Cotton carding machine falling impurity bimodal detection method and self-adjusting device

Abstract

The invention discloses a carding machine falling impurity bimodal detection method and a self-adjusting device, and relates to the technical field of textile machinery. Aiming at the limitation of a traditional gravimetric method in high-speed impurity detection, a bimodal detection technology integrating optical imaging and differential pressure measurement is provided. By optimizing an optical sensor and combining a turn-back type optical lens and a composite light source, it is ensured that impurity images are clearly captured under high-speed cotton flow. Impurity features are extracted by using a YOLO target recognition algorithm, pressure fluctuation in the pipeline is captured based on a differential pressure sensor, and impurity quality is predicted in combination with a neural network. Real-time data processing is realized by means of a field programmable gate array circuit and GPU acceleration, and a mapping model of image data and impurity quality is established. Through localized deployment and a real-time data processing technology, an efficient data processing and linkage adjustment mechanism is constructed, a detection result is linked with blowing-carding process parameters, and high-precision and real-time control is realized through automatic adjustment.

Description

Technical Field

[0001] The present invention relates to the technical field of textile machinery, and particularly to a method for dual-mode detection of waste removal in a carding machine and a self-regulating device. Background Art

[0002] In the spinning process flow, the carding process is the core link in the preparation of fiber materials, and its impurity removal efficiency directly affects the quality of the spun yarn and the stability of subsequent processing. According to the GB 1103.1—2023 standard, natural impurities in cotton are defined as sand, branches, leaves, boll shells, cotton seeds, unginned cotton, etc. mixed in the process of cotton picking; foreign fibers are non-cotton fibers and non-natural color fibers mixed in cotton, such as chemical fibers, hair, plastic films, etc. The coexistence of natural impurities and foreign fibers will lead to an increase in the raw material loss coefficient, cause attenuation of the processing efficiency of the processing process, and deteriorate the quality index of the spun yarn.

[0003] The carding machine licker-in area undertakes more than 60% of the overall impurity removal load. This area realizes hierarchical impurity removal through three gradient impurity removal zones: the first impurity removal zone (feed plate - dust knife) mainly removes impurities with larger mass such as dust and cotton seeds; the second impurity removal zone (dust knife - small dust cage) focuses on removing light impurities with attached fibers on the surface; the third impurity removal zone (small dust cage) is dedicated to the separation of short fibers and fine impurities.

[0004] According to the GB / T 6499—2022 standard, the impurity content rate refers to the percentage of the mass of impurities in the raw cotton sample to the mass of the sample. The current experimental method for the impurity content rate of raw cotton is mainly sampling detection, which relies on an offline electronic balance device for measurement. According to the formula: impurity content rate = (mass of impurities / total mass of the sample) × 100%, this method has a reference error of 0.5 - 1.0 g and cannot distinguish the impurity components, and it is necessary to interrupt the production process for offline sampling, resulting in detection lag, which is essentially in conflict with the modern continuous production line.

[0005] At present, computer vision shows certain advantages in the field of carding waste removal detection, and there are already patents for a number of optical detection schemes, such as: DE102019115138B3; CN112048785A; EP3951033A1; CN117120839A; WO2024104691A1; WO2024104692A1; GB2334532A. However, the above-mentioned existing technologies do not involve three major technical bottlenecks faced by waste removal detection: (1) The high-speed cotton flow (>25 m / s) leads to an increase in image acquisition blur, and conventional area array sensors are difficult to meet the dynamic capture requirements; (2) The feature extraction of micron-sized impurities (<0.5 mm) is limited by traditional image processing algorithms, and the false detection rate is as high as 28% - 35%; (3) There is a lack of real-time data interaction between the detection system and the process execution mechanism, and the parameter adjustment delay exceeds 500 ms, and a closed-loop control cannot be formed. Summary of the Invention

[0006] The object of the present invention is to overcome the deficiencies existing in the prior art. Aiming at the problem of insufficient real-time detection accuracy of waste removal in carding machines, an optical-mechanical dual-modal detection method and a self-regulating device adapted to the detection of high-speed cotton flow environment (>25 m / s) and tiny impurities (>0.5 mm) are provided.

[0007] In the first aspect, the present invention provides a dual-modal detection method for waste removal in a carding machine, including the following steps:

[0008] Step 1: Deploy an optical sensor at the intersection of the dust suction pipelines to capture impurity images on the surface of the cotton flow in real time; the optical sensor adopts a composite light source and analyzes each frame through the YOLO object recognition algorithm to identify the type of impurities and extract the projected area A of the impurities imp =A i ×w i , where A i is the pixel area of the i-th type of impurity, and w i is the category weight coefficient; the projected area A of the dropped cotton cotton and the impurity content rate P imp =A imp / A cotton ;

[0009] Step 2: Set up a multi-differential pressure sensor array at the interception of the dust suction pipelines to measure the change in differential pressure ΔP caused by the dropped cotton flow in the pipeline in real time; combined with the flow velocity v measured by a thermal anemometer, through a neural network to fuse optical data (A imp , A cotton , P imp ) and mechanical data (ΔP, v), predict the impurity mass M imp and the total dropped cotton mass M total , calculate the impurity content rate η = M imp / M total , and realize the continuous quantitative evaluation of the dropped cotton mass;

[0010] Step 3: Deploy an interception device in the dust suction pipeline; the electric actuator dynamically adjusts the interception period according to the dropped cotton mass M total predicted by the neural network and the dropped cotton density D t obtained by the image sensor to optimize the impurity separation efficiency. The interception period T is calculated by the formula T = K·(σ + 1) / (|D t -D t-1 | + ε), where ε = 0.001 is a tiny deviation, D t and D t-1 are the current and previous dropped cotton densities respectively, σ is the standard deviation of the past 30 measurements, K is the adjustment coefficient, when the impurity density increases rapidly, i.e., |D t -D t-1When M > 0.1, the interception period T decreases, and the electric actuator increases the opening and closing frequency to prevent pipeline blockage; when M total > 100g, the system synchronously adjusts the closing duration according to the predicted value; the edge computing unit processes the image and differential pressure data in real time, and sends the closing instruction to the actuator to achieve dynamic synchronization with the change of cotton flow. The closed-loop control module adjusts process parameters such as the distance between the saw-tooth cylinder and the dust removal knife according to the impurity content η, η = M imp / M total × 100%. If abnormal blockage of the interception net is detected, reverse pulse cleaning is performed to ensure continuous operation.

[0011] In a second aspect, the present invention provides a dual-mode self-adjusting device for waste removal of a carding machine, including: a sensing module, an optical module, a computing and control module, and an auxiliary module;

[0012] The sensing module includes an optical sensor and a differential pressure sensor. The optical sensor is installed at the intersection of the dust suction pipelines and is used to collect images of impurities on the surface of the cotton flow; the differential pressure sensor array is arranged along the axial direction of the pipeline to measure the differential pressure ΔP before and after the interception device;

[0013] The optical module includes a polarized light source, an ultraviolet light source, a polarization filter, a UV filter, and a folding mirror. The polarized light source and the ultraviolet light source enhance the capture of impurity signals through the polarization filter and the UV filter, and the folding mirror optimizes the optical path to increase the signal-to-noise ratio to ≥ 20dB;

[0014] The computing and control module includes an edge computing unit and a closed-loop control module. The edge computing unit supports real-time image processing and data fusion, and the closed-loop control module is connected to the carding machine control system for process parameter adjustment;

[0015] The auxiliary module includes an adjustable bracket, a self-cleaning air curtain, and a pulse modulation light source controller. The adjustable bracket supports the adjustment of the detection angle and focal length; the self-cleaning air curtain consists of a micro air compressor and a porous nozzle array and is used to reduce the adhesion of fiber dust; the pulse modulation light source controller synchronizes the light source and the sensor exposure and supports high-speed cotton flow.

[0016] The waste removal dual-mode detection method and self-adjustment device of the present invention optimize the optical sensor, combine a composite light source (wavelength range 700-900nm, main wavelength 850nm) and a folded optical lens to improve the imaging quality and data fusion accuracy, and solve the limitations of the traditional gravimetric method in high-speed waste removal detection. The system accurately extracts impurity features through the YOLO object recognition algorithm, establishes an impurity mass prediction model in combination with differential pressure data, and realizes real-time processing (response time <0.02s) using the synchronization function of the field-programmable gate array (FPGA) circuit and GPU acceleration. Further, by constructing a mechanical sensing unit, capturing the dynamic pressure fluctuations in the pipeline based on a differential pressure sensor, and establishing a mapping model of image data-impurity mass in combination with a neural network. Combining local deployment and real-time data processing technologies, an efficient data processing and linkage adjustment mechanism is constructed to ensure that the system can process detection data in real time and automatically adjust the carding and cleaning process parameters according to the detection results, realizing high-precision and real-time control of the carding and cleaning process parameters.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] 1. Through the optical-mechanical dual-mode detection design, the present invention combines an optical sensor and a high-precision differential pressure sensor to fuse multi-modal data to accurately predict the impurity mass. Compared with the single weight detection of the prior art, the detection accuracy is significantly improved. The traditional technology is limited by single-modal data, and the misdetection rate is as high as 28%-35%. However, through the neural network fusion algorithm of the present invention (training samples ≥100000 groups, the error is reduced to ±1.8%), it can still accurately identify tiny impurities (≥0.5mm) under high-speed cotton flow (≥25m / s), overcoming the limitations of the traditional method in a dynamic environment.

[0019] 2. Through the modular designed interception device (butterfly valve structure, the preferred aperture of the carbon fiber mesh is 0.3mm, and the tensile strength is 850MPa) and the multi-drive actuator (the electric precision is ±0.01mm, and the pneumatic opening and closing time is 80-60ms), rapid replacement and dynamic adjustment are realized, significantly extending the service life of the equipment and reducing the maintenance frequency. Compared with the fixed interception structure of the prior art, the present invention is equipped with an RFID chip to track the mesh parameters, and the self-cleaning air curtain (the preferred injection angle is 15°) periodically blows the mirror surface to reduce the adhesion of fiber dust, improving the adaptability of the equipment under complex working conditions.

[0020] 3. The present invention adopts the field programmable gate array (FPGA) synchronization and GPU acceleration technologies to achieve a system response time of <0.02 s and a sampling frequency of ≥50 Hz, effectively improving the real-time detection performance under a high-speed cotton flow (≥25 m / s). Aiming at the high false detection rate caused by blurred image acquisition in the high-speed cotton flow in the prior art, the present invention improves the imaging quality through a folded-back optical path optimization (installation depth ≤200 mm) and a composite light source (wavelength 700 - 900 nm, main wavelength 850 nm), and combines the data processing accelerated by the GPU to ensure the detection efficiency and accuracy, meeting the requirements of modern continuous production lines.

[0021] 4. The present invention integrates a closed-loop control module, which interacts with the carding machine control system in real time and supports the dynamic adjustment of process parameters. Compared with the defect of the lack of real-time data interaction (adjustment delay >500 ms) in the prior art, the carding and cleaning efficiency is significantly improved. The traditional technology only provides detection results and cannot form a closed-loop optimization. However, the present invention realizes intelligent control through an edge computing unit and a self-diagnosis mechanism (automatically alarm and clean under abnormal working conditions), reduces the loss of high-quality fibers, and improves the flexibility and stability of the spinning process. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic installation layout diagram of the carding machine waste removal dual-mode detection self-adjustment device provided by the present invention.

[0023] Figure 2 It is a schematic diagram of the carding machine licker-in waste removal area provided by the present invention.

[0024] Figure 3 It is a schematic diagram of the optical sensor functional module provided by the present invention.

[0025] Figure 4 It is an exploded schematic diagram of the internal components of the optical sensor provided by the present invention.

[0026] Figure 5 It is a schematic mechanical structure diagram of the dust suction pipe interception device provided by the present invention.

[0027] Figure 6 It is a flowchart of the carding machine waste removal dual-mode detection method provided by the present invention.

[0028] Figure 7 It is a flowchart of the dual-mode data acquisition program in the carding machine waste removal dual-mode detection method provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] A carding machine waste removal dual-mode detection method includes the following steps:

[0030] Step 1: Deploy an optical sensor at the intersection of the dust suction pipelines to capture the images of impurities on the surface of the cotton flow at the outlet of the card doffer in real time. The optical sensor uses a composite light source (600 - 900nm, with a main wavelength of 850nm), and through the YOLO object recognition algorithm, it analyzes each frame to identify the types of impurities (such as sand, branches and leaves, boll shells, cotton seeds), and extracts the projected area A of the impurities imp = A i × w i (unit: m 2 ), where A i is the pixel area of the i-th type of impurity, and w i is the category weight coefficient; the projected area A of the doffer cotton (unit: m 2 ) and the impurity rate P of the image imp = A imp / A cotton .

[0031] Step 2: Set up a multi-differential pressure sensor array at the interception of the dust suction pipelines to measure the change in differential pressure ΔP (unit: Pa) caused by the doffer flow in the pipeline in real time. Combining with the flow velocity c measured by the thermal anemometer, through the neural network to fuse the optical data (A imp , A cotton , P imp ) and mechanical data (ΔP, v), predict the impurity mass M imp (unit: g) and the total doffer mass M total (unit: g), calculate the impurity content η = M imp / M total , and realize the continuous quantitative evaluation of the doffer mass.

[0032] Step 3: Deploy an interception device (carbon fiber mesh with a pore size of 0.3mm) in the dust suction pipeline. The electric actuator dynamically adjusts the interception period according to the doffer mass M total predicted by the neural network and the doffer density D t (defined as the proportion of the doffer in each frame of the image) obtained by the image sensor to optimize the impurity separation efficiency. The interception period T is calculated by the formula T = K·(σ + 1) / (|D t - D t-1 | + ε), where ε = 0.001 is a small deviation, D t and D t-1 are the current and previous doffer densities respectively, σ is the standard deviation of the past 30 measurements, and K is the adjustment coefficient (range 0.01 - 0.05s, calibrated by experiments). When the doffer density increases rapidly (i.e., |D t - D t-1 |> 0.1), the interception period T decreases, and the electric actuator increases the opening and closing frequency to prevent pipeline blockage; when (M total> 100, g), the system synchronously adjusts the closing duration according to the predicted value to ensure the interception efficiency. The edge computing unit processes the image and differential pressure data in real time and sends the closing instruction to the actuator to achieve dynamic synchronization with the cotton flow change. The closed-loop control module adjusts process parameters such as the distance between the saw cylinder and the dust removing knife according to the impurity content η (calculation formula η = M imp / M total × 100%). If an abnormal blockage of the interception net is detected (judged by the change of differential pressure gradient), reverse pulse cleaning is performed to ensure continuous operation.

[0033] Preferably, a YOLO model optimized based on fiber impurity characteristics is locally deployed in the edge computing unit. YOLO is pre-trained through a cotton flow impurity data set (sample size ≥ 100000, including cottonseed hulls, leaf debris, etc.), with a classification accuracy rate ≥ 95% and a positioning error ≤ 0.3 mm. The neural network adopts a multi-layer perceptron (256, 128, 64 neurons) combined with efficient channel attention (ECA), with the training error controlled within ± 2% and the sampling frequency ≥ 50 Hz. The model is adapted to materials such as cotton and hemp through transfer learning, with a classification accuracy rate ≥ 95% and an impurity positioning error ≤ 0.3 mm.

[0034] A dual-mode detection self-adjusting device for waste removal in a carding machine, comprising: a sensing module (optical sensor, differential pressure sensor), an optical module (polarized light source, ultraviolet light source, polarizing filter, UV filter, folding mirror), a computing and control module (edge computing unit, closed-loop control module), and an auxiliary module (adjustable bracket, self-cleaning air curtain, pulse modulation light source controller).

[0035] In the sensing module, the optical sensor (resolution 3840 × 2160, sampling rate 120 fps) is installed at the interception point of the dust suction pipeline to collect impurity images on the surface of the cotton flow; the differential pressure sensor array (accuracy ± 0.25% FS, spacing 10 cm) is arranged along the axial direction of the pipeline to measure the differential pressure ΔP before and after the interception device.

[0036] The optical module uses a polarized light source (wavelength 700 - 900 nm, main wavelength 850 nm) and an ultraviolet light source (wavelength 365 nm), enhances the capture of impurity signals through a polarizing filter and a UV filter, and optimizes the optical path with a folding mirror (installation depth ≤ 200 mm) to increase the signal-to-noise ratio ≥ 20 dB.

[0037] The computing and control module includes an edge computing unit and a closed-loop control module. The edge computing unit supports real-time image processing and data fusion, and the closed-loop control module is connected to the carding machine control system for process parameter adjustment.

[0038] In the auxiliary module, the adjustable bracket supports the adjustment of the detection angle (0° - 90°) and focal length (50 - 500 mm); the self-cleaning air curtain consists of a micro air compressor and a porous nozzle array (spray angle 15°, periodic pulse purging), reducing the adhesion of fiber dust; the pulse modulation light source controller synchronizes the light source and the sensor exposure (frequency ≥ 50 Hz), supporting high-speed cotton flow (≥ 25 m / s).

[0039] Preferably, the interception device uses a carbon fiber mesh (aperture 0.3 mm, tensile strength 850 MPa), which can be quickly replaced through an aviation aluminum alloy slot, and is equipped with an RFID chip to record the mesh parameters and service life. The actuator is designed as a butterfly valve structure, including an electric drive module (repeated positioning accuracy ±0.01 mm, thrust 300 N) and a pneumatic drive module (opening and closing time 60 - 80 ms), and the interface complies with the cylinder installation standard and the waterproof electrical standard.

[0040] Preferably, a folding mirror is used to install the optical sensor on the side wall or top of the pipeline, reducing the equipment occupation space (installation depth ≤ 200 mm); supporting multi-sensor segmented deployment, covering a pipeline width ≥ 1.2 m, each sensor independently detects a local area, and supports the detection of the global impurity distribution.

[0041] Preferably, the optical module configures a polarized light source (wavelength 700 - 900 nm, main wavelength 850 nm) to vertically incident, uses the transmitted light to penetrate the fiber layer, and combines an orthogonal polarization filter to receive the transmitted signal; the ultraviolet light source (wavelength 365 nm) excites the fluorescence signal of non-translucent impurities, filters out the environmental light interference through a UV filter, and improves the detection accuracy.

[0042] Preferably, the self-cleaning air curtain periodically purges the mirror surface with compressed air (spray frequency 1 - 2 Hz), extending the maintenance cycle; the adjustable bracket enables rapid calibration to ensure installation flexibility.

[0043] In order to make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with the attached drawings.

[0044] Embodiment 1

[0045] This embodiment provides a self-regulating device for detecting impurities in carding waste in a dual-mode manner. By integrating an optical sensor 1 and a fiber impurity interception device 2, the on-line detection of impurities in high-speed cotton flow is realized. A composite light source 102 and a folding optical path 103b are adopted to optimize the accuracy of impurity image acquisition and quality estimation.

[0046] As Figures 1-5As shown in the figure, a dual-mode detection self-adjusting device for waste removal in a carding machine includes an optical sensor 1 and a dust suction pipe interception device 2. The dust suction pipe 4 adopts an orthogonal pipe structure, where the horizontal dust suction pipe is arranged below the impurity separation area formed by the licker-in 3 and the dust knife 5, and the vertical dust suction pipe is connected to the central dust suction system. The optical sensor 1 is installed at the intersection of the horizontal and vertical dust suction pipes 4 through a flange positioning structure for collecting the free waste flow in the horizontal pipe 4. The dust suction pipe interception device 2 is installed in the vertical pipe through a clamp, and it has the dual functions of impurity interception and dynamic pressure difference sensing.

[0047] As Figure 3 and Figure 4 shown in the figure, the optical sensor 1 includes a self-cleaning air curtain unit 101, a composite light source module 102, and an image acquisition module 103. The self-cleaning air curtain unit 101 consists of a micro air compressor 101a and a porous nozzle array 101b. The axis of the nozzle forms a 15° injection angle with the mirror surface of the image acquisition module 103, and dynamic pulse purging is achieved through a controller. The composite light source module 102 adopts a coaxial ring layout, including a polarized light source 102a, an ultraviolet light source 102b, and a light source driver. The image acquisition module 103 includes a band-pass filter 103a, a folded optical lens 103b, and a CMOS sensor 103c. Specifically, the self-cleaning air curtain unit 101 is used to clean the fiber dust on the image acquisition module 103, and it purges the mirror surface periodically through compressed air. The composite light source module 102 is used to provide clear optical contrast to enable the image acquisition module 103 to effectively capture impurity signals. The image acquisition module 103 is responsible for collecting waste image information, and the folded optical lens 103b effectively shortens the straight-line distance between the CMOS sensor 103c and the focus. The image processing unit uses a locally deployed YOLO model to achieve impurity classification and impurity content calculation.

[0048] As Figure 5 shown in the figure, the dust suction pipe interception device 2 includes a carbon fiber interception net 201, a pressure sensor 202, and an interception net actuator 203. The carbon fiber interception net 201 is opened and closed by the interception net actuator 203. The interception net actuator 203 adopts a butterfly valve structure to achieve the opening and closing control of the interception net through rapid rotation. During operation, the carbon fiber interception net 201 is closed to intercept fiber impurities, while the air is not affected. The pressure sensor 202 measures the pressure change of the carbon fiber interception net 201 in real time to obtain the waste mass data per unit time. The optical sensor 1 detects the image information on the carbon fiber interception net 201.

[0049] When the carding machine is working, the licker-in 3 makes the cotton tufts loose and separates the impurities from the fibers through high-speed hitting and carding, uses the centrifugal force to throw the impurities to the outside, and the dust knife 5 installed at a specific position and angle below the licker-in, with the assistance of the air flow (25 - 30 m / s), guides the impurities into the dust suction pipe 4. The optical sensor 1 captures the optical images of the impurities on the surface of the falling cotton flow in real time, and the dust suction pipe interception device 2 measures the differential pressure change caused by the falling cotton flow in the pipe in real time.

[0050] The calculation and control module includes an edge computing unit, an analog-to-digital conversion module, a controller, a power supply module, a touch display module, a data transmission module, and a program download module; the analog-to-digital conversion module converts the signals of the sensors into digital signals that can be read by the controller; the power supply module is used to supply power to the controller; the touch display module is used for program mode selection and system parameter display; the data transmission module is used to transmit the data measured by the sensors and the controller instructions; the program download module is used for subsequent software upgrades.

[0051] As Figure 6 and Figure 7 shown, a method for dual-modal detection of waste removal in a carding machine includes the following steps.

[0052] Step S1, system initialization: Input the thresholds for waste removal in the carding machine through the touch display screen, including the upper limit of impurity content (set to 5%, based on the economic threshold of 2% - 5% in the textile industry) and the upper limit of the falling cotton quality (100 g).

[0053] Step S2, start judgment: The touch liquid crystal display screen shows two controls, "start" and "exit". If the "start" control is pressed, it enters step S3; if the "exit" is pressed, it enters step S7.

[0054] Step S3, dual-modal data acquisition of the waste removal detection program: This program includes three sub-programs, namely the image detection sub-program, the pressure detection sub-program, and the interception device control sub-program; these three sub-programs are carried out synchronously.

[0055] Step S3.1, image detection sub-program: The optical sensor captures the optical images of the impurities on the surface of the falling cotton flow in real time (resolution 3840×2160, sampling rate 120 fps), analyzes frame by frame through the YOLO object recognition algorithm, identifies the types of impurities (such as sand, branches and leaves, boll shells, cotton seeds), and extracts the projected area A imp (unit: m 2 ) and the projected area A cotton (unit: m 2 ) of the falling cotton, and calculates the impurity content rate P imp =A imp / A cotton .

[0056] Step S3.2, Pressure Detection Subroutine: The fiber impurity pipeline interception device measures the differential pressure change ΔP (unit: Pa, accuracy ±0.25% FS) caused by the falling cotton flow in the pipeline in real time. Through neural network fusion of the image data (A imp 、A cotton 、P imp ) and pressure data (ΔP, flow velocity v) in Step S3.1, predict the impurity mass M imp (unit: g) and the total falling cotton mass M total (unit: g) per unit time, and calculate the impurity content η = M imp / M total . Among them, the flow velocity v (unit: m / s, accuracy ±0.1 m / s) is measured by a thermal anemometer. The neural network model adopts a multi-layer perceptron structure (three hidden layers, 256, 128, 64 neurons, with an efficient channel attention mechanism ECA). The training samples are ≥100000 groups, and the error is controlled within ±1.8%.

[0057] Step S3.3, Interception Device Control Subroutine: Based on the impurity density D t (defined as the proportion of falling cotton in each frame of image) captured by the optical sensor and the falling cotton mass M total and impurity content η predicted in Step S3.2, control the pneumatic actuator to realize the dynamic opening and closing of the interception net. The interception period T (unit: s) is calculated by the formula T = K·(σ + 1) / |D t - D t-1 + ε|, where D t and D t-1 are the current and previous falling cotton densities respectively, σ is the standard deviation of the past 30 measurements, K is an adjustment coefficient (range 0.01 - 0.05 s, experimentally calibrated), and ε = 0.001 is a small deviation correction value to prevent the denominator from being zero.

[0058] Step S4, Parameter Adjustment Judgment: If the impurity content η predicted in Step S3.2 is greater than the set threshold of 5% or the falling cotton mass M total is greater than the set value of 100 g, then enter Step S5; if both are less than the set value, then jump to Step S6.

[0059] Step S5, Self-Adjustment of Waste Removal Control Parameters: Based on the detection data returned by Step S3.1 and Step S3.2, adjust the relative position and angle between the saw-tooth cylinder and the dust removal knife; the above data is transmitted to the controller through the data transmission module and displayed on the touch liquid crystal display screen.

[0060] Step S6, Display System Parameters: Display the set values of the relative position and angle between the current saw-tooth cylinder and the dust removal knife through the touch display screen.

[0061] Step S7. Determine whether the waste removal is completed: If the predicted waste cotton quality M in Step S3.2 total is less than the minimum set value (e.g., 5 g, experimentally calibrated) within 300 consecutive seconds, it is determined that there is no cotton input in the cotton conveying pipeline, and Step S8 is entered; if it is greater than the set value, return to Step S3; in addition, the touch display screen is provided with "pause" and "emergency stop" controls: if "pause" is pressed, the system pauses and waits for the operator's instruction; if "emergency stop" is pressed, directly enter Step S8; if there is no operation, return to Step S3.

[0062] Step S8. End: Exit.

[0063] Embodiment 2

[0064] This embodiment provides a dual-mode self-regulating device for waste removal in a carding machine based on multi-sensor fusion and neural network prediction, aiming to improve the accuracy and real-time performance of impurity quality prediction by fusing optical images and differential pressure data. Different from Embodiment 1, this embodiment introduces a multi-layer perceptron neural network and combines FPGA and GPU acceleration technologies to achieve synchronous data acquisition and processing.

[0065] As Figures 1-5 shown, a dual-mode detection and self-regulating device for waste removal in a carding machine, based on multi-sensor fusion technology, combines an optical sensor 103c and a differential pressure sensor 202, and uses a neural network model to comprehensively analyze optical and pressure data, thereby predicting the waste removal rate in the cotton flow (defined as the percentage of impurity mass in the total waste cotton mass). This system aims to overcome the limitation of the traditional gravimetric method in being unable to real-time monitor the waste removal quality in a high-speed flow field and achieve the precise coupling of image data and quality information.

[0066] Specifically, the optical sensor 1 is installed upstream of the carbon fiber interception net 201, coaxial with the transverse dust suction pipeline 4, covering the upstream impurity area of the fiber. The sensor uses a composite light source 102 (wavelength range 700 - 900 nm, main wavelength 850 nm), which can penetrate the fiber layer to detect shallow impurities and provide uniform illumination. The optical sensor 1 uses a multi-spectral image sensor 103c with a resolution of 3840×2160 pixels, supporting 120 fps high-speed sampling to ensure the image clarity in the fast cotton flow. The interception device is located in the pipeline and uses a carbon fiber interception net 201 with an aperture of 0.3 mm to intercept the smallest impurities but allow the air flow to pass through.

[0067] To measure the mass change on the interception device 2, the interception device is configured with a differential pressure sensor array 202, and four groups of sensors are arranged equidistantly along the axial direction of the pipeline with a spacing of 10 cm and an accuracy of ±0.25% FS. The sensors monitor the differential pressure change before and after the interception device, and combine with an air flow velocity sensor (thermal anemometer, accuracy ±0.1 m / s) to separate the pressure effects of cotton flow interception and air flow, providing mass calculation data.

[0068] Data acquisition uses a field-programmable gate array (FPGA) to achieve nanosecond-level synchronous control, ensuring the precise alignment of optical and pressure data. The image sensor 103c captures cotton flow images at a resolution of 3840×2160 pixels and a sampling rate of 120fps. After removing noise through the dark current correction algorithm, the images are input into the YOLO (You Only Look Once) object recognition model to extract the impurity projection area A imp (unit: m 2 ), the waste projection area A cotton unit: m 2 ), and their ratio P imo =A imp / A cotton , and the recognition confidence S imp (range 0 - 1). Meanwhile, the differential pressure sensor array 202 (accuracy ±0.25% FS) measures the differential pressure ΔP (unit: Pa) before and after the interception device 2. The data is preprocessed by Kalman filtering (filter gain 0.3) to eliminate short-term fluctuations, and environmental correction is performed in combination with the flow velocity v (unit: m / s, accuracy ±0.1m / s) and temperature T (unit: °C) measured by the thermal anemometer. The temperature correction formula is P corr =P raw +C×(T0 / T), where C is the sensor coefficient and T0 = 25°C.

[0069] The mass inversion model uses a neural network to comprehensively process optical and pressure data and directly predicts the impurity mass M imp (unit: g). The network input features include: optical data (A imp , A cotton , S imp ), pressure data (ΔP, v), and temperature T, and the output is M imp . The mass inversion model adopts a multi-layer perceptron structure, including three hidden layers (256, 128, and 64 neurons in each layer), and embeds an efficient channel attention (ECA) mechanism to enhance the response to key features (such as the flow velocity v and the impurity ratio P imo) Sensitivity. The neural network is trained with standard cotton flow samples containing cotton flow data with known impurity masses. The loss function is mean squared error (MSE). After training, the error on the validation set is controlled within ±1.8%. The number of samples is not less than 100,000 groups, covering flow velocities of 5 - 30 m / s and impurity types (such as cottonseed hulls, leaf debris). Experimental verification includes test conditions with different flow velocities and temperature ranges. To meet the real-time requirement, network inference is accelerated on the GPU, and the image processing time per frame is about 0.01 seconds. Combined with FPGA synchronization, the total system response time is controlled within 0.02 seconds. Experiments show that the model exhibits high robustness under different working conditions, with significant improvement compared to the single-mode error of traditional linear formulas (optical ±6.2%, pressure ±8.5%).

[0070] The system includes temperature and humidity compensation modules, and humidity correction is based on humidity sensor data. Daily zero calibration and weekly span calibration (using standard test blocks) ensure long-term stability. Error analysis shows that in the range of 0 - 100 g, the absolute error is ±1.2 g and the relative error ≤1.5%; in the range of 100 - 500 g, the absolute error is ±5.8 g and the relative error ≤1.2%. The number of test samples is not less than 50 groups, and the conditions include different flow velocities and temperature ranges.

[0071] Embodiment 3

[0072] This embodiment provides a dual-mode detection method for waste removal in a carding machine, aiming to calculate the impurity content rate through single-mode mass estimation and data fusion. Compared with Embodiment 2, the feature of this embodiment is that it combines the impurity projection area identified by the YOLO algorithm and differential pressure data to estimate the total mass, optimizes the estimation accuracy of impurity mass through a weighted fusion model, and introduces a self-diagnosis mechanism to monitor data deviation, improving the robustness of the system in high-speed cotton flow.

[0073] Combined with the attached Figures 1-5 , a dual-mode detection method for waste removal in a carding machine combines the data collected by the image sensor 103c and the data of the differential pressure sensor 202 to estimate the mass of impurities in the cotton flow and calculate the waste removal rate (defined as the percentage of impurity mass in the total intercepted mass). The algorithm process starts with the data acquisition and preprocessing stage. The image sensor 103c captures cotton flow images at a high sampling rate of 3840×2160 pixel resolution and 120 fps, and then processes the images through the YOLO (You Only Look Once) object recognition algorithm. YOLO is pre-trained on a cotton flow impurity dataset (such as cottonseed hulls, leaf debris) and a cotton mass region dataset to real-time identify the impurity projection area A imp and the projection area A cotton of the fallen cotton, and calculate the impurity content rate P imp =A imp / A cotton, approximating the proportion of approximate reaction impurities in the actual waste cotton. The image data is corrected for dark current to remove image noise and ensure data reliability. At the same time, four groups of sensors (spacing 10 cm, accuracy ±0.25% FS) of the differential pressure sensor 202 array are arranged along the axial direction of the pipeline to measure the differential pressure ΔP before and after the interception device 2. raw . The differential pressure data is smoothed by Kalman filtering (filtering gain 0.3), and the temperature correction formula ΔP corr = ΔP raw × (T0 / T) (T0 = 25 °C) is applied to adjust to the standard conditions, and the flow velocity v (accuracy ±0.1 m / s) measured by the thermal anemometer is used as the dynamic input.

[0074] In the single-modal mass estimation stage, the image sensor 103c estimates the impurity mass based on the YOLO recognition result. Using the projected area A of the impurity imp and the experimentally calibrated average impurity density ρ imp (such as about 0.46 g / cm for cotton husks 3 , about 0.15 g / cm for cotton branches 3 , 0.64 g / cm for leaves 3 ), the optically estimated impurity mass is M opt = k opt · A imp · ρ imp , where k opt is the calibration coefficient (adjusting the deviation between the projected area and the actual mass through experiments), which needs to be determined by training with standard samples. The differential pressure sensor 202 detects the total mass M total = k press · ΔP corr · v, where k press is the pressure calibration coefficient (unit: g·s / Pa·m), determined by experiments with standard cotton flow samples, reflecting the correlation between the differential pressure and the total mass of the intercepted objects (cotton fibers plus impurities). The pressure data provides an independent measurement of the total mass, complementary to the dynamic proportional estimation of the image data.

[0075] In the fusion calculation stage, the single-modal estimation results are weighted and combined, and the impurity mass is calculated by the formula M imp = w opt · M opt + w press · M total · η imp , where M opt is the impurity mass estimated by the image, η imp = (∑A imp · ρ imp ) / (∑A imp · ρ imp + A cotton · ρ cotton) is the quality ratio adjusted based on density, ρ cotton is the density of cotton fibers, ρ imp is the density of impurities. The weight is adjusted according to the flow rate and YOLO confidence: w opt = 0.5 + 0.2·(v / v max ) + 0.2·S imp , w press = 1 - w opt , where v max = 30m / s, S imp is the average confidence of YOLO recognition (range 0 - 1). The impurity removal rate is then calculated as R imp = M imp / M total ×100%. To meet the real-time requirement, the FPGA synchronizes the image and pressure data with nanosecond-level precision. The YOLO algorithm processes one frame of image every 0.01 seconds under GPU acceleration, and the total system response time is controlled within 0.02 seconds. The fusion result is verified through experiments, and the single-modal error (optical ±6.2%, pressure ±8.5%) is reduced to ±1.8%, showing a significant improvement.

[0076] The algorithm robustness is ensured by the self-diagnosis mechanism, which monitors the deviation of M opt and M total ·η imp , and the calculation formula is e rd = |M opt - M total ·η imp | / M imp . If the deviation exceeds 15%, the system starts the self-diagnosis program to check for abnormal fluctuations in the YOLO recognition confidence or differential pressure data (such as caused by air flow disturbances), and switches to a single reliable mode (such as relying only on optical data) as needed. To adapt to the high-speed cotton flow and coefficient distribution, YOLO needs to be trained to recognize the waste cotton area (A cotton ), ensuring that P imp reflects the dynamic waste cotton state to improve the accuracy of impurity recognition.

[0077] Example 4

[0078] This example provides a multi-light-source optical sensor for cotton and linen fibers, aiming to adapt to the impurity detection of various fiber types such as cotton and linen fibers by optimizing the spectral configuration. Compared with the previous examples, the characteristics of this example lie in multi-light-source optical detection and application expansion. Combining near-infrared and infrared light sources, adopting a dual-light-path switching mechanism and a polarization analysis model, the recognition ability for impurities such as cottonseed hulls and hemp skins is expanded, and the reliability of the system in a high-dust environment is improved through IP68 protection design.

[0079] As Figures 3-4As shown, a multi - light - source optical sensor 1 for cotton and linen fibers combines a dual - spectrum light - source module 102 with an optical detection strategy, aiming to improve the detection accuracy and adaptability of impurities in cotton and linen fibers. The system optimizes for the special properties of cotton fibers and linen fibers respectively through spectral feature extraction and mathematical models, providing efficient and accurate impurity identification and classification.

[0080] The core component of the optical sensor 1 is the multi - spectrum light - source module 102. For the detection of cotton fibers, a near - infrared LED array with a wavelength of 850 nm is used to identify impurities such as cottonseed hulls and leaf debris in the transmission polarization mode. This wavelength can effectively penetrate the epidermal layer of cotton fibers (thickness < 200 μm). Combining with polarization analysis technology, precise detection is carried out by analyzing the relationship between the polarization change caused by the sample and the thickness of the impurities. For the detection of linen fibers, a laser diode with a wavelength of 1700 nm is used to identify impurities such as linen bark and lignin through the reflection confocal mode. Based on the absorption characteristics of lignin in the near - infrared spectrum, this wavelength combines with the normalized difference vegetation index (NDVI). Based on the difference in reflectance between the 1700 nm and 680 nm bands, the identification accuracy of impurities in linen fibers is effectively improved.

[0081] In terms of the optical detection strategy, the system has been specifically optimized. For cotton fibers, the transmission imaging of the 850 nm light source combined with the polarization analysis model can accurately identify impurities such as cottonseed hulls and leaf debris by measuring the change of polarized light. For linen fiber mode, based on the reflection spectral characteristics at a wavelength of 1700 nm, combined with the modified NDVI index, the impurity characteristics such as grass impurities in linen fibers are extracted by interpolating the reflectance at 1700 nm and 680 nm, realizing efficient detection.

[0082] In terms of the mechanical structure, the mechanical structure of this system adopts a dual - optical - path switching mechanism and uses a 45° beam splitter to optimize the optical path design to ensure the stable operation of the system under complex working conditions. To adapt to high - dust environments such as textile factories, the protection level of the system reaches IP68, ensuring the long - term reliability of the equipment.

[0083] Compared with traditional methods, in the detection of cotton fibers by this system, the detection accuracy rates of cottonseed hulls and leaf debris are increased to 98.1% and 93.4% respectively; in the detection of linen fibers, the detection accuracy rates of linen bark fibers and grass - impurity mixtures are increased to 96.8% and 89.7% respectively. In addition, the system has strong adaptability to interference factors (such as the moisture content of cotton fibers, environmental temperature changes, etc.), the false - detection rate is greatly reduced, and high detection stability is achieved through a compensation algorithm.

[0084] Example 5

[0085] This embodiment provides a configurable fiber impurity interception device, aiming to optimize the impurity separation efficiency through flexible hardware design and intelligent control strategies. Different from the previous embodiments, this embodiment adopts a replaceable carbon fiber interception mesh 201 and a multi-drive actuator 203, combined with a dynamic adjustment algorithm based on the image sensor 103c. By real-time monitoring the impurity density and differential pressure changes, the interception period is optimized and the adaptability of the device under complex working conditions is improved.

[0086] As Figure 5 shown, a configurable fiber impurity interception device adopts a modular design and can flexibly adjust the specifications and operation modes of the filter screen according to different impurity types and working environments. The system integrates a standardized carbon fiber interception mesh 201, a multi-drive compatible actuator 203, and an intelligent control strategy based on the image sensor 103c, aiming to improve the fiber impurity interception efficiency, extend the service life of the mesh, and reduce the system energy consumption. Compared with the pipeline flow sensor, the image sensor 103c directly shoots the cotton flow, reducing the impurity escape rate by 62%. It can reliably measure the impurity density, thereby optimizing the dynamic adjustment of the interception period and effectively improving the adaptability of the system under complex working conditions.

[0087] The carbon fiber interception mesh 201 adopts a standardized design to meet the impurity interception requirements for different particle size ranges. The mesh preferably has three models: Model FN03 is suitable for ultra-fine impurities (particle size ≤ 0.5mm), with a pore size of 0.3 ± 0.02mm and a tensile strength of 850MPa; FN05 is suitable for standard working conditions (particle size 0.5 - 1mm), with a pore size of 0.5 ± 0.03mm and a tensile strength of 780MPa; FN10 is suitable for high-flow modes (impurity concentration < 5%), with a pore size of 1.0 ± 0.05mm and a tensile strength of 700MPa. The mesh is quickly replaced through an aviation aluminum alloy card slot structure and is equipped with an RFID chip to record the mesh parameters and service life, ensuring efficient management and accurate tracking.

[0088] The interception mesh actuator 203 adopts a butterfly valve structure, preferably with two driving modes: electric and pneumatic. The electric drive module uses a ball screw module, with a repeat positioning accuracy of ±0.01mm and a maximum thrust of 300N, suitable for precise control. The pneumatic drive module uses a cylinder, with short opening and closing times (80ms and 60ms respectively), suitable for working conditions requiring quick response. All interfaces comply with the cylinder installation standard and the waterproof electrical interface standard, ensuring the stability and long-term reliability of the system under different working conditions.

[0089] In terms of the intelligent control strategy, the system uses the image sensor 103c to directly shoot the cotton flow to obtain the falling cotton density D t(Defined as the proportion of waste cotton in each frame of the image), the image sensor 103c can capture the impurity situation in the cotton flow more accurately. The opening and closing control algorithm of the intercepting net is dynamically adjusted based on the change of impurity density. The interception period T is calculated by the formula T = K·(σ + 1) / (|D t -D t-1 | + ε), where ε = 0.001 is a small value, D t and D t-1 are the current and previous waste cotton densities respectively, σ is the standard deviation of the past 30 measurements, and K is the adjustment coefficient. When the waste cotton density increases rapidly (i.e., |D t -D t-1 | is large), T becomes smaller and the opening and closing frequency of the intercepting net increases to prevent pipeline blockage.

[0090] To cope with abnormal working conditions, the system is provided with multiple protection mechanisms. When the standard deviation σ of the continuously measured images exceeds 0.15 three times in a row, the system automatically triggers an alarm and switches to the safe mode. If it is detected that the intercepting net is abnormally blocked (judged by the abnormal increase of impurity density or the change of pressure difference gradient), the intercepting net actuator 203 executes reverse pulse cleaning to ensure continuous and efficient operation.

[0091] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting impurities in a carding machine with dual modes, characterized in that It includes the following steps: Step 1: Deploy an optical sensor at the intersection of the dust suction pipelines to capture the images of impurities on the surface of the cotton flow in real time. The optical sensor uses a composite light source and analyzes each frame through the YOLO object recognition algorithm to identify the type of impurities and extract the projected area A of the impurities imp = A i × w i , where A i is the pixel area of the i-th type of impurities, and w i is the class weight coefficient; the projected area A of the waste cotton and the impurity content rate P of the image imp = A imp / A cotton ; Step 2: Set up a multi-differential pressure sensor array at the interception of the dust suction pipeline to measure the change in differential pressure ΔP caused by the falling cotton flow in the pipeline in real time; combine the flow velocity v measured by the thermal anemometer, and through a neural network to fuse the optical data (A imp 、A cotton 、P imp ) and mechanical data (ΔP, v) to predict the impurity mass M imp and the total falling cotton mass M total , calculate the impurity content η = M imp / M total , and realize the continuous quantitative evaluation of the falling cotton mass; Step 3: Deploy an interception device in the dust suction pipeline; the electric actuator adjusts the interception period dynamically according to the predicted lint mass M total and the lint density D obtained by the image sensor t , so as to optimize the impurity separation efficiency. The interception period T is calculated by the formula T = K·(σ + 1) / (|D t -D t-1 | + ε), where ε = 0.001 is a small deviation, D t and D t-1 are the current and previous lint densities respectively, σ is the standard deviation of the past 30 measurements, and K is an adjustment coefficient. When the impurity density increases rapidly, i.e., |D t -D t-1 | > 0.1, the interception period T decreases, and the electric actuator increases the opening and closing frequency to prevent pipeline blockage; when M total > 100 g, the system synchronously adjusts the closing duration according to the predicted value; The edge computing unit processes the image and differential pressure data in real time, sends a closing instruction to the actuator, and realizes dynamic synchronization with the cotton flow change; the closed-loop control module adjusts process parameters such as the distance between the saw-tooth cylinder and the dust removing knife according to the impurity content η, η = M imp / M total × 100%; if abnormal blockage of the intercepting net is detected, reverse pulse cleaning is performed to ensure continuous operation.

2. The carding machine waste removal dual-mode detection method according to claim 1, characterized in that In the above step S3, a YOLO model optimized based on fiber impurity characteristics is locally deployed in the edge computing unit. YOLO is pre-trained with a cotton flow impurity dataset, with a classification accuracy of ≥95% and a positioning error of ≤0.3 mm. The neural network uses a multi-layer perceptron combined with an efficient channel attention mechanism, with the training error controlled within ±2% and a sampling frequency of ≥50 Hz. The model is adapted to materials such as cotton and hemp through transfer learning, with a classification accuracy of ≥95% and an impurity positioning error of ≤0.3 mm.

3. An automatic adjustment device for implementing the carding machine waste removal dual-mode detection method according to claim 1 or 2, characterized in that It includes: A sensing module, an optical module, a computing and control module, and an auxiliary module; The sensing module includes an optical sensor and a differential pressure sensor. The optical sensor is installed at the intersection of the dust suction pipelines and is used to collect the images of impurities on the surface of the cotton flow. The differential pressure sensor array is arranged along the axial direction of the pipeline to measure the pressure difference ΔP before and after the interception device. The optical module includes a polarized light source, an ultraviolet light source, a polarization filter, a UV filter, and a folded mirror. The polarized light source and the ultraviolet light source enhance the capture of impurity signals through the polarization filter and the UV filter, and the folded mirror optimizes the optical path to increase the signal-to-noise ratio to ≥20 dB. The computing and control module includes an edge computing unit and a closed-loop control module. The edge computing unit supports real-time image processing and data fusion, and the closed-loop control module is connected to the carding machine control system for process parameter adjustment. The auxiliary module includes an adjustable bracket, a self-cleaning air curtain, and a pulse modulation light source controller. The adjustable bracket supports the adjustment of the detection angle and focal length. The self-cleaning air curtain consists of a micro air compressor and a porous nozzle array and is used to reduce the adhesion of fiber dust. The pulse modulation light source controller synchronizes the light source and the sensor exposure and supports high-speed cotton flow.

4. A waste removal dual-mode self-adjusting device for a carding machine according to claim 3, characterized in that It also includes an interception device using a carbon fiber mesh, which can be quickly replaced through an aviation aluminum alloy slot and is equipped with an RFID chip to record the mesh parameters and service life.

5. A waste removing dual-mode self-adjusting device for a carding machine according to claim 3, characterized in that It also includes an actuator designed as a butterfly valve structure, including an electric drive module for driving an electrical device or a pneumatic drive module for driving a cylinder.

6. The self - regulating device for waste removal in a carding machine with dual - mode according to claim 3, wherein The optical sensor is installed on the side wall or top of the pipeline using a folded mirror, supporting the segmented deployment of multiple sensors, covering a pipeline width of ≥1.2 m. Each optical sensor independently detects a local area and supports the detection of the global impurity distribution.

7. The self-adjusting device for waste removal in a carding machine according to claim 3, characterized in that The optical module is configured with the polarized light source incident vertically, using the transmitted light to penetrate the fiber layer, and combining with an orthogonal polarization filter to receive the transmitted signal. The ultraviolet light source excites the fluorescence signal of non-transparent impurities and filters out the ambient light interference through the UV filter.

8. The self - regulating device for waste discharge in a carding machine according to claim 3, wherein The self-cleaning air curtain periodically purges the mirror surface with compressed air; the adjustable bracket enables quick calibration.

Images