US 20080306665 A1
A vehicle powertrain modelling system includes a powertrain model (20), a vehicle model (30), a seat model (55), a driver model (60) and a correlation element (50) comprising a neural network. As a result vehicle development is enhanced, using the neural network (50) to correlate modelled shift aspects with previous subjectively obtained ratings.
11. An AMT vehicle transmission control system comprising a transmission shift parameter control element, a vehicle manoeuvre detection element and a controller, wherein the controller is arranged derive control instructions from the detected vehicle manoeuvre to control the shift parameter control element.
12. A system as claimed in
13. A system as claimed in
14. A system as claimed in
15. A vehicle control system comprising an AMT vehicle transmission control system as claimed in
16. A vehicle including a control system as claimed in
17. A method of calibrating a system as claimed in any preceding claim 1 comprising the steps of running the system through one or more predetermined manoeuvre and correlating a modelled rating against a driver subjective rating for the same manoeuvre.
18. A method as claimed in
20. A method of controlling an AMT vehicle transmission comprising detecting a vehicle manoeuvre, deriving shift control instructions for the detected manoeuvre and controlling a shift parameter control element based on the control instructions.
21. A method of developing an AMT vehicle transmission comprising the steps of measuring a shift profile, determining the contributory factors and varying the contributory factors to obtain an improved shift profile.
22. A method as claimed in
23. A vehicle powertrain modelling system comprising a plurality of powertrain model elements and an interaction monitor arranged to monitor interaction between the elements, in which the model elements include a time-deterioration model and the interaction monitor monitors interaction of the elements over time.
24. A vehicle development system comprising a modelling system as claimed in
The invention relates to vehicle transmission shift quality, in particular in relation to automated manual transmission (AMT) shift quality.
In Europe there is considerable growth in the application of automated manual transmissions (AMT) predominantly for reasons of cost and CO2 efficiency. There are a range of such transmissions available, and they can generally be grouped into single clutch systems (AMT-1) and twin path systems (AMT-2). Yet more sophisticated versions are under development. However, the poor shift quality of this “first wave” of manual-based transmissions, that is to say, the performance of the vehicle during gear changes, particularly as perceived by the driver is a significant problem, due to interruption of the propulsive torque. This differs from conventional planetary automatic transmissions in which the propulsive torque remains positive throughout the shift process.
Various attempts have been made to assess AMT shift quality but these have tended to be inappropriately based on corresponding tests for automatic transmissions, which have significantly different considerations and indeed expectations, or based on simplistic metrics.
The invention is set out in the claims.
The invention gives rise to various advantages over known arrangements. The modelling of the various aspects of the vehicle powertrain or transmission, including a driver model, allows a full and meaningful set of data to be obtained relevant to the shift quality and taking into account, for example, the effects on the driver. As a result the development process can be significantly focused and enhanced as a result of the more advanced metrics adopted. It will be appreciated that references to AMT transmissions extends to alternative stepped transmissions such as automated stepped ratio transmissions and other appropriate transmissions. The invention is also applicable to other transient vehicle phenomena for which object/subjective correlation is relevant.
Embodiments of the invention will now be described, by way of example, with reference to the drawings, of which:
In overview the invention provides an improved approach to measurement, modeling and control of shift quality in AMTs. This is based on a more detailed analysis of the factors affecting the quality of the shift both physically and in terms of driver perception. As a result the invention provides an assessment of the subjective shift quality obtained from measured or predicted time-based data. In addition the system can be used to define cause/effect maps which direct changes to the control variables in an off-line environment. In addition, in on-line operation the cause/effect maps can direct changes within the control algorithms. Yet further the system is used to predict shift quality and flag a diagnostic for service remedy if a given threshold is exceeded.
Analysis of shift quality requires many factors to be considered in combination including: longitudinal acceleration, vehicle pitch and acoustic frequencies and time derivatives of these factors.
In each case the technique must account for two key requirements: identification of features of the shift which allow ready mathematical description of a recorded or predicted shift profile, and a means of correlation which accounts for the inevitable interactions between the sensory inputs to the driver in the formulation of opinion.
According to a first approach any deviation from his ideal shift can be assigned a mathematical “cost” which would act to reduce the ideal rating from a value of 10. The shift profile is idealized by piece-wise approximations using the following line portions:
Alternative piece-wise elements may also be considered according to the nature of the metric: sports or a comfort shift as shown in
A cost function comprises weighted factors associated with the magnitude of the deviation from the “ideal” shift for each of the line portions. Additionally, factors relating to the areas marked as area A and B are also included. The units of these areas is velocity (m.s−2×time), therefore the smaller the combined area (A+B), the smaller the loss of associated quantities such as kinetic energy and momentum. Area B would have a higher weighting than area A since if sustained it would relate to a potential condition in which the driver rocked in the seat due to the change in sign of acceleration.
The damping ratio of the oscillatory line portion 5 is taken as a measure of cost for the final portion of the shift.
As can be seen from
An alternative or additional approach is illustrated by the overall correction scheme shown in
In the modelling stage, block 30 represents a vehicle model of any appropriate type representing the dynamics/physical characteristics of the vehicle. In the present embodiment the model 30 has 3 degrees of freedom, X-axis, Z-axis and pitch.
The raw variables affecting shift quality are fed to parameter definition block 40 which converts the data to data representative of shift quality (for example mean acceleration, peak to peak jerk, and a Fast Fourier Transform of the acceleration in the range 0-10 Hz). In order to assess the quality of the shift a correlation is made with a predicted rating (using for example a rating of 1—very bad to 10—very good) based on a subjective rating made by an experienced vehicle tested carrying out the manoeuvre. Any appropriate subjective rating scheme can be used. To achieve the rating correlation, a neural network 50 is employed, which is found to be a particularly efficient manner of obtaining the correlation.
In addition to the vehicle/powertrain inputs, the driver subjectivity can be modelled using a seat model 55 and a multi-element human model generally designated 60 in
The relevant parameters are derived from a dynamic model of the driver. The transformation of these variables would include the following:
In addition conventional parameters relate to variables referenced to the seat rail of the vehicle, where appropriate.
Related metrics would also be derived as dosage metrics associated with Power Spectral Densities (from FFTs) of vibrations in longitudinal and pitch modes.
The above transformations would be applied to the following variables:
The raw variables thus obtained are also fed to the parameter definition block 40 after sensitivity weightings are applied as appropriate at block 70. These weightings may be pre-set or themselves learned in the neural network 50. The parameters are fed to the neural network 50, to obtain correlation with the predicted rating.
Accordingly the method applies the following process:
The effectiveness of the technique arises because the greater the number of direct influences included in the model, the more the effect of extraneous influences (such as brand image etc) will be reduced, but not eliminated. It is inevitable that the same gear-shift (objectively identical) will give different subjective rating when assessed in a two different vehicle platforms.
As a result a Correlation Model is provided in the form of the algorithm which has inputs of system parameters (transforms of variables) and outputs a predicted value of shift quality rating. The model can be used in the vehicle development process once the neural network has been trained in a calibration period to ascertain the effects of changes to the vehicle on shift quality and to tailor the design to improve shift quality.
Although a look up table could be used to store the correlation data, or a regression model adopted, Neural Networks provide a simple and robust mechanism to convert system parameters into a predicted rating involving the following components;
Furthermore neural networks win account for main effects and multiple interactions. Any appropriate Neural network can be utilized and indeed toolboxes exist which may enable the tools to be applied without expert understanding. Appropriate techniques of linear/non-linear regression may be adopted in addition or alternatively.
It will be appreciated that once the system is trained in a calibration phase using correlations against real subjective shift ratings, further development can be carried out for a specific vehicle or class of vehicles using the model alone. Yet further, the system can be implemented in a vehicle control system of the type shown in
The control system includes a vehicle manoeuvre sensor 100 which detects the manoeuvre parameters (gear change, speed, acceleration) and inputs them to a controller 108, for example implemented in the ECU. The controller derives control parameters from a look-up table or neural network as obtained in the calibration/development process and feeds the control parameters to the shift control block 102 which controls the shift to provide the desired shift quality. The controller 108 can further compensate for long term changes in the vehicle, providing diagnostics to trigger maintenance by implementing a control loop with additional sensor/model blocks 104, 106. Block 104 comprises a driver module which can comprise sensors to monitor driver perception (seat rail acceleration for example) and/or a model based on predetermined values derived in the calibration/development stage to assess human model parameters such as neck moment. Block 106 comprises shift parameter sensors (which can alternatively be measured and returned to the control block 108 from the shift parameter block 102) to monitor continuing operation of the vehicle. Based on this feedback the ECU can adjust control parameters over the long term and/or flag up the need for repair or adjustment. In addition to adjusting direct control parameters the ECU can also liaise with other engine components under its control (e.g active suspension) to adjust shift quality.
It will be seen further that, using a trained neural network as a component of the controller block 108, the system can continue to learn about shift quality based on feedback from the driver module 104.
A further possible system providing improved shift quality metrics uses system identification to aid the understanding of transmission shift characteristics and establish dynamic relationships for the vehicle system. Simple techniques for estimating parametric system descriptions given step and impulse response assumptions and data have been applied as a first stage.
The time history for the sprung mass longitudinal acceleration during a transmission shift can be separated into torque disruption, re-application and transient phases, see
The results shown in
By considering the above-mentioned first and second order components of the impulse response, engineering insight can be gained about shift quality and the vehicle. For example, the first order component gives an indication of the shift time constant and statistical measures of power loss during shifts; the second order component can be used to estimate the natural frequency and damping ratio of the driveline (i.e. parameters associated with the shunt phenomenon).
It will be seen that the methodology as described above can be applied in various instances. First of all they can be used for the collection of objective and subjective data which in itself can form basis for useful analysis as well as the foundation for automating the subjective rating process. Off-line simulation, normally undertaken for the development of mechanical and control elements relating to powertrain, drive-line and vehicle systems can be extended according to the invention. In particular the trained neural network or other selective means of correlation between objective and subjective data allows the predictions likely objective assessment of the system or relevant part of the system response allowing appropriate design changes to be quickly identified and implemented in the development stage.
Once design parameters have been arrived at in the simulation phase these can be assessed on dynamic test rigs. Such rigs will be fitted with system elements to simulate certain aspects of the system. For example a real transmission can be driven by an electrical dynanometer with the vehicle load applied to the output of the transmission loaded by another dynanometer.
Yet further, as discussed above, sensitivity studies can be undertaken to establish the relevant effect and interaction between system changes due to wear and degradation during the life of the vehicle. This can be in simulation which will be more straightforward for reasons of practicality but can also be in actual operation of the vehicle as indicated above. The information derived can be applied for two main purposes. First of all the information would be used to guide systems designed to achieve the minimum desired shift quality performance in the most efficient way in terms of weight/cost/package envelope and so forth. This can be based on the system specification for the vehicle, and taken into account controllable chassis devices and powertrain devices for example. As a result a tolerance can be built in to accommodate likely system degradation. Secondly the derived information can be used as stored information to guide the control systems of the vehicle to adapt control parameters to maintain shift quality during the life of the vehicle as mechanical elements of the chassis and powertrain degrade. In other words dynamic adaptation can be applied as an alternative to introducing tolerances ab initio.
It will be appreciated that the invention can be applied to any appropriate AMT scheme, and used to calibrate and develop vehicles and vehicle transmissions, based on a shift profile appropriate to the vehicle or market (eg smooth or sporty profiles as appropriate). The invention extends not only to the features specified in the claims that follow but to any other novel feature, method or technique or combination disclosed herein.