System to detect misfire for an internal combustion engine of a vehicle

The present invention relates to a system for detecting a misfire of a combustion engine of a motor vehicle.

In incomplete combustion, hydrocarbons, considered as pollutants and whose emissions are regulated, are emitted in excess into the atmosphere.

In order to limit these emissions, anti-pollution legislation requires the detection of misfires and the modification of engine operation control parameters when their number exceeds predefined thresholds.

In addition, when a misfire leads to the release of unburnt hydrocarbons into the catalytic converter, the deterioration thereof can accelerate.

It is known, in particular from document US Pat. No. 6,434,541, a system for determining a misfire. The system includes a neural processor executing a reconfigurable network topology comprising a plurality of hidden layers containing interconnected neurons in a recursive configuration.

The object of the invention is to propose an alternative system for detecting a misfire.

For this purpose, the subject of the invention is a system for detecting a misfire of at least one cylinder of a combustion engine of a motor vehicle, comprising:<ul id="ul0001" list-style="dash" compact="compact"><li>motor load information acquisition means,</li><li>means for acquiring crankshaft speed information,</li><li>means for calculating the average rotational speed of the engine from the crankshaft speed information,</li><li>means for determining information representative of the indicated torque of the engine cylinder,</li><li>neural means for processing the engine load information, information representative of the indicated torque, and engine rotational speed information for detecting a misfire, characterized in that:</li><li>the information determining means representative of the indicated torque are adapted to deliver n elementary information representative of the indicated torque corresponding to n successive steps of the indicated torque generated during a combustion in a cylinder, and in that</li><li>the neural means comprise a first and second neuronal stage independent of each other, the first stage receiving as input the n elementary information representative of the indicated torque and the second stage receiving as input the load information of the motor and the speed motor rotation means to optimize the information processing and the detection of a misfire.</li></ul>

Such a system is simple and makes it possible to detect a misfire with a relatively short calculation time.

According to particular embodiments, the system for detecting a misfire includes one or more of the following features:<ul id="ul0002" list-style="dash" compact="compact"><li>the information determining means representative of the indicated torque comprise means for calculating speed fluctuation information and the n elementary information comprises n speed fluctuation information, fluctuation information being calculated from the difference between a crankshaft speed over a predefined crankshaft rotation stroke and a crankshaft speed averaged over a stroke corresponding to a half revolution of crankshaft rotation;</li><li>the information determining means representative of the indicated torque comprise an accelerometer and the n elementary information comprises n acceleration information measured during the rotation of the crankshaft over a predetermined stroke;</li><li>the information determining means representative of the indicated torque comprise an engine cylinder exhaust gas pressure sensor and the n elementary information comprises n pressure variation information, a pressure variation information being calculated from the difference between an exhaust pressure over a predefined crankshaft rotation stroke and an averaged exhaust pressure over a stroke corresponding to a half revolution of crankshaft rotation;</li><li>the crankshaft speed information acquisition means comprise means for acquiring information of the angular position of the crankshaft;</li><li>the means for acquiring speed information and angular position of the crankshaft comprise a coded wheel connected to the crankshaft and a corresponding probe;</li><li>the neuronal means are composed of an input neuron layer, a single hidden neuron layer and an output neuron layer;</li><li>the first and second stages each comprise neurons of the input layer and the hidden layer;</li><li>the neuronal means are static;</li><li>the first stage further receives at input an inverse signal of the engine rotational speed, the sign of which varies as a function of the angular segment of the coded wheel used to make the measurement in order to correct the dissymmetry thereof;</li><li>the first stage further receives at input an inverse signal of the mean rotation speed of the motor squared whose sign varies as a function of the angular segment of the coded wheel used to make the measurement in order to correct the dissymmetry;</li><li>the neural means and the calculation means are located in a multifunction computer of the engine.</li></ul>

The invention will be better understood on reading the description which follows, given solely by way of example and with reference to the drawings, in which:<ul id="ul0003" list-style="dash" compact="compact"><li>Figure 1 is a schematic representation of a system for detecting a misfire of a cylinder of an engine, according to the invention;</li><li>Figure 2 is a schematic representation of a first embodiment of neural means used in the system according to the invention; and</li><li>Figure 3 is a schematic representation of a second embodiment of neural means used in the system according to the invention.</li></ul>

Figure 1 shows schematically a system 2 for detecting a misfire of a cylinder 3 of a motor vehicle engine.

This cylinder 3 comprises an exhaust valve 4 of the residual combustion gases and an intake valve 6 of air and fuel. It contains a movable piston 8, sliding by reciprocating movement, during the operating phases of the engine. This piston 8 is connected by an articulated link mechanism 10 to a crankshaft schematized by its main axis AA.

Conventionally, the four operating times of a four-stroke internal combustion engine include a phase of admission of air and fuel, a phase of air compression and ignition of the mixture for example to the using a spark plug in the case of a spark ignition engine, a phase of explosion and expansion of the burnt gases and a phase of exhaust gas residual combustion.

In a four or six cylinder engine, each cylinder 3 drives the crankshaft in rotation during its expansion phase and is driven by the other cylinders during the other phases.

During the expansion phase, the piston 8 of the cylinder 3 is able to move from a position called top dead center TDC to a position called low dead point PMB and to drive the crankshaft in rotation by half a turn around. its axis AA.

The detection system 2 according to the invention is able to detect a misfire produced during the expansion phase of a cylinder 3 of a motor by detecting a drop in the indicated torque transmitted by the piston.

The indicated torque of a cylinder of an engine is defined as the mechanical torque produced on the crankshaft by the pressure generated during combustion.

The detection system 2 according to the invention comprises a measuring sensor 14 of the engine load CH, acquisition means 12 of the crankshaft speed and a multifunction computer 16 of the engine.

The measurement sensor 14 of the load CH of the engine is located at the confluence of the intake manifolds of the cylinders 3 of the engine. It is able to measure the pressure of the air admitted into the cylinder 3.

The crankshaft speed information acquisition means 12 comprise a sensor for measuring the angular position of the crankshaft. This sensor is formed in the example described by a coded wheel 18 and a corresponding probe 20.

The coded wheel 18 is integral with the crankshaft. It can for example be fixed to the flywheel. This wheel comprises for example fifty-eight teeth arranged radially at its periphery at preset angles α of 6 degrees and a location 22 where two teeth are missing to establish a reference point with respect to the position of the top dead center TDC of each piston of cylinders.

The probe 20 is for example an inductive type probe. It is fixed close to and facing the teeth of the wheel 18 so as to detect their passages. During the rotation of the crankshaft and the coded wheel, the probe 20 is able to generate a signal transformable into slots. The duration between two edges of the generated slots is equal to the passage time of each angular segment in front of the probe.

The multifunction computer 16 of the engine comprises, in particular, calculating means 26, 28 for the average speed of rotation of the engine, determining means 26, 28 for information representative of the indicated torque of the engine cylinder and neural means 30.

The means for calculating the average rotation speed of the motor comprise an internal clock 26 and a calculation unit 28.

The internal clock 26 is able to receive the signal in the form of a slot emitted by the probe 20 and to measure the duration of each slot. During a relaxation phase, the clock is able to measure the n elementary values ​​of duration T_i each corresponding to a displacement of the crankshaft of the predefined angle α.

From these n values ​​of duration T_i, the computing unit 28 is able to calculate the average rotation speed of the RPM engine by using the following algorithm:<maths id="math0001" num=""><img file="EP1522841A1_D0001.tif" /></maths>    in which T_i is a measure of the duration in rotation seconds of the crankshaft on a stroke corresponding to an angle α of 6 °.

The average rotational speed of the RPM engine is defined as the average rotational speed of the crankshaft. It is calculated during a relaxation phase of the cylinder 3 corresponding to a half-turn stroke of the crankshaft.

The means for determining information representative of the indicated torque comprise the clock 26 and the calculation unit 28. From the n values ​​of duration T_i measured by the clock 26, the calculation unit 28 is able to calculate n elementary information of speed fluctuation F_i crankshaft. For this, the calculation unit 28 uses the following algorithms:<maths id="math0002" num=""><img file="EP1522841A1_D0002.tif" /></maths><maths id="math0003" num=""><math display="block"><mrow><mtext mathvariant="italic">Fi = Ti - Tmoy</mtext></mrow></math><img file="EP1522841A1_D0003.tif" /></maths>    wherein :<ul id="ul0004" list-style="dash" compact="compact"><li>T_i is a measure of duration in second rotation of the crankshaft on a stroke corresponding to an angle α of 6 °;</li><li>T_Avg is a measure of the duration in second rotation of the crankshaft averaged over a stroke of one half-turn of the crankshaft;</li><li>F_i is an elementary information in second of fluctuation of the speed of the crankshaft;</li></ul>

Every elementary information of speed fluctuation F_i is defined as the difference between a rotational speed of the crankshaft on a stroke corresponding to an angle α and an average speed T_Avg crankshaft averaged over a stroke corresponding to a half turn of rotation of the crankshaft.

In this embodiment of the invention, the clock 26 is able to measure thirty values ​​of duration T_i corresponding to the passage of thirty teeth of the coded wheel 18, that is to say alternately thirty teeth or twenty-eight teeth and the two missing teeth depending on the angular segment used for the measurement. The duration of passage of the missing teeth is estimated using the durations measured during the passage of the preceding and following teeth.

Alternatively, the means for determining information representative of the indicated torque comprise an accelerometer integral with the engine block. This accelerometer is capable of delivering n elementary acceleration information measured during the rotation of the crankshaft over a half-rotation stroke of the crankshaft.

Alternatively also, the information determining means representative of the indicated torque comprise an exhaust gas pressure sensor of the cylinder 3 of the engine and the calculation unit 28. The pressure sensor is attached to the outlet of the exhaust pipes and is able to generate n information on the exhaust gas exhaust gas pressure. The calculation unit 28 is able to calculate n elementary information of variation of the exhaust pressure. Each elementary variation information is calculated from the difference between an exhaust pressure on a stroke corresponding to an angle α and a pressure averaged over a stroke corresponding to a half-turn of rotation of the crankshaft.

The neural means 30 are formed by a network as illustrated in FIG. 2. This network is static and is composed of basic treatment elements called S neurons._e1, ..., S_e30. The neurons are arranged in an input layer 32, a single hidden layer 34 and an output layer 36.

Each neuron of one layer is associated with each neuron of another layer via weight w_ij. The output of a neuron is a linear combination of the neurons that are connected to that neuron and all the weights w_ij of these connections.

Each neuron further comprises an activation function such as, for example, the sigmoid function:<maths id="math0004" num=""><math display="block"><mrow><mtext mathvariant="italic">f</mtext><mtext>(</mtext><mtext mathvariant="italic">x</mtext><mtext>) = </mtext><mfrac><mrow><mtext>1</mtext></mrow><mrow><mtext>1+</mtext><msup><mrow><mtext mathvariant="italic">e</mtext></mrow><mrow><mtext mathvariant="italic">-x</mtext></mrow></msup></mrow></mfrac></mrow></math><img file="EP1522841A1_D0004.tif" /></maths>    for x belonging to real numbers.

The input layer 32 and the hidden layer 34 are composed of a first 38 and a second 40 stages.

The first stage 38 of the input layer 32 comprises thirty S neurons_e1, .., S_e30 which receive as input signals corresponding to the information representative of the indicated torque of the engine cylinder. In the case shown in FIG. 1, this information comprises n elementary information of speed fluctuation F_i crankshaft for i ranging from 1 to 30.

The first stage 38 of the hidden layer 34 comprises a neuron S_c2, and an additional entry S_c1 called bias. This input is equivalent to a constant equal to 1. The neuron S_c2 delivers an output signal of the form:<maths id="math0005" num=""><img file="EP1522841A1_D0005.tif" /></maths><ul id="ul0005" list-style="none" compact="compact"><li>where w_{2, i} represent the weights of the interconnections between the neuron S_c2 and the input neurons Fi for i ranging from 1 to 30;</li><li>W_{2, c1} represents the weight of the connection between the neuron S_c2 and S bias_c1 the hidden layer 34; and</li><li>f is the sigmoid function (described above).</li></ul>

The second stage 40 of the input layer 32 comprises two S neurons_e31, S_e32 which receive as input signals respectively corresponding to the average speed information RPM of the engine and engine load CH.

The second stage 40 of the hidden layer 34 comprises two neurons S_c4, S_c6 and two S bias_c3, S_c5. The two biases are each equal to 1. The neuron S_c4 is connected to both S neurons_e31, S_e32 of the input layer and S bias_c3. The neuron S_c6 is also connected to both S neurons_e31, S_e32 of the input layer and S bias_c5. S neurons_c4, S_c6 deliver an output signal of the form:<ul id="ul0006" list-style="none"><li><i>S</i>_<i>c</i>₄<i>= f</i>(<i>w</i>_4.31<i>RPM + w</i>_4.32<i>LOAD + w</i>_{4, C3})</li><li><i>S</i>_<i>c</i>₆<i>= f</i>(<i>w</i>_<i>6</i>_,_<i>31</i><i>RPM + W</i>_6.32<i>CHARGE +</i> W_{6, C5})</li><li>where w_4.31 represents the weight of the interconnection between the neuron S_c4 and the neuron S_e31,</li><li>W_4.32 represents the weight of the interconnection between the neuron S_c4 and the neuron S_e32,</li><li>W_{4, C3} represents the weight of the interconnection between the neuron S_c4 and S bias_c3,</li><li>W_6.31 represents the weight of the interconnection between the neuron S_c6 and the neuron S_e31 ;</li><li>W_6.32 represents the weight of the interconnection between the neuron S_c6 and the neuron S_e32 ;</li><li>W_{6, C5} represents the weight of the interconnection between the neuron S_c6 and S bias_c5 ; and</li><li>f is the sigmoid function.</li></ul>

The first 38 and the second 40 stages are independent of each other. Weights w_ij associated with the relationships between S neurons_e1, S_e2, S .._e30 of the input layer 32 of the first stage 38 and the neurons S_c3, S_c4, S_c5, S_c6 the hidden layer 34 of the second stage 40 are zero. Similarly, the weights associated with the relationships between S neurons_e31, S_e32 of the input layer 32 of the second stage 40 and the neurons S_c1, S_c2 of the hidden layer 34 of the first stage 38 are zero.

The neurons of the first 38 and second 40 stages of the hidden layer 34 are coupled to the final neuron S_f of the output layer 36. This neuron S_f is able to generate a response signal to the input signals supplied to the neurons of the input layer. It is connected to neurons S_c2, S_c4, S_c6 the first and second stages of the hidden layer 38. It delivers an output signal of the form:<ul id="ul0007" list-style="none" compact="compact"><li>S_f = <i>f</i>(<i>W</i>_<i>1</i><i>S</i>_<i>C2</i><i>+ W</i>_<i>2</i><i>S</i>_<i>C4</i><i>+ W</i>_<i>3</i><i>S</i>_<i>c6</i>)</li><li>where W₁ represents the weight of the interconnection between the neuron S_c2 and the neuron S_f ;</li><li>W₂ represents the weight of the interconnection between the neuron S_c4 and the neuron S_f;</li><li>W₃ represents the weight of the interconnection between the neuron S_c6 and the neuron S_f ; and</li><li>f is the sigmoid function.</li></ul>

The neuron S_f of the output layer 36 delivers an output signal between 0 and 1. This signal is compared in a comparator to a threshold value Vs to indicate in case of exceeding the existence of a misfire. This threshold value is for example equal to 0.5.

Weights w_ij and W_i neurons are obtained in a conventional manner during a learning phase from a load-stabilized heuristic set of the engine CH and average speed of the motor RPM. They are calculated during the learning phase by an optimization algorithm quasi Newton type or Levenberg Marquardt type. Weights w_ij and W_i are stored in a non-volatile memory ROM type not shown of the multifunction computer of the engine 16 and represent the calibration data of the algorithm.

The detection system 2 of a misfire also includes a cylinder phasing sensor not shown motor. This sensor is capable of sending to the multifunction computer of the engine 16 identification information of the cylinder in the expansion phase. The computing unit 28 is able to correlate this information with the information indicating the existence of a misfire emitted by the neural means 30 to identify the cylinder that has generated a misfire.

Alternatively, a plurality of neurons may be disposed in the first stage 38 of the hidden layer.

Alternatively also, several neurons may be arranged in the second stage 40 of the hidden layer.

Alternatively also, the output layer 36 includes a bias.

FIG. 3 represents a second exemplary structure of the neural means 30 according to the invention.

The input layer 32 of the first stage 38 comprises an additional neuron S_e0 which receives as input an inverse signal of the average speed of the 1 / RPM motor, the sign of which is variable as a function of the position of the coded wheel, that is to say essentially as a function of the angular segment used for the measurement. The neuron S_e0 makes it possible to compensate for the asymmetries generated by the machining tolerances of the teeth of coded wheel 18 and by its off-centering.

As a variant, another additional neuron S_e0 receives as input an inverse signal of the average engine speed squared 1 / RPM² whose sign is variable according to the position of the coded wheel.

As a variant, the wheel 18 of the angular position measuring sensor 12 comprises a different number of teeth and defines a different number of elementary acceleration information F_i. In this case, the first stage 38 of the input layer 32 of the neural means 30 comprises a different number of neurons. Of course, it is also possible to use a coded wheel of another type.

Alternatively, one can use a different angular segment, ie not covering a single tooth, but several, and may be for example equal to 12 ° or 18 °, etc .... The skilled person will adjust the size of the angular sector to reduce the number of neurons of the input layer in order to simplify the calculations while keeping a good accuracy of the algorithm.

Alternatively also, a computer dedicated to the function can be used in place of the multifunction computer of the engine 16.

The detection system of a misfire according to the invention can be used for engines of different types with internal combustion and in particular for gasoline or diesel engines, for four- or six-cylinder engines. However, the weights of the neurons vary from one vehicle to another and it is necessary to determine them by learning for a given engine and for a given range of vehicles.