JPH05270413A - Vibration reducing device for steering wheel - Google Patents

Abstract

PURPOSE:To reduce vibration of a steering wheel by outputting a signal of specifying an excitation vibration waveform for canceling the vibration, and vibration exciting the steering wheel by a vibratory exciter, by neural network arithmetic operation of using the vibration of the steering wheel detected by a vibration sensor. CONSTITUTION:A vibration sensor 11 formed of an acceleration sensor is mounted on a steering wheel, to input a detection signal of the sensor to a level detector 17 through BPF14 to 16. A level signal of representing each amplitude level a1 to a3 of the extracted detection signal is output from this level detector 17, and the respective signals are input to the first neural network 20 of estimating a car speed V, engine speed N and a functional value S and to the second neural network 30 of determining a phase phi, gain (g) and a frequency (f) of specifying an exciting vibration waveform for canceling vibration of the steering wheel. By an exciting vibration waveform signal obtained by arithmetically processing outputs of each neural network 20, 30, a vibration exciter 70 additionally provided in the steering wheel is controlled to suppress vibration of the steering wheel.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a steering wheel vibration reducing apparatus for reducing the vibration generated in a steering wheel of a vehicle.

[0002]

2. Description of the Related Art Conventionally, an apparatus of this type is disclosed in, for example, Japanese Patent Laid-Open No.
As disclosed in Japanese Patent Application Laid-Open No. 3-20269, the rigidity of the steering operation system is increased by detecting the vehicle speed and the steering wheel steering angle, and when the vehicle speed is in the vibration generation region and the steering wheel steering angle is in the minute range. Try to prevent the steering wheel from vibrating.

[0003]

However, in the above conventional device, when the steering wheel may vibrate, the rigidity of the steering operation system is simply increased. The vibration once generated cannot be positively suppressed. Further, even if vibration does not occur in the steering wheel, the rigidity of the steering operation system may be increased, and the steering torque of the steering wheel may be unnecessarily increased. The present invention has been made to solve the above problem, and an object of the present invention is to provide a handle vibration reducing apparatus that can appropriately reduce the handle vibration by using a neural network operation.

[0004]

In order to achieve the above-mentioned object, a structural feature of the present invention is that a vibration sensor is attached to a handle or in the vicinity thereof and detects a vibration of the handle and outputs a detection signal indicating the vibration. And neural network means for storing a calculation parameter and performing a neural network calculation of the detection signal from the vibration sensor using the calculation parameter to output a waveform defining signal defining a vibration waveform for canceling the vibration of the handle. , An excitation waveform signal generating means for forming an excitation waveform signal based on the output waveform regulation signal, and a handle which is mounted on or near a handle and is driven by the formed excitation waveform signal to apply the handle. The shaker to be shaken and the calculation parameters stored in the neural network means are updated to In that a learning control means to optimize the vibration state of dollars.

[0005]

In the present invention configured as described above, the neural network means uses the vibration of the steering wheel detected by the vibration sensor to perform a neural network calculation to cancel the vibration. In addition to forming and outputting the waveform defining signal that defines the waveform, the exciting waveform signal generating means forms and outputs the exciting waveform signal defined by the waveform defining signal, and the exciter operates the handle based on the exciting waveform signal. Since the vibration is applied, the vibration generated in the steering wheel is directly offset by this vibration, and the vibration is positively and accurately suppressed. Further, since the learning control means updates the calculation parameters stored in the neural network means to optimize the vibration state of the handle by the vibration exciter, the neural network means always defines the optimum vibration waveform. It will output a signal. As a result, even if the steering operation system changes over time due to long-term use of the vehicle, the vibration generated in the steering wheel is always accurately suppressed.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a vibration reduction device for a handle according to the embodiment.

This vibration reducing device is equipped with a vibration sensor 11. As shown in FIG. 2, the vibration sensor 11 is composed of an acceleration sensor mounted on the steering shaft 13 in the vicinity of the steering wheel 12, detects a vibration acting on the steering wheel 12, and outputs a detection signal representing the vibration. To do. The vibration sensor 11 may be directly provided on the handle 12.
The vibration sensor 11 includes three band pass filters 14 to 16
It is connected to the level detector 17 via. The center frequencies of the bandpass filters 14 to 16 are set to different values, and the filters 14 to 16 extract and output detection signals of about one octave adjacent to each other. The level detector 17 detects each amplitude level of the extracted detection signal and outputs a level signal representing each amplitude level a1, a2, a3 (see FIG. 3).

The level detector 17 determines a vehicle speed V, an engine speed N and a sensory value S, and a first neural network 20 for determining a phase φ, a gain g and a frequency f for defining a vibration waveform. 2 connected to the neural network 30. As shown in FIG. 4, the first neural network 20 includes input units 21-1 to 21-1.
21-3, the input layer 21, the intermediate units 22-1 to 22
-m intermediate layer 22 and output units 23-1 to 23-23
-3 and the output layer 23. The input units 21-1 to 21-3 respectively input the amplitude levels a1, a2 and a3 as input data IDi (i = 1 to 3), and the output units 23-1 to 23-3 respectively output the vehicle speed V and the engine. Output data Zk (k = 1 to 3) representing the rotation speed N and the sensory value S
Are output respectively. The sensory value S is an index indicating the comfort that the driver feels, and the larger the same value S, the more comfortable the driver is. Input data IDi (i = 1 to 3), output values Xi (i = 1 to 3) of the input units 21-1 to 21-3,
Output values Yj of the intermediate units 22-1 to 22-m (j = 1 to 1
m) and the output values Zk of the output units 23-1 to 23-3
The respective relationships of (k = 1 to 3) are defined as the following mathematical expressions 1 to 3.

[0009]

[Equation 1] Xi = f (ω0i · IDi−θ0i)

[0010]

[Equation 2]

[0011]

[Equation 3]

The respective coupling coefficients ω0i, ω1ij, ω2jk and the respective thresholds θ0i, θ1j, θ2k (calculation parameters) are constants representing the learning result in advance, and the function f (X) is a step function or a sigmoid function. And the coupling coefficients ω0i, ω1ij, ω2jk, the thresholds θ0i, θ1j, θ2k, and the function f (X) are the units 21-1 to 21-3 and 22-1 to 22.
-m, which is stored in advance in a non-rewritable memory provided in 23-1 to 23-3.

This pre-learning and the coupling coefficients ω0i, ω1ij, ω2
Each unit 21-1 for jk and thresholds θ0i, θ1j, θ2k
The storage for 21-3, 22-1 to 22-m, and 23-1 to 23-3 will be described. First, a neural network 20 'configured in software or hardware as in the above embodiment is prepared and mounted in the same vehicle as this embodiment. However, in this neural network 20 ',
Each coupling coefficient ω0i, ω1ij, ω2jk and each threshold θ0
A rewritable memory device such as a RAM for storing i, θ1j, and θ2k is prepared. Then, the amplitude levels a1, a2, a3 measured by the vibration sensor 11 and detected by the level detector 17 are input to the neural network 20 ', and output as the vehicle speed V, the engine speed N, and the sensory value S. , While updating the coupling coefficients ω0i, ω1ij, ω2jk and the thresholds θ0i, θ1j, θ2k by the back-propagation method so that the respective differences between the actual vehicle speed V, the engine speed N and the sensory value S are minimized. Stored in the memory device. These attempts are repeated many times while the engine speed N is variously changed while the vehicle is stopped, and the vehicle speed V and the engine speed N are variously changed, and the coupling coefficient stored in the memory device is stored. ω0i, ω1ij, ω2jk and the thresholds θ0i, θ1j, θ2k are used as learning results, and each unit 21-1 to 21-1 in the neural network 20 of the present embodiment.
21-3, 22-1 to 22-m, and 23-1 to 23-3.

The second neural network 30 is shown in FIG.
As shown in FIG. 3, an input layer 31 including input units 31-1 to 31-5, an intermediate layer 32 including intermediate units 32-1 to 32-n, and an output layer 33 including output units 33-1 to 33-3. It is composed by. Each input unit 31-1 ~
Reference numeral 31-5 inputs each amplitude level a1, a2, a3, vehicle speed V and engine speed N as input data IDi (i = 1 to 5), and each output unit 33-1 to 33-3 applies an excitation waveform. The output data Wk (k = 1 to 3) representing the phase φ, the gain g, and the frequency f that define Input data IDi (i = 1 to 5), input units 31-1 to 31
-5 output value Ui (i = 1 to 5), intermediate unit 32-1 to
The relationship between the output value Vj of 32-n (j = 1 to n) and the output value Wk of the output units 33-1 to 32-3 (k = 1 to 3) is similar to that of the first neural network 20. It is defined as in the following expressions 4 to 6.

[0015]

[Formula 4] Ui = f (α0i · IDi−β0i)

[0016]

[Equation 5]

[0017]

[Equation 6]

Each of the coupling coefficients α0i, α1ij, α2jk and the thresholds β0i, β1j, β2k (calculation parameters) are constants representing the learning result in advance, but are sequentially updated by the subsequent learning. It is a thing. Therefore, each of these coupling coefficients α0i, α1ij, α2jk and each threshold β0i,
β1j and β2k are each unit 31-1 to 31-3, 32-1 to 32-
It is stored in a rewritable memory provided in n, 33-1 to 33-3. The function f (X) is the same as that of the first neural network 20.

This pre-learning and the coupling coefficients α0i, α1ij, α2
jk and thresholds β0i, β1j, β2k units 31-1 ~
The initial storage for 31-3, 32-1 to 32-n, and 33-1 to 33-3 will be described. First, a neural network 30 'configured in software or hardware as in the above embodiment is prepared and mounted on the same vehicle as this embodiment. While the vehicle is stopped, the engine speed N is changed variously, and the vehicle speed V and the engine speed N are also changed while the amplitude levels a1, a2, a3, the vehicle speed V and the engine speed N are changed. To output the phase φ, the gain g and the frequency f. On the other hand, in this state, the phase φ, the gain g and the frequency f of the sine wave signal are applied to the vibration exciter 70 while changing respectively, and at that time, the phase φ, the gain g and the frequency f of the sine wave having the highest sensory value S are given. Is measured, and the measurement result and the phase φ, gain g and frequency f output from the neural network 30 ′ are measured.
The back-propagation method is used to minimize the deviation from and the coupling coefficients α0i, α1ij, α2jk and the threshold β
Neural network 3 while updating 0i, β1j, β2k
Store in memory device 0 '. Then, the coupling coefficients α0i, α1ij, α2jk and the thresholds β0i, β1j, β2k stored in the memory device are used as learning results, and each unit 31-1 to 31- in the neural network 30 of the present embodiment.
5, 32-1 to 32-n, 33-1 to 33-3.

First and second neural networks 2
A central control circuit 40 is connected to 0 and 30. The central control circuit 40 is composed of a microcomputer including a CPU, a ROM, a RAM, an I / O, etc., and executes a program corresponding to the flowcharts of FIGS.
The sine wave defining signal representing the phase φ, the gain g and the frequency f from the neural network 30 is output to the sine wave signal generating circuit 50. The sine wave signal generation circuit 50 synthesizes and outputs a sine wave signal of g · sin (2πf + φ) based on the sine wave defining signal. The amplifier 60 current-amplifies this sine wave signal and outputs it to the vibration exciter 70. The vibrator 70 is composed of a permanent magnet and a coil, and as shown in FIG.
It is attached to the steering shaft 13 in the vicinity of 2. The vibrator 70 may be directly attached to the handle 12.
A learning control circuit 80 is also connected to the central control circuit 40. The learning control circuit 80 controls the learning of the second neural network 30 in response to the learning command signal from the central control circuit 40.

Next, the operation of the embodiment configured as described above will be described. At the same time when the engine is started, the vibration reduction device for the handle of the present embodiment operates. By this operation, the vibration sensor 11 detects the vibration of the steering shaft 13 and outputs a detection signal indicating the vibration. This detection signal exhibits a frequency distribution as shown in FIG. 3 (A), for example. The band pass filters 14 to 16 extract signals belonging to each frequency band from the detected signals and supply them to the level detector 17, respectively. The level detector 17 detects the amplitude levels a1, a2, a3 (see FIG. 3 (B)) of each extracted signal, and outputs the same level.
The signals representing a1, a2, a3 are output to the first and second neural networks 20, 30.

The first neural network 20 uses the input amplitude levels a1, a2, a3 to determine the coupling coefficients ω0i, ω1.
ij, ω2jk and the thresholds θ0i, θ1j, θ2k are used to execute the neural network operations of the above equations 1 to 3, and the vehicle speed V, the engine speed N, and the sensory value S are calculated as the second value. Neural network 3
0 and output to the central control circuit 40. The second neural network 30 inputs the calculated vehicle speed V and engine speed N in addition to the amplitude levels a1, a2, a3,
Based on each of these values, the neural network operation of the above equations 4 to 6 using the operation parameters consisting of the coupling coefficients α0i, α1ij, α2jk and the threshold values β0i, β1j, β2k is executed, and the sine wave is obtained as the operation result. The sine wave defining signal representing the phase φ, the gain g, and the frequency f is output to the central control circuit 40.

The central control circuit 40 starts the execution of the program at step 100 of FIG. 6 at the same time when the engine is started, and repeatedly executes the circulation processing consisting of steps 101 to 111. In this circulation processing, in step 101, the vehicle speed V and the engine speed N from the first neural network 20 are input as the vehicle speed value V ₁ and the engine speed value N ₁ , and in step 102, the vehicle speed value V ₁ is a predetermined value. It is determined whether V ₀ or less (V ₁ ≤V ₀ ) and whether the engine speed N ₁ is a value between a predetermined lower limit value N _L and upper limit value N _H (N _L ≤N ₁ ≤N _H ). Thus, it is determined whether the engine is in the idle rotation state. In this case, if the engine is not in the idle rotation state, it is determined to be “NO” in step 102, and the control signal for inhibiting the generation of the sine wave signal is output to the sine wave signal generation circuit 50 in step 111. As a result, the sine wave signal generation circuit 50 stops the generation of the sine wave signal and the vibrator 70 does not operate.

On the other hand, if the engine is in the idle rotation state, it is judged "YES" at step 102 and the program is advanced to step 103 and thereafter. In step 103, the phase φ, the gain g and the frequency f from the second neural network 30 are input as the phase value φ ₁ , the gain value g ₁ and the frequency value f ₁ , and the same phase value φ ₁ and the gain value are input in step 104. The g ₁ and the frequency value f ₁ are output to the sine wave signal generating circuit 50 as a sine wave defining signal. The sine wave signal generation circuit 50 synthesizes the sine wave signal g ₁ · sin (2πf ₁ + φ ₁ ) on the basis of the sine wave regulation signal, and supplies the synthesized signal to the vibration exciter 70 via the amplifier 60. The vibration exciter 70 vibrates the steering shaft 13 by the sine wave signal g ₁ · sin (2πf ₁ + φ ₁ ) and the engine 12 rotates to rotate the steering wheel 12 (steering shaft 1
The vibration generated in 3) is canceled out.

After the processing in step 104, in steps 105 to 107, the "optimum phase update routine",
The respective processes of the "optimal gain update routine" and the "optimal frequency update routine" are executed. The "optimum phase update routine" is executed in step 20 as shown in detail in FIG.
0 is started in, as well as sets the sum minute value Δφ in the phase value phi ₁ as a phase value phi ₂ at step 201 and 202, the sine wave signal generating circuit 5 to the phase value phi ₂
Output to 0. The sine wave signal generation circuit 50 changes the phase of the sine wave signal being generated to the phase value φ ₂ and the amplifier 60
To the vibration exciter 70, the steering shaft 13 is excited by the sine wave signal g ₁ · sin (2πf ₁ + φ ₂ ). Next, in step 203, the sensory value S is input from the first neural network 20, and it is determined whether the same value S has become larger than the previous value, that is, the same value S has been improved.

In this case, if the sensory value S is improved,
If YES in step 203, the program proceeds to step 204. In steps 204 to 206, similar to the processing in steps 201 to 203,
On condition that the sensory value S is improved, the phase value φ ₂ is increased by a minute value Δφ. On the other hand, if the sensory value S is not improved due to the increase of the phase value φ ₂ , step 206
Is determined as “NO” in step 207, and the phase value φ is determined in step 207.
_The phase value φ _{2 is set} to 1 by subtracting the small value Δφ from ₂
The value is returned to the previous value, the same phase value φ ₂ is output to the sine wave signal generation circuit 50 in step 208, and the processing of this “optimal phase update routine” is ended in step 219.

On the other hand, if "NO" is determined in the step 203, it is determined in the steps 209 to 211 that the phase value φ ₂ is controlled to be decreased by the minute value Δφ to improve the sensory value S. Is determined. Also in this case, if the sensory value S is improved, in step 211, “YE
S ”, and the phase value φ ₂ is a minute value Δφ, provided that the sensory value S is improved in steps 212 to 214.
It is reduced by one. On the other hand, due to the decrease of this phase value φ ₂ ,
When the sensory value S is not improved, it is determined to be “NO” in step 214, and the phase value φ ₂ is returned to the previous value by adding the minute value Δφ to the phase value φ ₂ in step 215. In step 216, the same phase value φ ₂ is output to the sine wave signal generation circuit 50, and in step 219, the process of this “optimal phase update routine” is completed.

Further, even if the phase value φ ₂ is increased by the processing of steps 201 and 202, steps 209 and 2
If the sensory value S is not improved even by the decrease of the phase value φ ₂ by the process of 10, both are determined as “NO” in steps 203 and 211, and the phase value φ is determined in step 217.
₂ is returned to the previous phase value φ ₁ , the same phase value φ ₂ is output to the sine wave signal generation circuit 50 in step 218, and the processing of this “optimal phase update routine” is ended in step 219. As a result of the execution of such "optimum phase update routine", the optimum phase φ of the sine wave signal for suppressing the vibration of the handle 12 is determined.

The "optimal gain update routine" is started in step 300 as shown in detail in FIG. 8, and the same as in the case of the "optimal phase update routine", step 3
By the processing of 01 to 303, the gain value g ₁ is increased by a minute value Δg, and it is determined whether or not the sensory value S is improved.
In this case, if the sensory value S is improved, step 30
It is determined to be “YES” in Step 3, and Steps 304 to 308
By the processing of 1, the gain value g ₂ is increased by a minute value Δg until the sensory value S is not improved. If the sensory value S is not improved due to the increase of the gain value g ₁ in steps 301 to 303, is the sensory value S improved by reducing the gain value g ₁ by the minute value Δg by the processing in steps 309 to 311? Determine whether or not. In this case, the sensory value S
Is improved, the gain value g _{2 is} increased until the sensory value S is not improved by the processing of steps 312 to 316.
Is reduced by a small value Δg.

Further, if the sensory value S is not improved by increasing or decreasing the gain value g ₁ , step 3
03 and 311 both determine "NO", and step 3
In step 17, the gain value g ₂ is returned to the previous gain value g ₁ , and in step 318, the gain value g _{2 is changed} to the sine wave signal generation circuit 5
0 is output, and the processing of this "optimal gain update routine" is ended in step 319. As a result of the execution of the "optimal gain update routine", the optimal gain g of the sine wave signal for suppressing the vibration of the handle 12 is determined.

As shown in detail in FIG. 9, the "optimal frequency updating routine" is started in step 400, and like the "optimal phase updating routine", step 4 is executed.
By the processing of 01 to 403, it is determined whether or not the sensory value S is improved by increasing the frequency value f ₁ by the minute value Δf.
In this case, if the sensory value S is improved, step 40
It is determined to be “YES” in step 3, and steps 404 to 408 are performed.
The frequency value f ₂ is increased by a minute value Δf until the sensory value S is no longer improved by the processing of. If the sensory value S is not improved due to the increase in the frequency value f ₁ in steps 401 to 403, is the sensory value S improved by reducing the frequency value f ₁ by the minute value Δf by the processing in steps 409 to 411? Determine whether or not. In this case, the sensory value S
If is improved, the frequency value f _{2 is} reduced until the sensory value S is not improved by the processing of steps 412 to 416.
Is reduced by a small value Δf.

Further, if the sensory value S is not improved by increasing or decreasing the frequency value f ₁ , step 4
03 and 411 both determine "NO" and step 4
In step 17, the frequency value f ₂ is returned to the previous frequency value f ₁ , and in step 418, the frequency value f _{2 is set} to the sine wave signal generation circuit 5
0 is output, and the processing of this "optimal frequency updating routine" is ended in step 419. As a result of the execution of such "optimum frequency update routine", the optimum frequency f of the sine wave signal for suppressing the vibration of the handle 12 is determined.

Again, returning to the flowchart of FIG. 6,
After the processing of step 107, the first
The vehicle speed V and the engine speed N are input from the neural network 20 as the vehicle speed value V ₂ and the engine speed value N ₂ . Next, in step 109, the above step 10
Absolute of respective differences V ₁ -V ₂ and N ₁ -N ₂ between the vehicle speed value V ₁ and the engine speed value N ₁ input in ₁ and the vehicle speed value V ₂ and the engine speed value N ₂ input in step 108. The values | V ₁ -V ₂ |, | N ₁ -N ₂ | are predetermined small values ΔV, ΔN
It is determined whether or not the following. In this case, the absolute value | V ₁
−V ₂ | is larger than the minute value ΔV, or the absolute value | N ₁ −
If N ₂ | is larger than the minute value ΔN, it is determined in step 109 that “NO”, that is, the state of the vehicle has changed,
The program returns to step 101. On the other hand, absolute value | V
_{If 1−} V ₂ | is a minute value ΔV or less and absolute value | N ₁ −N ₂ | is a minute value ΔN or less, “YE
S ”, that is, it is determined that the state of the vehicle has not changed, and the learning command signal is sent together with the optimum phase value φ ₂ , gain value g ₂ and frequency value f ₂ set by the processing of steps 105 to 107 in step 110. Output to the learning control circuit 80.

The learning control circuit 80 receives the phase value φ ₂ , the gain value g ₂ and the frequency value f ₂ and the phase value φ ₁ , the gain value g ₁ and the frequency value f ₁ output from the second neural network 30. Deviations φ ₂ −φ ₁ , g ₂ −g ₁ , f ₂ −f
_{Using the} backpropagation method, the coupling coefficients α0i, α1ij, α2jk and the thresholds β0i, β1 are minimized so that ₁ is minimized.
Update j, β2k. After that, the second neural network 30 performs the neural network operation using the updated coupling coefficients α0i, α1ij, α2jk and the thresholds β0i, β1j, β2k, and the phase φ, the gain g, and the frequency that define the sine wave signal. Output f.

As can be understood from the above description of the operation, according to the above embodiment, the vibration generated in the handle 12 causes the first and second neural networks 20 and 3 to operate.
0, the sine wave signal generation circuit 50 and the vibrator 70,
Actively and accurately reduced. In the neural network 30, the coupling coefficients α0i, α1ij, α2jk and the thresholds β0i, β1j, β2k are updated by the central control circuit 40 and the learning control circuit 80 based on the learning result of the vibration state of the handle 12. Therefore, even if the steering operation system changes with time due to long-term use of the vehicle, the vibration generated in the steering wheel is always properly suppressed.

Although the signal detected by the vibration sensor 11 is divided into three bands in the above embodiment, it may be divided into five bands as shown in FIG. 3 (B). .. Also, it may be divided into more bands.

In the above embodiment, the first and second neural networks 20 and 30 are configured as hardware, but the networks 20 and 30 may be configured as software using a microcomputer. Good.

Further, in the above embodiment, the vibration of the steering wheel 12 in the idling state is reduced, but the vibration (flutter) of the steering wheel 12 in a predetermined vehicle speed range (for example, about 100 km / h) is reduced. Alternatively, the vibration (shaking) of the handle 12 in the high vehicle speed region may be further reduced. in this case,
When the vehicle speed V represents the predetermined vehicle speed range and the higher vehicle speed range in step 102 of FIG. 6, step 10
The vibrations of the steering wheel 12 may be controlled to be reduced by the processes of 3 to 110, and the learning of the second neural network 30 may be controlled.

[Brief description of drawings]

FIG. 1 is an overall block diagram of a vibration reducing device for a handle according to an embodiment of the present invention.

FIG. 2 is a schematic view of a handle portion showing the mounting positions of the vibration sensor and the vibration exciter of FIG.

FIG. 3A is a distribution diagram for frequencies of a vibration level detected by a vibration sensor, and FIG. 3B is a distribution diagram for frequencies of the same vibration level divided for each frequency band.

4 is a detailed view of the first neural network of FIG. 1. FIG.

5 is a detailed view of the second neural network of FIG.

FIG. 6 is a flowchart of a “main program” executed by the central control circuit of FIG.

FIG. 7 is a detailed flowchart of an “optimum phase update routine” shown in FIG.

FIG. 8 is a detailed flowchart of an “optimal gain update routine” shown in FIG.

FIG. 9 is a detailed flowchart of the “optimal frequency update routine” of FIG.

[Explanation of symbols]

11 ... Vibration sensor, 12 ... Handle, 14-16 ... Band pass filter, 17 ... Level detector, 20 ... First neural network, 30 ... Second neural network, 40 ... Central control circuit, 50 ... Sine wave signal generating circuit ,
60 ... Amplifier, 70 ... Exciter, 80 ... Learning control circuit.

Claims (1)

[Claims] 1. A vibration sensor which is mounted on a handle or in the vicinity thereof and which detects a vibration of the handle and outputs a detection signal representing the vibration, and a calculation parameter stored in the calculation parameter. Neural network means for outputting a waveform defining signal that defines an exciting waveform for canceling the vibration of the handle by performing a neural network operation using the above, and forming an exciting waveform signal based on the output waveform defining signal. Exciting waveform signal generating means, an exciter that is mounted on or near the handle and that excites the handle by being driven by the formed exciting waveform signal, and calculation parameters stored in the neural network means. And learning control means for optimizing the vibration state of the handle by the vibration exciter. A vibration reduction device for the handle, which is characterized by