JP4769111B2 - Automatic transmission simulation apparatus and program thereof - Google Patents

Description

  The present invention relates to an automatic transmission simulation apparatus for performing a simulation for obtaining a shift control value in an automatic transmission, and a program therefor.

  Conventionally, an automatic transmission is mounted on many vehicles, and the speed change operation in the automatic transmission is controlled so that the output shaft torque and the turbine rotational speed change as intended. And in order to obtain this control value, it is necessary to perform an actual vehicle running test under various conditions.

  However, since such a running test requires a large amount of time and work, a simulation using an automatic transmission model has been performed as an alternative.

For example, Patent Document 1, the torque ratio of the clutch-to-clutch shift, the input torque T _in, the torque capacity T ₁ of the engagement side frictional element, the torque capacity of the disengagement side frictional element T _2, T _in the T ₁ In accordance with the equation of motion T _in = AT ₁ + BT ₂ + C set by the correction term C, A representing the torque ratio between A and T _in representing T, and T ₂ balanced in the gear stage before the shift. It has been shown to determine the hydraulic command for either one of the elements.

  Moreover, in patent document 2, the feedforward control amount is calculated | required with the inverse model of a common rail apparatus, and the parameter of an inverse model is updated by performing a system identification sequentially in that case.

  In Non-Patent Document 1, the trajectory tracking performance of a robot is improved by repetitively obtaining a command value correction amount by paying attention to system dynamics and kinematics.

  Patent Document 3 discloses that the trajectory tracking performance of a robot is improved by using a learning control device that simulates a biological neural circuit. Furthermore, Patent Document 4 shows that a control constant of a controller that satisfies a plurality of use items is determined by a sequential quadratic programming method.

JP 2004-340287 A JP 2004-108216 A JP 2000-203891 A JP-A-8-22307 Masaru Uchiyama, Kazuaki Asai, Tsuyoshi Dainaga "Proceedings of the Society of Instrument and Control Engineers", vol. 33, no. 8, pp. 858-860 “Inverse kinematics solution of flexible manipulator using learning algorithm”

  In Patent Document 1, it is necessary to obtain in advance a hydraulic command for either the disengagement side or the engagement side friction element at the time of shifting. Further, the response to the performance variation is performed by on-board feedback control and learning control (these are included in the correction term), and no attempt is made to cope with the performance variation by feedforward control.

  Moreover, in patent document 2, in order to correct | amend the parameter of an inverse model one by one, an actual apparatus is required and there exists a problem that only the command value corresponding to a specific individual can be generated.

  Further, in Non-Patent Document 1, Patent Documents 3 and 4, etc., there is no description about efficiently deriving a control value for feedforward control that is resistant to performance variations.

The present invention is a shift simulation device for an automatic transmission using a model, and reduces an error between a simulation result using the model and a target for an output shaft torque waveform and a turbine angular acceleration waveform of the automatic transmission. The simulation result is obtained by repeating the learning calculation as described above, and calculating the engagement control command value for each engagement friction element for torque transmission of the automatic transmission and the feedforward command value for the input turbine torque command to the automatic transmission. The model is derived based on repeated calculations, and two models are prepared: a reference model in which design specification values are set and a variation model that includes performance variations, and the command values are corrected for both models. repeated learning operation by, to the error of the simulation results of both models with the predetermined or less And obtaining the command value.

  Further, it is preferable that the learning gain can be adjusted independently between the reference model and the variation model.

  In addition, it is preferable to express the gear train system of the automatic transmission as a multi-input multi-output system, and obtain the correction amount of the feedforward command value from the pseudo inverse matrix of the matrix representing this system.

Further, the present invention is a shift simulation program for an automatic transmission that causes a computer to execute a shift simulation of an automatic transmission using a model. The computer is configured to output an output shaft torque waveform and a turbine angular acceleration waveform of the automatic transmission. The simulation results using the model and the learning calculation are repeatedly performed so as to reduce the error between the target, the engagement control command value to each engagement friction element for automatic transmission torque transmission, and the automatic transmission A feedforward command value for an input turbine torque command to the machine is repeatedly derived based on the simulation result, and the model includes a reference model in which design specification values are set, and performance variations Prepare two variable models and modify the command values for both models. It was repeated a learning calculation by, the error of the simulation results of both models characterized by causing prompted the command value to a predetermined or less.

  By incorporating the concept of learning control when designing the feedforward command value, the command value can be automatically generated off-line on the model. As a result, the number of man-hours can be greatly reduced as compared with the conforming work by the actual vehicle running test. Also, the feedforward command value is designed after assuming the performance variation in the system. As a result, the shift shock can be suppressed and the ride comfort can be improved under a wide range of conditions. In addition, the dependence on feedback control can be reduced.

  Further, by taking the automatic transmission as a multi-input multi-output system, it is possible to obtain a time-series command value that realizes a smooth power transfer operation.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
The present embodiment relates to an adaptation operation of an automatic transmission in which a gear ratio is set according to the engagement / release states of a plurality of friction engagement devices.

  The automatic transmission to be adapted is configured by a plurality of planetary gears and a hydraulic friction engagement device that is frictionally engaged by a plurality of hydraulic actuators, and the hydraulic actuators are electronically controlled by an in-vehicle computer. .

"overall structure"
FIG. 1 shows an outline of the adaptation work on a CAE (computer aided engineering) model. The CAE model includes an engine / torque converter model 10, a gear train model 12 composed of planetary gears of an automatic transmission, a release-side brake model 14, and an engagement-side brake model 16, and these are input to these models. The correction amount of the command value is calculated by the correction amount calculation unit 18.

  An initial value for the engine retard amount is supplied to the adder 20. Here, a correction amount for the engine retardation amount is supplied from the correction amount calculation unit 18, and the engine retardation amount is corrected by adding both. The corrected engine retard amount is supplied to the engine / torque converter model 10. Accordingly, the engine output is controlled according to the supplied engine retard amount, and the output turbine torque from the torque converter is determined.

The initial value of the release side hydraulic pressure command is supplied to the adder 22, and the correction amount of the release side hydraulic pressure command from the correction amount calculation unit 18 is supplied here. Accordingly, the release side hydraulic pressure command corrected by the adder 22 is obtained and supplied to the release side brake model 14, and the corresponding release side torque capacity T ₁ is output from the release side brake model 14.

Further, the initial value of the engagement side hydraulic pressure command is supplied to the adder 24, and the correction amount of the engagement side hydraulic pressure command from the correction amount calculation unit 18 is supplied thereto. Therefore, the engagement hydraulic pressure command that is modified is obtained at the adder 24, which is supplied to the engaging-side brake model 16, the engagement side torque capacity T ₂ corresponding the engaging-side brake model 16 is output.

The turbine torque output from the engine torque converter model 10, the release side torque capacity T ₁ output from the release side brake model 14, and the corresponding engagement side torque capacity T ₂ from the engagement side brake model 16 are the gear train model. 12 is supplied. Thus, from this gear train model 12 output shaft torque T _o and turbine angular acceleration (d / dt) ω _t is output.

Then, the output shaft torque T _o is supplied as a negative value to the adder 26 where it is subtracted from the target value, resulting errors are supplied to the correction amount calculation unit 18. Further, the turbine angular acceleration (d / dt) ω _t is supplied to the adder 28 as a negative value, wherein the target value T _od, (d / dt) is subtracted from each of omega _td, resulting error is corrected amount It is supplied to the calculation unit 18.

  Based on the supplied error, the correction amount calculation unit 18 calculates and calculates correction amounts for the engine retardation amount, the release side hydraulic pressure command, and the engagement side hydraulic pressure command, and these are added to the adders 20, 22, and 24, respectively. Supply.

Command value generation
For example, a feedforward command value (disengagement side and engagement side brake hydraulic pressure / engine retardation amount) that realizes a target output shaft torque waveform and a target turbine angular acceleration waveform at the time of upshift from the second speed to the third speed is obtained by learning calculation. derive. Here, in an actual automatic transmission, there are performance variations due to individual differences of constituent members, aging deterioration, and the like. In the present embodiment, both the learning calculation for obtaining the feedforward command value for the normative model having no performance variation and the learning calculation for obtaining the feedforward command value for the variation model considering the performance variation are performed.

  That is, as shown in FIG. 2, first, a simulation is performed using only the reference model, and learning is performed to obtain a feedforward command value (step S1). Thereafter, simulation is performed using both the norm model and the variation model, and learning is executed (step 2).

  Here, FIG. 3 shows a learning flow in S1 in FIG. First, initial values for the brake oil pressure (release side and engagement side brake oil pressure) waveforms and the engine retard angle command value are input (step S11). Based on this initial command value, a shift simulation is executed using a reference model (step S12). It is determined whether or not the simulation result satisfies the learning termination condition (S13). If the simulation result is satisfied, the process is terminated. The learning end condition is that the number of repetitions exceeds a predetermined number, or that the error between the output shaft torque and turbine angular acceleration and the target value of these outputs is sufficiently smaller than the predetermined value (below the predetermined value). And so on.

  If the end condition is not satisfied, the torque correction amount of each part is calculated (S14), the correction amount for each hydraulic waveform and the retard amount is calculated based on the calculated torque correction amount, and the correction amount is calculated by the calculated correction amount. Each command value is corrected (S15). And each corrected command value is employ | adopted and it returns to step S12.

`` Calculation of correction amount ''
Next, calculation of torque and control value correction amount in steps S14 and S15 will be described.

"Torque correction amount"
First, the equation of motion at the time of clutch-to-clutch shift of the automatic transmission is shown below.

Here, T _o is the output shaft torque, (d / dt) ω _t is turbine angular acceleration, T _B1 torque capacity of the disengagement side brake, T _B2 is the torque capacity of the on-coming brake, T _t is turbine torque, T _w represents the running load torque. Each coefficient A ₁₁ , A ₁₂ ,..., A ₂₄ in the above expression represents a constant determined by the gear ratio of the gear train system constituted by the planetary gears and the moment of inertia of each part. Using the matrix representation, rewrite the above equation as shown below. The bold character A is a matrix represented by the rightmost side.

As input two brake torque capacity and turbine torque, an error between the case of executing the speed change simulation in engine torque converter model, a target value of the output shaft torque T _o and turbine angular acceleration (d / dt) ω _t [delta] t _o, delta when there is (d / dt) ω _t, pseudo-inverse a ⁺ of the matrix a, and the correction amount of the brake torque capacity using the gain matrix K for adjusting the convergence to the target value [delta] t _B1, [delta] t _B2 , a correction amount ΔT _t of the turbine torque is obtained as follows.

As for the brake torque capacities T _B1 and T _B2 , a new brake torque capacity is obtained by adding these correction amounts ΔT _B1 and ΔT _B2 to the brake torque capacities so far. The running load torque _Tw is a load on the output side of the automatic transmission and is not input to the model.

`` Control value correction amount ''
Here, with respect to the brake torque capacities T _B1 and T _B2 , the correction amount is determined as follows. First, the brake on the engagement side is converted to a brake hydraulic pressure that generates torque capacity through the torque phase / inertia phase according to the following relational expression.

Here, P _B2 is the brake hydraulic pressure, μ _B2 is the friction coefficient, N _B2 is the number of friction materials, R _B2 is the friction material effective radius, S _B2 is the piston pressure area, and F _B2 is the return spring load.

Further, during the torque phase, with respect to the correction ΔT _B2 of the torque capacity T _B2 of the engagement side brake, the tie-up and the engine speed blowing phenomenon from the relationship between the torque capacity of the engagement side brake and the turbine torque. The hydraulic waveform of the release side brake hydraulic pressure T _B2 is corrected so as not to occur.
Further, during the inertia phase, the retard control of the engine is performed in order to suppress the peak value of the output shaft torque. From the turbine torque correction amount ΔT _t , the engine retardation amount correction amount Δα is obtained according to the following equation.

  Here, α represents a retard amount, Δα represents a retard amount correction amount, t represents a torque ratio between the engine torque and the turbine torque, and k represents a constant.

  In this way, a plurality of time-series feedforward command values that realize smooth power transfer work between the engagement friction elements by repeatedly calculating the correction amount of each feedforward command value according to the equation of motion of the system. It becomes possible to derive theoretically.

"Processing using a variation model"
Here, in an actual automatic transmission, there is a case where a torque capacity as designed cannot always be generated due to individual variation of constituent members or aging deterioration. As a result, the timing at which the equation of motion during the shift changes is shifted, and a large shift shock may occur.

  Therefore, in the present embodiment, as described above, not only the reference model but also a variation model that includes a system performance variation such as a variation in hydraulic pressure in advance is also a subject. And, by correcting the feedforward command value alternately for both the norm and the fluctuation model, it is possible to reduce the influence of performance variation and obtain the feedforward command value that completes the shift operation under a wide range of conditions. Can do.

  FIG. 7 shows a flow of learning using both the basic model and the variation model corresponding to step S2 in FIG.

  First, the control value obtained using the reference model obtained by the process of FIG. 3 is set as an initial value, and a simulation is performed using a variation model (step S21). Here, the simulation using the variation model adopts the one that can obtain the worst result by the variation model. Then, a correction amount for each hydraulic pressure waveform and retardation amount is obtained from the error between the output (output shaft torque and turbine angular acceleration) obtained by the simulation and the target value of these outputs, and each command value is determined by the obtained correction amount. Is corrected (step S22).

  In this way, when the simulation using the variation model is completed, the simulation is performed using the normative model using each corrected command value (step S23). Then, a correction amount for each command value is obtained from the error between the obtained output and the target value, and is corrected (step S24).

  Next, it is determined whether or not the learning end condition is satisfied (step S25). If the learning end condition is satisfied, the process ends. If not, the process returns to step S21. As the termination condition, the number of repetitions of the learning calculation or whether the error is within a predetermined value in the norm / variation model obtained in step S24 and step S22 is adopted.

  Further, in correcting the feedforward command value, it is preferable to change the values of the learning gain in the reference model in step S24 and the learning gain in the variation model in step S22. This makes it possible to determine which model gives priority to the response. For example, it is preferable to increase the learning gain in step S22 and prioritize the response improvement with the variation model.

"Example of processing"
FIG. 4 shows the shift simulation result when the initial command value is input in the reference model, and FIG. 5 shows the simulation result by the command value corrected by the learning calculation. From these figures, it is understood that the shift shock is improved by learning calculation, and a command value for realizing the target waveform is derived.

  However, automatic transmissions have performance variations due to individual differences in components and deterioration over time. For this reason, when a shift simulation is performed by intentionally adding a minus offset to the hydraulic waveform of the engagement-side brake obtained in step S1 in FIG. 2, the shift does not end within the expected time as shown in FIG. A shock occurs.

  FIG. 8 shows a transition of an error from the target output shaft torque waveform during the learning calculation by the processing of FIG. Case 1 in FIG. 8 shows the simulation result with the reference model, and Case 2 shows the simulation result in a state where the hydraulic pressure of the engagement side brake is offset to the minus as the worst fluctuation model. From this figure, it can be seen that the error is increased and the response is worsened for case 1, but the error value is reduced and the response is improved for case 2. In this example, the correction of the feedforward command value is finished when the errors in these two cases become substantially the same value.

  FIGS. 9 and 10 show the shift simulation results based on the feedforward command value derived in this way. FIG. 9 shows the result when the obtained feedforward command value is inputted as it is, and FIG. 10 shows the result when the hydraulic waveform to the engagement side brake is inputted with a minus offset.

  Furthermore, FIG. 11 shows an engagement-side hydraulic waveform derived by executing learning calculation for the variation model shown in FIG. In FIG. 11, waveform 1 represents a learning calculation result using only the norm of step S1 in FIG. 2, and waveform 2 represents a learning result using both the norm and variation model of step S2 in FIG. Since the fluctuation model is also targeted, the hydraulic waveform during the shift slightly increases.

  From FIG. 9 and FIG. 10, it is found that such a correction requires a feedforward command value that can be shifted without causing an excessive shock even when the brake hydraulic pressure as designed is not generated due to performance variations in the system. I understand.

It is a figure which shows the outline | summary of the adaptation work on a model. It is a flowchart of command value generation. It is a flowchart which shows the learning calculation in a reference | standard model. It is a figure which shows the simulation result using an initial command value. It is a figure which shows the simulation result in a reference | standard model. It is a figure which shows the simulation result in case there exists a hydraulic pressure fluctuation. It is a flowchart which shows the learning calculation in a norm and a fluctuation model. It is a figure which shows transition of the error by repetition learning. It is a figure which shows the simulation result in case there is no oil pressure fluctuation | variation. It is a figure which shows the simulation result in case there exists a hydraulic pressure fluctuation. It is a figure which shows the corrected engagement hydraulic pressure waveform.

Explanation of symbols

  10 Engine / torque converter model, 12 Gear train model, 14 Release side brake model, 16 Engagement side brake model, 18 Correction amount calculation unit, 20, 22, 24, 26, 28 Adder.

Claims (4)

A shift simulation device for an automatic transmission using a model,
For the output shaft torque waveform and the turbine angular acceleration waveform of the automatic transmission, learning calculation is repeatedly performed so as to reduce an error between the simulation result using the model and the target, and each factor for torque transmission of the automatic transmission is determined. Deriving by repeatedly obtaining the feed control command value for the engagement control command value to the combined friction element and the input turbine torque command to the automatic transmission based on the simulation results ;
Two models are prepared: a reference model in which design specification values are set, and a variation model that includes performance variations. The command values are corrected for both models, and the learning calculation is repeated. A simulation apparatus for an automatic transmission, characterized in that the command value for obtaining an error of a result or less is obtained. In the automatic transmission simulation device according to claim 1,
The automatic transmission simulation device, wherein the learning gain can be adjusted independently between the reference model and the variation model. In the automatic transmission simulation device according to claim 1 or 2,
An automatic transmission simulation apparatus characterized in that a gear train system of an automatic transmission is expressed as a multi-input multi-output system, and a correction amount of a feedforward command value is obtained from a pseudo inverse matrix of a matrix representing the system. A shift simulation program for an automatic transmission that causes a computer to execute a shift simulation of an automatic transmission using a model,
On the computer,
With respect to the output shaft torque waveform and the turbine angular acceleration waveform of the automatic transmission, the learning calculation is repeatedly performed so as to reduce the error between the simulation result using the model and the target. In addition to deriving the engagement control command value to the engagement friction element and the feedforward command value for the input turbine torque command to the automatic transmission based on the simulation results ,
As the model, two models, a reference model in which design specification values are set and a variation model including performance variation, are prepared, the command value is corrected for both models, and the learning calculation is repeatedly performed . A simulation program for an automatic transmission, characterized in that the command value for obtaining an error of a simulation result below a predetermined value is obtained.