DE19511863C2 - Automatic transmission control with adaptive shift behavior - Google Patents

Abstract

The transmission has a neuronal network (1) to whose input side are fed parameters indicative of the operating mode and which outputs a signal defining the selected gear. The operating mode indicating parameters include at least the previously selected gear (G(t-1)), the vehicle speed (v(t)) and a parameter (DK(t)) influencing the engine torque. The indicative parameters can also include the time variation of the vehicle speed and/or of the parameter influencing the torque, a driver induced gear selection, driver type indicating parameters and/or road characteristic parameters.

Description

The invention relates to automatic transmission control with operating adaptive switching behavior according to the Oberbe handle of claim 1 or 2, as is particularly the case in motor vehicles witness is used.

There are already automatic transmission controls for motor vehicles knows, for which different gearboxes to adapt the driving style Switching modes are provided, such as sporty or economical Driving style, mountain and / or winter trip. Each switching mode are corresponding switching characteristics in these known systems assigned, between which each by hand or automatically vice versa can be switched or interpolated. Suitable difference Application criteria for this are either by means of algorithms or again explicitly determined via maps. From the patent US 5,089,963 is such an automatic transmission control known for a motor vehicle in which the switching behavior over suitably specified switching maps between an operation wise in normal traffic and a mode of operation in heavy Traffic can be switched. For automatic adjustment of the switching ver suitable for the current driving style Holding is, among other things, a unit with associative logic intended. This unit is diverse as input variables  Indications of driving operations supplied, e.g. B. the throttle valve opening angle, vehicle speed, engine ge speed and the steering angle as well as the time derivation against these sizes. The associative logic unit generates in ei a tutorial pattern, each one of the two gears switching modes can be assigned. If subsequently in driving mode certain input operation values indicative of driving operation are available, the associative logic unit decides which of the two given switching behavior best fits this condition, if the driver has not made his own choice in this regard. Subsequent units then control the transmission in accordance with the predefined shift characteristics for the selected transmission switching mode.

The use is in the published patent application DE 42 09 150 A1 a neural network within a control system for described a motor vehicle in which the neural network for automatic estimation of the current vehicle weight on the basis of driving operation data supplied on the input side.

In the company contribution "The transmission thinks with", Daimler Benz HighTechReport 2/1995, page 61 is generally the use development of a learnable neural network in an automat Transmission control of an automobile specified to the transmission switching behavior to the driving situation and the driving style itself to be able to adapt daily. In published patent application EP 0 645 559 A1 is an automatic transmission control for a more detailed Motor vehicle described in which to achieve a be mode-adaptive switching behavior a neural network on the output side a signal determining the switching stage depending of input variables that affect an engine torque send size, brake pressure, steering angle, vehicle ge dizziness, the current as well as the previously engaged Gangstu fe, the driving distance and the road friction coefficient as well if necessary, include the temporal change in these quantities. Everyone Input size is an input of the neural network, i. H. a neuron is assigned to an input layer of the same. The neural network contains at least one intermediate layer and  an output layer consisting of a single neuron, which emits the switching stage determining signal.

The invention is a technical problem of providing based on an automatic transmission control, its shifting behavior comparatively sensitive to each one Adapts operating mode.

This problem is solved by automatic transmission control with the Features of claim 1 or 2 solved.

The use of the neural network as a switching unit determining unit the specification of a limited number of predefined gears shift modes or gear shift characteristics are no longer required Lich, because the desired, sensitively coordinated switching behavior suitable from the neural network in a training phase input, operating mode-indicative input variables learned becomes. Every user becomes an automatic transmission after a certain time be with switchgear specially adapted to its operating mode keep provided by the neural network adapts its behavior according to user behavior. The output signal of the neural network is used directly than the switching stage determining signal. Opposite a ver equally rough adaptation to different driver types Specification of a finite number of switching modes between which can be switched, the use of the neural allows Network in a motor vehicle automatic transmission the setting individual switching behavior for the respective Driver.

Accuracy increases the control after on saying 1 from that for each input and / or output variable a separate connection for each digit of a given number lens systems is provided in which the values of these quantities are encoded.

In the control according to claim 2 is one for each switching stage separate neural network provided, all of which the same input information is supplied and their Output signal shows how cheap the choice of the supplied  audible switching stage is currently. Via a so-called winner Linking these output signals then becomes the one to be set Switching level determined.

In the automatic transmission tax developed according to claim 3 tion are the features of claims 1 and 2 advantageous combi kidney.

A training according to claim 4 enables dynamics in to bring in the static neural network as such.

An increase in accuracy can be achieved by using it seized transmission interventions by the driver and / or by the driver type-indicative and / or roadway-indicative quantities as achieve further input variables in the neural network, such as specified in claim 5.

Preferred embodiments of the invention are in the drawing voltage and are described below. Here demonstrate:

Fig. 1 is a schematic block diagram of a uniform neural network as a switching stage determining unit of an automatic transmission with driving style adaptive switching behavior in a motor vehicle and

Fig. 2 is a schematic block diagram of a multi-part neural network as a shift stage determining unit for an automatic transmission with shifting behavior adaptive behavior in a motor vehicle.

The neural networks ( 1 , 2 ) shown in the two figures work periodically according to a predefined system cycle and have the function of engaging with each system cycle the most favorable shift stage for the current driving situation, ie the correct gear, the driving situation being fed based on the input side , driving operation-indicative quantities is recorded. These input variables, which are symbolically combined in the example of FIG. 2 to form an input variable set (E), include in particular the current vehicle speed (v (t)) at the cycle time point (t) and the position (DK (t)) of the throttle valve as a torque influencing variable at this time (t). In addition, the neural network ( 1 , 2 ) is supplied on the input side with the information about the gear (G (t-1)) previously engaged, ie determined in the previous time cycle (t-1). In order to simulate a dynamic network behavior of the static neural network ( 1 , 2 ), the positions (DK (t-1), DK (t-2)) of the throttle valve in the last (t-1) are used as further input variables. and the penultimate time cycle (t-2) and the vehicle speeds (v (t-1), v (t-2)) at these times are used as input variables, which takes into account the time profile of the vehicle speed (v) and the throttle valve position ( DK) corresponds. Alternatively, time derivatives of these variables (v, DK) can be calculated in a preprocessing stage and used as additional input variables.

Depending on the further information available in the vehicle electronics, suitable further input variables for the neural network ( 1 , 2 ) can be taken into account, as indicated by dotted lines in FIG. 1. In particular, this is information about another gear desired by the driver, e.g. B. recognized by a driver-induced shift request in a higher or lower gear, information about the Fahrbahnver run, z. B. the slope and / or curvature of the road, as well as driver-specific information, ie taking into account a rather calm or rather sporty driving style. This driver type information can either be done manually, e.g. B. be specified via a corresponding switching element, or the neural network ( 1 , 2 ) input variables such as steering wheel angle and lateral acceleration are supplied, from which this can determine the driver type or the driver type is used by a dedicated algorithm for gradual Driving style classification determined.

At its output, the network ( 1 , 2 ) emits a signal which specifies the switching stage (G (t)) to be currently set. Alternatively, the output signal can also consist of specifying a required switching stage change upwards or downwards. By using the neural network ( 1 , 2 ), the shift behavior of the automatic transmission is learned on the basis of suitable measurement data in a training phase, as is known to the person skilled in the art for neural networks and therefore requires no further description here. The specification of fixed shift modes and the shift characteristic curves assigned to them is therefore not necessary, which simplifies the design of the transmission, since not all aspects or special cases need to be explicitly examined. Rather, it is sufficient to record the corresponding data and present it to the neural network ( 1 , 2 ) for learning. Another advantage of using the neural network ( 1 , 2 ) is its ability to continue learning. If the driver does not agree with the behavior of the transmission, he intervenes, for example, using the gear selector. The neural network ( 1 , 2 ) registers such interventions and uses them to further improve the adaptation of the transmission shift behavior to the driver behavior. This makes it possible to deliver an automatic transmission to the customer, in which the neural network has pre-learned the basic shift functions and which can continue to adapt to the respective driver by continuing to learn from delivery of the vehicle, while maintaining the shift behavior that has already been learned.

The two neural networks ( 1 , 2 ) shown differ in the structure described below. The neural network ( 1 ) of Fig. 1 is structured as a uniform network with the required number of inputs and the switching stage-determining output. The suitable network structure in detail can be realized in a conventional manner by a person skilled in the art with knowledge of the above-mentioned functional requirements, which need not be explained further here.

The neural network ( 2 ) shown in FIG. 2 consists of several individual networks (NN1, NN2, NN3, NN4, ...) arranged in parallel, the number of which corresponds to that of the shift stages present in the automatic transmission. With three switching stages, the three individual networks (NN1, NN2, NN3) shown with solid lines are available. If the transmission has a fourth shift stage, a fourth individual network (NN4) is added, as shown in dotted lines. The same applies to any additional switching stages in the transmission. All individual networks (NN1 to NN4) are supplied with the same set of input variables (E) as described in more detail above. Each one network (NN1 to NN4) is assigned to a switching stage and indicates with its output signal how cheap the choice of the assigned switching stage is in the operating conditions detected at the momentary driving conditions. The outputs of all individual networks (NN1 to NN4) are brought together at a connection point ( 3 ) and are subjected to a so-called winner formation, with the aid of which it is decided which current switching stage (G (t)) is to be set. A corresponding output signal is emitted from the node ( 3 ) as the output signal of the entire neural network ( 2 ). The structure of the individual networks (NN1 to NN4) can in turn be implemented by the person skilled in the art without any problems on the basis of the specified functional requirements and therefore requires no further discussion here.

In order to increase the accuracy, the inputs and / or the outputs of the respective network ( 1 , 2 ) or of the parallel individual networks (NN1 to NN4) can be coded in any number system and put in one input or output for each number of the number system. For example, three inputs can be used for the current vehicle speed (v (t)) in the decimal system, namely one for the hundreds, one for the tens and one for the ones.

The above description of two neural networks ( 1 , 2 ) which can be used to adapt the gear shift behavior to the driving mode makes it clear that an automatic transmission control equipped with such a neural network can automatically adapt itself very sensitively and learnable to a particular driving style without the need to specify fixed switching characteristics. Of course, the invention can also be used for automatic transmission controls with multiple switching stages used outside of a motor vehicle, for which purpose the appropriate input variables are to be selected and the neural network has to be suitably structured.

Claims (5)

1. Automatic transmission control with mode-adaptive shifting behavior, especially for a motor vehicle, with

• 1. a neural network, to the input-side indicative quantities (v (t), DK (t), G (t-1)) are supplied and which outputs a switching stage-determining signal (G (t), where
• 2. the values for the input variables and / or for the output variable of the neural network ( 1 ) are values coded in a predetermined number system,

characterized in that

• 1. each number system location each of the values for the input variables and / or for the output variable is assigned a separate input or output of the neural network.

2. Automatic transmission control with mode-adaptive shifting behavior, with

• 1. a neural network which is supplied with indicative variables (v (t), DK (t), G (t-1)) on the input side and which outputs a signal (G (t)) determining the switching stage,

characterized in that

• 1. the neural network ( 2 ) consists of ne parallel arranged neural individual networks (NN1, NN2, NN3, NN4, ...) for each one gear shift stage, the outputs of the individual networks being brought together in a node ( 3 ), in which they are subjected to a winner formation for generating the switching stage-determining output signal (G (t)).

3. Automatic transmission control according to claim 1, further characterized in that

• 1. the neural network ( 2 ) consists of ne parallel arranged neural individual networks (NN1, NN2, NN3, NN4, ...) for each one gear shift stage, the outputs of the individual networks being brought together in a node ( 3 ), in which they are subjected to a winner formation for generating the switching stage-determining output signal (G (t)).

4. Automatic transmission control according to one of claims 1 to 3 for a motor vehicle, further characterized in that as operating mode-indicative input variables at least the to gear position (G (t-1)), the vehicle speed (v (t)) and a motor torque influencing variable (DK (t)) as well the change in vehicle speed over time and / or serve to influence the motor torque. 5. Automatic transmission control according to claim 4, further characterized in that a driver as additional operating mode-indicative input variables induced switching stage setting, driver-type indicative quantities and / or parameters that are indicative of the course of the roadway.