Laser beam straightness measuring machine and laser beam straightness measuring method

Abstract

To provide a straightness measuring instrument capable of achieving high accuracy even with a simple configuration.SOLUTION: According to a straightness measuring instrument 10, various two dimensional intensity distributions Iri (X, Y) of a laser beam Lr for which air fluctuation errors (δXri, δYri) and straightness errors (2Xrti, 2Yrti) are known are defined as input data, air fluctuation errors (δXri, δYri) and straightness errors (2Xrti, 2Yrti) corresponding to the input data are defined as output data, and supervised learning is performed in advance. With respect to unknown two-dimensional intensity distributions Imi (X, Y) detected by a two-dimensional image sensor 11, air fluctuation errors (δXmi, δYmi) and straightness errors (2Xmti, 2Ymti) are estimated based on results of the supervised learning.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a laser beam straightness measuring instrument for measuring the straightness of a laser beam and a method for measuring the straightness of a laser beam.

Background Art

[0002] In the fields of precision measurement and precision machining, there is a method of utilizing the straight traveling property of a laser for straightness, which is one of the important measurement items (see, for example, Patent Documents 1 and 2). Although various methods for measuring straightness using a laser have been studied, currently, a type using laser interference and a type directly detecting fluctuations in the beam position on a common optical path distance sensor are commercially available as measuring instruments, respectively. In the present invention, the latter measuring instrument is dealt with. Since many of these laser-based measurements are performed in the atmosphere, the air refractive index fluctuates due to environmental factors and appears as a systematic error in the measurement results. This temporal and spatial change in the air refractive index is called air turbulence. This air turbulence causes fluctuations in the wavefront of the laser beam (generation of wavefront aberration), causes changes in the beam profile, and impairs the straight traveling property of the laser beam.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] Methods for compensating for the effects of atmospheric turbulence are known as adaptive optics in astronomy and have seen remarkable development. In the field of precision measurement, methods such as digital optical phase conjugation have been proposed. In digital optical phase conjugation, the wavefront aberration of the optical path is measured with a wavefront sensor, and the wavefront aberration is removed by applying an aberration conjugate to the measured wavefront aberration to the beam (for example, by reflection with a spatial optical modulator). However, digital optical phase conjugation has the problem of being complex in configuration, such as requiring a wavefront sensor.

[0005] Therefore, the object of the present invention is to provide a laser beam straightness measuring machine and the like that can achieve high accuracy with a simple configuration by focusing on the beam profile (intensity distribution of the laser beam) due to aberrations. [Means for solving the problem]

[0006] The laser beam straightness measuring device according to the present invention is A two-dimensional image sensor that receives a laser beam and detects the two-dimensional intensity distribution of the laser beam, A machine learning device that estimates the air fluctuation error and straightness error of the laser beam based on the two-dimensional intensity distribution detected by the two-dimensional image sensor, It is equipped with. The aforementioned machine learning device is A means for performing supervised learning in advance, using various two-dimensional intensity distributions, each with known air fluctuation error and straightness error, as input data, and the corresponding air fluctuation error and straightness error as output data. A means for estimating the air fluctuation error and the straightness error in a two-dimensional intensity distribution, where the air fluctuation error and the straightness error are unknown, based on the results of supervised learning. It holds.

[0007] The laser beam straightness measurement method according to the present invention is: A two-dimensional image sensor that receives a laser beam and detects the two-dimensional intensity distribution of the laser beam, A machine learning device that estimates the air fluctuation error and straightness error of the laser beam based on the two-dimensional intensity distribution detected by the two-dimensional image sensor, This involves using the following steps, and includes the next step. (1) A variety of two-dimensional intensity distributions, each with known air fluctuation error and straightness error, are used as input data, and the corresponding air fluctuation error and straightness error are used as output data. The machine learning device is then subjected to supervised training in advance. (2) Using the supervised machine learning device described above, the air fluctuation error and the straightness error are estimated for the two-dimensional intensity distribution for which the air fluctuation error and the straightness error are unknown. [Effects of the Invention]

[0008] According to the present invention, the two-dimensional intensity distribution of various laser beams with known air fluctuation error and straightness error is used as input data, and the corresponding air fluctuation error and straightness error are used as output data. Supervised learning is performed in advance, and the air fluctuation error and straightness error are estimated for an unknown two-dimensional intensity distribution detected by a two-dimensional image sensor based on the results of supervised learning. In conventional numerical calculations, it has been extremely difficult to accurately separate and determine the air fluctuation error and straightness error in a short time for an unknown two-dimensional intensity distribution in which air fluctuation error and straightness error are mixed. In contrast, according to the present invention, the air fluctuation error and straightness error can be immediately and accurately estimated for an unknown two-dimensional intensity distribution, so it is possible to provide a laser beam straightness measuring machine that can achieve high accuracy with a simple configuration. [Brief explanation of the drawing]

[0009] [Figure 1] This is a block diagram showing the straightness measuring machine of Embodiment 1. [Figure 2] This is a block diagram showing a machine learning device in Embodiment 1. [Figure 3] This is a block diagram showing the straightness measuring machine of Embodiment 2. [Figure 4]It is a block diagram showing a machine learning device in Embodiment 2. [Figure 5] It is a flowchart showing the straightness measurement method of Embodiment 3. [Figure 6] It is a flowchart showing a detailed example of the steps in FIG. 5. FIG. 6[A] shows step S1 and FIG. 6[B] shows step S2. [Figure 7] It is a conceptual diagram for explaining air fluctuations and wavefront aberration. [Figure 8] It is a conceptual diagram for explaining the deviation of straightness. [Figure 9] It is a conceptual diagram for explaining the case where air fluctuation error and straightness error are mixed.

Embodiments for Carrying Out the Invention

[0010] Hereinafter, embodiments for carrying out the present invention (hereinafter referred to as "embodiments") will be described with reference to the accompanying drawings. In the following description, the "laser beam straightness measuring machine" and the "laser beam straightness measuring method" are simply referred to as the "straightness measuring machine" and the "straightness measuring method".

[0011] <Air Fluctuations and Wavefront Aberration> FIG. 7 shows a conceptual diagram of the influence of air fluctuations. Since a complex refractive index distribution exists on the optical path due to air fluctuations, a phase difference (wavefront aberration) occurs in each part of the laser beam 72 emitted from the laser light source 71. As described above, when there is air fluctuation 73 between the A plane and the B plane, a phase difference occurs in the wavefront, which is the equiphase surface of the laser beam 72, and the wavefront is deformed (wavefront aberration). Also, the light ray as the normal line of the wavefront passes through a point different from when there is no deformation, causing deformation to the beam profile 74. The deformation of the beam profile 74 affects the calculation of the luminance center of gravity as the air fluctuation error δ and becomes a factor that causes a systematic error in straightness measurement.

[0012] <Deviation of Straightness> FIG. 8 shows a conceptual diagram of the deviation in straightness. The straightness measuring machine 80 in FIG. 8 includes a laser light source 71, a corner cube 81, a moving stage 82, and a CCD sensor 83. The moving stage 82 is, for example, an X-Y stage and supports the corner cube 81 movably. The laser beam 72 emitted from the laser light source 71 is reflected by the corner cube 81 and received by the CCD sensor 83. Here, it is assumed that the corner cube 81 moves by ΔL in the traveling direction of the laser beam 72 and moves by S perpendicular to the traveling direction. At this time, the light reception position moves by 2S in the CCD sensor 83. This 2S is hereinafter referred to as the straightness error t.

[0013] <When the air fluctuation error and the straightness error are mixed> FIG. 9 shows a conceptual diagram when the air fluctuation error and the straightness error are mixed. The laser beam 91 emitted from the laser light source 71 and reflected by the corner cube 81 generates a straightness error t in the corner cube 81 and a phase difference (wavefront aberration) in each part due to air fluctuation. As described above, when there is air fluctuation 73 between the A plane and the B plane, a phase difference occurs in the wavefront, which is the equiphase surface of the laser beam 91, and the wavefront is deformed (wavefront aberration). Also, the light ray as the normal line of the wavefront passes through a point different from when there is no deformation, causing deformation in the beam profile 92. The deformation of the beam profile 92 affects the calculation of the luminance center of gravity as the air fluctuation error δ together with the straightness error t, becoming a factor that causes a systematic error in the straightness measurement.

[0014] <Embodiment 1> FIG. 1 is a block diagram showing the straightness measuring machine of Embodiment 1. FIG. 2 is a block diagram showing the machine learning device in Embodiment 1. Hereinafter, description will be made based on FIGS. 1 and 2.

[0015] The meanings of the main subscripts used hereinafter are as follows. i: measurement number r: learning data (known) m: estimation data (unknown) δ: air fluctuation t: straightness

[0016] The straightness measuring device 10 is equipped with a two-dimensional image sensor 11 and a machine learning device 12. The two-dimensional image sensor 11 receives the laser beam Lm and detects the two-dimensional intensity distribution Imi(X,Y) of the laser beam Lm. Based on the two-dimensional intensity distribution Imi(X,Y) detected by the two-dimensional image sensor 11, the machine learning device 12 estimates the air fluctuation error (δXmi,δYmi) and the straightness error (2Xmti,2Ymti) of the laser beam Lm.

[0017] The machine learning device 12 includes learning means and estimation means. In Figure 1, the machine learning device 12 operates as an estimation means, and in Figure 2, the machine learning device 12 operates as a learning means.

[0018] The learning method uses various two-dimensional intensity distributions Iri(X,Y) with known air fluctuation errors (δXri,δYri) and straightness errors (2Xrti,2Yrti) as input data, and the corresponding air fluctuation errors (δXri,δYri) and straightness errors (2Xrti,2Yrti) as output data, and performs supervised learning in advance. The estimation method estimates the air fluctuation errors (δXmi,δYmi) and straightness errors (2Xmti,2Ymti) for two-dimensional intensity distributions Imi(X,Y) with unknown air fluctuation errors (δXmi,δYmi) and straightness errors (2Xmti,2Ymti) based on the results of supervised learning.

[0019] The two-dimensional image sensor 11 is, for example, a CMOS sensor or a CCD sensor. Here, "two-dimensional" refers to the XY plane perpendicular to the laser beam.

[0020] The machine learning device 12 is, for example, a neural network 120 in which each means is constructed by a computer program within a computer. The neural network 120 is a general type in which many neurons are connected to each other by input and output, and consists of an input layer 121, a hidden layer 122, and an output layer 123. In this embodiment 1, the machine learning device 12 is pre-trained using various two-dimensional intensity distributions Iri(X,Y) in which the air fluctuation error (δXri,δYri) and straightness error (2Xrti,2Yrti) are known. Although the hidden layer 122 is shown as multiple layers in Figure 2, it may be a single layer. As the neural network 120, a convolutional neural network (CNN) consisting of an input layer, a convolutional layer, an activation function, a pooling layer, a fully connected layer, and an output layer may be used.

[0021] The straightness measuring machine 10 of this embodiment 1 is equipped with a laser light source 13 and a retroreflector 14. The laser light source 13 can be a general-purpose one such as a semiconductor laser or a He-Ne laser. The retroreflector 14 is a device or surface that reflects incident light parallel to and opposite to the direction of incidence, such as a corner cube prism. The retroreflector 14 is attached to the object to be measured, for example.

[0022] The operation of the straightness measuring device 10 is as follows:

[0023] First, the laser beam Lo emitted from the laser light source 13 undergoes a shift in straightness (Xmti, Ymti) before and after reflection by the retroreflector 14, and passes through the region 15 where air fluctuations exist, resulting in a laser beam Lm with waveform distortion. The laser beam Lm is received by the two-dimensional image sensor 11, which detects a two-dimensional intensity distribution Imi(X,Y) that is affected by the shift in straightness and air fluctuations.

[0024] Next, the two-dimensional intensity distribution Imi(X,Y) data is input to the machine learning device 12, which outputs data for the air turbulence error (δXmi,δYmi) and straightness error (2Xmti,2Ymti) estimated from the input data. This input-to-output process is extremely short because it does not require complex calculations compared to calculating the luminance centroid using the input data. Since air turbulence changes moment by moment, higher processing speed leads to higher accuracy. The sum of the air turbulence error (δXmi,δYmi) and straightness error (2Xmti,2Ymti) is the luminance centroid (X,Y)mi.

[0025] The effects of the straightness measuring device 10 are as follows:

[0026] The straightness measuring device 10 takes the two-dimensional intensity distribution Iri(X,Y) of various laser beams Lr for which the air fluctuation error (δXri,δYri) and straightness error (2Xrti,2Yrti) are known as input data, and the air fluctuation error (δXri,δYri) and straightness error (2Xrti,2Yrti) corresponding to the input data as output data. Supervised learning is performed in advance, and for an unknown two-dimensional intensity distribution Imi(X,Y) detected by the two-dimensional image sensor 11, the air fluctuation error (δXmi,δYmi) and straightness error (2Xmti,2Ymti) are estimated based on the results of supervised learning.

[0027] In conventional numerical calculations, it was extremely difficult to quickly and accurately separate the air fluctuation error (δXmi,δYmi) and the straightness error (2Xmti,2Ymti) from an unknown two-dimensional intensity distribution Imi(X,Y) which is a mixture of air fluctuation error and straightness error. In contrast, the straightness measuring device 10 can immediately and accurately estimate the air fluctuation error (δXmi,δYmi) and straightness error (2Xmti,2Ymti) from an unknown two-dimensional intensity distribution Imi(X,Y), thus achieving high accuracy despite its simple configuration.

[0028] <Embodiment 2> Figure 3 is a block diagram of the straightness measuring machine of Embodiment 2. Figure 4 is a block diagram of the machine learning device in Embodiment 2. The following explanation will be based on Figures 3 and 4. In Embodiment 2, the same reference numerals are used for the same components as in Embodiment 1 to avoid redundant explanations.

[0029] The straightness measuring device 20 further includes a wavefront distortion application device 21 that applies arbitrary wavefront distortion to a laser beam Lo passing through a region 26 where no air fluctuations exist, and a computing device 22 that calculates the luminance centroid (X,Y)mi of the two-dimensional intensity distribution Imi(X,Y) detected by the two-dimensional image sensor 11 for the laser beam Lm to which the wavefront distortion has been applied. The air fluctuation error (δXri,δYri) used in supervised learning is the luminance centroid (X,Y)mi calculated by the computing device 22.

[0030] The straightness error (2Xrti,2Yrti) used in supervised learning is applied to the two-dimensional intensity distribution Imi(X,Y) by an image processing means that translates the two-dimensional intensity distribution Imi(X,Y). Translation means that all points on the figure move in the same direction by the same distance. The straightness error (2Xrti,2Yrti) is applied to the two-dimensional intensity distribution Imi(X,Y) to become the two-dimensional intensity distribution Iri(X,Y). The image processing means is provided in the computing device 22, but it may also be provided in the machine learning device 12.

[0031] The two-dimensional intensity distribution Imi(X,Y) of a laser beam Lm with wavefront distortion is given a known straightness error (2Xrti,2Yrti) to obtain the two-dimensional intensity distribution Iri(X,Y), which is used as the input data for supervised learning. The luminance centroid (X,Y)mi calculated by the computing device 22 is used as the air fluctuation error (δXri,δYri), which is used as the output data (training data) corresponding to the input data. The known straightness error (2Xrti,2Yrti) is used as the output data (training data) corresponding to the input data.

[0032] The wavefront distortion device 21 includes a shape-variable mirror 23 having a mirror surface 231 that reflects the laser beam Lo and a driving element 232 that deforms the shape of the mirror surface 231, and a control device 24 that controls the driving element 232 using a Zernike approximation polynomial. The shape-variable mirror 23 can be a general type, and the driving element 232 is, for example, a piezoelectric element. A phase modulator or the like may be used instead of the shape-variable mirror 23. Furthermore, in order to realize a region 26 in which there are no air fluctuations, the shape-variable mirror 23, the retroreflector 14, and the two-dimensional image sensor 11 are housed in a housing 27. Note that "a region in which there are no air fluctuations" means, in other words, a region in which air fluctuations can be ignored.

[0033] The control device 24, along with the computing device 22 and the machine learning device 12, is built within the computer 25 using a computer program. This computer program may be recorded on a non-temporary recording medium (e.g., optical disc, semiconductor memory, etc.), read from the recording medium, and executed by the computer 25. The computer 25 is, for example, a personal computer. The computer 25 can also be replaced with an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0034] While the straightness measuring machine of Embodiment 1 is configured for actual measurement, the straightness measuring machine 20 is characterized by a configuration for supervised learning. The operation of the straightness measuring machine 20 is as follows. However, since it is well known in fields such as adaptive optics that a mirror surface 231 of a desired shape can be obtained by controlling the driving element 232 using a Zernike approximation polynomial, a detailed explanation will be omitted.

[0035] First, the laser beam Lo emitted from the laser light source 13 is subjected to wavefront distortion when reflected by the mirror surface 231 of the shape-variable mirror 23. The laser beam Lm reflected from the mirror surface 231 passes through a region 26 where there are no air fluctuations before and after reflection by the retroreflector 14, and is received by the two-dimensional image sensor 11 while maintaining the applied waveform distortion. The two-dimensional image sensor 11 outputs data of the two-dimensional intensity distribution Imi(X,Y) affected by the applied wavefront distortion to the computing device 22. The computing device 22 calculates the luminance centroid (X,Y)mi for the two-dimensional intensity distribution Imi(X,Y). Calculating the luminance centroid is equivalent to, for example, finding the centroid of a flat plate in which the light intensity in the XY plane is replaced by mass. The luminance centroid (X,Y)mi is the air fluctuation error (δXri,δYri).

[0036] Next, the two-dimensional intensity distribution Imi(X,Y) is translated by (2Xrti,2Yrti) to obtain the two-dimensional intensity distribution Iri(X,Y) by adding a straightness error (2Xrti,2Yrti) to the two-dimensional intensity distribution Imi(X,Y).

[0037] Next, the two-dimensional intensity distribution Iri(X,Y) and its pair of data, consisting of the air fluctuation error (δXri,δYri) and straightness error (2Xrti,2Yrti), are input to the machine learning device 12. The machine learning device 12 adjusts the weighting of each neuron so that, when the two-dimensional intensity distribution Iri(X,Y) is input data, the training data, consisting of the air fluctuation error (δXri,δYri) and straightness error (2Xrti,2Yrti), becomes the output data. General deep learning can be used for this learning.

[0038] The above operation is repeated, for example, tens of thousands of times, while gradually changing the shape of the mirror surface 231 and gradually changing the straightness error (2Xrti, 2Yrti), to collect tens of thousands of pairs of data of the two-dimensional intensity distribution Iri(X,Y) and its air turbulence error (δXri, δYri) and straightness error (2Xrti, 2Yrti), and these data are used to train the machine learning device 12. At this time, it is also possible to first acquire tens of thousands of two-dimensional intensity distributions Iri(X,Y) and then calculate their luminance centroids (X,Y)mi to collect tens of thousands of pairs of data.

[0039] The effects of the straightness measuring device 20 are as follows:

[0040] According to the straightness measuring device 20, it is further equipped with a wavefront distortion application device 21 that applies arbitrary wavefront distortion to a laser beam Lo passing through a region 26 where no air fluctuations exist, and a computing device 22 that calculates the luminance centroid (X,Y)mi of the two-dimensional intensity distribution Imi(X,Y) detected by the two-dimensional image sensor 11, thereby enabling accurate collection of input and output data used in supervised learning.

[0041] Furthermore, if the wavefront distortion imparting device 21 is configured with a shape-variable mirror 23 that deforms the shape of the mirror surface 231 by a driving element 232, and a control device 24 that controls the driving element 232 using a Zernike approximation polynomial, various waveform distortions can be easily realized, allowing for the collection of diverse data. For example, when artificial air fluctuations are created using a fan or similar device to impart wavefront distortion to a laser beam, only a limited number of wavefront distortions can be imparted.

[0042] In addition, since the machine learning device 12 can be trained with a wide variety of data, the performance of the machine learning device 12 can be improved. In other words, its ability to obtain correct answers even for unknown data is improved. Therefore, the straightness measuring machine 20 can achieve even higher accuracy despite its simple configuration.

[0043] When a straightness error (2Xrti,2Yrti) is added to a two-dimensional intensity distribution Imi(X,Y) by translating it using image processing, it is possible to easily add the straightness error (2Xrti,2Yrti) using only software without requiring special hardware, compared to when the straightness error (2Xrti,2Yrti) is added by moving the retroreflector 14 on an XY stage or the like. The other operations and effects of this second embodiment are the same as those of the first embodiment.

[0044] <Embodiment 3> Embodiment 3 considers the present invention as a method invention. Figure 5 is a flowchart of the straightness measurement method of Embodiment 3. Figure 6 is a flowchart of a detailed example of steps S1 and S2 in Figure 5. The following explanation will be based on Figures 1 to 6.

[0045] In the straightness measurement method of this third embodiment, a two-dimensional image sensor 11 and a machine learning device 12 are used. The two-dimensional image sensor 11 receives the laser beam Lm and detects the two-dimensional intensity distribution Imi(X,Y) of the laser beam Lm. Based on the two-dimensional intensity distribution Imi(X,Y) detected by the two-dimensional image sensor 11, the machine learning device 12 estimates the air fluctuation error (δXmi,δYmi) and the straightness error (2Xmti,2Ymti) of the laser beam Lm.

[0046] The straightness measurement method of this third embodiment includes the following steps (see Figure 5). A variety of two-dimensional intensity distributions Iri(X,Y) with known air fluctuation errors (δXri,δYri) and straightness errors (2Xrti,2Yrti) are used as input data, and the corresponding air fluctuation errors (δXri,δYri) and straightness errors (2Xrti,2Yrti) are used as output data. The machine learning device 12 is pre-trained using supervised learning (steps S1 to S3). Using a supervised machine learning device 12, the air fluctuation error (δXmi, δYmi) and straightness error (2Xmti, 2Ymti) are estimated for a two-dimensional intensity distribution Imi(X,Y) where the air fluctuation error (δXmi, δYmi) and straightness error (2Xmti, 2Ymti) are unknown, based on the results of supervised learning (Step S4).

[0047] Furthermore, the straightness measurement method of this third embodiment also uses a wavefront distortion application device 21 and a calculation device 22. The wavefront distortion application device 21 applies an arbitrary wavefront distortion to the laser beam Lo passing through a region 26 where no air fluctuations exist. The calculation device 22 calculates the luminance centroid (X,Y)mi of the two-dimensional intensity distribution Imi(X,Y) detected by the two-dimensional image sensor 11 for the laser beam Lm to which wavefront distortion has been applied.

[0048] Here, the air turbulence error (δXri, δYri) used in supervised learning is the luminance centroid (X,Y)mi calculated by the computing device 22. The straightness error (2Xrti, 2Yrti) used in supervised learning is added to the two-dimensional intensity distribution Imi(X,Y) by image processing that translates the two-dimensional intensity distribution Imi(X,Y).

[0049] Next, we will explain each step in Figure 5.

[0050] (Step S1) Wavefront distortion is applied to the laser beam Lo passing through the region 26 where no air fluctuations exist via the wavefront distortion application device 21, and the wavefront-distorted laser beam Lm is received by the two-dimensional image sensor 11 to acquire data of various two-dimensional intensity distributions Imi(X,Y).

[0051] (Step S2) For the various two-dimensional intensity distributions Imi(X,Y) obtained, the computing device 22 calculates each luminance centroid (X,Y)mi. These luminance centroids (X,Y)mi are the air turbulence errors (δXri,δYri) and become the output data (training data) in supervised learning. On the other hand, by shifting the two-dimensional intensity distribution Imi(X,Y) by (2Xrti,2Yrti), a straightness error (2Xrti,2Yrti) is added to the two-dimensional intensity distribution Imi(X,Y). These straightness errors (2Xrti,2Yrti) become the output data (training data) used in supervised learning. The two-dimensional intensity distribution Iri(X,Y) to which the straightness error (2Xrti,2Yrti) has been added becomes the input data used in supervised learning.

[0052] (Step S3) The various two-dimensional intensity distributions Iri(X,Y) obtained are used as input data, and the air volatility error (δXri,δYri) and straightness error (2Xrti,2Yrti) corresponding to the input data are used as output data (training data) to perform supervised learning on the machine learning device 12.

[0053] (Step S4) The laser beam Lm that has passed through the region 15 where air fluctuations exist is received by the two-dimensional image sensor 11 to detect an unknown two-dimensional intensity distribution Imi(X,Y), and the air fluctuation error (δXmi,δYmi) and straightness error (2Xrti,2Yrti) corresponding to the detected unknown two-dimensional intensity distribution Imi(X,Y) are estimated by the machine learning device 12.

[0054] Thus, step S1 is the acquisition of a two-dimensional intensity distribution, step S2 is the acquisition of output data (training data) and input data, step S3 is supervised learning, and step S4 is the measurement.

[0055] As shown in Figure 6[A], a detailed example of step S1 is as follows: First, the shape of the mirror surface 231 is changed (step S101), the laser beam Lm with wavefront distortion is received (step S102), data of the two-dimensional intensity distribution Imi(X,Y) is output (step S103), and it is determined whether the number of data points has reached a predetermined value (step S104). If the number of data points has reached the predetermined value, the process proceeds to step S2; if the number of data points has not reached the predetermined value, the process returns to step S101.

[0056] As shown in Figure 6[B], a detailed example of step S2 is as follows. First, the computing device 22 calculates each luminance centroid (X,Y)mi, i.e., the air turbulence error (δXri,δYri), for the two-dimensional intensity distribution Imi(X,Y) (step 201). Next, the straightness error (2Xrti,2Yrti) is added to the two-dimensional intensity distribution Imi(X,Y) by translating it by (2Xrti,2Yrti) using image processing (step S202). This gives us a two-dimensional intensity distribution Iri(X,Y) to which the air turbulence error (δXri,δYri) and the straightness error (2Xrti,2Yrti) have been added (step S203).

[0057] The operation and effects of the straightness measurement method of this third embodiment are the same as those of the straightness measurement devices of embodiments 1 and 2.

[0058] <Other> Although the present invention has been described above with reference to the embodiments described above, the present invention is not limited to the embodiments described above. Various modifications can be made to the structure and details of the present invention that will be understood by those skilled in the art, and such modified technologies are also included in the present invention. Furthermore, the present invention also includes combinations of some or all of the embodiments described above. [Industrial applicability]

[0059] The present invention can be used, for example, in fields such as semiconductor manufacturing and ultra-precision microfabrication, for displacement measurement technology with resolution on the order of sub-picometers. [Explanation of symbols]

[0060] 10. Straightness measuring machine 11. Two-dimensional image sensor 12 Machine Learning Devices 120 Neural Networks 121 Input Layer 122 Middle layer 123 Output Layer 13 Laser light source 14 Retroreflector 15 Regions where air fluctuations exist 20 Straightness measuring machine 21 Wavefront distortion imparting device 23 Shape-changing mirror 231 Mirror surface 232 Drive element 24 Control device 22 Computing equipment 25 Computers 26 Regions where air fluctuations do not exist 27 cabinets Lo, Lr, Lm laser beam Io(X,Y),Iri(X,Y),Imi(X,Y) Two-dimensional intensity distribution (X,Y)o,(X,Y)ri,(X,Y)mi Brightness centroid (δXri,δYri),(δXmi,δYmi) Air turbulence error (2Xrti,2Yrti),(2Xmti,2Ymti) Straightness error

Claims

1. A two-dimensional image sensor that receives a laser beam and detects the two-dimensional intensity distribution of the laser beam, A machine learning device that estimates the air fluctuation error and straightness error of the laser beam based on the two-dimensional intensity distribution detected by the two-dimensional image sensor, Equipped with, The aforementioned machine learning device is A means for performing supervised learning in advance, using various two-dimensional intensity distributions, each with known air fluctuation error and straightness error, as input data, and the corresponding air fluctuation error and straightness error as output data. A means for estimating the air fluctuation error and the straightness error in a two-dimensional intensity distribution, where the air fluctuation error and the straightness error are unknown, based on the results of supervised learning. Having, Laser beam straightness measuring machine.

2. By image processing that translates the two-dimensional intensity distribution by the magnitude of the straightness error, the two-dimensional intensity distribution after the translation to which the straightness error has been added is used as the input data for supervised learning as the two-dimensional intensity distribution with a known straightness error. The air turbulence error and the straightness error are estimated based on the results of the supervised learning described above. However, the two-dimensional intensity distribution for which the air turbulence error and the straightness error are unknown is not subjected to the image processing that involves translation. The laser beam straightness measuring device according to claim 1.

3. A wavefront distortion application device that applies arbitrary wavefront distortion to a laser beam passing through a region where air fluctuations do not exist, A computing device for calculating the luminance centroid of the two-dimensional intensity distribution detected by the two-dimensional image sensor for the laser beam to which the wavefront distortion has been applied, Furthermore, The luminance centroid calculated by the aforementioned computing device is used as the air fluctuation error, and the two-dimensional intensity distribution for the laser beam to which the wavefront distortion used in the calculation of the luminance centroid is applied is used as the input data for the supervised learning as the two-dimensional intensity distribution for which the air fluctuation error is known. The laser beam that generates the two-dimensional intensity distribution in which the air fluctuation error and the straightness error are estimated based on the results of the supervised learning described above, and in which the air fluctuation error and the straightness error are unknown, is not subjected to wavefront distortion by the wavefront distortion application device. A laser beam straightness measuring device according to claim 1 or 2.

4. The wavefront distortion device is A shape-changing mirror having a mirror surface that reflects the laser beam and a driving element that deforms the shape of the mirror surface, A control device that controls the drive element using a Zernike approximation polynomial, including, The laser beam straightness measuring device according to claim 3.

5. A two-dimensional image sensor that receives a laser beam and detects the two-dimensional intensity distribution of the laser beam, A machine learning device that estimates the air fluctuation error and straightness error of the laser beam based on the two-dimensional intensity distribution detected by the two-dimensional image sensor, Using, The machine learning device is pre-trained using various two-dimensional intensity distributions, each with known air turbulence error and straightness error, as input data, and the corresponding air turbulence error and straightness error as output data. Using the supervised machine learning device described above, the air fluctuation error and the straightness error are estimated for the two-dimensional intensity distribution in which the air fluctuation error and the straightness error are unknown. Method for measuring the straightness of a laser beam.

6. By image processing that translates the two-dimensional intensity distribution by the magnitude of the straightness error, the two-dimensional intensity distribution after the translation to which the straightness error has been added is used as the input data for supervised learning as the two-dimensional intensity distribution with a known straightness error. The air turbulence error and the straightness error are estimated based on the results of the supervised learning described above. However, the two-dimensional intensity distribution for which the air turbulence error and the straightness error are unknown is not subjected to the image processing that involves translation. The method for measuring the straightness of a laser beam according to claim 5.

7. A wavefront distortion application device that applies arbitrary wavefront distortion to a laser beam passing through a region where air fluctuations do not exist, A computing device for calculating the luminance centroid of the two-dimensional intensity distribution detected by the two-dimensional image sensor for the laser beam to which the wavefront distortion has been applied, Furthermore, The luminance centroid calculated by the aforementioned computing device is used as the air fluctuation error, and the two-dimensional intensity distribution for the laser beam to which the wavefront distortion used in the calculation of the luminance centroid is applied is used as the input data for the supervised learning as the two-dimensional intensity distribution for which the air fluctuation error is known. The laser beam that generates the two-dimensional intensity distribution in which the air fluctuation error and the straightness error are estimated based on the results of the supervised learning described above, and in which the air fluctuation error and the straightness error are unknown, is not subjected to wavefront distortion by the wavefront distortion application device. The method for measuring the straightness of a laser beam according to claim 5 or 6.

Images