Method for real-time determination of clutch torque of an electronic limited-slip differential

Abstract

A method for real-time determination of a clutch torque of an electronic limited-slip differential (eLSD) (39), comprising: Real-time receipt of vehicle data, wherein the vehicle data includes a torque request; Determining a preliminary eLSD clutch torque using a neural network and the vehicle data; Determining eLSD clutch torque bounds using a physics-based model; Determine whether the preliminary eLSD clutch torque is outside the eLSD clutch torque limits; in response to determining that the preliminary eLSD clutch torque is outside the clutch torque limits of the eLSD (39), adjusting the preliminary eLSD clutch torque using the clutch torque limits to determine a final eLSD clutch torque; and Commanding the eLSD (39) in real time to apply the final clutch torque to the clutch (43) of the eLSD (39).

Description

INTRODUCTION

[0001] The present invention relates to a method for real-time determination of a clutch torque of an electronic limited-slip differential.

[0002] Electronic limited-slip differential (eLSD) systems include a clutch. Some eLSD systems use a physics-based clutch torque estimation algorithm to estimate clutch torque. However, when the eLSD system is locked, physics-based clutch torque estimation algorithms are not always accurate due to inherent limitations of the physics-based model. In particular, the physics-based model makes certain assumptions, such as that the vehicle is on a surface with a high coefficient of friction and does not experience split-coefficient friction conditions, which can lead to inaccurate estimates under other conditions. These inaccuracies can cause confusion for the eLSD controller and may result in erroneous actions being taken.Consequently, it is desirable to consider the limitations of the physics-based model used in the locked-state clutch torque estimation algorithm and apply other methods to improve the accuracy of the estimation.

[0003] For further background information, reference is made to the documents DE 10 2022 107 911 A1, DE 11 2018 001 436 T5 and WO 2023 / 002 405 A1. SUMMARY

[0004] According to the invention, a method for real-time determination of a clutch torque of an electronic limited-slip differential is presented, which is characterized by the features of claim 1.

[0005] Also described is a vehicle including sensors, an electromechanical steering system, and a controller in communication with the sensors and the electromechanical steering system. The controller is programmed to perform the method described above. The present invention further describes a tangible, non-transitory, machine-readable medium containing machine-readable instructions that, when executed by a processor, cause the processor to perform the method described above.

[0006] Further areas of applicability of the present invention will become apparent from the detailed description provided below. It is to be understood that the detailed description and specific examples are provided for purposes of illustration only.

[0007] The above-described features and advantages and other features and advantages of the presently disclosed system and method will be apparent from the detailed description, including the claims and exemplary embodiments, when taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The present invention will be more fully understood from the detailed description and the accompanying drawings, in which: Fig. 1 is a block diagram showing one embodiment of a vehicle including an eLSD. Fig. 2 a flowchart of a method for estimating the clutch torque of the vehicle's eLSD. DETAILED DESCRIPTION

[0009] Reference will now be made in detail to some examples of the invention illustrated in the accompanying drawings. Where possible, the same or similar reference numerals are used throughout the drawings and the description to refer to the same or similar sections or steps.

[0010] With reference to Fig. 1, a vehicle 10 generally includes a chassis 12, a body 14, front and rear wheels 17, and may be referred to as a vehicle system. In the illustrated embodiment, the vehicle 10 includes two front wheels 17a and two rear wheels 17b. The body 14 is disposed on the chassis 12 and substantially encloses components of the vehicle 10. The body 14 and the chassis 12 may together form a frame. The wheels 17 are each rotationally coupled to the chassis 12 near a respective corner of the body 14. The vehicle 10 includes a front axle 19 coupled to the front wheels 17a and a rear axle 25 coupled to the rear wheels 17b.

[0011] The vehicle 10 is an autonomous vehicle, and a control system 98 is incorporated into the vehicle 10. The system 98 may be referred to as the system. The vehicle 10 is, for example, a vehicle that is automatically controlled to convey passengers from one location to another. The vehicle 10 is shown as a pickup truck in the illustrated embodiment, but it should be appreciated that other vehicles, including trucks, sedans, coupes, sport utility vehicles (SUVs), recreational vehicles (RVs), etc., may also be used. In one embodiment, the vehicle 10 may include a so-called level two, level three, level four, or level five driving automation system.A level four system indicates "high automation," which refers to the drive-mode-specific performance by an automated driving system of aspects of the dynamic driving task, even when a human driver does not appropriately respond to a request to intervene. A level five system indicates "full automation," which refers to the full-time performance by an automated driving system of aspects of the dynamic driving task under multiple road and environmental conditions that can be managed by a human driver. In Level 3 vehicles, the system 98 performs the entire dynamic driving task (DDT) within the range in which it is designed to do so. In Level 2 vehicles, systems provide steering, braking / acceleration assistance, lane centering, and adaptive cruise control.However, even when these systems are activated, the vehicle operator must be behind the wheel and constantly monitor the automated features.

[0012] As shown, the vehicle 10 generally includes a propulsion system 20, a transmission system 22, a steering system 24, a braking system 26, a sensor system 28, an actuator system 30, at least one data storage device 32, at least one controller 34, and a communication system 36. The steering system 24 is an electromechanical steering system. The propulsion system 20, in various embodiments, may include an electric machine such as a traction motor and / or a fuel cell propulsion system. The vehicle 10 may further include a battery (or battery pack) 21 electrically connected to the propulsion system 20. Accordingly, the battery 21 is configured to store electrical energy and to provide electrical energy to the propulsion system 20. In certain embodiments, the propulsion system 20 may include an internal combustion engine.The transmission system 22 is configured to transfer power from the propulsion system 20 to the vehicle wheels 17 according to selectable speed ratios. According to various embodiments, the transmission system 22 may include a stepped-ratio automatic transmission, a continuously variable transmission, or other suitable transmission. The braking system 26 is configured to impart braking torque to the vehicle wheels 17. In various embodiments, the braking system 26 may include friction brakes, electromechanical braking, a regenerative braking system such as an electric machine, and / or other suitable braking systems. The steering system 24 influences the position of the vehicle wheels 17 and may include a steering wheel 33. While illustrated as including a steering wheel 33 for illustrative purposes, in some embodiments, the steering system 24 may not include a steering wheel 33.

[0013] The sensor system 28 includes one or more sensors 40 (i.e., measuring devices) that sense observable conditions of the external environment and / or the internal environment of the vehicle 10. The sensors 40 are in communication with the controller 34 and may include one or more steering wheel sensors 45, one or more radar devices, one or more light detection and ranging sensors (lidar sensors), one or more proximity sensors, one or more odometers, one or more ground penetrating radar (GPR) sensors, one or more steering angle sensors, global navigation satellite system (GNSS) transceivers (e.g., one or more global positioning systems (GPS) transceivers), one or more tire pressure sensors, one or more cameras 41 (e.g.,Eye tracking), one or more gyroscopes, one or more accelerometers, one or more tilt sensors, one or more speed sensors, one or more ultrasonic sensors, one or more inertial measurement units (IMUs), one or more night vision devices, thermal imaging sensors, and / or other sensors. Each sensor 40 is configured to generate a signal indicative of the sensed observable conditions of the external environment and / or the internal environment of the vehicle 10. Since the sensor system 28 provides data to the controller 34, the sensor system 28 and its sensors 40 are considered information sources (or simply sources).

[0014] The actuator system 30 includes one or more actuators 42 that control one or more vehicle features, such as the propulsion system 20, the powertrain system 22, the steering system 24, and the braking system 26. According to various embodiments, the features of the vehicle may further include interior and / or exterior features of the vehicle, such as doors, a trunk, and features of the vehicle interior, such as ventilation, music, lighting, etc. The actuators 42 may be part of the steering system 24 and include one or more wheel actuators (RWAs) and a handwheel actuator (HWA).

[0015] The data storage device 32 stores data for use in automatically controlling the vehicle 10. In various embodiments, the data storage device 32 stores defined maps of the navigable environment. According to various embodiments, the defined maps may be predefined by and obtained from a remote system. For example, the defined maps may be compiled by the remote system and communicated to the vehicle 10 (wirelessly and / or in a wired manner) and stored in the data storage device 32. The data storage device 32 may be part of the controller 34, separate from the controller 34, or part of the controller 34 and part of a separate system.

[0016] The vehicle 10 may further include one or more airbags 35 in communication with the controller 34 or another controller of the vehicle 10. The airbag 35 includes an inflatable bladder and is configured to transition between a stowed configuration and a deployed configuration to cushion the effects of an external force applied to the vehicle 10. The sensors 40 may include an airbag sensor, such as an IMU, configured to detect an external force and generate a signal indicative of the magnitude of such an external force. The controller 34 is configured to command the airbag 35 to deploy based on the signal from one or more sensors 40, such as the airbag sensor. Accordingly, the controller 34 is configured to determine when the airbag 35 has deployed.

[0017] The controller 34 includes at least one processor 44 and a non-transitory computer-readable storage device or storage medium 46. The processor 44 may be a custom-built or off-the-shelf processor, a central processing unit (CPU), a graphics processing unit (GPU), an auxiliary processor among multiple processors associated with the controller 34, a semiconductor-based microprocessor (in the form of a microchip or chipset), a macroprocessor, a combination thereof, or generally a device for executing instructions. The computer-readable storage device or storage medium 46 may include, for example, volatile and non-volatile memory in read-only memory (ROM), random access memory (RAM), and retained memory (RAM).KAM is persistent or non-volatile memory that can be used to store various operating variables while the processor 44 is powered off. The computer-readable storage device or medium 46 can be implemented using a number of storage devices such as PROMs (programmable read-only memories), EPROMs (electrical PROMs), EEPROMs (electrically erasable PROMs), flash memory, or other electrical, magnetic, optical, or combination storage devices that can store data, some of which represents executable instructions used by the controller 34 in controlling the vehicle 10. The controller 34 of the vehicle 10 may be referred to as a vehicle controller and may be programmed to implement a method 100 (. Fig. 2), which is described in detail below.

[0018] The instructions may include one or more separate programs, each comprising an ordered list of executable instructions for implementing logical functions. The instructions, when executed by the processor 44, receive and process signals from the sensor system 28, perform logic, calculations, methods, and / or algorithms to automatically control the components of the vehicle 10, and generate control signals to the actuator system 30 to automatically control the components of the vehicle 10 based on the logic, calculations, methods, and / or algorithms. Although in Fig. 1, embodiments of the vehicle 10 may include multiple controllers 34 that communicate via a suitable communication medium or combination of communication media and that cooperate to process the sensor signals, perform logic, calculations, methods, and / or algorithms, and generate control signals to automatically control features of the vehicle 10. In various embodiments, one or more commands of the controller 34 are embodied in the control system 98.

[0019] The vehicle 10 includes a user interface 23, which may be a touchscreen in the instrument panel. The user interface 23 may include an alarm such as one or more speakers 27 to provide an audible sound, haptic feedback in a vehicle seat or other object, one or more display devices 29, one or more microphones 31, and / or other devices suitable for providing notification to the vehicle user of the vehicle 10. The user interface 23 is in electronic communication with the controller 34 and is configured to receive inputs from a vehicle occupant 11 (e.g., a vehicle driver or a vehicle passenger). For example, the user interface 23 may include a touchscreen and / or buttons configured to receive inputs from a vehicle occupant 11.Accordingly, the controller 34 is configured to receive inputs from the user via the user interface 23. The vehicle 10 may include one or more display devices 29 configured to display information about the vehicle occupant 11 (e.g., the vehicle operator or passenger) and may be a head-up display (HUD).

[0020] The communication system 36 is in communication with the controller 34 and is configured to wirelessly communicate information to and from other remote vehicles 48, such as other vehicles ("V2V" communication), infrastructure ("V2I" communication), remote systems at a remote call center (e.g., ON-STAR from GENERAL MOTORS), and / or personal electronic devices such as a cellular phone. In the present invention, the term "remote vehicle" refers to a vehicle, such as a passenger car, that is configured to send one or more signals to the vehicle 10 while not physically connected to the vehicle 10. In certain embodiments, the communication system 36 is a wireless communication system configured to communicate via a wireless local area network (WLAN) using IEEE 802.11 standards or using cellular data communication.However, additional or alternative communication methods, such as a dedicated short-range communication channel (DSRC channel), are also contemplated. DSRC channels refer to unidirectional or bidirectional short- to medium-range wireless communication channels specifically designed for use in motor vehicles, and a corresponding set of protocols and standards. Accordingly, the communication system 36 may include one or more antennas and / or communication transceivers 37 for receiving and / or transmitting signals such as cooperative sensing messages (CSMs). The communication transceivers 37 may be considered sensors 40. The communication system 36 is configured to wirelessly communicate information between the vehicle 10 and another vehicle.Furthermore, the communication system 36 is configured to wirelessly communicate information between the vehicle 10 and infrastructure or other vehicles.

[0021] The vehicle 10 includes one or more electronically controlled limited-slip differentials (eLSD) 39. The eLSD 39 includes a slip control algorithm that, through the use of an electrically actuated clutch 43, provides controlled, variable coupling of the vehicle's drive torque to two or more of the vehicle's wheels. Under normal driving conditions, the eLSD 39 functions as an open differential, distributing torque equally between the wheel shafts. However, if a loss of traction is detected at a differential output, the clutch 43 of the eLSD 39 can be activated via a closed-loop control to maintain a speed difference between the differential outputs that would result if the wheels were rotating at their natural speeds, thereby improving the stability and comfort of the vehicle 10.

[0022] Fig.2 shows a flowchart of a method 100 for real-time determination of clutch torque of an electronic limited-slip differential (eLSD). The method 100 begins in block 102. Block 102 includes developing a physics-based model to estimate the clutch torque of the eLSD 39 both when the eLSD 39 is locked and when the eLSD 39 is open. However, the physics-based clutch torque estimate for the open state is more reliable than the physics-based clutch torque estimate for the locked state. For the open state of the eLSD 39, the clutch torque is a function of the demanded torque, the speed differences between the wheels 17, and the clutch torque capacity. For the open state of the eLSD 39, the controller 34 may use the following equation to estimate the clutch torque of the eLSD 39: Tc_act=1τc s+1p1 tanh(p2(dω)p3)Tc_rqst where: Tc_act the clutch torque of the eLSD 39 is in the open state; T c_rqst the required torque; τ c is the model time constant; s is the Laplace domain; p1 is a first model coefficient; p2 is a second model coefficient; (dω) is the speed difference between the rear wheels; and p3 is a third model coefficient.

[0023] When the clutch 43 of the eLSD 39 is engaged, the wheel speeds are identical. For the locked state of the eLSD 39, the controller 34 can use the following equations to estimate the clutch torque of the eLSD 39: Reff,rFxrl+Reff,rFxrr=Tr_act−Iwr−Iwrω˙rr Tc_act=Reff,r(Fxrl−Fxrr) where: R eff,r is the effective rolling radius of the rear tires; F xrl is the traction force of the left rear wheel; T r_actthe rear engine actuation is; I wr is the moment of inertia of the right rear wheel; and ώ rr is the angular acceleration of the right rear wheel.

[0024] If both wheels 17 are within the linear range, the following equations can be used: Fxrlμrl Fzrl−Fxrrμrr Fzrr=Knorm Δκ Δκ=lwrReff,rωrr where: T c_act the clutch operation is; F xrr the traction force on the right rear wheel; µ rl is the friction coefficient of the left rear wheel; F zrl is the normal force on the left rear wheel; µ rr is the friction coefficient of the right rear wheel; F zrr is the normal force on the right rear wheel; F xrl the traction force on the left rear wheel; K normis the longitudinal stiffness of the tire normalized to the normal force; l w the wheelbase of the vehicle is 10; r is the yaw rate of the vehicle 10; R eff,r is the effective rolling radius of the rear tires; and ω rr is the angular velocity of the right rear wheel.

[0025] If both wheels 17 are within the non-linear range, the following equation can be used: FxrlFxrr=μrlFzrlμrrFzrr where: F xrl the rear left traction force; F xrr the rear right traction force is; µ rl is the rear left friction coefficient; µ rr is the rear right friction coefficient; F zrl is the rear left normal force; and F zrr is the rear right normal force.

[0026] If a wheel 17 is in a linear region, the following equation can be used: Fxrlμrl Fzrl−Fxrrμrr Fzrr=1−Knorm κrr where: F xrl the rear left traction force; F xrr the rear right traction force is; µ ri is the rear left friction coefficient; µ rr is the rear right friction coefficient; F zrl is the rear left normal force; and F zrr is the rear right normal force.

[0027] After executing block 102, the method 100 proceeds to block 104. Block 104 involves analyzing vehicle data to understand which signals may be useful for predicting the clutch torque of the eLSD 39. Certain input features (i.e., vehicle data) are selected for offline training. To reduce the complexity of the neural network, various combinations of possible features were chosen to train the neural network. A final set of input features was then selected based on the following criteria: (a) the performance of the neural network (i.e., the features resulting in the most accurate eLSD clutch torque estimate were selected); and (b) the ease of implementation in the production model (i.e., input features that are more difficult to implement in the production model are not selected as inputs to the neural network).After analyzing the vehicle data, the neural network requires only the following inputs for offline training and subsequent real-time execution, namely: the torque clutch capacity of the eLSD 39, the speed of a left rear wheel of the vehicle 10, the speed of the right rear wheel 17 of the vehicle 10, the axle torque of the vehicle 10, the longitudinal speed of the vehicle 10, the yaw rate of the vehicle 10, the normal force of the left rear wheel 17 of the vehicle 10, and the normal force of the right rear wheel 17 of the vehicle 10. The method 100 then proceeds to block 106.

[0028] In block 106, the neural network model for the eLSD clutch torque estimation is developed using the previously discussed inputs. The structure of the neural network can be based on the complexity of the data and the physical phenomena. The features (i.e., the inputs to the neural network) significantly impact the output (i.e., the eLSD clutch torque estimation). In block 106, the appropriate type of neural network is evaluated and selected (e.g., feedforward, feedback). In block 106, the size and complexity of the neural network are determined. Specifically, the minimum number of layers and neurons that can represent the data is selected. As an example, the neural network can contain two hidden layers and a switch activation function. Each layer can have thirteen neurons. After developing the neural network, the neural network can be trained offline.As an example, two hundred twenty-five vehicle data sets can be used to train the neural network. The vehicle data can be collected in many different driving scenarios while the vehicle 10 is operated at surface friction coefficients from zero to one. The training data can be divided into training data, validation data, and test data with five-fold cross-validation data. Since sign accuracy is essential for eLSD control, the cost function for training the neural network was adjusted to include the sign of the clutch torque. The cost function of the neural network can be expressed as follows: W‖TC−T^C‖2+‖sign(TC)−tanh(T^C100)‖2 where: W is the weight of the cost function; T C is the grid output coupling torque; and T C is the ground truth clutch torque.

[0029] The neural network therefore uses the vehicle data as inputs and outputs a preliminary eLSD clutch torque. Method 100 then proceeds to block 108.

[0030] In block 108, a constraint model (i.e., clutch torque bounds) is used to limit the clutch torque estimation to a physics-based model during the open state and to a data-driven approach during the locked state, with a seamless transition between the two. To ensure that the clutch torque estimation does not exhibit erratic behavior when the neural network is exposed to information it has never analyzed before, constraint logic is used. The constraint logic differs based on whether the eLSD 39 is in a locked or an open state. When the eLSD 39 is in the open state, the neural network is constrained by an open physics-based estimate, which can be expressed by the following equation: Max=Tc+margin Min=Tc−margin where: Tc is the clutch torque capacity; margin is a predefined margin value; Max is a maximum limit; and Min is a minimum barrier.

[0031] When the eLSD 39 is in the locked state, the locked, physics-based estimator is used to constrain the neural network output. The physics-based model calculates the maximum and minimum possible eLSD clutch torque using the following equations: Tcest_max_final=Reff(Fzr+Fzl) Tcest_min_final=−Reff(Fzr+Fzl) Max=Min(Tcest_max, Tc) Min=Max(Tcest_max, Tc) where: R eff is the effective rolling radius of the tire; F zl is the normal force of the left rear wheel; F zr is the normal force of the right rear wheel; Max is a maximum limit; Min is a minimum barrier. T cis the clutch torque capacity; T cest_max is a maximum torque factor; and T cest_min_final is a minimum torque factor.

[0032] Therefore, the minimum and maximum limits of the clutch torque constraints are each a function of the clutch torque capacity of clutch 43 of the eLSD 39. Next, the controller 34 determines whether the preliminary eLSD clutch torque determined using the neural network is outside the eLSD clutch torque constraints. If the preliminary eLSD clutch torque determined by the neural network is outside the clutch torque constraints, the controller 34 adjusts the preliminary eLSD clutch torque using the clutch torque constraints to determine a constrained eLSD clutch torque and ultimately the final clutch torque of the eLSD 39. For example, if the preliminary eLSD clutch torque is above the maximum limit, the constrained eLSD clutch torque is set as the maximum limit.If the preliminary eLSD clutch torque is below the minimum threshold, the restricted eLSD clutch torque is set as the minimum threshold. If the preliminary eLSD clutch torque is within the clutch threshold, no adjustment is necessary. Method 100 then proceeds to block 110.

[0033] In block 110, a digital low-pass filter is used to ensure that the final signal sent to the eLSD control algorithm does not contain a significant amount of noise. The selected coefficient of the digital low-pass filter must be chosen in such a way as to reduce noise by a sufficient amount, but also not cause a significant delay in the estimation. Optimizations using algorithms such as pattern search have shown to provide the best results for calculating the filter coefficients. Additionally, discontinuities in the clutch torque estimation signal may exist between the open and locked limiting schemes. The eLSD 39 is in a transient state when the eLSD switches between the locked state and the open state.When the eLSD switches between the locked and open states, heavy filtering is added for a calibratable period of time. The controller 34 thus determines whether the eLSD 39 is in the transient state. If the eLSD 39 is in the transient state, the controller 34 sets the time constant of the digital low-pass filter to a first predetermined value. If the eLSD 39 is not in the transient state, the controller 34 sets the time constant of the digital low-pass filter to a second predetermined value. The first predetermined value is greater than the second predetermined value to filter out the interruptions in the clutch torque estimation signal when the eLSD 39 is in the transient state. The digital low-pass filter then outputs the final clutch torque of the eLSD 39.The controller 34 may then instruct the vehicle 10 in real time to execute a control action taking into account the final clutch torque of the eLSD 39. For example, the controller 34 may instruct the eLSD 39 in real time to apply the final clutch torque to the clutch 43 of the eLSD 49.

Claims

[1] A method for real-time determination of a clutch torque of an electronic limited-slip differential (eLSD) (39), comprising: Real-time receipt of vehicle data, wherein the vehicle data includes a torque request; Determining a preliminary eLSD clutch torque using a neural network and the vehicle data; Determining eLSD clutch torque bounds using a physics-based model; Determine whether the preliminary eLSD clutch torque is outside the eLSD clutch torque limits; in response to determining that the preliminary eLSD clutch torque is outside the clutch torque limits of the eLSD (39), adjusting the preliminary eLSD clutch torque using the clutch torque limits to determine a final eLSD clutch torque; and Commanding the eLSD (39) in real time to apply the final clutch torque to the clutch (43) of the eLSD (39). [2] The method of claim 1, further comprising filtering the final clutch torque using a digital filter, the digital filter having a time constant. [3] The method of claim 2, further comprising: Determining whether the eLSD (39) is in a transient state, wherein the eLSD (39) is in a transient state when the eLSD (39) switches between a locked state and an open state; and in response to determining that the eLSD (39) is in the transient state, setting the time constant to a first predetermined value. [4] The method of claim 3, further comprising, in response to determining that the eLSD (39) is not in the transient state, setting the time constant to a second predetermined value, wherein the first predetermined value is greater than the second predetermined value. [5] The method of claim 4, wherein the vehicle data includes a clutch torque capacity of the eLSD (39) of a vehicle (10), a speed of a left wheel of the vehicle (10), a speed of a right wheel of the vehicle (10), an axle torque of the vehicle (10), a longitudinal speed of the vehicle (10), a yaw rate of the vehicle (10), a normal force of the left wheel of the vehicle (10), and a normal force of the right wheel of the vehicle (10), and the method further comprises determining the preliminary eLSD clutch torque using the neural network and exclusively the clutch torque capacity of the eLSD (39) of the vehicle (10), the speed of the left wheel of the vehicle (10), the speed of the right wheel of the vehicle (10), the axle torque of the vehicle (10), the longitudinal speed of the vehicle (10), the yaw rate of the vehicle (10),the normal force of the left wheel of the vehicle (10) and the normal force of the right wheel of the vehicle (10)., [6] The method of claim 5, wherein the eLSD clutch torque limits comprise a maximum limit and a minimum limit, and the maximum limit and the minimum limit are each a function of the eLSD clutch torque capacity. [7] The method of claim 6, wherein the neural network is trained offline.

Images