Method for generating digital mask correction model for DLP projector for 3D printing, and digital mask correction method and digital mask correction apparatus using same
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-01-26
- Publication Date
- 2026-08-13
Smart Images
Figure KR2026001478_13082026_PF_FP_ABST
Abstract
Description
Method for generating a digital mask correction model for a DLP projector for 3D printing, and a digital mask correction method and digital mask correction device using the same
[0001] The present invention relates to a method for generating a digital mask correction model for 3D printing using DLP projection, a digital mask correction method using the same, and a digital mask correction device.
[0002] 3D printing is an innovative manufacturing method that produces three-dimensional shapes by layering materials. It is a technology expected to have a significant ripple effect, generating high added value and creating new jobs across various industrial sectors such as robotics, automotive, aerospace, defense, and healthcare. Because this technology builds complex 3D shapes layer by layer based on digital models, it enables the easy production of intricate structures that are difficult to realize using traditional manufacturing methods. While initially used primarily for prototyping, advancements in materials and additive manufacturing technologies have confirmed its utility and potential for creating future value, leading to a rapid global expansion of its application fields and market size. In particular, it is attracting attention as a core technology in diverse future industries—including semiconductor processes, secondary batteries, hydrogen energy, electric vehicles, urban unmanned aerial vehicles, the biomedical industry, and construction—and active utilization is being sought.
[0003] However, the precision of current 3D printing outputs is lower than that of conventional subtractive machining technology, which limits its industrial application. Accordingly, various technical approaches are required to improve precision.
[0004] Digital Light Processing (DLP) printing is a notable method among these 3D printing technologies that fabricates three-dimensional shapes by projecting light in area units to induce a photocuring reaction. This technique is receiving significant interest from academia and industry due to its advantages, such as fast process speed, minimal constraints on processing shapes, and the ability to use various materials. DLP printing uses images cut along the layer direction based on a digital model as a digital mask, and projects the shape through this digital mask to form an image on the focal plane.
[0005] However, in conventional DLP printing, there is a problem where the distribution of light intensity irradiated at the focal plane differs from that of the digital mask due to the characteristics of the optical system. To solve this, methods such as changing the alignment of the optical system or using high-performance optical lenses have been attempted, but these have the disadvantage of significantly increasing the system configuration time and cost.
[0006] Recently, various approaches have been attempted to overcome these problems. For example, there have been attempts to improve precision by predicting the output shape through simulation; however, simulations have limitations in that they do not adequately reflect the complexity of the actual environment because they model the printing system in a simplified manner. As another approach, research has been conducted to correct digital masks by learning the correlation between the actual output shape and the digital mask through machine learning. However, this method also faces the difficulty of securing sufficient data because the process of collecting and measuring the actual output shape requires a significant amount of time and effort. To resolve these technical limitations, there is a need for the development of more effective and general digital mask correction methods and devices capable of implementing them.
[0007] The problem that the present invention aims to solve is to provide a method for generating a digital mask correction model for a DLP projector that can improve the precision of 3D printing.
[0008] Another problem that the present invention aims to solve is to provide a digital mask correction method using such a digital mask correction model.
[0009] Another problem that the present invention aims to solve is to provide a digital mask correction device using such a digital mask correction model.
[0010] The problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned problems will be clearly understood by those skilled in the art to which the present invention belongs from the description below.
[0011] A method for generating a digital mask correction model according to an embodiment of the present invention for achieving the above objective is a method performed by a computing device for generating a digital mask correction model for correcting a digital mask of a DLP projector for 3D printing, comprising: preparing a light distribution image generated from light projected from the DLP projector; converting the light distribution image to generate a threshold profile image composed of a binary level image; training a first image conversion model using the threshold profile image as an input variable and the light distribution image as an output variable; and training a second image conversion model using the light distribution image as an input variable and the digital mask as an output variable.
[0012] The above light distribution image can be composed of an 8-bit grayscale image.
[0013] The above binary level image can be generated by converting pixels having a grayscale value greater than or equal to a predetermined threshold (T) from the image to be converted to have a grayscale value of the threshold (T), and converting pixels having a grayscale value less than the threshold (T) to have a grayscale value of 0.
[0014] By controlling the threshold value (T) according to the light energy projected from the DLP projector, the threshold profile image can match the shape of the print result produced by the DLP projector.
[0015] The first image transformation model and the second image transformation model may be models based on a conditional generative adversarial network (cGAN).
[0016] The step of generating the threshold profile image may include: generating a degree of curation image by applying a Gaussian convolution operation to the light distribution image to simulate optical scattering and chemical diffusion occurring within the photocurable resin; and generating the threshold profile image by converting pixels having a grayscale value greater than or equal to a predetermined threshold value (T) from the degree of curation image to have a grayscale value of the threshold value (T), and converting pixels having a grayscale value less than the threshold value (T) to have a grayscale value of 0.
[0017] The above first image transformation model may include a first-1 image transformation model that uses the threshold profile image as an input variable and the hardening degree image as an output variable; and a first-2 image transformation model that uses the hardening degree image as an input variable and the light distribution image as an output variable.
[0018] A digital mask correction method according to an embodiment of the present invention for achieving the above other objectives is a method performed by a computing device that corrects a digital mask of a DLP projector for 3D printing using a digital mask correction model, comprising: a step of preparing a light distribution image generated from light projected from the DLP projector; a step of converting the light distribution image to generate a threshold profile image composed of a binary level image; a step of training a first image conversion model using the threshold profile image as an input variable and the light distribution image as an output variable; a step of training a second image conversion model using the light distribution image as an input variable and the digital mask as an output variable; a step of inputting an original digital mask corresponding to a desired print result into the first image conversion model; and a step of inputting the output of the first image conversion model into the second image conversion model to output a modified digital mask.
[0019] A digital mask correction device according to an embodiment of the present invention for achieving the above-mentioned additional objective comprises: a DLP projector for 3D printing; an image sensor that detects light projected from the DLP projector and generates a light distribution image; and a computing device that generates a digital mask correction model for correcting the digital mask of the DLP projector using the light distribution image. Herein, the computing device converts the light distribution image to generate a threshold profile image consisting of a binary level image; trains a first image conversion model using the threshold profile image as an input variable and the light distribution image as an output variable; and trains a second image conversion model using the light distribution image as an input variable and the digital mask as an output variable.
[0020] The computing device can input an original digital mask corresponding to a desired print result into the first image conversion model; and can input the output of the first image conversion model into the second image conversion model to output a modified digital mask.
[0021] The image sensor can be placed on the focal plane of the DLP projector.
[0022] It may further include a neutral density filter disposed between the DLP projector and the image sensor to adjust the intensity of the light so that the intensity of the light projected from the DLP projector is included within the measurement range of the image sensor.
[0023] Specific details of other embodiments are included in the specific contents and drawings.
[0024] As described above, the method for generating a digital mask correction model for a 3D printing DLP projector according to the present invention, and the digital mask correction method and digital mask correction device using the same, have the following excellent effects.
[0025] The present invention overcomes the limitations of conventional technology, such as the lack of data and the failure to systematically analyze the intermediate stages of the printing process leading from the arrival of the light focal plane to the curing of the photocurable resin, by efficiently applying machine learning or deep learning techniques to improve the precision of 3D printing. Existing technology has focused only on the relationship between the digital mask and the printed output, and there have been few attempts to systematically analyze the optical changes occurring during the intermediate stages of the printing process or to utilize them as data.
[0026] The device proposed in this invention aims to fundamentally solve the problem of reduced precision in the final printing result due to optical characteristics by focusing on the relationship between the digital mask and the light distribution in the focal plane. In particular, this invention dramatically improves the data collection limitations faced by existing machine learning-based methods. By constructing a dataset that combines digital data, such as a digital mask, with light distribution data measured in the focal plane, the data collection process can be performed quickly and in large quantities. This maximizes the learning efficiency of machine learning or deep learning models and enables high performance to be expected even under various conditions.
[0027] In addition, the present invention has the universality of being applicable to all types of DLP printers using digital image generation devices and optical devices. Through this, optical distortion or deformation can be minimized to significantly improve the precision of printing results, and furthermore, it is expected that the industrial applicability of 3D printing technology can be further expanded.
[0028] FIG. 1 is a schematic diagram showing a digital mask correction device for 3D printing according to one embodiment of the present invention.
[0029] FIG. 2 is a diagram illustrating an exemplary digital mask correction device for 3D printing of FIG. 1.
[0030] Figure 3 is a schematic diagram showing the configuration of the computing device of Figure 1.
[0031] FIG. 4 is a flowchart sequentially illustrating a method for generating a digital mask correction model for 3D printing according to an embodiment of the present invention.
[0032] Figure 5 is a diagram conceptually illustrating the process of generating a critical profile image in Figure 4.
[0033] Figure 6 is a diagram conceptually explaining the process of training the first image conversion model and the second image conversion model in Figure 4.
[0034] FIG. 7 is a graph illustrating the similarity between a critical profile image and a print result according to one embodiment of the present invention.
[0035] FIG. 8 is a diagram showing an experimental example demonstrating the similarity between a critical profile image and a print result according to one embodiment of the present invention.
[0036] FIG. 9 is a flowchart sequentially illustrating a digital mask correction method for 3D printing according to one embodiment of the present invention.
[0037] FIG. 10 is a diagram comparing light distribution images before and after digital mask correction using a digital mask correction device according to one embodiment of the present invention.
[0038] FIG. 11 is a drawing comparing print results before and after digital mask correction using a digital mask correction device according to an embodiment of the present invention.
[0039] FIG. 12 is a flowchart sequentially illustrating a method for generating a digital mask correction model for 3D printing according to another embodiment of the present invention.
[0040] FIG. 13 is a diagram conceptually illustrating the process of training the first-1 image conversion model, the first-2 image conversion model, and the second image conversion model in FIG. 12.
[0041] FIG. 14 is a flowchart sequentially illustrating a digital mask correction method for 3D printing according to another embodiment of the present invention.
[0042] FIG. 15 is a drawing comparing print results before and after digital mask correction according to another embodiment of the present invention.
[0043] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.
[0044]
[0045] A digital mask correction device for 3D printing according to an embodiment of the present invention will be described in detail below with reference to FIGS. 1 to 3. FIG. 1 is a schematic diagram showing a digital mask correction device for 3D printing according to an embodiment of the present invention. FIG. 2 is a diagram exemplifying the digital mask correction device for 3D printing of FIG. 1. FIG. 3 is a schematic diagram showing a computing device of FIG. 1. In the illustrated embodiment, each component may have different functions and capabilities in addition to those described below, and may include additional components in addition to those described below.
[0046] The digital mask correction device (100) includes a DLP projector (20) for 3D printing, a neutral density filter (30), an image sensor (40), a cooling fan (50), and a computing device (10).
[0047] A DLP projector (20) is a device used for 3D printing that uses a Digital Micromirror Device (DMD) to manipulate light and project a 2D image. The DLP projector (20) includes a light source, a DMD, and an optical lens. The light source provides light to cure photocurable resin in 3D printing and projects light onto the DMD. The light source can be composed of, for example, an LED, a laser, a lamp, etc. The DMD consists of millions of micromirrors that precisely manipulate the light projected from the light source. Individual micromirrors are electronically controlled and switch between an on state and an off state. When the micromirrors are on, they reflect light in a specific direction, and when they are off, they reflect light in a different direction to create a desired image overall. A digital mask is digital data that defines an image or pattern to be projected by a DLP projector (20), and represents cross-sectional image data generated by slicing 3D modeling data in stacking thickness units. The DLP projector (20) controls the on or off state of the micro-mirror based on the digital mask. An optical lens projects an image onto a focal plane by manipulating light reflected in a specific direction from the DMD.
[0048] A neutral density filter (30), also known as an ND (Neutral Density) filter, is placed between a DLP projector (20) and an image sensor (40) to adjust the intensity of light so that the intensity of light projected from the DLP projector (20) is included within the measurement range of the image sensor (40). The neutral density filter (30) is attached in front of the optical lens of the DLP projector (20) to uniformly reduce the intensity of light. The neutral density filter (30) can reduce all wavelengths at the same rate without filtering specific wavelengths of light so that color distortion does not occur.
[0049] The image sensor (40) detects light projected from the DLP projector (20) and converts it into an electrical signal to generate a light distribution image. Specifically, each pixel constituting the image sensor (40) has a transistor and an amplifier built into it individually, so the image sensor (40) generates an electric charge after detecting light and converts it into an electrical signal to generate a digitized original image. The image sensor (40) converts the original image into an 8-bit grayscale image to generate a light distribution image. However, the present invention is not limited thereto, and the computing device (10) may generate a light distribution image using the original image. It is preferable that the image sensor (40) be placed on the focal plane of the DLP projector (20). For example, the image sensor (40) may be made of a CMOS image sensor.
[0050] A cooling fan (50) is placed at the bottom of the image sensor (40) to prevent the image sensor (40) from overheating and degrading in quality, and circulates air around the sensor module or device to disperse heat.
[0051] The computing device (10) is an artificial intelligence computing device that generates a digital mask correction model for correcting the digital mask of a DLP projector (20) using an original image or a light distribution image received from an image sensor (40). The computing device (10) includes at least one processor (2), memory (4), a bus (1), an input / output unit (6), and a communication unit (8). The computing device (10) may be a computing device programmed in software or a computing device implemented in hardware. Below, a computing device (10) implemented in software is described as an example.
[0052] The processor (2) prepares a light distribution image generated from light projected from a DLP projector, and to do so, it may execute one or more programs (5) stored in memory (4). Here, the processor (2) may receive a light distribution image from an image sensor (40) or receive an original image from an image sensor (40) and convert it into an 8-bit grayscale image to generate a light distribution image. Additionally, the processor (2) may convert the light distribution image to generate a threshold profile image consisting of a binary level image, and to do so, it may execute one or more programs (5) stored in memory (4). Additionally, the processor (2) may use the threshold profile image as an input variable and the light distribution image as an output variable to train a first image conversion model, and to do so, it may execute one or more programs (5) stored in memory (4). Additionally, the processor (2) may use the light distribution image as an input variable and a digital mask as an output variable to train a second image conversion model, and to do so, it may execute one or more programs (5) stored in memory (4). Additionally, the processor (2) inputs an original digital mask corresponding to a desired print result into a first image conversion model using an input variable corresponding to the threshold profile image, and for this purpose, it may execute one or more programs (5) stored in memory (4). Additionally, the processor (2) inputs an output variable of the first image conversion model into the second image conversion model to output a modified digital mask, and for this purpose, it may execute one or more programs (5) stored in memory (4). The one or more programs (5) may include one or more computer-executable instructions, and the computer-executable instructions may be configured so that when executed by the processor (2), the computing device (10) performs operations according to an embodiment of the present invention.
[0053] Memory (4) stores computer-executable instructions, data, and / or other suitable forms of information. A program (5) stored in memory (4) includes a set of instructions executable by a processor (2). For example, memory (4) may be volatile memory such as random access memory, non-volatile memory, one or more magnetic disk storage devices, optical disk storage devices, flash memory devices, other forms of storage media that can be accessed by a computing device (10) and store desired information, or a suitable combination thereof.
[0054] The bus (1) interconnects various components of the computing device (10), including the processor (2) and memory (4).
[0055] Examples of various components include an input / output unit (6) and a communication unit (8), which are connected to the bus (1). For example, the input / output unit (6) may include a pointing device (such as a mouse or trackpad), a keyboard, a touch input device (such as a touchpad or touchscreen), a voice or sound input device, various types of sensor devices, a camera, a display device, a printer, a speaker, a network card, etc. The communication unit (8) connects the computing device (10) to an external device via a wired or wireless network. The input / output unit (6) and / or the communication unit (8) may be included inside the computing device (10) as a component constituting the computing device (10) as in the present embodiment, or they may be connected to the computing device (10) as a separate device distinct from the computing device (10).
[0056] The present invention can generate tens of thousands of pairs of digital masks, light distribution images, and threshold profile images within minutes using a DLP projector (20) and an image sensor (40) and use them as training data sets for a digital mask correction model.
[0057]
[0058] Hereinafter, with reference to FIGS. 4 to 8, a method for generating a digital mask correction model for correcting a digital mask of a DLP projector for 3D printing according to an embodiment of the present invention will be described in detail. FIG. 4 is a flowchart sequentially illustrating a method for generating a digital mask correction model for 3D printing according to an embodiment of the present invention. FIG. 5 is a diagram conceptually illustrating the process of generating a threshold profile image in FIG. 4. FIG. 6 is a diagram conceptually illustrating the process of training a first image conversion model and a second image conversion model in FIG. 4. FIG. 7 is a graph illustrating the similarity between a threshold profile image and a print result according to an embodiment of the present invention. FIG. 8 is a diagram showing an experimental example demonstrating the similarity between a threshold profile image and a print result according to an embodiment of the present invention.
[0059] First, the computing device (10) prepares a light distribution image (120) generated from light projected from a DLP projector (20) and passed through a neutral density filter (30) (S10). Specifically, an image sensor (40) generates a light distribution image (120) from light projected from a DLP projector (20), and the computing device (10) can receive and prepare the light distribution image (120) from the image sensor (40). The light distribution image (120) is a black and white image in which brightness information for each pixel is expressed in 8 bits, and is composed of brightness (intensity) without color information (RGB), and the grayscale value of each pixel has a range from 0 to 255. For example, a pixel with a grayscale value of 0 is represented as black, and a pixel with a grayscale value of 255 is represented as white.
[0060] The computing device (10) converts the light distribution image (120) to generate a threshold profile image (130) composed of a binary level image (S12). Specifically, the computing device (10) generates the threshold profile image (130) by representing pixels having a grayscale value greater than or equal to a predetermined threshold value (T) in the light distribution image (120) composed of an 8-bit grayscale image with a brightness corresponding to the grayscale value of the threshold value (T), and representing pixels having a grayscale value less than the threshold value (T) with a brightness corresponding to the grayscale value of 0. That is, the binary level image can be generated by converting pixels having a grayscale value greater than or equal to a predetermined threshold value (T) from the image to be converted (e.g., light distribution image) to have a grayscale value of the threshold value (T), and converting pixels having a grayscale value less than the threshold value (T) to have a grayscale value of 0. For example, each pixel constituting a binary level image may be represented by either of two grayscale values: a first value (e.g., 0) representing the background and a second value (e.g., T) representing the active area. Pixels represented by brightness corresponding to the grayscale value of the threshold (T) are areas where the photocurable resin is cured during subsequent 3D printing, and pixels represented by brightness corresponding to the grayscale value of 0 are areas where the photocurable resin is not cured during subsequent 3D printing. For example, if the threshold (T) is 128, and the grayscale value of a pixel constituting the light distribution image (120) is 200, then that pixel is represented by brightness corresponding to the grayscale value of 128 in the threshold profile image (130). If the grayscale value of a pixel constituting the light distribution image (120) is 50, then that pixel is represented by brightness corresponding to the grayscale value of 0 in the threshold profile image (130).
[0061] Here, the critical profile image (130) is the shape of the cured product or printed result expected to be produced after the photocuring of the photocurable resin. Referring to the top of the drawing in FIG. 7, the energy distribution curve of the irradiated light may have a distribution that is strongest at the center and weaker towards the periphery, for example, a Gaussian distribution. In the light distribution image (120), the energy (E) corresponding to the grayscale value of the critical value (T) T Regions receiving energy greater than )(hereinafter referred to as 'critical profile energy') are converted into regions having a grayscale value of the threshold (T) in the critical profile image (130). Additionally, in the light distribution image (120), the critical profile energy (E T Regions receiving less than ) are converted into regions having a grayscale value of 0 in the critical profile image (130). Referring to the bottom of FIG. 7, the photocurable resin is cured by receiving more than a specific critical curing energy (Ec). The width at which the resin is cured at the resin surface (i.e., z = 0) is such that the energy distribution curve [is at] the critical profile energy (E T It matches the width of the point where it meets ). That is, the grayscale threshold (T) or the critical profile energy (E T By appropriately controlling ), the critical profile image (130) can be matched with the cross-sectional shape of the printed result or cured material.
[0062] Generally, light energy (E) for light intensity (I) and irradiation time (t) is calculated by the following mathematical formula 1.
[0063]
[0064] In 3D printing, as light intensity (I) or time (t) increases, the region of the photocurable resin that receives energy greater than the critical curing energy (Ec) expands. To simulate this phenomenon in the critical profile image (130), a grayscale threshold (T) or critical profile energy (E T ) must decrease. Conversely, if the light intensity (I) or time (t) decreases, the region in the photocurable resin that receives energy greater than the critical curing energy (Ec) narrows. To simulate this phenomenon in the critical profile image (130), the grayscale threshold (T) or the critical profile energy (E T ) must increase. Therefore, to match the critical profile image (130) with the print result, the relationship of the following mathematical formula 2 can be established.
[0065]
[0066] That is, critical profile energy (E T An inverse relationship can be established between ) and the actual projected light energy (I·t). Here, the critical profile energy (E T ) can be expressed as a function of the grayscale threshold (T). For example, the critical profile energy (E T ) can be represented as a linear function of the grayscale threshold (T) or as an exponential non-linear function. Consequently, by controlling the grayscale threshold (T) according to the light energy projected from the DLP projector (20), the threshold profile image (130) can match the shape of the print result produced by the DLP projector (20).
[0067] Figure 8 shows the change in printed results according to the light intensity of the DLP projector, the change in the threshold profile image according to the grayscale threshold (T), and the measurement of the similarity between the printed results and the threshold profile image through a merged profile. In each experimental example, the product of the light intensity and the threshold (T) was set to be constant. That is, the light intensity was set to 100 for Experimental Example 1 and 30 for Experimental Example 2. The threshold (T) was set to 15 for Experimental Example 1 and 50 for Experimental Example 2. Therefore, the product of the light intensity and the threshold (T) is the same at 1500 for both experimental examples. The Dice coefficient, which indicates the similarity between the printed results and the threshold profile image, was found to be 0.9873 for Experimental Example 1 and 0.9935 for Experimental Example 2. Therefore, it was confirmed that in both experimental examples, the threshold profile image accurately replicates the actual printed results by more than 99%.
[0068] In this way, a predetermined threshold value (T) can be selected so that the threshold profile image (130) matches the shape of the cured material or the printed result. Therefore, in this embodiment, even without physical output, the actual printed result can be accurately simulated in the threshold profile image (130) by controlling the threshold value (T), and the digital mask correction model learned using the threshold profile image (130) can produce a modified digital mask that generates the shape of the desired printed result.
[0069] Meanwhile, the threshold value (T) may be predefined according to the DLP projector (20). Alternatively, to generate an optimal digital mask correction model, the computing device (10) may arbitrarily pre-set a plurality of threshold values and generate a digital mask correction model for each threshold value, and then compare the 3D print results according to the plurality of digital mask correction models to select the optimal threshold value and the corresponding digital mask correction model.
[0070] A computing device (10) trains a first image transformation model (M1) using a threshold profile image (130) as an input variable and a light distribution image (120) as an output variable (S14). The first image transformation model (M1) may be a machine learning model or a deep learning model optimized for image data processing and matching. For example, the first image transformation model (M1) may be a model based on a conditional generative adversarial network (cGAN). Preferably, the first image transformation model (M1) may be a pix2pix model. Given a dataset consisting of pairs of input variables (i.e., threshold profile image) and output variables (i.e., light distribution image), the first image transformation model (M1) includes a generator that generates an image similar to the output variable (referred to as the 'generated image') based on the input variable, and a discriminator that determines how similar the generated image is to the output variable.
[0071] A computing device (10) trains a second image transformation model (M2) using a light distribution image (120) as an input variable and a digital mask (110) as an output variable (S16). The second image transformation model (M2) may be a machine learning model or a deep learning model optimized for image data processing and matching. For example, the second image transformation model (M2) may be a model based on a conditional generative adversarial network (cGAN). Preferably, the second image transformation model (M2) may be a pix2pix model. Given a dataset consisting of pairs of input variables (i.e., light distribution image) and output variables (i.e., digital mask), the second image transformation model (M2) includes a generator that generates an image similar to the output variable (referred to as a 'generated image') based on the input variable, and a discriminator that determines how similar the generated image is to the output variable.
[0072] In this way, the computing device (10) independently trains the first image conversion model (M1) and the second image conversion model (M2) to generate a digital mask correction model (ML) in which the first image conversion model (M1) and the second image conversion model (M2) are sequentially combined. Here, the output of the first image conversion model (M1) is used as the input of the second image conversion model (M2).
[0073]
[0074] Hereinafter, with reference to FIGS. 6 and FIGS. 9, a method for correcting a digital mask of a DLP projector for 3D printing using a digital mask correction model according to an embodiment of the present invention will be described in detail. FIGS. 9 is a flowchart sequentially illustrating a method for correcting a digital mask for 3D printing according to an embodiment of the present invention.
[0075] After the training of the digital mask correction model (ML) is completed, the computing device (10) inputs the original digital mask corresponding to the desired print result into the first image transformation model (M1) as an input variable corresponding to the threshold profile image (S20). The original digital mask may be composed of the same shape as the desired print result. Since the original digital mask represents the shape of the desired print result, it can be considered as input data with the same attributes as the threshold profile image in the training stage. The original digital mask may be composed of a binary level image.
[0076] The computing device (10) inputs the output of the first image conversion model (M1) into the second image conversion model (M2) and outputs a modified digital mask (S22, S24).
[0077]
[0078] Hereinafter, with reference to FIGS. 10 and 11, a light distribution image and a 3D print result obtained using a digital mask correction device according to an embodiment of the present invention will be described. FIG. 10 is a drawing comparing light distribution images before and after digital mask correction using a digital mask correction device according to an embodiment of the present invention. FIG. 11 is a drawing comparing print results before and after digital mask correction using a digital mask correction device according to an embodiment of the present invention.
[0079] As shown in FIG. 10, when the DLP projector (20) projects light using the original digital mask (112) of the triangle pattern, it can be seen that the light distribution image (122) differs somewhat from the original digital mask (112). On the other hand, when the original digital mask (112) is corrected into a modified digital mask (114) using the digital mask correction model (ML) of the present invention and then the DLP projector (20) projects light using this, it can be seen that the light distribution image (124) is almost identical to the original digital mask (112).
[0080] As shown in FIG. 11, it can be seen that the print result (146) obtained by 3D printing with the original digital mask (116) of the de-pattern is somewhat different from the original digital mask (116). On the other hand, when the original digital mask (116) is corrected into a modified digital mask (118) using the digital mask correction model (ML) of the present invention and then 3D printed using it, it can be seen that the print result (148) is almost identical to the original digital mask (116).
[0081]
[0082] Hereinafter, a method for generating a digital mask correction model for correcting a digital mask of a DLP projector for 3D printing according to another embodiment of the present invention will be described in detail with reference to FIGS. 12 and 13. FIG. 12 is a flowchart sequentially illustrating a method for generating a digital mask correction model for 3D printing according to another embodiment of the present invention. FIG. 13 is a diagram conceptually illustrating the process of training the first-1 image conversion model, the first-2 image conversion model, and the second image conversion model in FIG. 12. For convenience of explanation, components having the same function as each component shown in the drawings of the previous embodiment ( FIGS. 1 to 11) are indicated by the same reference numerals, and their descriptions are omitted; the following description will focus on the differences.
[0083] In the same way as in the previous embodiment, the computing device (10) prepares a light distribution image (120) (S10).
[0084] Next, the computing device (10) converts the light distribution image (120) to generate a threshold profile image (130) composed of binary level images (S12). Specifically, the computing device (10) generates a degree of curation image (125) by applying a Gaussian convolution operation to the light distribution image to simulate optical scattering and chemical diffusion occurring within the photocurable resin. Next, the computing device (10) generates a threshold profile image (130) by converting pixels from the degree of curation image (125) that have grayscale values greater than or equal to a predetermined threshold value (T) to have grayscale values of the threshold value (T), and converting pixels that have grayscale values less than the threshold value (T) to have grayscale values of 0.
[0085] Generally, the DLP printing process is divided into an <optical stage> in which a digital mask (110) is projected through an optical system, and a <physical-chemical stage> in which the projected light penetrates into the resin and causes curing. In this embodiment, to improve the physical alignment between the critical profile image (130) and the actual print result, a curing degree image (125) reflecting the scattering of light and the diffusion of reactive substances occurring within the resin is introduced as an intermediate parameter. This curing degree image (125) can bridge the gap between the light distribution image (120), which is the measurement data of the optical system, and the actual print result.
[0086] The curing degree image (125) has a blurred energy distribution characteristic in which energy spreads outward compared to the light distribution image (120) due to the combined action of optical scattering and chemical diffusion. The curing degree image (125) is data that simulates physical-chemical diffusion phenomena using an image processing technique based on the light distribution image (120). Specifically, to mathematically simulate the energy redistribution caused by the optical scattering and chemical diffusion, a convolution operation using, for example, a Gaussian filter may be used. The Gaussian function has a bell-shaped distribution in which the value decreases smoothly from the center to the periphery, which is very similar to the physical profile of energy incident at a specific point within the resin spreading outward. Accordingly, by convolving a Gaussian kernel onto the measured light distribution image (120), a curing degree image (125) reflecting physical diffusion can be produced. At this time, the standard deviation of the Gaussian filter is a parameter that determines the degree of diffusion. In this embodiment, the optimal standard deviation value can be determined through experimental or iterative matching based on the similarity between the critical profile image (130) converted from the curing degree image (125) and the actual print result. However, the present invention is not limited to the above-mentioned Gaussian filter, and various diffusion functions that consider the physical properties of the resin (viscosity, photoinitiator reactivity, etc.) may also be applied.
[0087] Next, the computing device (10) trains a first image transformation model using a threshold profile image (130) as an input variable and a light distribution image (120) as an output variable. Here, the first image transformation model includes a first-1 image transformation model (M1-1) using a threshold profile image (130) as an input variable and a hardening image (125) as an output variable; and a first-2 image transformation model (M1-2) using a hardening image (125) as an input variable and a light distribution image (120) as an output variable. Accordingly, the computing device (10) trains the first-1 image transformation model (M1-1) using a threshold profile image (130) as an input variable and a hardening image (125) as an output variable (S14-1). In addition, the computing device (10) trains a first-second image conversion model (M1-2) using a hardness image (125) as an input variable and a light distribution image (120) as an output variable (S14-2).
[0088] The first-1 image transformation model (M1-1) and / or the first-2 image transformation model (M1-2) may be a machine learning model or a deep learning model optimized for image data processing and matching. For example, the first-1 image transformation model (M1-1) and / or the first-2 image transformation model (M1-2) may be a model based on a Conditional Generative Adversarial Network (cGAN). Preferably, the first-1 image transformation model (M1-1) and / or the first-2 image transformation model (M1-2) may be a pix2pix model.
[0089] Next, the computing device (10) uses a light distribution image (120) as an input variable and a digital mask (110) as an output variable to train a second image transformation model (M2) (S16).
[0090] In this way, the computing device (10) independently trains the first-1 image conversion model (M1-1), the first-2 image conversion model (M1-2), and the second image conversion model (M2) to generate a digital mask correction model (ML) in which the first-1 image conversion model (M1-1), the first-2 image conversion model (M1-2), and the second image conversion model (M2) are sequentially combined. Here, the output of the first-1 image conversion model (M1-1) is used as the input of the first-2 image conversion model (M1-2), and the output of the first-2 image conversion model (M1-2) is used as the input of the second image conversion model (M2).
[0091]
[0092] Hereinafter, with reference to FIG. 14, a method for correcting a digital mask of a DLP projector for 3D printing using a digital mask correction model according to another embodiment of the present invention will be described in detail. FIG. 14 is a flowchart sequentially illustrating a method for correcting a digital mask for 3D printing according to another embodiment of the present invention.
[0093] After the training of the digital mask correction model (ML) according to FIGS. 12 and 13 is completed, the computing device (10) inputs the original digital mask corresponding to the desired print result into the first-1 image transformation model (M1-1) as an input variable corresponding to the threshold profile image (S30). The original digital mask may be formed with the same shape as the desired print result. Since the original digital mask represents the shape of the desired print result, it can be considered as input data with the same attributes as the threshold profile image in the training stage. The original digital mask may be formed as a binary level image.
[0094] Next, the computing device (10) inputs the output of the first-1 image conversion model (M1-1) to the first-2 image conversion model (M1-2) (S32).
[0095] The computing device (10) inputs the output of the first-second image conversion model (M1-2) into the second image conversion model (M2) and outputs a modified digital mask (S34, S36).
[0096]
[0097] Referring to FIG. 15, a 3D printed result obtained using a digital mask correction method according to another embodiment of the present invention will be described. FIG. 15 is a drawing comparing printed results before and after digital mask correction according to another embodiment of the present invention.
[0098] As illustrated in FIG. 15, it can be seen that the print result (147) obtained by 3D printing with the original digital mask (117) of the bird shape pattern differs somewhat from the original digital mask (117). On the other hand, when the original digital mask (117) is corrected into a modified digital mask (119) using the digital mask correction model (ML) of the present invention, added pixels and subtracted pixels are generated compared to the original digital mask (117). When 3D printing is performed using this modified digital mask (119), it can be seen that the print result (149) is almost identical to the original digital mask (117).
[0099]
[0100] The aforementioned systems, functional parts, modules, etc., may be computer programs and may consist of multiple program modules implemented by one or more processors, for example, stored or loaded on a computer-readable storage medium. Additionally, each step mentioned in the flowchart may consist of a computer program, a program module, or a computer-executable instruction.
[0101] Although embodiments of the present invention have been described above with reference to the attached drawings, those skilled in the art will understand that the present invention may be implemented in other specific forms without changing its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive.
Claims
1. A method performed by a computing device for generating a digital mask correction model for correcting a digital mask of a DLP projector for 3D printing, A step of preparing a light distribution image generated from light projected from the above DLP projector; A step of converting the light distribution image to generate a critical profile image composed of a binary level image; A step of training a first image transformation model using the above threshold profile image as an input variable and the above light distribution image as an output variable; and A method for generating a digital mask correction model, comprising the step of training a second image transformation model using the light distribution image as an input variable and the digital mask as an output variable.
2. In Paragraph 1, A method for generating a digital mask correction model, characterized in that the light distribution image above is composed of an 8-bit grayscale image.
3. In Paragraph 1, A method for generating a digital mask correction model, characterized in that the above binary level image is generated by converting pixels having a grayscale value greater than or equal to a predetermined threshold (T) from a target image to have a grayscale value of the threshold (T), and converting pixels having a grayscale value less than the threshold (T) to have a grayscale value of 0.
4. In Paragraph 3, A method for generating a digital mask correction model, characterized in that the threshold value (T) is controlled according to the light energy projected from the DLP projector, so that the threshold profile image matches the shape of the print result produced by the DLP projector.
5. In Paragraph 1, A method for generating a digital mask correction model, characterized in that the first image transformation model and the second image transformation model are models based on a conditional generative adversarial network (cGAN).
6. In paragraph 1, the step of generating the threshold profile image is, A step of generating a Degree of Cure image by applying a Gaussian convolution operation to the light distribution image to simulate optical scattering and chemical diffusion occurring within the photocurable resin; and A method for generating a digital mask correction model, comprising the step of converting pixels having a grayscale value greater than or equal to a predetermined threshold value (T) from the hardening degree image to have a grayscale value of the threshold value (T), and converting pixels having a grayscale value less than the threshold value (T) to have a grayscale value of 0 to generate the threshold profile image.
7. In Paragraph 6, A method for generating a digital mask correction model, wherein the first image conversion model comprises: a first-1 image conversion model that uses the threshold profile image as an input variable and the hardening degree image as an output variable; and a first-2 image conversion model that uses the hardening degree image as an input variable and the light distribution image as an output variable.
8. A method performed by a computing device that corrects the digital mask of a DLP projector for 3D printing using a digital mask correction model, wherein A step of preparing a light distribution image generated from light projected from the above DLP projector; A step of converting the light distribution image to generate a critical profile image composed of a binary level image; A step of training a first image transformation model using the above threshold profile image as an input variable and the above light distribution image as an output variable; A step of training a second image transformation model using the light distribution image as an input variable and the digital mask as an output variable; A step of inputting an original digital mask corresponding to a desired print result into the first image conversion model; and A digital mask correction method comprising the step of inputting the output of the first image conversion model into the second image conversion model to output a modified digital mask.
9. DLP projector for 3D printing; An image sensor that detects light projected from the above DLP projector and generates a light distribution image; and A computing device that generates a digital mask correction model for correcting the digital mask of the DLP projector using the light distribution image above, The above computing device is, The above light distribution image is transformed to generate a critical profile image consisting of a binary level image; Training a first image transformation model using the above threshold profile image as an input variable and the above light distribution image as an output variable; A digital mask correction device characterized by using the light distribution image above as an input variable and the digital mask above as an output variable to train a second image conversion model.
10. In Paragraph 9, The above computing device is, Input the original digital mask corresponding to the desired print result into the first image conversion model; A digital mask correction device characterized by inputting the output of the first image conversion model into the second image conversion model to output a modified digital mask.
11. In Paragraph 9, A digital mask correction device characterized by the image sensor being positioned on the focal plane of the DLP projector.
12. In Paragraph 9, A digital mask correction device further comprising a neutral density filter disposed between the DLP projector and the image sensor, which adjusts the intensity of light so that the intensity of light projected from the DLP projector is included within the measurement range of the image sensor.