Description:Complete Specification:
The following specification describes and ascertains the nature of this invention and the manner in which it is to be performed.

Field of the invention:
[0001] The present invention relates to a controller to estimate intake air mass for an engine of a vehicle and method thereof.

Background of the invention:
[0002] According to a prior art AT522649, a method and system for determining the amount of air supplied to an internal combustion engine is disclosed. System for the determination of the amount of air supplied to an internal combustion engine, created by a process comprising the following steps: - operating a virtual or real reference internal combustion engine with a virtual or real air flowmeter in different operating conditions, wherein the virtual internal combustion engine and of the virtual air flowmeter by a mathematical model of the real reference internal combustion engine, and possibly of the real air mass sensor are formed, or determining input values in the individual operating states of the internal combustion engine, reference, the amount of air supplied to the internal combustion engine in the individual values of the reference operating conditions, - training of the artificial neural network, so that it is in the case of entry of the same input values are substantially the same amount of air as the reference values of the internal combustion engine.

Brief description of the accompanying drawings:
[0003] An embodiment of the disclosure is described with reference to the following accompanying drawings,
[0004] Fig. 1 illustrates a block diagram of a controller to estimate intake air mass for an engine of a vehicle, according to an embodiment of the present invention;
[0005] Fig. 2 illustrates a plot of air mass against time with and without using the air mass sensor, according to an embodiment of the present invention, and
[0006] Fig. 3 illustrates a method for estimating intake air mass for the engine of the vehicle, according to the present invention.

Detailed description of the embodiments:
[0007] Fig. 1 illustrates a block diagram of a controller to estimate intake air mass for an engine of a vehicle, according to an embodiment of the present invention. The controller 110 configured to receive input parameters 102 comprising engine speed, injection quantity, torque, accelerator pedal position, environmental pressure, boost pressure, exhaust manifold pressure (or exhaust manifold absolute pressure), actuator position, ambient temperature and coolant temperature. The controller 110 processes the input parameter 102 through a Machine Learning (ML) model 118. The ML model 118 is pretrained using samples of the input parameters 102, and estimate the intake air mass using the ML model 118, characterized in that, the input parameter 102 further comprises an exhaust manifold delta pressure 104 as additional parameter to improve the accuracy of the estimated intake air mass.

[0008] According to an embodiment of the present invention, the ML model 118 is a Recurrent Neural Network (RNN), and specifically a Long Short-Term Memory (LSTM) based Recurrent Neural Network (RNN). The neural network consists of interconnected layers of artificial neurons, where each neuron receives inputs, applies a nonlinear activation function, and passes the output to the next layer. The neural networks are usable for both time-independent and time-dependent problems. However, the neural networks are not explicitly designed to handle time dependencies unless additional structures like recurrent layers or temporal convolutions are incorporated. The neural networks are trained using various optimization algorithms, such as gradient descent or its variants, to minimize a specified loss function. The weights and biases of the network are adjusted during training to improve its performance on the given task. The neural networks typically work with fixed-size input vectors or tensors. Each input is usually considered independent of the others, and the model's architecture does not inherently account for temporal dependencies.

[0009] The RNNs are unique on account of the ability to process both past data and input data with memory and overcomes the weaknesses of the feed-forward network. Due to an internal memory, the RNN makes relatively precise predictions. The RNN is useful for solving sequential data-based problems. The RNN is the neural network with memory and double data processing. The RNN maps out several inputs and outputs. Unlike other algorithms that deliver one output for one input, the benefit of RNN is mapping out many to many, one to many, and many to one input and outputs.

[0010] The training of the ML model 118 is explained briefly. The required data to train the ML model 118 is collected from an engine test bench or data recorded from a vehicle in real drive scenario. The machine learning concept used is RNN capable of learning order dependence in sequence prediction problems which considers the relationship with all the inputs provided and gives the output. The data comprising values of input parameter 102 is fed into the ML model 118 to be trained. The input parameter 102 provided for calculation are taken from the engine mapping and the transient cycles run on the engine testbench or the real time collected data. The parameters considered are engine speed, injection quantity, torque, accelerator pedal position, environmental pressure, boost pressure, exhaust manifold pressure, actuator position, ambient temperature and coolant temperature, and additionally exhaust manifold delta pressure measured from exhaust delta pressure sensor 124. Since the input also contains the environmental pressure, boost pressure and the exhaust manifold delta pressure, any disturbances in the open loop system like intake depression and back pressure increase due to soot loading is captured and intake air mass 120 is calculated accordingly.

[0011] According to an embodiment of the present invention, the data from the engine mapping and transient cycles comprises input parameters without Exhaust Gas Recirculation (EGR) and Throttle Valve Actuation (TVA) and with EGR and TVA, with significant soot loaded in the DPF 122 (to train for back pressure impact), and with clogged air filter (to train for intake depression impact).

[0012] According to an embodiment of the present invention, a volumetric efficiency is usable to validate the output of the ML model 118. The volumetric efficiency is a measure of consumption of intake air compared to maximum potential of the engine. The volumetric efficiency represents the ratio of the actual amount of air entering the engine's cylinders to the theoretical maximum amount of air that the engine could draw in under the same conditions. By comparing the predicted air mass of the ML model 118 with the volumetric efficiency air mass, the controller 110 assess the accuracy of the estimated/predicted air mass. If the predicted air mass closely aligns with the volumetric efficiency air mass, then the controller 110 ascertains that the ML model 118 predictions are reliable and accurate. On the other hand, significant discrepancies between the two values indicates issues with the ML model 118 or the data used for training and testing. This comparison helps ensure the effectiveness of the ML model 118 in estimating air mass, which is crucial for engine performance optimization and emissions control. The calculation of the volumetric efficiency involves intake manifold pressure and intake manifold temperature. The volumetric efficiency factor calibrated using the map engine speed versus injection quantity.

[0013] Once the training of the ML model 118 with required accuracy is obtained, the ML model 118 is shifted to the memory element 108 of the controller 110. With the estimation of ML model 118 in open loop, the predicted/estimated air mass deviation within is ±5% in relative frequency >95% of the data points under base, back pressure simulated data and intake depression simulated data. The ML model 118 is deployable to three-wheeler vehicle, buses and the OFF Highway vehicles segment as well.

[0014] According to the present invention, the controller 110 is provided with necessary signal detection, acquisition, and processing circuits. The controller 110 is a control unit which comprises memory element 108 such as Random Access Memory (RAM) and/or Read Only Memory (ROM), Analog-to-Digital Converter (ADC) and a Digital-to-Analog Convertor (DAC), clocks, timers, counters and at least one processor (capable of implementing machine learning) connected with each other and to other components through communication bus channels. The memory element 108 is pre-stored with logics and/or instructions and/or programs and/or applications and/or ML models/module 118 and/or threshold values, which is/are accessed by the at least one processor as per the defined routines. The internal components of the controller 110 are not explained for being state of the art, and the same must not be understood in a limiting manner. The controller 110 may also comprise communication units to communicate with an external devices such as the cloud, a remote server, etc., through wireless or wired means such as Global System for Mobile Communications (GSM), 3G, 4G, 5G, Wi-Fi, Bluetooth, Ethernet, serial networks, and the like. The controller 110 is implementable in the form of System-in-Package (SiP) or System-on-Chip (SOC) or any other known types.

[0015] According to an embodiment of the present invention, the ML model 118 is a Nonlinear Autoregressive Network (NARX) with Exogenous inputs. The NARX model is a specific type of network that is designed for time series prediction. The NARX model includes an autoregressive component, which indicates usage of past values of the target variable as inputs, along with exogenous inputs that capture additional information related to the target variable. The NARX models are specifically designed to capture time dependencies in time series data. They incorporate the autoregressive component, which allows them to consider the influence of past values on the current prediction. This makes the NARX model particularly suitable for time-dependent problems. The NARX models are trained using similar optimization algorithms as neural networks. However, due to their specific time-dependent structure, they may require additional considerations during training, such as defining appropriate lagged inputs and handling time delays. The NARX models, being designed for time series data, can handle inputs that have a temporal structure. They can incorporate lagged versions of the target variable (autoregressive inputs) and exogenous inputs, allowing them to model the influence of past values on the current prediction. Hence, similar results as shown in figure 2 can be obtained with NARX model also.

[0016] According to an embodiment of the present invention, the exhaust manifold delta pressure 104 is measured using pressure sensor 124 mounted across a Diesel Particulate Filter (DPF) 122. Since, the pressure sensor 124 near the exhaust port of the engine is not usable as it is subject to high temperature, the pressure sensor(s) 124 mounted across the DPF 122 is used to measure/estimate the exhaust manifold delta pressure 104. The pressure sensor 124 is available in the vehicle layout and provides live inputs for the ML model 118.

[0017] According to an embodiment of the present invention, the controller 110 is configurable to consider the boost pressure which is the pressure in the intake manifold at various stages of clogging of the air filter. In the conventional solution, the boost pressure is considered but with air filter without any clogging. However, the ML model 118 of the present invention is trained on air filter with different levels of clogging, thus yielding even more accurate results.

[0018] In yet another embodiment of the present invention, the actuator is at least one selected from a group comprising a throttle actuator located in the intake manifold of the engine, a turbocharger with electronically controlled vanes, , the Exhaust Gas Recirculation (EGR) valve for the exhaust gas, and an exhaust flap component used in some automotive exhaust systems to control the flow of exhaust gases, exhaust temperatures and for engine braking. The turbocharger with electronically controlled vanes also known as Variable Geometry Turbocharger. VGT is a type of forced induction system used in internal combustion engines to improve their performance and efficiency. The VGT allows the vanes in the turbine or compressor housing to be adjusted or varied electronically based on the engine's operating conditions. This adjustability enables the turbocharger to optimize its performance at different engine speeds and loads

[0019] According to an embodiment of the present invention, the controller 110 is part of or is at least one of an internal device or an external device. The internal device is an Electronic Control Unit (ECU) of the vehicle 106. The external device is at least one selected from a group comprising a portable device 114 and a cloud device 116. The portable device 114 is selected from a group comprising a smartphone, a smartwatch, a tablet, a laptop and the like. The ECU of the vehicle is at least one selected from a group comprising an Engine Control Unit or Exhaust/emission control unit, or other control unit within the vehicle 106. When the controller 110 is part of the ECU or is the ECU itself, the input parameters are received by in-vehicle network such as Controller Area Network (CAN) as known in the art. Similarly, when the controller 110 is part of the portable device 116, the input parameters are received from the vehicle 106 through a communication unit 112 such as a Telematic Control Unit (TCU), a Universal Serial Bus (USB), an On-Board Diagnostic (OBD) port and the like. The TCU comprises wireless connectivity means such as (GSM), 3G, 4G, 5G, Wi-Fi, Bluetooth etc. Similarly, the OBD port is connectable with dongle with wireless connectivity means. When the controller 110 is part of the cloud device 116, then the input parameters 102 are either received through the communication unit 112 or through the portable device connected to the communication unit 112.

[0020] Fig. 2 illustrates a plot of air mass against time with and without using the air mass sensor, according to an embodiment of the present invention. The Y-axis 202 denotes air mass and X-axis 204 denotes time, both in respective suitable units. There are three plots, a first plot 206 represented by solid line, a second plot 208 represented by dotted line and a third plot 210 represented by dashed line. The first plot 206 is actual air mass as measured by an air mass flow sensor or mass air flow sensor. The plot 208 is the estimated intake air mass without exhaust manifold pressure signal and the plot 210 is estimated intake air mass with exhaust manifold pressure signal. As can be seen, the third plot 210 is closer to the first plot 206 than the second plot 208 indicating that the significance of the additional parameter, i.e. exhaust manifold delta pressure 104 which is measured by pressure sensor 124 mounted across the DPF 122.

[0021] According to the present invention, a working of the controller 110 is envisaged. Consider the vehicle 102 is provided with the controller 110 which is installed with the ML model 118 as the virtual sensor as a replacement of the actual air mass flow sensor. The controller 110 is the Engine Control Unit of the vehicle 110. The ML model 118 is provided with input parameters 102 as measured/derived/received or estimated by the controller 110. The ML model 118 processes the input parameter 102 and provides intake air mass as the output 120. The output 120 is then used by the Engine Control Unit (ECU) for various functions such as fuel injection quantity, soot mass calculation and other functions known in the art.

[0022] In another working scenario, consider the controller 110 being part of the cloud device 116. The ECU of the vehicle 106 which receives the data through the in-vehicle network is transmitted to the cloud device 116 through the communication unit 112. The controller 110 in the cloud device 116 receives the data containing the input parameters 102, processes through the ML model 118 and outputs estimated intake air mass. The output 120 is then transmitted back to the ECU through the communication unit 112 for further processing.

[0023] In yet another working scenario, the estimation of the intake air mass is shared among the internal device and the external device or within the external devices as per the possible load sharing.

[0024] Fig. 3 illustrates a method for estimating intake air mass for the engine of the vehicle, according to the present invention. The method comprises plurality of steps of which a step 302 comprises receiving, by the controller 110, real-time input parameters 102 comprising engine speed, injection quantity, torque, accelerator pedal position, environmental pressure, boost pressure, exhaust manifold pressure, actuator position, ambient temperature and coolant temperature. The input parameters 102 are detected/measured/derived by sensors present in the vehicle 102. A step 304 comprises processing, by the controller 110, the input parameters 102 through the ML model 118. The ML model 118 is pretrained using samples of the input parameter 102. A step 306 comprises estimating, by the controller 110, the intake air mass using the ML model 118. The method is characterized by step 302 where the input parameter 102 further comprises the exhaust manifold delta pressure 104 as additional parameter which improves the accuracy of estimation of the intake air mass.

[0025] According to the method, the step 304 where processing through ML model 118 is mentioned, the ML model 118 is the Recurrent Neural Network (RNN), and specifically the Long Short-Term Memory (LSTM) network. Alternatively, the ML model 118 is the Nonlinear Autoregressive Network (NARX) with Exogenous inputs. In the step 306, the exhaust manifold delta pressure is measured using pressure sensor 124 mounted across Diesel Particulate Filter (DPF) 122.

[0026] According to the present invention, the method is performed by the controller 110 which is part of or is at least one selected from the group comprising the internal device and the external device. The internal device comprises an Electronic Control Unit within the vehicle 102, and the external device is at least one selected from the portable device 114, and the cloud device 116. The intake air mass calculation considering the external disturbances like exhaust back pressure and intake depression when the naturally aspirated engine of the vehicle 102 does not have a dedicated air mass sensor is provided, i.e. open loop system. The machine learning model 118 of the present invention learns from the input parameter 102 given to the controller 110 and calculates the intake air mass based on the behavior of the input parameters 102. The intake air mass calculated is within the accepted Key Performance Indicator of air mass accuracy and hence is indeed an accurate input to the smoke limitation, soot mass calculation, and other similar functions. The controller 110 considers the input parameter 102 from the already available inputs in the vehicle 102.

[0027] According to the present invention, the method also comprises checking the correctness of the predicted/estimated air mass of the ML model 118 with the volumetric efficiency air mass. The method comprises comparing the estimated air mass of the ML model 118 with the volumetric efficiency air mass, if the difference is within acceptable range, then the method validates the correctness of the estimated air mass and indicates that the ML model 118 is reliable and accurate.

[0028] According to an embodiment of the present invention, machine learning based air mass modeling for Off-Highway and other vehicles with naturally aspirated engine is provided. The similar approach is implementable to the system layout with turbocharger. The ML model 118 is usable as virtual sensor either as a replacement of mass air flow sensor or as a redundant means in case the mass air flow sensor becomes faulty. The ML model 118 is also possible to be used to correct any drift in the mass air flow sensor as well.

[0029] It should be understood that the embodiments explained in the description above are only illustrative and do not limit the scope of this invention. Many such embodiments and other modifications and changes in the embodiment explained in the description are envisaged. The scope of the invention is only limited by the scope of the claims.
, Claims:We claim:
1. A controller (110) to estimate intake air mass for an engine of a vehicle (106), said controller (110) configured to:
receive real-time input parameters (102) comprising engine speed, injection quantity, engine torque, accelerator pedal position, environmental pressure, boost pressure, exhaust manifold pressure, actuator position, ambient temperature and coolant temperature;
process said input parameter (102) through a Machine Learning (ML) model (118), said ML model (118) is pretrained using samples of said input parameters (102), and
estimate said intake air mass (120) using said ML model (118), characterized in that, said input parameter (102) further comprises an exhaust manifold delta pressure (104).

2. The controller (110) as claimed in claim 1, wherein said ML model (118) is a Recurrent Neural Network (RNN), and specifically a Long Short-Term Memory (LSTM) network.

3. The controller (110) as claimed in claim 1, wherein said ML model (118) is a Nonlinear Autoregressive Network (NARX) with Exogenous inputs.

4. The controller (110) as claimed in claim1, wherein said exhaust manifold delta pressure (104) is measured using a pressure sensor (124) mounted across Diesel Particulate Filter (DPF) (122) in an exhaust conduit of said vehicle (106).

5. The controller (110) as claimed in claim 1 is part of at least one selected from a group comprising an internal device and an external device, wherein said internal device comprises an Electronic Control Unit within said vehicle (106), and said external device comprises a portable device (116), and a cloud device (114).

6. A method for estimating intake air mass (120) for an engine of a vehicle (106), said method comprising the steps of:
receiving real-time input parameters (102) comprising engine speed, injection quantity, engine torque, accelerator pedal position, environmental pressure, boost pressure, exhaust manifold pressure, actuator position, ambient temperature and coolant temperature;
processing said input parameters (102) through a Machine Learning (ML) model (118), said ML model (118) is pretrained using samples of said input parameters (102), and
estimating said intake air mass (120) using said ML model (118), characterized by, said input parameter (102) further comprises an exhaust manifold delta pressure (104).

7. The method as claimed in claim 6, wherein said ML model (118) is a Recurrent Neural Network (RNN), and specifically a Long Short-Term Memory (LSTM) network.

8. The method as claimed in claim 6, wherein said ML model (118) is a Nonlinear Autoregressive Network (NARX) with Exogenous inputs.

9. The method as claimed in claim 6, wherein said exhaust manifold delta pressure (104) is measured using a pressure sensor (124) mounted across a Diesel Particulate Filter (DPF) (122) in an exhaust conduit of said vehicle (106).

10. The method as claimed in claim 6 is performed by at least one selected from a group comprising an internal device and an external device, wherein said internal device comprises an Electronic Control Unit within said vehicle (106), and said external device comprises a portable device (116), and a cloud device (114).