JP7731940B2 - Discharge pressure monitoring device, discharge pressure monitoring method, computer program, and coating device - Google Patents

Description

The subject matter disclosed in this specification relates to a discharge pressure monitoring device, a discharge pressure monitoring method, a computer program, and an application device.

In the manufacturing process of flat panel displays, a device called a coater is used. A coater uses a pump to eject processing liquid from a slit nozzle, coating the entire surface of the substrate as it is being transported. With the recent trend toward higher product quality, coaters are required to apply processing liquid so that the film thickness of the processing liquid is uniform across the entire substrate. To maintain consistent film thickness uniformity, it is important to monitor the processing liquid's ejection pressure (ejection flow rate) during production. For example, Patent Document 1 attempts to optimize parameters related to the ejection pressure.

However, it is difficult to detect minute changes in the discharge pressure obtained from a pressure sensor using a simple threshold. To address this issue, a number of common techniques and equipment systems have been proposed that use machine learning to detect anomalies.

For example, Patent Document 2 proposes building a system that can determine the type (mode) of anomaly from given sensor information. Building such a system not only makes it possible to plan specific countermeasures such as part replacement, but also functions as a failure prediction system that can detect minor changes (signs) that could potentially develop into a breakdown.

Japanese Patent Application Laid-Open No. 2020-040046 Japanese Patent Application Laid-Open No. 2022-125288

Patent Document 2 proposes a method using unsupervised learning and a method using supervised learning as systems for predicting failures. However, unsupervised learning is a method that learns data from normal states and evaluates the degree of deviation from the normal state of newly measured data. Therefore, while it is possible to detect abnormalities, it is not possible to determine the type of abnormality. Furthermore, supervised learning is able to determine the type of abnormality by learning from training data that has been pre-labeled with the type of abnormality, but because any abnormality is classified into a type of abnormality that has been pre-assumed during learning, it is not possible to respond to unknown abnormalities.

The object of the present invention is to provide technology that can appropriately detect unknown abnormalities in discharge pressure that have not been anticipated in advance.

In order to solve the above problem, a first aspect is a coating apparatus comprising: a substrate holding unit that holds a substrate; a nozzle that discharges a processing liquid toward the substrate held by the substrate holding unit; a pressure sensor that measures a pressure inside the nozzle; and a discharge pressure monitoring device, wherein the discharge pressure monitoring device includes a pressure data acquisition unit that acquires pressure data indicating a time change in the pressure inside the nozzle using the pressure sensor; a feature calculation unit that calculates a feature of the pressure data; an anomaly degree calculation unit that receives the feature as an input and calculates the anomaly degree for the feature calculated by the feature calculation unit using an unsupervised learning model that outputs an anomaly degree that indicates a degree of deviation from a normal distribution of the feature; and an anomaly degree calculation unit that receives the feature as an input and calculates the anomaly degree for the feature calculated by the feature calculation unit using an unsupervised learning model that outputs an anomaly degree that indicates a degree of deviation from a normal distribution of the feature. The apparatus includes an abnormality probability calculation unit that calculates the abnormality probability for the feature calculated by the feature calculation unit using a supervised learning model that always outputs an abnormality probability that indicates a corresponding probability, and an abnormality determination unit that acquires the degree of abnormality calculated by the abnormality degree calculation unit and the abnormality probability calculated by the abnormality probability calculation unit, and determines whether the pressure data is in a normal state, an abnormality close to a normal distribution, an unknown abnormality, or an abnormal state based on a comparison of the degree of abnormality and the abnormality probability with a threshold value, wherein the feature is the degree of overshoot that occurs during a rise period of the ejection pressure, which is the period from the time when ejection of the treatment liquid from the nozzle starts to the time when the ejection pressure reaches a target pressure .

A second aspect is the coating apparatus of the first aspect, wherein the unsupervised learning model is a model obtained by unsupervised learning using, as input data, feature quantities of only the plurality of normal pressure data.

A third aspect is the coating apparatus of the first or second aspect, wherein the unsupervised learning model outputs a Mahalanobis distance as the degree of abnormality.

A fourth aspect is the coating apparatus of the first or second aspect, wherein the supervised learning model outputs probabilities of occurrence of a plurality of types of specific abnormalities.

A fifth aspect is a coating apparatus comprising: a substrate holding unit that holds a substrate; a nozzle that discharges a processing liquid toward the substrate held by the substrate holding unit; a pressure sensor that measures a pressure inside the nozzle; and a discharge pressure monitoring device, wherein the discharge pressure monitoring device comprises a pressure data acquisition unit that acquires pressure data indicating a time change in the pressure inside the nozzle using the pressure sensor ; a feature amount calculation unit that calculates a feature amount of the pressure data; an abnormality degree calculation unit that calculates an abnormality degree indicating a degree of deviation from a normal feature amount for the feature amount calculated by the feature amount calculation unit; and a supervised learning algorithm that receives the feature amount as an input and outputs an abnormality probability indicating a probability that the feature amount corresponds to a specific abnormality. an abnormality probability calculation unit that calculates the abnormality probability for the feature calculated by the feature calculation unit using a learning model; and an abnormality determination unit that acquires the degree of abnormality calculated by the abnormality degree calculation unit and the abnormality probability calculated by the abnormality probability calculation unit, and determines whether the pressure data is in a normal state, an abnormality close to a normal distribution, an unknown abnormality, or an abnormal state based on a comparison of the degree of abnormality and the abnormality probability with a threshold value, wherein the feature is the degree of overshoot that occurs during a rise period of the discharge pressure, which is the period from the time when the discharge of the processing liquid from the nozzle starts to the time when the discharge pressure reaches a target pressure .

According to the first to fifth aspects, when a previously anticipated anomaly is determined, the anomaly can be determined by calculating the anomaly probability. Furthermore, even when an unknown anomaly that has not been previously anticipated occurs, the unknown anomaly can be appropriately detected by calculating the anomaly degree.

According to the coating device of the second aspect, only the distribution of normal feature amounts can be learned, and therefore, abnormality detection can be performed appropriately.

According to the coating device of the third aspect, the degree of abnormality can be calculated by taking into account the correlation between each dimension of the feature values indicated by the pressure data, thereby making it possible to more accurately capture characteristic patterns and abnormal behavior in the pressure data.

According to the coating apparatus of the fourth aspect, the probability of a plurality of types of specific abnormalities can be calculated, and therefore it is possible to determine whether the pressure data corresponds to any of the plurality of types of abnormalities.

1 is a diagram schematically illustrating an overall configuration of a coating apparatus according to an embodiment. FIG. 2 is a diagram showing the configuration of a coating liquid supply mechanism. FIG. 4 is a diagram showing a pressure waveform of a discharge pressure. FIG. 2 is a block diagram showing an example of the configuration of a control unit. FIG. 10 is a diagram illustrating an example of a feature amount. FIG. 10 is a diagram for explaining another example of feature amounts. FIG. 1 is a block diagram showing a control unit that performs machine learning. FIG. 10 is a diagram showing the flow of machine learning processing executed by the control unit. FIG. 10 is a block diagram showing a control unit that monitors abnormalities in pressure data. FIG. 10 is a diagram showing a flow of a monitoring process for monitoring a discharge pressure executed by a control unit. FIG. 10 is a diagram conceptually illustrating how the degree of abnormality (Mahalanobis distance) is calculated. FIG. 10 is a diagram conceptually illustrating the output of an abnormality probability calculation unit. FIG. 10 is a diagram conceptually illustrating an example of comprehensive determination of an abnormality using an abnormality degree and an abnormality probability.

Embodiments of the present invention will now be described with reference to the accompanying drawings. Note that the components described in these embodiments are merely examples and are not intended to limit the scope of the present invention. For ease of understanding, the dimensions and number of parts may be exaggerated or simplified as necessary in the drawings.

Figure 1 is a diagram showing a schematic diagram of the overall configuration of a coating apparatus 1 according to an embodiment. The coating apparatus 1 is a substrate processing apparatus that applies a coating liquid to the upper surface Sf of a substrate S. The substrate S is, for example, a glass substrate for a liquid crystal display device. The substrate S may also be a semiconductor wafer, a glass substrate for a photomask, a glass substrate for a plasma display, a glass or ceramic substrate for a magnetic or optical disk, a glass substrate for an organic electroluminescence (EL) display, a glass or silicon substrate for a solar cell, or various other substrates for electronic devices such as flexible substrates and printed circuit boards. The coating apparatus 1 is, for example, a slit coater.

In Figure 1, an XYZ coordinate system is defined to explain the relative positions of the elements of the coating device 1. The transport direction of the substrate S is the "X direction." The direction in which the substrate S advances in the X direction (toward downstream in the transport direction) is the +X direction, and the opposite direction (toward upstream in the transport direction) is the -X direction. The direction perpendicular to the X direction is the Y direction, and the direction perpendicular to both the X and Y directions is the Z direction. In the following explanation, the Z direction is the vertical direction, and the X and Y directions are the horizontal directions. In the Z direction, the +Z direction is the upward direction, and the -Z direction is the downward direction.

The coating device 1 comprises, in order in the +X direction, an input conveyor 100, an input transfer section 2, a floating stage section 3, an output transfer section 4, and an output conveyor 110. The input conveyor 100, the input transfer section 2, the floating stage section 3, the output transfer section 4, and the output conveyor 110 form a transport path along which the substrate S passes. The coating device 1 also comprises a substrate transport section 5, a coating mechanism 7, a coating liquid supply mechanism 8, and a control unit 9.

The substrate S is transported to the input conveyor 100 from a device upstream of the coating device 1. The input conveyor 100 comprises a roller conveyor 101 and a rotation drive mechanism 102. The rotation drive mechanism 102 rotates each roller of the roller conveyor 101. As each roller of the roller conveyor 101 rotates, the substrate S is transported downstream (+X direction) in a horizontal position. A "horizontal position" refers to 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 section 2 is equipped with a roller conveyor 21 and a rotation/lift drive mechanism 22. The rotation/lift drive mechanism 22 rotates each roller of the roller conveyor 21 and raises and lowers the roller conveyor 21. As the roller conveyor 21 rotates, the substrate S is transported downstream (+X direction) in a horizontal position. Additionally, as the roller conveyor 21 is raised and lowered, 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 stage section 3 via the input transfer section 2.

As shown in FIG. 1, the levitation stage unit 3 is generally flat. The levitation stage unit 3 is divided into three sections along the X direction. The levitation stage unit 3 comprises, in order along the +X direction, an entrance levitation stage 31, a coating stage 32, and an exit levitation stage 33. The top surfaces of the entrance levitation stage 31, the coating stage 32, and the exit levitation stage 33 are all flush with each other. The levitation stage unit 3 further comprises a lift pin drive mechanism 34, a levitation control mechanism 35, and an elevation drive mechanism 36. The lift pin drive mechanism 34 raises and lowers the multiple lift pins arranged on the entrance levitation stage 31. The levitation control mechanism 35 supplies compressed air to the entrance levitation stage 31, the coating stage 32, and the exit levitation stage 33 to lift the substrate S. The elevation drive mechanism 36 raises and lowers the exit levitation stage 33.

A large number of nozzle holes are arranged in a matrix on the upper surface of the entrance levitation stage 31 and the upper surface of the exit levitation stage 33, which spray compressed air supplied from the levitation control mechanism 35. When compressed air is sprayed from each nozzle hole, the substrate S levitates upward relative to the levitation stage unit 3. The lower surface Sb of the substrate S is then separated from the upper surface of the levitation stage unit 3 and supported in a horizontal position. When the substrate S is levitated, the distance (levitation amount) between the lower surface Sb of the substrate S and the upper surface of the levitation stage unit 3 is, for example, 10 μm or more and 500 μm or less.

On the upper surface of the coating stage 32, jet holes that eject compressed air supplied from the levitation control mechanism 35 and suction holes that suck in gas are arranged alternately in the X and Y directions. The levitation control mechanism 35 controls the amount of compressed air ejected from the jet holes and the amount of air sucked in from the suction holes. This precisely controls the amount of levitation of the substrate S relative to the coating stage 32 so that the position in the Z direction of the upper surface Sf of the substrate S passing above the coating stage 32 becomes a specified value. The amount of levitation of the substrate S relative to the coating stage 32 is calculated by the control unit 9 based on the detection results of sensor 61 or sensor 62, described below. The amount of levitation of the substrate S relative to the coating stage 32 is preferably adjustable with high precision by airflow control.

The substrate S carried into the floating stage section 3 is imparted with a driving force in the +X direction by the roller conveyor 21 and transported onto the entrance floating stage 31. The entrance floating stage 31, coating stage 32, and exit floating stage 33 support the substrate S in a floating state. For example, the configuration described in Japanese Patent No. 5346643 may be used as the floating stage section 3.

The substrate transport unit 5 is disposed below the floating stage unit 3. The substrate transport unit 5 comprises a chuck mechanism 51 and a suction/travel control mechanism 52. The chuck mechanism 51 comprises a suction pad (not shown) attached to a suction member. The chuck mechanism 51 supports the substrate S from below by bringing the suction pad into contact with the peripheral edge of the lower surface Sb of the substrate S. The suction/travel control mechanism 52 applies negative pressure to the suction pad, thereby suctioning the substrate S to the suction pad. The suction/travel control mechanism 52 also causes the substrate transport unit 5 to travel back and forth in the X direction.

The chuck mechanism 51 holds the substrate S with the lower surface Sb of the substrate S positioned higher than the upper surface of the floating stage unit 3. With the peripheral edge of the substrate S held by the chuck mechanism 51, the buoyancy applied by the floating stage unit 3 maintains the substrate S in a horizontal position.

As shown in FIG. 1, the coating device 1 is equipped with a sensor 61 for measuring plate thickness. The sensor 61 is located near the roller conveyor 21. The sensor 61 detects the position in the Z direction of the upper surface Sf of the substrate S held by the chuck mechanism 51. Furthermore, by positioning a chuck (not shown) that is not holding the substrate S directly below the sensor 61, the sensor 61 can detect the position in the vertical direction Z of the suction surface, which is the upper surface of the suction member.

The chuck mechanism 51 moves in the +X direction while holding the substrate S that has been loaded into the floating stage section 3. This transports the substrate S from above the entrance floating stage 31, past above the coating stage 32, and to above the exit floating stage 33. The substrate S is then moved from the exit floating stage 33 to the output transfer section 4.

The output transfer unit 4 moves the substrate S from a position above the exit floating stage 33 to the output conveyor 110. The output transfer unit 4 includes a roller conveyor 41 and a rotation/lifting drive mechanism 42. The rotation/lifting drive mechanism 42 drives the roller conveyor 41 to rotate and raises and lowers the roller conveyor 41 in the Z direction. As each roller of the roller conveyor 41 rotates, the substrate S moves in the +X direction. Additionally, as the roller conveyor 41 rises and falls, the substrate S is displaced in the Z direction.

The output conveyor 110 includes a roller conveyor 111 and a rotation drive mechanism 112. The output conveyor 110 transports the substrate S in the +X direction by rotating each roller of the roller conveyor 111, and discharges the substrate S outside the coating device 1. The input conveyor 100 and output conveyor 110 are part of the coating device 1. However, the input conveyor 100 and output conveyor 110 may be incorporated into a device separate from the coating device 1.

The coating mechanism 7 applies a coating liquid to the upper surface Sf of the substrate S. The coating mechanism 7 is disposed above the transport path of the substrate S. The coating mechanism 7 has a nozzle 71. The nozzle 71 is a slit nozzle with a slit-shaped outlet on its underside. The nozzle 71 is connected to a positioning mechanism (not shown). The positioning mechanism moves the nozzle 71 between a coating position above the coating stage 32 (the position indicated by the solid line in Figure 1) and a maintenance position, which will be 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, causing the coating liquid to be ejected from an outlet disposed on the underside of the nozzle 71.

Figure 2 shows the configuration of the coating liquid supply mechanism 8. The coating liquid supply mechanism 8 includes a pump 81, piping 82, a coating liquid replenishment unit 83, piping 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 supplies the coating liquid by changing its volume. The pump 81 may be, for example, a bellows-type pump as described in Japanese Patent Application Laid-Open No. 10-61558. As shown in Figure 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 replenishment unit 83 via piping 82. The other end of the flexible tube 811 is connected to the nozzle 71 via piping 84.

The pump 81 has a bellows 812 that is elastically deformable in the axial direction. The bellows 812 has a small bellows section 813, a large bellows section 814, a pump chamber 815, and an actuation disk section 816. The pump chamber 815 is disposed between the flexible tube 811 and the bellows 812. An incompressible medium is sealed in the pump chamber 815. The actuation disk section 816 is connected to the drive section 87.

The coating liquid replenishment unit 83 has a storage tank 831 that stores the coating liquid. The storage tank 831 is connected to the pump 81 via piping 82. An on-off valve 833 is inserted in the piping 82. The on-off valve 833 opens and closes in response to commands from the control unit 9. When the on-off valve 833 is opened, the coating liquid can be replenished from the storage tank 831 to the flexible tube 811 of the pump 81. When the on-off valve 833 is closed, the replenishment of the coating liquid from the storage tank 831 to the flexible tube 811 of the pump 81 is restricted.

The pipe 84 is connected to the output side of the pump 81. The on-off valve 85 is inserted in the pipe 84. The on-off valve 85 opens and closes in response to commands from the control unit 9. The opening and closing of the on-off valve 85 switches between sending and stopping the application liquid to the nozzle 71. The pressure sensor 86 is disposed in the pipe 84. The pressure sensor 86 detects the pressure (discharge pressure) of the application liquid sent to the nozzle 71, and outputs a signal indicating the detected pressure value to the control unit 9.

Figure 3 shows the pressure waveform of the discharge pressure. In Figure 3, the horizontal axis represents time, and the vertical axis represents pressure value. In the coating device 1, an optimization process is performed as appropriate to approximate the pressure waveform of the coating liquid discharged from the nozzle 71 to an ideal pressure waveform by adjusting various parameters (acceleration time, steady-state speed, steady-state speed time, deceleration time, etc.) that define the movement of the operating disk portion 816.

As shown in FIG. 3, the discharge pressure is measured from before the nozzle 71 starts discharging the coating liquid until the nozzle 71 stops discharging the coating liquid. In the example shown in FIG. 3, the discharge pressure at time ta when the nozzle 71 starts discharging the coating liquid and the discharge pressure at time te when the nozzle 71 stops discharging the coating liquid are the initial pressure Pi. However, the pressures at the start and end of discharge do not always match the initial pressure Pi. As shown in FIG. 3, the discharge period is divided into a rise period T1, a transition period T2, a steady period T3, and a fall period T4.

The rise period T1 is the period from time ta when the coating liquid supply mechanism 8 starts discharging the coating liquid from the nozzle 71 (i.e., the time when the coating liquid supply mechanism 8 starts moving the operating disk portion 816) to time tb when the discharge pressure reaches the target pressure Pt. In other words, when the nozzle 71 starts discharging the coating liquid at time ta, the discharge pressure increases from the initial pressure Pi to the target pressure Pt between time ta and time tb.

The transition period T2 is the period from time tb to time tc, when a predetermined vibration damping period has elapsed. This vibration damping period is the period required for the time change in the discharge pressure to stabilize, and is set, for example, by the user through an input operation on the input device 97 and stored in the memory unit 93.

The steady-state period T3 is the period from time tc to time td when the coating liquid supply mechanism 8 begins to reduce the discharge pressure (i.e., the time when the coating liquid supply mechanism 8 begins to decelerate the operating disc portion 816 from the target speed). In other words, the coating liquid supply mechanism 8 moves the operating disc portion 816 at a constant speed from time tc to time td, and begins to decelerate the operating disc portion 816 at time td. Note that during the steady-state period T3, the discharge pressure basically stabilizes at the target pressure Pt. However, even during the steady-state period T3, the change in the discharge pressure over time includes slight vibrations, and the discharge pressure may be greater or less than the target pressure Pt.

The fall period T4 is the period from time td to time te when the coating liquid supply mechanism 8 stops discharging the coating liquid from the nozzle 71 (i.e., time te when the coating liquid supply mechanism 8 stops the operating disk portion 816). In other words, the discharge pressure decreases to the initial pressure Pi between time td and time te, and at time te, discharging of the coating liquid from the nozzle 71 stops.

As shown in Figures 1 and 2, a sensor 62 is disposed on the nozzle 71 to which the coating liquid is supplied from the coating liquid supply mechanism 8. The sensor 62 detects the height of the substrate S in the Z direction in a non-contact manner. The sensor 62 is electrically connected to the control unit 9. Based on the detection results of the sensor 62, the control unit 9 measures the distance (separation distance) between the floating substrate S and the upper surface of the coating stage 32. The control unit 9 then adjusts the coating position of the nozzle 71 using the positioning mechanism based on the measured separation distance. Note that an optical sensor or an ultrasonic sensor can be used as the sensor 62.

The coating mechanism 7 is equipped with a nozzle cleaning standby unit 72. The nozzle cleaning standby unit 72 performs predetermined maintenance on the nozzle 71 positioned at the maintenance position. The nozzle cleaning standby unit 72 has a roller 721, a cleaning section 722, and a roller butt 723. The nozzle cleaning standby unit 72 cleans the nozzle 71 and forms a liquid pool, thereby preparing the nozzle 71's discharge port for coating processing. Furthermore, in the coating device 1, in order to evaluate the discharge pressure applied to the coating liquid, a simulated discharge is performed in which the nozzle 71 is discharged from the nozzle 71 while the nozzle 71 is positioned at the maintenance position.

Figure 4 is a block diagram showing an example configuration of the control unit 9. The control unit 9 controls the operation of each element of the coating device 1. The control unit 9 includes a calculation unit 91, a memory unit 93, a display 95, and an input device 97. The control unit 9 can be, for example, a desktop, laptop, or tablet computer. As described below, the control unit 9 functions as a discharge pressure monitoring device that monitors for abnormalities in the discharge pressure.

The calculation unit 91 is a processor composed of a CPU (Central Processing Unit) and other components. The storage unit 93 is composed of a transient storage device such as RAM (Random Access Memory) and a non-transient auxiliary storage device such as an HDD (Hard Disk Drive) or SSD (Solid State Drive).

The display 95 is a device that displays information to the user, and is specifically an LCD display. The input device 97 is a device that accepts input operations by the user, and is a mouse, keyboard, etc.

The storage unit 93 stores a computer program 931. The computer program 931 is provided by a recording medium M. That is, the recording medium M records the computer program 931 so that it can be read by the control unit 9, which is a computer. The recording medium M may be, for example, a USB (Universal Serial Bus) memory, an optical disk such as a DVD (Digital Versatile Disc), or a magnetic disk.

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 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 (feeding operation) of the pump 81 that supplies the coating liquid to the nozzle 71 based on preset parameters.

The discharge pressure measurement unit 911 measures the discharge pressure. Specifically, the discharge pressure measurement unit 911 periodically acquires the discharge pressure measured by the pressure sensor 86 at a predetermined sampling period. The discharge pressure measurement unit 911 acquires the time change (time series data) of the discharge pressure applied to the coating liquid during the period in which the coating liquid is discharged from the nozzle 71, and stores the acquired data as pressure data in the memory unit 93. The pressure data is data indicating each time and the pressure measured at each time. The discharge pressure measurement unit 911 is an example of a pressure data acquisition unit.

The feature calculation unit 913 derives a feature from the discharge pressure measured by the discharge pressure measurement unit 911. The feature may be, for example, one of several feature values described in Japanese Patent Application Laid-Open No. 2022-138109.

Figure 5 is a diagram illustrating an example of a feature quantity. In this example, the degree of overshoot that occurs when the discharge pressure rises is calculated as the feature quantity F1. Specifically, the feature quantity calculation unit 913 calculates the sign (positive or negative) of the second derivative Dif2 of the discharge pressure at time t11, when the discharge pressure reaches its maximum value Pmax. The feature quantity calculation unit 913 then calculates time t12, when the sign of the second derivative switches twice from the sign at time t11. The feature quantity calculation unit 913 then calculates the feature quantity F1 of the time change in the discharge pressure during the initial vibration period T2_s, from time t11 to t12.

For example, the feature calculation unit 913 selects the smaller of the minimum value Pmin of the discharge pressure during the initial vibration period T2_s and the steady-state pressure Pm (the average value of the discharge pressure during the steady-state period T3) as the target pressure Pg. The feature calculation unit 913 may then calculate the difference between the maximum pressure Pmax and the target pressure Pg (= Pmax - Pg) as the feature F1.

This characteristic quantity F1 becomes relatively larger as the overshoot of the discharge pressure over time increases due to the momentum of the discharge pressure rise.

Figure 6 is a diagram illustrating another example of a feature quantity. In this example, the stability of the time change in 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 steady pressure Pm, which is the average value of the discharge pressure during the steady period T3, as the feature quantity F2. The feature quantity F2 becomes relatively larger as the ringing (waveform oscillation) of the time change in the discharge pressure during the transition period T2 increases.

The anomaly calculation unit 915 inputs the features calculated by the feature calculation unit 913 into the unsupervised learning model M1 to calculate the anomaly level of the pressure data.

The abnormality probability calculation unit 917 inputs the features calculated by the feature calculation unit 913 into the supervised learning model M2 to calculate the probability of an abnormality. In more detail, pressure data can be classified into a normal state and one or more abnormal states assumed in advance. The feature calculation unit 913 calculates the probability that the pressure data corresponds to each of the one or more abnormal states assumed in advance (hereinafter also referred to as the "abnormality probability"). For example, if two types of abnormalities ("abnormality A" and "abnormality B") are assumed in advance during learning of the supervised learning model M2, the pressure data is classified into a normal state, an abnormality A state, and an abnormality B state. The abnormality probability calculation unit 917 then calculates the abnormality probability indicating the probability that the pressure data corresponds to abnormality A or B.

The abnormality determination unit 919 determines whether the pressure data is abnormal using the degree of abnormality calculated by the abnormality degree calculation unit 915 and the abnormality probability calculated by the abnormality probability calculation unit 917.

<Learning stage>
Fig. 7 is a block diagram showing the control unit 9 that performs machine learning. Fig. 8 is a diagram showing the flow of the machine learning process executed by the control unit 9. Note that the machine learning shown in Figs. 8 and 9 may be performed by a computer device separate from the control unit 9.

First, as shown in Figures 7 and 8, in order to perform machine learning, the discharge pressure measurement unit 911 acquires in advance a considerable number of normal pressure data D1, a considerable number of pressure data D2 corresponding to abnormality A, and a considerable number of pressure data D3 corresponding to abnormality B (Figure 8: step S11). The acquired pressure data D1 to D3 are stored in the memory unit 93. The feature calculation unit 913 then acquires the feature values of each of the pressure data D1 to D3 (Figure 8: step S12).

The anomaly degree calculation unit 915 performs machine learning (unsupervised learning) to obtain the unsupervised learning model M1 (Figure 8: step S13). The anomaly degree calculation unit 915 stores the unsupervised learning model M1 (more specifically, the learned parameters) obtained by machine learning in the memory unit 93 (step S14).

Specifically, in unsupervised learning, a set of features from multiple normal pressure data D1 is prepared as learning data. Then, machine learning based on the k-nearest neighbor method is performed using this learning data. The unsupervised learning model M1 takes the features as input and outputs an anomaly level indicating the degree of deviation from the distribution of the features of normal pressure data. The anomaly level is preferably the Mahalanobis distance from the k-nearest neighbors, but may also be the Euclidean distance or Manhattan distance.

The anomaly probability calculation unit 917 performs machine learning (supervised learning) to obtain the supervised learning model M2 (Figure 8: step S15). The anomaly probability calculation unit 917 also stores the supervised learning model M2 (more specifically, the learned parameters) obtained by machine learning in the memory unit 93 (Figure 8: step S16).

In supervised learning, first, multiple feature values of normal pressure data, multiple feature values of abnormal pressure data A, and multiple feature values of abnormal pressure data B are prepared as input data. Furthermore, the basis of supervised learning model M2 is, for example, a linear regression model. In supervised learning, optimal parameters for predicting target values (labels indicating "normal," "abnormal A," and "abnormal B") from the input data are found. Specifically, to distinguish between the three states of normal, abnormal A, and abnormal B, the target value for normal is set to "0," the target value for abnormal A is set to "1," and the target value for abnormal B is set to "-1," and learning proceeds so that the model outputs the target values of the input data.

It is also possible to consider using the rising period T1 as a feature quantity. However, if there are several ejection waveform patterns, such as when the target pressure Pt is different, the rising period T1 may vary from pattern to pattern even under normal conditions. For this reason, if there is variation in the rising period T1, it is preferable not to use it as a feature quantity.

<Prediction stage>
Fig. 9 is a block diagram showing the control unit 9 for monitoring abnormalities in the pressure data, and Fig. 10 is a diagram showing the flow of a monitoring process executed by the control unit 9 for monitoring the discharge pressure.

In this monitoring process, first, the discharge pressure measurement unit 911 of the control unit 9 acquires pressure data (Figure 10: step S21). Next, the feature calculation unit 913 calculates the feature of the acquired pressure data (Figure 10: step S22).

The anomaly degree calculation unit 915 inputs the features calculated by the feature calculation unit 913 into the unsupervised learning model M1 to calculate the anomaly degree (Figure 10: step S23). Furthermore, the anomaly probability calculation unit 917 inputs the features calculated by the feature calculation unit 913 into the supervised learning model M2 to calculate the probability of each anomaly (Figure 10: step S24).

Figure 11 is a conceptual diagram showing how the degree of anomaly (Mahalanobis distance) is calculated. In the example shown in Figure 11, multiple open circles ("○") indicate the position (distribution) in feature space of normal pressure data used in machine learning. Open triangles ("△") indicate the position in feature space of newly acquired pressure data. As shown in Figure 11, the anomaly calculation unit 915 calculates the Mahalanobis distance between the newly acquired pressure data (△) and the k-nearest (here, nearest) normal pressure data (○) as the degree of anomaly. The degree of anomaly is an index that indicates the degree to which the newly acquired pressure data (△) deviates from the normal pressure data (○).

Figure 12 is a diagram conceptually illustrating the output of the abnormality probability calculation unit 917. It illustrates examples of pressure waveforms for normal, abnormal A, and abnormal B. In the example shown in Figure 12, abnormality A is a state in which the discharge pressure rises excessively higher than in a normal state, resulting in an overshoot. On the other hand, abnormality B is a state in which the discharge pressure does not rise sufficiently compared to a normal state, and it takes a long time for the discharge pressure to reach steady-state pressure Pm.

The abnormality probability calculation unit 917 is configured to output a numerical value between -1 and +1 for newly acquired pressure data. As described above, machine learning is performed so that normal is 0, the output value for abnormality A is +1, and the output for 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.

Returning to Figure 9, the degree of abnormality calculated by the abnormality degree calculation unit 915 and the abnormality probability (a value between -1 and 1) calculated by the abnormality probability calculation unit 917 are sent to the abnormality determination unit 919. The abnormality determination unit 919 compares the degree of abnormality and the abnormality probability with thresholds to determine whether the newly obtained pressure data is normal or abnormal (Figure 10: Step S25).

Specifically, the abnormality determination unit 919 determines whether the degree of abnormality exceeds a predetermined threshold value Th1 (see FIG. 11). If the degree of abnormality exceeds threshold value Th1, the abnormality determination unit 919 determines that the pressure data is abnormal.

The abnormality determination unit 919 also determines whether the abnormality probability exceeds a predetermined threshold value. For example, as shown in FIG. 12, when determining whether the data corresponds to abnormality A or abnormality B, threshold values Th2 and Th3 are determined in advance. Threshold value Th2 is a numerical value greater than 0 and less than 1, and threshold value Th3 is a numerical value greater than -1 and less than 0. If the abnormality probability exceeds threshold value Th2 on the positive side (i.e., if the abnormality probability value is greater than threshold value Th2), the abnormality determination unit 919 determines that the new pressure data corresponds to abnormality A. If the abnormality probability exceeds threshold value Th3 on the negative side (i.e., if the abnormality probability falls below 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 Figure 12, the abnormality probability of the first pressure data (data 1) does not exceed thresholds Th2 and Th3, so the pressure data is determined to be normal. The abnormality probability of the second pressure data (data 2) exceeds threshold Th2 on the positive side, so the pressure data is determined to be abnormal A. The abnormality probability of the third pressure data (data 3) exceeds threshold Th3 on the negative side, so the pressure data is determined to be abnormal B.

Furthermore, the anomaly determination unit 919 may combine the results of the anomaly determinations using the anomaly level and the anomaly probability to make a final anomaly determination for the pressure data. Figure 13 is a conceptual diagram illustrating an example of a comprehensive anomaly determination using the anomaly level and the anomaly probability. In Figure 13, the horizontal axis represents the anomaly probability, and the vertical axis represents the anomaly level. If neither the anomaly level nor the anomaly probability exceeds the threshold (i.e., both are normal), the comprehensive determination of the pressure data is determined to be "normal" (corresponding to the third quadrant in Figure 13). If both the anomaly level and the anomaly probability exceed the threshold (i.e., both are abnormal), the comprehensive determination of the new pressure data is determined to be "anomaly." Furthermore, if the anomaly probability is high and the anomaly level is low, the comprehensive determination of the new pressure data is determined to be "anomaly close to a normal distribution." Conversely, if the anomaly probability is low and the anomaly level is high, the comprehensive determination of the new pressure data is determined to be "unknown anomaly," i.e., an anomaly not anticipated 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 may print out the determination result using a printer. The abnormality determination unit 919 may also output the determination result by turning on a lamp or outputting an alarm sound from a speaker, etc.

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 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 change over time in the pressure inside the nozzle 71 that discharges the treatment liquid. The feature calculation unit 913 calculates feature amounts of the pressure data. The abnormality degree calculation unit 915 inputs the feature amounts and calculates the abnormality degree for the feature amounts calculated by the feature calculation unit 913 using an unsupervised learning model M1 that outputs an abnormality degree indicating the degree of deviation from the distribution of normal feature amounts. The abnormality probability calculation unit 917 inputs the feature amounts and calculates the abnormality probability for the feature amounts calculated by the feature calculation unit 913 using a supervised learning model M2 that inputs the feature amounts and outputs an abnormality probability indicating the probability that the feature amounts correspond to a predetermined abnormality. The abnormality determination unit 919 determines whether the pressure data is abnormal using the degree of abnormality calculated by the abnormality degree calculation unit 915 and the abnormality probability calculated by the abnormality probability calculation unit 917.

The control unit 9 also executes a pressure monitoring method. The pressure monitoring method includes the steps of: a) acquiring pressure data indicating changes over time in pressure within the nozzle 71 that ejects the processing liquid; b) calculating feature quantities for the pressure data; c) calculating the degree of anomaly for the feature quantities calculated in step b) using an unsupervised learning model M1 that takes the feature quantities as input and outputs an anomaly degree indicating the degree of deviation from a normal distribution of feature quantities; d) calculating the anomaly probability for the feature quantities calculated in step b) using a supervised learning model M2 that takes the feature quantities as input and outputs an anomaly probability indicating the probability that the feature quantities correspond to a specific anomaly; and e) determining an anomaly in the pressure data using the anomaly degree calculated in step c) and the anomaly probability calculated in step d).

With this configuration, when a previously anticipated abnormality is detected, the abnormality can be identified by calculating the abnormality probability. Furthermore, even if an unknown abnormality that was not previously anticipated occurs, the unknown abnormality can be properly detected by calculating the abnormality degree.

Furthermore, the unsupervised learning model M1 is a model obtained by unsupervised learning using only the features of multiple normal pressure data as input data.

This configuration allows for learning only the distribution of normal features, enabling appropriate anomaly detection.

The unsupervised learning model M1 outputs the Mahalanobis distance as the degree of anomaly.

With this configuration, the degree of anomaly is calculated by taking into account the correlation between each dimension of the feature values indicated by the pressure data, making it possible to more accurately capture characteristic patterns and abnormal behavior in the pressure data.

The supervised learning model M2 outputs the probability of multiple types of specific anomalies (e.g., anomalies A and B).

This configuration allows for the calculation of multiple types of probabilities, making it possible to determine whether the pressure data corresponds to one of multiple types of abnormalities.

<Other embodiments>
In the above embodiment, the anomaly degree calculation unit 915 calculates the anomaly degree using the unsupervised learning model M1. However, it is not necessary to calculate the anomaly degree using the unsupervised learning model M1. For example, the anomaly degree calculation unit 915 may calculate the difference between a normal feature amount and a feature amount of newly obtained pressure data using a predetermined calculation formula. The difference corresponds to an anomaly degree indicating the degree of deviation of the feature amount of the newly obtained pressure data from the normal feature amount. Furthermore, the normal feature amount to be compared may be defined from the feature amounts of multiple normal pressure data, or may be the feature amount of a single normal pressure data.

The abnormality determination unit 919 compares the degree of abnormality (difference) calculated by the abnormality degree calculation unit 915 with a preset threshold value. If the degree of abnormality exceeds the threshold value, the abnormality determination unit 919 determines that the newly acquired pressure data is abnormal.

For example, Japanese Patent Application Laid-Open Publication No. 2012-098133 describes calculating the degree of similarity between the characteristic quantities of the impulse component under normal conditions and the characteristic quantities of the impulse component being judged when detecting an abnormality in an inverter device based on the impulse component of a current waveform. Furthermore, Japanese Patent Application Laid-Open Publication No. 2006-026584 describes calculating the difference between the voltage under normal conditions and the voltage being judged when detecting an ink ejection abnormality based on the voltage applied to an actuator that ejects ink. These similarities and differences may be used as the degree of abnormality described above.

Although this invention has been described in detail, the above description is illustrative in all respects and does not limit the invention. It is understood that countless variations not illustrated can be envisioned without departing from the scope of this invention. The configurations described in the above embodiments and variations can be combined or omitted as appropriate, as long as they are not mutually inconsistent.

1 Coating device 51 Chuck mechanism (substrate holding part)
52 Travel control mechanism 7 Application mechanism 71 Nozzle 86 Pressure sensor 9 Control unit (discharge pressure monitoring device)
91 Calculation unit 911 Discharge pressure measurement unit (pressure data acquisition unit)
913 Feature amount calculation unit 915 Abnormality degree calculation unit 917 Abnormality probability calculation unit 919 Abnormality determination unit 931 Computer program M1 Unsupervised learning model M2 Supervised learning model

Claims (5)

A coating device comprising:
a substrate holder for holding a substrate;
a nozzle that ejects a processing liquid toward the substrate held by the substrate holder;
a pressure sensor for measuring the pressure in the nozzle;
a discharge pressure monitoring device ;
Equipped with
The discharge pressure monitoring device
a pressure data acquisition unit that acquires pressure data indicating a time change in pressure inside the nozzle using the pressure sensor ;
a feature amount calculation unit that calculates a feature amount of the pressure data;
an anomaly degree calculation unit that calculates the anomaly degree for the feature calculated by the feature calculation unit using an unsupervised learning model that receives the feature as an input and outputs an anomaly degree that indicates the degree of deviation from a normal distribution of the feature;
an abnormality probability calculation unit that uses the feature amount as an input and calculates the abnormality probability for the feature amount calculated by the feature amount calculation unit using a supervised learning model that outputs an abnormality probability indicating the probability that the feature amount corresponds to a specific abnormality;
an abnormality determination unit that acquires the degree of abnormality calculated by the abnormality degree calculation unit and the abnormality probability calculated by the abnormality probability calculation unit, and determines whether the pressure data is in a normal state, an abnormality close to a normal distribution, an unknown abnormality, or an abnormal state based on a comparison of the degree of abnormality calculated by the abnormality degree calculation unit and the abnormality probability with threshold values;
Including,
The coating apparatus , wherein the characteristic amount is the degree of overshoot that occurs during a rise period of the discharge pressure, which is the period from the time when the discharge of the treatment liquid from the nozzle starts to the time when the discharge pressure reaches a target pressure . The coating device according to claim 1 ,
The unsupervised learning model is a model obtained by unsupervised learning using, as input data, feature quantities of only the plurality of normal pressure data. The coating device according to claim 1 or 2,
The coating apparatus, wherein the unsupervised learning model outputs a Mahalanobis distance as the degree of abnormality. The coating device according to claim 1 or 2,
The supervised learning model outputs probabilities of multiple types of specific anomalies. A coating device comprising:
a substrate holder for holding a substrate;
a nozzle that ejects a processing liquid toward the substrate held by the substrate holder;
a pressure sensor for measuring the pressure in the nozzle;
a discharge pressure monitoring device ;
Equipped with
The discharge pressure monitoring device
a pressure data acquisition unit that acquires pressure data indicating a time change in pressure inside the nozzle using the pressure sensor ;
a feature amount calculation unit that calculates a feature amount of the pressure data;
an abnormality degree calculation unit that calculates an abnormality degree indicating a degree of deviation from a normal feature amount for the feature amount calculated by the feature amount calculation unit;
an abnormality probability calculation unit that uses the feature amount as an input and calculates the abnormality probability for the feature amount calculated by the feature amount calculation unit using a supervised learning model that outputs an abnormality probability indicating the probability that the feature amount corresponds to a specific abnormality;
an abnormality determination unit that acquires the degree of abnormality calculated by the abnormality degree calculation unit and the abnormality probability calculated by the abnormality probability calculation unit, and determines whether the pressure data is in a normal state, an abnormality close to a normal distribution, an unknown abnormality, or an abnormal state based on a comparison of the degree of abnormality calculated by the abnormality degree calculation unit and the abnormality probability with threshold values;
Including,
The coating apparatus , wherein the characteristic amount is the degree of overshoot that occurs during a rise period of the discharge pressure, which is the period from the time when the discharge of the treatment liquid from the nozzle starts to the time when the discharge pressure reaches a target pressure .