CN115494077A - Gas cylinder detection device based on single sample learning AI algorithm and application method - Google Patents
Gas cylinder detection device based on single sample learning AI algorithm and application method Download PDFInfo
- Publication number
- CN115494077A CN115494077A CN202211270298.9A CN202211270298A CN115494077A CN 115494077 A CN115494077 A CN 115494077A CN 202211270298 A CN202211270298 A CN 202211270298A CN 115494077 A CN115494077 A CN 115494077A
- Authority
- CN
- China
- Prior art keywords
- hydrogen cylinder
- electric
- data
- hydrogen
- camera
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/90—Investigating the presence of flaws or contamination in a container or its contents
- G01N21/909—Investigating the presence of flaws or contamination in a container or its contents in opaque containers or opaque container parts, e.g. cans, tins, caps, labels
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/01—Arrangements or apparatus for facilitating the optical investigation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8883—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges involving the calculation of gauges, generating models
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/30—Hydrogen technology
- Y02E60/32—Hydrogen storage
Abstract
The gas cylinder detection device based on the single sample learning AI algorithm comprises an electric roller type conveyor, an electric linear sliding table, an electric telescopic rod, a lighting lamp, a wireless camera, a motor speed reducing mechanism, a power supply module, a PLC, a photoelectric switch and a conductive bearing; also has a control circuit; the electric roller type conveyor is provided with a supporting plate, an electric linear sliding table, an electric telescopic rod, a lighting lamp, a wireless camera, a motor speed reducing mechanism and a photoelectric switch are arranged below the supporting plate, and two conductive bearings, the camera and the lighting lamp are arranged at the side end of a power output shaft of the motor speed reducing mechanism; the power supply module, the PLC and the control circuit are arranged in the electric cabinet and are electrically connected; the application method of the gas cylinder detection device based on the single-sample learning AI algorithm comprises three steps. The invention can effectively realize the acquisition of the pictures without dead angles in the hydrogen cylinder, achieve good acquisition effect and effectively identify the quality problem of the hydrogen cylinder on the premise of low cost.
Description
Technical Field
The invention relates to the technical field of quality detection equipment and application methods, in particular to a gas cylinder detection device based on a single-sample learning AI algorithm and an application method.
Background
The high-pressure container hydrogen storage has the advantages of simple structure, high hydrogen storage density, high charging and discharging speed and the like, and becomes a main vehicle-mounted hydrogen storage mode of the hydrogen fuel cell automobile. Hydrogen in the hydrogen cylinder is high pressure state, in case the hydrogen cylinder takes place fatigue damage or emergency and damages, causes high-pressure hydrogen to leak, will bring serious potential safety hazard, consequently need detect hydrogen cylinder inside (hydrogen cylinder outward appearance detects the staff and can obviously see specific problem with the eye, consequently does not need special check out test set to detect).
Along with the development of science and technology, the AI-based hydrogen cylinder internal quality detection system is developed at a rapid pace, and in specific application, the camera acquires image data in the gas cylinder, and then the data is output to the upper computer, and the upper computer intelligently judges the quality of the gas cylinder based on internal application software. Because the quality of the hydrogen cylinder does not need to be judged manually in the detection, the detection data is more real and effective, convenience is brought to workers, and the detection working efficiency is improved. In the quality detection of the hydrogen cylinder based on the AI technology, the collection of pictures in the hydrogen cylinder is the basis for judging the quality of the hydrogen cylinder (before the hydrogen cylinder is used, a detection system establishes a single-sample learning model based on a plurality of collected pictures of defects of the hydrogen cylinder, and in the subsequent practical application, the system compares the collected image data of the hydrogen cylinder with the model to obtain the result of whether the hydrogen cylinder has defects). In the prior art, a worker generally holds a handle for installing a camera in a manual mode, and then inserts the camera into a hydrogen cylinder through a plug opening or a bottle opening valve opening of the hydrogen cylinder (before detection, the plug or the bottle opening valve is detached), so as to acquire image data. The manual mode of people gathers the image data of hydrogen cylinder not only for the staff has brought inconvenience, has increased staff's intensity of labour, and owing to in order to realize good data acquisition effect, the back in the camera gets into the hydrogen cylinder, needs the manual twist grip of staff, and then makes the camera rotate, realizes 360 degrees picture collections in the hydrogen cylinder, consequently more can bring adverse effect for the staff collection picture data. Finally, in the existing hydrogen cylinder AI detection, the following problems exist due to technical limitations, namely, the data marking cost is high, and a large amount of defect data needs to be marked and classified manually; secondly, the model (adopting the sample data of the hydrogen cylinder defect) can not well solve the situation of unknown defect; for the reasons, the application of the existing AI technology to the quality detection of the hydrogen cylinders has great restrictions.
Disclosure of Invention
In order to overcome the defects of the existing AI-based hydrogen cylinder internal quality detection system and the corresponding identification method, due to the technical limitations of the background, the invention provides a single-sample learning AI algorithm-based hydrogen cylinder detection device and an application method, wherein under the combined action of related mechanisms and circuits, the manual operation is not needed, an electric roller type conveyor is used for circularly conveying hydrogen cylinders to be detected to a detection station, after the corresponding hydrogen cylinders reach the detection station, related equipment can control a wireless camera to enter the hydrogen cylinders for picture acquisition, and a middle camera can be in a rotating mode and can acquire pictures in the hydrogen cylinders at 360-degree dead angles, so that the picture acquisition in the hydrogen cylinders can be effectively realized, a good acquisition effect can be achieved, convenience is brought to workers, the data acquisition cost is saved, and the defects of the hydrogen cylinders can be effectively and accurately identified through the functions of related unit modules.
The technical scheme adopted by the invention for solving the technical problems is as follows:
the gas cylinder detection device based on the single sample learning AI algorithm comprises an electric roller type conveyor, an electric linear sliding table, an electric telescopic rod, a lighting lamp, a wireless camera, a motor speed reducing mechanism, a power supply module, a PLC, a photoelectric switch and a conductive bearing; the device is characterized by also comprising a control circuit, a data acquisition unit, a data analysis unit, a database unit and a prompt unit; the electric roller type conveyor is provided with a supporting plate, an electric telescopic rod is arranged at the upper end in the supporting plate, and an electric linear sliding table is arranged at the lower part of the electric telescopic rod; the motor reducing mechanism is arranged on the lower part of a sliding block of the electric linear sliding table; an insulating pipe is sleeved on one side of a power output shaft of the motor speed reducing mechanism, two conductive bearings are tightly sleeved on the outer side of the insulating pipe at intervals, a wire groove is formed in the outer side of the insulating pipe, wires connected with the inner sides of the inner rings of the two conductive bearings are led out through the wire groove, outer insulating pipes are tightly sleeved on the outer sides of the outer rings of the two conductive bearings, the wires connected with the outer sides of the outer rings of the two conductive bearings are led out through the outer insulating pipes, and one end of each outer insulating pipe is installed on the other side end of a shell of the motor speed reducing mechanism; the camera and the illuminating lamp are arranged at the side end of a power output shaft of the motor speed reducing mechanism; the photoelectric switch is arranged in the supporting plate, and the power supply module, the PLC and the control circuit are arranged in the electric cabinet; the power output end of the control circuit is electrically connected with the power input end of the PLC, and the multi-path power output end of the PLC is electrically connected with the outer sides of the two conductive bearing outer rings, the power input end of the motor reducing mechanism, the power input end of the electric telescopic rod and the power input end of the electric linear sliding table respectively; the inner sides of the inner rings of the two conductive bearings are electrically connected with the power input ends of the camera and the illuminating lamp respectively; the trigger signal output end of the photoelectric switch is electrically connected with the trigger signal input end of the control circuit, and the control power supply output end of the control circuit is electrically connected with the power supply input end of the electric roller conveyor; the data acquisition unit, the database unit, the data analysis unit and the prompt unit are application software installed in a PC.
Furthermore, the hydrogen cylinder is positioned behind the upper part of the conveying roller of the electric roller type conveyor, the screw plug hole, the middle part of the bottle opening valve hole and the wireless camera and the illuminating lamp of the hydrogen cylinder are positioned on a vertical plane, and the outer diameters of the wireless camera and the illuminating lamp are smaller than the inner diameters of the screw plug hole and the bottle opening valve hole of the hydrogen cylinder.
Further, the control circuit comprises a time relay module and a relay which are electrically connected, a negative pole trigger signal input end of the time relay module, a negative pole power supply input end and a negative pole power supply input end of the relay are connected, and a positive pole power supply output end of the time relay module is connected with a positive pole power supply input end of the relay.
The application method of the gas cylinder detection device based on the single-sample learning AI algorithm is characterized by comprising the following steps of: the data acquisition unit receives picture data in the gas cylinder acquired by the camera in real time, and outputs the data to the data analysis unit after the data is primarily processed; and B: the data analysis unit is used for comparing the data input by the data acquisition unit with the qualified hydrogen cylinder picture data in the database unit, and outputting a trigger signal to the prompt unit when the quality of the hydrogen cylinders in the corresponding detection area is poor; and C: after receiving the trigger signal, the prompting unit can prompt staff that the quality problem exists at the corresponding position of the hydrogen cylinder in a display interface character and alarm sound mode.
Furthermore, the model constructed by the data acquisition unit, the database unit, the data analysis unit and the prompt unit does not need defect data of the hydrogen cylinder to construct an algorithm; the model specifically uses a pre-training model to extract features from the input qualified hydrogen cylinder images, and a memory base is uniformly established to store a representative feature map of qualified hydrogen cylinder samples.
Further, before the database unit is applied, an Imagenet pre-training model RESNET64 is downloaded, then image features of qualified hydrogen cylinders are extracted, feature maps of down-sampling layers are obtained, then down-sampling feature maps of all qualified hydrogen cylinder samples are collected, dimension reduction is carried out by using a Johnson-Lindenstauss and Coreset sub-sampling method, and therefore the required storage space is compressed.
Further, in the application of the data analysis unit, in the analysis stage, the real-time new picture obtained by the camera is extracted in the same way as the hydrogen cylinder feature map by using the data acquisition unit, then the extracted feature map is compared with the qualified sample data set subjected to dimensionality reduction in the training stage in the database unit, and when the difference is greater than a set threshold value, the data analysis unit is regarded as abnormal.
The invention has the beneficial effects that: under the action of a control circuit, a photoelectric switch and the like, when the hydrogen cylinder is output to the lower part of the supporting plate, the electric roller type conveying machine can lose electricity temporarily and does not work, then the PLC controls the working modes of the electric telescopic rod and the electric linear sliding table, the camera and the illuminating lamp automatically enter the hydrogen cylinder to collect pictures of the hydrogen cylinder, the collected picture data are transmitted in a wireless mode, and after the PC and the matched wireless receiving module thereof receive wireless picture signals, the AI system can judge the quality in the hydrogen cylinder. The invention particularly has the advantages that the camera can be in a rotating mode during collection, and can collect pictures in the hydrogen cylinder in 360 degrees without dead angles, so that the collection of the pictures in the hydrogen cylinder without dead angles can be effectively realized, a good collection effect can be achieved, convenience is brought to workers, and the data collection cost is saved. In the invention, when the software analysis unit analyzes the quality of the hydrogen cylinder, if a new defect occurs, the characteristic of the software analysis unit is obviously different from the picture characteristic of a qualified hydrogen cylinder sample, so that the software analysis unit cannot be mistakenly identified as a qualified hydrogen cylinder, and the problem that the defect type data faced by the conventional visual detection technology needs to be inexhaustible is effectively solved on the premise of a small sample. In conclusion, the invention has good application prospect.
Drawings
The invention is further illustrated by the following figures and examples.
Fig. 1 is a schematic diagram of the overall structure and a partial enlarged structure of the gas cylinder detection device based on the single-sample learning AI algorithm.
Fig. 2 is a circuit diagram of the gas cylinder detection device based on the single sample learning AI algorithm.
Fig. 3 is a block diagram of an application method software architecture of the gas cylinder detection device based on the single-sample learning AI algorithm of the present invention.
Detailed Description
As shown in fig. 1 and 2, the gas cylinder detection device based on the single sample learning AI algorithm includes an electric roller conveyor M, an electric linear sliding table M1, an electric telescopic rod M2, an illuminating lamp H, a wireless camera SX, a motor speed reduction mechanism M3, a power module A1, a PLCA4, a photoelectric switch A2, and a copper bearing X; also has a control circuit 1; the middle part of the electric roller type conveyor M is provided with an n-shaped support plate 2, two sets of electric telescopic rods M2 are arranged, the two sets of electric telescopic rods M2 are distributed at intervals left and right, the upper ends of the cylinders are vertically arranged on the left and right sides of the middle part of the upper end in the support plate 2 through screw nuts, the electric linear sliding tables M1 are distributed left and right, the upper ends of the shells of the electric linear sliding tables are arranged on the lower parts of the movable rods of the two sets of electric telescopic rods M2, and the sliding blocks of the electric linear sliding tables are positioned on the lower parts; the shell of the motor reducing mechanism M3 is distributed left and right and is arranged on the lower part of the sliding block of the electric linear sliding table M1 through a screw nut; the right end of a power output shaft of the motor speed reducing mechanism M3 is tightly sleeved with an insulating rubber tube 3 (the right side end is spaced from a shell of the motor speed reducing mechanism by a certain distance), inner rings of two bearings X are tightly sleeved at the outer side end of the rubber tube 3 by a certain distance at the left and right sides, the upper end of the outer side of the rubber tube 3 is transversely provided with a through wire groove, wires connected with the inner sides of the inner rings of the two bearings X are led out to the left outer end of the rubber tube through the wire groove, the outer sides of the outer rings of the two bearings X are tightly sleeved with an outer rubber tube 4, the wires connected with the outer sides of the outer rings of the two bearings X are led out through the upper end of the outer rubber tube 4, and a flange at the right end of the outer rubber tube 4 is arranged at the left end of the shell of the motor speed reducing mechanism M3 through a screw nut; the camera SX and the illuminating lamp H are respectively arranged at the left end of a shaft lever 5 through screw nuts, and the right end of the shaft lever 5 is welded with the leftmost end of a power output shaft of the motor speed reducing mechanism M3; the photoelectric switch A2 is arranged in the front of the middle part in the supporting plate 2, and a detection head of the photoelectric switch A faces downwards; the power supply module A1, the PLCA4 and the control circuit 1 are arranged in an electric cabinet 6 of the electric roller type conveyor.
As shown in figures 1 and 2, the distance between the baffle plates at the front side and the rear side of the electric roller type conveyor M is slightly larger than the outer diameter of the hydrogen cylinders by 5 mm. After the hydrogen cylinder 7 is positioned on the upper part of the conveying roller of the electric roller type conveyor M, the screw plug hole of the hydrogen cylinder 7, the middle part of the bottle opening valve hole and the front and the back of the wireless camera SX and the illuminating lamp H are positioned in a vertical straight line, and the outer diameters of the wireless camera SX and the illuminating lamp H are smaller than the inner diameters of the screw plug hole and the bottle opening valve hole of the hydrogen cylinder. When the movable rod of the electric telescopic rod M2 is positioned at the top dead center, the lower ends of the motor reducing mechanism M3, the wireless camera SX, the illuminating lamp H and the photoelectric switch A2 are higher than the upper end of the hydrogen cylinder 7. The working voltage of the electric roller type conveyor M is 380V and the power is 6KW; the electric linear sliding table M1 is a screw rod type electric linear sliding table finished product, the power is 150W, and the working voltage is direct current 48V; the electric telescopic rod M2 is a finished product of a reciprocating electric push rod with working voltage of direct current of 48V and power of 50W; the working voltage of the illuminating lamp H is 48V direct current, and the power is 10W; the motor reducing mechanism M3 is a finished product of a coaxial motor gear reducer with working voltage of direct current of 48V and power of 20W; the power module A1 is a finished product of a 220V/48V/1KW AC 220V-to-DC 48V switching power module. The control circuit comprises a time relay module A3 and a relay K1 which are connected through circuit board wiring, wherein a negative pole trigger signal input end 4 pin and a negative pole power input end 2 pin of the time relay module A3 are connected with a negative pole power input end of the relay K1, and a positive pole power output end of the time relay module A3 is connected with a positive pole power input end of the relay K1.
As shown in fig. 1 and 2, pins 1 and 2 of a power input terminal of a power module A1 and two poles of an alternating current 220V power supply are respectively connected through a wire, three control power input terminals of a 380V power supply and a control power input terminal relay K1 of a control circuit are respectively connected through a wire, pins 3 and 4 of a power output terminal of the power module A1, pins 1 and 2 of a power input terminal of a photoelectric switch A2, and pins 1 and 2 of a power input terminal time relay module A3 of the control circuit are respectively connected through a wire; pins 9 and 2 of a power output end time relay module A3 of the control circuit and pins 1 and 2 of a power input end of a PLCA4 are respectively connected through leads, and a plurality of power output ends 7 and 8, pins 5 and 6, pins 3 and 4 of the PLC, the outer sides of outer rings of two bearings X and the power input end of a motor speed reducing mechanism M3, an electric telescopic rod M2 and the power input end of an electric linear sliding table M1 are respectively connected through leads; the inner sides of the inner rings of the two bearings X are respectively connected with the power input ends of the camera SX and the illuminating lamp H through wires; the 3 feet of the trigger signal output end of the photoelectric switch A2 are connected with the 3 feet of the trigger signal input end time relay module A3 of the control circuit through leads, and the three normally closed contact ends of the control power output end relay K1 of the control circuit are connected with the power input end of the electric roller type conveyor M through leads.
As shown in fig. 1 and 2, after a 220V power supply enters the power supply input terminal of the power supply module A1, pins 3 and 4 of the power supply module A1 will output a stable dc48V power supply to enter the power supply input terminals of the photoelectric switch A2 and the control circuit, and the above circuits are powered on. After the electric roller conveyor M works by electricity, the rotating conveying rollers push hydrogen cylinders 7 continuously placed on the electric roller conveyor M by workers towards the detection station (the plurality of hydrogen cylinders are distributed from front to back, and the distance between the hydrogen cylinders is greater than the distance between the electric linear sliding table and the camera, which drives the camera to move back and forth), and the discharge station at the rear end of the electric conveying rollers is used for detecting the hydrogen cylinders 7. Before front end hydrogen cylinder 7 approached the detection station, it can block the infrared light beam that photoelectric switch A2 emitter launches, and then, 3 feet output high level of photoelectric switch A2 get into 3 feet of time relay module A3, and 9 feet of time relay module A3 can be separated 1 second and export 40 seconds power (time is adjustable) and get into PLC's power input end and relay K1 positive power input end. After the relay K1 is electrified and sucked, the control power supply input end and the normally closed contact end are opened, so that the electric roller type conveyor M can lose electricity temporarily and does not work, and the front-end hydrogen cylinder 7 stops at the detection station to wait for detection. After the PLC is electrified to work, a power supply of 40 seconds is output to the power input end of the motor reducing mechanism M3 by 7 and 8 pins of the PLC, and the power supply is supplied to the camera SX and the illuminating lamp H by the inner ring and the outer ring of the bearing X. After the PLCA4 is electrified, the 5 and 6 pins of the PLCA4 can output 4-second positive and negative pole power supplies to the positive and negative pole power supply input ends of the two sets of electric telescopic rods M2, the two sets of electric telescopic rods M2 are electrified to work, the movable rods of the two sets of electric telescopic rods push the motor speed reducing mechanism and the electric linear sliding table to move downwards to a stop point to stop moving, and at the moment, the camera SX is transversely positioned on the right side of a screw hole of the hydrogen cylinder; then, pins 4 and 5 of the PLCA4 output a 16-second positive and negative pole power supply to the positive and negative pole power supply input end of the electric linear sliding table M1, after the electric linear sliding table M1 is powered on, the sliding block of the electric linear sliding table pushes the camera to enter the hydrogen cylinder from right to left, and then the camera gradually collects images from right to left in the hydrogen cylinder (the illuminating lamp H emits light to provide a light source for the camera). When the motor reducing mechanism M3 drives the camera to rotate, the power supply supplies power to the camera SX and the illuminating lamp H through the outer rings of the two non-rotating bearings X, the rolling balls of the two bearings X and the inner rings of the two bearings X, so that the rotating camera can rotate to acquire picture data in the hydrogen cylinder (the 16-second camera can move to the 2 cm position at the left end in the hydrogen cylinder, and therefore, the picture data in the hydrogen cylinder can be acquired completely). After 16 seconds, the pins 4 and 5 of the PLCA4 output a 16-second negative and positive two-pole power supply to the negative and positive two-pole power supply input end of the electric linear sliding table M1, the sliding block of the electric linear sliding table M1 drives the camera to withdraw from the left to the right in the hydrogen cylinder 7, in the process, the camera SX continues to acquire images from the left to the right in the hydrogen cylinder, and more image data samples are acquired. When the pins 4 and 5 of the PLCA4 stop outputting power, the camera SX is already positioned at the right outer end of the hydrogen cylinder, then the pins 5 and 6 of the P LCA3 can output 4-second negative and positive pole power to the negative and positive pole power input ends of the two sets of electric telescopic rods M2, the two sets of electric telescopic rods M2 are electrified to work, the movable rods of the two sets of electric telescopic rods drive the motor speed reducing mechanism and the electric linear sliding table to move upwards to a stop point to stop moving, and at the moment, the height of the camera SX is higher than that of the hydrogen cylinder 7, so that preparation is made for next detection. After 40 seconds, after the power supply output of the pin 9 of the time relay module A2 is stopped, the PLCA4 loses power, the relay K1 loses power at the same time and controls the power supply input end and the normally closed contact end to be closed, the electric roller conveyor M is powered on again to convey the detected hydrogen cylinders 7 to the rear unloading station (unloading personnel unload the hydrogen cylinders), meanwhile, the next hydrogen cylinders to be detected are conveyed to the detection station, after the next hydrogen cylinders 7 approach the photoelectric switch A2 again, the photoelectric switch A2 outputs high level at intervals (intervals are formed between the two hydrogen cylinders), the pin 3 of the time relay module A3 is triggered again, the pin 9 of the time relay module A3 outputs power supply to the PLC again, the process is completely consistent with the process, and the invention can automatically detect the next hydrogen cylinders. Through the combined action of all the mechanisms, after the hydrogen cylinder is output to the supporting plate, the electric roller type conveying machine can lose power temporarily and does not work, then the PLC controls the working modes of the electric telescopic rod and the electric linear sliding table, the camera and the illuminating lamp automatically enter the hydrogen cylinder, the picture is collected on the hydrogen cylinder, the collected picture data is transmitted in a wireless mode, after the PC and the matched wireless receiving module receive a wireless picture signal, the AI system judges the quality in the hydrogen cylinder, particularly, the camera can be in a rotating mode during collection, the picture in the hydrogen cylinder is collected in 360-degree dead angles, therefore, the dead angle-free picture collection in the hydrogen cylinder can be effectively realized, a good collection effect can be achieved, convenience is brought to workers, and the data collection cost is saved. It should be noted that the time of the time relay module and the time of the PLC output power supply are not fixed, and technicians can adjust the time according to needs, so that it is guaranteed that effective pictures of the hydrogen cylinder are collected. The relay K1 is a DC48V relay; the photoelectric switch A2 is a PNP type infrared reflection photoelectric switch finished product with the model E3K100-100, the working voltage is direct current 48V, the photoelectric switch is provided with three connecting wires, two pins 1 and 2 are power supply input wires, the other pin 3 is a signal output wire, when the photoelectric switch works, an emitting head of a detecting head of the photoelectric switch can emit infrared beams, when the infrared beams emitted by the detecting head are blocked by objects and received by receiving heads arranged in parallel on the detecting head within the farthest 100 cm range, the pin 3 of the signal output wire can output high level, when the objects are not blocked, the pin 3 of the signal output wire can not output high level, an adjusting knob is arranged outside the upper end of a shell of the photoelectric switch, the adjusting knob adjusts the detecting distance of the detecting head to the left to become close, and adjusts the detecting distance of the detecting head to the right to become far (the embodiment adjusts to about 50 cm); the time controller module A3 is a time controller module finished product with the model YYC-2S, the working voltage of the time controller module finished product A31 is direct current 48V, the time controller module finished product A3 is provided with a digital LED tube with a four-digit time display, two power supply input ends 1 and 2 pins, two trigger signal input ends 3 and 4 pins, a setting key 5 pin, an emergency stop key 6 pin, a time adding key 7 pin, a time reducing key 8 pin and a normally open power supply output end 9 pin, after the positive and negative two-pole power supply input ends of the time controller module finished product A3 are electrified, an operator presses the setting key, the time adding key and the time reducing key are respectively operated through the digital display of a nixie tube, the normally open power supply output end can output a positive power supply in a required time period, after the set time period, the normally open power supply output end stops outputting the power supply, and the 3 and 4 pins can output a period of power supply after inputting a trigger signal once and the 9 pins; PLCA4 model is FX3U-16MR/ES-A; the wireless camera SX model is y-202 (the wireless camera takes pictures, and the wireless receiving module matched with the PC receives the pictures is the existing mature technology, and this embodiment does not protect the technical point).
As shown in fig. 1, 2 and 3, the application method of the gas cylinder detection device based on the single-sample learning AI algorithm comprises the following steps: the data acquisition unit receives image data in the gas cylinder acquired by the camera in real time through the wireless receiving module, and outputs the data to the data analysis unit after primary processing; and B: the data analysis unit is used for comparing the data input by the data acquisition unit with the qualified hydrogen cylinder picture data in the database unit, and outputting a trigger signal to the prompt unit when the quality of the hydrogen cylinders in the detection area is poor; step C: after receiving the trigger signal, the prompting unit can prompt staff that the quality problem exists at the corresponding position of the hydrogen cylinder through displaying interface characters and an alarm sound mode. The model constructed by the data acquisition unit, the database unit, the data analysis unit and the prompt unit does not need a large amount of defect data of the hydrogen cylinders to construct an algorithm; the model specifically uses a pre-training model to extract features from input qualified hydrogen cylinder images, and a memory base is established in a unified manner to store a representative feature map of qualified hydrogen cylinder samples; when the method is applied, if a new defect occurs, the characteristic of the defect is obviously different from that of a qualified hydrogen cylinder sample, so that the defect cannot be mistakenly identified as a qualified product, and the problem that a defect type picture sample set cannot be exhausted in the conventional visual detection technology is effectively solved on the premise of a small sample. Specifically, before the database unit is applied, an Imagenet (large-scale computer vision recognition data set) pre-training model RESNET64 (residual error network) is downloaded, then image features of qualified hydrogen cylinders are extracted, feature maps of down-sampling layers are obtained, then down-sampling feature maps of all qualified hydrogen cylinder samples are collected, and dimension reduction is performed by using Johnson-Lindenstaus (Johnson-Linden Schott's theorem) and Coreset (kernel set) sub-sampling methods, so that the required storage space is compressed. In the application of the data analysis unit, in the analysis stage, the real-time hydrogen cylinder new picture obtained by the camera is extracted in the same way as the hydrogen cylinder characteristic picture extracted by the data acquisition unit, the hydrogen cylinder picture characteristic is extracted, then the extracted characteristic picture is compared with a qualified sample data set subjected to dimensionality reduction in the training stage in the database unit, and when the difference is greater than a set threshold value (0.05), the hydrogen cylinder picture is considered as abnormal. Through the above, in the invention, when the software analysis unit analyzes the quality of the hydrogen cylinder, if a new defect occurs, the characteristic of the software analysis unit is obviously different from the picture characteristic of a qualified hydrogen cylinder sample, so that the software analysis unit cannot be mistakenly identified as a qualified hydrogen cylinder, and the problem that the type of a defective picture sample set faced by a conventional visual detection technology is inexhaustible is effectively solved on the premise of a small sample.
While there have been shown and described what are at present considered the fundamental principles and essential features of the invention and its advantages, it will be apparent to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, but is capable of other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.
Furthermore, it should be understood that although the present description refers to embodiments, the embodiments do not include only one independent technical solution, and such description is only for clarity, and those skilled in the art should take the description as a whole, and the technical solutions in the embodiments may be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims (7)
1. The gas cylinder detection device based on the single sample learning AI algorithm comprises an electric roller type conveyor, an electric linear sliding table, an electric telescopic rod, a lighting lamp, a wireless camera, a motor speed reducing mechanism, a power supply module, a PLC, a photoelectric switch and a conductive bearing; the device is characterized by also comprising a control circuit, a data acquisition unit, a data analysis unit, a database unit and a prompt unit; the electric roller type conveyor is provided with a supporting plate, an electric telescopic rod is arranged at the upper end in the supporting plate, and an electric linear sliding table is arranged at the lower part of the electric telescopic rod; the motor reducing mechanism is arranged on the lower part of a sliding block of the electric linear sliding table; an insulating pipe is sleeved on one side of a power output shaft of the motor speed reducing mechanism, two conductive bearings are tightly sleeved on the outer side of the insulating pipe at intervals of an inner ring, a wire groove is formed in the outer side of the insulating pipe, wires connected with the inner sides of the inner rings of the two conductive bearings are led out through the wire groove, an outer insulating pipe is tightly sleeved on the outer sides of the outer rings of the two conductive bearings, wires connected with the outer sides of the outer rings of the two conductive bearings are led out through the outer insulating pipe, and one end of the outer insulating pipe is installed on the other side end of a shell of the motor speed reducing mechanism; the camera and the illuminating lamp are arranged at the side end of a power output shaft of the motor speed reducing mechanism; the photoelectric switch is arranged in the supporting plate, and the power supply module, the PLC and the control circuit are arranged in the electric cabinet; the power output end of the control circuit is electrically connected with the power input end of the PLC, and the multi-path power output end of the PLC is electrically connected with the outer sides of the two conductive bearing outer rings, the power input end of the motor reducing mechanism, the power input end of the electric telescopic rod and the power input end of the electric linear sliding table respectively; the inner sides of the inner rings of the two conductive bearings are respectively and electrically connected with the camera and the power input end of the illuminating lamp; the control power supply output end of the control circuit is electrically connected with the power supply input end of the electric roller type conveyor; the data acquisition unit, the database unit, the data analysis unit and the prompt unit are application software installed in a PC.
2. The device for detecting the gas cylinder based on the single-sample learning AI algorithm as claimed in claim 1, wherein the hydrogen cylinder is positioned behind the upper part of the delivery rollers of the electric roller conveyor, the middle part of the screw plug hole and the cylinder opening valve hole of the hydrogen cylinder and the front and back parts of the wireless camera and the illuminating lamp are positioned on a vertical plane, and the outer diameters of the wireless camera and the illuminating lamp are smaller than the inner diameters of the screw plug hole and the cylinder opening valve hole of the hydrogen cylinder.
3. The single-sample learning AI algorithm based gas cylinder detection device of claim 1 wherein the control circuit comprises an electrically connected time relay module and relay, the time relay module negative trigger signal input, the negative power input and the relay negative power input being connected, the time relay module positive power output and the relay positive power input being connected.
4. The application method of the gas cylinder detection device based on the single-sample learning AI algorithm as claimed in claim 1, characterized by comprising the following steps, step A: the data acquisition unit receives picture data in the gas cylinder acquired by the camera in real time, and outputs the data to the data analysis unit after the data is primarily processed; and B: the data analysis unit is used for comparing the data input by the data acquisition unit with the qualified hydrogen cylinder picture data in the database unit, and outputting a trigger signal to the prompt unit when the quality of the hydrogen cylinders in the corresponding detection area is poor; step C: after receiving the trigger signal, the prompting unit can prompt staff that the quality problem exists at the corresponding position of the hydrogen cylinder in a display interface character and alarm sound mode.
5. The application method of the gas cylinder detection device based on the single-sample learning AI algorithm is characterized in that the model constructed by the data acquisition unit, the database unit, the data analysis unit and the prompt unit does not need defect data of the hydrogen cylinder to construct the algorithm; the model specifically uses a pre-training model to extract features from the input qualified hydrogen cylinder images, and a memory base is uniformly established to store a representative feature map of qualified hydrogen cylinder samples.
6. The application method of the gas cylinder detection device based on the single-sample learning AI algorithm is characterized in that before the database unit is applied, an Imagenet pre-training model RESNET64 is downloaded, then image features of qualified hydrogen cylinders are extracted to obtain feature maps of a down-sampling layer, then the down-sampling feature maps of all qualified hydrogen cylinder samples are collected, and dimension reduction is performed by using a Johnson-lindenstruuss and Coreset sub-sampling method, so that the required storage space is compressed.
7. The application method of the gas cylinder detection device based on the single-sample learning AI algorithm, according to the claim 4, is characterized in that in the application of the data analysis unit, in the analysis stage, the real-time new picture obtained by the camera is extracted in the same way as the hydrogen cylinder feature map extracted by the data acquisition unit, then the extracted feature map is compared with the qualified sample data set after dimension reduction in the training stage in the database unit, and when the difference is larger than a set threshold value, the data analysis unit is regarded as abnormal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211270298.9A CN115494077A (en) | 2022-10-18 | 2022-10-18 | Gas cylinder detection device based on single sample learning AI algorithm and application method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211270298.9A CN115494077A (en) | 2022-10-18 | 2022-10-18 | Gas cylinder detection device based on single sample learning AI algorithm and application method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115494077A true CN115494077A (en) | 2022-12-20 |
Family
ID=84475283
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211270298.9A Pending CN115494077A (en) | 2022-10-18 | 2022-10-18 | Gas cylinder detection device based on single sample learning AI algorithm and application method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115494077A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117437224A (en) * | 2023-12-20 | 2024-01-23 | 山东特联信息科技有限公司 | Gas cylinder defect image data processing system and method based on artificial intelligence |
| KR102627233B1 (en) * | 2023-09-26 | 2024-01-23 | (주) 웨다 | Ultra-high voltage cable surface quality inspection system using image based anomaly detection model |
-
2022
- 2022-10-18 CN CN202211270298.9A patent/CN115494077A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR102627233B1 (en) * | 2023-09-26 | 2024-01-23 | (주) 웨다 | Ultra-high voltage cable surface quality inspection system using image based anomaly detection model |
| CN117437224A (en) * | 2023-12-20 | 2024-01-23 | 山东特联信息科技有限公司 | Gas cylinder defect image data processing system and method based on artificial intelligence |
| CN117437224B (en) * | 2023-12-20 | 2024-03-29 | 山东特联信息科技有限公司 | Gas cylinder defect image data processing system and method based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115494077A (en) | Gas cylinder detection device based on single sample learning AI algorithm and application method | |
| CN109188161A (en) | Capacitor detection device and its detection method | |
| CN112678733A (en) | Wind turbine generator system maintenance platform | |
| CN205300898U (en) | Automated inspection associated working platform is examined to car light light, gas | |
| CN101865970B (en) | Automatic voltage-withstanding grounding test system for display and voltage-withstanding grounding test method thereof | |
| CN207117573U (en) | A kind of difunctional on-line detection device of defects for photovoltaic plant battery component | |
| CN106370658A (en) | Automatic keyboard conducting film visual detection system | |
| CN107961991B (en) | Image inspection device suitable for telephone shell | |
| CN110207745A (en) | A kind of laminated batteries x-ray on-line detector and its application method | |
| CN201892538U (en) | LAMP-LED (light-emitting diode) stand glue content detector | |
| CN208953506U (en) | The finished product mark checking device and mark check system of detector | |
| CN211740276U (en) | Mobile phone screen detection device | |
| CN206208769U (en) | One kind automation keyboard conducting film vision detection system | |
| CN111491160B (en) | Automatic detection device of camera for passenger flow statistics | |
| CN216622203U (en) | Multi-station detection equipment | |
| CN205506276U (en) | Automatic detect device of electric connector contact confining force | |
| CN216150436U (en) | NGOK connects sieving mechanism of platform of plugging into | |
| CN215613346U (en) | SenSOR test machine defective products automatic classification device | |
| CN220139100U (en) | Power transmission line overhauling device | |
| CN220568878U (en) | Aviation storage battery detection device | |
| CN216815861U (en) | Automobile parts gas tightness detection device | |
| CN218982382U (en) | Radio frequency coaxial connector detection equipment | |
| CN215542910U (en) | Quality inspection platform for manual detection of battery cap | |
| CN214851608U (en) | Camera module checkout stand capable of automatically classifying | |
| CN216561891U (en) | Intelligent mobile inspection device for wind power plant |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |