentration .8 . Method according to any one of the previous claims , wherein said statistical module is defined by an arti ficial intelligence architecture , preferably including a recurrent neural network and more preferably a Long Short-Term Memory ( LSTM) neural network, trained such that to extract said NH3 and NOx concentrations from temperature and Nox sensors and exhaust gas mass flow signal s to predict Nox and NH3concentration levels in a ATS tailpipe.

9. Method for training the artificial intelligence architecture according to claim 8, including a step of training the artificial intelligence architecture by means of patterns including uniform distribution of NOx mass flow values, temperature values, NH3 / N0x ratios, exhaust gas mass flows, and preferably NH3 peak concentrations.

10. Method according to claim 9, wherein said training step includes also transients of NOx concentration signals, mass flow calculations and temperature signals.

11. Method for controlling water urea injection means arranged immediately upstream of a SCR catalytic converter of a vehicle, the method comprising the steps: acquiring NH3 concentration downstream from the SCR catalytic converter according to the step of the method of claims 1 - 10 and- adjusting water urea injection in order to minimize the NOx and NH3 concentrations.

12. Control unit for calculating NH3 and NOx concentrations downstream from an SCR system of an internal combustion engine (E) of a vehicle implementing a statistical module wherein the control unit is suitable to be operatively connected with temperature sensors (Tl, T2) arranged upstream (Tl) anddownstream ( T2 ) of the SCR in order to receive first input signals andNOx sensors (Nl , N2 ) arranged upstream (Nl ) and downstream (N2 ) o f the SCR in order to receive second input signals , exhaust mass flow sensor or estimator as third input signal , the control unit being configured to calculate NH3 and NOx concentrations downstream from the SCR according to all the steps of any one of the previous claims from 1 to 11 .13 . The control unit (ECU) according to claim 12 further suitable to be operatively connected with a water urea inj ection means arranged immediately upstream of said SCR, Wherein the control unit is configured to calculate NH3 and NOx concentrations downstream from the SCR as a function of the first and second inputs and to control said water urea inj ection means in order to minimi ze the NH3 and NOx concentrations .14 . A system for the abatement of pollutants (ATS ) produced by an internal combustion engine , the system comprising- at least a first SCR Catalyst including first urea water inj ection means arranged immediately upstream of the firstSCR,- temperature sensors (Tl, T2) arranged upstream (Tl) and downstream (T2) of the SCR in order to generate first input signals and- NOx mass flow sensors (Nl, N2) arranged upstream (Nl) and downstream (N2) of the SCR Catalyst in order to generate second input signals, a control unit (ECU) operatively connected with said temperature sensors, said NOx mass flow sensors and said water urea injection means, wherein, the control unit is according to claim 13.

15. The system according to claim 14, further comprising at least one of the following devices: a Diesel Oxidation Catalyst (DOC) , optionally electrically heated, arranged upstream of said first SCR or immediately downstream of the first SCR, a Diesel Oxidation Catalyst (DOC) with NOx storage properties, arranged upstream of the said first SCR,- a Diesel Particulate Filter (DPF) , arranged between said DOC and said first SCR or between said first and second SCRs and / or embedded in one of said first or second SCR to define an SCR on filter,- an NH3 oxidation catalyst,- an NH3 oxidation catalyst arranged downstream of said second SCR.

16. A computer program comprising instructions to cause the control unit of claims 12 or 13 to execute the steps of the method of claim 1 .17 . A computer-readable medium having stored thereon the computer program of claim 16.18 . Internal combustion engine (E ) provided of a system for the abatement of pollutants (ATS ) produced by the internal combustion engine , according to any one of the claims 14 or

Images