State determination method and state determination system

Abstract

To provide a technique capable of determining a state of a device even when a variable related to the state of the device cannot be actually measured.SOLUTION: A machine learning model M in which a first variable that can be actually measured and a second variable that cannot be actually measured are explanatory variables and a third variable that can be actually measured is an objective variable is used. The measured value of the first variable and the second variable in the normal state are input to the machine learning model M, and the estimated value of the third variable output from the machine learning model M is acquired. Then, the state of the apparatus is determined by comparing the estimated value and the measured value of the third variable. Accordingly, even when the second variable itself cannot be actually measured, the state of the device can be determined.SELECTED DRAWING: Figure 4

Description

[Technical Field]

[0001] The present invention relates to a state determination method and a state determination system for determining the state of a device. [Background technology]

[0002] Conventionally, there is known a technique for monitoring the state of a device and determining whether the state of the device is normal. For example, Patent Document 1 describes a technique for obtaining estimated and measured temperatures of the bearings of a wind turbine, and comparing the estimated and measured values ​​to determine whether the wind turbine is abnormal. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7288794 Summary of the Invention [Problem to be solved by the invention]

[0004] As in Patent Document 1, when the variable to be judged is measurable, it is possible to judge whether the device is abnormal based on the difference between the estimated value and the actual measured value. However, when the variable to be judged is not measurable, it is not possible to judge whether the device is abnormal using the above method.

[0005] Therefore, an object of the present invention is to provide a technique that can determine the state of a device even when variables related to the state of the device cannot be actually measured. [Means for solving the problem]

[0006] A first invention of the present application is a state determination method for determining the state of an apparatus, the method comprising: an estimation step of inputting an actual measured value of a first variable that can be measured in the apparatus and a second variable that cannot be measured and that changes depending on the state of the apparatus as explanatory variables, and a third variable that can be measured in the apparatus as a target variable, into a machine learning model; an actual measurement step of acquiring the actual measured value of the third variable; and a determination step of determining the state of the apparatus by comparing the estimated value of the third variable with the actual measured value.

[0007] A second invention of the present application is a state determination method of the first invention, wherein in the determination step, if the difference between the estimated value and the actual measured value of the third variable is less than a threshold value, the state of the device is determined to be normal, and if the difference between the estimated value and the actual measured value of the third variable is equal to or greater than the threshold value, the state of the device is determined to be abnormal.

[0008] A third invention of the present application is a state determination method of the second invention, further comprising a simulation step of obtaining learning data for the explanatory variables and the target variable by simulation, and a learning step of creating the machine learning model based on the learning data.

[0009] A fourth invention of the present application is a state determination method of the third invention, further comprising a second variable determination step of inputting an actual measured value of the first variable into the machine learning model when the device is in a normal state, and determining the second variable in the normal state based on the difference between the estimated value of the third variable output from the machine learning model and the actual measured value of the third variable.

[0010] A fifth invention of the present application is a state determination method of the fourth invention, in which a plurality of data sets of actual measured values ​​of the first variable and the third variable are prepared, and the second variable determination step is executed for each of the data sets to determine a fluctuation range of the second variable under normal conditions and to determine the threshold value based on the fluctuation range.

[0011] A sixth aspect of the present invention is a state determination system for determining the state of an apparatus, the system including a machine learning model having a first variable that can be actually measured in the apparatus and a second variable that cannot be actually measured and that changes depending on the state of the apparatus as explanatory variables, and a third variable that can be actually measured in the apparatus as a target variable, and an estimation unit that inputs an actual measured value of the first variable and the second variable in a normal state into the machine learning model and acquires an estimated value of the third variable output from the machine learning model, an actual measured value acquisition unit that acquires the actual measured value of the third variable, and a determination unit that determines the state of the apparatus by comparing the estimated value of the third variable with the actual measured value. [Effects of the Invention]

[0012] According to the first to sixth inventions, the machine learning model outputs an estimated value of the third variable based on the second variable under normal conditions. In contrast, the actual measured value of the third variable is a value that reflects the second variable that has changed depending on the state of the device. Therefore, by comparing the estimated value and the actual measured value of the third variable, the state of the device can be determined even if the second variable itself cannot be actually measured.

[0013] In particular, according to the third aspect of the present invention, learning data for a second variable that cannot be actually measured by the device is obtained by simulation, thereby enabling the creation of a machine learning model.

[0014] In particular, according to the fourth aspect of the present invention, the second variable in a normal state can be determined without actually measuring the second variable.

[0015] In particular, according to the fifth aspect of the present invention, the threshold value for the difference between the estimated value and the actually measured value of the third variable can be determined in consideration of the fluctuation range of the second variable under normal conditions. [Brief explanation of the drawings]

[0016] [Figure 1] FIG. [Figure 2] FIG. 2 is a control block diagram of the exposure apparatus. [Figure 3] FIG. 1 is a diagram showing the configuration of a laser device and a state determination system. [Figure 4]FIG. 1 is a diagram showing input and output of a machine learning model. [Figure 5] 1 is a flowchart showing a procedure for constructing and using a state determination system. [Figure 6] FIG. 10 is a diagram showing an example of a thermal circuit used in a simulation. [Figure 7] FIG. 10 is a diagram showing the configuration of a main scanning device and a state determination system according to a second embodiment. [Figure 8] FIG. 10 is a diagram illustrating input and output of a machine learning model according to the second embodiment. [Figure 9] FIG. 10 is a diagram showing the configuration of a valve device and a state determination system according to a third embodiment. [Figure 10] FIG. 11 is a diagram illustrating input and output of a machine learning model according to the third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0017] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0018] 1. First Embodiment <1-1. Configuration of exposure equipment> FIG. 1 is a perspective view of an exposure apparatus 1 equipped with a state determination system according to one embodiment of the present invention. This exposure apparatus 1 is an apparatus that irradiates light onto the upper surface of a substrate 9, on which a photosensitive material is applied, to expose a pattern onto the upper surface of the substrate 9. The substrate 9 is, for example, a semiconductor wafer or a printed circuit board. As shown in FIG. 1, the exposure apparatus 1 includes a base 10, a gantry 15, a stage 20, a moving mechanism 25, an exposure unit 30, and a control unit 40.

[0019] The base 10 is a support platform that supports the gantry 15, the stage 20, the moving mechanism 25, and the exposure unit 30. The base 10 has a flat plate-like outer shape that extends along a horizontal plane. The base 10 is made of stone material such as quartz. The base 10 is fixed to the floor of the factory.

[0020] The base 10 has a rectangular shape when viewed from above. In the following, among the horizontal directions, the direction along the long sides of the base 10 will be referred to as the "main scanning direction," and the direction along the short sides of the base 10 will be referred to as the "sub-scanning direction." The main scanning direction and the sub-scanning direction are perpendicular to each other.

[0021] The gantry 15 is fixed to the upper surface of the base 10. The gantry 15 has a pair of legs 16 and a bridge portion 17. The pair of legs 16 are spaced apart in the sub-scanning direction. Each leg 16 extends upward from the upper surface of the base 10. The bridge portion 17 connects the upper ends of the legs 16 to each other in the sub-scanning direction. A through-hole 18 is formed between the upper surface of the base 10 and the gantry 15, allowing the substrate 9 to pass through in the main scanning direction.

[0022] The stage 20 is a holder on which the substrate 9 is placed. The stage 20 has a flat plate-like outer shape. When viewed from above, the stage 20 has a rectangular shape that is smaller than the base 10. The stage 20 is placed above the base 10 in a substantially horizontal position. The substrate 9 is supported on the upper surface of the stage 20 in a substantially horizontal position. The stage 20 may have chuck pins for fixing the substrate 9 and multiple suction holes for suctioning the substrate 9.

[0023] The movement mechanism 25 is a mechanism that moves the stage 20 relative to the base 10 and the gantry 15. The movement mechanism 25 has a main scanning device 26, a sub-scanning device 27, and an intermediate plate 28. The intermediate plate 28 is a flat member that is disposed between the base 10 and the stage 20.

[0024] The main scanning device 26 moves the intermediate plate 28 in the main scanning direction relative to the base 10. The main scanning device 26 has a pair of guide rails and a linear motor. The pair of guide rails are rails for guiding the intermediate plate 28 in the main scanning direction. The pair of guide rails are provided on the upper surface of the base 10 with a gap between them in the sub-scanning direction. Each guide rail extends linearly along the main scanning direction.

[0025] The linear motor of the main scanning device 26 has a stator provided on the upper surface of the base 10 and a slider fixed to the lower surface of the intermediate plate 28. When a drive signal is supplied to the linear motor from the control unit 40, the slider moves in the main scanning direction along the stator due to magnetic attractive and repulsive forces generated between the stator and slider. As a result, the intermediate plate 28 moves in the main scanning direction relative to the base 10.

[0026] The main scanning device 26 may use a mechanism that converts the rotational motion of the motor into linear motion via a ball screw instead of a linear motor to move the intermediate plate 28 in the main scanning direction.

[0027] The sub-scanning device 27 moves the stage 20 in the sub-scanning direction relative to the intermediate plate 28. The sub-scanning device 27 has a pair of guide rails and a linear motor. The pair of guide rails are rails for guiding the stage 20 in the sub-scanning direction. The pair of guide rails are provided on the upper surface of the intermediate plate 28 at a distance from each other in the main scanning direction. Each guide rail extends linearly in the sub-scanning direction.

[0028] The linear motor of the sub-scanning device 27 has a stator provided on the upper surface of the intermediate plate 28 and a slider fixed to the lower surface of the stage 20. When a drive signal is supplied to the linear motor from the control unit 40, the slider moves in the sub-scanning direction along the stator due to magnetic attractive and repulsive forces generated between the stator and slider. As a result, the stage 20 moves in the sub-scanning direction relative to the intermediate plate 28.

[0029] The sub-scanning device 27 may use a mechanism that converts the rotational motion of a motor into linear motion via a ball screw, instead of a linear motor, to move the stage 20 in the sub-scanning direction.

[0030] In the exposure apparatus 1, the substrate 9 is placed on the stage 20, and by operating the movement mechanism 25, the substrate 9 is moved together with the stage 20 in the main scanning direction and the sub-scanning direction.

[0031] The moving mechanism 25 may further include a mechanism for moving the stage 20 in the vertical direction, and a mechanism for rotating the stage 20 around an axis extending in the vertical direction.

[0032] The exposure unit 30 is a unit that exposes the substrate 9 supported by the stage 20. As shown in Fig. 1, the exposure unit 30 has a head 31, an illumination optical system 32, and a laser device 33. The head 31, the illumination optical system 32, and the laser device 33 are fixed to the bridge portion 17 of the gantry 15.

[0033] The laser device 33 emits laser light in accordance with a drive signal supplied from the control unit 40. The laser light emitted from the laser device 33 is introduced into the head 31 via the illumination optical system 32. A spatial modulator 34 is provided inside the head 31. For example, a GLV (Grating Light Valve) (registered trademark), which is a diffraction grating type spatial light modulator, is used as the spatial modulator 34. The laser light introduced into the head 31 is modulated into a predetermined pattern by the spatial modulator 34 and irradiated onto the upper surface of the substrate 9. As a result, the photosensitive material on the upper surface of the substrate 9 is exposed.

[0034] The exposure device 1 irradiates the substrate 9 with laser light from the head 31 while moving the substrate 9 in the main scanning direction and the sub-scanning direction using the movement mechanism 25. As a result, a pattern is exposed onto the upper surface of the substrate 9.

[0035] Control unit 40 is a unit for controlling the operation of each part of exposure apparatus 1. Figure 2 is a control block diagram of exposure apparatus 1. As shown in Figure 2, control unit 40 is made up of a computer having a processor 41 such as a CPU (Central Processing Unit), a memory 42 such as a RAM (Random Access Memory), and a storage unit 43 such as a hard disk drive.

[0036] The storage unit 43 stores a computer program P for performing the exposure process. As shown in FIG. 2, the control unit 40 is electrically connected to the main scanning device 26, the sub-scanning device 27, the laser device 33, and the spatial modulator 34. The control unit 40 controls the operation of each of the above-mentioned units in accordance with the computer program P. This allows the exposure process of the substrate 9 in the exposure apparatus 1 to proceed.

[0037] <1-2. Determining the status of the laser device> Fig. 3 is a diagram showing the configuration of the above-mentioned laser device 33 and a state determination system 50 that determines the state of the laser device 33. As shown in Fig. 3, the laser device 33 has a laser element 331, a temperature sensor 332, a cooling element 333, and a driver circuit 334. The laser element 331, the temperature sensor 332, and the cooling element 333 are electrically connected to the control unit 40 via the driver circuit 334.

[0038] The laser element 331 is an element that emits laser light in response to a drive current supplied from the control unit 40 via a driver circuit 334. The characteristics of the laser element 331, such as the threshold current and output wavelength, change depending on the temperature. For this reason, the driver circuit 334 controls the temperature of the laser element 331 using a temperature sensor 332 and a cooling element 333.

[0039] Temperature sensor 332 is a sensor that measures the temperature of laser element 331. Temperature sensor 332 is fixed to the surface of laser element 331 with adhesive or fasteners such as screws. Temperature sensor 332 measures the temperature of laser element 331 and outputs a detection signal indicating the measurement value to control unit 40 via driver circuit 334.

[0040] The cooling element 333 is an element that absorbs heat from the laser element 331 in response to the current supplied by the driver circuit 334. For example, a Peltier element is used as the cooling element 333. The cooling element 333 is fixed to the surface of the laser element 331 with adhesive or a fastener such as a screw. The driver circuit 334 controls the cooling element 333 so that the temperature measured by the temperature sensor 332 falls within an allowable range.

[0041] In such a laser device 33, long-term use causes changes over time in the temperature sensor 332, changes in the state of fixing the temperature sensor 332 to the laser element 331, changes over time in the cooling element 333, and changes in the state of fixing the cooling element 333 to the laser element 331. When these changes occur, an abnormality may occur in the temperature control function of the laser device 33. However, the changes over time and changes in the fixing state cannot be measured directly.

[0042] Therefore, the exposure apparatus 1 of this embodiment is equipped with a state determination system 50 that determines whether the state of the temperature adjustment function of the laser device 33 is normal or not, without directly measuring the above-mentioned changes over time or changes in the fixed state. The state determination system 50 will be described below.

[0043] The condition determination system 50 is mounted, for example, on the control unit 40 of the exposure apparatus 1. As shown in Fig. 3, the condition determination system 50 has an actual measurement value acquisition unit 51, an estimation unit 52, and a determination unit 53. The functions of the actual measurement value acquisition unit 51, the estimation unit 52, and the determination unit 53 are realized by the computer operating as the control unit 40 in accordance with the computer program P described above.

[0044] The actual measurement value acquiring unit 51 is a processing unit that acquires various actual measurement values ​​related to the operation of the laser device 33. For example, the actual measurement value acquiring unit 51 acquires the actual measurement value of the temperature of the laser element 331 from the temperature sensor 332 via the driver circuit 334. The actual measurement value acquiring unit 51 also acquires the actual measurement value of the output of the laser element 331 from the driver circuit 334. The output of the laser element 331 is, for example, the value of current supplied from the driver circuit 334 to the laser element 331. The actual measurement value acquiring unit 51 also acquires the actual measurement value of the output of the cooling element 333 from the driver circuit 334. The output of the cooling element 333 is, for example, the value of current supplied from the driver circuit 334 to the cooling element 333. The actual measurement value acquiring unit 51 also acquires the actual measurement value of the environmental temperature from an environmental thermometer (not shown).

[0045] The estimation unit 52 is a processing unit that estimates the temperature of the laser element 331 using a machine learning model M. The machine learning model M is a regression-type inference model created by a learning process described later using a machine learning algorithm. The machine learning algorithm used to obtain the machine learning model M is, for example, a so-called supervised machine learning algorithm, such as a neural network including a single-layer neural network or deep learning, a decision tree algorithm including a random forest or gradient boosting, or a support vector machine.

[0046] Fig. 4 is a diagram showing the input and output of the machine learning model M. As shown in Fig. 4, the machine learning model M uses the first variable and the second variable as explanatory variables.

[0047] The first variables are variables that can be measured in the laser device 33. In the example of FIG. 4, the first variables include the temperature of the laser element 331, the ambient temperature, the output of the laser element 331, and the output of the cooling element 333. The state determination system 50 acquires the actual measured values ​​of these first variables using the actual measured value acquisition unit 51. Then, the estimation unit 52 inputs the actual measured values ​​of the first variables acquired by the actual measured value acquisition unit 51 into the machine learning model M.

[0048] The first variables input to the machine learning model M may be the most recent measured values ​​or time-series data from a certain period in the past. Also, the first variables input to the machine learning model M may be only some of the four types of first variables shown in FIG.

[0049] The second variables are variables that cannot be measured in the laser device 33. In the example of Fig. 4, the second variables include the thermal resistance between the temperature sensor 332 and the laser element 331, the thermal resistance between the temperature sensor 332 and the outside air, the heat absorption efficiency of the cooling element 333, and the thermal resistance between the cooling element 333 and the laser element 331.

[0050] The second variable changes depending on the state of the laser device 33. For example, if the state of fixing the temperature sensor 332 to the laser element 331 changes due to long-term use, the thermal resistance between the temperature sensor 332 and the laser element 331 and the thermal resistance between the temperature sensor 332 and the outside air will change. Furthermore, the heat absorption efficiency of the cooling element 333 will change due to changes in the cooling element 333 over time. Furthermore, if the state of fixing the cooling element 333 to the laser element 331 changes due to long-term use, the thermal resistance between the cooling element 333 and the laser element 331 will change.

[0051] The estimation unit 52 inputs the second variables in the normal state determined in step S3, which will be described later, to the machine learning model M. Note that the second variables input to the machine learning model M may be only some of the four types of second variables shown in FIG.

[0052] As shown in Fig. 4, the machine learning model M uses a third variable as the objective variable. The third variable is a variable that can be actually measured in the laser device 33. In the example of Fig. 4, the third variable is the temperature of the laser element 331. More specifically, the third variable is the temperature of the laser element 331 at a time later than the temperature of the laser element 331 included in the first variable.

[0053] The third variable changes depending on the first variable and the second variable. The estimation unit 52 inputs the actual measured value of the first variable and the second variable in a normal state into the machine learning model M. Then, the machine learning model M outputs an estimated value of the third variable. This allows the estimation unit 52 to obtain an estimated value of the temperature of the laser element 331.

[0054] The determination unit 53 is a processing unit that determines the state of the laser device 33 by comparing the estimated value and actual measurement value of the third variable. The determination unit 53 acquires the estimated value of the temperature of the laser element 331 output from the above-mentioned machine learning model M from the estimation unit 52. The determination unit 53 also acquires the actual measurement value of the temperature of the laser element 331 from the actual measurement value acquisition unit 51. The determination unit 53 then compares the estimated value acquired from the estimation unit 52 with the actual measurement value acquired from the actual measurement value acquisition unit 51 to determine the state of the laser device 33.

[0055] Specifically, a threshold value for the difference between the estimated value and the actual measurement value of the temperature of the laser element 331 is stored in advance in the memory unit 43 of the control unit 40. The determination unit 53 determines that the state of the temperature adjustment function of the laser device 33 is abnormal when the difference between the estimated value acquired from the estimation unit 52 and the actual measurement value acquired from the actual measurement value acquisition unit 51 is equal to or greater than the threshold value.

[0056] Next, a description will be given of the procedure for constructing and using the above-described state determination system 50. Fig. 5 is a flowchart showing the procedure for constructing and using the above-described state determination system 50.

[0057] When constructing the state determination system 50, first, learning data is acquired to create the machine learning model M. However, among the explanatory variables of the machine learning model M, an actual measured value cannot be obtained for the second variable. Therefore, in this embodiment, learning data is acquired by simulation (step S1: simulation step).

[0058] Figure 6 is a diagram showing an example of the thermal circuit H used in the simulation of step S1. The thermal circuit H in Figure 6 represents the behavior of heat in the laser device 33 by imitating it with an electric circuit. In the thermal circuit H, temperature corresponds to voltage, and heat movement corresponds to current. In the thermal circuit H in Figure 6, the laser element 331 is the heat generating part, and the cooling element 333 is the heat absorbing part. In addition, in the thermal circuit H, the thermal resistance R, heat capacity C, outside air temperature G, and temperature sensor 332 are represented as a resistor, capacitor, ground, and voltmeter, respectively.

[0059] By using such a thermal circuit H, it is possible to simulate the temperature measurement value by the temperature sensor 332 when changing not only the output of the laser element 331, the output of the cooling element 333, and the outside air temperature G, which are measurable, but also the thermal resistance R and heat capacity C, which are difficult to measure. In step S1, learning data for the first and second variables, which are explanatory variables, and the corresponding third variable, which is the objective variable, is obtained by simulation using the thermal circuit H. This makes it possible to obtain learning data including the second variable, which is unmeasurable in the laser device 33.

[0060] In step S1, the behavior of heat may be simulated by a method other than the thermal circuit H. For example, learning data for explanatory variables and objective variables may be obtained by a simulation using analytical formulas or three-dimensional CAE (Computer Aided Engineering).

[0061] Next, a machine learning model M is created based on the learning data acquired in step S1 (step S2: learning process). In this step S2, the parameters of the machine learning model M are adjusted so that when learning data for the first variable and the second variable are input, corresponding learning data for the third variable is output. This results in a machine learning model M that can output an estimated value of the third variable in response to the input of the first variable and the second variable. In other words, a machine learning model M that can output an estimated value of the temperature of the laser element 331 is obtained.

[0062] Next, the second variable under normal conditions is determined based on the machine learning model M (step S3: second variable determination step). In this step S3, in the laser device 33 under normal conditions, a large number of actual measured values ​​of the first variable and the third variable are obtained while performing normal temperature control. Then, when the obtained actual measured values ​​of the first variable under normal conditions and the second variable are input into the machine learning model M, the second variable under normal conditions that minimizes the difference between the estimated value of the third variable output from the machine learning model M and the actual measured value of the third variable under normal conditions is determined as the second variable under normal conditions. In this way, the second variable under normal conditions can be determined without actually measuring the second variable.

[0063] Next, a threshold value for determining the state of the laser device 33 is determined (step S4: threshold value determination step). In this step S4, the many actual measured values ​​of the first variable and the third variable acquired in step S3 are divided into multiple sets of data sets. Then, for each of the data sets, a process similar to that of step S3 described above is performed to acquire multiple second variables under normal conditions. Then, based on the multiple second variables obtained, the fluctuation range of the second variable under normal conditions is determined. For example, the standard deviation or variance of the second variable is calculated based on the multiple second variables.

[0064] In this embodiment, the variation range of the second variable is considered to be a normal range, and a threshold value for the difference between the estimated value and the actual measured value of the third variable is determined. For example, when the actual measured value of the first variable and the maximum or minimum value of the variation range of the second variable are input to the machine learning model M, the difference between the estimated value of the third variable output from the machine learning model M and the actual measured value of the third variable is determined as the threshold. This makes it possible to determine the threshold value for the difference between the estimated value and the actual measured value of the third variable, taking into account the variation range of the second variable under normal conditions.

[0065] Thereafter, while using the laser device 33 in the exposure apparatus 1, the state determination system 50 uses the machine learning model M to acquire an estimated value of the third variable (step S5: estimation step). Specifically, while controlling the temperature of the laser device 33, the actual measurement value of the first variable acquired by the actual measurement value acquisition unit 51 and the second variable under normal conditions determined in step S3 are input to the machine learning model M. Then, an estimated value of the third variable output from the machine learning model M is acquired.

[0066] Furthermore, while using the laser device 33 in the exposure apparatus 1, the actual measurement value acquisition unit 51 acquires the actual measurement value of the third variable (step S6: actual measurement step). Then, the determination unit 53 compares the estimated value of the third variable acquired in step S5 with the actual measurement value of the third variable acquired in step S6 to determine the state of temperature adjustment control of the laser device 33 (step S7: determination step).

[0067] In step S7, if the difference between the estimated value of the third variable and the actual measured value of the third variable is less than the threshold determined in step S4, the determination unit 53 determines that the state of the temperature control of the laser device 33 is normal. Furthermore, if the difference between the estimated value of the third variable and the actual measured value of the third variable is equal to or greater than the threshold determined in step S4, the determination unit 53 determines that the state of the temperature control of the laser device 33 is abnormal. If the determination unit 53 determines that the state of the temperature control of the laser device 33 is abnormal, the determination unit 53 may display the fact that an abnormality has occurred on a display connected to the control unit 40.

[0068] As described above, the machine learning model M outputs an estimated value of the third variable based on the second variable under normal conditions. In contrast, the actual measured value of the third variable is a value that reflects the second variable, which changes depending on the state of the laser device 33. Therefore, by comparing the estimated value and the actual measured value of the third variable, the state of temperature adjustment control of the laser device 33 can be determined even if the second variable itself cannot be actually measured.

[0069] 2. Second Embodiment Next, a second embodiment of the present invention will be described. The second embodiment is an example in which the present invention is applied to determining the state of the main scanning device 26 of the exposure apparatus 1. Figure 7 is a diagram showing the configuration of the main scanning device 26 and a state determination system 50 according to the second embodiment.

[0070] 7, the main scanning device 26 has a pair of guide rails 261, a linear motor 262, an encoder 263, and a driver circuit 264. The linear motor 262 moves the intermediate plate 28 in the main scanning direction along the pair of guide rails 261. The encoder 263 detects the position of the intermediate plate 28 in the main scanning direction. The driver circuit 264 controls the operation of the linear motor 262 based on the position in the main scanning direction obtained from the encoder 263.

[0071] In such a main scanning device 26, as the device is used, changes occur in the frictional force of the intermediate plate 28 relative to the guide rails 261 and in the center of gravity of the intermediate plate 28. These changes may cause abnormalities in the operation of the main scanning device 26. However, the changes in the frictional force and the center of gravity cannot be measured directly.

[0072] Therefore, the condition determination system 50 of this embodiment determines whether the condition of the main scanning device 26 is normal without directly measuring the changes in frictional force or the changes in the center of gravity. As in the first embodiment, the condition determination system 50 has an actual measurement value acquisition unit 51, an estimation unit 52, and a determination unit 53.

[0073] The actual measurement value acquisition unit 51 acquires from the driver circuit 264 the actual measurement value of the acceleration of the intermediate plate 28 in the main scanning direction, the actual measurement value of the velocity of the intermediate plate 28 in the main scanning direction, the actual measurement value of the position of the intermediate plate 28 in the main scanning direction, the actual measurement value of the position of the stage 20 in the sub-scanning direction relative to the intermediate plate 28, and the actual measurement value of the thrust of the linear motor 262.

[0074] The estimation unit 52 estimates the thrust of the linear motor 262 using the machine learning model M. Fig. 8 is a diagram showing input and output of the machine learning model M according to the second embodiment. As shown in Fig. 8, the machine learning model M uses the first variable and the second variable as explanatory variables and the third variable as a response variable.

[0075] In this embodiment, the first variable, which is an explanatory variable, is the acceleration of the intermediate plate 28 in the main scanning direction, the velocity of the intermediate plate 28 in the main scanning direction, the position of the intermediate plate 28 in the main scanning direction, and the position of the stage 20 in the sub-scanning direction relative to the intermediate plate 28. In addition, in this embodiment, the second variable, which is an explanatory variable, is the friction force of the intermediate plate 28 with respect to the guide rail 261 and the center of gravity of the intermediate plate 28. In addition, in this embodiment, the third variable, which is a response variable, is the thrust of the linear motor 262.

[0076] In this embodiment, similarly to the above-described embodiment, the following steps are performed: acquiring learning data through simulation (step S1), creating a machine learning model M (step S2), determining a second variable in a normal state (step S3), determining a threshold value (step S4), acquiring an estimated value of a third variable (step S5), and acquiring an actual measured value of the third variable (step S6).The estimated value and the actual measured value of the third variable are then compared to determine the state of the main scanning device 26 (step S7).

[0077] The machine learning model M outputs an estimated value of the third variable based on the second variable under normal conditions. In contrast, the actual measured value of the third variable is a value that reflects the second variable, which changes depending on the state of the main scanning device 26. Therefore, by comparing the estimated value and the actual measured value of the third variable, the state of temperature regulation control of the main scanning device 26 can be determined even if the second variable itself cannot be actually measured.

[0078] 3. Third Embodiment Next, a third embodiment of the present invention will be described. The third embodiment is an example in which the present invention is applied to determining the state of a valve device 60 that controls the flow rate of a liquid. Fig. 9 is a diagram showing the configuration of a valve device 60 and a state determination system 50 according to the third embodiment.

[0079] As shown in Fig. 9, the valve device 60 has a flow meter 61, a pressure meter 62, a needle valve 63, and a driver circuit 64. The flow meter 61 measures the flow rate of the liquid flowing through a pipe 65. The pressure meter 62 measures the pressure of the liquid in the pipe 65. The needle valve 63 opens and closes the flow path in the pipe 65 by operating a valve 632 with a linear motor 631. The driver circuit 64 controls the operation of the needle valve 63 based on the measurement values ​​obtained from the flow meter 61 and the pressure meter 62. In this way, the flow rate of the liquid in the pipe 65 is controlled to be constant.

[0080] In such a valve device 60, the shape of the tip of the valve 632 gradually changes with use. When the shape of the tip of the valve 632 changes, abnormalities may occur in the control of the flow rate by the valve device 60. However, the change in the shape of the tip of the valve 632 cannot be measured directly.

[0081] Therefore, the state determination system 50 of this embodiment determines whether the state of the valve device 60 is normal or not without directly measuring the change in the shape of the valve 632. As in the first and second embodiments, the state determination system 50 has an actual measurement value acquisition unit 51, an estimation unit 52, and a determination unit 53.

[0082] The actual measurement value acquisition unit 51 acquires the actual measurement value of the liquid flow rate, the actual measurement value of the liquid pressure, and the actual measurement value of the opening of the needle valve 63 from the driver circuit 64. The estimation unit 52 outputs an estimated value of the liquid flow rate using the machine learning model M. FIG. 10 is a diagram showing input and output of the machine learning model M according to the third embodiment. As shown in FIG. 10, the machine learning model M uses the first variable and the second variable as explanatory variables and the third variable as a target variable.

[0083] In this embodiment, the first variable, which is an explanatory variable, is the flow rate of the liquid, the pressure of the liquid, and the opening of the needle valve 63. In addition, in this embodiment, the second variable, which is an explanatory variable, is the shape of the tip of the valve 632 of the needle valve 63. In addition, in this embodiment, the third variable, which is an objective variable, is the flow rate of the liquid. More specifically, the third variable is the flow rate of the liquid at a time later than the flow rate of the liquid included in the first variable.

[0084] In this embodiment, similarly to the above-described embodiment, the steps of acquiring learning data by simulation (step S1), creating a machine learning model M (step S2), determining a second variable in a normal state (step S3), determining a threshold value (step S4), acquiring an estimated value of a third variable (step S5), and acquiring an actual measured value of the third variable (step S6) are performed. Then, the estimated value and the actual measured value of the third variable are compared to determine the state of the valve device 60 (step S7).

[0085] The machine learning model M outputs an estimated value of the third variable based on the second variable under normal conditions. In contrast, the actual measured value of the third variable is a value that reflects the second variable, which changes depending on the state of the valve device 60. Therefore, by comparing the estimated value and the actual measured value of the third variable, the state of temperature regulation control of the valve device 60 can be determined even if the second variable itself cannot be actually measured.

[0086] <4. Modifications> Although one embodiment of the present invention has been described above, the present invention is not limited to the above embodiment.

[0087] In the above embodiment, an example has been described in which the present invention is applied to the laser device 33, the main scanning device 26, and the valve device 60. However, the present invention may also be applied to devices other than these. For example, the present invention may be used to determine the status of the ink temperature control function in an inkjet printing device.

[0088] The first, second, and third variables are not limited to those shown in the above embodiment, and appropriate parameters may be used depending on the object of determination.

[0089] Furthermore, the elements appearing in the above-described embodiments and modifications may be selected as appropriate within the scope of not causing any contradiction. [Explanation of symbols]

[0090] 1: Exposure equipment 9: Circuit board 10: Bass 15: Gantry 20: Stage 25: Movement mechanism 26: Main scanning device 27: Sub-scanning device 28: Intermediate plate 30: Exposure area 31: Head 32: Illumination optical system 33: Laser device 34: Spatial modulator 40: Control section 50: Status determination system 51: Actual measurement value acquisition unit 52: Estimation part 53: Judgment section 60: Valve device 61:Flowmeter 62: Pressure gauge 63: Needle valve 64: Driver circuit 261: Guide rail 262: Linear motor 263: Encoder 264: Driver circuit 331: Laser element 332: Temperature sensor 333: Cooling element 334: Driver circuit 631: Linear motor 632: Valve H: Heat circuit M: Machine learning model

Claims

1. A state determination method for determining a state of a device, comprising: an estimation step of inputting the actual measured value of the first variable and the second variable in a normal state into a machine learning model having a first variable that can be actually measured in the device and a second variable that cannot be actually measured and that changes depending on the state of the device as explanatory variables and a third variable that can be actually measured in the device as a response variable, and acquiring an estimated value of the third variable output from the machine learning model; an actual measurement step of acquiring an actual measured value of the third variable; a determining step of determining a state of the device by comparing the estimated value and the actual measured value of the third variable; A state determination method comprising:

2. 2. The state determination method according to claim 1, In the determination step, determining that the state of the device is normal when the difference between the estimated value and the actual measured value of the third variable is less than a threshold value; a state determination method for determining that the state of the device is abnormal when a difference between the estimated value and the actual measured value of the third variable is equal to or greater than the threshold value;

3. The state determination method according to claim 2, a simulation step of acquiring learning data for the explanatory variables and the objective variables by simulation; a learning process for creating the machine learning model based on the learning data; The state determination method further comprises:

4. The state determination method according to claim 3, a second variable determination step of inputting an actual measurement value of the first variable to the machine learning model when the device is normal, and determining the second variable when the device is normal based on a difference between the estimated value of the third variable output from the machine learning model and the actual measurement value of the third variable; The state determination method further comprises:

5. The state determination method according to claim 4, A state determination method comprising: preparing a plurality of data sets of actual measured values ​​of the first variable and the third variable; and executing the second variable determination step for each of the data sets to determine a fluctuation range of the second variable under normal conditions, and determining the threshold value based on the fluctuation range.

6. A state determination system for determining a state of a device, comprising: an estimation unit that inputs the actual measured value of a first variable that can be measured in the device and a second variable that cannot be measured and that changes depending on the state of the device as explanatory variables, and a third variable that can be measured in the device as a response variable, into a machine learning model, and acquires an estimated value of the third variable output from the machine learning model; an actual measurement value acquisition unit that acquires an actual measurement value of the third variable; a determination unit that determines a state of the device by comparing the estimated value and the actual measured value of the third variable; A state determination system having:

Images