System and method for peak load prediction

Abstract

A method of predicting a peak load in response to a test scenario comprises: for each of a plurality of configurations of a vehicle 104, obtaining a peak load value for that configuration of the vehic

Description

TECHNICAL FIELD The present disclosure relates to a method and system for predicting peak loads in a vehicle. Aspects of the invention relate to a method, a system and a computer program product for predicting a peak load for a vehicle configuration in response to a test scenario. BACKGROUND New vehicle designs are subject to extensive validation and testing during development. Historically, testing may have been performed using one or more prototype vehicles to allow measurements and experiments to be performed on a physical vehicle. However, such one off prototype vehicles are expensive to produce. Furthermore, even minor changes during development may require significant changes to the prototype vehicle increasing costs and extending development time. In order to reduce the costs and time involved in developing new vehicle designs, significant effort has been placed into simulating aspects of a vehicle for virtual validation and testing of the design. Peak, or extreme, loads experienced by components of a vehicle may have a significant impact on durability of the components and of the vehicle as a whole. Historically, such peak loads, for example as experienced by suspension components due to poor road surfaces, e.g. potholes, are determined though real world testing of a vehicle. However, predicting such peak loads at an early stage of a vehicle design process and / or without access to a physical prototype may be difficult. It is an aim of the present invention to address one or more of the disadvantages associated with the prior art. SUMMARY OF THE INVENTION Aspects and embodiments of the invention provide a method, system and computer program product as claimed in the appended claims. According to an aspect of the invention, there is provided a method of predicting a peak load experienced in a vehicle in response to a test scenario based on a machine learning algorithm trained to predict the peak load based on one or more vehicle configuration parameters. According to an aspect of the invention, there is provided a method of predicting a peak load in response to a test scenario, the method comprising for each of a plurality of configurations of a vehicle, obtaining a peak load value for that configuration of the vehicle in response to the test scenario, wherein the test scenario comprises applying a predefined input to the vehicle to cause a peak load value to be generated at a component of the vehicle, generating a training data set comprising input / output pairs of a plurality of parameters defining a vehicle configuration and a corresponding peak load value, training a machine learning model based on the training data set, inputting a plurality of parameters defining a modified vehicle configuration to the trained machine learning model to generate a predicted peak load value for the modified vehicle configuration. Advantageously, time series fatigue loads in a configuration of a vehicle that has not been simulated can be generated based on a predicted frequency response function (FRF) for the vehicle. A machine learning algorithm can be trained based on a number of simulation runs covering ranges of interest for vehicle design parameters, such as spring rate, damping rate, tyre size, etc. Once trained, the machine learning algorithm has been found capable of accurately predicting an FSF for modified configurations of the vehicle in significantly less time than would be required to perform a multibody simulation for that configuration, allowing more efficient exploration of the design space. In embodiments, the predefined input comprises an input displacement value or a input load value applied to a further component of the vehicle. In embodiments, the method further comprises determining a design parameter of a component of the vehicle based on the predicted peak load value. Advantageously, the result of the fatigue analysis based on the output of the machine learning algorithm allows for selection of design parameters for the vehicle, such as spring rate, damper rate, etc. to meet a particular fatigue life requirement, with lower processing and engineer time requirements than full simulation. In embodiments, obtaining the peak load value comprises performing a multibody simulation of a corresponding vehicle configuration in response to the predefined input. Advantageously, the machine learning algorithm may be trained on frequency response functions determined from simulation of a relatively small number of vehicle configurations, avoiding the need to test multiple physical configurations of a vehicle. In embodiments, the plurality of parameters defining the vehicle configurations comprises an operating status of one or more active suspension components of the vehicle configuration. Advantageously, the machine learning algorithm may be operable to predict the effect of active suspension components on the FRF, allowing for complex scenarios including the response of the active systems to input displacements / loads to be taken into account. In embodiments, the plurality of parameters defining the vehicle configuration comprises one or more of: a ride height; a spring rate; one or more tyre parameters; a damping rate; a vehicle mass; an unsprung mass; and a location of centre of gravity. Advantageously, the machine learning model is able to take account of a large range of vehicle design parameters when training to predict the FRF, allowing the method to be used to predict the effect of changes in these design parameters. In embodiments, test scenario comprises a kerb drive over event. Advantageously, the method is able to predict responses to standard test scenarios such as kerb drive over events, allowing easy incorporation into current design processes and comparison / reuse of existing test and simulation data. In embodiments, the peak load value comprises a load value associated with a suspension component of the vehicle. Advantageously, the machine learning model may be trained using predicted load data for specific components or hard points in the suspension system to allow the effects at these points to be predicted. In embodiments, performing a multibody system simulation comprises simulating a response of a multibody system model in response to an input displacement value. Advantageously, the simulation data may be generated by simulating response to a number of impulse events which may be relatively simple simulations requiring less processing time / power than more complicated scenarios. In embodiments, training the machine learning model based on the training set comprises inputting the parameters defining the vehicle configuration from one or more of the input / output pairs of the training data set to obtain a predicted peak load value, characterizing an error between the predicted peak load value and the predicted peak load value corresponding to the parameters defining the vehicle configuration of an input / output pair, using an optimization algorithm to update weights of the machine learning algorithm based on the characterized error. According to a further aspect of the invention, there is provided a system comprising at least one processor, and a memory coupled to the at least one processor, the memory storing instructions that when executed by the at least one processor cause the system to implement a method as described above. According to another aspect of the invention, there is provided a computer program product comprising computer program instructions that when executed on a processor implements a method as described above. Within the scope of this application it is expressly intended that the various aspects, embodiments, examples and alternatives set out in the preceding paragraphs, in the claims and / or in the following description and drawings, and in particular the individual features thereof, may be taken independently or in any combination. That is, all embodiments and / or features of any embodiment can be combined in any way and / or combination, unless such features are incompatible. The applicant reserves the right to change any originally filed claim or file any new claim accordingly, including the right to amend any originally filed claim to depend from and / or incorporate any feature of any other claim although not originally claimed in that manner. BRIEF DESCRIPTION OF THE DRAWINGS One or more embodiments of the invention will now be described, byway of example only, with reference to the accompanying drawings, in which: Figure 1 illustrates a computer implemented method according to an embodiment of the present invention; Figure 2 illustrates a method of according to embodiments of the invention; Figure 3 shows a method of training a machine learning model to generate diagnostic signatures according to embodiments of the invention; Figure 4 shows a method of generating a training data set according to embodiments of the invention; Figure 5 illustrates a system suitable for performing the method of Figure 2 according to embodiments of the invention. DETAILED DESCRIPTION New vehicle designs are subject to extensive validation and testing during development. Historically, testing may have been performed using one or more prototype vehicles to allow measurements and experiments to be performed on a physical vehicle. However, such one off prototype vehicles are expensive to produce. Furthermore, even minor changes during development may require significant changes to the prototype vehicle increasing costs and extending development time. In order to reduce the costs and time involved in developing new vehicle designs, significant effort has been placed into simulating aspects of a vehicle for virtual validation and testing of the design. Peak, or extreme, loads experienced by components of a vehicle may have a significant impact on durability of the components and of the vehicle as a whole. Historically, such peak loads, for example as experienced by suspension components due to poor road surfaces, e.g. potholes, are determined though real world testing of a vehicle. For example, a kerb roll over scenario is a standard test scenario used to obtain road load data including peak loads experienced by a vehicle, that may be used when designing vehicle components to ensure that they are sufficiently robust. One approach for modelling peak loads is using a combination of a simulation model, such as a multibody simulation model and road load data measured using a physical prototype vehicle. To allow for changes in vehicle tune, or configuration, a linear correction factor may be applied to the data. However, while this approach provides useful results for small changes in vehicle configuration, it has been found to be less accurate for large vehicle changes. Furthermore, while it is not necessary to have a prototype vehicle in a particular state of tune or configuration to be subject to physical testing, this approach still relies on a physical vehicle in order to measure the road load data. In future, it may be desired to be able to determine loads, including peak loads, placed experienced by components in vehicle designs in a range of vehicle configurations without requiring a physical prototype. In particular, it may be beneficial to be able to predict the loads experienced by suspension components of the vehicle in response to predefined test scenarios. Embodiments of the invention provide a method of predicting peak load data for use in testing a vehicle design or configuration without requiring a physical prototype of the design. FIG. 1 illustrates a method of predicting peak loads for m modified vehicle configurations according to embodiments of the invention. According to the method illustrated in FIG. 1, a plurality of configurations of the vehicle are defined at block 104 with configuration parameters spanning a range of interest. A simulation model, such as a multibody simulation model, is used to model 106 a vehicle design for the plurality of configurations of the vehicle. For each configuration of the vehicle, the simulation is used to determine 108 a peak load experienced at one or more predefined locations, or components, of the vehicle being modelled in response to a particular test scenario. For example, the peak load value may be indicative of a peak load experienced at a hardpoint or component in a suspension of the vehicle in response to a predefined input, such as a tyre input displacement value. The simulation of the vehicle model is performed for each of the plurality of configurations to generate a data set comprising input configuration parameter values defining the particular configuration of the vehicle model being simulated, and an output value corresponding to the determined peak load value for that configuration. According to embodiments, the resultant data set is then divided into a training data set for training a machine learning model and a validation data set to be used to validate the machine learning model. In embodiments, the machine learning model may comprise an artificial neural network (ANN), such as a feedforward ANN model. Relevant design parameters may include a ride height of the vehicle; a spring rate of a suspension of the vehicle; one or more tyre parameters such as a size or pressure associated with the tyre; a damping rate of the vehicle suspension; a vehicle mass; an unsprung mass of the vehicle; a location of centre of gravity (CoG), etc. According to embodiments vehicle parameters used as input to define a vehicle configuration may comprise one or more of: Damper Scaling; Knee Velocity; High Speed Compression; Low Speed Damping; High Speed Rebound; Spring Rate; Tyre Property File / Size; CoG Height; CoG Fore / Aft; Vehicle Mass (Sprung); Unsprung Mass; Rebound gap; Rebound spring scale; Spring Aid Scale by energy; Spring aid gap; Axle Type; stiffness of one or more bushings in X, Y and / or Z directions; Lower Control Arm Stiffness; Toe Link Stiffness; Upper Control Arm Stiffness; and Knuckle Stiffness. In embodiments, for each of the plurality of configurations of the vehicle, the peak load value may be modelled based on a predefined input displacement event, for example tyre displacements, as might be experienced in a kerb roll-over scenario. The machine learning model is then trained at block 110 using a suitable optimization algorithm (e.g., stochastic gradient descent) by feeding it input parameters from the training set and comparing its predictions with corresponding target output values, i.e., actual peak load values as determined by the simulation. During training, prediction accuracy of the machine learning model may be evaluated using the validation data set to assess the prediction accuracy of the trained ANN, for example using metrics like mean squared error (MSE) between predicted and actual peak load values from the validation data set. Once the machine learning model has been trained to accurately predict a peak load value for an input vehicle configuration, the model may be used to predict a peak load value for new vehicle configurations as may be required for testing of a vehicle design space. This allows a new peak load value to be predicted fora vehicle configuration that has not been modelled using multibody simulation of the vehicle design. As illustrated in FIG. 1, a new, or modified, vehicle configuration comprising a plurality of parameters defining the vehicle configuration is obtained at block 116. The obtained modified vehicle configuration is applied to the trained machine learning model at block 112 to generate a predicted peak loads value corresponding to the modified vehicle configuration which is output at block 114. FIG. 2 illustrates a method 200 of generating a machine learning model that may be used in the prediction of peak load values experienced by a vehicle in response to a particular test scenario according to embodiments. According to the method 200 of FIG. 2, in block 202 for each of a plurality of configurations of a vehicle, a peak load value associated with a configuration of a vehicle defining a load value generated in a component of the vehicle when a predefined input corresponding to the test scenario is applied to vehicle. For example, a plurality of peak load values each corresponding to one of a plurality of vehicle configurations may be obtained through multibody simulation of a model of the vehicle as discussed above. In block 204, a training data set comprising input / output pairs of a plurality parameters defining a vehicle configuration and a corresponding peak load value is generated, for example as discussed above. In some embodiments, a validation data set may also be generated from the obtained peak load value data, wherein the validation data set is used to determine that the machine learning model has been trained to accurately predict a peak load value for an input configuration. In block 206, the machine learning model, such as an ANN, trained based on the training data set, as discussed above, to predict a peak load value based on an input vehicle configuration. Once trained, the machine learning model may be used to predict peak load values for new, or modified, vehicle configurations. For example, the trained model may be used to predict changes to peak loads in response to specification of stiffer suspension springs and / or dampers or changes in tyre profile. The 6 predicted peak load value may then be used in analysis of a vehicle configuration to allow determination of loads experienced by different components of the vehicle for a particular configuration. In embodiments, one or more design parameters of the vehicle may be automatically determined based on the analysis to ensure that the component is specified to be sufficiently robust to expected road load events. FIG. 3 illustrates a method 300 of generating a training data set for training a machine learning algorithm, such as a neural network, for predicting a peak load value for a configuration of the vehicle according to embodiments. According to the method 300 of FIG. 3, a series of test configurations are defined at method 300 each test configuration defining a particular vehicle configuration to be tested. For example, each vehicle configuration may comprise a number of vehicle parameters defining particular aspects of the vehicle design such as suspension spring rates, damping rates, suspension geometry, wheel diameter, tyre size, tyre pressure, vehicle mass, unsprung mass, vehicle ride height, a position of the centre of gravity of the vehicle, etc. The defined test configurations are designed to cover a range of possible vehicle parameters to “envelope” the available design space for the vehicle. For each test configuration, a multibody simulation (MBS) is performed at block 302. For example, the vehicle configuration parameters can be applied to a model of the vehicle and simulated using a commercially available multibody simulation software package such as Simscape Multibody by Mathworks®. As an output of the multibody simulation a peak load value is determined that corresponds to the peak load calculated to be exerted at a particular hardpoint or component in the simulation model in response to a particular event forming part of a test scenario, such as a tyre displacement or road load, for example the peak load may correspond to a maximum load that would be measured at a top of a suspension tower on the vehicle. In embodiments, the input parameters defining the configuration of the vehicle may define a presence and / or operating status of one or more active suspension components, wherein the machine learning model is trained to take account of the operation of the one or more active suspension components when predicting a peak load value for the vehicle configuration. The peak load value determined in block 302 is then combined with the vehicle parameters defining the vehicle configuration forthat simulation to generate an input / output pair 304. An input / output pair 304 is generated for each of the vehicle test configurations defined at method 300, and the input / output pairs are stored to generate a training data set 306 comprising a plurality of vehicle configurations and corresponding peak load values. FIG. 4 illustrates a method 400 according to an embodiment of the present invention to train first neural network for use in predicting peak load values. As illustrated in FIG 4, the training data set 402 is provided comprising training example input / output pairs 304 as generated by the method 300 of FIG. 3, each input / output pair 304 associating a parameters defining a vehicle configuration with a corresponding peak load value, for example determined through multibody simulation of that vehicle configuration. For each input / output pair of the training data set 402, the input value of an input / output pair, i.e. parameters defining a vehicle configuration, are input to a machine learning model 404 which outputs a predicted peak load value. The predicted peak load value is compared at comparator 406 with an expected peak load value 408 corresponding to the output value of the input / output pair to determine an error between the predicted and expected peak load values. The error is provided to a optimization algorithm 410 which adjusts parameters of the machine learning model 404. The method of FIG. 4 is then iterated until the machine learning model learns to accurately predict peak load values based on input vehicle configuration parameters. Certain methods and systems as described herein may be implemented by one or more processors that processes program code that is retrieved from a non-transitory storage medium. FIG. 5 shows an example 502 of a device comprising a computer-readable storage medium 506 coupled to at least one processor 504. The computer-readable media 506 can be any media that can contain, store, or maintain programs and data for use by or in connection with an instruction execution system. Computer-readable media can comprise any one of many physical media such as, for example, electronic, magnetic, optical, electromagnetic, or semiconductor media. More specific examples of suitable machine-readable media include, but are not limited to, a hard drive, a random-access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory, ora portable disc. In FIG. 5, the computer-readable storage medium comprises program code to perform a method corresponding to the embodiment shown in FIG. 2, that is: obtaining 202 a peak load value for each of a plurality of configurations of the vehicle; generating 204 a training data set comprising input / output pairs of a plurality of parameters defining a vehicle configuration and a corresponding peak load values; and training 206 a machine learning model based on the training data set. In embodiments, the computer-readable storage medium may comprise program code to perform a method corresponding to the embodiment of any of FIG. 1, FIG. 2, FIG. 3, and / or FIG. 4. It will be appreciated that various changes and modifications can be made to the present invention without departing from the scope of the present application.

Images