System for Active Pollution Prevention in Power Plant Facilities
The real-time physics-AI-control system generates optimal soundproof and vibration-damping geometric patterns for power plants, addressing non-linear and complex noise and vibration challenges by integrating AI with physical trial and error, achieving efficient and adaptive noise and vibration reduction.
Patent Information
- Application Number
- KR1020260110495
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2026-06-17
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2046-03-04
AI Technical Summary
Conventional methods for reducing noise and vibration in power plants are ineffective in responding to non-linear and complex pollution patterns that change in real-time, particularly in thermal power plants, due to limitations in frequency band specificity and computational complexity.
A real-time physics-AI-control integrated system that utilizes a pre-trained latent diffusion model to generate optimal soundproof and vibration-damping geometric patterns, converting them into variable surface structures through a closed-loop control system, combining physical trial and error with AI pattern synthesis.
The system effectively adapts to variable and non-linear noise and vibration environments, providing stable soundproofing and vibration damping across a wide frequency band, enhancing environmental safety and equipment lifespan.
Smart Images

Figure 112026073673547-PAT00001_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a technology for reducing noise and vibration in power plant facilities, and more specifically, to a real-time physics-AI-control integrated system and method that actively controls pollution by visualizing physical state data, such as temperature, pressure, or vibration of the facility surface, in real time, discovering optimal soundproof and vibration-damping geometric patterns through a generative AI based on a pre-trained latent diffusion model that utilizes this as conditioning data, and physically transferring the generated patterns to a variable surface structure. Background Technology
[0002] The following description merely provides background information related to the present embodiment and does not constitute prior art.
[0003] Thermal power plants are core infrastructure for national energy supply, but the high-intensity noise and vibration inevitably generated during the operation of the facilities have been pointed out as serious environmental pollution problems. In particular, high-frequency roars generated from HP bypass valves through which high-pressure steam is released into the atmosphere, low-frequency vibrations transmitted to the ground by coal mills with massive millstone structures, and intermittent impact noises and airborne dust generated from air lock feeders at the bottom of electrostatic precipitators threaten the safety of workers and cause massive damage to nearby communities.
[0004] To address these issues, conventional passive methods, such as installing fixed sound insulation, sound absorption, or vibration damping pads, have been primarily used. However, these methods are limited in their effectiveness to reducing noise and vibration within specific frequency bands and have clear limitations in that they cannot actively respond to non-linear and complex pollution patterns that change in real-time according to variable operating conditions, such as load fluctuations in power plants. Another approach is Active Noise Control (ANC) technology, which cancels noise by generating anti-phase sound waves; however, this is only effective to a limited extent for low-frequency noise with simple and predictable sound sources. It is virtually impossible to apply to complex turbulent environments, such as those generated in power plants, or to broadband high-frequency noise.
[0005] Meanwhile, although there have been attempts to design optimal soundproof and vibration-damping structures using numerical analysis techniques such as Computational Fluid Dynamics (CFD) or the Finite Element Method (FEM), solving the Helmholtz equation or the Navier-Stokes equation in real time under the complex boundary conditions of power plants is virtually impossible even with current computing technology. This has acted as a fundamental bottleneck in the implementation of real-time adaptive control systems.
[0006] Therefore, there is an urgent need for a new paradigm of active pollution control technology that moves beyond the conventional limitations of directly calculating physical laws and responds in real-time to the actual physical state of power plant facilities to autonomously identify and physically implement optimal soundproofing and vibration-damping structures. The problem to be solved
[0007] The present invention has been devised to solve the problems of the aforementioned prior art, and its purpose is to provide an active pollution prevention method and system capable of maximizing reduction efficiency by responding in real time to non-linear, atypical, and broadband noise and vibration occurring in the harsh and variable operating environment of a coal-fired power plant.
[0008] Another objective of the present invention is to provide a novel approach for real-time 'discovery' of optimal soundproof and vibration-damping shapes, even in non-computable environments, by combining the pattern synthesis capabilities of a pre-trained image generation model with physical trial and error, instead of computing complex physical equations. In particular, it utilizes the fact that AI can present physical solutions through the exploration of the latent space of vast geometric data (fractals, mazes, grids, etc.), even if it has not directly learned physical laws.
[0009] Another objective of the present invention is to implement a closed-loop control system that immediately converts a digital pattern generated from visual data representing the physical state of an equipment into an actual physical structure (surface irregularities or internal stiffness), measures the effect using a sensor, and continuously evolves the pattern in a direction that improves performance.
[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 a person skilled in the art from the description below. means of solving the problem
[0011] According to one aspect of the present disclosure, a method performed by an active pollution control system to reduce noise or vibration generated in a power plant facility comprises: acquiring physical state data in real time, including a surface temperature distribution, a pressure distribution, or a vibration distribution of the power plant facility, through a sensor installed in the power plant facility; generating a plurality of candidate geometric pattern images through a pre-learned latent diffusion-based image generation model by utilizing a two-dimensional energy distribution map image converted from the acquired physical state data as conditioning data; converting pixel information of each generated candidate geometric pattern image into physical driving parameters including one or more of voltage, current, or light intensity; sequentially deforming a variable surface structure provided in the power plant facility according to the converted driving parameters to realize each candidate geometric pattern image as a physical shape or an internal stiffness distribution; and evaluating the level of noise or vibration reduction by each realized physical shape or internal stiffness distribution through a sensor installed in the power plant facility. The present invention provides a method comprising the process of obtaining an optimal geometric pattern image that brings about an optimal soundproofing or dustproofing effect by comparing the above-mentioned reduction levels.
[0012] In some examples, the image generation model may be induced to synthesize a pattern of higher geometric complexity corresponding to a region of high energy level in the energy distribution map image.
[0013] In some examples, the variable surface structure may include a pin array installed to wrap around the outer wall of a high-pressure steam discharge valve and composed of a plurality of metal pins and a piezoelectric element that individually drives each of the metal pins to form a fractal surface that scatters high-frequency noise by protruding the metal pins to a height proportional to the pixel information. Additionally or alternatively, the variable surface structure may include an MR fluid damper installed between the bottom support and the floor foundation of a coal pulverizer and composed of a plurality of cells filled with magnetorheological fluid whose viscosity changes according to the strength of the magnetic field and an electromagnetic coil grid that controls each cell, forming an acoustic bandgap that blocks the transmission of low-frequency vibrations by controlling the local stiffness distribution by applying a magnetic field proportional to the pixel information. Additionally or alternatively, the variable surface structure may include a photosensitive polymer skin that is attached to a coal ash discharge pipe and comprises a photosensitive polymer coating that changes the local Young's modulus when exposed to light of a specific wavelength, and a projector that projects a pattern image onto the photosensitive polymer coating, thereby forming an acoustic trap that absorbs impact sound by controlling the stiffness distribution through the projection of light of an intensity proportional to the pixel information.
[0014] In some examples, the process of generating the plurality of candidate geometric pattern images includes a process of performing preprocessing on the pixel values of the energy distribution map image, including quantization that converts them into a plurality of discrete levels or thresholding that extracts only regions above a specific threshold, and the image generation model can convert the preprocessed energy distribution map image into a depth map or a Canny edge map and use it as an input to a ControlNet for conditional control.
[0015] In some examples, the process of determining the optimal geometric pattern image may include: a process of selecting a candidate geometric pattern image that exhibits the highest reduction level among the candidate geometric pattern images as a parent image for the evolution of the next generation; and a process of searching for the optimal geometric pattern image by generating new variant pattern images through an image-to-image transformation using the selected parent image as input. For example, the searching process may search for new variant patterns that are structurally different from the parent image by setting the denoising strength of the image-to-image transformation to a first threshold or higher when the reduction level evaluated for the selected parent image is less than a preset target value, and search for finely tuned variant patterns that maintain the structural features of the parent image by setting the denoising strength to a second threshold or lower when the reduction level evaluated for the selected parent image is greater than the target value.
[0016] In some examples, the method may further include the process of constructing a digital twin having the same physical model as the power plant facility in a virtual space; and, prior to real-world implementation of the physical shape or internal stiffness distribution, simulating the candidate geometric pattern images on the digital twin to pre-evaluate the expected reduction effect, and selecting some of the candidate geometric pattern images based on the evaluation results to perform the real-world implementation.
[0017] According to another aspect of the present disclosure, an active pollution control system for reducing noise or vibration generated in a thermal power plant facility comprises: a sensor installed on the surface of said power plant facility to measure physical state data, including surface temperature distribution, pressure distribution, or vibration distribution, in real time; and a controller connected communicably to said sensor and comprising a processor and a memory. An active pollution prevention system can be provided, comprising a variable surface structure in which a physical shape or internal stiffness distribution is deformed in real time according to a control signal of the controller, wherein the controller includes a GPU (Graphics Processing Unit) capable of large-scale parallel computation to convert physical state data acquired from the sensor into a two-dimensional energy distribution map image, and utilizes the energy distribution map image as conditional guide data to generate a plurality of candidate geometric pattern images in parallel through an image generation model based on a pre-trained latent diffusion model, and includes a mapping unit based on an FPGA (Field Programmable Gate Array) or ASIC to convert pixel information of each generated pattern image into physical driving parameters in real time to sequentially deform the variable surface structure, evaluate the level of noise or vibration reduction according to each deformation through the sensor, and perform a trial-and-error-based generative evolution process to generate new variant patterns through image-to-image conversion by selecting the pattern with the highest measured reduction level as the parent image. Effects of the invention
[0018] According to the present invention, the following effects can be expected.
[0019] First, it can maximize pollution reduction efficiency by adapting in real time to the variable and non-linear noise and vibration environment of power plants. Unlike conventional fixed or passive pollution control facilities, this invention immediately reflects the actual physical state of the equipment to autonomously form an optimal physical response structure, thereby maintaining stable soundproofing and vibration damping performance even in situations where operating conditions change rapidly.
[0020] Second, it presents a new paradigm for solving complex turbulence and vibration problems that are impossible to analyze computationally. Instead of directly solving physical equations, engineering challenges can be practically solved without computation by combining the random yet vast pattern synthesis capabilities of image generation models with selections based on physical feedback. In particular, by adjusting denoising intensity, it is possible to effectively search for a global optimal solution without getting stuck in a local optimal.
[0021] Third, a single system can simultaneously address broadband noise and vibration. Since the fractal or asymmetric lattice structures generated by AI inherently contain geometric features of various scales, they can effectively attenuate pollution across a wide frequency band, ranging from high-frequency noise from HP bypass valves to low-frequency vibrations from differentiators.
[0022] Fourth, it effectively responds to strengthening environmental regulations, such as the Noise and Vibration Control Act, and contributes to improving the working environment and extending the lifespan of equipment. By reducing the accumulation of structural fatigue caused by noise and vibration and simultaneously suppressing secondary pollution such as fugitive dust, it supports the sustainable operation of power plants and can fundamentally prevent the occurrence of complaints.
[0023] The effects of the present disclosure are not limited to those mentioned above, and other unmentioned effects will be clearly understood by a person skilled in the art from the description below. Brief explanation of the drawing
[0024] FIG. 1 is a diagram showing the schematic configuration of a thermal power generation system to which an embodiment of the present invention can be applied. FIG. 2 is a block diagram showing the overall architecture of an active pollution prevention system according to one embodiment of the present invention. FIG. 3 is a flowchart illustrating the process of preprocessing heatmap data and generating conditional geometric patterns according to an embodiment of the present invention. FIGS. 4a to 4c are conceptual diagrams for explaining a control mechanism in which pixel information of a generated pattern image is mapped to physical parameters of a variable surface structure according to an embodiment of the present invention. FIG. 5 is a conceptual diagram illustrating a trial-and-error-based generative evolution feedback loop according to one embodiment of the present invention. FIG. 6 is an illustrative diagram illustrating a metal fin array-based fractal acoustic scattering system for reducing high-frequency noise of an HP bypass valve as a first embodiment of the present invention. FIG. 7 is an illustrative diagram explaining a vibration damping system based on a magnetorheological fluid (MR Fluid) damper for blocking low-frequency ground vibrations of a differential, as a second embodiment of the present invention. FIG. 8 is an illustrative diagram illustrating a photosensitive polymer skin-based impact sound and dust prevention system for simultaneously controlling impact sound and dust scattering of an air lock feeder as a third embodiment of the present invention. FIG. 9 shows the configuration of an active pollution prevention system linked with a digital twin according to one embodiment of the present invention. Specific details for implementing the invention
[0025] Some embodiments of the present disclosure are described in detail below with reference to exemplary drawings. It should be noted that in assigning reference numerals to the components of each drawing, the same components are given the same reference numeral whenever possible, even if they are shown in different drawings. Furthermore, in describing the present disclosure, if it is determined that a detailed description of related known components or functions could obscure the essence of the present disclosure, such detailed description is omitted.
[0026] In describing the components of the embodiments according to the present disclosure, symbols such as first, second, i), ii), a), b), etc., may be used. These symbols are intended only to distinguish the components from other components, and the essence, order, or sequence of the components is not limited by the symbols. When a part in the specification is described as 'comprising' or 'having' a component, this means that, unless explicitly stated otherwise, it does not exclude other components but may include additional components.
[0027] The detailed description set forth below, together with the accompanying drawings, is intended to describe exemplary embodiments of the present disclosure and is not intended to represent the only embodiment in which the present disclosure can be practiced.
[0028] FIG. 1 is a diagram showing the schematic configuration of a thermal power generation system to which an embodiment of the present invention can be applied.
[0029] A thermal power generation system is broadly composed of a fuel supply system that stores and transports raw materials, a fuel processing system that processes the input fuel into a form suitable for combustion, a combustion air control system that preheats and supplies the air required for combustion, a power generation system that converts the thermal energy of fuel combustion into mechanical and electrical energy, and an exhaust system that processes and discharges exhaust gases generated after combustion.
[0030] Referring to FIG. 1, the fuel supply system includes a coal hopper (10), a coal feeder (11), and a coal chute (13). The coal hopper (10) receives and temporarily stores coal fuel to be used for power generation and performs a buffer function to stably supply fuel to a downstream process through a bottom discharge port. The coal feeder (11) is a device connected to the bottom of the coal hopper (10) that physically transports the stored coal toward the pulverizer, and plays a role in precisely adjusting the amount of coal supplied according to power generation load conditions. The coal chute (13) functions as a closed transport passage that guides the coal discharged from the coal feeder (11) to be smoothly fed into the downstream facility by its own weight, and prevents the leakage of dust generated during fuel transport.
[0031] The fuel processing system includes a pulverizer (30) and a rotary classifier (31). The pulverizer (30) performs a process of increasing the surface area of the fuel by crushing the coal fed through the coal chute (13) into pulverized coal, which has maximized combustion efficiency. The rotary classifier (31) controls the process so that the pulverized coal, which has been processed in the pulverizer (30) and has become sufficiently fine, is transferred to the next process, and the coal particles that have been crushed coarser than the standard are sorted and returned to the pulverizer (30) for re-crushing.
[0032] The combustion air control system includes an air preheater (50) and a blower (51). The blower (51) performs a pumping function by forcibly drawing in external air to supply oxygen required for the combustion reaction of the boiler (100) and pushing it into the air preheater (50) inside the system. The air preheater (50) functions as a heat exchanger that significantly improves the combustion efficiency of the boiler by preheating the combustion air flowing in from the blower (51) using the heat held by the high-temperature exhaust gas heading toward the chimney (90).
[0033] The power generation system includes a boiler (100), a steam turbine (70), and a generator (71). The boiler (100) is equipped with a combustion furnace and heat exchange tubes that generate high-temperature thermal energy by burning pulverized coal supplied from a rotary classifier (31) together with preheated air in an internal combustion chamber. The boiler (100) is responsible for heating water flowing inside the pipes using the high heat generated during combustion and converting it into high-temperature, high-pressure superheated steam. The steam turbine (70) converts the energy of the steam into mechanical rotational force by rotating internal blades at high speed using the energy of the high-temperature, high-pressure steam supplied from the boiler (100). At this time, the steam expanded in the steam turbine (70) can form a reheat cycle in which it is recirculated into a reheater inside the boiler (100), reheated, and then supplied back to the steam turbine (70). A high-pressure bypass valve may be installed between the boiler (100) and the steam turbine (70) to bypass steam in case of emergency. The high-pressure bypass valve serves to send ultra-high temperature and ultra-high pressure steam generated in the superheater to the condenser by bypassing the steam turbine (70) when the steam turbine (70) stops or the load changes rapidly. The generator (71) is mechanically connected to the rotating shaft of the steam turbine (70) and produces power by converting the rotational motion of the turbine into final electrical energy through the phenomenon of electromagnetic induction.
[0034] The exhaust system includes a chimney (90). The chimney (90) is a vertical structure that diffuses and discharges the final exhaust gas, which has passed through an air preheater (50) and completed the energy recovery process, to a high place in the atmosphere. The chimney (90) facilitates the smooth discharge of exhaust gas. Meanwhile, the gas discharged after combustion in the boiler is purified by passing through an electrostatic precipitator (ESP), and the fly ash collected during this process can be discharged to the outside through an air lock feeder.
[0035] The present invention can be intensively applied to sources of severe noise and vibration among these major facilities, which were extremely difficult to control using conventional passive methods. A pollution source according to one embodiment of the present invention is as follows.
[0036] First, there is the HP bypass valve. When the steam turbine (70) stops or the load changes rapidly, the extreme high-frequency noise generated as the high-temperature and high-pressure steam passes through the narrow valve is one of the worst noise sources in the power plant. This noise has non-linear and broadband frequency characteristics, making it difficult to effectively block with simple sound barriers.
[0037] Second, there is a pulverizer (30). This is a rotating machine that pulverizes coal using large rollers or balls, and generates massive low-frequency vibrations during operation. These vibrations are transmitted to the ground through the foundation structure of the power plant, causing vibration pollution in nearby areas and accelerating fatigue failure of the equipment itself.
[0038] Third, there is the air lock feeder. This is a device installed at the bottom of the electrostatic precipitator to discharge the collected coal ash to the outside. The impact noise generated as the coal ash falls intermittently and the fugitive dust caused by air vortices during the discharge process are complex sources of pollution that worsen the working environment.
[0039] The present invention is installed individually or integrally at pollution sources having specific and distinct physical characteristics, thereby fundamentally solving the problem by generating and applying an optimal physical response structure tailored to the characteristics of each pollution source in real time.
[0040] FIG. 2 is a block diagram showing the overall architecture of an active pollution prevention system according to one embodiment of the present invention.
[0041] The active pollution prevention system (20) can operate by forming an organic closed-loop of sensing-analysis-generation-implementation-feedback. To this end, the active pollution prevention system (20) is largely composed of a sensor (200), a controller (220), and a driving device (240).
[0042] The sensor (200) serves to acquire physical state data of the target equipment (22) in real time. Depending on the type of pollution, the sensor (200) may include an infrared thermal imaging camera, an acceleration sensor grid, a sound pressure sensor, or a particle concentration meter. For example, in the case of an HP bypass valve, an infrared thermal imaging camera may be used to capture a heatmap image of the point where steam pressure is concentrated and noise is generated. As another example, in the case of a pulverizer, an acceleration sensor grid may be placed on the outer wall of the equipment and the foundation of the ground to analyze the strong low-frequency vibrations generated during the rotation and grinding processes of the equipment in three dimensions. Through this, the amplitude and frequency characteristics of the vibrations can be identified in real time to track the path of vibration propagation to the surrounding ground and to diagnose the mechanical fatigue state of the equipment. As yet another example, in the case of an air lock feeder, a sound pressure sensor may be installed near the equipment to capture intermittent impact noise and air vortex noise generated during coal ash discharge. At the same time, by operating a particle concentration meter in parallel to measure the diffusion range of fine dust generated around the emission outlet, the complex pollution situation involving noise and dust can be converted into quantitative data and managed.
[0043] The controller (220) acts as the core brain of the system and includes a heatmap conversion module (222), an image generation module (224), a mapping module (226), and a feedback control module (228).
[0044] The heatmap conversion module (222) converts raw data (e.g., temperature, acceleration values) received from the sensor (200) into a two-dimensional visual energy distribution map, i.e., a heatmap image, that the image generation module (224) can understand. For example, local temperature and pressure change data measured at the HP bypass valve are mapped into color information on a two-dimensional coordinate system. This expresses the point where turbulence and noise caused by the rapid expansion of steam occur most intensively as a visual energy density. In the case of the differential (30), minute vibration signals entering from the acceleration sensor grid placed throughout the facility are converted into location-specific intensity and processed through interpolation to generate a vibration distribution map that allows the pattern of vibration propagation across the entire surface and foundation of the facility to be grasped at a glance. Sound pressure and particle concentration data collected around the air lock feeder are also converted into image data that spatially visualizes the noise intensity and dust dispersion density centered on the source.
[0045] The image generation module (224) may be implemented as an image generation model based on a pre-trained latent diffusion model, for example, stable diffusion. Specifically, the image generation module (224) may be a conditional text-to-image generation model that receives a heatmap generated by the heatmap transformation module (222) as conditional guide data and generates multiple candidate geometric pattern images according to text prompts such as 'complex fractal pattern', 'asymmetric grid', etc. Instead of explicitly learning physical laws regarding pollution sources, the image generation model may utilize geometric diversity within the latent space already secured through hundreds of millions of visual data. That is, the image generation model does not calculate a solution for noise engineering, but can function as a 'pattern source' that infinitely generates complex patterns optimal for physically scattering waves or forming air gaps, such as "fractals," "mazes," and "lattices." The image generation module (224) is executed on a GPU capable of large-scale parallel computing (e.g., NVIDIA A100, etc.) and can simultaneously generate multiple candidate images in batch units.
[0046] The mapping module (226) converts pixel information (e.g., brightness, color) of a digital pattern image output by the image generation module (224) into physical driving signals (e.g., voltage, current, light intensity) for controlling a driving device (240). The mapping module (226) is implemented as a Field Programmable Gate Array (FPGA) or an ASIC, which can minimize the delay time for converting image data into actuator control signals. The mapping module (226) can convert pixel information into physical driving signals based on a pre-stored look-up table. As another example, the mapping module (226) can convert pixel information into physical driving signals by using a simple linear function as a mapping function. Even if the conversion logic by the mapping module (226) is configured in such a simplified manner instead of precise physical equations, the objective of the present invention can be achieved. This is because the system does not rely on the absolute numerical accuracy of individual patterns, but is based on a 'generative evolution' mechanism that directly identifies the optimal soundproofing and vibration damping shape by testing a vast pool of synthesized candidates in real time using the latent space of the image generation model, as described below. In other words, minute physical errors that may arise from simplified mapping are naturally corrected through iterative pattern variations via feedback loops and fit evaluations based on sensor data; consequently, the physical actuation value that delivers the highest reduction performance in actual field environments can be discovered in real time through trial and error.
[0047] The feedback control module (228) evaluates the actual amount of noise and vibration reduction resulting from the operation of the driving device (240) based on newly received data from the sensor (200). The feedback control module (228) selects the most effective pattern based on the information and intervenes in the next-generation pattern generation process (e.g., Img2Img conversion) of the image generation module (224) to progressively optimize the performance of the system.
[0048] The driving device (240) is a variable surface structure that deforms the actual physical structure according to the command of the controller (220). It is composed of an actuator suitable for the characteristics of the target facility (22), such as a metal pin array, a magnetorheological fluid (MR Fluid) damper, or a photosensitive polymer skin.
[0049] Through this architecture, the active pollution prevention system (20) can detect changes in the physical state of the target facility (22) in real time, and self-draw and sculpt an optimal physical barrier to respond to pollution.
[0050] FIG. 3 is a flowchart illustrating the process of preprocessing heatmap data and generating conditional geometric patterns according to an embodiment of the present invention.
[0051] FIG. 3 exemplarily illustrates a process that is the core of the present invention, in which physical data is converted into a visual guide to enable AI to generate an optimal pattern. This process can be understood as a concept of replacing physical calculations with visual pattern matching.
[0052] The process begins at the raw data acquisition (S310) stage. The sensor (200) acquires analog or digital signals, such as heat and vibration, from the surface of the equipment. For example, a raw heatmap with a smooth gradient is acquired from a thermal imaging camera, and vibration intensity values at each point are acquired from an acceleration sensor grid.
[0053] Next, the preprocessing and boundary clarification (S320) step is performed. The raw data must be processed into a form that can be effectively recognized by a conditional control module, such as the ControlNet of an image generation model. Specifically, a quantization process is performed to convert a smooth gradient heatmap into 5 to 10 distinct color levels. This process produces the effect of creating virtual boundary lines at points where energy levels change, much like contour lines on a map. Additionally, or alternatively, a clear outline (mask) of the core area where pollution occurs can be extracted through thresholding, which converts only the area above a certain threshold (e.g., a specific temperature or vibration intensity) to white and the rest to black.
[0054] The preprocessed image is converted into a final input to guide the image generation model during the conditional control data generation (S330) stage. For example, the quantized contour image can be converted into a depth map where bright areas (i.e., high energy) are recognized as protruding and dark areas as recessed. This is the most intuitive method to induce the image generation model to draw more three-dimensional and complex structures in areas of energy concentration. The converted map is used as input to the ControlNet, which forcibly establishes a logical correlation between physical phenomena and generation patterns by providing visual guidelines to place higher geometric complexity in areas of concentrated energy. This implements the principle of complexity matching in real-time, which aligns the energy distribution of the noise source with the complexity of the reduction structure (sink). Additionally, or alternatively, the outline information of the binarized image extracted during the preprocessing process can be converted into a Canny edge map and utilized. The Canny edge map provides physical structural skeleton information into which the image generation model must fill in complex geometric patterns. This serves to guide the image generation model to synthesize intricate patterns by using boundaries where rapid energy changes occur or specific vibration node points as guidelines.
[0055] In the latent diffusion model inference (S340) step, the image generation model receives preprocessed control data (e.g., depth map and / or Canny edge map) as input to the ControlNet and simultaneously receives a text prompt such as "ultra-fine porous fractal pattern". The image generation model synthesizes a new image within the latent space that satisfies both conditions. This process can be understood as drawing the most plausible image probabilistically based on visual similarity and constraints, rather than interpreting physical meaning.
[0056] In the final geometric pattern image generation (S350) step, the image inferred within the latent space is converted into pixel space to output a high-resolution black-and-white or grayscale pattern image, which serves as the final design blueprint to be used for noise and vibration reduction. This image can serve as a direct command for the next step, physical implementation.
[0057] FIGS. 4a to 4c are conceptual diagrams for explaining a control mechanism in which pixel information of a generated pattern image according to various embodiments of the present invention is mapped to physical parameters of a variable surface structure.
[0058] FIGS. 4a through 4c exemplarily show how the digital pattern image generated in FIG. 3 leads to changes in the actual physical world. This can be understood as a process of translating the product of the image generation model into a language that real-world actuators can understand and execute.
[0059] The mapping module (226) serves to convert information contained in each pixel of the generated two-dimensional pattern image (40) into quantitative physical control parameters. The most common information is the pixel brightness value (Grayscale Value), which is usually an 8-bit integer, i.e., has a value from 0 (black) to 255 (white).
[0060] According to an embodiment of the present invention, this pixel brightness value can be mapped to different physical quantities as follows.
[0061] FIG. 4a shows the relationship in which pixel brightness is converted into the protrusion height of a pin. The position (x, y) of each pixel in the image (40) corresponds one-to-one with a specific pin position in the pin array (42). Pixel brightness value (I xy ) is the voltage (V) to be applied to the piezoelectric element (402) at the bottom of the corresponding pin through the voltage conversion circuit (400). xy It is converted into ). Voltage (V xy ) is the protrusion height of the pin (H xy Forms ). For example, V xy ∝I xy Depending on the relationship, brightness 0 is set to correspond to 0V (no protrusion), and brightness 255 is set to correspond to maximum voltage (e.g., 50mm protrusion). This allows the brightness and contrast of the image to be sculpted directly into a physical texture terrain.
[0062] FIG. 4b illustrates the concept of pixel brightness being converted into local stiffness. Each cell location of the MR fluid grid (44) corresponds to a pixel location in the image (40). Pixel brightness value (I xy ) is the current (A) flowing through the electromagnetic coil at the bottom of the cell. xy It determines the size of ). When a strong current flows (bright pixel), the MR fluid in the corresponding cell has internal iron particles aligned by the magnetic field, and momentarily acquires high stiffness close to that of a solid. On the other hand, when the current is weak (dark pixel), it maintains a liquid state and becomes flexible. Through this, a stiffness map is formed in which the hardness of the floor changes in real time according to the pattern drawn by the image generation model.
[0063] Fig. 4c shows the concept of pixel brightness being converted into elasticity.
[0064] In the case of photosensitive polymer skin control, pixel brightness can be mapped to the intensity or wavelength of light projected by the projector (46). Strong ultraviolet (UV) light is irradiated onto the area corresponding to the bright pixel to promote the photo-crosslinking reaction of the polymer, rapidly increasing the Young's modulus (elastic modulus) of that part and making it hard. The dark area does not receive light and remains soft.
[0065] Through this pixel-physical mapping mechanism, abstract patterns generated by the image generation model are immediately realized as tangible structures that physically obstruct the propagation of noise and vibration waves. Meanwhile, the modification of the physical shape by the variable surface structure is equivalent to modifying the boundary conditions of the wave equation in real time. Since the behavior of acoustic and elastic waves is determined by the boundary conditions of the medium along their propagation path, waves react immediately to changed boundaries at the speed of sound or faster. Therefore, if the control loop of this system is sufficiently fast, a physically immediate pollution reduction effect can be achieved.
[0066] FIG. 5 is a conceptual diagram illustrating a trial-and-error-based generative evolution feedback loop according to one embodiment of the present invention.
[0067] Figure 5 conceptually illustrates a generative evolutionary feedback loop, which is a core strategy of the present invention for addressing complex system problems where the answer is unknown. Similar to Darwin's theory of evolution, this process finds an optimal solution through 'random variation' and 'selection by the environment'.
[0068] The loop starts at the initial candidate generation (S510) stage. Through the process described in FIG. 3, the controller (220) generates N different initial pattern images in parallel by using the same heatmap guide and prompt but assigning different random seed values. This is like children with different personalities born from the same parent. To ensure real-time performance, acceleration models such as SD Turbo or LCM (Latent Consistency Models) can be applied to shorten the inference step to 1 to 4 steps (approximately 50 to 100 ms). Additionally, search efficiency can be increased by generating 4 to 8 candidate patterns simultaneously through GPU batch processing.
[0069] In the fitness evaluation (S520) step, N patterns are sequentially physicalized by the driving device (240). At the moment each pattern is implemented, the feedback control module (228) evaluates the reduction amount based on the noise or vibration measured through the sensor (200). The reduction amount can be considered as the fitness of each entity (i.e., pattern). The driving sequence undergoes sensing, heat map conversion, pattern generation, FPGA signal conversion, actuator driving, and stabilization processes, and the total loop time is set to approximately 0.2 to 0.3 seconds to enable real-time control.
[0070] In the best object selection (S530) step, the feedback control module (228) selects the pattern with the highest measured reduction amount as the survivor or the parent image of the next generation.
[0071] The selected parent image is used as input to the image-to-image (Img2Img) conversion function of the image generation module (224) in the image-to-image mutation generation (S550) step. This function is a process of adding a small amount of noise to the original image and then reconstructing it into a new image, which is equivalent to causing mutation. An important parameter at this stage is the denoising strength.
[0072] If the performance of the current best specimen falls short of the target value, it is determined that the current solution has fallen into a local optimum, and the system enters Exploration mode with a high denoising intensity (e.g., 0.7–0.9). This allows for the exploration of a wider solution space by drastically altering the shape of the parent image to generate completely new child patterns. Conversely, if performance is improving satisfactorily, the system enters Exploitation mode with a low denoising intensity (e.g., 0.2–0.4). This is a process of further refining the current solution by finely adjusting only the detailed textures while largely maintaining the superior traits (structures) of the parent.
[0073] While conventional genetic algorithms (GAs) rely on random bit flips, which have a high probability of generating physically meaningless shapes, this invention generates only structurally meaningful patterns within the latent space of a pre-trained model, thereby significantly compressing the search range. Theoretically, this structural search provides a logical basis for approaching an optimal solution with only a few generations of evolution, unlike existing methods that require thousands of trials and errors, and a flexible search strategy utilizing denoising intensity plays a decisive role in escaping local optima.
[0074] Meanwhile, a prompt mutation (S540) may occur depending on the frequency characteristics of the residual noise measured by the sensor. For example, if high-frequency noise is still measured strongly, keywords such as "micro-scale pores" and "dense texture" are added to the existing prompt to induce the next generation pattern to have a finer structure.
[0075] This evolutionary loop is repeated until the amount of noise and vibration reduction converges. Through this, even while utilizing an image generation model that has not learned spatial physical laws, it is possible to improve the generated output in a positive direction and discover an optimal physical solution that is perfectly adapted to the field environment.
[0076] FIG. 6 is an illustrative diagram illustrating a metal fin array-based fractal acoustic scattering system for reducing high-frequency noise of an HP bypass valve as a first embodiment of the present invention.
[0077] In this embodiment, the driving device (240) is composed of a pin array (640) that wraps around the outer wall surface of the valve body and the connecting pipe like clothing. The pin array (640) is a variable surface structure and may include tens of thousands of fine metal pins. The driving device (240) may further include a piezoelectric element (642) that controls each pin individually. The spacing of the pins is designed to be dense, 5 mm or less, to effectively respond to short wavelengths of high-frequency noise (e.g., 3.4 cm or less).
[0078] A thermal imaging camera (600) may be used as the sensor (200) to capture the temperature distribution on the valve surface and identify the core noise-generating area where high-temperature steam is concentrated. The controller (220) converts this thermal heat map into a depth map and generates a pattern image using a "micro-porous fractal pattern" prompt. The controller (220) converts the pixel brightness of the generated image into a voltage signal and transmits it to each piezoelectric element (642). The piezoelectric element (642) protrudes a pin vertically from the valve surface in proportion to the applied voltage. As a result, physical irregularities are formed on the valve surface that have the same shape as the fractal pattern drawn by the image generation model.
[0079] The soundproofing mechanism is broadband acoustic scattering based on fractal structures. Fractal structures possess self-similarity, containing complex geometric features at all scales. The scattering effect is maximized when the wavelength corresponding to the frequency of the noise matches the size of the surface irregularities; since fractals simultaneously possess irregularities of various sizes, they can dissipate energy by scattering broadband noise—rather than specific frequencies—simultaneously. High-frequency sound waves generated inside the valve and attempting to radiate into the air through the surface scatter in tens of thousands of directions the moment they strike these complex and irregular irregular sections. The scattered sound waves interfere with each other, causing their energy to dissipate rapidly.
[0080] FIG. 7 is an illustrative diagram explaining a vibration damping system based on a magnetorheological fluid (MR Fluid) damper for blocking low-frequency ground vibrations of a differential, as a second embodiment of the present invention.
[0081] In this embodiment, the driving device (240) is implemented as a magnetorheological fluid in the form of a large active base installed between the lower support of the pulverizer and the bottom foundation of the power plant. For example, the driving device (240) may include an MR fluid damper consisting of a grid of small electromagnetic coils (742) installed at the bottom of each cell, and a plurality of cells (740) filled with magnetorheological fluid that undergoes a phase change from liquid to solid state depending on the strength of the magnetic field.
[0082] As the sensor (200), an acceleration sensor grid including multiple acceleration sensors (700) may be applied, and a two-dimensional vibration distribution map of the bottom of the differential (30) is generated in real time based on data collected from the acceleration sensors (700). The controller (220) observes this vibration map and generates an asymmetric Voronoi lattice pattern image capable of effectively dispersing or blocking vibration energy. The controller (220) sends a strong current to the electromagnetic coil (742) corresponding to the bright pixels of the generated image. The MR fluid in the cell (740) where a strong magnetic field is formed rapidly increases in viscosity within a few milliseconds and becomes hard like a solid.
[0083] The vibration isolation mechanism is the formation of an acoustic / elastic bandgap. Regular low-frequency vibration waves attempting to propagate into the ground encounter an irregular and asymmetric stiffness map drawn by the image generation model. This non-uniform stiffness distribution acts as a vibration exclusion zone where vibration waves in specific frequency bands cannot physically pass through. In other words, vibration energy is blocked by this artificial terrain, or dispersed into multiple paths where the energy is attenuated and cannot be transmitted to the ground.
[0084] FIG. 8 is an illustrative diagram illustrating a photosensitive polymer skin-based impact sound and dust prevention system for simultaneously controlling impact sound and dust scattering of an air lock feeder as a third embodiment of the present invention.
[0085] In this embodiment, the driving device (240) may be implemented as a polymer film in the form of a Smart Skin attached to the discharge pipe (80) where coal ash falls. This may be a photosensitive polymer film such as azobenzene, which changes its elastic modulus in response to light of a specific wavelength (e.g., UV). Azobenzene molecules possess reversible photoisomerization characteristics in which the molecular structure changes variably depending on light irradiation conditions. For example, upon irradiation with ultraviolet (UV) light, it changes from a stable Trans form to a bent Cis form, inducing a local change in elastic modulus (stiffness). Subsequently, it can be restored to its original Trans form by irradiating visible light or undergoing a thermal relaxation process. Through this photoreversibility mechanism, active control is enabled to immediately erase and reconstruct the physical stiffness distribution of the surface in accordance with the real-time variation pattern generated by the controller (220).
[0086] A projector (842) for projecting a pattern image onto a polymer film (840) may be installed on the outside of the pipe. The polymer film (840) may be attached in a manner that wraps around the outer wall of the discharge pipe (80) or may be attached to the inner wall of the discharge pipe (80). In this case, at least a portion of the discharge pipe (80) may include a projection window formed of a transparent material (e.g., reinforced quartz or high-strength polymer) with high ultraviolet (UV) transmittance so that the light from the projector (842) can reach the polymer film (840) on the inner wall of the pipe. As another example, the projector (842) may be positioned to project an image in the direction of the internal conduit through an upper or lower opening of the discharge pipe (80), and the controller (220) may perform a geometric correction algorithm to correct image distortion according to the projection angle in real time.
[0087] A sound pressure sensor (800) and a particle concentration meter (802) may be applied as the sensor (200). The controller (220) synthesizes the sound pressure distribution and particle density data inside the outlet measured by the sensor (200) to generate a pollution map. The controller (220) uses a dual-purpose prompt, such as a streamlined labyrinth, to generate a hybrid pattern image in which dust slides down easily while only sound is trapped. A projector (842) projects this pattern image onto a polymer film (840) on the inner wall of the pipe. The parts exposed to strong light undergo a change in molecular structure and harden, while the parts not exposed to light maintain their original flexible state.
[0088] The dual pollution prevention mechanism is as follows: As the elastic modulus (stiffness) of the exhaust pipe changes non-uniformly according to the complex maze pattern drawn by the image generation model, a type of acoustic metamaterial structure is formed inside. Impact sound waves generated by falling coal ash are trapped within this complex stiffness structure, undergoing infinite reflection until they dissipate as thermal energy. At the same time, the macroscopic shape of the pattern is designed to be streamlined to smoothly guide the airflow. This functions as an acoustic trap that physically suppresses the scattering phenomenon where coal ash creates vortices and flies out from the exhaust port.
[0089] FIG. 9 shows the configuration of an active pollution prevention system linked with a digital twin according to one embodiment of the present invention.
[0090] To increase operational stability and economic efficiency, a digital twin can be integrated into the active pollution prevention system (20d).
[0091] For example, to enable digital twin integration, a virtual digital model corresponding 1:1 with the real-world target equipment (22) is constructed on a computer. Before operating the actual drive unit (240), heat map data received from the sensor (200) is first input into the digital twin. The twin controller (920) generates patterns within a virtual space, implements virtual deformations through the twin drive unit (940), and simulates the resulting noise and vibration reduction effects at high speed. Through this simulation, a few of the most promising initial candidate groups can be selected in advance from among hundreds or thousands of candidate patterns. By applying these pre-optimized patterns to the actual drive unit (240), the initial search time can be significantly reduced, and unstable behavior that may occur during the trial-and-error process can be minimized, thereby ensuring the stability of the system.
[0092] Each component of the device or method according to the present invention may be implemented in hardware or software, or in a combination of hardware and software. Additionally, the function of each component may be implemented in software, and a microprocessor may be implemented to execute the function of the software corresponding to each component.
[0093] Various embodiments of the systems and techniques described herein may be realized as digital electronic circuits, integrated circuits, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include being implemented as one or more computer programs executable on a programmable system. A programmable system comprises a storage system, at least one input device, and at least one programmable processor (which may be a special-purpose processor or a general-purpose processor) coupled to receive data and instructions from and transmit data and instructions to at least one output device. Computer programs (which are also known as programs, software, software applications, or code) include instructions for the programmable processor and are stored on a "computer-readable recording medium."
[0094] Computer-readable recording media include all types of recording devices in which data that can be read by a computer system is stored. Such computer-readable recording media may be non-volatile or non-transitory media such as ROM, CD-ROM, magnetic tape, floppy disk, memory card, hard disk, magneto-optical disk, and storage device, and may also include transitory media such as data transmission media. Additionally, computer-readable recording media may be distributed across networked computer systems, and computer-readable code may be stored and executed in a distributed manner.
[0095] Although the flowcharts and timing diagrams in this specification describe each process as being executed sequentially, this is merely an illustrative explanation of the technical concept of one embodiment of the present disclosure. In other words, a person skilled in the art to which one embodiment of the present disclosure belongs may modify and adapt the flowcharts and timing diagrams in various ways, such as changing the order described in the flowcharts and timing diagrams or executing one or more of the processes in parallel, without departing from the essential characteristics of one embodiment of the present disclosure; therefore, the flowcharts and timing diagrams are not limited to a chronological order.
[0096] The above description is merely an illustrative explanation of the technical concept of the present embodiment, and a person skilled in the art to which the present embodiment belongs would be able to make various modifications and variations within the scope of the essential characteristics of the present embodiment. Accordingly, the present embodiments are intended to explain, not limit, the technical concept of the present embodiment, and the scope of the technical concept of the present embodiment is not limited by these embodiments. The scope of protection of the present embodiment shall be interpreted by the claims below, and all technical concepts within an equivalent scope shall be interpreted as being included within the scope of rights of the present embodiment. Explanation of the symbols
[0097] 10: Coal hopper 11: Coal feeder 13: Coal Chute 30: Differentiator 31: Rotary classifier 50: Air preheater 51: Blower 70: Steam turbine 71: Generator 90: Chimney 100: Boiler
Claims
Claim 1 An active pollution control system for reducing noise generated in power plant facilities comprises: a sensor that measures physical state data, including the surface temperature distribution of the power plant facilities, in real time through a sensor installed in the power plant facilities; a controller that is communicably connected to the sensor; and a variable surface structure whose physical shape is deformed according to a control signal of the controller, wherein the controller utilizes a two-dimensional energy distribution map image converted from the physical state data acquired from the sensor as conditional guide data, generates a plurality of candidate geometric pattern images through a pre-learned latent diffusion-based image generation model, and converts pixel information of each generated candidate geometric pattern image into physical driving parameters including voltage; An active pollution prevention system configured to sequentially deform the variable surface structure according to the converted driving parameters to realize each candidate geometric pattern image into a physical shape, evaluate the noise reduction level by each realized physical shape through the sensor, and compare the evaluated reduction levels to obtain an optimal geometric pattern image that brings about an optimal soundproofing effect, wherein the variable surface structure is installed to wrap around the outer wall of a high-pressure steam discharge valve and comprises a plurality of metal pins and a piezoelectric element that individually drives each metal pin, and includes a pin array that forms a fractal surface that scatters high-frequency noise by protruding each metal pin to a height proportional to the pixel information.
Citation Information
Patent Citations
Apparatus and method for detecting fault of structure considering noise environment
KR1020220093942A
