Conveying device and conveying method

Abstract

The control section (40) of the conveyance device has an angle acquisition section (91) and a conveyance control section (92). The conveyance control section (92) has a first learned model (M1) obtained by reinforcement learning for maintaining a deflection angle (θ) of the object table at a desired state. The conveyance control section (92) inputs first input data (d1) including the deflection angle acquired by the angle acquisition section (91) to the first learned model (M1). Further, the conveyance control section (92) controls the pair of straight line mechanisms based on a control value v output by the first learned model (M1). Thus, the object table can be conveyed appropriately without the need for a person to set parameters for control.

Description

Technical Field

[0001] This invention relates to a conveying device and method for conveying a flat platform in a horizontal direction via a pair of straight-moving mechanisms. Background Technology

[0002] Previously, it was known to perform various processes on a substrate held by a flat stage while conveying it. For example, Patent Document 1 describes an apparatus that draws an exposure pattern on the upper surface of a substrate (W) while moving a stage (10) on which a substrate (W) is placed by a stage moving mechanism (20).

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent document 1: Japanese Patent Application Publication No. 2016-72434. Summary of the Invention

[0006] The problem the invention aims to solve

[0007] The conveying device of the platform mounted on such a device sometimes has a pair of linear mechanisms. Specifically, a mechanism is known to convey the platform in a predetermined direction by means of a pair of linear motors arranged in parallel to each other.

[0008] In this conveying device, to move the stage in a certain posture, a pair of linear mechanisms need to operate equally. However, sometimes minute drive errors in the pair of linear mechanisms, air pressure fluctuations in the gaps of the linear motor guides, or machining errors can cause slight variations in the rotation angle (so-called "deflection angle") of the stage about the vertical axis. When such variations in the deflection angle occur, it is difficult to perform precise processing on the substrate held by the stage.

[0009] Therefore, in this type of conveying device, the deflection angle of the platform is measured, and the movement of the conveying device is controlled by feedback based on the measurement result. However, the parameters used for feedback control, such as proportional gain, integral gain, and derivative gain, need to be set manually through continuous experimentation, resulting in a heavy workload.

[0010] The present invention was made in view of this situation, and its object is to provide a conveying device and a conveying method that can properly convey a platform without requiring human setting of parameters for control.

[0011] means for solving problems

[0012] To address the aforementioned issues, the first invention of this application is a conveying device that conveys a flat platform along a horizontal direction using a pair of straight-moving mechanisms. The device includes: an angle acquisition unit for acquiring the deflection angle of the platform; and a conveying control unit for controlling the pair of straight-moving mechanisms based on the deflection angle acquired by the angle acquisition unit. The conveying control unit inputs first input data, including the deflection angle acquired by the angle acquisition unit, into a first learned model obtained through reinforcement learning for maintaining the deflection angle in a desired state, and controls the pair of straight-moving mechanisms based on a control value output by the first learned model.

[0013] The second invention of this application is a conveying device as described in the first invention, wherein it further includes an angle measuring device for measuring the deflection angle of the platform, and the angle acquisition unit acquires the measurement result of the angle measuring device on the deflection angle.

[0014] The third invention of this application is a conveying device as described in the first invention, wherein it further comprises an angle estimation unit that estimates the deflection angle of the stage based on measurements output by the conveying mechanism including the pair of straight-moving mechanisms. The angle estimation unit uses a second learned model to estimate the deflection angle, which is obtained by supervised machine learning for estimating the deflection angle of the stage based on second input data including the measurements. The angle acquisition unit acquires the estimation result of the angle estimation unit for the deflection angle.

[0015] The fourth invention of this application is a conveying device as described in any one of the first to third inventions, wherein the control value is the torque value of the pair of straight-line mechanisms.

[0016] The fifth invention of this application is a conveying device as described in any one of the first to fourth inventions, wherein the reinforcement learning is machine learning performed with the goal of maintaining the deflection angle within a specified range.

[0017] The sixth invention of this application is a conveying device as described in any one of the first to fifth inventions, wherein the first input data includes the deflection angle and the angular velocity of the deflection angle.

[0018] The seventh invention of this application is a method for conveying a flat platform along a horizontal direction using a pair of straight-moving mechanisms, comprising: step a), acquiring the deflection angle of the platform; and step b), controlling the pair of straight-moving mechanisms based on the deflection angle acquired in step a), wherein in step b), first input data including the deflection angle acquired in step a) is input into a first learned model obtained by reinforcement learning for maintaining the deflection angle in a desired state, and the pair of straight-moving mechanisms are controlled based on the control value output by the first learned model.

[0019] The effects of the invention

[0020] According to the first to seventh inventions of this application, a first learned model obtained through reinforcement learning is used for control to maintain the deflection angle of the stage in a desired state. Therefore, the stage can be moved appropriately without requiring manual setting of control parameters.

[0021] In particular, according to the third invention of this application, the deflection angle of the stage can be obtained without always setting up a large-scale angle measuring device. Attached Figure Description

[0022] Figure 1 It is a three-dimensional drawing of a device with a conveying mechanism.

[0023] Figure 2 It is a schematic top view of the device with a conveying mechanism.

[0024] Figure 3 It is a block diagram showing the electrical connections between the control unit and the various parts within the drawing device.

[0025] Figure 4 This is a partial cross-sectional view when a portion of the conveying device is cut off by a plane perpendicular to the main scanning direction.

[0026] Figure 5 It is a block diagram schematically illustrating the functions of the control unit used to control the main scanning mechanism.

[0027] Figure 6 This is a flowchart representing the process of prior reinforcement learning.

[0028] Figure 7 This is a graph representing an example of changes in evaluation values ​​during prior reinforcement learning.

[0029] Figure 8 It is a flowchart representing the motion control process of the straight-moving mechanism.

[0030] Figure 9 This is a block diagram schematically representing the function of the control unit in the first modified example. Detailed Implementation

[0031] The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0032] It should be noted that, in the following, the direction in the horizontal direction in which the stage is moved by a pair of straight-line mechanisms is defined as the "main scanning direction", and the direction orthogonal to the main scanning direction is defined as the "secondary scanning direction".

[0033] <1. Describe the structure of the device>

[0034] Figure 1 This is a perspective view of a drawing device 1 having a conveying device 10 according to an embodiment of the present invention. Figure 2 This is a schematic top view of the drawing apparatus 1. The drawing apparatus 1 is a device that illuminates the upper surface of a substrate W, such as a semiconductor substrate or a glass substrate coated with a photosensitive material, with spatially modulated light, thereby drawing an exposure pattern on the upper surface of the substrate W. Figure 1 and Figure 2 As shown, the drawing device 1 includes a conveying device 10, a frame 20, a drawing processing unit 30, and a control unit 40.

[0035] The transfer device 10 is a means of horizontally transferring a flat stage 12 in a generally defined posture on the upper surface of the base 11. The transfer device 10 has a transfer mechanism including a main scanning mechanism 13 and a sub-scanning mechanism 14. The main scanning mechanism 13 is used to transfer the stage 12 along the main scanning direction. The sub-scanning mechanism 14 is used to transfer the stage 12 along the sub-scanning direction. The substrate W is held horizontally on the upper surface of the stage 12 and moves together with the stage 12 along the main scanning direction and the sub-scanning direction.

[0036] The more detailed structure of the conveying device 10 is described below.

[0037] The frame 20 is a structure for holding the drawing processing unit 30 above the base 11. The frame 20 has a pair of support sections 21 and a bridging section 22. The pair of support sections 21 are spaced apart in the sub-scanning direction. Each support section 21 extends upward from the upper surface of the base 11. The bridging section 22 extends in the sub-scanning direction between the upper ends of the two support sections 21. The stage 12 holding the substrate W can pass between the pair of support sections 21 and below the bridging section 22.

[0038] The drawing processing unit 30 includes two optical heads 31, an illumination optical system 32, a laser oscillator 33, and a laser drive unit 34. The two optical heads 31 are fixed to the bridging unit 22 at a distance from each other in the sub-scanning direction. The illumination optical system 32, the laser oscillator 33, and the laser drive unit 34 are housed, for example, within the internal space of the bridging unit 22. The laser drive unit 34 is electrically connected to the laser oscillator 33. When the laser drive unit 34 is activated, pulsed light is emitted from the laser oscillator 33. Furthermore, the pulsed light emitted from the laser oscillator 33 is guided into the optical heads 31 through the illumination optical system 32.

[0039] An optical system including a spatial modulator is provided inside the optical head 31. For example, a GLV (Grating Light Valve) (registered trademark) is used as a diffraction grating type spatial light modulator. Pulsed light introduced into the optical head 31 is modulated into a predetermined pattern by the spatial modulator and irradiates the upper surface of the substrate W. As a result, photosensitive materials such as resist coated on the upper surface of the substrate W are exposed.

[0040] When the drawing apparatus 1 is operating, the exposure of the optical head 31 and the transport of the substrate W by the transport device 10 are repeatedly performed. Specifically, while the stage 12 is transported along the sub-scanning direction by the sub-scanning mechanism 14, pulsed light from the optical head 31 is applied to expose a strip-shaped area (swath) extending along the sub-scanning direction. Then, the stage 12 is transported along the main scanning direction by the main scanning mechanism 13 in the amount of one swath. By repeatedly performing this exposure in the sub-scanning direction and transport of the stage 12 in the main scanning direction, the drawing apparatus 1 draws a pattern on the entire upper surface of the substrate W.

[0041] The control unit 40 is a mechanism for controlling the operation of various parts of the drawing device 1. Figure 3 This is a block diagram showing the electrical connections between the control unit 40 and the various parts within the drawing device 1. For example... Figure 3 As schematically shown, the control unit 40 is composed of a computer having a processor 41 such as a CPU, a memory 42 such as RAM, and a storage unit 43 such as a hard disk drive. The storage unit 43 stores a computer program P for controlling the operation of the drawing device 1.

[0042] In addition, such as Figure 3As shown, the control unit 40 is electrically connected to the drawing processing unit 30 (including the aforementioned optical head 31 and laser drive unit 34), the main scanning mechanism 13 (including the linear motor 61 and air guide 62 described later), the sub-scanning mechanism 14 (including the linear motor 71 described later), the angle measuring device 80 described later, and various sensors 50. The control unit 40 reads the computer program P and data D stored in the storage unit 43 into the memory 42. The processor 41 performs calculations based on the computer program P and data D, thereby controlling the operation of the aforementioned parts within the drawing device 1. Thus, drawing processing is performed in the drawing device 1.

[0043] <2. Structure of the conveying device>

[0044] Next, the detailed structure of the conveying device 10 will be explained. Figure 4 This is a partial cross-sectional view when a portion of the conveying device is cut off by a plane perpendicular to the main scanning direction. For example... Figures 1-4 As shown, the conveying device 10 includes a base 11, a platform 12, a main scanning mechanism 13, a secondary scanning mechanism 14, a support plate 16, an angle measuring device 80, and the aforementioned control unit 40.

[0045] The base 11 is a support platform that supports the various parts of the conveying device 10. The base 11 has a flat, plate-like shape that extends along the main scanning direction and the sub-scanning direction. Four feet 111 and two dampers 112 are provided on the lower surface of the base 11. The lengths of the feet 111 and the dampers 112 can be adjusted individually. Therefore, by adjusting the lengths of the feet 111 and the dampers 112, the posture of the base 11 can be adjusted to be horizontal.

[0046] The support plate 16 and the stage 12 each have a flat, plate-like shape. The support plate 16 is supported on the base 11 and is movable along the main scanning direction via the main scanning mechanism 13. The stage 12 is supported on the support plate 16 and is movable along the sub-scanning direction via the sub-scanning mechanism 14. The stage 12 has an upper surface capable of holding the substrate W. Furthermore, the upper surface of the stage 12 is provided with chuck pins for holding the substrate W or multiple suction holes for adsorbing the substrate W.

[0047] The main scanning mechanism 13 is a mechanism that moves the support plate 16 relative to the base 11 along the main scanning direction. The main scanning mechanism 13 has a pair of linear mechanisms 60. The pair of linear mechanisms 60 are disposed at both ends of the upper surface of the base 11 in the secondary scanning direction. Figure 2 and Figure 4 As shown, a pair of straight-line mechanisms 60 each have a linear motor 61 and an air guide 62.

[0048] The linear motor 61 has a stator 611 and a mover 612. The stator 611 is laid on the upper surface of the base 11 along the main scanning direction. That is, a pair of stators 611 are arranged parallel to each other. The mover 612 is fixed to the support plate 16 by an air bearing 622 described later.

[0049] Additionally, the main scanning mechanism 13 has a control board 63 for controlling the operation of the linear motor 61. For example, Servopack (registered trademark) is used as the control board 63. The control board 63 is electrically connected to the control unit 40. When driving the linear motor 61, the control board 63 calculates the torque to be generated in the linear motor 61 according to the instructions from the control unit 40. Furthermore, a drive signal corresponding to the calculated torque is supplied to the stator 611 of each linear motor 61. In this way, utilizing the magnetic attraction and reaction forces generated between the stator 611 and the mover 612, the mover 612 moves along the stator 611 in the main scanning direction.

[0050] The air guide 62 includes a guide rail 621 and an air bearing 622. The guide rail 621 is laid on the upper surface of the base 11 along the main scanning direction. That is, the stator 611 of the linear motor 61 and the guide rail 621 of the air guide 62 are arranged parallel to each other. The air bearing 622 is fixed to the support plate 16 and the mover 612. In addition, the air bearing 622 is disposed above the guide rail 621.

[0051] like Figure 4 As shown, a gas outlet 623 is provided on the lower surface of the air bearing 622. During operation of the conveying device 10, gas is continuously supplied to the air bearing 622 from the equipment in the factory, and pressurized gas is blown out from the gas outlet 623 onto the upper surface of the guide rail 621. Thus, the air bearing 622 is non-contactly supported on the guide rail 621. Therefore, when the linear motor 61 is driven, the support plate 16 moves smoothly along the main scanning direction with low friction while being supported by the air guide 62.

[0052] The sub-scanning mechanism 14 is a mechanism that moves the stage 12 relative to the support plate 16 along the sub-scanning direction. The sub-scanning mechanism 14 has a linear motor 71 and a pair of guide mechanisms 72.

[0053] A linear motor 71 is positioned approximately at the center of the main scanning direction on the upper surface of the support plate 16. The linear motor 71 has a stator 711 and a mover 712. The stator 711 is laid along the sub-scanning direction on the upper surface of the support plate 16. The mover 712 is fixed to the stage 12. When the linear motor 71 is driven, the mover 712 moves along the stator 711 in the sub-scanning direction by utilizing the attractive and reactive forces generated between the stator 711 and the mover 712.

[0054] A pair of guide mechanisms 72 are disposed at both ends of the upper surface of the support plate 16 in the main scanning direction. Each guide mechanism 72 includes a guide rail 721 and a ball bearing 722. The guide rail 721 is laid along the sub-scanning direction on the upper surface of the support plate 16. The ball bearing 722 is fixed to the lower surface of the stage 12. Furthermore, the ball bearing 722 is movable along the guide rail 721 in the sub-scanning direction. Therefore, when the linear motor 71 is driven, the stage 12 moves relative to the support plate 16 in the sub-scanning direction.

[0055] In this way, the stage 12 can move relative to the base 11 along the main scanning direction and the sub-scanning direction via the main scanning mechanism 13 and the sub-scanning mechanism 14.

[0056] An angle measuring device 80 is used to measure the deflection angle θ of the stage 12. The angle measuring device 80 includes a reflector 81 fixed to the stage 12 and a laser interferometer 82. The reflector 81 is fixed to the end edge of the stage 12 in the main scanning direction. The laser interferometer 82 is fixed to the upper surface of the base 11. The laser interferometer 82 illuminates two laser beams towards the reflector 81. Furthermore, the optical path difference between the two laser beams is detected based on the interference between the two laser beams reflected from the reflector 81. The deflection angle θ of the stage 12 is then measured based on this optical path difference.

[0057] <3. Transport Control of the Main Scanning Mechanism>

[0058] <3-1. Structure of the Control Unit>

[0059] Figure 5 This is a block diagram schematically illustrating the functions of the control unit 40 used to control the main scanning mechanism 13 described above. For example... Figure 5 As shown, the control unit 40 includes an angle acquisition unit 91 and a transport control unit 92. The functions of the angle acquisition unit 91 and the transport control unit 92 are implemented by a computer, which is the control unit 40, performing operations according to a computer program P.

[0060] The angle acquisition unit 91 inputs the measurement result of the deflection angle θ measured by the angle measuring device 80. The angle acquisition unit 91 sends first input data d1, including the acquired deflection angle θ, to the transport control unit 92. The first input data d1 may only contain the deflection angle θ, or it may include variables other than the deflection angle θ. For example, the first input data d1 may include the deflection angle θ and the angular velocity dθ / dt of the deflection angle θ. In addition, the angle acquisition unit 91 may perform processing such as filtering on the deflection angle θ, and include the processed deflection angle θ in the first input data d1.

[0061] The transport control unit 92 has a first learned model M1. The first learned model M1 is an inference program whose parameters have been adjusted through prior reinforcement learning using a reinforcement learning algorithm. For example, reinforcement learning algorithms used to obtain the first learned model M1 can be Q-learning, DQN (Deep Q Network), SARSA, Monte Carlo methods, epsilon-greedy methods, Boltzmann QPolicy methods, etc.

[0062] The transport control unit 92 inputs the first input data d1 sent by the angle acquisition unit 91 to the first learned model M1. Based on the first input data d1, the first learned model M1 outputs a control value v for controlling a pair of straight-line mechanisms 60. The control value v can, for example, be set to the torque value of the linear motor 61 to be input to each straight-line mechanism 60. However, the control value v output by the first learned model M1 can also be a value other than the torque value used to operate each straight-line mechanism 60.

[0063] The transport control unit 92 controls a pair of linear motors 60 based on the control value v output by the first learned model M1. Specifically, the transport control unit 92 sends the control value v to a pair of linear motors 61 via the control board 63. The pair of linear motors 60 operate according to the control value v sent by the transport control unit 92.

[0064] <3-2. Prior reinforcement learning>

[0065] Next, the pre-reinforcement learning performed in the conveying device 10 will be explained. Figure 6 This is a flowchart representing the process of prior reinforcement learning.

[0066] Pre-reinforcement learning is a machine learning process that autonomously learns control rules to maintain the deflection angle θ of the stage 12 within a specified range. During pre-reinforcement learning, the control unit 40 prepares a learning model Mo, which serves as the master model for the first learned model M1, based on the reinforcement learning program. Furthermore, the prepared learning model Mo learns the relationship between the control value v and the first input data d1, including the deflection angle θ.

[0067] Specifically, firstly, the learning model Mo outputs a control value v (step S11). In the initial stage of reinforcement learning, the control value v output in step S11 is a random value. The transport control unit 92 operates a pair of straight-line mechanisms 60 based on the control value v output by the learning model Mo (step S12). Furthermore, the deflection angle θ input from the angle measuring device 80 to the angle acquisition unit 91 is acquired (step S13).

[0068] Next, the control unit 40 assigns an evaluation value corresponding to the deflection angle θ based on the reinforcement learning program (step S14). At this time, the closer the deflection angle θ is to the preset target value, the higher the evaluation value assigned by the control unit 40.

[0069] Next, the control unit 40 determines whether to end the pre-reinforcement learning (step S15). If the evaluation value assigned in step S14 does not reach the desired level, the control unit 40 continues the pre-reinforcement learning (step S15: No).

[0070] While continuing pre-learning, the angle acquisition unit 91 generates first input data d1 including the deflection angle θ acquired in step S13 and sends it to the transport control unit 92. Furthermore, the transport control unit 92 inputs this first input data d1 to the learning model Mo. In this way, the learning model Mo outputs a new control value v (step S11). Then, the processes described in steps S12 to S15 are executed again.

[0071] As described above, in the initial stage of reinforcement learning, the learning model Mo outputs a random control value v. However, during the repeated processing of steps S11 to S15, the learning model Mo attempts to increase, maintain, or decrease the control value v based on the first input data d1. Thus, it automatically learns the relationship between the control value v and the first input data d1, including the deflection angle θ. Furthermore, the learning model Mo gradually obtains a higher evaluation value. That is, the learning model Mo gradually becomes able to output a control value v based on the first input data d1 to maintain the deflection angle θ within a specified range.

[0072] Figure 7 This is a diagram illustrating an example of how repeated processing steps S11 to S15 causes changes in the evaluation value. Figure 7 The horizontal axis indicates the number of times the process in steps S11 to S15 is repeated. Figure 7 The vertical axis shows the evaluation values. For example... Figure 7 As shown, the above evaluation value gradually increases as the model Mo learns during the learning process.

[0073] Finally, when it is determined that the evaluation value assigned in step S14 has reached the desired level (step S15: Yes), the control unit 40 ends the pre-reinforcement learning. Furthermore, the learning model Mo, strengthened through the above pre-reinforcement learning, is taken as the first completed learning model M1. It should be noted that the control unit 40 may also end the pre-reinforcement learning when the number of repetitions in steps S11 to S15 reaches a preset upper limit.

[0074] <3-3. Motion control of a pair of straight-line mechanisms>

[0075] Next, the motion control of a pair of straight-line mechanisms 60, which is executed after reinforcement learning is completed in advance, will be explained. Figure 8 This is a flowchart representing the motion control process of the straight-moving mechanism 60.

[0076] When controlling the operation of a pair of straight-moving mechanisms 60, firstly, the deflection angle θ output by the angle measuring device 80 is input to the angle acquisition unit 91 (step S21). Next, the angle acquisition unit 91 generates first input data d1 including the acquired deflection angle θ and sends it to the transport control unit 92 (step S22).

[0077] The first input data d1 sent from the angle acquisition unit 91 to the transport control unit 92 is the same type of data as the first input data d1 input to the learning model Mo in the aforementioned pre-learning process. For example, when the first input data d1 input to the learning model Mo in the aforementioned pre-learning process includes the deflection angle θ and the angular velocity dθ / dt of the deflection angle θ, the first input data d1 sent to the transport control unit 92 in this step S22 also includes the deflection angle θ and the angular velocity dθ / dt of the deflection angle θ.

[0078] The transport control unit 92 inputs the first input data d1 sent by the angle acquisition unit 91 to the first learned model M1 (step S23). Then, the first learned model M1 outputs a control value v corresponding to the first input data d1 (step S24). The output control value v is a value determined by the first learned model M1 according to the control rules optimized through the aforementioned pre-reinforcement learning. Therefore, the control value v is a value that can maintain the deflection angle θ within a specified range.

[0079] The transport control unit 92 controls a pair of straight-line mechanisms 60 based on the control value v output by the first learned model M1 (step S25). As a result, the pair of straight-line mechanisms 60 operate appropriately, transporting the platform 12 while maintaining the deflection angle θ of the platform 12 within a specified range.

[0080] As described above, in this conveying device 10, a first learned model M1 obtained through prior reinforcement learning is used for control to maintain the deflection angle θ of the stage 12 at the desired state. Therefore, the manufacturer or user of the conveying device 10 does not need to set parameters for control. As a result, the stage 12 can be conveyed appropriately while reducing the workload of setting parameters for control.

[0081] <4. Variations>

[0082] The above describes one embodiment of the present invention, but the present invention is not limited to the above embodiment. Hereinafter, various modifications will be described focusing on the differences from the above embodiment.

[0083] <4-1. First Variation>

[0084] Figure 9 A block diagram schematically illustrating the function of the control unit 40 in the first modified example. Figure 9 In this example, the control unit 40 has an angle estimation unit 93. The function of the angle estimation unit 93 is achieved by the computer, which is the control unit 40, performing actions according to the computer program P.

[0085] The angle estimation unit 93 has a second learned model M2. The second learned model M2 is an inference program whose parameters have been adjusted through prior learning using a machine learning algorithm. As the machine learning algorithm used to obtain the second learned model M2, for example, a neural network including a single-layer neural network, deep learning, etc., a decision tree algorithm including random forest, gradient advancement, etc., and a supervised machine learning algorithm such as support vector machine can be used.

[0086] The measured values ​​output by various sensors 50 are input into the angle estimation unit 93. The measured values ​​may be, for example, the torque values ​​of a pair of linear motors 61 of the main scanning mechanism 13, or the values ​​of the air pressure of the air bearing 622, the temperature of the guide rail 621, the drive sound of the conveying device 10, the vibration of the stage 12, the position of the stage 12, etc.

[0087] The angle estimation unit 93 inputs second input data d2, which includes the input measurement values, into the second learned model M2. The second input data d2 may consist only of the measurement values, or it may include variables other than the measurement values.

[0088] The second learned model M2 learns the relationship between the second input data d2 and the deflection angle θ of the stage 12 through pre-learning. In the pre-learning, the measurement results of the angle measuring device 80 are used as training data, and the aforementioned machine learning algorithm is used to learn the relationship between the second input data d2 and the deflection angle θ. Therefore, after the pre-learning is completed, when the second input data d2 is input into the second learned model M2, the second learned model M2 outputs an estimated result of the deflection angle θ.

[0089] The angle estimation unit 93 inputs the estimated deflection angle θ output by the second learned model M2 to the angle acquisition unit 91. In this way, the angle measuring device 80 only needs to be set up during the pre-learning period, and can be removed and used after the pre-learning is completed. Therefore, a pair of linear mechanisms 60 can be controlled based on the deflection angle θ of the stage 12 without always setting up a large number of angle measuring devices 80.

[0090] <4-2. Other variations>

[0091] The conveying device 10 described in the above embodiment has not only a main scanning mechanism 13, but also a secondary scanning mechanism 14. However, the present invention can also be applied to conveying devices that do not have a secondary scanning mechanism 14.

[0092] Furthermore, the conveying device 10 of the above embodiment is mounted on the drawing device 1. However, the present invention can also be applied to conveying devices mounted on devices other than the drawing device 1. For example, the conveying device can also be mounted on a device for coating a substrate held by a stage with a processing liquid. In addition, the conveying device can also be mounted on a device for printing on a recording medium held on a stage.

[0093] Furthermore, the angle measuring device 80 of the above embodiment measures the deflection angle θ of the stage 12 using a laser interferometer 82. However, the deflection angle θ of the stage 12 can also be measured by other methods. For example, the deflection angle θ of the stage 12 can also be measured based on an image of the stage 12 acquired by a camera.

[0094] Furthermore, the linear mechanism 60 in the above embodiment includes a linear motor 61. However, a mechanism that converts the rotational motion output by a rotary motor into linear motion via a ball screw can be used instead of the linear motor 61.

[0095] Furthermore, without creating contradictions, the elements appearing in the above-described implementation methods and variations can be appropriately combined.

[0096] Explanation of reference numerals in the attached figures

[0097] 1: Drawing device

[0098] 10: Transport device

[0099] 11: Abutment

[0100] 12: Stage

[0101] 13: Main scanning mechanism

[0102] 14: Sub-scanning mechanism

[0103] 16: Support plate

[0104] 20: Framework

[0105] 30: Drawing Processing Department

[0106] 40: Control Department

[0107] 50: Sensor

[0108] 60: Straight-through agency

[0109] 61: Linear Motor

[0110] 62: Air guide

[0111] 63: Control board

[0112] 71: Linear Motor

[0113] 72: Guiding Organization

[0114] 80: Angle measuring device

[0115] 91: Angle Acquisition Department

[0116] 92: Transport Control Department

[0117] 93: Angle estimation section

[0118] M1: First learned model

[0119] M2: Second learned model

[0120] W: substrate

[0121] d1: First input data

[0122] d2: Second input data

[0123] θ: Deflection angle

[0124] v: Control value

Claims

1. A conveying device for conveying a flat platform horizontally via a pair of linear mechanisms, wherein, have: Angle acquisition unit acquires the deflection angle of the stage; as well as The conveying control unit controls the pair of straight-line mechanisms based on the deflection angle obtained by the angle acquisition unit. The conveying control unit inputs first input data, including the deflection angle acquired by the angle acquisition unit, into a first learned model obtained through reinforcement learning for maintaining the deflection angle in the desired state, and controls the pair of straight-line mechanisms based on the control value output by the first learned model.

2. The conveying device as claimed in claim 1, wherein, It also includes an angle measuring device for measuring the deflection angle of the stage. The angle acquisition unit acquires the measurement result of the deflection angle by the angle measuring device.

3. The conveying device as described in claim 1, wherein, It also includes an angle estimation unit that estimates the deflection angle of the stage based on measurements output by the conveying mechanism, which includes the pair of straight-moving mechanisms. The angle estimation unit uses a second learned model to estimate the deflection angle. This second learned model is obtained through supervised machine learning for estimating the deflection angle of the stage based on second input data including the measured values. The angle acquisition unit acquires the estimation result of the angle estimation unit for the deflection angle.

4. The conveying device according to any one of claims 1 to 3, wherein, The control value is the torque value of the pair of straight-line mechanisms.

5. The conveying device according to any one of claims 1 to 3, wherein, The reinforcement learning is a machine learning process performed with the goal of maintaining the deflection angle within a specified range.

6. The conveying device according to any one of claims 1 to 3, wherein, The first input data includes the deflection angle and the angular velocity of the deflection angle.

7. A conveying method for conveying a flat platform horizontally via a pair of linear mechanisms, comprising: Step a) Obtain the deflection angle of the stage; as well as Step b), controlling the pair of straight-line mechanisms based on the deflection angle obtained through step a). In step b), the first learned model, obtained through reinforcement learning for maintaining the deflection angle in the desired state, is input with first input data including the deflection angle obtained through step a), and the pair of straight-line mechanisms are controlled based on the control value output by the first learned model.

Images