Hollow fiber membrane dry-state online visual detection system based on voxelization multi-resolution clustering

Abstract

The invention relates to a voxelization multi-resolution clustering-based hollow fiber membrane dry-state online visual inspection system, which belongs to the technical field of combined visual inspection and comprises an air source, a pressure self-adaptive regulator, a PE (polyethylene) membrane, a hollow fiber membrane component fixing device, a combined visual system and a computer. The gas source is used for continuously inputting helium or hydrogen into the hollow fiber membrane module to be detected, and compared with other gas molecules, the helium or hydrogen is smaller in molecule and higher in permeability; the pressure self-adaptive regulator is used for self-regulation of air pressure in the hollow fiber membrane module; the PE film is a thin film which can be firmly attached to the end surface of the detected end of the hollow fiber membrane module and has tensile property; the hollow fiber membrane module fixing device is used for fixedly placing a hollow fiber membrane module; the combined vision system comprises a depth sensor and three cameras; and the computer is used for monitoring data on line. A combined visual detection technology is adopted, information acquired by a depth sensor is used as an initial value, three groups of binocular stereoscopic vision systems are combined, a voxelization multi-resolution clustering matching method is adopted to obtain an accurate measurement value, small molecule gas is used as a detection medium, the whole detection process is kept in a dry state, liquid intervention is not needed, and the detection accuracy is high. And the position of the defective membrane wire is positioned on line, so that the false detection problem occurring in off-line detection and wet detection is reduced, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of joint visual detection, and in particular relates to an online visual detection system for dry-state hollow fiber membranes based on voxelized multi-resolution clustering. Background Art

[0002] Hollow fiber membrane is a kind of ultrafiltration membrane and the most mature and advanced form of ultrafiltration technology. Currently, it is basically tested in wet state or offline. The demand for full dry state testing, especially online testing, is becoming more and more urgent.

[0003] A Chinese patent discloses a hollow fiber membrane module integrity detection device, application number: 201510264009.8, the device includes a reactor, an aeration device, a water pump, two pairs of platinum wire probes are installed on the water inlet pipe connected to the water pump, and the two circuits are connected in parallel to a 5V power supply; each pair of platinum wire probes is connected in series with a fixed resistor in the circuit, each resistor is also connected in parallel with a fixed capacitor, and each fixed resistor is connected to the acquisition card through a wire, and the acquisition card collects the voltage at both ends of the fixed resistor and connects to the computer through a data line and judges the integrity of the hollow fiber membrane module according to the change of the online display voltage. According to the electrical principle, the device uses a data acquisition device to collect the voltage change at both ends of the fixed resistor, and judges the integrity of the membrane module according to the change law of the voltage signal. However, the detection method for the integrity of the hollow fiber membrane module mainly adopts a contact measurement method, or can only detect the integrity of a single membrane wire but cannot detect the entire hollow fiber membrane module, the detection cost and the subsequent maintenance cost are high, the detection process may cause secondary damage to the membrane module, and the position of the defective wire in the module cannot be accurately located, and the quality control of the membrane module is difficult.

[0004] A Chinese patent discloses a non-contact hollow fiber membrane integrity detection device, application number: 202311719585.8, which includes a gas transmission device, a film, an optical detection kit, a detection displacement platform and a host computer; after the gas is introduced, a line laser is used to scan the hollow fiber membrane group as a whole, and the defect position coordinates are obtained based on the light spot information reflected back to the camera by the straight line laser. However, although this is a non-contact measurement method, the measurement time is still long, the operation process is cumbersome, and the uneven size of the gas molecules used is difficult to control.

[0005] Therefore, it is necessary to design a fully dry, online detection system to solve the above problems.

[0006] The present invention adopts joint visual detection technology, uses the information obtained by the depth sensor as the initial value, and combines three groups of binocular stereo vision systems to adopt the voxelized multi-resolution clustering matching method to obtain accurate measurement values. Small molecule gas is used as the detection medium. The entire detection process remains dry and no liquid intervention is required. The position of the defective membrane filament is located online, which reduces the false detection problem in offline detection and wet detection and improves the detection efficiency. Summary of the invention

[0007] The present invention adopts joint visual detection technology, uses the information obtained by the depth sensor as the initial value, and combines three groups of binocular stereo vision systems to adopt the voxelized multi-resolution clustering matching method to obtain accurate measurement values. Small molecule gas is used as the detection medium. The whole detection process remains dry, and the position of the defective membrane filament is located online, which reduces the false detection problem in offline detection and wet detection and improves the detection efficiency.

[0008] To achieve the above object, the present invention is implemented through the following technical solutions:

[0009] The hollow fiber membrane dry state online visual inspection system based on voxelized multi-resolution clustering is characterized by: comprising an air source, a pressure adaptive regulator, a PE membrane, a hollow fiber membrane assembly fixing device, a joint visual system and a computer; the air source is used to continuously input helium or hydrogen into the hollow fiber membrane assembly to be inspected, and compared with other gas molecules, its molecules are smaller and have higher permeability; the pressure adaptive regulator is used to self-regulate the air pressure in the hollow fiber membrane assembly; the PE membrane is a thin film with tensile properties that can be firmly attached to the end face of the hollow fiber membrane assembly to be inspected; the hollow fiber membrane assembly fixing device is used to fix and place the hollow fiber membrane assembly; the joint visual system includes a depth sensor and three cameras; the computer is used to monitor three-dimensional data online.

[0010] Preferably, the hollow fiber membrane assembly fixing device can fix the hollow fiber membrane assembly horizontally so that the end face of the detection end faces the depth sensor in the joint vision system.

[0011] Preferably, the joint vision system includes a depth sensor and three cameras, wherein cameras C1 and C3 are fixed on the detection displacement platform at a certain angle to the left and right of the same horizontal plane, camera C2 is above the detection platform, shooting the end face of the hollow fiber membrane group at position, and the optical centers of the three cameras are located in the same plane.

[0012] Preferably, the three cameras are C1, C2, and C3, and are combined into three binocular stereo vision systems in pairs. The rotation matrices and translation vectors of the three binocular stereo vision systems are:

[0013] Rotation Matrix Translation Vector <![CDATA[C1、C2]]> <![CDATA[R 12 ]]> <![CDATA[t 12 ]]> <![CDATA[C1、C3]]> <![CDATA[R 13 ]]> <![CDATA[t 13 ]]> <![CDATA[C2、C3]]> <![CDATA[R 23 ]]> <![CDATA[t 23 ]]>

[0014] Preferably, if there is a measured point P, three sets of binocular stereo vision systems are fused to solve the three-dimensional coordinates (x p ,y p ,z p ), that is, when the error F = min(||P-P1||+||P-P2||+||P-P3||) takes the minimum value, point P is the optimal estimate of the actual coordinates, where P is the measured point, P1, P2, and P3 are the measurement values ​​of P by three sets of binocular stereo vision systems, and the conversion relationship between P1, P2, and P3 is P2 = R 12 P1+t 12 、P3=R 13 P1+t 13 =R 23 P2+t 23 ,Right now P3=R 13 P1+t 13 =R 23 P2+t 23 .

[0015] Preferably, the depth sensor and the three sets of binocular stereo vision systems adopt a nonlinear optimization method to solve the mutual spatial transformation relationship T, where where q i are the coordinates of the measured point under the depth sensor.

[0016] Preferably, a voxelized multi-resolution clustering matching algorithm is used to fuse the real-time data obtained by the depth sensor and three sets of binocular stereo vision systems to quickly and accurately remove external points and achieve relocation, that is, first obtain the maximum value X in the three directions of X, Y, and Z. max , Y max , Z max , minimum value X min , Y min , Z min , and then use the maximum value minus the minimum value to calculate the bounding box side length l x , l y , l z , set the side length of the voxel cube cell, and cut it into M, N, and L parts along the three directions of X, Y, and Z coordinate axes. Label each voxel cube as (x, y, z) to get the voxel cube corresponding to each data point Calculate the center of gravity of each point cloud, keep only the center of gravity point cloud of each voxel cube, and remove other point clouds. Among them C ijkis the center of gravity of the voxel cube, and k is the number of point clouds. According to the different values ​​of the side length cell of the voxel cube, different voxel cube center of gravity C will be calculated. ijk In this method, three groups of cell values ​​are set up from large to small, that is, three groups of C ijk , cluster matching is performed layer by layer.

[0017] Preferably, the maximum joint probability (MJP) is used as the basis for judging the matching result. Where F(P i ) is the maximum joint probability, U(P i ) is point P i The set of all neighbors, P j YesP i Points in the neighborhood, P E is the joint probability, P E (P i , P j )=P d (P i ) d (P j )LGC(P i , P j ), where P d is the probability of the feature descriptor, LGC(P i , P j ) is local set consistency.

[0018] Preferably, the pressure adaptive regulator comprises a micro-pressure sensor and an automatic control device, and the pressure in the hollow fiber membrane assembly is automatically balanced by the pressure adaptive regulator during the detection process, and the pressure range is 8000±200Pa.

[0019] Preferably, the PE membrane needs to be tightly attached to the detection end face of the hollow fiber membrane assembly before ventilation. It has a certain degree of stretchability. When gas molecules penetrate from the pores of the membrane fibers, the PE membrane will be lifted up to form a deformation.

[0020] Preferably, the hollow fiber module detection step comprises:

[0021] Step S1, fixing the hollow fiber membrane assembly to a hollow fiber membrane assembly fixing device;

[0022] Step S2, the hollow fiber membrane assembly is connected to an air source and a pressure adaptive regulator;

[0023] Step S3, covering the detection end of the hollow fiber membrane assembly with a PE film;

[0024] Step S4, placing and fixing the hollow fiber membrane assembly horizontally, with the end surface facing the joint vision system;

[0025] Step S5, turning on the joint vision system, transmitting the images received online to the computer for processing to obtain real-time surface change data and the position of the defective film wire;

[0026] Step S6, closing the gas source, the pressure adaptive regulator and the combined visual system;

[0027] Step S7: detection ends.

[0028] Preferably, the basic properties are as follows:

[0029] This method Wet test method Detection time ＜10s 30s General defect detection accuracy 99.7％ 95％ Micro defect detection accuracy 99.3％ 30％ Damage to components none Breeding bacteria Detection method Dry non-contact Wet contact

[0030] The beneficial effects of the present invention are:

[0031] 1. The present invention adopts joint visual detection technology, uses the information obtained by the depth sensor as the initial value, and combines three groups of binocular stereo vision systems to adopt voxelized multi-resolution cluster matching method to obtain accurate measurement values;

[0032] 2. Using small molecule gas as the detection medium, the whole detection process remains dry without the need for liquid intervention;

[0033] 3. The position of defective membrane filaments can be located online, which reduces the false detection problems in offline detection and wet detection and improves the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The present invention is further described in detail below with reference to the accompanying drawings.

[0035] Figure 1 It is a schematic diagram of the process of the present invention.

[0036] Figure 2 It is a schematic diagram of the structure of the present invention.

[0037] Among them: 1. Camera; 2. Depth sensor; 3. Hollow fiber membrane assembly; 4. Air source; 5. Pressure adaptive regulator; 6. PE membrane; 7. Computer; 8. Hollow fiber membrane assembly fixing device; DETAILED DESCRIPTION

[0038] like Figure 2As shown, a hollow fiber membrane dry-state online visual inspection system based on voxelized multi-resolution clustering includes a gas source, a pressure adaptive regulator, a PE membrane, a hollow fiber membrane assembly fixture, a joint visual system and a computer; the gas source 4 is used to continuously input helium or hydrogen into the hollow fiber membrane assembly to be inspected, and compared with other gas molecules, its molecules are smaller and have a higher permeability; the pressure adaptive regulator 5 is used to self-regulate the air pressure in the hollow fiber membrane assembly; the PE membrane 6 is a thin film with tensile properties that can be firmly attached to the end face of the hollow fiber membrane assembly to be inspected; the hollow fiber membrane assembly fixture 8 is used to fix and place the hollow fiber membrane assembly; the joint visual system includes a depth sensor 2 and three cameras 1; the computer 7 is used to monitor three-dimensional data online.

[0039] The operation steps of the online visual inspection system for dry hollow fiber membranes based on voxelized multi-resolution clustering include:

[0040] Step S1, fixing the hollow fiber membrane assembly to a hollow fiber membrane assembly fixing device;

[0041] Step S2, the hollow fiber membrane assembly is connected to an air source and a pressure adaptive regulator;

[0042] Step S3, covering the detection end of the hollow fiber membrane assembly with a PE film;

[0043] Step S4, placing and fixing the hollow fiber membrane assembly horizontally, with the end surface facing the joint vision system;

[0044] Step S5, turning on the joint vision system, transmitting the images received online to the computer for processing to obtain real-time surface change data and the position of the defective film wire;

[0045] Step S6, closing the gas source, the pressure adaptive regulator and the combined visual system;

[0046] Step S7: detection ends.

Claims

1. An online visual inspection system for dry hollow fiber membranes based on voxelized multi-resolution clustering, characterized by: It includes an air source, a pressure adaptive regulator, a PE membrane, a hollow fiber membrane assembly fixing device, a joint vision system and a computer; the air source is used to continuously input helium or hydrogen into the hollow fiber membrane assembly to be inspected, and compared with other gas molecules, its molecules are smaller and have a higher permeability; the pressure adaptive regulator is used to self-regulate the air pressure in the hollow fiber membrane assembly; the PE membrane is a thin film with tensile properties that can be firmly attached to the end face of the hollow fiber membrane assembly to be inspected; the hollow fiber membrane assembly fixing device is used to fix the hollow fiber membrane assembly in place; the joint vision system includes a depth sensor and three cameras; the computer is used to monitor three-dimensional data online.

2. The online visual inspection system for dry state of hollow fiber membrane based on voxelized multi-resolution clustering according to claim 1 is characterized in that: The hollow fiber membrane component fixing device can fix the hollow fiber membrane component horizontally so that the end face of the detection end faces the depth sensor in the joint vision system.

3. The online visual inspection system for dry state of hollow fiber membrane based on voxelized multi-resolution clustering according to claim 1, characterized in that: The joint vision system includes a depth sensor and three cameras C1, C2, and C3. Cameras C1 and C3 are fixed on the detection displacement platform at a certain angle to the left and right of the same horizontal plane. Camera C2 is above the detection platform and shoots the end face of the hollow fiber membrane group at its position. The optical centers of the three cameras are located in the same plane.

4. The online visual inspection system for dry state of hollow fiber membrane based on voxelized multi-resolution clustering according to claim 3 is characterized in that: The three cameras are C1, C2, and C3, which are combined into three binocular stereo vision systems. The rotation matrix and translation vector of the three binocular stereo vision systems are calculated as follows:

5. The online visual inspection system for dry state of hollow fiber membrane based on voxelized multi-resolution clustering according to claims 3 and 4 is characterized in that: If there is a measured point P, the three sets of binocular stereo vision systems are fused to solve the three-dimensional coordinates (x p ,y p ,z p ), that is, when the error F = min(||P-P1||+||P-P2||+||P-P3||) takes the minimum value, point P is the optimal estimate of the actual coordinates, where P is the measured point, P1, P2, and P3 are the measurement values ​​of P by three sets of binocular stereo vision systems, and the conversion relationship between P1, P2, and P3 is P2 = R 12 P1+t 12 、P3=R 13 P1+t 13 =R 23 P2+t 23 ,Right now P3=R 13 P1+t 13 =R 23 P2+t 23 .

6. The online visual inspection system for dry state of hollow fiber membrane based on voxelized multi-resolution clustering according to claims 3, 4 and 5, characterized in that: The depth sensor and three sets of binocular stereo vision systems use nonlinear optimization to solve the mutual spatial transformation relationship T, where where q i are the coordinates of the measured point under the depth sensor.

7. The online visual inspection system for dry state of hollow fiber membrane based on voxelized multi-resolution clustering according to claims 3, 4, 5 and 6, characterized in that: Using the data obtained by the depth sensor as the initial value, the voxylated multi-resolution clustering algorithm is used in combination with three sets of binocular stereo vision systems to fuse the real-time data obtained by the depth sensor and the three sets of binocular stereo vision systems, quickly and accurately remove the external points, achieve relocation, and obtain the maximum value X in the three directions of X, Y, and Z. max , Y max , Z max , minimum value X min， Y min , Z min , and then use the maximum value minus the minimum value to calculate the bounding box side length l x , l y , l z , set the side length of the voxel cube cell, and cut it into M, N, L parts in the three directions of X, Y, and Z coordinate axes. Label each voxel cube as (x, y, z) to get the voxel cube corresponding to each data point Calculate the center of gravity of each point cloud, keep only the center of gravity point cloud of each voxel cube, and remove other point clouds. Among them C ijk is the center of gravity of the voxel cube, and k is the number of point clouds.

8. The online visual inspection system for dry state of hollow fiber membrane based on voxelized multi-resolution clustering according to claims 3, 4, 5 and 6, characterized in that: The maximum joint probability (MJP) is used as the basis for judging the matching results. Where F(P i ) is the maximum joint probability, U(P i ) is point P i The set of all neighbors, P j YesP i Points in the neighborhood, P E is the joint probability, P E (P i , P j )=P d (P i ) d (P j )LGC(P i , P j ), where P d is the probability of the feature descriptor, LGC(P i , P j ) is the local geometric consistency.

9. The online visual inspection system for dry hollow fiber membranes based on voxelized multi-resolution clustering according to claim 1, characterized in that: The pressure self-adapting regulator includes a micro-pressure sensor and an automatic control device. The pressure self-adapting regulator automatically balances the air pressure in the hollow fiber membrane assembly during the detection process, and the pressure range is 8000±200Pa.

10. The online visual inspection system for dry state of hollow fiber membrane based on voxelized multi-resolution clustering according to claim 1, characterized in that: The PE membrane needs to be tightly attached to the detection end face of the hollow fiber membrane assembly before ventilation. It has a certain degree of stretchability. When gas molecules penetrate from the pores of the membrane fibers, the PE membrane will be lifted up to form a deformation.

11. The online visual inspection system for dry hollow fiber membranes based on voxelized multi-resolution clustering according to claim 1, characterized in that: The hollow fiber module detection step comprises: Step S1, fixing the hollow fiber membrane assembly to a hollow fiber membrane assembly fixing device; Step S2, the hollow fiber membrane assembly is connected to an air source and a pressure adaptive regulator; Step S3, covering the detection end of the hollow fiber membrane assembly with a PE film; Step S4, placing and fixing the hollow fiber membrane assembly horizontally, with the end surface facing the joint vision system; Step S5, turning on the combined visual measurement device, transmitting the online received image to the computer for processing to obtain real-time surface change data, and obtaining the defective film wire position; Step S6, closing the gas source, the pressure adaptive regulator and the combined visual measuring device; Step S7: detection ends.

12. The online visual inspection system for dry hollow fiber membranes based on voxelized multi-resolution clustering according to claim 1, characterized in that:

Images