DE212005000024U1 - Loading of particulate filter determination method for diesel engine involves using the artificial neural network from measured variables characterizing the engine operating state - Google Patents

Abstract

The method involves training an artificial neural network using the characteristic diagram for a particulate filter in a known state of loading, and loading of particulate filter is determined using the artificial neural network from measured variables characterizing engine operating state. The characteristic diagram indicates pressure difference as a function of exhaust gas temperature and mass. An independent claim is included for a regeneration control system of particulate filter.

Description

The The invention relates to a method for the design of regeneration strategies a particulate filter, in particular of quality-controlled internal combustion engines.

particulate filter systems for engines show promising potential for reducing exhaust emissions. Filteryteme can Retain solid components (as well as substances attached to them) in the exhaust gas. Coatings on the filter (i.d.R. catalytic) can also gaseous Reactivate components reactively. It can thus - depending on the filter efficiency - solids be separated out of the exhaust. The remaining exhaust gas escapes on the back side of the filter while the solids of the appropriate size remain on or in the filter. Change the filter properties through this accumulation or incorporation of solids. While at the flow empty filters already show a drop in pressure increases this as the loading of separated solids progresses. In general, the degree of loading of a filter is just this Size of the pressure drop (dp) described. Will the pressure drop and thus the work must be applied to flow through the filter for the respective Application too high, so there is a need to clean it. ever according to chemical nature of the deposited solids can this cleaning process will vary. For solids such as diesel particulates, for the most part consist of carbon (C), this cleaning process z. B. from a temperature increase the flowing through Mediums exist. By the increase the temperature and in compliance with certain boundary conditions (such e.g. sufficient oxygen content) conditions are created under which the particles oxidized for the most part can be. The products of this oxidation - mainly carbon dioxide (CO2) and carbon monoxide (CO) - lie in gaseous form and escape during of the process, which is called regeneration, at the exit of the Filter.

Becomes a particle filter used for exhaust aftertreatment, it is mandatory necessary to regenerate this. The regeneration can be active, as well passively triggered and operated. Passive regeneration is when, without external intervention, oxidation conditions enter. This would be for example, a high-speed drive with a vehicle the highway.

at the active regeneration, the oxidation conditions by generates such an active intervention in the overall system. It could be here for Example, a burner ignited in the exhaust system.

In the vast majority Part of the operating map of a diesel engine are the necessary Regeneration conditions not achieved without action. It must therefore Energy is spent, which, of course, the operating costs of the Motors is attributable. Overall, the use of a particulate filter system rate of deterioration, as the back pressure of such System acts on the charge cycle of the engine and this can "breathe" almost less badly. The regeneration in itself participates also negatively in the total consumption of Engine. In order to outweigh the positive properties, namely the reduction of the emitted particle mass to let, will be worked intensively on the inevitable Consumption deterioration as low as possible.

Around to ensure optimal processes (loading and regeneration) is a process monitoring unavoidable. As a rule, different variables are used such process monitoring allow. Necessary sensors for generating the measured quantities are either before, i. they have already been used to monitor, control or Regulation of other processes of the overall system applied, or must be specially added become. additional Sensors for the overall system cause of course Costs that must always be kept as low as possible. Also in the case of the overall system engine and exhaust aftertreatment means Diesel particulate filter, can be accessed on the size of the Operation of the engine anyway, on more sensors can However, not be waived. There are numerous developments at sensors available, which make a process description of the particle filter subsystem possible. However, these differ greatly by price and functional stability. Established Here are pressure and temperature sensors.

In one of today's common Process monitoring, as described in DE-A-42 30 180, the pressure before (and possibly also after) the particle filter measured. This is especially true because of this, because (as above) the back pressure created by the filter directly affects its efficiency. The pressure increase while Loading phases, as well as the pressure drop during regeneration can be measured and compared with limits. By the temperature measurement before the filter is monitored whether in designated regeneration phases also according to necessary thermal conditions are present. In practice, throws one purely pressure-based process monitoring but some problems.

The exhaust back pressure generated by filters depends on several factors. One of the main influencing factors is the exhaust volume, wel It always changes under unsteady engine operating conditions. Furthermore, pulsations occur in an exhaust system, which influence the measurement result. In sum, therefore, the pressure is a very unsteady measure. Usually, an upper and lower limit characteristic (oGk, uGk) is defined, the achievement of which triggers loading or regeneration end.

1 should clarify the problem: depending on the exhaust gas volume flow and the filter load a differential pressure dp can be measured. Like every measured variable, this is subject to errors. While at high flow rates the measurement is relatively uncritical, it becomes clear at low flow rates that, regardless of the load, but depending on the random measurement error, both the lower and the upper limit value can be achieved. Specifically, this means that the process control depending on the measurement error, could give both the command for regeneration, but also to cancel the regeneration. In the case of the loading phase would therefore start too early with the regeneration, which may have an adverse effect on fuel consumption. Considering a regeneration phase, the emptying would be stopped too early and on the filter would remain a residual amount. This case can also have a negative impact on fuel consumption. In practice, the regeneration is maintained for a period of time after the termination signal of the monitoring to ensure the safe emptying. This necessary time delay of the process can only be controlled, not regulated, and therefore also leads to increased fuel consumption.

It is therefore an object of the present invention, an improved Method and an improved system for controlling the regeneration of a in the exhaust gas purification system of a diesel engine arranged particulate filter provide. In particular, the increase in fuel consumption due to the regeneration of the particulate filter can be reduced or minimized.

These Tasks are characterized by the features contained in the claims solved.

The The present invention is based on the basic idea of loading a Particulate filter using an artificial neural network to determine. The neural network is inventively through a characteristic diagram of the particulate filter, preferably by the Filter caused pressure difference as a function of the exhaust gas temperature and the one flowing through the filter Exhaust mass describes trained at a known load. That trained neural network can then be based on metrics that the Characterize the operating condition of the engine, the current load of the filter. These measures are preferably by the filter caused pressure difference, the exhaust gas temperature and the flowing through the filter Exhaust gas mass.

The The invention provides a system for controlling the regeneration of a particulate filter ready with the start time of the regeneration of a particulate filter can be determined. This is the current loading of the particulate filter determined using the neural network described above and this compared with a predetermined threshold. This way will initiated the regeneration phase of the particulate filter only if the load is actually this necessary.

With The system of the invention may further comprise the duration of the substantially complete regeneration of a particulate filter Regeneration phase determines and this phase, which is an additional Fuel consumption is reduced to the necessary minimum.

Of the measured differential pressure dp at particle filters, is at constant Volume flow, inter alia, a function of the loading x of the filter. The process monitoring or control on the causative Size x too in particular has the advantage that the loading x is semistationary. Although rising the loading x, depending the volume flow and the particle concentration contained therein, in time, however, in correspondingly small time intervals This size can be considered stationary.

Problematic however, the detection of the filter load during the Engine operation. in principle Different methods are conceivable, in practice they fail However, the effort, respectively, the price, as well as the reliability. In this context, it should be noted that on a particle filter in extreme cases, temperatures between - 30 ° C and 1400 ° C can be measured.

is however, the relationship between loading x and the resulting Pressure loss dp known, can be from the simple (inexpensive) pressure measurement be closed to the loading. Although you are subject to the same Restrictions on the pressure measurement. The loading is easy as a semistationary output to rate.

Another advantage of load-based process monitoring occurs during regeneration. Based on the mass x can be predicted with knowledge of the regeneration conditions the necessary duration, which is necessary for the emptying of the filter. Also, based on the relevant input variables, the regeneration online with be calculated, whereby at any time the current load value is available. It is therefore not dependent on the problematic pressure measurement during regeneration.

The The invention will now be explained in more detail with reference to the drawings. It demonstrate:

1 Limit values and pressure readings as a function of the volume flow;

2 Types of Filtration in Particulate Filters: Depth Filtration and Surface Filtration; [Source: www.dieselnet.com]

3 the schematic course of the pressure loss dp over time (mDPF = x);

4 exemplified the arrangement of pressure ( 1 . 2 ) and temperature sensors ( 10 . 11 . 12 ) on the particle filter;

5 the projection surface in the particle filter map;

6 the typical result of a characteristic loop;

7 for example, the time course of a Kermfeldschleife;

8th the area interpolation of the data points measured during the map loop;

9 an example of a section of the dp surface at a constant temperature;

10 the pressure loss as a function of the exhaust mass, the temperature and the load (not extrapolated);

11 schematically the actual measured and the extrapolated data space;

12 schematically the system for controlling the regeneration;

13 for example the typical loading in a stationary test; and

14 the loading and regeneration under realistic conditions.

Around the load-based process monitoring to be able to operate is the knowledge of the relationship between the load mass x and the pressure loss dp essential. There is a strict distinction between loading and regeneration, since only during loading homogeneous conditions in the Are given filters that allow a description and / or calculation.

Basically, there are different types of filtration that rely on different mechanisms and are just as different in their effect on pressure loss. 2 shows the types of filtration that can both occur in particulate filters.

During the loading of particle filters, deep filtration usually occurs initially (see Phase I, 3 ), which causes the pressure loss dp to increase relatively strongly. The filtration passes with increasing loading in the surface filtration, which can further increase the pressure loss dp (see Phase II, 3 ). In 3 schematically shows the course of a load with the size dp over time t.

As already mentioned change the filter properties with the load. In addition to the quantity change, can also be the quality change the load yourself. This results, for example, from different particle size distribution and composition at various engine operating points.

Capture, storage and providing the relationship between dp and filter loading in loading phases

The operating range of an engine is usually described by its engine map. Depending on the engine speed n and the output torque Md all resulting variables can be displayed. For a particle filter or the pressure loss dp caused by the particle filter, on the other hand, the volume flow V of the medium flowing through is decisive. The volume flow V in turn is composed of the temperature T of the exhaust gas, the prevailing pressure p and the exhaust gas mass mABG. While the pressure and temperature measurement on the particulate filter must be provided separately, the size mABG is already present in modern engines in the sum of the supplied air mass mL and the current injection quantity mK. They can be taken from the engine control unit (ECU). In experiments, a separate map for the particle filter can now be found. Depending on the exhaust gas mass, the temperature and the load, the resulting pressure loss dp is recorded. With the data, one (or more) so-called artificial neural networks (KNN) can now be trained, which store these three-dimensional relationships in an appropriate manner and provide for retrieval. The problem here is the inability of neural networks for extrapolation. If, for example, the temperature presented to the network as an input variable is outside the training range, then no ne reliable output variables are generated. Thus, in practice, each engine-particulate filter combination would first have to be examined before the process could be used. Therefore, a procedure has been developed that allows the use of known filter systems on any different engines. The necessary extrapolation is made possible even before the actual network training.

In the development stage, the necessary data records were generated on the engine test bench in so-called CMS measurements. For this purpose, the engine is provided with the appropriate particulate filter in the exhaust system and mounted the sensor. Specifically, these are the in 4 shown sensors, which are not all needed in the vehicle. To determine the pressure loss before and after the particulate filter 4 pressure sensors 1 . 2 arranged. The exhaust gas temperature can be before, in or after the particulate filter 4 with temperature sensors 12 . 11 . 10 be measured. In addition, the exhaust gas mass mABG in the exhaust stream 3 certainly.

To generate a "particulate filter characteristic map" with the pressure loss dp as a dependent variable of temperature T and the exhaust gas mass mABG, the engine is operated so that as far as possible all extreme points are reached or in 5 shown projection area is maximized.

For example, maximum exhaust mass with minimum temperature (point P1 in 5 ), the motor is towed at maximum speed n. To represent point P2, maximum load is driven at maximum speed n. The exhaust gas mass mABG thus depends mainly on the speed, while the temperature naturally increases with the load, so the injection quantity.

To maximize the projection area A, a special driving cycle - referred to below as the map loop - was developed on the engine test, which is based on the generation of the training data. 6 shows a typical result of such a map loop for a loading level of 11g. The time course of such a characteristic loop can be off 7 be seen.

you recognizes that only 350 seconds to produce such Kennfeldschleife are necessary. Thus, even in the development phase, it becomes the necessary test bench time preferably kept low.

In the next step, the data points are connected using area interpolation, as shown 8th becomes apparent. To allow extrapolation, the surface is cut at constant temperatures. 9 shows an example of such a section.

The blue line in 9 shows the exact section of the surface 8th , The curve thus generated is approximated quadratically (Polyfit) and described mathematically, so that an extrapolation of the curve is easily possible: f (x) = p0 + (p1) x 1 + (p2) · x 2 with f (x) = y = dp, and x = mABG

The cuts at constant temperatures also provide a means of control: if the filter is not flown through, the pressure loss must be dp = 0. For the equation, this means that the coefficient p0 should always be equal to 0. Considering the individual coefficients of the totality of the sections for different loadings, further findings can be obtained. Applying the coefficients individually over different loading levels, a trend can be identified here, which allows the extrapolation on the load. Specifically, this means that not all load levels for the generation of the training data set actually have to be measured in the experiment. 10 shows measurements at different loading levels.

From the data set generated by the entirety of the measurements, the treatment Cutting, Polyfit, Extrapolation can thus be used to create a synthetic data record that can cover considerably more areas than the one originally measured. 11 illustrates this schematically.

Each The data points of the data space become the corresponding values assigned. The assignment and storage is done by means of Trainings of the neural network (s). Once trained, put the Networks then the desired output size in dependence different, independent inputs, for Available. For the Application Particulate filters are considered as artificial neural networks Input variables of the current exhaust gas mass flow mABG, the temperature of the exhaust gas in the filter, as well as the pressure loss dp presented. By classification, the associated load is recognized and output.

Of course, all errors in the measured quantities of the mains input also influence the output. Since, however, the size of the load x may not decrease under corresponding conditions compared to the pure pressure measurement, the output size can be evaluated. With knowledge of the particle emission of the engine, the output size can be further evaluated: increasing the load beyond the level of raw emissions can be identified as obviously wrong. additionally the output signal can be averaged or segmented, since a value update with the input frequency of 1 Hz is not necessary. Rather, it is sufficient to output the load values "by the minute", especially considering that under real conditions "golf class vehicles" can drive about 500 km until a particle filter has to be regenerated.

calculation the filter load x in regeneration phases

In Regeneration phases fail the mathematical-physical description the relationship between filter loading x and pressure loss dp. Even the empirically found connection can not be applied are, as opposed to the loading here no homogeneous conditions to rule. The process of soot oxidation, i.e. regeneration is much more of a coincidence in time and place embossed. In general, soot ignition conditions are achieved locally in the filter and thus initiating the regeneration of these so-called hot spots. In temporal As a result of regeneration, the filter can be partially completely burned free be while wide, usually cooler Areas that are still loaded. For catalytically coated particle filters become common the loading particles are first oxidized, which are in direct contact with the catalyst material. Naturally, therefore, those particles, which were deposited in the depth of the filter or near the surface.

As mentioned above the depth filtration has a strong influence on the pressure loss. Behaves accordingly This relationship in the regeneration, that is, the pressure loss initially decreases strongly upon oxidation of the deep-filtered particles.

In The rule will be a complete one Aimed for regeneration while this information about the loading condition of the filter is not absolutely necessary. Important However, it is a reliable one information about the regeneration end or the complete discharge of the filter around the expenditure of energy for the the regeneration needed will be as low as possible to keep. For the reasons mentioned above this information from the measured variable pressure loss dp is insufficient to deduce exactly.

Got to the desired complete Regeneration should be stopped prematurely (e.g., the driver stops the vehicle), the knowledge of the remaining on the filter Mass of advantage. Based on the calculated, remaining on the filter Residual quantity can be decided whether a resumption of the Regeneration after vehicle or engine start makes sense or the Filter is reloaded.

mit:

k

= Oxidationsrate

A

= reaktionsabhängige Konstante

R

= Gaskonstante

T

= Temperatur

Ea

= Aktivierungsenergie

This is where the approach of load-based process monitoring comes in. With the temperature T on the filter, the main variable influencing factor for the regeneration is known. According to the equation of Arrhenius, which generally describes chemical reactions, the reaction rate depends only on the temperature under the same boundary conditions.

With:

k

= Oxidation rate

A

= reaction-dependent constant

R

= Gas constant

T

= Temperature

Ea

= Activation energy

With in test bench trials and the so-called DTA (differential thermal analysis) method empirically determined sizes of Activation energy Ea and the reaction-dependent constant A can thus the Rußumsatz or the oxidation rate k can be determined as a temperature-dependent variable. The chemical Composition of the flowing through Medium, in this case diesel exhaust, has just as much influence but usually during an active regeneration or approximately constant over the operating time of the engine. Are there different modes in the regeneration strategy, so can on information from a corresponding regeneration map resorted become.

At the end of a loading phase, the value of the load x is transferred from the neural network to another computing tool. This first calculates the current particulate emission on the basis of the measured input variables temperature T, speed n, the current injection quantity and the air mass. This must be known, ie already measured and stored in a suitable manner. For this purpose, a neural network is also used. The same happens with the concentration of oxygen in the exhaust gas. The oxidation rate k is now calculated, which is optionally corrected for differing oxygen contents. It is thus possible both the active regeneration, as well as random partial regeneration, the z. B. from full load trips, to calculate. The actual loading loss dx at the particle filter is finally calculated as follows: Oxidation rate k [g / h] + particle emission [g / h] = loading loss dx [g / h]

The size dx is a momentary magnitude calculated for the respective current conditions. Starting from the start load of the filter at the Regeneration (= the final value of the load = x) is calculated via the load loss dx the current load:

The Loading x becomes so during the regeneration calculated as described above and not as during the Loading, calculated based on the pressure loss dp. In addition, can the loading loss dx over Exothermicity can be determined more accurately. There must be two temperature sensors for this purpose and past the particulate filter.

There the calculation of the loading x in the individual phases is fundamental is done differently, is the exchange of information about the respective values are essential. While a loading phase is constantly the load value x is output and the tool, which during the regeneration makes the calculations provided. It can be so at any time, both at random, as well as actively initiated regeneration start, his work do. If the regeneration is completed, then the Loading value x = 0 and in the subsequent loading phase first compared with the current calculation of the loading tool. Becomes repeated, after supposedly complete regeneration, a difference between the output values of the two tools are determined so subsequent regeneration phases each extended in time. Become after this measure still detects differences, so this phenomenon is non-regenerative Deposited deposits and issued as a characteristic value. The deposits can for example, from ashes, which constantly load a particle filter as well as pure soot particles. Of the Ash content is when using low-S, non-additive diesel fuels relatively low. While with the addition of additives that increase the ash content of a filter life (indicated i.d.R in vehicle mileage) of 80,000 km can go out Systems that do without additives already extend the life of an engine which is usually stated to be around 240,000 km. Due to the pressure loss caused by this non-regenerative deposit, is the regeneration strategy, but also the calculation of the total load myself, adapt. This circumstance will be in the interaction between both calculation tools.

On the Characteristic value of the non-regenerative loading, is also an evaluation variable for the filter performance to disposal. It can therefore be shown that a filter can be exchanged must because the intended thermal cleaning is no longer sufficient to the claims to meet moderate pressure loss.

In summary, the development consists of two tools (see also 12 ), one of which (Tool 1) carries out the calculation of the soot loading mass. Based on dynamic input variables, corrected with corresponding factors, the semistationary size of the load mass x is output. The tool has the option to recalculate recorded loads offline. But it can also online, ie during the current engine operation, spend the current load, which corresponds to the desired application in vehicles.

The second tool (Tool 2) calculates, as soon as mentionable oxidation processes on take place the filter, starting from the input variables temperature (tE), air mass flow (mL), fuel mass flow (mK), particle concentration (kPM), oxygen content in the exhaust gas (kO2) and Rußumsatz- or oxidation rates (k), the mass loss of soot particles in the filter.

Both Tools work in parallel, with the first only under loading conditions is used, the second works under regeneration conditions. An exchange of information takes place when of loading into regeneration or from regeneration to loading becomes.

following a concrete example: during a load becomes the size x1 for the load mass issued by Tool 1. Reach the difference between the top Limit value Xo and the size x1, the Value 0 (Xo-x1 = 0) initiates regeneration. It will be the Transfer value x1 from Tool 1 to Tool 2 and continuously calculates the conversion dm / dt of soot particles. It is calculated the necessary regeneration time tR (under current conditions) for the full Emptying of the filter, but also the instantaneous value of the load x2 depending the prevailing regeneration conditions. Thus, also with incomplete Regeneration and premature transition to re-loading the current load value x2 available, the then taken from Tool 1 and until the setting of homogeneous conditions on the filter for evaluation and costing is used. Any ashing, or not regenerable deposits are recognized and flow Formation of a characteristic xNR in further calculations. See full ad thus always come the currently deposited on the filter soot mass x1 or x2, as well as the characteristic value of the ashing xNR. Furthermore, the Values appropriately indexed to the system, but also the driver to communicate the confidence level of the instantaneous values.

Experiments to validate the virtual load sensor were performed. 13 shows a characteristic load test performed for steady state conditions (constant load and constant speed). The test shown was performed at medium speed (about 2000 rpm) and medium load (about 100 N / m). The load was interrupted three times and the filter weighed as indicated by the diamonds at 0 minutes, 120 minutes and about 235 minutes. The load increase between these points is assumed to be linear, as shown by the solid line. This was confirmed by a particle measurement. The triangles show the results of the virtual load sensor.

Under Under real circumstances, any regeneration that has started must be interrupted with carbon black remaining. For regeneration strategies is the knowledge of the laggard soot essential. The behavior of the virtual load sensor was also tested under these real conditions.

14 shows load and regeneration of a diesel particulate filter under real circumstances. The results of the virtual load sensor represented by diamonds are contrasted with the results of a gravimetric measurement of the soot mass marked by the squares and the solid line. Overall, the largest absolute deviation in the load mass is a value of only 1.1 g, which occurs after the first regeneration.

Claims (9)

System for controlling the regeneration of an im Emission control system of a diesel engine arranged particulate filter, the system comprising means for determining the start time a regeneration phase by comparing the loading of the particulate filter having a predetermined limit, the loading of the Particulate filter during the loading phase is determined by: a. Provide a Characteristic map of the particle filter with known loading; b. Work out an artificial one neural network with the provided map; c. Determine the loading of the particulate filter through the artificial neural network Measurands that the Characterize the operating condition of the engine. The system of claim 1, further comprising means for determining the duration of a substantially complete regeneration of the particulate filter by kinetic kinetic calculation of the current temporal Course of loading on the basis of empirically found factors and the loading of the particulate filter of a previous loading phase. System according to claim 1 or 2, wherein the map the pressure difference caused by the filter depending on from the exhaust gas temperature and the exhaust gas mass flowing through the filter. A system according to any one of the preceding claims, wherein the operating condition of the engine caused by the filter Pressure difference, the exhaust gas temperature and the exhaust gas mass flowing through the filter is characterized. A system according to any one of the preceding claims, wherein Measure the map through a drive cycle or a map loop becomes. The system according to claim 5, wherein the signal loop through a map loop measured characteristic map refined by interpolation or extrapolation or extended. The system of claim 1, wherein before and after Particle filter per a pressure sensor is arranged. The system of claim 1, wherein the particulate filter having a temperature sensor for measuring the temperature. The system of claim 8, wherein additionally in front of and behind the particulate filter in each case a temperature sensor for determining the exotherm during regeneration is arranged.