High-speed steel strip laser selected area cleaning and deoiling method and system based on UV / NIR detection and AI control

Abstract

The invention relates to the technical field of cleaning equipment, in particular to a high-speed steel strip laser selective cleaning and deoiling method and system based on UV / NIR detection and AI control. The method comprises a UV / NIR-AI-laser three-in-one closed loop, specifically, laser area selection and grading energy cleaning are driven through an energy demand map, full-width excessive irradiation is avoided, high speed and low energy consumption are both considered, the real-time speed of a steel belt is combined, laser cleaning parameters are dynamically decided through an AI model, and a high-speed galvanometer system is driven to implement precise energy projection on an oil-containing area. The method has the beneficial effects that no chemical cleaning agent is used in the whole process of the technical scheme, no waste water or waste liquid is generated, the generated oil gas waste can be uniformly collected and treated through an integrated extraction system, and the environmental protection requirement of modern industrial green manufacturing is completely met.

Description

Technical Field

[0001] This invention relates to the field of cleaning equipment technology, and in particular to a high-speed steel strip laser selective cleaning and degreasing method based on UV / NIR detection and AI control. Background Technology

[0002] In the field of flux-cored welding wire manufacturing, the surface of the steel strip is typically coated with a 50–80 mg / m² anti-rust oil film (such as palm oil or mineral oil) to prevent oxidation and corrosion during transportation and storage. However, this oil film decomposes under heat during subsequent welding, releasing gases such as hydrogen, carbon monoxide, and carbon dioxide, which can easily cause defects such as weld porosity, metal spatter, and slag inclusions, severely damaging the mechanical properties and appearance quality of the welded joint. To ensure welding quality, the residual oil content must be reduced to below 20 mg / m² before the steel strip enters the forming welding process. Although traditional alkaline cleaning or electrolytic degreasing processes can effectively remove oil, they generate large amounts of oily wastewater, facing increasingly stringent environmental regulations, and the cost of building or modifying cleaning lines is high and difficult to implement.

[0003] Dry laser cleaning technology is considered a potential alternative due to its advantages of producing no chemical reagents and no wastewater discharge. Its basic principle is to use high-energy laser pulses to instantly vaporize or break down the oil film, thereby achieving surface cleaning. However, in actual operation on high-speed continuous production lines (belt speeds of 50–70 m / s), the full-width uniform irradiation laser cleaning mode faces severe challenges: on the one hand, to cover the entire steel belt and ensure thorough removal of heavy oil areas, a total power of 8–12 kW is required, leading to a significant increase in equipment investment and energy costs; on the other hand, the oil film distribution on the steel belt surface is uneven, with localized heavy oil areas becoming cleaning bottlenecks, while light oil areas are prone to overheating or even annealing deformation due to excess energy, resulting in material performance degradation.

[0004] In the prior art, patent document publication number TWM673695U discloses a laser alignment cleaning device. This laser cleaning system generally lacks real-time sensing and dynamic response capabilities to the surface oil film state. Cleaning parameters (such as power density and spot size) are mostly fixed settings, making precise control based on the spatial distribution of the oil film impossible. This not only leads to energy waste and the risk of thermal damage but also makes it difficult to consistently achieve the process target of residual oil <20 mg / m². Especially under high-speed operating conditions, the traditional open-loop control mode cannot simultaneously consider cleaning efficiency, energy economy, and substrate thermal safety, significantly restricting the large-scale application of laser cleaning technology in high-end welding wire production lines.

[0005] Therefore, it is necessary to design a high-speed steel strip laser selective cleaning and degreasing method based on UV / NIR detection and AI control to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a high-speed steel strip laser selective cleaning and degreasing method based on UV / NIR detection and AI control, so as to overcome the above-mentioned shortcomings of the existing technology.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A laser selective cleaning and degreasing method for high-speed steel strip based on UV / NIR detection and AI control includes: A multispectral sensing array is arranged along the width direction of the high-speed steel strip. The fluorescence response intensity data of the surface of the high-speed steel strip under preset ultraviolet light excitation and the reflection and absorption data of preset near-infrared spectrum are collected synchronously through the multispectral sensing array to obtain the original multimodal spectral data stream characterizing the oil film distribution state. The original multimodal spectral data stream is input into a data fusion and quantization analysis model. The data fusion and quantization analysis model is based on a preset oil film fluorescence characteristic calibration curve and a near-infrared absorptivity-thickness calibration model. The fluorescence response intensity data and the near-infrared reflection absorption data are fused at the pixel level to generate a real-time, high-resolution two-dimensional quantitative oil film thickness map. The map accurately characterizes the oil film mass density of each micro-element region on the surface of the high-speed steel strip. The two-dimensional quantitative oil film thickness map and the real-time running speed value of the high-speed steel belt obtained by the speed encoder are used as inputs and fed into a pre-trained artificial intelligence laser parameter decision model. The artificial intelligence laser parameter decision model infers and outputs a laser cleaning operation parameter matrix that is precisely corresponding to the oil film thickness map in spatial coordinates in real time based on the input oil film thickness distribution and steel belt speed. The laser cleaning operation parameter matrix is ​​transmitted to a laser control command generation unit. The laser control command generation unit parses each parameter set in the matrix into a specific laser control sequence for a specific spatial coordinate point. The laser control sequence defines the target output power, pulse frequency, pulse width, and dwell time of the laser beam at that coordinate point. The laser control sequence is sent to a high-speed laser selective area execution system. The system drives its internal high dynamic response galvanometer scanning system and laser power modulation system to precisely project laser pulse energy with specific parameters onto the oil-containing area on the surface of the high-speed steel strip marked by the oil film thickness map, and performs vaporization stripping and cleaning of the oil film. For areas where the oil film thickness is lower than the preset cleaning threshold, no laser energy is applied. A cleaning effect verification array is deployed downstream of the high-speed laser selective area execution system. The verification array is used to generate a residual oil film thickness map after cleaning. A model adaptive optimization unit compares the residual oil film thickness map with a preset cleaning quality target value and generates an error feedback signal. The error feedback signal is used to fine-tune the artificial intelligence laser parameter decision model online or retrain it offline to continuously optimize the cleaning strategy.

[0008] Preferably, acquiring the raw multimodal spectral data stream characterizing the oil film distribution includes: A uniform ultraviolet light band with a wavelength of 365 nanometers is projected onto the surface of the high-speed steel strip through an ultraviolet light linear light source array; under the synchronous triggering of the ultraviolet light source array, a linear array complementary metal-oxide-semiconductor image sensor equipped with a 450-nanometer center wavelength bandpass filter is used to collect the fluorescence signal generated by the excitation of aromatic hydrocarbon compounds in the oil film, forming a fluorescence response intensity matrix. A near-infrared light band with a center wavelength of 1720 nm is projected onto the surface of the high-speed steel strip through a near-infrared linear light source array. Under the synchronous triggering of the near-infrared linear light source array, a linear array indium gallium arsenide image sensor equipped with a 1720 nm center wavelength bandpass filter collects the near-infrared light signal reflected from the surface of the high-speed steel strip, forming a near-infrared reflection intensity matrix. The fluorescence response intensity matrix and the near-infrared reflection intensity matrix together constitute the original multimodal spectral data stream.

[0009] Preferably, generating a two-dimensional quantitative oil film thickness map includes: The acquired near-infrared reflectance intensity matrix is ​​normalized and converted into a preliminary oil film absorptivity distribution map according to the Beer-Lambert law. The acquired fluorescence response intensity matrix is ​​then subjected to dark current correction and flat-field correction, and the fluorescence intensity nonlinear saturation phenomenon caused by excessive oil film thickness is corrected according to a preset fluorescence quenching effect compensation algorithm, resulting in a corrected fluorescence intensity distribution map. A built-in three-dimensional lookup table is used to perform a lookup operation on each pixel. The input of the three-dimensional lookup table is the preliminary oil film absorptivity and the corrected fluorescence intensity, and the output is the precise oil film mass density value in milligrams per square meter, ultimately generating the two-dimensional quantitative oil film thickness map.

[0010] Preferably, the artificial intelligence laser parameter decision model is a deep convolutional neural network structure, specifically an encoder-decoder architecture; the encoder part consists of cascaded convolutional layers and pooling layers, used to extract multi-scale spatial features from the input two-dimensional quantized oil film thickness map.

[0011] Preferably, the laser control command generation unit parses the laser cleaning operation parameter matrix into a laser control sequence by: mapping image pixel coordinates to steel strip physical coordinates; performing region segmentation based on an oil film thickness threshold, and generating control commands only for regions exceeding the threshold.

[0012] Preferably, the model adaptive optimization unit generates error feedback signals by: calculating the mean and variance of residual oil; if the mean exceeds 15 mg / m² or the local area exceeds 20 mg / m², it is determined to be insufficient cleaning; if substrate oxidation or melting marks appear, it is determined to be over-cleaning; and feeding the determination results back to the artificial intelligence decision control module in the form of gradients to trigger the online fine-tuning or offline retraining process of model parameters.

[0013] A high-speed steel strip laser selective cleaning and degreasing system based on UV / NIR detection and AI control includes: an oil film multispectral sensing module, which is set upstream of the steel strip cleaning production line to acquire the original multimodal spectral data stream of the surface of the high-speed steel strip in real time and non-contact. The oil film quantization and analysis module is electrically connected to the oil film multispectral sensing module. It is used to receive the original multimodal spectral data stream and convert it into a real-time two-dimensional quantized oil film thickness map through the built-in data fusion and quantization analysis algorithm. The artificial intelligence decision control module has its data input end connected to the oil film quantification and analysis module and a steel strip speed sensor. The module is equipped with a pre-trained artificial intelligence laser parameter decision model, which is used to generate a laser cleaning operation parameter matrix based on the received oil film thickness map and real-time steel strip speed, and further parse the matrix into a specific high-speed laser control command sequence. The high-speed laser selective area execution module is located downstream of the oil film multispectral sensing module and receives control commands from the artificial intelligence decision control module. The module includes a high-power pulse laser, a beam shaping and transmission system, a high-speed two-dimensional galvanometer scanning system, and an exhaust gas extraction system, and is used to perform precise energy projection and cleaning operations on a designated area of ​​the steel strip surface according to the commands. The cleaning effect verification and model adaptive optimization module is located downstream of the high-speed laser selective execution module. It is used to collect residual oil film data on the surface of the steel strip after cleaning and compare it with a preset target to generate an error signal. The error signal is fed back to the artificial intelligence decision control module for continuous iterative optimization of its internal model.

[0014] Preferably, the oil film multispectral sensing module includes: An ultraviolet light excitation unit is composed of multiple ultraviolet light-emitting diodes arranged closely along a straight line perpendicular to the direction of movement of the steel strip. Its emission wavelength is 365 nanometers, and the light is focused into a uniform bright line spanning the full width of the steel strip by a cylindrical mirror. An ultraviolet fluorescence acquisition unit consists of a high-speed linear array complementary metal-oxide-semiconductor camera and a front-end bandpass filter. The field of view of the camera is precisely aligned with the ultraviolet light line. The center wavelength of the bandpass filter is 450 nanometers and the bandwidth is 40 nanometers. A near-infrared detection unit consists of a near-infrared broadband light source, a beam splitter, and a high-speed linear array indium gallium arsenide camera. The light emitted by the broadband light source is shaped by a cylindrical mirror and then illuminates the surface of the steel strip. The reflected light is guided into the indium gallium arsenide camera through the beam splitter. The bandpass filter in front of the camera has a center wavelength of 1,720 nanometers and a bandwidth of 50 nanometers. A synchronous control and encoder interface is provided to receive pulse signals from the rotary encoder on the steel belt drive roller, and to strictly synchronize the flashing of the light source and the exposure acquisition of the camera based on these signals.

[0015] The beneficial effects of this invention are: This technical solution realizes the transformation from full-area indiscriminate cleaning to selective area precise cleaning based on real-time detection. Laser energy is only applied to the area where there is an oil film and is distributed as needed according to the thickness of the oil film, which greatly reduces the total energy consumption of the system and reduces equipment investment and operating costs. By constructing a multimodal sensing system of ultraviolet fluorescence and near-infrared absorption and combining it with a data fusion algorithm, high-precision and quantitative online measurement of oil film thickness distribution was achieved. This solved the problem that single detection methods are easily affected by oil type and steel strip surface condition, and provided a reliable data foundation for precise control. By introducing an artificial intelligence decision-making model, the nonlinear mapping relationship between the complex oil film distribution and laser parameters is modeled, realizing the collaborative optimization and instantaneous decision-making of laser power, scanning speed and spot morphology under high-speed dynamic conditions. This ensures the uniformity and consistency of the cleaning effect and effectively avoids welding defects caused by under-cleaning and thermal damage to the substrate caused by over-cleaning. The entire process does not use any chemical cleaning agents and does not generate wastewater or waste liquid. The generated oil and gas waste can be collected and treated uniformly through an integrated extraction system, which fully meets the environmental protection requirements of modern industrial green manufacturing. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system structure of a high-speed steel strip laser selective cleaning and degreasing method based on UV / NIR detection and AI control according to the present invention. Detailed Implementation

[0017] Reference Figure 1A laser selective cleaning and degreasing method for high-speed steel strip based on UV / NIR detection and AI control, the method comprising the following steps: S1. A multispectral sensing array is deployed along the width direction of the high-speed steel strip. The multispectral sensing array synchronously collects fluorescence response intensity data of the surface of the high-speed steel strip under preset ultraviolet light excitation and reflection and absorption data of preset near-infrared spectrum, thereby obtaining the original multimodal spectral data stream characterizing the oil film distribution state.

[0018] S2, the original multimodal spectral data stream is input into a data fusion and quantization analysis model. The data fusion and quantization analysis model is based on a preset oil film fluorescence characteristic calibration curve and a near-infrared absorptivity-thickness calibration model. The fluorescence response intensity data and the near-infrared reflection absorption data are fused at the pixel level to generate a real-time, high-resolution two-dimensional quantized oil film thickness map. The map accurately characterizes the oil film mass density of each micro-element region on the surface of the high-speed steel strip.

[0019] S3, the two-dimensional quantitative oil film thickness map and the real-time running speed value of the high-speed steel belt obtained by the speed encoder are used as inputs and fed into a pre-trained artificial intelligence laser parameter decision model; the artificial intelligence laser parameter decision model, based on the input oil film thickness distribution and steel belt speed, infers in real time and outputs a laser cleaning operation parameter matrix that is precisely corresponding to the oil film thickness map in spatial coordinates.

[0020] S4, the laser cleaning operation parameter matrix is ​​transmitted to a laser control command generation unit, which parses each parameter set in the matrix into a specific laser control sequence for a specific spatial coordinate point. The laser control sequence defines the target output power, pulse frequency, pulse width, and dwell time of the laser beam at that coordinate point.

[0021] S5, the laser control sequence is sent to a high-speed laser selective area execution system, which drives its internal high dynamic response galvanometer scanning system and laser power modulation system to precisely project laser pulse energy with specific parameters onto the oil-containing area on the surface of the high-speed steel strip marked by the oil film thickness map, and performs vaporization stripping and cleaning of the oil film. For areas where the oil film thickness is lower than the preset cleaning threshold, no laser energy is applied.

[0022] S6. Downstream of the high-speed laser selective area execution system, a cleaning effect verification array is deployed. The structure and working principle of the verification array are the same as those of the multispectral sensing array. It is used to generate a residual oil film thickness map after cleaning. A model adaptive optimization unit compares the residual oil film thickness map with the preset cleaning quality target value and generates an error feedback signal. The error feedback signal is used to fine-tune the artificial intelligence laser parameter decision model online or retrain it offline to continuously optimize the cleaning strategy and ensure the long-term stability of the cleaning effect.

[0023] In step S1, acquiring the raw multimodal spectral data stream characterizing the oil film distribution specifically includes: projecting a uniform ultraviolet light band with a wavelength of 365 nm onto the surface of the high-speed steel strip through an ultraviolet linear light source array; under synchronous triggering of the ultraviolet linear light source array, acquiring fluorescence signals generated by the excitation of aromatic hydrocarbon compounds in the oil film through a linear array complementary metal-oxide-semiconductor image sensor configured with a 450 nm center wavelength bandpass filter, forming a fluorescence response intensity matrix; projecting a near-infrared light band with a center wavelength of 1720 nm onto the surface of the high-speed steel strip through a near-infrared linear light source array, the wavelength corresponding to the characteristic absorption peak of carbon-hydrogen bonds in organic matter in the oil film; under synchronous triggering of the near-infrared linear light source array, acquiring near-infrared light signals reflected from the surface of the high-speed steel strip through a linear array indium gallium arsenide image sensor configured with a 1720 nm center wavelength bandpass filter, forming a near-infrared reflection intensity matrix; the fluorescence response intensity matrix and the near-infrared reflection intensity matrix together constitute the raw multimodal spectral data stream.

[0024] The ultraviolet (UV) linear light source array consists of multiple UV light-emitting diodes (LEDs) with a peak wavelength of 365 nm arranged closely along a straight line perpendicular to the direction of the steel strip's movement. The spacing between adjacent light-emitting units does not exceed 0.5 mm to ensure a gapless excitation light band across the entire width of the steel strip. The UV LEDs are powered by a constant current drive circuit, and their pulse activation sequence is precisely controlled by a field-programmable gate array (FPGA) controller based on the steel strip speed encoder signal. The pulse width is set to 10 to 50 microseconds to match high-speed imaging requirements and suppress thermal drift. The linear array complementary metal-oxide-semiconductor (CMOS) image sensor has a pixel count of no less than 4,096 and a line frequency of up to 20 kHz. Its photosensitive surface is focused onto the steel strip surface by a cylindrical lens imaging system, ensuring that each pixel corresponds to a physical width of approximately 0.1 mm on the steel strip surface. The 450 nm bandpass filter has a full width at half maximum (FWHM) of 40 nm to effectively filter out excitation light scattering components and environmental stray light interference, retaining only the oil film fluorescence emission signal.

[0025] The near-infrared linear light source array uses halogen lamps or LED broadband light sources. Its output light is filtered by a monochromator or narrowband interference filter to obtain quasi-monochromatic light with a center wavelength of 1720 nm and a bandwidth of less than 10 nm. This light is then shaped by a cylindrical mirror into an illumination band spanning the full width of the steel strip. The linear array indium gallium arsenide (IGaAs) image sensor has a response band covering 900 to 1700 nm, a pixel count of no less than 2560, and a line frequency of no less than 15 kHz. Its front-mounted 1720 nm bandpass filter has a half-width of 50 nm, used to accurately capture the reflection intensity changes of the oil film at this characteristic absorption wavelength. All light sources and sensors are uniformly scheduled by the same synchronous controller to ensure that the data acquisition time of each line strictly corresponds to the displacement of the steel strip, with a spatial positioning error of less than 0.05 mm.

[0026] In step S2, generating a two-dimensional quantitative oil film thickness map specifically includes: First, normalizing the acquired near-infrared reflectance intensity matrix to convert it into a preliminary oil film absorptivity distribution map; Second, performing dark current correction and flat-field correction on the acquired fluorescence response intensity matrix, and correcting the nonlinear saturation phenomenon of fluorescence intensity caused by excessive oil film thickness according to a preset fluorescence quenching effect compensation algorithm, to obtain a corrected fluorescence intensity distribution map; Finally, the data fusion and quantitative analysis model incorporates a three-dimensional lookup table, which is pre-established by measuring standard samples with oil films of known thickness gradients. Its inputs are the preliminary oil film absorptivity and the corrected fluorescence intensity, and its output is the precise oil film mass density value in milligrams per square meter; by performing a lookup operation on each pixel, the two-dimensional quantitative oil film thickness map is finally generated.

[0027] In step S3, the artificial intelligence laser parameter decision model is a deep convolutional neural network structure, specifically an encoder-decoder architecture. The encoder part consists of cascaded convolutional and pooling layers, responsible for extracting multi-scale spatial features from the input two-dimensional quantized oil film thickness map, capturing the morphology, boundary, and aggregation degree of oil film contamination. The decoder part consists of cascaded upsampling and deconvolutional layers, responsible for restoring the abstract features extracted by the encoder to the resolution of the original input image, and combining it with the input real-time running speed value of the steel strip to generate the laser cleaning operation parameter matrix. The laser cleaning operation parameter matrix is ​​a multi-channel tensor, whose channel dimensions correspond to the average laser power, laser pulse repetition frequency, galvanometer scanning speed, and spot size, respectively.

[0028] In step S4, after receiving the four-channel parameter matrix, the laser control command generation unit first performs coordinate system transformation, mapping the image pixel coordinates to the physical coordinates of the steel strip; then, it performs region segmentation based on the oil film thickness threshold (set to five milligrams per square meter), generating control commands only for regions above the threshold; next, it calculates the optimal scanning path according to the galvanometer dynamics model; finally, it packages the path point sequence with the corresponding power, frequency, and pulse width parameters into a hardware-executable command stream, which is sent to the laser execution system through the fiber optic communication interface.

[0029] In step S5, the selected area laser cleaning specifically includes: using a nanosecond pulse fiber laser with a wavelength of 1064 nanometers as the energy source; the laser control command generation unit calculates an optimal galvanometer scanning path based on the laser cleaning operation parameter matrix. This path prioritizes covering the area with the highest oil film thickness and automatically reduces the scanning speed in the thick oil area to increase the energy density, while completely skipping clean areas where the oil film thickness is below the cleaning threshold; the high-speed dynamic response galvanometer scanning system drives the reflective mirror to deflect at high speed according to the optimal scanning path, guiding the laser beam to accurately "draw" the cleaning trajectory on the high-speed moving steel strip; simultaneously, the acoustic-optical modulator modulates the output pulse sequence of the laser in real time according to the power and frequency commands in the laser cleaning operation parameter matrix, ensuring that the energy deposition at each point of action is precisely matched with the oil film thickness at that point.

[0030] The laser outputs an average power of 1 kilowatt, with pulse width adjustable from 10 to 100 nanoseconds and repetition frequency adjustable from 50 kilohertz to 500 kilohertz. The galvanometer scanning system has a positioning accuracy better than 10 microradians and a maximum scanning speed of no less than 3 kilometers per second. Combined with a dynamic zoom optical element, the spot diameter can be adjusted in real time within the range of 0.2 to 2 millimeters. The acousto-optic modulator has a response time of less than 1 microsecond and a modulation depth greater than 30 decibels, ensuring precise and on-demand pulse energy delivery. The exhaust gas extraction system has an airflow of no less than 500 cubic meters per hour, with a negative pressure dust collection hood tightly attached to the galvanometer's light outlet, effectively capturing oil fumes and carbonized particles generated during cleaning.

[0031] In step S6, the cleaning effect verification array reuses the hardware structure of the upstream sensing module but runs independently, generating a residual oil film thickness map in real time. The model adaptive optimization unit calculates the mean and variance of the residual oil. If the mean exceeds 15 mg / m² or the local area exceeds 20 mg / m², it is determined to be insufficient cleaning; if substrate oxidation or melting marks are found, it is determined to be over-cleaned. The error signal is fed back to the artificial intelligence decision control module in the form of a gradient, triggering online fine-tuning of model parameters (learning rate set to 0.001) or triggering an offline retraining process.

[0032] A high-speed steel strip laser selective cleaning and degreasing system based on UV / NIR detection and AI control includes an oil film multispectral sensing module, an oil film quantification and analysis module, an artificial intelligence decision control module, a high-speed laser selective execution module, and a cleaning effect verification and model adaptive optimization module. The oil film multispectral sensing module integrates an ultraviolet light excitation unit, an ultraviolet fluorescence acquisition unit, a near-infrared detection unit, and a synchronous control and encoder interface. The artificial intelligence decision control module hardware includes a field-programmable gate array (FPGA), a graphics processing unit, and a multi-core central processing unit, which respectively handle high-speed data caching and preprocessing, model inference, and task scheduling. The high-speed laser selective execution module includes a kilowatt-level nanosecond pulse fiber laser, a dynamic zoom and beam shaping system, a micro-radius-level galvanometer scanning system, and a coaxial negative pressure dust collection hood.

[0033] An oil film multispectral sensing module, located upstream of the steel strip cleaning production line, is used to acquire the raw multimodal spectral data stream of the high-speed steel strip surface in real time and non-contactly. An oil film quantification and analysis module, electrically connected to the oil film multispectral sensing module, receives the raw multimodal spectral data stream and converts it into a real-time two-dimensional quantized oil film thickness map using a built-in data fusion and quantization analysis algorithm. An artificial intelligence decision control module, whose data input is connected to the oil film quantification and analysis module and a steel strip speed sensor, contains a pre-trained artificial intelligence laser parameter decision model. This model generates a laser cleaning operation parameter matrix based on the received oil film thickness map and real-time steel strip speed, and further analyzes this matrix into specific parameters. The system includes a high-speed laser control command sequence; a high-speed laser selective area execution module, located downstream of the oil film multispectral sensing module, receives control commands from the artificial intelligence decision control module. This module comprises a high-power pulsed laser, a beam shaping and transmission system, a high-speed two-dimensional galvanometer scanning system, and an exhaust gas extraction system, used to perform precise energy projection and cleaning operations on a designated area of ​​the steel strip surface according to the commands; and a cleaning effect verification and model adaptive optimization module, located downstream of the high-speed laser selective area execution module, used to collect residual oil film data on the steel strip surface after cleaning and compare it with a preset target to generate an error signal. This error signal is fed back to the artificial intelligence decision control module for continuous iterative optimization of its internal model.

[0034] The advantages of this invention are that this technical solution realizes the transformation from full-area indiscriminate cleaning to selective area precision cleaning based on real-time detection. Laser energy is only applied to the area where there is an oil film and is distributed as needed according to the thickness of the oil film, which greatly reduces the total energy consumption of the system and reduces equipment investment and operating costs. By constructing a multimodal sensing system of ultraviolet fluorescence and near-infrared absorption and combining it with a data fusion algorithm, high-precision and quantitative online measurement of oil film thickness distribution was achieved. This solved the problem that single detection methods are easily affected by oil type and steel strip surface condition, and provided a reliable data foundation for precise control. By introducing an artificial intelligence decision-making model, the nonlinear mapping relationship between the complex oil film distribution and laser parameters is modeled, realizing the collaborative optimization and instantaneous decision-making of laser power, scanning speed and spot morphology under high-speed dynamic conditions. This ensures the uniformity and consistency of the cleaning effect and effectively avoids welding defects caused by under-cleaning and thermal damage to the substrate caused by over-cleaning. The entire process does not use any chemical cleaning agents and does not generate wastewater or waste liquid. The generated oil and gas waste can be collected and treated uniformly through an integrated extraction system, which fully meets the environmental protection requirements of modern industrial green manufacturing.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for laser selective cleaning and degreasing of high-speed steel strip based on UV / NIR detection and AI control, characterized in that: include: A multispectral sensing array is arranged along the width direction of the high-speed steel strip. The fluorescence response intensity data of the surface of the high-speed steel strip under preset ultraviolet light excitation and the reflection and absorption data of preset near-infrared spectrum are collected synchronously through the multispectral sensing array to obtain the original multimodal spectral data stream characterizing the oil film distribution state. The original multimodal spectral data stream is input into a data fusion and quantization analysis model. The data fusion and quantization analysis model is based on a preset oil film fluorescence characteristic calibration curve and a near-infrared absorptivity-thickness calibration model. The fluorescence response intensity data and the near-infrared reflection absorption data are fused at the pixel level to generate a real-time, high-resolution two-dimensional quantitative oil film thickness map. The map accurately characterizes the oil film mass density of each micro-element region on the surface of the high-speed steel strip. The two-dimensional quantitative oil film thickness map and the real-time running speed value of the high-speed steel belt obtained by the speed encoder are used as inputs and fed into a pre-trained artificial intelligence laser parameter decision model. The artificial intelligence laser parameter decision model infers and outputs a laser cleaning operation parameter matrix that is precisely corresponding to the oil film thickness map in spatial coordinates in real time based on the input oil film thickness distribution and steel belt speed. The laser cleaning operation parameter matrix is ​​transmitted to a laser control command generation unit. The laser control command generation unit parses each parameter set in the matrix into a specific laser control sequence for a specific spatial coordinate point. The laser control sequence defines the target output power, pulse frequency, pulse width, and dwell time of the laser beam at that coordinate point. The laser control sequence is sent to a high-speed laser selective area execution system. The system drives its internal high dynamic response galvanometer scanning system and laser power modulation system to precisely project laser pulse energy with specific parameters onto the oil-containing area on the surface of the high-speed steel strip marked by the oil film thickness map, and performs vaporization stripping and cleaning of the oil film. For areas where the oil film thickness is lower than the preset cleaning threshold, no laser energy is applied. A cleaning effect verification array is deployed downstream of the high-speed laser selective area execution system. The verification array is used to generate a residual oil film thickness map after cleaning. A model adaptive optimization unit compares the residual oil film thickness map with a preset cleaning quality target value and generates an error feedback signal. The error feedback signal is used to fine-tune the artificial intelligence laser parameter decision model online or retrain it offline to continuously optimize the cleaning strategy.

2. The method for laser selective cleaning and degreasing of high-speed steel strip based on UV / NIR detection and AI control according to claim 1, characterized in that: Obtaining the raw multimodal spectral data stream characterizing the oil film distribution includes: A uniform ultraviolet light band with a wavelength of 365 nanometers is projected onto the surface of the high-speed steel strip through an ultraviolet light linear light source array; under the synchronous triggering of the ultraviolet light source array, a linear array complementary metal-oxide-semiconductor image sensor equipped with a 450-nanometer center wavelength bandpass filter is used to collect the fluorescence signal generated by the excitation of aromatic hydrocarbon compounds in the oil film, forming a fluorescence response intensity matrix. A near-infrared light band with a center wavelength of 1720 nm is projected onto the surface of the high-speed steel strip through a near-infrared linear light source array. Under the synchronous triggering of the near-infrared linear light source array, a linear array indium gallium arsenide image sensor equipped with a 1720 nm center wavelength bandpass filter collects the near-infrared light signal reflected from the surface of the high-speed steel strip, forming a near-infrared reflection intensity matrix. The fluorescence response intensity matrix and the near-infrared reflection intensity matrix together constitute the original multimodal spectral data stream.

3. The method for laser selective cleaning and degreasing of high-speed steel strip based on UV / NIR detection and AI control according to claim 1, characterized in that: The generation of a two-dimensional quantitative oil film thickness map includes: The acquired near-infrared reflectance intensity matrix is ​​normalized and converted into a preliminary oil film absorptivity distribution map according to the Beer-Lambert law. The acquired fluorescence response intensity matrix is ​​then subjected to dark current correction and flat-field correction, and the fluorescence intensity nonlinear saturation phenomenon caused by excessive oil film thickness is corrected according to a preset fluorescence quenching effect compensation algorithm, resulting in a corrected fluorescence intensity distribution map. A built-in three-dimensional lookup table is used to perform a lookup operation on each pixel. The input of the three-dimensional lookup table is the preliminary oil film absorptivity and the corrected fluorescence intensity, and the output is the precise oil film mass density value in milligrams per square meter, ultimately generating the two-dimensional quantitative oil film thickness map.

4. The method for laser selective cleaning and degreasing of high-speed steel strip based on UV / NIR detection and AI control according to claim 1, characterized in that: The artificial intelligence laser parameter decision model is a deep convolutional neural network structure, specifically an encoder-decoder architecture. The encoder part consists of cascaded convolutional layers and pooling layers, used to extract multi-scale spatial features from the input two-dimensional quantized oil film thickness map.

5. The method for laser selective cleaning and degreasing of high-speed steel strip based on UV / NIR detection and AI control according to claim 1, characterized in that: The laser control command generation unit parses the laser cleaning operation parameter matrix into a laser control sequence, including: mapping image pixel coordinates to steel strip physical coordinates; performing region segmentation based on an oil film thickness threshold, and generating control commands only for regions above the threshold.

6. The method for laser selective cleaning and degreasing of high-speed steel strip based on UV / NIR detection and AI control according to claim 1, characterized in that: The model adaptive optimization unit generates error feedback signals by: calculating the mean and variance of residual oil; if the mean exceeds 15 mg / m² or the local area exceeds 20 mg / m², it is determined to be insufficient cleaning; if substrate oxidation or melting marks appear, it is determined to be over-cleaning; the determination results are fed back to the artificial intelligence decision control module in the form of gradients to trigger the online fine-tuning or offline retraining process of model parameters.

7. A high-speed steel strip laser selective cleaning and degreasing system based on UV / NIR detection and AI control according to claims 1-6, characterized in that: It includes: The oil film multispectral sensing module is located upstream of the steel strip cleaning production line and is used to acquire the original multimodal spectral data stream of the surface of the high-speed steel strip in real time and without contact. The oil film quantization and analysis module is electrically connected to the oil film multispectral sensing module. It is used to receive the original multimodal spectral data stream and convert it into a real-time two-dimensional quantized oil film thickness map through the built-in data fusion and quantization analysis algorithm. The artificial intelligence decision control module has its data input end connected to the oil film quantification and analysis module and a steel strip speed sensor. The module is equipped with a pre-trained artificial intelligence laser parameter decision model, which is used to generate a laser cleaning operation parameter matrix based on the received oil film thickness map and real-time steel strip speed, and further parse the matrix into a specific high-speed laser control command sequence. The high-speed laser selective area execution module is located downstream of the oil film multispectral sensing module and receives control commands from the artificial intelligence decision control module. The module includes a high-power pulse laser, a beam shaping and transmission system, a high-speed two-dimensional galvanometer scanning system, and an exhaust gas extraction system, and is used to perform precise energy projection and cleaning operations on a designated area of ​​the steel strip surface according to the commands. The cleaning effect verification and model adaptive optimization module is located downstream of the high-speed laser selective execution module. It is used to collect residual oil film data on the surface of the steel strip after cleaning and compare it with a preset target to generate an error signal. The error signal is fed back to the artificial intelligence decision control module for continuous iterative optimization of its internal model.

8. A high-speed steel strip laser selective cleaning and degreasing system based on UV / NIR detection and AI control according to claim 7, characterized in that: The oil film multispectral sensing module includes: An ultraviolet light excitation unit is composed of multiple ultraviolet light-emitting diodes arranged closely along a straight line perpendicular to the direction of movement of the steel strip. Its emission wavelength is 365 nanometers, and the light is focused into a uniform bright line spanning the full width of the steel strip by a cylindrical mirror. An ultraviolet fluorescence acquisition unit consists of a high-speed linear array complementary metal-oxide-semiconductor camera and a front-end bandpass filter. The field of view of the camera is precisely aligned with the ultraviolet light line. The center wavelength of the bandpass filter is 450 nanometers and the bandwidth is 40 nanometers. A near-infrared detection unit consists of a near-infrared broadband light source, a beam splitter, and a high-speed linear array indium gallium arsenide camera. The light emitted by the broadband light source is shaped by a cylindrical mirror and then illuminates the surface of the steel strip. The reflected light is guided into the indium gallium arsenide camera through the beam splitter. The bandpass filter in front of the camera has a center wavelength of 1,720 nanometers and a bandwidth of 50 nanometers. A synchronous control and encoder interface is provided to receive pulse signals from the rotary encoder on the steel belt drive roller, and to strictly synchronize the flashing of the light source and the exposure acquisition of the camera based on these signals.

Images