FR2879243A1 - Exhaust gas temperature estimation system for internal combustion engine, has control unit with neural networks receiving measured data and delivering data with treated exhaust gas temperature, and module to reroute latter data - Google Patents

Abstract

The system has a control unit (6) with several neural networks (8) receiving measured data and delivering a data including a treated exhaust gas temperature (Ts). A module (11) stores and reroutes the data having a value of the temperature (Ts) into inputs of the networks for subsequent calculation. Learning data bases (12) of the networks calibrate a weight or coefficients of weight and a bias of the networks. An independent claim is also included for a method for estimating a treated exhaust gas temperature of a post-treatment exhaust gas device of an internal combustion engine of a motor vehicle.

Description

System and method for estimating the temperature of

  BACKGROUND OF THE INVENTION The invention relates to a system and a method for estimating the temperature of the outlet gases of a device. aftertreatment of the exhaust gases of an internal combustion engine of a motor vehicle.

  As the standards for emissions of gaseous pollutants by motor vehicles are becoming more stringent, exhaust aftertreatment systems are placed on the exhaust line of internal combustion engines.

  As the motor is used, the pollutants accumulate in the after-treatment devices, for example the particles accumulate in the particulate filter, and end up causing significant back-pressure to the filter. engine exhaust, which significantly decreases its performance.

  In order to restore the performance of the engine, it is known to practice a regeneration of the after-treatment devices of the engine exhaust gases.

  These devices are particularly present to reduce the emissions of particles and nitrogen oxides. These devices operate in a discontinuous manner, in cycles comprising two phases. In a first phase, the pollutants are stored, then, in a second phase, the post-treatment device is regenerated.

  The triggering of a regeneration phase of a post-treatment device depends directly on the mass of pollutant stored in the latter. It is then important to be able to correctly estimate the pollutant mass in order to be able to launch a regeneration phase of the post-treatment device if necessary. The regeneration phases take place when the engine is running, without the driver of the vehicle being aware of it.

  The estimation of the mass of pollutants present in a device for the after-treatment of the exhaust gases of an engine strongly depends on the quality of determination of the mass stored at each instant by the device. This instantaneous mass depends strongly on the temperature in these post-treatment devices. It is therefore important to know as precisely as possible the temperature of the gas output of a post-treatment device.

  There are, for the temperature at the output of an exhaust aftertreatment device, temperature sensors, or estimation devices. However such solutions are not very precise.

  Indeed, such a temperature sensor has a precision inversely proportional to its field of use. In other words, the more it is desired to measure the temperature over a wide range of temperature values, the less accurate the measurements are. In addition, over time, the sensor clogs and ages, further reducing the accuracy of its measurements. In addition, the use of such a sensor has a high cost.

  As for the estimation devices, the results are correct in steady state, but mediocre in transient state, and the input parameters necessary for these devices are often difficult to measure. In addition, these devices require too much computing time and memory for a control unit on a motor vehicle, or require a high overhead at the control unit.

  The object of the invention is therefore to be able to accurately estimate the gas outlet temperature of an aftertreatment device of the exhaust gas of a motor vehicle internal combustion engine, without additional cost of memory or operating time. calculation.

  Also, according to one aspect of the invention, there is provided a system for estimating the temperature of the exhaust gases of an aftertreatment device of the exhaust gases of a motor vehicle internal combustion engine comprising a unit of ordered. The control unit comprises one or more neural networks receiving as input estimated or measured data, and outputting data including the output gas temperature of the aftertreatment device. Data output from the neural network (s), including the output gas temperature of the post-treatment device, is redirected to the input of the neural network (s) for the following calculation.

  The use of one or more neural networks for estimating the temperature of the exhaust gases of an exhaust after-treatment device makes it possible to obtain an accurate estimate without additional cost for the control unit. . In addition, the accuracy of the estimation is greatly improved thanks to the highly dynamic aspect of the neural networks.

  In addition, the use, at the input of the neural network or networks, of data estimated during the previous calculation makes it possible to take account of the history of the system, and to obtain an accurate estimate in steady state and in steady state. transient. In addition, this substantially improves the stability of the neural network or networks.

  In an advantageous embodiment, the system further comprises means for storing and redirecting the value of the output gas temperature of the post-processing device estimated at the preceding calculation to the input of the neural network or networks for the following calculation. .

  It is then possible to take into account the second derivative of the output gas temperature of the post-treatment device with respect to time, in order to have a learning domain including the thermal acceleration of the exhaust gases, or in other words, the second derivative of the exhaust gas temperature with respect to time.

  In a preferred embodiment, the system comprises learning databases of the neural network or networks, the training data being established during prior testing of the vehicle.

  The training data makes it possible to optimize the determination of the weights associated with the inputs of the neural network or networks, in order to minimize the error on the estimate.

  In an advantageous embodiment, the system comprises a module for pre-processing the input data of the neural network or networks.

  The preprocessing module performs calculations based on known physical relationships, and makes it possible to limit the number of inputs of the neural network or networks by combining known physical parameters, as well as to filter certain outlier values of inputs.

  In a preferred embodiment, the system comprises a module for post-processing the temperature of the output gases of the post-processing device estimated by the neuron network or networks.

  The post-processing module makes it possible to perform a treatment of the temperature estimated by the neuron network or networks, for example a filtering, making it possible to eliminate outliers and a saturation avoiding a divergence of the estimate.

  In an advantageous embodiment, the system is adapted to redirect the input data of the pre-processing module from the output data of the system, comprising the output gas temperature of the post-processing device, for the following calculation.

  In a preferred embodiment, the data used at the input of the system comprises at least one of the following data: the mass of pollutant in the post-treatment device; the temperature of the gases entering the post-treatment device; the flow rate of the gases entering the post-treatment device; the pressure of the gases entering the post-treatment device; the gas pressure at the outlet of the post-treatment device; the speed of the vehicle; the atmospheric temperature.

  For example, the post-treatment device comprises a catalytic device for treating nitrogen oxides.

  For example, the post-treatment device comprises a particulate filter.

  In a preferred embodiment, the data output from the at least one neural network comprises the mass of soot present in the particulate filter.

  This value is important because it has a significant influence on the exit temperature.

  According to another aspect of the invention, there is also provided a method for estimating the temperature of the exhaust gases of an aftertreatment device for the exhaust gases of a motor vehicle internal combustion engine. One or more neural networks are used that receive, as input, estimated or measured data and output data comprising the temperature of the output gases of the post-processing device, and the data output at the output of the system is redirected to the input of the system. neural networks, including the output gas temperature of the post-processing device, for the following calculation.

  Other objects, features and advantages of the invention will become apparent on reading the following description, given solely by way of non-limiting example, and with reference to the appended drawings, in which: FIG. 1 is a block diagram of FIG. a system according to one aspect the invention; and FIG. 2 is a block diagram of a control unit comprising one or more neural networks according to one aspect of the invention.

  FIG. 1 shows an internal combustion engine 1, and its exhaust line 2, the exhaust gas outlet being symbolized by the arrow 3. In the exhaust line 2, two devices are mounted. 4 and 5 post-treatment devices. The after-treatment device devices 4 and 5 are a catalytic device 4 for treating the nitrogen oxides, and a particulate filter 5 downstream of the catalytic device 4. The particulate filter 5 is of conventional type and comprises means, for example electrostatic, for trapping soot and particles from the engine 1 and conveyed by the exhaust gas in the exhaust line 2.

  A control unit 6 operates the motor 1 and for this purpose receives a certain amount of information from a set of connections.

  The control unit 6 comprises one or more neural networks 8 receiving as input estimated or measured data, and outputting data comprising said temperature TS of the output gases of the post-processing devices 4 and 5. The basis of a neural network is the neuron, which is able to perform some elementary calculations. A neural network comprises a larger or smaller number of these units or neurons.

  A neuron receives several input data, each assigned a weighting coefficient related to the input. The neuron performs the weighted sum of the input data, and compares that sum with a threshold called bias. If the weighted sum is greater than the bias of the neuron, the neuron outputs a first value, for example 1, and else outputs a second value, for example 0.

  FIG. 2 shows the control unit 6. The neuron network (s) 8 receive data input by the set of connections. These connections come from the engine 1 and the exhaust line 2, which includes sensors and / or estimating devices capable of providing these data. For an aftertreatment device 4 or 5, the input data of the at least one neural network 8 comprises at least one of the following parameters: the temperature of the gases at the input of the post-processing device 4 or 5, the flow rate of the inlet gas of the after-treatment device 4 and 5, the inlet gas pressure of the after-treatment device 4 and 5, the outlet gas pressure of the after-treatment device, the vehicle speed, or the temperature atmospheric.

  The embodiment illustrated in FIG. 2 further comprises a preprocessing module 9 which performs a preprocessing of at least a portion of the input data of the neural network (s) 8.

  The preprocessing module 9 makes it possible to filter the value of certain entries, in case they take outliers. The preprocessing module 9 can also make it possible to perform calculations based on known physical relations, to combine several parameters provided by the set of connections, in order to reduce the number of input parameters of the connection network (s). neurons 8.

  The set of connections 7, or the set of data respectively transmitted at the input of the system by the connections of the set of connections 7, comprises at most three subsets of respective connections or data transmitted. A first subset 7a is formed of data transmitted to the preprocessing module 9. A second subset 7b is formed of data transmitted to the preprocessing module 9 and to the neuron network 8. Finally, a third subset 7c together is formed of data transmitted to the neuron network (s) 8.

  This embodiment also comprises a post-processing module 10 of the temperature Ts of the output gases of the post-processing device 4 or 5 estimated by the neuron network or networks 8. The post-processing module 10 makes it possible to process the value estimated by the neural network or networks 8, for example by performing a filtering for eliminating outliers and in such a case not to reinject them at the input of the neural network or networks 8.

  The system also comprises a module 11 for storing and redirecting the input of the neural network or networks 8 to the value of the temperature Ts of the output gases of the post-processing device 4, 5, estimated at the preceding calculation, for the following calculation. The module 11 makes it possible to store an estimated value of the temperature Ts of the output gases of the post-processing device 4, 5, during a computation step, and then to transmit it to the input of the neuron network or networks 8. other words, the estimated value of the temperature Ts of the output gases of the post-processing device 4, 5, by a computation step of order n is stored during the following computation step of order n + 1, and then is redirected to the input of the neuron network (s) 8 at the next computation step of order n + 2.

  It is then possible to take into account the derivative of order one and of the second order derivative with respect to the time of the temperature Ts of the exit gases of the post-treatment device 4, 5. values of these derivatives in the learning domain of the neural network or networks 8, which makes it possible to have a very wide learning domain, in which the neural network or networks 8 are very stable.

  The neuron network or networks 8 can also estimate other output data, and reinject them at the input of the neural network or networks.

  For example, when the post-processing device (4, 5) comprises a particulate filter (5), another data estimated at the output of the neuron network (s) 8 is the mass of soot may be the mass of soot present in In fact, the presence of soot in the particulate filter has a very strong influence on the temperature of TS of the outlet gases of the post-treatment device 4, 5.

  The system shown in FIG. 2 also comprises learning databases 12 of the neural network or networks 8. The learning databases 12 of the neural network or networks 8 serve to calibrate the weights or weighting coefficients, as well as the biases of the neural network (s) 8.

  The learning databases 12 make it possible to optimize the calibration of the weighting coefficients and biases of the neuron network (s) 8, in order to optimize the estimation.

  Indeed, learning is the penultimate phase of developing a network of neurons. It consists first of all in calculating the optimal weightings of the various links of the network, or inputs of the neurons, by using the samples contained in the learning databases. In general, the input data corresponding to these data are entered. samples, and the different weights are corrected according to the error obtained at the output. This step is repeated until we have the lowest error at the output of the neural network.

  Generally, the final validation tests of the neuron network (s) 8 are carried out from test databases 13 different from the learning data bases 12. Indeed, in order to have optimized tests, it is necessary to different tests of the data used to learn the neural network (s) 8.

  The quality of the learning databases 12 and tests 13 has an effect on the accuracy and robustness of the system. The learning database 12 must cover as completely as possible the variation domain of the input group and its derivatives in order to best describe the function to be estimated by the system. It is also important to have low noise learning data to prevent the network from integrating measurement noise during its learning.

  The invention makes it possible to estimate accurately and can be implemented in a conventional control unit of a motor vehicle, the temperature at the outlet of a device for aftertreatment of the exhaust gases of a vehicle internal combustion engine automobile.

Claims (1)

11 CLAIMS   1. System for estimating the temperature (Ta) of the exhaust gases of an aftertreatment device (4, 5) of the exhaust gases of an internal combustion engine (1) of a motor vehicle comprising a unit control unit (6), characterized in that the control unit (6) comprises one or more neural networks (8) receiving as input estimated or measured data, and outputting data comprising said temperature (Te), and in that data output from the at least one neural network (8), including said temperature (Te), is redirected to the input of the system for the next calculation.   2. System according to claim 1, characterized in that the system further comprises means (11) for storing and redirecting the input of the neuron network (s) (8) to the value of said temperature (Te) of the output gas of the device. of the post-processing (4, 5) estimated at the previous calculation, for the following calculation.   3. System according to claim 1 or 2, characterized in that it comprises learning databases (12) of the neural network or networks (8), the learning data (12) being established when preliminary tests of the vehicle.   4. System according to any one of claims 1 to 3, characterized in that it comprises a post-processing module (10) of the temperature (Ta) of the output gases of the post-treatment device (4, 5). ) estimated by the neural network or networks (8).   5. System according to any one of claims 1 to 4, characterized in that it comprises a preprocessing module (9) of input data of the neural network or networks (8).   6. System according to claim 5, characterized in that the system is adapted to redirect input of the preprocessing module (9) data outputted from the system, comprising said temperature (TS) for the following calculation.   7. System according to any one of claims 1 to 6, characterized in that the data used at the input of the system comprise at least one of the following data: the mass of pollutant in the post-treatment device 5 (4, 5); the temperature of the gases entering the post-treatment device (4, 5); the gas flow rate at the inlet of the post-treatment device (4, 5); the gas pressure at the inlet of the post-treatment device (4, 5); the gas pressure at the outlet of the post-treatment device (4, 5); the speed of the vehicle; the atmospheric temperature.   8. System according to any one of claims 1 to 7, characterized in that the after-treatment device (4, 5) comprises a catalytic device (4) for treating nitrogen oxides.   9. System according to any one of claims 1 to 8, characterized in that the after-treatment device (4, 5) comprises a particulate filter (5).   10. System according to claim 9, characterized in that the data output from the neuron network (s) (8) comprises the mass of soot present in the particulate filter (5).   11. A method for estimating the temperature (Ts) of the exhaust gases of an aftertreatment device (4, 5) of the exhaust gas of an internal combustion engine (1) of a motor vehicle, characterized in that using one or more neural networks (8) receiving as input data estimated or measured and outputting data comprising said temperature (TS), and in that data is input to the system input delivered at the output of the neuron network or networks (8), comprising said temperature (Ta), for the following calculation.