Coating apparatus

Abstract

This invention provides a technique for appropriately detecting unforeseen and unknown anomalies in discharge pressure. The pressure measurement unit 911 acquires pressure data. The feature quantity calculation unit 913 calculates the feature quantity of the pressure data. The anomaly calculation unit 915 uses an unsupervised learning model M1 to calculate the anomaly degree related to the feature quantity calculated by the feature quantity calculation unit 913. The unsupervised learning model M1 takes the feature quantity as input data and outputs the anomaly degree representing the degree of deviation from the distribution of the normal feature quantity. The anomaly probability calculation unit 917 uses a supervised learning model M2 to calculate the anomaly probability related to the feature quantity calculated by the feature quantity calculation unit 913. The supervised learning model M2 takes the feature quantity as input data and outputs the anomaly probability representing the probability corresponding to specific anomalies A and B. The anomaly determination unit 919 uses the anomaly degree calculated by the anomaly calculation unit 915 and the anomaly probability calculated by the anomaly probability calculation unit 917 to determine the anomaly of the pressure data.

Description

Discharge Pressure Monitoring Device, Discharge Pressure Monitoring Method, Computer Program, and Coating Device The subject matter disclosed by the present invention relates to a discharge pressure monitoring device, a discharge pressure monitoring method, a computer program, and a coating device. In the manufacturing steps of a flat panel display, a device called a coater is used. The coater discharges a processing liquid from a slit nozzle by driving a pump and coats the entire conveyed substrate with the processing liquid. In recent years, with the improvement of product quality, this coater is required to coat the processing liquid in such a way that the film thickness of the processing liquid is uniform over the entire substrate. In order to maintain the uniformity of the film thickness, it is very important to monitor the discharge pressure (discharge flow rate) of the processing liquid during production. For example, in Patent Document 1, an attempt is made to optimize the parameters associated with the discharge pressure. However, it is difficult to detect an abnormality in a minute change of the discharge pressure obtained from a pressure sensor using a simple threshold value. For this, as a general technique, many methods, device systems for performing anomaly detection using machine learning have been proposed. For example, in Patent Document 2, a system for discriminating the type (pattern) of an abnormality based on the provided sensor information is proposed. By constructing such a system, in addition to formulating specific countermeasure plans such as component replacement, it can also function as a failure prediction system that can grasp a slight change (omen) that has the possibility of developing into a failure. [Prior Art Documents] [Patent Documents] [Patent Document 1] Japanese Patent Application Laid-Open No. 2020-040046 [Patent Document 2] Japanese Patent Application Laid-Open No. 2022-125288 (Problems to be Solved by the Invention) In Patent Document 2, as a system for predicting a failure, a method using unsupervised learning and a method using supervised learning have been proposed. However, unsupervised learning is a method of learning data in a normal state and evaluating the degree of deviation of newly measured data from the normal state. Therefore, although it can detect an abnormality, it cannot discriminate the type of the abnormality. Also, supervised learning can discriminate the type of an abnormality by learning supervised data with labels of the type of abnormality pre-labeled. However, no matter what kind of abnormality it is, it will be classified as the type of abnormality预想 in advance during learning. Therefore, it cannot cope with an unknown abnormality. An object of the present invention is to provide a technique capable of appropriately detecting an unknown abnormality in the discharge pressure that has not been预想 in advance. (Technical Means for Solving the Problem) In order to solve the above problems, a first aspect of the present invention is a discharge pressure monitoring device, which includes: a pressure data acquisition unit that acquires pressure data representing the time change of the pressure in the nozzle that discharges the processing liquid; a feature amount calculation unit that calculates the feature amount of the pressure data; an abnormality degree calculation unit that uses an unsupervised learning model to calculate the abnormality degree related to the feature amount calculated by the feature amount calculation unit, where the unsupervised learning model uses the feature amount as input data and outputs the abnormality degree indicating the degree of deviation from the distribution of the normal feature amount; an abnormality probability calculation unit that uses a supervised learning model to calculate the abnormality probability related to the feature amount calculated by the feature amount calculation unit, where the supervised learning model uses the feature amount as input data and outputs the abnormality probability indicating the probability corresponding to a specific abnormality; and an abnormality determination unit that uses the abnormality degree calculated by the abnormality degree calculation unit and the abnormality probability calculated by the abnormality probability calculation unit to determine the abnormality of the pressure data. A second aspect is the discharge pressure monitoring device according to the first aspect, where the unsupervised learning model is a model obtained by unsupervised learning that uses only the feature amounts of a plurality of normal pressure data as input data. A third aspect is the discharge pressure monitoring device according to the first aspect or the second aspect, where the unsupervised learning model outputs the Mahalanobis distance as the abnormality degree. A fourth aspect is the discharge pressure monitoring device according to the first aspect or the second aspect, where the supervised learning model outputs the probabilities corresponding to a plurality of specific abnormalities. A fifth aspect is a discharge pressure monitoring device, which includes: a pressure data acquisition unit that acquires pressure data representing the time change of the pressure in the nozzle that discharges the processing liquid; a feature amount calculation unit that calculates the feature amount of the pressure data; an abnormality degree calculation unit that calculates the abnormality degree indicating the degree of deviation from the normal feature amount for the feature amount calculated by the feature amount calculation unit; an abnormality probability calculation unit that uses a supervised learning model to calculate the abnormality probability related to the feature amount calculated by the feature amount calculation unit, where the supervised learning model uses the feature amount as input data and outputs the abnormality probability indicating the probability corresponding to a specific abnormality; and an abnormality determination unit that uses the abnormality degree calculated by the abnormality degree calculation unit and the abnormality probability calculated by the abnormality probability calculation unit to determine the abnormality of the pressure data. The sixth aspect is a discharge pressure monitoring method, which includes: a) a step of obtaining pressure data representing the time change of the pressure in the nozzle that discharges the processing liquid; b) a step of calculating a feature amount of the aforementioned pressure data; c) a step of using an unsupervised learning model to calculate an abnormality degree related to the aforementioned feature amount calculated in step b) above, where the aforementioned unsupervised learning model uses the aforementioned feature amount as input data and outputs the aforementioned abnormality degree representing the degree of deviation from the distribution of the normal aforementioned feature amount; d) a step of using a supervised learning model to calculate an abnormality probability related to the aforementioned feature amount calculated in step b) above, where the aforementioned supervised learning model uses the aforementioned feature amount as input data and outputs the aforementioned abnormality probability representing the probability corresponding to a specific abnormality; and e) a step of using the aforementioned abnormality degree calculated in step c) above and the aforementioned abnormality probability calculated in step d) above to determine the abnormality of the aforementioned pressure data. The seventh aspect is a discharge pressure monitoring method, which includes: A) a step of obtaining pressure data representing the time change of the pressure in the nozzle that discharges the processing liquid; B) a step of calculating a feature amount of the aforementioned pressure data; C) a step of calculating, for the aforementioned feature amount calculated in step B) above, an abnormality degree representing the degree of deviation from the normal aforementioned feature amount; D) a step of using a supervised learning model to calculate an abnormality probability related to the aforementioned feature amount calculated in step B) above, where the aforementioned supervised learning model uses the aforementioned feature amount as input data and outputs the aforementioned abnormality probability representing the probability corresponding to a specific abnormality; and E) a step of using the aforementioned abnormality degree calculated in step C) above and the aforementioned abnormality probability calculated in step D) above to determine the abnormality of the aforementioned pressure data. The eighth aspect is a computer-executable computer program that causes the aforementioned computer to execute the discharge pressure monitoring method of the sixth or seventh aspect. The ninth aspect is a coating device, which includes: a substrate holding part that holds a substrate; a nozzle that discharges a processing liquid toward the aforementioned substrate held by the aforementioned substrate holding part; a pressure sensor that measures the pressure in the aforementioned nozzle; and the discharge pressure monitoring device of the first, second, or fifth aspect. (Effect compared with the prior art) According to the first to ninth aspects, when determining a pre-conceived abnormality, the abnormality can be discriminated by calculating the abnormality probability. Also, even when an unknown abnormality that has not been pre-conceived has occurred, the unknown abnormality can be appropriately detected by calculating the abnormality degree. According to the discharge pressure monitoring device of the second aspect, since only the distribution of normal feature amounts can be learned, abnormality detection can be appropriately performed. The ejection pressure monitoring device according to the third aspect can more accurately grasp the characteristic pattern or abnormal operation of the pressure data because it calculates the abnormality degree in consideration of the correlation relationship between the characteristic quantities representing the pressure data among the respective dimensions. The ejection pressure monitoring device according to the fourth aspect can determine whether the pressure data corresponds to one of a plurality of types of abnormalities because it can calculate the probabilities of a plurality of specific abnormalities. Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. Furthermore, the constituent elements described in these embodiments are merely illustrative, and the scope of the present invention is not limited only to these constituent elements. In the drawings, there are cases where the dimensions or quantities of respective parts are exaggerated or simplified as needed for easy understanding. FIG. 1 is a diagram schematically showing the overall configuration of the coating device 1 of the embodiment. The coating device 1 is a substrate processing device that coats a coating liquid on the upper surface Sf of a substrate S. For example, the substrate S is a glass substrate for a liquid crystal display device. Furthermore, the substrate S may also be various substrates to be processed for electronic devices such as a semiconductor wafer, a glass substrate for a photomask, a glass substrate for a plasma display, a glass substrate for a magnetic disk and an optical disk, or a ceramic substrate, a glass substrate for an organic EL, a glass substrate for a solar cell, or a silicon substrate, other flexible substrates, and a printed circuit board. The coating device 1 is, for example, a slit coater. In FIG. 1, a coordinate system XYZ is defined in order to explain the arrangement relationship of the respective elements of the coating device 1. The conveyance direction of the substrate S is the "X direction". The direction in which the substrate S travels in the X direction (the direction toward the downstream of the conveyance direction) is the +X direction, and the opposite direction (the direction toward the upstream of the conveyance direction) is the -X direction. Also, the direction orthogonal to the X direction is the Y direction, and the direction orthogonal to the X direction and the Y direction is the Z direction. In the following description, the Z direction is set as the vertical direction, and the X direction and the Y direction are set as the horizontal directions. In the Z direction, the +Z direction is set as the upward direction, and the -Z direction is set as the downward direction. The coating device 1 sequentially includes, in the +X direction: an input conveyor 100, an input transfer unit 2, a floating table unit 3, an output transfer unit 4, and an output conveyor 110. The input conveyor 100, the input transfer unit 2, the floating table unit 3, the output transfer unit 4, and the output conveyor 110 form a conveyance path through which the substrate S passes. Also, the coating device 1 further includes: a substrate conveyance unit 5, a coating mechanism 7, a coating liquid supply mechanism 8, and a control unit 9. The substrate S is transported from a device or the like on the upstream side of the coating device 1 to the input conveyor 100. The input conveyor 100 includes a roller conveyor 101 and a rotary drive mechanism 102. The rotary drive mechanism 102 rotates each roller of the roller conveyor 101. By the rotation of each roller of the roller conveyor 101, the substrate S is transported downstream (+X direction) in a horizontal posture. The "horizontal posture" means a state in which the main surface (the surface with the largest area) of the substrate S is parallel to the horizontal plane (XY plane). The input transfer unit 2 includes a roller conveyor 21 and a rotary lifting drive mechanism 22. The rotary lifting drive mechanism 22 rotates each roller of the roller conveyor 21 and lifts the roller conveyor 21. By the rotation of the roller conveyor 21, the substrate S is transported downstream (+X direction) in a horizontal posture. Also, by the lifting of the roller conveyor 21, the position of the substrate S in the Z direction is changed. The substrate S is transferred from the input conveyor 100 to the floating table portion 3 via the input transfer unit 2. As shown in FIG. 1, the floating table portion 3 is substantially flat. The floating table portion 3 is divided into three parts in the X direction. The floating table portion 3 sequentially includes, in the +X direction: an entrance floating table 31, a coating table 32, and an exit floating table 33. The upper surfaces of the entrance floating table 31, the coating table 32, and the exit floating table 33 are located on the same plane. The floating table portion 3 further includes a lift pin drive mechanism 34, a floating control mechanism 35, and a lift drive mechanism 36. The lift pin drive mechanism 34 raises and lowers a plurality of lift pins disposed on the entrance floating table 31. The floating control mechanism 35 supplies compressed air for floating the substrate S to the entrance floating table 31, the coating table 32, and the exit floating table 33. The lift drive mechanism 36 raises and lowers the exit floating table 33. On the upper surfaces of the entrance floating table 31 and the exit floating table 33, a plurality of ejection holes for ejecting the compressed air supplied from the floating control mechanism 35 are arranged in a matrix. When the compressed air is ejected from each ejection hole, the substrate S floats upward relative to the floating table portion 3. Thus, the lower surface Sb of the substrate S is separated from the upper surface of the floating table portion 3 and is supported in a horizontal posture. The distance (floating amount) between the lower surface Sb of the substrate S and the upper surface of the floating table portion 3 in the state where the substrate S is floating is, for example, 10 μm or more and 500 μm or less. On the upper surface of the coating table 32, ejection holes for ejecting compressed air supplied from the self-floating control mechanism 35 and suction holes for sucking gas are alternately arranged in the X direction and the Y direction. The floating control mechanism 35 controls the ejection amount of the compressed air from the ejection holes and the suction amount of the air from the suction holes. Thereby, it precisely controls the floating amount of the substrate S relative to the coating table 32 so that the position of the upper surface Sf of the substrate S passing above the coating table 32 in the Z direction becomes a specified value. Furthermore, the floating amount of the substrate S relative to the coating table 32 is calculated by the control unit 9 based on the detection results of the sensors 61 or 62 described later. Also, the floating amount of the substrate S relative to the coating table 32 is preferably adjustable with high precision by airflow control. The substrate S carried into the floating table portion 3 is applied with a propelling force in the +X direction from the roller conveyor 21 and is conveyed onto the inlet floating table 31. The inlet floating table 31, the coating table 32, and the outlet floating table 33 support the substrate S in a floating state. As the floating table portion 3, for example, the configuration described in Japanese Patent No. 5346643 can also be adopted. The substrate transfer unit 5 is disposed below the floating table portion 3. The substrate transfer unit 5 includes a chuck mechanism 51 and an adsorption transfer control mechanism 52. The chuck mechanism 51 includes an adsorption pad (not shown) provided on the adsorption member. The chuck mechanism 51 supports the substrate S from below by bringing the adsorption pad into contact with the peripheral portion of the lower surface Sb of the substrate S. The adsorption transfer control mechanism 52 adsorbs the substrate S to the adsorption pad by applying a negative pressure to the adsorption pad. Also, the adsorption transfer control mechanism 52 moves the substrate transfer unit 5 back and forth in the X direction. The chuck mechanism 51 holds the substrate S in a state where the lower surface Sb of the substrate S is at a position higher than the upper surface of the floating table portion 3. The substrate S maintains a horizontal posture by the buoyancy applied from the floating table portion 3 while being held at the peripheral portion by the chuck mechanism 51. As shown in FIG. 1, the coating apparatus 1 includes a sensor 61 for measuring the plate thickness. The sensor 61 is disposed near the roller conveyor 21. The sensor 61 detects the position of the upper surface Sf of the substrate S held by the chuck mechanism 51 in the Z direction. Also, since a chuck (not shown) in a state where the substrate S is not held is located directly below the sensor 61, the sensor 61 can detect the position of the upper surface of the adsorption member, that is, the adsorption surface, in the vertical direction Z. The chuck mechanism 51 holds the substrate S carried into the floating table portion 3 and moves in the +X direction. Thereby, the substrate S is conveyed from above the inlet floating table 31 via above the coating table 32 to above the outlet floating table 33. Then, the substrate S moves from the outlet floating table 33 to the output transfer unit 4. The output transfer unit 4 moves the substrate S from a position above the exit floating table 33 toward the output conveyor 110. The output transfer unit 4 includes a roller conveyor 41 and a rotary lifting drive mechanism 42. The rotary lifting drive mechanism 42 rotationally drives the roller conveyor 41 and raises and lowers the roller conveyor 41 in the Z direction. By the rotation of each roller of the roller conveyor 41, the substrate S moves in the +X direction. Also, by the raising and lowering of the roller conveyor 41, the substrate S is displaced in the Z direction. The output conveyor 110 includes a roller conveyor 111 and a rotary drive mechanism 112. The output conveyor 110 conveys the substrate S in the +X direction by the rotation of each roller of the roller conveyor 111 and discharges the substrate S out of the coating device 1. Furthermore, the input conveyor 100 and the output conveyor 110 are part of the coating device 1. However, the input conveyor 100 and the output conveyor 110 may also be assembled to a device different from the coating device 1. The coating mechanism 7 coats the coating liquid on the upper surface Sf of the substrate S. The coating mechanism 7 is disposed above the conveyance path of the substrate S. The coating mechanism 7 has a nozzle 71. The nozzle 71 is a slit nozzle having a slit-shaped discharge port on the lower surface. The nozzle 71 is connected to a positioning mechanism (not shown). The positioning mechanism moves the nozzle 71 between a coating position (the position shown by a solid line in FIG. 1) above the coating table 32 and a maintenance position described later. The coating liquid supply mechanism 8 is connected to the nozzle 71. The coating liquid supply mechanism 8 supplies the coating liquid to the nozzle 71 and discharges the coating liquid from the discharge port disposed on the lower surface of the nozzle 71. FIG. 2 is a diagram showing the configuration of the coating liquid supply mechanism 8. The coating liquid supply mechanism 8 includes: a pump 81, a pipe 82, a coating liquid replenishing unit 83, a pipe 84, an on-off valve 85, a pressure sensor 86, and a drive unit 87. The pump 81 is a supply source for supplying the coating liquid to the nozzle 71, and it supplies the coating liquid by a volume change. The pump 81 may also be a corrugated tube type pump described in, for example, Japanese Patent Laid-Open No. 10-61558. As shown in FIG. 2, the pump 81 has a flexible tube 811 that is elastically expandable and contractible in the radial direction. One end of the flexible tube 811 is connected to the coating liquid replenishing unit 83 via the pipe 82. The other end of the flexible tube 811 is connected to the nozzle 71 via the pipe 84. The pump 81 has a corrugated tube 812 that is elastically deformable in the axial direction. The corrugated tube 812 has a small corrugated tube portion 813, a large corrugated tube portion 814, a pump chamber 815, and an actuator disk portion 816. The pump chamber 815 is disposed between the flexible tube 811 and the corrugated tube 812. A non-compressible medium is enclosed in the pump chamber 815. The actuator disk portion 816 is connected to the drive unit 87. The coating liquid replenishing unit 83 has a storage tank 831 for storing the coating liquid. The storage tank 831 is connected to the pump 81 via a pipe 82. A shut-off valve 833 is inserted in the pipe 82. The shut-off valve 833 opens and closes according to an instruction from the control unit 9. When the shut-off valve 833 is open, it can replenish the coating liquid from the storage tank 831 to the flexible pipe 811 of the pump 81. Also, when the shut-off valve 833 is closed, it restricts the replenishment of the coating liquid from the storage tank 831 to the flexible pipe 811 of the pump 81. The pipe 84 is connected to the output side of the pump 81. A shut-off valve 85 is inserted in the pipe 84. The shut-off valve 85 opens and closes according to an instruction from the control unit 9. By opening and closing the shut-off valve 85, it switches the conveyance and the stop of the conveyance of the coating liquid to the nozzle 71. A pressure sensor 86 is disposed in the pipe 84. The pressure sensor 86 detects the pressure (discharge pressure) of the coating liquid being conveyed to the nozzle 71, and outputs a signal indicating the detected pressure value to the control unit 9. Figure 3 is a diagram showing the pressure waveform of the discharge pressure. In Figure 3, the horizontal axis represents time, and the vertical axis represents the pressure value. In the coating device 1, by adjusting various parameters (acceleration time, steady speed, steady speed time, deceleration time, etc.) of the movement of the specified actuator part 816, it can appropriately perform an optimization process in which the pressure waveform of the coating liquid discharged from the nozzle 71 approximates the ideal pressure waveform. As shown in Figure 3, the discharge pressure is measured during the period from before the coating liquid starts to be discharged from the nozzle 71 until after the coating liquid finishes being discharged from the nozzle 71. In the example shown in Figure 3, the discharge pressure at the time ta when the coating liquid starts to be discharged from the nozzle 71 and the discharge pressure at the time te when the coating liquid finishes being discharged from the nozzle 71 are the initial pressures Pi. However, the pressures at the start and end of discharge do not always coincide with the initial pressure Pi. As shown in Figure 3, the discharge period is divided into a rising period T1, a transition period T2, a steady period T3, and a falling period T4. The rising period T1 is the period from the time ta when the coating liquid supply mechanism 8 starts to discharge the coating liquid from the nozzle 71 (that is, the time when the coating liquid supply mechanism 8 starts to move the actuator part 816) until the time tb when the discharge pressure reaches the target pressure Pt. That is, when the coating liquid starts to be discharged from the nozzle 71 at the time ta, the discharge pressure increases from the initial pressure Pi to the target pressure Pt between the time ta and the time tb. The transition period T2 is the period from the time tb until the time tc after a predetermined vibration attenuation period has elapsed. This vibration attenuation period is the period required for the time change of the discharge pressure to stabilize, and is set, for example, by an input operation of the user to the input device 97, and is stored in the storage unit 93. The stabilization period T3 is the period from time tc to the time td when the coating liquid supply mechanism 8 starts to decrease the discharge pressure (i.e., the time when the coating liquid supply mechanism 8 starts to decelerate from the target speed of the turntable portion 816). That is, between time tc and time td, the coating liquid supply mechanism 8 moves the turntable portion 816 at a constant speed and starts to decelerate the turntable portion 816 at time td. Furthermore, during the stabilization period T3, the discharge pressure is basically stabilized at the target pressure Pt. However, even during the stabilization period T3, the time variation of the discharge pressure includes minute vibrations, and the discharge pressure becomes greater than or less than the target pressure Pt. The descent period T4 is the period from time td to the time te when the coating liquid supply mechanism 8 finishes discharging the coating liquid from the nozzle 71 (i.e., the time te when the coating liquid supply mechanism 8 stops the turntable portion 816). That is, the discharge pressure decreases to the initial pressure Pi between time td and time te, and at time te, the discharge of the coating liquid from the nozzle 71 stops. As shown in FIGS. 1 and 2, a sensor 62 is disposed at the nozzle 71 that supplies the coating liquid by the coating liquid supply mechanism 8. The sensor 62 non - contactedly detects the height of the substrate S in the Z direction. The sensor 62 is electrically connected to the control unit 9. Based on the detection result of the sensor 62, the control unit 9 measures the distance (separation distance) between the floated substrate S and the upper surface of the coating table 32. Then, the control unit 9 adjusts the coating position of the nozzle 71 by the positioning mechanism according to the measured separation distance. Furthermore, as the sensor 62, an optical sensor or an ultrasonic sensor can be suitably used. The coating mechanism 7 is provided with a nozzle cleaning standby unit 72. The nozzle cleaning standby unit 72 performs predetermined maintenance on the nozzle 71 disposed at the maintenance position. The nozzle cleaning standby unit 72 has a roller 721, a cleaning unit 722, and a roller groove 723. The nozzle cleaning standby unit 72 adjusts the discharge port of the nozzle 71 to a state suitable for the coating process by cleaning the nozzle 71 and forming accumulated liquid. Also, in the coating apparatus 1, in order to evaluate the discharge pressure applied to the coating liquid, a simulated discharge of the coating liquid from the nozzle 71 is performed in a state where the nozzle 71 is disposed at the maintenance position. FIG. 4 is a block diagram showing a configuration example of the control unit 9. The control unit 9 controls the operations of the respective components of the coating apparatus 1. The control unit 9 includes: an arithmetic unit 91, a storage unit 93, a display 95, and an input device 97. As the control unit 9, for example, a desktop, laptop, or tablet computer can be used. As described below, the control unit 9 functions as a discharge pressure monitoring device that monitors abnormalities in the discharge pressure. The calculation unit 91 is a processor composed of a CPU (Central Processing Unit), etc. The storage unit 93 is composed of a temporary storage device such as a RAM (Random Access Memory), and a non-temporary auxiliary storage device such as an HDD (Hard Disk Drive) and an SDD (Solid State Drive). The display 95 is a device that displays information to the user. Specifically, it is a liquid crystal display or the like. The input device 97 is a device that accepts the input operation of the user, and it is a mouse, a keyboard, etc. The storage unit 93 stores the computer program 931. The computer program 931 is provided by the recording medium M. That is, the recording medium M records the computer program 931 in a readable manner by the computer, i.e., the control unit 9. The recording medium M is, for example, a USB (Universal Serial Bus) memory, an optical disc such as a DVD (Digital Versatile Disc), a magnetic disk, etc. By executing the computer program 931, the calculation unit 91 functions as a discharge control unit 910, a discharge pressure measurement unit 911, a feature amount calculation unit 913, an abnormality degree calculation unit 915, an abnormality probability calculation unit 917, and an abnormality determination unit 919. The discharge control unit 910 controls the operation (supply operation) of the pump 81 that supplies the coating liquid to the nozzle 71 according to the preset parameters. The discharge pressure measurement unit 911 measures the discharge pressure. Specifically, the discharge pressure measurement unit 911 periodically obtains the discharge pressure measured by the pressure sensor 86 in a predetermined sampling period. The discharge pressure measurement unit 911 obtains the time change (time series data) of the discharge pressure applied to the coating liquid during the period when the coating liquid is discharged from the nozzle 71, and stores the obtained data as pressure data in the storage unit 93. The pressure data represents the data of each moment and the pressure measured at each moment. The discharge pressure measurement unit 911 is an example of a pressure data acquisition unit. The feature amount calculation unit 913 derives a feature amount based on the discharge pressure measured by the discharge pressure measurement unit 911. The feature amount can adopt, for example, several feature amounts described in Japanese Patent Laid-Open No. 2022-138109. FIG. 5 is a diagram for explaining an example of a feature quantity. In this example, the degree of overshoot that occurs during the rise of the discharge pressure is calculated as the feature quantity F1. Specifically, the feature quantity calculation unit 913 obtains the sign (positive or negative) of the second derivative Dif2 of the discharge pressure at the time t11 when the discharge pressure reaches the maximum value Pmax. Further, the feature quantity calculation unit 913 calculates the time t12 when the sign of the second derivative switches twice since the time t11. Then, the feature quantity calculation unit 913 obtains the feature quantity F1 of the time change of the discharge pressure during the initial vibration period T2_s from the time t11 to t12. For example, the feature quantity calculation unit 913 selects the pressure of the smaller one between the minimum value Pmin of the discharge pressure during the initial vibration period T2_s and the stable pressure Pm (the average value of the discharge pressure during the stable period T3) as the target pressure Pg. Further, the feature quantity calculation unit 913 may calculate the difference between the maximum pressure Pmax and the target pressure Pg (=Pmax - Pg) as the feature quantity F1. With the rising trend of the discharge pressure, the greater the overshoot of the time change of the discharge pressure, the relatively greater such a feature quantity F1. FIG. 6 is a diagram for explaining another example of a feature quantity. In this example, the stability of the time change of the discharge pressure during the transition period T2 is calculated as the feature quantity F2. Specifically, the feature quantity calculation unit 913 calculates the root mean square error RMSE(P_measure, Pm) between the discharge pressure during the transition period T2 and the stable pressure Pm which is the average value of the discharge pressure during the stable period T3 as the feature quantity F2. The greater the ringing (vibration of the waveform) of the time change of the discharge pressure during the transition period T2, the relatively greater such a feature quantity F2. The abnormality calculation unit 915 inputs the feature quantity calculated by the feature quantity calculation unit 913 to the unsupervised learning model M1 to calculate the abnormality degree of the pressure data. The abnormality probability calculation unit 917 inputs the feature quantity calculated by the feature quantity calculation unit 913 to the supervised learning model M2 to calculate the abnormality probability. Specifically, the pressure data can be classified into a normal state and one or more pre-conceived abnormal states. The feature quantity calculation unit 913 calculates the probability corresponding to one or more pre-conceived abnormalities (hereinafter also referred to as "abnormality probability"). For example, when learning the supervised learning model M2, in the case where two types of abnormalities ("abnormality A" and "abnormality B") are pre-conceived, the pressure data is classified into a normal state, a state of abnormality A, and a state of abnormality B. Then, the abnormality probability calculation unit 917 calculates the abnormality probability indicating the probability corresponding to abnormality A or B. The abnormality determination unit 919 uses the abnormality degree calculated by the abnormality calculation unit 915 and the abnormality probability calculated by the abnormality probability calculation unit 917 to determine the abnormality of the pressure data. <Learning Phase> Figure 7 is a block diagram showing the control unit 9 performing machine learning. Also, Figure 8 is a diagram showing the flow of the machine learning process executed by the control unit 9. Furthermore, the machine learning shown in Figures 8 and 9 can also be performed by other computer devices different from the control unit 9. First, as shown in Figures 7 and 8, in order to perform machine learning, a considerable number of normal pressure data D1, a considerable number of pressure data D2 corresponding to the abnormality A, and a considerable number of pressure data D3 corresponding to the abnormality B are respectively acquired in advance by the discharge pressure measurement unit 911 (Figure 8: Step S11). The acquired pressure data D1 to D3 are stored in the storage unit 93. Then, the feature quantity calculation unit 913 acquires the feature quantities of the respective pressure data D1 to D3 (Figure 8: Step S12). The abnormality degree calculation unit 915 performs machine learning (unsupervised learning) for obtaining the unsupervised learning model M1 (Figure 8: Step S13). The abnormality degree calculation unit 915 stores the unsupervised learning model M1 (specifically, the learned parameters) obtained by the machine learning in the storage unit 93 (Step S14). Specifically, in the unsupervised learning, a set of feature quantities of a plurality of normal pressure data D1 is prepared as learning data. Then, machine learning based on the k-nearest neighbor method using the learning data is performed. The unsupervised learning model M1 takes the feature quantity as input data and outputs the abnormality degree indicating the degree of deviation of the distribution of the feature quantities of the normal pressure data. The abnormality degree is preferably the Mahalanobis distance based on k-nearest neighbors, but may also be the Euclidean distance or the Manhattan distance. The abnormality probability calculation unit 917 performs machine learning (supervised learning) for obtaining the supervised learning model M2 (Figure 8: Step S15). In addition, the abnormality probability calculation unit 917 stores the supervised learning model M2 (specifically, the learned parameters) obtained by the machine learning in the storage unit 93 (Figure 8: Step S16). In the supervised learning, first, a plurality of feature quantities of normal pressure data, a plurality of feature quantities of pressure data of the abnormality A, and a plurality of feature quantities of pressure data of the abnormality B are prepared as input data. Also, the basis of the supervised learning model M2 is, for example, a linear regression model. In the supervised learning, the optimal parameters for predicting the target values (labels indicating "normal", "abnormality A", and "abnormality B") are obtained based on the input data. Specifically, in order to distinguish the three states of normal, abnormality A, and abnormality B, the normal target value is set to "0", the target value of the abnormality A is set to "1", and the target value of the abnormality B is set to "-1", and learning is performed in such a way that the model outputs the target values of the input data. As a feature quantity, it is also possible to consider using the rising period T1 as a feature quantity. However, when the target pressures Pt are different and there are several patterns of ejection waveforms, even in a normal state, there are cases where the rising periods T1 deviate for each pattern. Therefore, when there are deviations in the rising period T1, it is preferably not used as a feature quantity. <Prediction stage> FIG. 9 is a block diagram showing a control unit 9 that monitors abnormalities in pressure data. FIG. 10 is a diagram showing the flow of the monitoring process of the ejection pressure executed by the control unit 9. In this monitoring process, first, the ejection pressure measurement unit 911 of the control unit 9 acquires pressure data (FIG. 10: step S21). Next, the feature quantity calculation unit 913 calculates the feature quantity of the acquired pressure data (FIG. 10: step S22). The abnormality degree calculation unit 915 inputs the feature quantity calculated by the feature quantity calculation unit 913 to the unsupervised learning model M1 to calculate the abnormality degree (FIG. 10: step S23). Also, the abnormality probability calculation unit 917 inputs the feature quantity calculated by the feature quantity calculation unit 913 to the supervised learning model M2 to calculate each abnormality probability (FIG. 10: step S24). FIG. 11 is a diagram conceptually showing the situation of calculating the abnormality degree (Mahalanobis distance). In the example shown in FIG. 11, a plurality of hollow circles (「〇」) represent the positions (distributions) of normal pressure data for machine learning in the feature quantity space. The blank triangles (「△」) represent the positions of the newly acquired pressure data in the feature quantity space. As shown in FIG. 11, the abnormality degree calculation unit 915 calculates the Mahalanobis distance between the newly acquired pressure data (△) and the k nearest neighbors (here, the nearest neighbor) of the normal pressure data (〇) as the abnormality degree. The abnormality degree is an index indicating the degree to which the newly obtained pressure data (△) deviates from the normal pressure data (〇). FIG. 12 is a diagram conceptually showing the output of the abnormality probability calculation unit 917. It is a diagram illustrating the pressure waveforms of normal, abnormality A, and abnormality B. In the example shown in FIG. 12, the state where an overshoot occurs in which the ejection pressure rises too much compared to the normal state is set as abnormality A. On the other hand, the state where the ejection pressure does not rise sufficiently compared to the normal state and the period until the ejection pressure becomes the stable pressure Pm is long is set as abnormality B. The abnormality probability calculation unit 917 is configured to output a value between -1 and +1 for the newly acquired pressure data. As described above, in machine learning, machine learning is performed such that normal is 0, the output value of abnormality A is +1, and the output of abnormality B is -1. Therefore, the closer the output value is to +1, the higher the probability of abnormality A, and the closer the output value is to -1, the higher the probability of abnormality B. Referring back to FIG. 9, the abnormality degree calculated by the abnormality degree calculation unit 915 and the abnormality probability calculated by the abnormality probability calculation unit 917 (a value between -1 and 1) are sent to the abnormality determination unit 919. The abnormality determination unit 919 determines whether the newly obtained pressure data is normal or abnormal by comparing the abnormality degree and the abnormality probability with threshold values (FIG. 10: step S25). Specifically, the abnormality determination unit 919 determines whether the abnormality degree exceeds a predetermined threshold value Th1 (refer to FIG. 11). If the abnormality degree exceeds the threshold value Th1, the abnormality determination unit 919 determines that the pressure data is abnormal. Furthermore, the abnormality determination unit 919 determines whether the abnormality probability exceeds a predetermined threshold value. For example, as shown in FIG. 12, when determining whether it corresponds to abnormality A or abnormality B, threshold values Th2 and Th3 are predetermined. The threshold value Th2 is a value greater than 0 and less than 1, and the threshold value Th3 is a value greater than -1 and less than 0. When the abnormality probability exceeds the threshold value Th2 on the positive side (that is, when the abnormality probability value is greater than the threshold value Th2), the abnormality determination unit 919 determines that the new pressure data corresponds to abnormality A. In addition, when the abnormality probability exceeds the threshold value Th3 on the negative side (that is, when the abnormality probability is lower than the threshold value Th3), the abnormality determination unit 919 determines that the new pressure data corresponds to abnormality B. For example, in the example shown in FIG. 12, since the abnormality probability of the first pressure data (data 1) does not exceed the threshold values Th2 and Th3, it is determined that the pressure data is normal. Since the abnormality probability of the second pressure data (data 2) exceeds the threshold value Th2 on the positive side, it is determined that the pressure data is abnormality A. Since the abnormality probability of the third pressure data (data 3) exceeds the threshold value Th3 on the negative side, it is determined that the pressure data is abnormality B. Furthermore, the abnormality determination unit 919 can also comprehensively use the results of the abnormality determination of the abnormality degree and the abnormality probability to perform the final abnormality determination of the pressure data. FIG. 13 is a diagram conceptually showing an example of the comprehensive abnormality determination using the abnormality degree and the abnormality probability. In FIG. 13, the horizontal axis represents the abnormality probability, and the vertical axis represents the abnormality degree. When both the abnormality degree and the abnormality probability do not exceed the threshold values (that is, both are normal), the comprehensive determination of the pressure data is "normal" (corresponding to the third quadrant in FIG. 13). Also, when both the abnormality degree and the abnormality probability exceed the threshold values (that is, both are abnormal), the comprehensive determination of the new pressure data is determined to be "abnormal". Furthermore, when the abnormality probability is high and the abnormality degree is low, the comprehensive determination of the new pressure data is determined to be "abnormality close to the normal distribution". On the contrary, when the abnormality probability is low and the abnormality degree is high, the comprehensive determination of the new pressure data is determined to be "unknown abnormality", that is, an abnormality not envisioned during learning. The abnormality determination unit 919 outputs the determination result obtained in step S25 to the outside. As an example, the abnormality determination unit 919 displays the determination result on the display 95 (step S26). The abnormality determination unit 919 can also print and output the determination result by a printer. Also, the abnormality determination unit 919 can output the determination result by, for example, lighting a lamp or outputting an alarm sound from a speaker. As described above, the control unit 9 (discharge pressure monitoring device) includes: a discharge pressure measurement unit 911 as a pressure data acquisition unit, a feature quantity calculation unit 913, an abnormality degree calculation unit 915, an abnormality probability calculation unit 917, and an abnormality determination unit 919. The discharge pressure measurement unit 911 acquires pressure data indicating the time change of the pressure in the nozzle 71 that discharges the processing liquid. The feature quantity calculation unit 913 calculates the feature quantity of the pressure data. The abnormality degree calculation unit 915 uses the unsupervised learning model M1 to calculate the abnormality degree related to the feature quantity calculated by the feature quantity calculation unit 913, where the unsupervised learning model M1 takes the feature quantity as input data and outputs the abnormality degree indicating the degree of deviation from the distribution of normal feature quantities. The abnormality probability calculation unit 917 uses the supervised learning model M2 to calculate the abnormality probability related to the feature quantity calculated by the feature quantity calculation unit 913, where the supervised learning model M2 takes the feature quantity as input data and outputs the abnormality probability indicating the probability corresponding to a given abnormality. The abnormality determination unit 919 uses the abnormality degree calculated by the abnormality degree calculation unit 915 and the abnormality probability calculated by the abnormality probability calculation unit 917 to determine the abnormality of the pressure data. Also, the control unit 9 executes a pressure monitoring method. The pressure monitoring method includes: a) a step of acquiring pressure data indicating the time change of the pressure in the nozzle 71 that discharges the processing liquid; b) a step of calculating the feature quantity of the pressure data; c) a step of using the unsupervised learning model M1 to calculate the abnormality degree related to the feature quantity calculated in step b), where the unsupervised learning model M1 takes the feature quantity as input data and outputs the abnormality degree indicating the degree of deviation from the distribution of normal feature quantities; d) a step of using the supervised learning model M2 to calculate the abnormality probability related to the feature quantity calculated in step b), where the supervised learning model M2 takes the feature quantity as input data and outputs the abnormality probability indicating the probability corresponding to a specific abnormality; and e) a step of using the abnormality degree calculated in the previous step c) and the abnormality probability calculated in the previous step d) to determine the abnormality of the pressure data. According to this configuration, when determining a pre-conceived abnormality, it can discriminate the abnormality by calculating the abnormality probability. Also, even when an unknown abnormality that has not been pre-conceived occurs, it can appropriately detect the unknown abnormality by calculating the abnormality degree. Further, the unsupervised learning model M1 is a model obtained by unsupervised learning using only the feature amounts of a plurality of normal pressure data as input data. According to this configuration, since only the distribution of normal feature amounts can be learned, it can appropriately perform anomaly detection. The unsupervised learning model M1 outputs the Mahalanobis distance as the degree of anomaly. According to this configuration, since the degree of anomaly is calculated in consideration of the correlation between the feature amounts representing the pressure data in each dimension, it can more accurately grasp the characteristic pattern or abnormal behavior of the pressure data. The supervised learning model M2 outputs probabilities corresponding to a plurality of specific anomalies (for example, anomalies A and B). According to this configuration, since probabilities of more than one kind can be calculated, it can be determined whether the pressure data corresponds to one of the plurality of anomalies. <Other Embodiments> In the above embodiment, the anomaly degree calculation unit 915 calculates the anomaly degree using the unsupervised learning model M1. However, it does not necessarily need to calculate the anomaly degree using the unsupervised learning model M1. For example, the anomaly degree calculation unit 915 may also use a predetermined formula to calculate the difference between the feature amount of the normal feature amount and the feature amount of the newly obtained pressure data. This difference corresponds to the anomaly degree indicating the degree to which the feature amount of the newly obtained pressure data deviates from the normal feature amount. Further, the normal feature amount to be a comparison target may be defined by the feature amounts of a plurality of normal pressure data, or may also be the feature amount of a single normal pressure data. The anomaly determination unit 919 compares the anomaly degree (difference value) calculated by the anomaly degree calculation unit 915 with a preset threshold value. Then, when the anomaly degree exceeds the threshold value, the anomaly determination unit 919 determines that the newly acquired pressure data is abnormal. For example, in Japanese Patent Application Laid-Open No. 2012-098133, when detecting an anomaly of an inverter device based on the pulse component of a current waveform, it calculates the similarity between the feature amount of the pulse component in the normal state and the feature amount of the pulse component of the determination target. Also, in Japanese Patent Application Laid-Open No. 2006-026584, when detecting an ink ejection anomaly based on the voltage applied to an actuator that ejects ink, it calculates the difference between the voltage in the normal state and the voltage of the determination target. These similarities and difference values can also be used as the above-mentioned anomaly degree. The present invention has been described in detail above, but the above description is illustrative in all cases, and the present invention is not limited thereto. It should be understood that countless variations that are not illustrated can be conceived without departing from the scope of the present invention. Each configuration described in the above embodiments and each variation can be appropriately combined or omitted as long as they do not contradict each other. 1: Coating device 2: Input transfer unit 3: Floating table unit 4: Output transfer unit 5: Substrate transfer unit 7: Coating mechanism 8: Coating liquid supply mechanism 9: Control unit (discharge pressure monitoring device) 21, 41, 101, 111: Roller conveyor 22: Rotary lifting drive mechanism 31: Inlet floating table 32: Coating table 33: Outlet floating table 34: Lift pin drive mechanism 35: Floating control mechanism 36: Lifting drive mechanism 42: Rotary lifting drive mechanism 51: Chuck mechanism (substrate holding part) 52: Adsorption transfer control mechanism 61, 62: Sensor 71: Nozzle 72: Nozzle cleaning standby unit 81: Pump 82, 84: Pipe 83: Coating liquid replenishment unit 85, 833: On-off valve 86: Pressure sensor 87: Drive part 91: Calculation part 93: Storage part 95: Display 97: Input device 100: Input conveyor 102, 112: Rotary drive mechanism 110: Output conveyor 721: Roller 722: Cleaning part 723: Roller groove 811: Flexible tube 812: Bellows 813: Small bellows 814: Large bellows 815: Pump chamber 816: Actuator disk part 831: Storage tank 910: Discharge control part 911: Discharge pressure measurement part (pressure data acquisition part) 913: Feature quantity calculation part 915: Abnormality degree calculation part 917: Abnormality probability calculation part 919: Abnormality determination part 931: Computer program A, B: Abnormality D1, D2, D3: Pressure data Dif2: Second-order differential F1, F2: Feature quantity M: Recording medium M1: Unsupervised learning model M2: Supervised learning model Pg: Target pressure Pmax: Maximum pressure Pm: Stable pressure Pmin: Minimum value of discharge pressure Pi: Initial pressure Pt: Target pressure S: Substrate Sb: Lower surface Sf: Upper surface T1: Rising period T2: Transition period T3: Stable period T4: Falling period T2_s: Initial vibration period t11, t12, ta, tb, tc, td, te: Time FIG. 1 is a diagram schematically showing the overall configuration of the coating apparatus according to the embodiment. FIG. 2 is a diagram showing the configuration of the coating liquid supply mechanism. FIG. 3 is a diagram showing the pressure waveform of the ejection pressure. FIG. 4 is a block diagram showing a configuration example of the control unit. FIG. 5 is a diagram for explaining an example of a feature quantity. FIG. 6 is a diagram for explaining another example of the feature quantity. FIG. 7 is a block diagram showing the control unit performing machine learning. FIG. 8 is a diagram showing the flow of the machine learning process executed by the control unit. FIG. 9 is a block diagram showing the control unit monitoring the abnormality of the pressure data. FIG. 10 is a diagram showing the flow of the monitoring process of the ejection pressure executed by the control unit. FIG. 11 is a diagram conceptually showing the situation where the abnormality degree (Mahalanobis distance) is calculated. FIG. 12 is a diagram conceptually showing the output of the abnormality probability calculation unit. FIG. 13 is a diagram conceptually showing an example of the comprehensive determination of the abnormality using the abnormality degree and the abnormality probability. 86: Pressure sensor 95: Display 911: Ejection pressure measurement unit (pressure data acquisition unit) 913: Feature quantity calculation unit 915: Abnormality degree calculation unit 917: Abnormality probability calculation unit 919: Abnormality determination unit M1: Unsupervised learning model M2: Supervised learning model

Claims

1. A coating apparatus comprising: a substrate holding section for holding a substrate; a nozzle for dispensing a processing liquid toward the substrate held in the substrate holding section; a pressure sensor for measuring pressure within the nozzle; and a dispensing pressure monitoring device; the dispensing pressure monitoring device comprising: a pressure data acquisition section for acquiring pressure data representing the time-varying pressure within the nozzle by means of the pressure sensor; a feature quantity calculation section for calculating a feature quantity of the pressure data; and an anomaly calculation section for calculating an anomaly degree related to the feature quantity calculated by the feature quantity calculation section using an unsupervised learning model, wherein the unsupervised learning model takes the feature quantity as input data and outputs the anomaly degree representing the degree of deviation from the normal distribution of the feature quantity; The anomaly probability calculation unit uses a supervised learning model to calculate the anomaly probability related to the aforementioned feature quantity calculated by the aforementioned feature quantity calculation unit. The aforementioned supervised learning model takes the aforementioned feature quantity as input data and outputs the aforementioned anomaly probability, which represents the probability corresponding to a specific anomaly. The anomaly determination unit obtains the aforementioned anomaly degree calculated by the aforementioned anomaly degree calculation unit and the aforementioned anomaly probability calculated by the aforementioned anomaly probability calculation unit. Based on the comparison between the aforementioned anomaly degree and the aforementioned anomaly probability and the threshold value, it determines whether the aforementioned pressure data is normal, an anomaly close to normal distribution, an unknown anomaly, or any of the following states: the aforementioned feature quantity is the degree of overshoot during the period from the moment the treated liquid is discharged from the aforementioned nozzle until the moment the discharge pressure reaches the target pressure.

2. The coating apparatus as claimed in claim 1, wherein, The aforementioned unsupervised learning model is obtained by using only the features of a normal plurality of the aforementioned stress data as input data for unsupervised learning.

3. The coating apparatus as described in claim 1 or 2, wherein, The aforementioned unsupervised learning model outputs the Mahalanobis distance as the aforementioned anomaly score.

4. The coating apparatus as described in claim 1 or 2, wherein, The aforementioned supervised learning model outputs probabilities corresponding to a complex number of specific anomalies.

5. A coating apparatus comprising: a substrate holding section for holding a substrate; a nozzle for dispensing a processing liquid toward the substrate held in the substrate holding section; a pressure sensor for measuring the pressure inside the nozzle; and a dispensing pressure monitoring device; the dispensing pressure monitoring device comprising: a pressure data acquisition section for acquiring pressure data representing the time change of the pressure inside the nozzle by means of the pressure sensor; a feature quantity calculation section for calculating a feature quantity of the pressure data; and an anomaly calculation section for calculating an anomaly degree representing the degree of deviation from the normal feature quantity based on the feature quantity calculated by the feature quantity calculation section. The anomaly probability calculation unit uses a supervised learning model to calculate the anomaly probability related to the aforementioned feature quantity calculated by the aforementioned feature quantity calculation unit. The aforementioned supervised learning model takes the aforementioned feature quantity as input data and outputs the aforementioned anomaly probability, which represents the probability corresponding to a specific anomaly. The anomaly determination unit obtains the aforementioned anomaly degree calculated by the aforementioned anomaly degree calculation unit and the aforementioned anomaly probability calculated by the aforementioned anomaly probability calculation unit. Based on the comparison between the aforementioned anomaly degree and the aforementioned anomaly probability and the threshold value, it determines whether the aforementioned pressure data is normal, an anomaly close to normal distribution, an unknown anomaly, or any of the following states: an anomaly. The aforementioned feature quantity refers to the degree of overshoot during the period from the moment the treated liquid is discharged from the aforementioned nozzle until the discharge pressure reaches the target pressure.

Images