Vibration actuator control device and vibration actuator having the same, interchangeable lens, imaging device, and automatic stage

Abstract

To provide a control unit for a vibration type actuator or the like that has a control amount output unit different from a conventional PID controller as a main control amount output unit.SOLUTION: A control unit for a vibration type actuator moves a contact body in contact with a vibrator relative to the vibrator, and comprises a control unit that has a first learned model and a second learned model each having a neural network comprising an input layer, a hidden layer, and an output layer. When a target speed is input, the first learned model outputs a first control amount. When location deviation is input, the second learned model outputs a second control amount. A driving unit moves the contact body relative to the vibrator according to a value based on the first control amount and the second control amount. The location deviation is an amount of deviation between a target position and a detected position.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a control device for a vibration actuator, a vibration driving device having the same, an interchangeable lens, an imaging device, and an automatic stage. [Background technology]

[0002] As an example of a vibration type actuator, a vibration type motor will be described. The vibration type motor is a non-electromagnetically driven motor. Specifically, the vibration type motor is configured to apply an AC voltage to an electro-mechanical energy conversion element such as a piezoelectric element connected to an elastic body, thereby generating high-frequency vibration in the electro-mechanical energy conversion element, and extracting the vibration energy as continuous mechanical motion.

[0003] Compared to electromagnetically driven motors, vibration motors have excellent motor performance, such as being small and lightweight, highly accurate, and providing high torque when driven at low speeds. However, vibration motors have nonlinear motor characteristics, making them difficult to model. In addition, the controllability of vibration motors changes depending on the driving conditions and temperature environment, so the control system must be designed in a creative way. Furthermore, vibration motors have many control parameters, such as frequency, phase difference, and voltage, and adjusting these control parameters is also complicated.

[0004] FIG. 20(a) is a control block diagram of a vibration type driving device using a conventional general PID control (see Patent Document 1). A two-phase AC voltage (AC signal) is output from a driving circuit to which a control amount described later is input. The actual speed (detected speed) of a vibration type motor (vibration type actuator) can be controlled by controlling the frequency (1 / period), phase difference, and voltage amplitude (see FIG. 20(b)) of the two-phase AC voltage output from the driving circuit. The voltage amplitude is variable depending on the pulse width input from a PID controller to the driving circuit, which will be described later. Also, (1), (2), and (3) in FIG. 20(b) respectively represent the period, phase difference, and voltage amplitude of the two-phase AC voltage output from the driving circuit.

[0005] A position deviation, which is the difference (target position-detected position) between the target position of the vibration motor commanded by the position command unit and the actual position (detected position) of the vibration motor detected by the position detection unit, is input to a PID controller (control amount output unit). Then, the control amount (frequency, phase difference, and pulse width) PID-calculated according to the position deviation input to the PID controller is sequentially output from the PID controller for each control sampling period, and input to a drive circuit. Then, position feedback control is performed. In the following, the control sampling period is also simply referred to as the "sampling period."

[0006] FIG. 20(c) is a diagram showing the frequency-speed characteristics of a vibration actuator. Specifically, FIG. 20(c) shows that at a frequency (f1) in the high-speed range (low frequency range), the speed is high and the slope of the frequency-speed characteristics is large. Also, at a frequency (f2) in the low-speed range (high frequency range), the speed is low and the slope of the frequency-speed characteristics is small. The control performance (frequency-speed characteristics, phase difference-speed characteristics) of a vibration actuator varies depending on the speed range used, making it difficult to adjust the PID control gain. FIG. 20(d) is a diagram showing the phase difference-speed characteristics of a vibration actuator. Specifically, FIG. 20(d) shows that at a frequency (f1) in the high-speed range (low frequency range), the speed is high and the slope of the phase difference-speed characteristics is large. Also, at a frequency (f2) in the low-speed range (high frequency range), the speed is low and the slope of the phase difference-speed characteristics is small.

[0007] The control performance (frequency-speed characteristics, phase difference-speed characteristics) of a vibration actuator varies depending on the speed range used, making it difficult to adjust the PID control gain. FIG. 20(d) is a diagram showing a schematic diagram of the phase difference-speed characteristics of a vibration actuator. Specifically, FIG. 20(d) shows that at a frequency (f1) in the high-speed range (low frequency range), the speed is high and the slope of the phase difference-speed characteristics is large. Also, at a frequency (f2) in the low-speed range (high frequency range), the speed is low and the slope of the phase difference-speed characteristics is small.

[0008] In addition, when the environmental temperature changes, for example from room temperature to low temperature, the resonant frequency shifts from low to high frequency based on the temperature characteristics of the piezoelectric element. In this case, the speed corresponding to the same drive frequency and the slope of the frequency-speed characteristic corresponding to the drive frequency are different before and after the resonant frequency shifts from low to high frequency, so the control performance also changes depending on the environmental temperature.

[0009] Furthermore, the speed and inclination differ depending on the individual vibration motor, so the control performance also varies depending on the individual motor.

[0010] In addition, control performance also changes over time. It is necessary to consider all of these factors of change and adjust the PID control gains (proportional gain, integral gain, and differential gain of PID control) to ensure a gain margin and phase margin. [Prior art documents] [Patent documents]

[0011] [Patent Document 1] JP 2016-144262 A Summary of the Invention [Problem to be solved by the invention]

[0012] Therefore, there has been a demand for a control device for a vibration type actuator having a control amount output unit different from the conventional PID controller as the main control amount output unit. The present invention aims to provide a control device for a vibration type actuator having a control amount output unit different from the conventional PID controller as the main control amount output unit. [Means for solving the problem]

[0013] One aspect of the present invention is a control device for a vibration type actuator that moves a contact body in contact with a vibrator relative to the vibrator by vibrations generated in the vibrator, the control device including a first trained model and a second trained model, each of which has a neural network composed of an input layer, a hidden layer, and an output layer, and a drive unit that moves the contact body in contact with the vibrator relative to the vibrator by a control amount output from the control unit, wherein the first trained model outputs a first control amount for moving the contact body relative to the vibrator when a target speed for moving the contact body relative to the vibrator is input, and the second trained model outputs a second control amount for moving the contact body relative to the vibrator when a position deviation for moving the contact body relative to the vibrator is input, the control unit further includes a PID controller, and when the position deviation is input, the PID controller outputs a third control amount for moving the contact body relative to the vibrator; The position deviation is a deviation between a target position for moving the contact body relative to the vibrator and a detection position detected when the contact body is moved relative to the vibrator. the drive unit moves a contact body that contacts the transducer relative to the transducer using a value based on the first control amount, the second control amount, and the third control amount as the control amount. The present invention relates to a control device for a vibration type actuator. Effect of the Invention

[0014] According to the present invention, it is possible to provide a control device for a vibration type actuator having, as a main controlled variable output section, a controlled variable output section different from a conventional PID controller. [Brief description of the drawings]

[0015] [Figure 1] 1 is a diagram showing a control device for a vibration actuator according to a first embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram illustrating a configuration of a machine learning unit according to the first exemplary embodiment of the present invention. [Diagram 3] FIG. 2 is a diagram showing a neural network structure of a learning model according to the first embodiment of the present invention. [Figure 4] 1A and 1B are diagrams illustrating the driving principle of a linear drive vibration actuator. [Diagram 5] FIG. 2 is a diagram illustrating a lens drive mechanism of the lens barrel. [Figure 6] 1 shows a flowchart of machine learning and control using a trained model in a first embodiment of the present invention. [Figure 7] 11 is a timing chart illustrating batch learning and online learning in the machine learning unit. [Figure 8] 13 shows a flowchart for using Adam as a method for optimizing neural network parameters. [Figure 9] The calculation results of Adam, RMSprop, Momentum, and SGD are compared using the learning model of the first embodiment of the present invention and actually measured learning data. [Figure 10] 13 shows the results of feedback control of a vibration actuator using a predetermined target position pattern in the control device of the present invention. [Figure 11] This is a result showing the robustness of the control device of the present invention. [Figure 12] FIG. 4 is a diagram showing a control device for a vibration actuator in a second embodiment of the present invention. [Figure 13] FIG. 4 is a diagram showing the speed characteristic of a vibration actuator based on a control amount. [Figure 14] FIG. 11 is a diagram showing a control device for a vibration actuator in a third embodiment of the present invention. [Figure 15] This shows the neural network structure of a learning model that outputs phase difference, frequency, and pulse width. [Figure 16] FIG. 13 is a diagram showing a control device for a vibration actuator in a fourth embodiment of the present invention. [Figure 17] FIG. 13 is a diagram showing a control device for a vibration actuator in a fifth embodiment of the present invention. [Figure 18] 13A and 13B are a plan view showing the external appearance of an imaging device that is an application example (eighth embodiment) of a control device of the present invention, and a schematic diagram of an internal structure thereof. [Figure 19]FIG. 13 is a diagram showing the appearance of a microscope that is an application example (ninth embodiment) of a control device of the present invention. [Figure 20] 1 is a diagram showing an example of a control device for a vibration type actuator using a conventional general PID control. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] (First embodiment) Fig. 1 is a control block diagram of the vibration actuator of this embodiment. A vibration driving device 17 has a control device 15 and a vibration motor 13 (vibration actuator). In Fig. 1, the control device 15 is the vibration driving device 17 without the vibration actuator 13.

[0017] The control device 15 has a control unit 10 that controls the vibration type actuator, a machine learning unit 12 that generates a trained model, and a driving unit 11. The control device 15 also has a position detection unit 14 that detects a relative position (relative position) of the contact body 132 with respect to the vibrator 131, and a speed detection unit 16 that detects a relative speed (relative speed) of the contact body 132 with respect to the vibrator 13j. In the following, the relative position detected by the position detection unit is also referred to as a detected position. In the following, the relative speed detected by the speed detection unit is also referred to as a detected speed. As the position detection unit, a so-called absolute encoder or an incremental encoder is used, but is not limited to these. As the speed detection unit, it is not limited to one that directly detects speed information (a speed sensor). As the speed detection unit, one that indirectly detects speed information, for example, one that detects speed information by calculating position information, may be used.

[0018] The control unit 10 is configured to add the control amounts output by the two trained models so as to generate a signal for controlling the driving of the vibrator 131 (the relative movement of the contact body 132 with respect to the vibrator 131). That is, the target speed and the position deviation of the vibration type actuator are inputted individually to the trained model, and the output phase difference and frequency are used as the control amount. The target speed is a speed that is set so that the actual speed (detected speed) follows when the contact body 132 is moved relative to the vibrator 131. The position deviation is the difference between the target position and the actual position (detected position). The target position is a position that is set so that the actual position (detected position) follows when the contact body 132 is moved relative to the vibrator 131. Note that the pulse width for changing the voltage amplitude may be used as the control amount.

[0019] The control unit 10 is configured to generate a signal for controlling the driving of the vibrator 131 (the relative movement of the contact body 132 with respect to the vibrator 131). That is, the target speed and the position deviation are input to the learned model, and the output phase difference and frequency are used as the control amount of the vibration type actuator 13. The target speed is a speed that is set so that the actual speed (detected speed) follows when the contact body 132 is moved relative to the vibrator 131. The position deviation is the difference between the target position and the actual position (detected position). The target position is a position that is set so that the actual position (detected position) follows when the contact body 132 is moved relative to the vibrator 131. Note that a pulse width for changing the voltage amplitude may be used as the control amount.

[0020] The control unit 10 has a speed generating unit 101 that generates a target speed, and a position generating unit 102 that generates a target position. The control unit 10 also has a controlled variable output unit 103 having a first learned model that receives a target speed and outputs a phase difference and a frequency. The control unit 10 also has a controlled variable output unit 107 having a second learned model that receives a position deviation and outputs a phase difference and a frequency. In the following, the "controlled variable output unit having a learned model" is also simply referred to as the "learned model". The control unit 10 also has an adder 108 that adds the output from the first learned model 103 and the output from the second learned model 107.

[0021] The driving unit 11 includes an AC signal generating unit 104 and a boost circuit 105 .

[0022] The velocity generating unit 101 generates a target velocity for each time of the relative velocity between the vibrator 131 and the contact body 132. In addition, the position generating unit 102 generates a target position for each time of the relative position, and the difference between the detected position detected by the position detecting unit 14 and the target position is calculated as a position deviation (amount of deviation).

[0023] Here, the target speed is output from the speed generating unit for example, at each control sampling period as a unit time. Also, the target position is output from the position generating unit for example, at each control sampling period as a unit time. Specifically, one command value representing the target speed is output from the speed generating unit 101 for each control sampling period, and one command value representing the target position is output from the position generating unit 102 for each control sampling period. The command value may not be the target speed itself, but may be associated with the target speed. Also, the "speed generating unit" may be called a "speed command unit" since it outputs a command value (commands the target speed). Also, the "position generating unit" may be called a "position command unit" since it outputs a command value (commands the target position).

[0024] The control sampling period refers to one cycle from the acquisition of the position deviation in Fig. 1 through the output of the control amount, the application of AC voltage to the vibrator, and the detection of the actual speed (detected speed) and actual position (detected position) to just before the acquisition of the position deviation begins. In this cycle, the position or speed of the vibration type actuator is feedback-controlled.

[0025] The target velocity is a velocity given to make the vibration actuator follow a predetermined position, and may be generated by differentiating the target position over time. Conversely, the target position may be generated by integrating the target velocity.

[0026] The target speed is input to a first trained model 103, which outputs a first control amount. Meanwhile, the position deviation is input to a second trained model 107, which outputs a second control amount. The first control amount and the second control amount are added by an adder 108, and the sum is output to the drive unit 11 as a control amount (phase difference and frequency).

[0027] In the present invention, a control amount corresponding to a target speed is output from the first trained model 103, and a control amount capable of compensating for the control deviation caused by the driving conditions and the temperature environment is output from the second trained model 107. Therefore, highly accurate and robust control can be performed.

[0028] The control amount (phase difference, frequency) output from the control unit 10 is input to the AC signal generating unit 104, which controls the speed and driving direction of the vibration actuator. The AC signal generating unit 104 generates a two-phase AC signal based on the phase difference, frequency, and pulse width. The boost circuit 105 has, for example, a coil or a transformer, and the AC signal boosted to a desired driving voltage by the boost circuit 105 is applied to the piezoelectric element of the vibrator 131 and drives the contact body 132.

[0029] The machine learning unit 12 performs machine learning to generate a first trained model and a second trained model. The first learning unit 110 receives the detected speed detected by the speed detection unit 16 and the control amount (phase difference, frequency) output from the control unit 10, and uses them as training data (teacher data) to generate the first trained model 103. The second learning unit 111 receives the position deviation and the differential value of the control amount (phase difference, frequency) from the control unit 10, and uses them as training data to generate the second trained model 107.

[0030] The reason why the differential value of the controlled variable is input by the differentiator 112 is to convert it into data that is correlated with the position deviation. In other words, the differential value indicates the time change of the controlled variable, and is therefore related to the behavior of the position deviation.

[0031] Other learning data may be used in the second learning unit 111. For example, parameters (weights, thresholds) of a neural network (hereinafter also referred to as "NN") learned in the first learning unit 110 may be used after being adjusted at a predetermined ratio.

[0032] FIG. 2 is a diagram showing the configuration of the machine learning unit of this embodiment. In the first learning unit 110 and the second learning unit 111, machine learning using NN is performed, respectively. First, the first learning unit 110 will be described. The detection speed is input to a first learning model 901 which is a NN, and the calculated z is output. The error between the control amount and the output z is calculated, and machine learning 902 is performed. By repeating the learning, the weight and threshold of the first learning model 901 are updated and optimized. The second learning unit 111 is similar. The position deviation is input to a NN 903, and z is output. The error between the differential value of the control amount and the output z is calculated, and machine learning 904 is performed. The details of machine learning will be described later.

[0033] FIG. 3 is a diagram showing a neural network structure (hereinafter, also referred to as an "NN structure") of the learning model of this embodiment.

[0034] The first trained model 103 and the second trained model 107 are composed of the following NN. The NN is composed of an X layer as an input layer, an H layer as a hidden layer, and a Z layer as an output layer. In the first trained model 103 of this embodiment, the target speed is set to x1 as input data, and the phase difference is set to z1 and the frequency is set to z2 as output data. In the second trained model 107, the position deviation is set to x1 as input data, and the phase difference is set to z1 and the frequency is set to z2 as output data. The hidden layer is formed of seven neurons, and a general sigmoid function (see FIG. 3(b)) is used as the activation function. The number of neurons in the hidden layer may be other than seven, and is preferably in the range of 3 to 20, for example. The smaller the number of neurons, the lower the accuracy but the faster the learning converges, and the greater the number of neurons, the higher the accuracy but the slower the learning. In addition, while a sigmoid function or ReLU (ramp function) is generally used as the activation function for the output layer, a linear function (Figure 3(c)) was used to accommodate the negative sign of the phase difference, which is the control variable.

[0035] The weight connecting each neuron in the input layer and hidden layer is wh, the threshold of the neuron in the hidden layer is θh, the weight connecting each neuron in the hidden layer and output layer is wo, and the threshold of the neuron in the output layer is θo. Values ​​learned by the machine learning unit 12 are applied to all weights and thresholds. The trained NN can be regarded as a collection of common feature patterns extracted from the time-series data of the speed and control amount of the vibration actuator, and the output is a value obtained by a function with the weights and thresholds as variables.

[0036] In this example, the phase difference and frequency are used as the control variables, but other combinations such as pulse width and frequency, or pulse width and phase difference can also be applied. Also, the output layer of the NN may have one neuron, and may be designed to select either the phase difference, frequency, or pulse width.

[0037] An example of a vibration type actuator applicable to the present invention will be described with reference to the drawings. The vibration type actuator of this embodiment has a vibrator and a contact body. FIG. 4 is a diagram for explaining the driving principle of a linear drive type (direct acting type) vibration type actuator as an example of a vibration type actuator. The vibration type actuator 13 shown in FIG. 4(a) has a vibrator 131 having an elastic body 203 and a piezoelectric element 204 which is an electro-mechanical energy conversion element bonded to the elastic body 203, and a contact body 132 driven by the vibrator 131. By applying an AC voltage to the piezoelectric element 204, two vibration modes as shown in FIG. 4(c) and (d) are generated, and the contact body 132 which is in pressure contact with the protrusion 202 is moved in the direction of the arrow.

[0038] Fig. 4(b) is a diagram showing the electrode pattern of the piezoelectric element 204. For example, the piezoelectric element 204 of the vibrator 131 has electrode regions that are divided into two equal parts in the longitudinal direction. The polarization direction of each electrode region is the same (+). Of the two electrode regions of the piezoelectric element 204, an AC voltage (VB) is applied to the electrode region located on the right side of Fig. 4(b), and an AC voltage (VA) is applied to the electrode region located on the left side.

[0039] If VB and VA are AC voltages with frequencies near the resonance frequency of the first vibration mode and in phase, at one moment the entire piezoelectric element 204 (two electrode regions) expands, and at another moment the entire piezoelectric element 204 (two electrode regions) contracts. As a result, the vibrator 131 generates vibrations in the first vibration mode shown in Fig. 4(c) (hereinafter also referred to as "thrust-up vibrations"). This causes a displacement in the protrusion 202 in the thrust-up direction (Z direction).

[0040] Furthermore, if VB and VA are AC voltages with frequencies near the resonance frequency of the second vibration mode and out of phase with each other by 180°, then at a certain moment, the electrode area on the right side of the piezoelectric element 204 contracts while the electrode area on the left side of the piezoelectric element 204 expands. At another moment, the electrode area on the right side of the piezoelectric element 204 expands while the electrode area on the left side of the piezoelectric element 204 contracts. As a result, vibration of the second vibration mode (hereinafter also referred to as "feed vibration") shown in Fig. 4(d) is generated in the vibrator 131. This causes a displacement in the drive direction (feed direction, X direction) of the protrusion 202.

[0041] Therefore, by applying an AC voltage having a frequency near the resonant frequencies of the first vibration mode and the second vibration mode to the electrodes of the piezoelectric element 204, it is possible to excite a vibration that is a combination of the first vibration mode and the second vibration mode.

[0042] In this way, by combining the two vibration modes, the protrusion 202 performs an elliptical motion in a cross section perpendicular to the Y direction (direction perpendicular to the X and Z directions) in Fig. 4(d). The elliptical motion drives the contact body 132 in the direction of the arrow in Fig. 4(a). The direction in which the contact body 132 and the vibrator 131 move relative to each other, that is, the direction in which the contact body 132 is driven by the vibrator 131 (here, the X direction) is referred to as the drive direction.

[0043] In addition, the amplitude ratio R of the second vibration mode to the first vibration mode (amplitude of forward vibration / amplitude of thrust vibration) can be changed by changing the phase difference of the two-phase AC voltage input to the two equally divided electrodes. In this vibration actuator, it is possible to change the speed of the contact body by changing the amplitude ratio of the vibrations.

[0044] In the above description, the case where the vibrator 131 is stationary (fixed) and the contact body 132 is moved (driven) has been described as an example, but the present invention is not limited to this form. The positions of the contact parts of the contact body and the vibrator need only change relatively. For example, the contact body may be stationary (fixed) and the vibrator may move (drive). That is, in the present invention, "driving" means changing the relative position of the contact body with respect to the vibrator, and does not necessarily require the absolute position of the contact body (for example, the position of the contact body when the position of the housing containing the contact body and the vibrator is used as a reference) to change.

[0045] In the above description, a linear drive type (direct acting type) vibration actuator has been described as an example. That is, the case where the vibrator 131 or the contact body 132 moves (is driven) in a linear direction has been described as an example, but the present invention is not limited to this form. The contact body and the vibrator only need to change the positions of their contact parts relative to each other. For example, the vibrator and the contact body may move in a rotational direction. An example of a vibration actuator in which the vibrator and the contact body move in a rotational direction is a ring-type (rotating type) vibration actuator having a ring-shaped vibrator.

[0046] Vibration actuators are used, for example, in autofocus driving of cameras.

[0047] 5 is a diagram illustrating a driving mechanism for a lens in a lens barrel. The driving mechanism for a lens holder using a vibration actuator includes a vibrator, a lens holder, and a first guide bar and a second guide bar arranged in parallel to slidably hold the lens holder. In this embodiment, the second guide bar is a contact body, and the second guide bar is fixed to another member of the lens barrel, and the vibrator and the lens holder move together.

[0048] The vibrator generates a relative moving force between the vibrator and the second guide bar that contacts the protrusion of the vibrator (elastic body) due to the elliptical motion of the protrusion of the vibrator (elastic body) generated by applying a drive voltage to the electromechanical energy conversion element, thereby making it possible to move the lens holder fixed integrally with the vibrator along the first guide bar and the second guide bar.

[0049] Specifically, the contact body driving mechanism 300 mainly has a lens holder 302 which is a lens holding member, a lens 306, a vibrator 131 to which a flexible printed circuit board is connected, and a pressure magnet 305. The driving mechanism 300 further has two guide bars (a first guide bar 303 and a second guide bar 304) and a base body (not shown). Here, the vibrator 131 will be described as an example of the vibrator.

[0050] Both ends of each guide bar (first guide bar 303, second guide bar 304) are held and fixed to a base (not shown) so that the first guide bar 303 and the second guide bar 304 are arranged parallel to each other. The lens holder 302 has a cylindrical holder portion 302a, a holding portion 302b that holds and fixes the transducer 131 and the pressure magnet 305, and a first guide portion 302c that fits with the first guide bar 303 to act as a guide.

[0051] Pressure magnet 305, which constitutes the pressure means, has a permanent magnet and two yokes arranged on both ends of the permanent magnet. A magnetic circuit is formed between pressure magnet 305 and second guide bar 304, and an attractive force is generated between these members (pressure magnet 305, second guide bar 304). Pressure magnet 305 is arranged at a distance from second guide bar 304, and second guide bar 304 is arranged so as to contact vibrator 131.

[0052] The attractive force applies a pressure between second guide bar 304 and vibrator 131. Two protrusions of the elastic body come into pressure contact with second guide bar 304 to form the second guide section. The second guide section forms a guide mechanism using magnetic attractive force, and may cause a situation in which vibrator 131 and second guide bar 304 are pulled apart due to an external force or the like, but this is dealt with as follows.

[0053] That is, the lens holder 302 is adapted to return to a desired position when the fall prevention portion 302d provided on the lens holder 302 comes into contact with the second guide bar 304. By applying a desired AC voltage (AC signal) to the vibrator 131, a driving force is generated between the vibrator 131 and the second guide bar 304, and the lens holder is driven by this driving force.

[0054] An actual position (detected position) and an actual speed (detected speed) are detected by a position sensor (not shown) attached to the contact body 132 or the vibrator 131. The actual position (detected position) is fed back to the control unit 10 as a position deviation, and the vibration actuator is feedback-controlled so as to follow a target position per unit time. The actual speed (detected speed) is input to the machine learning unit 12 and used as learning data together with the control amount (phase difference and frequency) output from the control unit 10. The learning data is data consisting of a pair of input data and output data (correct answer data). In this embodiment, the learning data is data consisting of a pair of the detection speed as input data and the control amount (phase difference and frequency) as output data (correct answer data).

[0055] In this embodiment, a two-phase drive control device is used as an example to drive a piezoelectric element, which is an electromechanical energy conversion element, in two phases. However, the present invention is not limited to two-phase drive, and can also be applied to a vibration actuator having two or more phases.

[0056] Next, the machine learning unit 12 will be described in detail. The controlled variable output unit 106 having a learning model has an NN structure (see FIG. 3) that receives the actual speed (detected speed) from the speed detection unit 16 and the target deviation as inputs and outputs a phase difference and a frequency. The target deviation is a value that is set so that the position deviation follows when the contact body 132 is moved relatively to the oscillator 131. Here, the target deviation is a value in the same data format as the position deviation. The target deviation is set to zero, for example, but an offset value may be provided to compensate for backlash in the mechanical system.

[0057] The control amount (phase difference and frequency) output from the control unit 10 is used as correct data, and is compared with the control amount output from a control amount output unit 106 having a learning model that has not been learned or is being learned, and an error is calculated. In this example, the phase difference and frequency are used as the control amount, but other combinations such as a pulse width and frequency, or a pulse width and phase difference can also be used as the control amount. Also, the output layer of the NN may have one neuron, and may be designed to select any one of the phase difference, frequency, and pulse width. In the following, the "control amount output unit having a learning model" is also simply referred to as the "learning model."

[0058] Fig. 6 shows a control flowchart using machine learning and a trained model in this embodiment. In step 1 (S1), the weights and thresholds of the first trained model 103 and the second trained model 107 in the control unit are set to initial values. The initial values ​​are set based on a random function (untrained state), but previously trained parameters may also be set. In step 2 (S2), the vibration actuator is controlled using the untrained model (untrained NN).

[0059] In step 3 (S3), the sum of the control amounts output from the first trained model 103 and the second trained model 107 while the vibration actuator is being driven is acquired as learning data. In addition, time series data of the actual speed (detected speed) and the position deviation (deviation amount) detected by the speed detection unit 16 while the vibration actuator is being driven is acquired as learning data. In step 4 (S4), an optimization calculation of the learning model by machine learning is performed using the control amount of the learning data as correct answer data. Optimization refers to adjusting the parameters of the NN so that the output from the NN based on the input to the NN approaches the learning data, and is not limited to adjusting the parameters of the NN so that the output from the NN based on the input to the NN matches the learning data.

[0060] The weights and thresholds of the NN are optimized by machine learning, and the parameters of the first trained model 103 and the second trained model 107 are updated. The machine learning unit 12 has a program that causes a computer (not shown) to execute these steps (step 1 (S1) to step 4 (S4)). In step 5 (S5), the vibration actuator is controlled using the trained models (the first trained model 103 and the second trained model 107) whose weights and thresholds have been updated.

[0061] After the control, in order to respond to changes in the driving conditions and the temperature environment, the process returns to step 3 (S3) and learning data is acquired. The learning data can be acquired by batch learning, which learns while the driving is stopped, or online learning, which learns sequentially while the driving is in progress.

[0062] 7 is a timing chart for explaining batch learning and online learning (a manufacturing method for a control device for a vibration actuator) in the machine learning unit 12. The horizontal axis indicates time, and the vertical axis indicates a target position pattern given as a command value for feedback control of the vibration actuator.

[0063] FIG. 7(a) shows an example of batch learning in which learning is performed while driving is stopped (other than when the contact body is moved relatively to the vibrator). In this example, time-series data of the detected speed and control amount detected during the driving period of the vibration actuator are obtained as learning data, and machine learning and NN parameters (weights, thresholds) are updated using the stopped period. Note that machine learning does not necessarily need to be performed for each stopped period; for example, it is also possible to use a method in which learning is performed only when a change in the temperature environment or driving conditions is detected.

[0064] FIG. 7(b) shows an example of online learning in which sequential learning is performed during driving (when the contact body is moved relatively to the vibrator). In this example, machine learning is performed online in parallel with the driving period of the vibration actuator, and the parameters of the NN (weights, thresholds) are updated during the driving period. By applying online learning, it becomes possible to respond to load fluctuations that occur during the driving period.

[0065] The machine learning in step 4 (S4) described above will be further explained with reference to Fig. 8. Fig. 8 shows a flowchart when Adam is used as an optimization method (optimization algorithm) for the parameters of the NN. Steps 1 and 2 are as described in Fig. 6 above.

[0066] In step 3 (S3), the control amount (n) and the detected speed (n), which are the time-series learning data shown in FIG. 8(b), are acquired. In this example, the first learning unit will be explained, but the second learning unit is omitted because it is performed in a similar procedure. The detected speed (n) and the control amount (n) are measurement data when the vibration actuator 13 is controlled by an unlearned model, and the detected speed (n) is the speed detected by the speed detection unit 16 when the vibration actuator is driven by the control amount (n). The number of samples n of the detected speed and the control amount (phase difference) is 3400. This is actual measurement data when driving for 0.34 sec at a control sampling rate (1 / control sampling period) of 10 kHz.

[0067] It should be noted that the learning data does not necessarily need to be acquired at the controlled sampling rate, and thinning out the data can save memory and shorten the learning time. In this invention, the detection speed (n) is used as the input to the learning model, and the output z(n), which is the result calculated (derived) and output by the learning model, is compared with the controlled variable (n), which is the correct data of the learning data, to calculate the error e(n). Specifically, the error e(n) is given by: Error e(n) = (z(n) - controlled variable (n)) 2 In step 4 (S4), the error E (=Σe(n)=Σ(z(n)-control variable(n)) for 3400 pieces is 2 ) is calculated in the first loop, and the error gradient ∇E of the weights (wh, wo) and thresholds (θh, θo) is calculated.

[0068] Next, the error gradient ∇E is used to optimize the parameters using Adam, an optimization calculation method (optimization algorithm), as shown below.

[0069]

number

[0070] where wt is the parameter update amount, ∇E is the error gradient, Vt is the moving average of the error gradient, St is the moving average of the squared error gradient, η is the learning rate, and ε is an anti-zero division constant.

[0071] The parameters used are η = 0.001, β1 = 0.9, β2 = 0.999, and ε = 10e-12. Each time the optimization calculation is repeated, the weights and thresholds are updated, and the output z(n) of the learning model approaches the control amount (n) of the correct data, so the error E becomes smaller.

[0072] 8(c) shows the transition of the error E based on the number of calculation loops. Note that other methods (algorithms) may be used as the optimization method (optimization algorithm).

[0073] Fig. 9(a) shows a comparison of the calculation results of Adam, RMSprop, Momentum, and SGD using the learning model of this embodiment and the measured learning data. In terms of the number of calculations, stability, and final error, Adam produced the best results.

[0074] Figure 9(b) is an example of learning the control amount (phase difference) using Adam. It can be seen that the output z of the learning model in the first loop is significantly different from the control amount t of the correct data. After repeated calculations, the output z of the learning model in the 5000th loop almost matches the control amount t of the correct data. In this learning example, optimization was performed with the loop count set to 5000, but it is desirable to adjust the loop count appropriately depending on the convergence rate.

[0075] The above is the configuration of the control device of the present invention. The control unit 10 and the machine learning unit 12 are composed of digital devices such as a CPU and a PLD (including ASIC), and elements such as an A / D converter. The AC signal generating unit 104 of the driving unit 11 has, for example, a CPU, a function generator, and a switching circuit, and the boost circuit of the driving unit 11 is composed of, for example, a coil, a transformer, and a capacitor. The control unit 10, the machine learning unit 12, and the driving unit 11 may be composed of a plurality of elements or circuits, rather than being composed of one element or circuit. The processes in the control unit 10, the machine learning unit 12, and the driving unit 11 may be executed by any element or circuit.

[0076] The results of applying the control of the present invention to an actual machine will now be described.

[0077] FIG. 10 shows the results of feedback control of the vibration actuator with a predetermined target position pattern in the control device of the present invention.

[0078] The target speed is a trapezoidal drive with a maximum of 50 mm / s, and the pattern performs a reciprocating motion with a 5 mm stroke including positioning. The horizontal axis indicates time (sec), and the vertical axis on the top row indicates speed (mm / s). The vertical axis on the bottom row indicates the target position (encoder pulse count: 8000 pls per mm) (left axis) and the position deviation in μm (right axis).

[0079] FIG. 10(a) shows the results of PID control as a comparative example, and FIG. 10(b) shows the control results of the present invention. The start-up frequency was set to 93 kHz for both. Looking at the speed results in the upper row first, the detected speed roughly matches the target speed, and both have good tracking ability. The detected speed and speed deviation of the present invention are slightly noisy, but this is due to differences caused by filter processing and is not in keeping with the gist of the present invention, so it can be ignored. Note that the "target speed and detected speed" in FIG. 10(a) looks like a single line, but actually means that there are two lines, the target speed line and the detected speed line.

[0080] Next, we will explain the lower diagram showing the tracking ability of the detected speed to the target position. It can be seen that the present invention has significantly improved the position deviation compared to PID control. This is the effect of learning the speed characteristics of the vibration actuator relative to the control amount.

[0081] FIG. 11 shows the results indicating the robustness of the control device for the vibration actuator of the present invention.

[0082] Positioning was performed using trapezoidal drive at a maximum speed of 50 mm / s, and the position deviation during a reciprocating movement with a 5 mm stroke was calculated at 3σ. The horizontal axis is the starting frequency, and the vertical axis is the position deviation. The comparative example shows the results using conventional PID control. Compared to the comparative example, the control results of the present invention show a significant improvement in position deviation. Even when the drive frequency changes, there is little fluctuation in position deviation, and it can be seen that robustness has been improved.

[0083] As mentioned above, when control is performed with a different start-up frequency, the nonlinear characteristics of the vibration actuator cause the speed gradient to change, which was difficult to address with conventional PID control. The present invention outputs a control amount according to the target speed using the first trained model, and the control deviation caused by the drive conditions and temperature environment can be compensated for by the control amount of the second trained model, making it possible to perform highly accurate and robust control.

[0084] Second Embodiment Another embodiment of the present invention will now be described.

[0085] FIG. 12 is a diagram showing a control device for a vibration actuator in another embodiment of the present invention.

[0086] Machine learning is performed using the pulse width and frequency as control variables, and control is performed using the learned model. In this control block, position feedback control of the vibration actuator 13 is performed using the sum of the control variables output from the first learned model 1003 and the second learned model 1007, the pulse width, and the frequency. In the machine learning unit 12, the control variable (pulse width and frequency) output from the control unit 10 and the relative speed detected by the speed detection unit 16 are acquired as learning data, and machine learning of the first learned model 1003 is performed. In addition, the differential value of the control variable and the position deviation are acquired as learning data, and machine learning of the second learned model 1007 is performed.

[0087] FIG. 13 is a diagram showing the speed characteristic of the vibration actuator based on the controlled variable.

[0088] FIG. 13(a) shows a case where control is performed using phase difference and frequency (see Example 1). The horizontal axis indicates frequency, and the vertical axis indicates motor speed. As shown in the figure, the motor speed can be controlled by manipulating the phase difference and frequency, respectively. For example, when control is performed in the gray area, each control amount is output with a drive frequency of 88 to 93 kHz and a phase difference of 0 to ±120°. The first trained model 103 of the present invention performs control by outputting two control amounts in response to an input of a target speed.

[0089] FIG. 13(b) shows a case where control is performed using the pulse width and frequency of this embodiment. Similarly, the motor speed can be controlled by manipulating the pulse width and frequency, respectively. For example, when control is performed in the gray area, each control amount is output with a drive frequency of 88 to 93 kHz and a pulse width of 0 to 50%. Similarly, control is performed by outputting two control amounts according to the target speed input to the first trained model 1003.

[0090] (Third embodiment) Another embodiment of the present invention will now be described.

[0091] FIG. 14 is a control block diagram of a vibration type driving device in the case of performing control using a trained model in which machine learning is performed with the phase difference, frequency, and pulse width as control amounts. In FIG. 14, the vibration type actuator 13 is removed from the control device. In this control block, the position feedback control of the vibration type actuator 13 is performed based on each control amount (phase difference, frequency, and pulse width) output from each of the first trained model 1103 and the second trained model 1107. Specifically, the control amount is performed by adding the control amount output from the first trained model 1103 and the control amount output from the second trained model 1107. The addition of the control amounts is performed between the same types of control amounts. A target speed is input to the first trained model 1103, and the phase difference, frequency, and pulse width calculated by the NN are output to the driving unit 11. A position deviation is input to the second trained model 1107, and the phase difference, frequency, and pulse width calculated by the NN constituting the second trained model are output to the driving unit 11.

[0092] In the machine learning unit 12, the three control amounts output from the control unit 10 and the relative speed (detected speed) detected by the speed detection unit 16 are acquired as learning data, and machine learning of a first trained model 1103 is performed. In addition, the differential values ​​of the control amounts and the position deviation are acquired as learning data, and machine learning of a second trained model 1107 is performed.

[0093] FIG. 15 shows a NN structure that outputs a phase difference, a frequency, and a pulse width.

[0094] The first trained model 1103 and the second trained model 1107 have an NN structure that receives a target speed or a position deviation as an input and outputs three control amounts. The training data used for machine learning may be measurement data obtained when the vibration actuator is controlled by a trained model, or measurement data obtained when the vibration actuator is controlled by an untrained model in which parameters are set by a random function. In addition, measurement data obtained when the vibration actuator is controlled by open drive or measurement data obtained when the vibration actuator is controlled by PID control may be used.

[0095] When determining the weights and thresholds of the NN, parameters with optimal conditions may be selected from multiple learning data in terms of position deviation, power consumption, etc. This is because there are an infinite number of combinations of conditions under which a given speed of the vibration actuator can be obtained, i.e., phase difference, frequency, and pulse width.

[0096] By applying this embodiment, the number of parameters for operating the vibration actuator increases, so that by performing appropriate machine learning, it is possible to finely adjust the control performance.

[0097] (Fourth embodiment) FIG. 16 is a diagram showing a control device for a vibration actuator in another embodiment of the present invention.

[0098] In this control block, the position feedback control of the vibration actuator 13 is performed by connecting the PID controller 109 in parallel to the second trained model 1207. The position deviation is input to the PID controller 109, which performs PID calculation to output the control amount of the phase difference and the frequency. Note that a configuration other than the PID controller may be used, and for example, P control, PI control, PD control, etc. may be applied. The target speed is input to the first trained model 1203. The PID control amount (third control amount) output from the PID controller, the first control amount output from the first trained model 1203, and the second control amount output from the second trained model 1207 are added together and output to the drive unit 11. The machine learning unit 12 acquires the control amount (phase difference and frequency) output from the control unit 10 and the relative speed detected by the speed detection unit 16 as learning data, and machine learning of the first trained model 1203 is performed. In addition, the differential value of the controlled variable and the position deviation are acquired as learning data, and machine learning of the second learned model 1207 is performed.

[0099] By applying this embodiment, the first trained model 1203 outputs a control amount corresponding to the target speed, and the control deviation caused by the drive conditions and temperature environment can be compensated for by the control amount of the second trained model 1207. Therefore, highly accurate and robust control can be performed. In addition, by using a PID controller in combination, the transfer characteristics of the control loop can be flexibly adjusted, and the positioning accuracy and robustness can be further improved.

[0100] Fifth embodiment FIG. 17 is a diagram showing a control device for a vibration actuator in another embodiment of the present invention.

[0101] In this control block, the PID controller 109 is connected in parallel to the second trained model 1207 for control, and a switch (SW) selectively performs position feedback control of the vibration actuator 13. SW1 can select ON or OFF for the output of the PID controller 109. SW2 can select ON or OFF for the output of the sum of the first control amount and the second control amount. Therefore, it is possible to select control using only the PID controller or control using only the trained model depending on the conditions.

[0102] Both switches may be controlled to be ON. The same applies to machine learning. For example, if learning is performed using only a PID controller, a trained model can be stably generated even in a state where no learning has been performed at all.

[0103] By applying this embodiment, it is possible to perform highly accurate and robust control. In addition, by selectively using a PID controller, it is possible to improve the stability of control and learning.

[0104] Sixth embodiment In the above embodiments, a memory means may be provided for storing the parameters (first weight, second weight, threshold of the second neuron, and threshold of the third neuron) possessed by the NN. The NN may be machine-learned by replacing the parameters possessed by the NN with the parameters stored in the memory means.

[0105] Seventh embodiment In the above-described embodiments, an environmental sensor for detecting an environmental state may be provided. Then, the NN may perform machine learning when a change in the environment is detected by the environmental sensor. The environmental sensor may be at least one of a temperature sensor and a humidity sensor.

[0106] Eighth embodiment In the first embodiment, an example has been described in which the vibration actuator control device is used to drive a lens for autofocusing in an imaging device, but the application of the present invention is not limited to this. For example, as shown in Fig. 18, the control device can also be used to drive a lens or an imaging element during image stabilization. Fig. 18(a) is a plan view (top view) showing the exterior of an imaging device 60. Fig. 18(b) is a schematic diagram of the internal structure of the imaging device 60.

[0107] The imaging device 60 is roughly composed of a main body 61 and a lens barrel 62 that is detachable from the main body 61. The main body 61 is equipped with an imaging element 63 such as a CCD sensor or a CMOS sensor that converts an optical image formed by light passing through the lens barrel 62 into an image signal, and a camera control microcomputer 64 that controls the overall operation of the imaging device 60. A plurality of lenses L such as a focus lens and a zoom lens are arranged at predetermined positions in the lens barrel 62.

[0108] Further, the lens barrel 62 has an image blur correction device 50 built therein, which has a disk member 56 and an oscillator 131 provided on the disk member 56, and an image blur correction lens 65 is disposed in a hole formed in the center of the disk member 56. The image blur correction device 50 is disposed so as to be able to move the image blur correction lens 65 in a plane perpendicular to the optical axis of the lens barrel 62. In this case, by driving the oscillator 131 using the control device 15 of the vibration type actuator of the present invention, the oscillator 131 and the disk member 56 move relative to the contact body 132 fixed to the barrel, and the correction lens is driven.

[0109] The control device for the vibration actuator of the present invention can also be used to drive a lens holder for moving a zoom lens. Therefore, the control device of the present invention can be mounted on an interchangeable lens as well as an imaging device for lens drive.

[0110] Ninth embodiment The control device for the vibration type actuator shown in the first embodiment can also be used to drive an automatic stage. For example, as shown in FIG. 19, it can be used to drive an automatic stage of a microscope.

[0111] The microscope in Fig. 19 has an imaging section 70 incorporating an image sensor and an optical system, and an automatic stage 71 having a stage 72 that is mounted on a base and moved by a vibration actuator. An object to be observed is placed on the stage 72, and an enlarged image is captured by the imaging section 70. When the observation range is wide, the vibration actuator is driven using the control device 15 of the vibration actuator of the first or second embodiment to move the stage 72. This moves the object to be observed in the X and Y directions in the figure, and multiple captured images are acquired. The captured images are combined by a computer (not shown) to acquire a single image with a wide observation range and high resolution. [Explanation of symbols]

[0112] 10 Control section 11 Drive unit 12 Machine Learning Department 13 Vibration motor (vibration actuator) 14 Position detection unit 15 Control device 16 Speed ​​detection section 101 Speed ​​generation section 102 Position generator 103 First trained model 104 AC signal generation section 105 Boost circuit 107 Second trained model 108 Adder 131 Transducer 132 Contact body

Claims

1. 1. A control device for a vibration type actuator that causes a contact body in contact with a vibrator to move relative to the vibrator by vibration generated in the vibrator, comprising: A control unit having a first trained model and a second trained model, each of which has a neural network composed of an input layer, a hidden layer, and an output layer; a drive unit that moves a contact body that contacts the transducer relative to the transducer in accordance with a control amount output from the control unit, the first trained model outputs a first control amount for moving the contact body relative to the oscillator when a target velocity for moving the contact body relative to the oscillator is input; the second trained model outputs a second control amount for moving the contact body relative to the oscillator when a position deviation for moving the contact body relative to the oscillator is input; and The control unit further includes a PID controller, the PID controller outputs a third control amount for moving the contact body relative to the oscillator when the position deviation is input; the position deviation is a deviation amount between a target position for moving the contact body relative to the vibrator and a detection position detected when the contact body is moved relative to the vibrator, A control device for a vibration actuator, characterized in that the driving unit moves a contact body in contact with the vibrator relative to the vibrator using values ​​based on the first control amount, the second control amount, and the third control amount as the control amounts.

2. a detection speed detected when the contact body is moved relative to the vibrator is input, and a value based on the first control amount and the second control amount is output, and using learning data, the neural network to be included in the first trained model is machine-learned; 2. The control device for a vibration type actuator as described in claim 1, further comprising a machine learning unit that uses learning data in which the position deviation is input and a differential value of a value based on the first control amount and the second control amount is output to machine-train the neural network that will be possessed by the second trained model.

3. a detection speed detected when the contact body is moved relative to the vibrator is used as an input, and a value based on the first control amount, the second control amount, and the third control amount is used as an output to machine-learn the neural network to be included in the first trained model; 2. The control device for a vibration type actuator as described in claim 1, further comprising a machine learning unit that uses learning data in which the position deviation is input and a differential value of a value based on the first control amount, the second control amount, and the third control amount is output to machine-train the neural network that will be possessed by the second trained model.

4. A control device for a vibration type actuator that causes a contact body in contact with a vibrator to move relative to the vibrator by vibration generated in the vibrator, comprising: A control unit having a first trained model and a second trained model, each of which has a neural network composed of an input layer, a hidden layer, and an output layer; a drive unit that moves a contact body that contacts the transducer relative to the transducer in accordance with a control amount output from the control unit, the first trained model outputs a first control amount for moving the contact body relative to the oscillator when a target velocity for moving the contact body relative to the oscillator is input; the second trained model outputs a second control amount for moving the contact body relative to the oscillator when a position deviation for moving the contact body relative to the oscillator is input; The control unit further includes a PID controller, the PID controller outputs a third control amount for moving the contact body relative to the oscillator when the position deviation is input; the position deviation is a deviation amount between a target position for moving the contact body relative to the vibrator and a detection position detected when the contact body is moved relative to the vibrator, A control device for a vibration type actuator, characterized in that the driving unit moves a contact body in contact with the vibrator relative to the vibrator using a control amount that is a value based on the first control amount and the second control amount, or the third control amount.

5. a detection speed detected when the contact body is moved relative to the vibrator is used as an input, and a value based on the first control amount and the second control amount or the third control amount is used as an output of learning data to machine-learn the neural network to be included in the first trained model; The control device for a vibration type actuator as described in claim 4, characterized in that it has a machine learning unit that machine-learns the neural network that the second trained model will have using learning data that has the position deviation as an input and the differential value of a value based on the first control amount and the second control amount, or the differential value of the third control amount as an output.

6. 5. The control device for a vibration type actuator according to claim 1 or 4, characterized in that a value based on the first control amount and the second control amount is a sum of the first control amount and the second control amount, and a value based on the first control amount, the second control amount and the third control amount is a sum of the first control amount, the second control amount and the third control amount.

7. 7. The control device for a vibration type actuator according to claim 1, wherein the controlled variable is at least one of a phase difference, a frequency, and a pulse width.

8. A control device for a vibration type actuator as described in any one of claims 1, 3 to 7, characterized in that the neural network has the input layer having one or more first neurons, the hidden layer having a plurality of second neurons, and the output layer having one or more third neurons, a plurality of first weights connecting the first neuron and the second neuron, a plurality of second weights connecting the second neuron and the third neuron, a threshold value of the second neuron, and a threshold value of the third neuron.

9. The control device for a vibration type actuator as described in claim 2, characterized in that the neural network has the input layer having one or more first neurons, the hidden layer having a plurality of second neurons, and the output layer having one or more third neurons, a plurality of first weights connecting the first neuron and the second neuron, a plurality of second weights connecting the second neuron and the third neuron, a threshold value of the second neuron, and a threshold value of the third neuron.

10. The control device for a vibration actuator as described in claim 9, characterized in that the machine learning unit performs machine learning by optimizing the first weight, the second weight, the threshold of the second neuron, and the threshold of the third neuron.

11. The control device for a vibration actuator as described in claim 9, characterized in that the machine learning unit performs machine learning by optimizing the first weight, the second weight, the threshold of the second neuron, and the threshold of the third neuron, which are set by a random function.

12. The control device for a vibration actuator as described in claim 9, characterized in that the machine learning unit performs machine learning by optimizing the optimized first weight, the second weight, the threshold of the second neuron, and the threshold of the third neuron.

13. a storage means for storing the first weight, the second weight, the threshold of the second neuron, and the threshold of the third neuron that were included in the neural network; The control device for a vibration actuator as described in claim 9, characterized in that the machine learning unit performs machine learning by replacing the first weight, the second weight, the threshold of the second neuron, and the threshold of the third neuron of the neural network with the first weight, the second weight, the threshold of the second neuron, and the threshold of the third neuron stored in the memory means.

14. 14. The control device for a vibration type actuator according to claim 2, wherein the machine learning unit performs machine learning in cases other than when the contact body is moved relatively to the vibrator.

15. 14. The control device for a vibration type actuator according to claim 2, wherein the machine learning unit performs machine learning when the contact body is moved relatively to the vibrator.

16. An environmental sensor for detecting an environmental condition is provided.

14. The control device for a vibration type actuator according to claim 2, wherein the machine learning unit performs machine learning when a change in the environmental state is detected by the environmental sensor.

17. 17. The control device for a vibration type actuator according to claim 16, wherein the environmental sensor is at least one of a temperature sensor and a humidity sensor.

18. A speed generating unit that generates the target speed; 6. The control device for a vibration type actuator according to claim 2, further comprising a speed detection unit for detecting the detected speed.

19. A position generating unit that generates the target position; 18. The control device for a vibration type actuator according to claim 1, further comprising a position detection unit that detects the detection position.

20. a vibration actuator that moves a contact body that is in contact with a vibrator relative to the vibrator by vibration generated in the vibrator; A vibration type driving device comprising: a control device for a vibration type actuator according to any one of claims 1 to 19.

21. A vibratory driving device according to claim 20; a lens that is driven by moving the contact body relatively to the oscillator.

22. A vibratory driving device according to claim 20; an imaging element that is driven by moving the contact body relatively to the transducer.

23. A vibratory driving device according to claim 20; a stage that is driven by moving the contact body relative to the oscillator.

Images