Description:FIELD OF THE INVENTION
[001] The present invention relates to a system and a method for estimating air-fuel ratio of an internal combustion engine.

BACKGROUND OF THE INVENTION
[002] It is a known fact that, Air to Fuel Ratio (AFR) of an engine is a key index that impacts fuel economy and tailpipe emissions of the engine. Also, emission characteristics are extremely sensitive to AFR in engines, as three-way catalytic converters are effective only when a stoichiometric AFR is maintained. Therefore, proper control of AFR is critical over operating range of the engine.
[003] Most of the current production vehicles are installed with switched-type lambda sensor (otherwise known as ‘non-linear lambda sensors’) for measuring the AFR. However, the switched-type lambda sensor has certain limitations in performance and operating time, consequently limiting the AFR control in each cycle of operation of the engine.
[004] The current production vehicles may also employ a narrowband lambda sensor, which works on the principle of rich and lean mixture of the fuel that is inlet into the engine. A mixture which has less amount of air compared to the stoichiometric ratio is known as a rich mixture, and the mixture which contains more amount of air compared to the stoichiometric ratio is known as a lean mixture. When the mixture is in rich condition, the voltage output from the narrowband lambda sensor is around 0.8 to 0.9 volts, and when the air/fuel mixture is lean, voltage drops to 0.3 volts or less from the narrowband lambda sensor. Thus, the narrowband lambda sensor only determines whether the AFR is above or below the stoichiometric value. However, as the output from the narrowband lambda sensor is effectively one of the two levels of rich or lean of stoichiometry, the system’s performance is compromised by the use of the narrowband lambda sensor.
[005] To overcome the aforementioned limitations, wideband lambda sensors (otherwise known as ‘linear lambda sensors’) may be employed in the system. The wideband lambda sensor are typically more accurate, faster and provides a precise measurement of the actual exhaust AFR. However, linear lambda sensors are typically expensive, thereby restraining use in mass produced vehicles.
[006] Also, the AFR feedback control systems with the physical oxygen sensors are associated with time delays. Such time shift or delay may occur between injection timing and oxygen sensor measurement, which account for phasing of intake and exhaust valves of the engine. Such time delay could represent a significant problem for engine control applications. Further, such time delay may occur during a rich or a lean fuel injection in the engine, which may also affect the oxygen measurement by the lambda sensor. To overcome the aforementioned limitations pertaining to time delay, the sensors, such as the lambda sensors, can be installed close to exhaust ports of the engine. However, such an installation may expose the sensors to an exhaust temperature that significantly reduce longevity of the sensors, which is undesirable. Additionally, the sensors behave as a first order lag filter and thus, the signals generated from the sensors are filtered, which may lead to the time delay. Furthermore, during cold start-up condition of the engine, the lambda sensor is incapable of measuring the AFR, as the lambda sensor typically operate as an open loop. As such, a significant proportion of hydrocarbon emissions occur due to lack of sensing for feedback control during this period. To overcome the aforementioned limitations due to use of physical lambda sensor, mathematical methods of estimating the AFR are employed. However, these mathematical methods involve complex real-time computations which is cumbersome for engine control unit to process.
[007] In view of the above, there is a need for a system and a method for estimating air-fuel ratio of the engine, which addresses one or more limitations stated above.

SUMMARY OF THE INVENTION
[008] In one aspect, a system for estimating an air-fuel ratio (AFR) of an engine in a dynamic condition of the engine is disclosed. The system comprises an ion current measurement device in communication with an ignition coil. The ion current measurement device is configured to measure an ion current signal generated on a spark event in a spark plug of the engine. One or more sensors are coupled to the engine and are adapted to generate one or more engine parameters. A control unit is communicatively coupled to the ion current measurement device and the one or more sensors. The control unit is configured to receive the ion current signal from the ion current measurement device and the one or more engine parameters from the one or more sensors and estimate the AFR of the engine, the estimation being performed based on a trained machine learning model.
[009] In an embodiment, a data logging device in communication with the ion current measurement device, the one or more sensors and the control unit is provided. The data logging device is configured to receive the ion current signal from the ion current measurement device and the one or more engine parameters from the one or more sensors, and provide the ion current signal and the one or more engine parameters to the control unit.
[010] In an embodiment, the control unit is configured for training a machine learning model to obtain the trained machine learning model, and the control unit being communicably coupled to an analyzer unit. The analyzer unit is adapted to compare the estimated AFR with a reference AFR computed by a wideband lambda sensor.
[011] In an embodiment, the trained machine learning model is a non-linear autoregressive neural network.
[012] In an embodiment, the one or more engine parameters comprises a throttle position of a throttle body of the engine, a fuel pulse width of a fuel supply device; an instantaneous engine speed of the engine a manifold air pressure in the engine, an engine temperature and an output from a narrowband lambda sensor, the output indicating one of the AFR being a rich mixture or a lean mixture.
[013] In an embodiment, the control unit is configured to preprocess the ion current signal before sending the ion current signal to the trained machine learning model. The pre-processing involves interpolation of the ion current signal for identifying region of interest to be sampled in an analog signal, wherein the control unit is configured to perform a principal component analysis on the region of interest determined during the pre-processing.
[014] In an embodiment, the system comprises a calibration unit in communication with the at least one of the control unit and a data logging device. The calibration unit is configured to calibrate the one or more engine parameters and the ion current signal transmitted to the at least one of the control unit and the data logging device, as per required sampling rate or frequency.
[015] In an embodiment, the system comprises a regulating unit in communication with the one or more sensors. The regulating unit is configured to sample at least one of the engine parameters at a different frequency to overcome signal trimming of the at least one of the engine parameters.
[016] In an embodiment, the system comprise a processing unit in communication with a regulating unit and a data logging device. The processing unit is configured to compute and generate a fuel pulse width signal from state of a fuel supply device of the engine.
[017] In another aspect, a method for estimating the AFR for the engine. The method comprises generating, by the ion current measurement device, the ion current signal, the ion current signal being generated based on measurement of an ionization current supplied to the spark plug of an engine. The control unit 108 thereafter receives the ion current signal from the ion current measurement device and the one or more engine parameters from the one or more sensors. The control unit then estimates the AFR of the engine based on the ion current signal and the one or more engine parameters, the estimation being performed based on a trained machine learning module.

BRIEF DESCRIPTION OF THE DRAWINGS
[018] Reference will be made to embodiments of the invention, examples of which may be illustrated in accompanying figures. These figures are intended to be illustrative, not limiting. Although the invention is generally described in context of these embodiments, it should be understood that it is not intended to limit the scope of the invention to these particular embodiments.
Figure 1 is a block diagram of a system for estimating Air-Fuel Ratio (AFR) of an engine, in accordance with an exemplary embodiment of the present disclosure.
Figure 2 is a block diagram of a control unit of the system for estimating the AFR, in accordance with an exemplary embodiment of the present disclosure.
Figure 3 is a block diagram of a trained machine learning model of the control unit, in accordance with an exemplary embodiment of the present disclosure.
Figure 4a is a graphical representation of an ion current signal with respect to time determined by an ion current measurement device, in accordance with an exemplary embodiment of the present disclosure.
Figure 4b is a graphical representation of the ion current signal with respect to time determined by the control unit, in accordance with an exemplary embodiment of the present disclosure.
Figure 5 is a graphical representation depicting comparison of AFR ratio with respect to time obtained during a training phase of the machine learning model and during real time estimation by the trained machine learning model of the control unit, in accordance with an exemplary embodiment of the present disclosure.
Figure 6 is a flowchart of a method of estimating the AFR of the engine by the system, in accordance with an exemplary embodiment of the present invention.

DETAILED DESCRIPTION OF THE INVENTION
[019] Various features and embodiments of the present invention here will be discernible from the following further description thereof, set out hereunder. In the ensuing exemplary embodiments, the vehicle can be a multi-wheeled vehicle.
[020] Figure 1 illustrates a block diagram of a system 100 for estimating Air-Fuel Ratio (AFR) of an engine 102, in accordance with an exemplary embodiment of the present disclosure. The system 100 is adapted to estimate AFR of the engine 102 during a dynamic condition or running condition of the engine, in real-time based on a trained machine learning model. Thus, the system 100 is configured to eliminate the requirement of an expensive wideband lambda sensor (not shown) for determining the AFR of the engine 102. In the present embodiment, the engine 102 is a spark-ignited, fuel injection based internal combustion engine of a two-wheeled vehicle, or a three-wheeled vehicle or a multi wheeled vehicle (not shown).
[021] The system 100 comprises an ion current measurement device 104 in communication with an ignition coil (not shown) of the engine 102. The ion current measurement device 104 is adapted to measure an ion current signal generated due to flow of a spark current in the ignition coil. In an embodiment, the ion current measurement device 104 is an ion current circuit (not shown) that is connected to a secondary side of the ignition coil. In an embodiment, the ion current measurement device 104 comprises a capacitor (not shown) that is connected to the secondary side of the ignition coil. As such, the capacitor gets charged when a voltage is supplied to a spark plug (not shown) by the ignition coil during a spark event. The moment the spark event is completed, the capacitor provides a voltage to the spark plug through a spark electrode of the engine 102 for ionizing the air particles. The supply of voltage to the spark electrode causes flow of the ion current, which is indicated by the ion current signal. Thus, the ion current measurement device 102 is configured to determine the ion current signal based on the voltage supplied to the spark plug.
[022] The system 100 further comprises one or more sensors 112 coupled to the engine 102. The one or more sensors 112 are adapted to generate one or more engine parameters of the engine 102. In an embodiment, the one or more engine parameters are indicative of current operating conditions of the engine 102. The system 100 is adapted to monitor the one or more engine parameters for estimation of the AFR during real-time conditions or operating conditions of the engine 102.
[023] In an embodiment, the one or more engine parameters may be the operating parameters of the engine 102, such as a throttle position of a throttle body of the engine 102, a fuel pulse width of a fuel supply device (not shown) of the engine 102, an instantaneous speed of the engine 102, a manifold air pressure in the engine 102, and a fuel intake temperature.
[024] In an embodiment, a narrowband lambda sensor 112e also may be installed in an exhaust (not shown) of the engine 102. The narrowband lambda sensor generates a signal indicative of whether the AFR entering the engine 102 is lean or rich. This indicative signal may also constitute as the one or more engine parameters of the engine 102. In an embodiment, spark voltage may also be used as another parameter along with the 5 inputs as shown in a neural network depicted in Figure 3.
[025] In the present embodiment, the one or more sensors 112 comprise a throttle position sensor 112a, an engine speed sensor 112b, a temperature manifold absolute pressure (TMAP) sensor 112c and an engine temperature sensor 112d. In an embodiment, the one or more sensors 112 may also comprise of the narrowband lambda sensor 112e. The narrowband lambda sensor 112e is configured to monitor whether the AFR of the engine 102 is rich or lean. Based on the AFR, the narrowband lambda sensor 112e is configured to generate a signal indicative of whether the AFR is rich or lean.
[026] In an embodiment, the throttle position sensor 112a is located in the throttle body coupled to the engine 102. The throttle position sensor 112a is configured to monitor a degree of opening of the throttle body. Based on the degree of opening of the throttle body, the throttle position sensor 112a is configured to generate a throttle position signal.
[027] In an embodiment, the engine speed sensor 112b is a Rotation Per Minute (RPM) sensor, mounted adjacent to a crankshaft (not shown) of the engine 102. The engine speed sensor 112b is configured to monitor the speed of rotation of the crankshaft. Based on the speed of rotation of the crankshaft, the engine speed sensor 112b is configured to generate an engine speed signal.
[028] In an embodiment, the TMAP sensor 112c is located in an air intake pipe (not shown) of the engine 102. The TMAP sensor 112b is configured to monitor manifold air pressure in the engine 102. Accordingly, based on the determined air pressure, the TMAP sensor 112c generates a manifold air pressure signal and fuel intake temperature signal.
[029] In an embodiment, the engine temperature sensor 112d is mounted in an engine cylinder (not shown). The engine temperature sensor 112d is configured to monitor temperature of the engine 102. Based on the engine temperature, the engine temperature sensor 112d is configured to generate an engine temperature signal. For the determination of the fuel pulse width/ fuel injection time, the fuel injection ON instant and the fuel Injection OFF instant are communicated to a processing unit (118) through a regulating unit (116) of the system. In an embodiment, the fuel pulse width signal may be available as an engine parameter from a fuel supply device of the engine (102). The fuel pulse width signal usually ranges in microseconds.
[030] Referring to Figure 2 in conjunction with Figure 1, the system 100 comprises a control unit 108 communicably coupled to the one or more sensors 112 and to the ion current measurement device 104. In an embodiment, the control unit 108 is communicably coupled to the one or more sensors 112 through a wired connection or a wireless connection as per design feasibility and requirement. The control unit 108 is adapted to estimate the AFR of the engine 102, based on inputs signals received from the one or more sensors 112 and the ion current signal provided by the ion current measurement device 104. As such, the control unit 108 is adapted to estimate the AFR of the engine 102, based on the one or more engine parameters and the ion current signal.
[031] The control unit 108 is configured to receive input signal from each of the one or more sensors 112 along with the ion current signal. In other words, the control unit 108 is adapted to receive the throttle position signal from the throttle position sensor 112a, a fuel pulse width signal from the fuel supply device, the engine speed signal from the engine speed sensor 112b, the manifold air pressure signal from the TMAP sensor 112c, the temperature signal from the engine temperature sensor 112d along with the ion current signal for estimating the AFR. In an embodiment, the control unit 108 is also configured to receive input signal from the narrowband lambda sensor 112e. The control unit 108 is also configured to determine operating conditions or real-time conditions of the engine 102 based on the input signals received from the sensors 112.
[032] In an embodiment, the control unit 108 can be in communication with at least one vehicle control unit (not shown) of the vehicle 200. In an embodiment, the control unit 108 may comprise one or more additional components such as, but not limited to, an input/output module 126, a processing module 120, and an analytic module 122.
[033] The control unit 108 is in communication with the components such as the processing module 120 and the analytic module 122. In an embodiment, the processing module 120 and the analytic module 122 are configured within the control unit 108. In another embodiment, the control unit 108 may be embodied as a multi-core processor, a single core processor, or a combination of one or more multi-core processors and one or more single core processors. For example, the control unit 108 is embodied as one or more of various processing devices or modules, such as a coprocessor, a microprocessor, a controller, a digital signal processor (DSP), a processing circuitry with or without an accompanying DSP, or various other processing devices including integrated circuits such as, for example, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a microcontroller unit (MCU), a hardware accelerator, a special-purpose computer chip, or the like. In yet another embodiment, the control unit 108 may be configured to execute hard-coded functionality. In still another embodiment, the control unit 108 may be embodied as an executor of instructions, where the instructions are specifically configured to the control unit 108 to perform steps or operations described herein for estimating the AFR.
[034] Further, the control unit 108 is communicably coupled to a memory unit 128. The memory unit 128 is capable of storing information processed by the control unit 108 and also the data received from each of the sensors 112 and the ion current signal provided by the ion current measurement device 104. The memory unit 128 is embodied as one or more volatile memory devices, one or more non-volatile memory devices and/or combination thereof, such as magnetic storage devices, optical-magnetic storage devices and the like as per design feasibility and requirement. The memory unit 128 communicates with the control unit 108 via suitable interfaces such as Advanced Technology Attachment (ATA) adapter, a Serial ATA [SATA] adapter, a Small Computer System Interface [SCSI] adapter, a network adapter or any other component enabling communication between the memory unit 128 and the control unit 108. In an embodiment, the control unit 108 may be connected to a power supply such as a battery module (not shown) of the vehicle, for receiving electrical power. In an embodiment, the control unit 108 may have an inbuilt power supply 124 for drawing power from the battery module of the vehicle.
[035] In an embodiment, the control unit 108 or the analytic module 122 of the control unit 108 is adapted to estimate the AFR of the engine 102, based on the ion current signal and the one or more engine parameters received from the one or more sensors 112. The control unit 108 or the analytic module 122 of the control unit 108 estimates the AFR of the engine 102 based on a trained machine learning model, which may be stored in the memory unit 128 or the analytic module 122.
[036] In an embodiment, the trained machine learning model may be a recurrent type dynamic neural network (as shown in Figure 3), the non-linear autoregressive network with exogeneous inputs (NARX) which helps in prediction of non-linear time series, for estimating the output, i.e. the AFR. The trained machine learning model is configured to predict a time series when previous values of the time series are provided as a feedback input along with an external or exogeneous time series. In other words, the trained machine learning model is a virtual wideband lambda sensor model using one or more engine parameters and the ion current signal, and provides prediction or estimation of AFR to subsequently control the emissions of the engine 102. The output from the trained machine learning model is dependent on both current input to the neural network and also the previous input and output of the network. In an embodiment, the trained machine learning model may be a time series predicting data driven model such as genetic algorithm.
[037] In the present embodiment, the control unit 108 or the analytic module 122 of the control unit 108 is configured to preprocess the ion current signal before sending the ion current signal to the trained machine learning model. The preprocessing involves interpolation or extrapolation of the ion current signal per engine cycle for identifying region of interest to be sampled in an analog ion current signal. The interpolation or extrapolation of the ion current signal ensures to have uniform number of samples per engine cycle. The interpolation or extrapolation of the ion current signal values are shown in Figures 4a and 4b. Upon interpolation or extrapolation, the region of interest determined from the curves is processed via a Principal Component Analysis (PCA). This processing step reduces the number of inputs that are provided to the neural network, thereby reducing the number of processing steps involved for obtaining output from the network. It can be seen in Figure 4b that the region of interest is identified as time 0-t7 secs as the ion current signal shows variation in this region.
[038] In the present embodiment, the PCA for the determined region of interest is performed by firstly equating the ion current values with a product of the number of engine cycles and a 34 row matrix of the ion current. Thereafter, a covariance matrix and eigen decomposition of the covariance matrix is carried out, and the entire ion current values are projected onto the eigen values, resulting in a PCA score as mentioned in below eq. 1. In an embodiment, 34 matrix is considered, since 34 samples are obtained in an engine drive cycle within a predefined sampling time, which can be 100 ms. As such, the number of rows of the matrix of the ion current is considered based on the number of samples obtained for a given drive cycle and the sampling time.
PCA score = Number of cycles x 34 ….. (eq. 1)
[039] Further, the first five principal components of the ion current contribute to 95% of variance, which results in PCA score of the ion current as mentioned in below eq. 2. The determination of PCA score narrows or reduces the number of inputs to be processed by the neural network, for determining the AFR. In the present embodiment, the number of inputs to be processed by the neural network has been reduced to 5. Thus, the PCA score is multiplied by 5 in the below eq. 2.
PCA score of ion current signal = Number of cycles x 5 …. (eq. 2)
[040] In an embodiment, the number of inputs to be processed by the neural network is considered when the input signals contribute to the variance. In the above example, 5 signals that contribute to the variance out of the 34 input signals are considered for processing by the neural network. The five inputs being the one or more engine parameters along with the ion current signal. In other words, the 5 inputs processed by the neural network are the degree of opening of the throttle body, speed of rotation of the crankshaft (i.e. the engine speed), manifold air pressure in the engine 102, the engine temperature and the ion current signal. In an embodiment, the neural network may also receive and process a fuel-ignition timing (as shown in Figure 3) of the engine 102 for processing. In an embodiment, another parameter- narrow band lambda sensor output signal may also be used with these 5 inputs. Use of an additional parameter only improves the prediction accuracy of the machine learning model, provided the machine learning model is trained on the 5 inputs and the additional parameter.
[041] Subsequently, the inputs narrowed down through the PCA are transmitted to the neural network in real-time. That is, the engine parameters are input along with the narrowed ion current signal to the neural network (as shown in Figure 3) for processing. In the present embodiment, the neural network is trained prior via scaled conjugate gradient technique to obtain the trained machine learning model. In another embodiment, the neural network also comprises three hidden layers, with a predetermined number of neurons in each hidden layer. In an embodiment, the number of neurons in a first hidden layer is 23, the number of neurons in a second hidden layer is 21 and the number of neurons in a third hidden layer is 27. Further, an output layer of the neural network comprises one neuron. The network model is trained upto the present known values in a closed-loop mode as per accuracy requirement for predicting the output values i.e. the AFR. The output obtained from the neural network are the AFR values of the current drive cycle or dynamic condition of the engine 102. The output signal generated by the neural network is input to an engine controller/ engine control unit (ECU) (not shown) that adjusts injection timing of the engine 102 and thus the fuel injected during each cycle reaches the optimum stoichiometric condition or ratio.
[042] Further, the trained machine learning model is obtained on training and testing a machine learning model, that is the neural network on a large set of training data and testing data. On successful training of the machine learning model on the combination of inputs- engine parameters and the ion current signal, and/or narrow band lambda sensor, it is found that accuracy of prediction on the training data set was about 97% while accuracy of prediction on the testing data set is about 88% with a low mean square error.
[043] In another embodiment, for training the machine learning model in the control unit 108, the system 100 comprises an analyzer unit 110 (as shown in Figure 1). The analyzer unit 110 is incorporated in the system 100 during the training phase of the neural network. The analyzer unit 110 is adapted to determine a reference AFR value through a wideband lambda sensor (not shown) that is provided at the exhaust (not shown) of the engine 102. The analyzer unit 110 is also in communication with the control unit 108. The analyzer unit 110 is adapted to receive the AFR determined by the machine learning model in the control unit 108, and compare the determined AFR with the reference AFR generated by the wideband lambda sensor. When the variation of the determined AFR is greater than a preset value, the analyzer unit 110 is adapted to alert the control unit 108, for re-training of the machine learning model. As such, the analyzer unit 110 is adapted to ensure that the engine 102 operated through the determined AFR complies with the emission regulations set by local authorities.
[044] In an embodiment, the system 100 comprises a data logging device 106 (as shown in Figures 1 and 2) that is in communication with the one or more sensors 112, the ion current measurement device 104 and the control unit 108. The data logging device 106 may be adapted to receive the ion current signal from the ion current measurement device 104 and the one or more engine parameters from the one or more sensors 112. The data logging device 106 is configured to transmit the ion current signal and the one or more engine parameters to the control unit 108 for estimating the AFR.
[045] In an embodiment, the system 100 comprises a calibration unit 114 in communication with the control unit 108 and/or the data logging device 106. The calibration unit 114 is configured to calibrate the one or more engine parameters and the ion current signal transmitted to the control unit 108 and/or the data logging device 106 as per required sampling rate or frequency. In an embodiment, the calibration unit 114 is configured to calibrate the one or more engine parameters and the ion current signal transmitted to the control unit 108 and/or the data logging device 106 at a sampling frequency of 1kHZ.
[046] In an embodiment, at the sampling frequency of 1kHZ, there are instances where number of ion current samples may be low or high depending on operating cycle of the engine 102. As such, a variance persists due to dynamic nature of operation of the engine 102. However, the neural network requires constant number of samples on which it is trained and when it is online. Thus, PCA is performed on ion current signal samples to ensure that the neural network is trained for 5 samples or inputs of ion current signal along with the one or more engine parameters. Therefore, 5 samples are provided to the neural network in real-time for estimating the AFR.
[047] In an embodiment, the system 100 comprises a regulating unit 116 (as shown in Figure 1) and a processing unit 118 (as shown in Figure 1) that are in communication with the one or more sensors 112. The processing unit 118 determine the fuel pulse width from the fuel injection ON instant and fuel injection OFF instant received from a fuel supply device of the engine 102. Since the fuel pulse width signal ranges in microseconds while the sampling frequency of the calibration unit 114 and the control unit 108 are in milliseconds, the regulating unit 116 is adapted to overcome signal trimming of the fuel pulse width signal/ fuel injection time signal by sampling the fuel injection ON instant and fuel injection OFF instant at a different frequency to overcome signal trimming by a regulating unit 116. In an embodiment, the regulating unit 116 may also be used to prevent trimming of the other engine parameters generated by each of the one or more sensors 112, in case of mismatch in the sampling frequencies and rate of change of the engine parameters itself. In the present embodiment, the regulating unit 116 and the processing unit 118 are provided in order to obtain an accurate fuel pulse width signal.
[048] In an embodiment, the trained machine learning model may employ techniques such as Genetic algorithm technique, Convolution neural network technique and the like, as per requirement.
[049] In an embodiment, for extracting relevant data from the inputs may be carried out by statistical methods and machine learning techniques known in the art. In the present embodiment, the statistical method employed for extracting and narrowing down the data is the PCA.
[050] At the testing phase of the machine learning model, prior to obtaining the trained machine learning model, the analyzer unit 110 is adapted to receive the AFR determined by the machine learning model in the control unit 108, and compare the determined AFR with the reference AFR generated by the wideband lambda sensor as shown in Figure 5. Referring to Figure 5, a graphical representation of AFR values determined by the control unit 108 (referenced as ‘132’ in Figure 5) with respect to the reference AFR value (referenced as ‘130’ in Figure 5) computed by the analyzer unit 110 is depicted. As depicted, the AFR values determined by the control unit 108 are overlapping with the reference AFR value from the wideband lambda sensor. Thus, the determined AFR by the control unit 108 matches with the reference AFR. Therefore, the system 100 mitigates the need for a separate wideband lambda sensor or any other lambda sensor for determining the AFR of the engine 102. Thus, successfully the trained machine learning model is obtained and is deployed in real-time in the control unit 108 for determined AFR in the dynamic condition of the vehicle using the one or more engine parameters and the ion current signal.
[051] Figure 6 illustrates a flow diagram of a method 600 for estimating the AFR of the engine 102.
[052] At step 602, the ion current measurement device 104, based on the voltage supplied to the spark plug of the engine 102, generates the ion current signal. The ion current signal is thereafter transmitted to the control unit 108 from the ion current measurement device 104 at step 604. In an embodiment, the ion current measurement device 104 transmits the ion current signal to the data logging device 106. Accordingly, the data logging device 106 transmits the ion current signal to the control unit 108.
[053] At step 606, the control unit 108 then receives the one or more engine parameters from the one or more sensors 112. In an embodiment, the one or more sensors 112 may transmit the one or more engine parameters to the data logging device 106. Accordingly, the control unit 108 may receive the one or more engine parameters from the data logging device 106. In an embodiment, the ion current signal and the one or more engine parameters are processed through the calibration unit 114 for calibration, before sending to the control unit 108. At this scenario, the method 600 moves to step 608. In an embodiment, the ion current signal is preprocessed before sending to the trained machine learning model in the control unit 108.
[054] At step 608, the control unit 108 processes the ion current signal and the one or more engine parameters via the PCA for narrowing the input signals to be transmitted to the neural network. The neural network upon receiving the input signal, estimates the AFR of the engine 102.
[055] The claimed invention as disclosed above is not routine, conventional or well understood in the art, as the claimed aspects enable the following solutions to the existing problems in conventional technologies. Specifically, the claimed aspect provides the system 100 having the control unit 108 for estimating the AFR based on the ion current signal and the one or more engine parameters. Thus, mitigating the need for mounting of a wideband lambda sensor in the engine 102 or in the vehicle. Consequently, the cost associated with the installation and maintenance of the wideband lambda sensor is eliminated. Also, the limitations arising due to mounting of wideband lambda sensor in the engine 102 are eliminated in the system 100. Moreover, due to use of the trained machine learning model the system 100 is capable of capturing the non-linear dynamic behavior of the engine 102 with acceptable level of accuracy. Additionally, the system 100 ensures that the determined AFR operates the engine 102 within the emission norms set by local authorities, while also enhancing thermal efficiency and fuel economy of the engine 102. Further, the trained machine learning model of the present invention is non-recursive and parameters considered by the trained machine learning model are easy to estimate. Also, the trained machine learning model has a series-parallel architecture, which has a high training speed and is more accurate for training as the trained machine learning model receives inputs and actual target for learning. Furthermore, to present a dynamic system with some moderate assumption, the trained machine learning model can be represented by a NARX model, which yields better performance, higher accuracy and more stability that other non-linear model schemes. Moreover, the trained machine learning model employing the NARX network is converted to Network Operation Engine (NOE) network and is validated through multi-step prediction of the output from the system 100. Additionally, the trained machine learning model is configured to overcome the drawback or transport delay or installation of the sensors 112 in proximity to exhaust ports of the engine 102. Thus, the technique employed in the trained machine learning model is capable of being implemented in multi-cylinder engines for monitoring an engine cylinder that operates away from the optimum stoichiometric ratio. Furthermore, the system 100 is capable of measuring the AFR during the cold start-up condition of the vehicle, and thus eliminating problems associated during cold start-up condition of the vehicle. Also, as the trained machine learning model is trained to adhere to emission regulations, the system 100 reduces pollutant emissions and also the fuel consumption during fast transients of engine operation. Also, the system 100 ensures fine control of AFR input to the engine 102, thereby enhancing thermal efficiency and fuel economy of the engine 102. Additionally, the system 100 is also capable of predicting the age of the lambda sensor, based on the value of the AFR.

Reference numerals

100 System for estimating AFR
102 Engine
104 Ion current measurement device
106 Data logging unit
108 Control unit
110 Analyzer unit
112 One or more sensors
114 Calibration unit
116 Regulating unit
118 Processing unit
120 Processing module
122 Analytic module
124 Power supply
126 Input/output module
128 Memory unit

, Claims:WE CLAIM:
1. A system (100) for estimating an air-fuel ratio (AFR) of an engine (102) in a dynamic condition of the engine (102), the system (100) comprising:
an ion current measurement device (104) in communication with an ignition coil, the ion current measurement device (104) being configured to measure an ion current signal generated on a spark event in a spark plug of the engine (102);
one or more sensors (112) coupled to the engine (102), the one or more sensors (112) adapted to generate one or more engine parameters; and
a control unit (108) communicatively coupled to the ion current measurement device and the one or more sensors (112), the control unit (108) being configured to:
receive the ion current signal from the ion current measurement device and the one or more engine parameters from the one or more sensors (112), and
estimate the AFR of the engine (102) based on the ion current signal and the one or more engine parameters, the estimation being performed based on a trained machine learning model.

2. The system (100) as claimed in claim 1 comprises a data logging device (106) in communication with the ion current measurement device (104), the one or more sensors (112) and the control unit (108), the data logging device (106) being configured to receive the ion current signal from the ion current measurement device (104) and the one or more engine parameters from the one or more sensors (112) and provide the ion current signal and the one or more engine parameters to the control unit (108).

3. The system (100) as claimed in claim 1, wherein the control unit (108) is configured for training a machine learning model to obtain the trained machine learning model, and the control unit (108) being communicably coupled to an analyzer unit (110), the analyzer unit (108) being adapted to compare the estimated AFR with a reference AFR computed by a wideband lambda sensor.

4. The system (100) as claimed in claim 1, wherein the trained machine learning model is a non-linear autoregressive neural network.

5. The system (100) as claimed in claim 1, wherein the one or more engine parameters comprises:
a throttle position of a throttle body of the engine (102);
a fuel pulse width of a fuel supply device;
an instantaneous engine speed of the engine (102);
a manifold air pressure in the engine (102); and
an engine temperature, and
an output from a narrowband lambda sensor (112e), the output indicating one of the AFR being a rich mixture or a lean mixture.

6. The system (100) as claimed in claim 1, wherein the control unit (108) is configured to pre-process the ion current signal before sending the ion current signal to the trained machine learning model, the pre-processing involving interpolation of the ion current signal for identifying region of interest to be sampled in an analog ion current signal, wherein the control unit (108) is configured to perform a principal component analysis on the region of interest determined during the preprocessing.

7. The system (100) as claimed in claim 1, comprises a calibration unit (114) in communication with the at least one of the control unit (108) and a data logging device (106), the calibration unit (114) being configured to calibrate the one or more engine parameters and the ion current signal transmitted to the at least one of the control unit (108) and the data logging device (106), as per required sampling rate or frequency.

8. The system (100) as claimed in claim 1, comprises a regulating unit (116) in communication with the one or more sensors (112) and the regulating unit (116) being configured to sample at least one of the engine parameters at a different frequency to overcome signal trimming of the at least one of the engine parameters.

9. The system as claimed in claim 1, comprises a processing unit (118) in communication with a regulating unit (116) and a data logging device (106) and the processing unit (118) being configured to compute and generate a fuel pulse width signal from state of a fuel supply device of the engine (102).

10. A method (600) for estimating an air-fuel ratio (AFR) for an engine (102), the method comprising:
generating (602), by an ion current measurement device (104), an ion current signal, the ion current signal being generated based on measurement of an ionization current supplied to a spark plug of an engine (102);
receiving (604), by a control unit (108), the ion current signal from the ion current measurement device (104);
receiving (606), by the control unit (108), one or more engine parameters from one or more sensors (112);
and
estimating (608), by the control unit (108), the AFR of the engine (102) based on the ion current signal and the one or more engine parameters, the estimation being performed based on a trained machine learning model.

11. The method (600) as claimed in claim 10 comprising, pre-processing, by the control unit (108), the ion current signal before sending the ion current signal to the trained machine learning model, the pre-processing involving interpolation of the ion current signal for identifying region of interest to be sampled in an analog ion current signal, wherein the control unit (108) is configured to perform a principal component analysis on the region of interest determined during the preprocessing.

12. The method (600) as claimed in claim 10, wherein the trained machine learning model is a non-linear autoregressive neural network.

13. The method (600) as claimed in claim 10 comprising, training, by the control unit (108), a machine learning model to obtain the trained machine learning model, and comparing the estimated AFR with a reference AFR computed by a wideband lambda sensor, by the analyzer unit (108).

14. The method (600) as claimed in claim 10 comprising, receiving by a data logging device (106) the ion current signal from the ion current measurement device (104) and the one or more engine parameters from the one or more sensors (112), the data logging device (106) being configured to provide the ion current signal and the one or more engine parameters to the control unit (108).

15. The method (600) as claimed in claim 10, wherein the one or more engine parameters comprises:
a throttle position of a throttle body of the engine (102);
a fuel pulse width of a fuel supply device;
an instantaneous engine speed of the engine (102);
a manifold air pressure in the engine (102); and
an engine temperature, and
an output from a narrowband lambda sensor (112e), the output indicating one of the AFR being a rich mixture or a lean mixture.

16. The method (600) as claimed in claim 10, comprising calibrating the one or more engine parameters and the ion current signal transmitted to the at least one of the control unit (108) and the data logging device (106), as per required sampling rate by a calibration unit (114), wherein the calibration unit (114) is in communication with the at least one of the control unit (108) and a data logging device (106).

17. The method (600) as claimed in claim 10, comprising sampling at least one of the engine parameters at a different frequency to overcome signal trimming by a regulating unit (116), wherein the regulating unit is in communication with the one or more sensors (112).

18. The method (600) as claimed in claim 10, comprising computing and generating a fuel pulse width signal from state of a fuel supply device of the engine (102) by a processing unit (118), wherein the processing unit (118) is in communication with a regulating unit (116) and a data logging device (106).

Dated this 09th day of September 2022
TVS MOTOR COMPANY LIMITED
By their Agent & Attorney

(Nikhil Ranjan)
of Khaitan & Co
Reg No IN/PA-1471