CN114708584A - Big data based cigarette product quality defect prevention and control learning system and method - Google Patents
Big data based cigarette product quality defect prevention and control learning system and method Download PDFInfo
- Publication number
- CN114708584A CN114708584A CN202210333292.5A CN202210333292A CN114708584A CN 114708584 A CN114708584 A CN 114708584A CN 202210333292 A CN202210333292 A CN 202210333292A CN 114708584 A CN114708584 A CN 114708584A
- Authority
- CN
- China
- Prior art keywords
- module
- defect
- data
- image
- learning
- 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.)
- Granted
Links
- 230000007547 defect Effects 0.000 title claims abstract description 168
- 235000019504 cigarettes Nutrition 0.000 title claims abstract description 69
- 238000000034 method Methods 0.000 title claims abstract description 25
- 230000002265 prevention Effects 0.000 title claims abstract description 20
- 238000004519 manufacturing process Methods 0.000 claims abstract description 30
- 238000004458 analytical method Methods 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 22
- 238000007405 data analysis Methods 0.000 claims abstract description 20
- 238000013500 data storage Methods 0.000 claims abstract description 11
- 238000004891 communication Methods 0.000 claims description 40
- 230000002950 deficient Effects 0.000 claims description 23
- 238000000605 extraction Methods 0.000 claims description 18
- 230000006870 function Effects 0.000 claims description 14
- 238000007781 pre-processing Methods 0.000 claims description 12
- 238000012795 verification Methods 0.000 claims description 12
- 238000007726 management method Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 8
- 238000001514 detection method Methods 0.000 claims description 6
- 230000005611 electricity Effects 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 238000012423 maintenance Methods 0.000 claims description 6
- 238000012372 quality testing Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 238000013527 convolutional neural network Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 230000006399 behavior Effects 0.000 claims description 3
- 230000002708 enhancing effect Effects 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000013508 migration Methods 0.000 claims description 3
- 230000005012 migration Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000013024 troubleshooting Methods 0.000 claims description 3
- 238000003908 quality control method Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 7
- 238000007689 inspection Methods 0.000 description 6
- 230000009471 action Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 241000208125 Nicotiana Species 0.000 description 1
- 235000002637 Nicotiana tabacum Nutrition 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/53—Querying
- G06F16/535—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B5/00—Electrically-operated educational appliances
- G09B5/08—Electrically-operated educational appliances providing for individual presentation of information to a plurality of student stations
- G09B5/12—Electrically-operated educational appliances providing for individual presentation of information to a plurality of student stations different stations being capable of presenting different information simultaneously
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Business, Economics & Management (AREA)
- Educational Administration (AREA)
- Educational Technology (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of quality control learning systems, and discloses a cigarette product quality defect prevention and control learning system and method based on big data, which comprises a data acquisition unit, a data processing unit, a data storage unit and a learning unit; the data acquisition unit comprises a camera; the data processing unit comprises a first processor, the first processor is provided with an analysis system, and the analysis system comprises an image processing unit and a data analysis unit; the learning unit comprises a second processor, the second processor is electrically connected with a first display, a voice playing module and a first control module, and a multimedia interface is arranged on the second processor. The invention can automatically collect the problems of cigarette physical quality, appearance defects and the like on the cigarette production line, analyze reasons and solutions, form a cigarette product quality defect prevention and control learning database for workers to learn and improve the service capability of the workers.
Description
Technical Field
The invention relates to the technical field of quality control learning systems, in particular to a cigarette product quality defect prevention and control learning system and method based on big data.
Background
In the production process of cigarettes, in order to avoid quality defects caused by the fact that produced cigarette products flow to the hands of consumers and rejection rate is increased due to excessive defective products in production; therefore, at present, various cigarette production enterprises pay great attention to the problem of producing high-quality cigarette products with stable quality and continuously meeting the industry requirements.
In the field of cigarette physical quality and cigarette appearance quality monitoring of cigarette quality control, at present, a plurality of links of inspection such as post self-inspection, workshop quality personnel spot inspection, special inspection of factory quality management departments and the like are generally adopted to control the defect rate of cigarettes, but all the inspection is off-line inspection and cannot effectively represent the whole processing quality effect. Moreover, because the proficiency of workers on the operation at each production post is different, part of workers with poor experience in identifying and analyzing the physical quality and appearance quality defects of the cigarettes can miss or misdetect the defects, and the tobacco industry has high requirements on detecting the product defects, so that each production worker on the cigarette production line can timely and accurately identify various defects, accurately analyze and judge the possible causes of the problems and solve the quality defect problems. However, such a requirement is very difficult for the average manufacturer, and a great deal of work experience must be accumulated to achieve such a level of skill. However, if only the staff is relied on to accumulate experience in the working practice, a long time is needed, the quality defect of the cigarette products on the cigarette production line is increased, the control of the production cost of enterprises is not facilitated, and the cigarette with the quality defect flowing into the market also has adverse effect on the cigarette brand.
Disclosure of Invention
In view of the above, the present invention provides a big data-based system and method for preventing and controlling learning of quality defects of cigarette products, which can automatically collect the problems of physical quality, appearance defects, etc. of cigarettes on a cigarette production line, analyze reasons and solutions, and form a database for preventing and controlling learning of quality defects of cigarette products for workers to learn, thereby improving the business ability of the workers.
The invention solves the technical problems through the following technical means:
a big data-based cigarette product quality defect prevention and control learning system comprises a data acquisition unit, a data processing unit, a data storage unit and a learning unit; the data acquisition unit comprises a camera arranged on a cigarette production line; the data processing unit comprises a first processor, the first processor is provided with an analysis system, and the analysis system comprises an image processing unit and a data analysis unit; the data storage unit comprises a standard quality feature memory, a shot image feature memory and a defect data memory which are electrically connected with the first processor, and the defect data memory comprises a historical quality accident module, an A-type defect storage module, a B-type defect storage module and a C-type defect storage module; the learning unit comprises a second processor, the second processor is electrically connected with a first display, a voice playing module and a first control module, and a multimedia interface is arranged on the second processor; the second processor is provided with a learning system, the learning system comprises a first login verification unit and a retrieval unit, the retrieval unit comprises an A-type defect retrieval module, a B-type defect retrieval module, a C-type defect retrieval module and a recommended learning module, and the A-type defect retrieval module, the B-type defect retrieval module, the C-type defect retrieval module and the recommended learning module respectively comprise a defect example display module, a reason generation module, a history defect analysis module and a scheme solution module capable of displaying solutions of troubleshooting and equipment adjustment.
Further, the image processing unit comprises an image acquisition module for acquiring images shot by the camera, an image preprocessing module for preprocessing the images and an image feature extraction module for extracting features of the images; the data analysis unit comprises a feature comparison module for comparing the extracted wrapping features with the standard quality storage module, a data statistics module for counting the total detection number and the defect number, and a data analysis grading module for analyzing and classifying the generated quality defects; the data analysis grading module comprises a non-defective module, a type A defective module, a type B defective module and a type C defective module.
Furthermore, the camera is electrically connected with a first communication sending module for transmitting images shot by the camera, the first processor is electrically connected with a first communication receiving module and a second communication sending module, the first communication sending module is in communication connection with the first communication receiving module, the second processor is electrically connected with a second communication receiving module, and the second communication receiving module is in communication connection with the second communication sending module.
Further, the data acquisition unit further comprises a second display, the second display is electrically connected with the first processor, a USB interface and a second control module are arranged on the second display, the second control module is electrically connected with the first processor, and the analysis system further comprises a second login verification module.
Further, first treater still electricity is connected with board display screen, production team group display screen, maintenance team group display screen, quality testing group display screen and workshop management display screen, all install the buzzer siren on board display screen, production team group display screen, maintenance team group display screen, quality testing group display screen and the workshop management display screen.
Further, the learning unit further comprises a support frame, the first display is fixedly mounted on the support frame through a screw, the support frame comprises a support plate, a telescopic rod is fixedly connected to the support plate, a base is arranged at one end, away from the support plate, of the telescopic rod, the telescopic rod comprises a first support rod and a second support rod, the first support rod and the second support rod are both of hollow structures, the first support rod is slidably arranged in the hollow structure of the second support rod, a fixing nut is fixedly connected in the hollow structure of the first support rod, a screw rod is connected to the fixing nut through a thread, a servo motor is fixedly mounted at the hollow bottom of the second support rod, and an output shaft of the servo motor is fixedly connected with the screw rod; the servo motor is electrically connected with the second processor. The support frame supports the first display, and the support height of the first display is adjusted through the telescopic rod, so that workers with different heights can operate and learn conveniently; during specific operation, the first control module can be controlled, the servo motor is controlled through the second processor, the servo motor drives the screw rod to rotate, the height of the fixing nut is adjusted, the height of the first supporting rod in the second supporting rod is adjusted, and the purpose of adjusting the height of the supporting plate is achieved.
A cigarette product quality defect prevention and control learning method based on big data comprises the following steps:
s1, the camera shoots the cigarettes on the production line to form a shot image, the image is transmitted to the image processing unit through the first communication transmitting module and the first communication receiving module, and the image is collected by the image collecting module;
s2, manually inputting the cigarette defect problem images found in work into a defect data storage through a USB interface and a second control module, and filling the whole defect storage analysis database;
s3, the image preprocessing module carries out filtering, enhancing, smoothing and sharpening on the collected electronic image signals to make the images clear and ensure the image quality;
s4, the image feature extraction module performs self-training and feature extraction on the preprocessed image by using a convolutional neural network to form an image feature signal;
s5, comparing the image characteristic signal formed by the image characteristic extraction module with the standard quality storage module by using the characteristic comparison module to obtain comparison data;
s6, the data analysis grading module analyzes and grades the comparison data obtained by the characteristic comparison module, and the grading is respectively as follows: the defects are not defective, A-type defects, B-type defects and C-type defects, and the generation reason and the solution of each type of defects are input through a second control module;
s7, the data statistics module counts the total detection number, the A-type defect number, the B-type defect number and the C-type defect number, and stores the data in the defect data memory;
s8, the learner logs in the learning system through the first login verification unit, searches through the A-type defect search module, the B-type defect search module, the C-type defect search module or the recommended learning module of the search unit, and displays the defect image, the generation reason and the solution for the learner to learn through the first display by the defect example display module, the reason generation module, the previous defect analysis module and the solution module.
Further, the method for recommending learning modules to automatically match learning contents in step S8 includes the following steps:
s801, constructing a data set comprising learner embedding, learner post information, multiple defect image embedding, different post defect image sharing context embedding and marked click behaviors, and obtaining post embedding of each post specific user by adopting a hierarchical difference attention algorithm;
s802, embedding and fusing the individual embedding of the user with the learner post information of different posts respectively by adopting a double-gate control network responsible for controlling the information inflow;
s803, according to different post defect image embedding, each type of context embedding is assigned with a weight, and the context embedding of the corresponding field is obtained through weighting fusion;
s804, inputting the individual embedding, the context embedding and the defect image embedding of the learners of each post into a shared characteristic cross type double MMoE network, and screening to obtain the recommended learning cases of each post based on the dynamic learning integration strategy.
Further, in the step S804, a multitask learning model MMoE is used for post-crossing recommendation, the multitask learning model adopts a dual-in dual-out dual-MMoE network architecture, a special MMoE network is set at each post, and the post networks share a feature cross layer; each MMoE network realizes cross-domain recommendation through two special post Tower networks, wherein one is responsible for the recommendation and learning of a target post, the other is responsible for the migration of post sharing information, and a final prediction result of each post is obtained through each post fusion layer; setting weights according to performances of different MMoE networks at different posts, and adjusting the weights through an optimized loss function; the feature crossing layer in the step S804 adopts a plurality of feature crossing modes, including Concat, Inner-product, and Outer-product, wherein the Inner-product operation is an Inner product operation of vectors, and the Outer-product operation is a pairwise crossing of each dimension of input feature vectors to generate a feature crossing matrix; the Tower network adopts a common full-connection layer, and carries out prediction according to the specific position representation transmitted by the gating network.
Further, the loss function in step S804 adopts a binary cross entropy calculation method, which isMMoE network loss function of each post and integrated MMoE network loss function LensembleThe sum of (W) and the regularization term R (θ) is specifically:
wherein k is used for marking the input and output of MMoE networks with different positions, M is the number of the positions, N is,
the total number of defects learned, W, is the weight of the MMoE network on each post.
The invention has the beneficial effects that:
1. according to the invention, the camera is arranged on the cigarette production line, a large number of physical quality or appearance defect images of cigarette production on the production line can be collected by using the camera, then the defect image data is stored to form a cigarette quality defect database, and the cigarette quality defect database is arranged, so that workers can learn various collected quality defects through the learning unit, and the skill of the workers and the quality management level of cigarette production enterprises are favorably improved.
2. The invention can search the contents to be learned through the search unit, and can automatically match the learning contents through the recommended learning module according to the login account number of the learner, so that the learner can learn more purposefully and independently, thereby improving the learning efficiency of the learner.
3. According to the method, the cross-post recommendation is carried out through the multi-task learning model MMoE, a special MMoE network is arranged for each post when the learning task recommendation is carried out, the recommended learning content has very high accuracy, the algorithm is simple, and the learning content can be efficiently and accurately recommended to the learner.
Drawings
FIG. 1 is a schematic diagram of a big data based cigarette product quality defect prevention and control learning system of the present invention;
FIG. 2 is a schematic diagram of the analysis system of FIG. 1;
FIG. 3 is a schematic diagram of the data storage unit of FIG. 1;
FIG. 4 is a schematic diagram of the retrieval unit of FIG. 1;
FIG. 5 is a schematic structural diagram of the first display and the supporting frame;
FIG. 6 is a schematic sectional view of the extension pole of FIG. 5;
FIG. 7 is a flow chart of a big data based cigarette product quality defect prevention and control learning method of the present invention;
fig. 8 is a flowchart of the step S8 in fig. 7;
fig. 9 is a picture with defects learned in example 2.
The system comprises a camera 1, a second display 2, a first communication sending module 3, a USB interface 4, a second control module 5, a first processor 6, a first communication receiving module 7, a second communication sending module 8, a second login verification module 9, an image processing unit 10, a data analysis unit 11, an image acquisition module 12, an image preprocessing module 13, an image feature extraction module 14, a feature comparison module 15, a data statistics module 16, a data analysis grading module 17, a non-defective module 18, a class A defective module 19, a class B defective module 20, a class C defective module 21, a buzzer alarm 22, a standard quality feature memory 23, a shot image feature memory 24, a defective data memory 25, a second processor 26, a second communication receiving module 27, a first login verification unit 28, a retrieval unit 29, a class A defect retrieval module 30, a class B defect retrieval module 31, The system comprises a C-type defect retrieval module 32, a recommended learning module 33, a support frame 34, a first display 35, a first support rod 36, a second support rod 37, a fixing nut 38, a screw 39, a servo motor 40, a multimedia interface 41, a bottom plate 42, a voice playing module 43 and a first control module 44.
Detailed Description
The invention will be described in detail below with reference to the following drawings:
examples 1,
The embodiment is a big data-based cigarette product quality defect prevention and control learning system, as shown in fig. 1-6, comprising a data acquisition unit, a data processing unit, a data storage unit and a learning unit;
wherein, the data acquisition unit is including installing camera 1, second display 2 on the cigarette production line, and 1 electricity of camera is connected with the first communication sending module 3 with 1 transmission of shooting image of camera, and second display 2 is connected with first treater 6 electricity, is provided with USB interface 4 and second on the second display 2 and controls module 5, and the second is controlled module 5 and is connected with first treater 6 electricity.
The data processing unit comprises a first processor 6, the first processor 6 is electrically connected with a first communication receiving module 7 and a second communication sending module 8, the first communication sending module 3 is in communication connection with the first communication receiving module 7, the first processor 6 is provided with an analysis system, and the analysis system comprises a second login verification module 9, an image processing unit 10 and a data analysis unit 11; the image processing unit 10 comprises an image acquisition module 12 for acquiring images shot by the camera 1, an image preprocessing module 13 for preprocessing the images, and an image feature extraction module 14 for extracting features of the images; the data analysis unit 11 comprises a feature comparison module 15 for comparing the extracted wrapping features with a standard quality storage module, a data statistics module 16 for counting the total detection number and the defect number, and a data analysis grading module 17 for analyzing and classifying the generated quality defects; data analysis grading module 17 includes a non-defective module 18, a type A defective module 19, a type B defective module 20, and a type C defective module 21. First treater 6 still electricity is connected with board display screen, production team group display screen, maintenance team group display screen, quality testing group display screen and workshop management display screen, all installs buzzer siren 22 on board display screen, production team group display screen, maintenance team group display screen, the quality testing group display screen and the workshop management display screen.
The data storage unit comprises a standard quality feature memory 23, a shot image feature memory 24 and a defect data memory 25 which are electrically connected with the first processor 6, wherein the defect data memory 25 comprises a historical quality accident module, an A-type defect storage module, a B-type defect storage module and a C-type defect storage module;
the learning unit comprises a second processor 26, the second processor 26 is electrically connected with a second communication receiving module 27, the second communication receiving module 27 is in communication connection with a second communication sending module 8, the second processor 26 is electrically connected with a first display 35, a voice playing module 43 and a first control module 44, and the second processor 26 is provided with a multimedia interface 4; the second processor 26 is provided with a learning system, the learning system comprises a first login verification unit 28 and a retrieval unit 29, the retrieval unit 29 comprises a class A defect retrieval module 30, a class B defect retrieval module 31, a class C defect retrieval module 32 and a recommendation learning module 33, and the class A defect retrieval module 30, the class B defect retrieval module 31, the class C defect retrieval module 32 and the recommendation learning module 33 respectively comprise a defect example display module, a reason generation module, a historical defect analysis module and a scheme solution module capable of displaying solutions of troubleshooting and equipment adjustment.
The learning unit further comprises a support frame 34, the first display 35 is fixedly mounted on the support frame 34 through screws, the support frame 34 comprises a support plate, an expansion rod is fixedly connected onto the support plate, a base is welded at the lower end of the expansion rod, the expansion rod comprises a first support rod 36 and a second support rod 37, the first support rod 36 and the second support rod 37 are both of a hollow structure, the first support rod 36 is slidably arranged in the hollow structure of the second support rod 37, a fixing nut 38 is fixedly connected into the hollow structure of the first support rod 36, a screw rod 39 is connected into the fixing nut 38 through internal threads, a servo motor 40 is fixedly mounted at the hollow bottom of the second support rod 37, and an output shaft of the servo motor 40 is fixedly connected with the screw rod 39; the servo motor 40 is electrically connected to the second processor 26.
Examples 2,
The embodiment is a method for preventing, controlling and learning quality defects of cigarette products based on big data, as shown in fig. 7-8, and the method comprises the following steps:
s1, the camera shoots the cigarettes on the production line to form a shot image, the image is transmitted to the image processing unit through the first communication transmitting module and the first communication receiving module, and the image is collected by the image collecting module;
s2, manually inputting the cigarette defect problem images found in work into a defect data storage through a USB interface and a second control module, and filling the whole defect storage analysis database;
s3, the image preprocessing module carries out filtering, enhancing, smoothing and sharpening on the collected electronic image signals to make the images clear and ensure the image quality, wherein the image preprocessing method comprises the following steps: s301, taking a pixel of an image to be processed as a center, and making an m multiplied by m action template;
s302, selecting K pixels with the minimum gray difference with pixels of an image to be processed from the m multiplied by m action templates;
and S303, replacing the original pixel value with the gray average value of the K pixels to obtain the filtered, enhanced, smoothed and sharpened image.
S4, the image feature extraction module performs self-training and feature extraction on the preprocessed image by using a convolutional neural network to form an image feature signal, wherein the self-training and feature extraction comprises the following steps:
s401, dividing the preprocessed cigarette image into image block sets with the size of m multiplied by m, and then using I for each image block set blocki(i 1, 2.., n), where i denotes a certain set of image blocks;
s402, collecting the images of each block IiInputting the training data into a convolutional neural network for training for multiple times;
s403, a convolution-check image I with a size of 6 × 6 and a number of steps of 3 is output as 36iPerforming convolution processing, and outputting each convolution layer through a ReLU activation function, wherein the ReLU activation function is as follows:
and f (x) max (0, x), the output is 0 when the input signal is less than 0, and the output is equal to the input when the input signal is greater than 0.
S404, performing pooling operation on the characteristic diagram output by the step S404 by 2 units, wherein the step number is 2, and the output is 36;
s405, performing convolution processing by using a convolution kernel with the size of 4 x 4 and the step number of 2 and the output of 96;
s406, performing pooling operation on the characteristic diagram output in the step S405 by 2 units, wherein the step number is 2, and the output is 96;
and S407, performing image feature extraction through four times of training, and classifying results output in the neural network to form a picture feature signal.
S5, comparing the image characteristic signal formed by the image characteristic extraction module with the standard quality storage module by using the characteristic comparison module to obtain comparison data;
s6, the data analysis grading module analyzes and grades the comparison data obtained by the characteristic comparison module, and the grading is respectively as follows: the defect-free method comprises the steps of defect-free defect, class A defect-free defect-type B and defect-type C defect-free defect;
s7, the data statistics module counts the total detection number, the A-type defect number, the B-type defect number and the C-type defect number and stores the data in the defect data memory;
s8, the learner logs in the learning system through the first login verification unit, searches through the A-type defect search module, the B-type defect search module, the C-type defect search module or the recommended learning module of the search unit, and displays the defect image, the generation reason and the solution for the learner to learn through the first display by the defect example display module, the reason generation module, the previous defect analysis module and the solution module.
The method for recommending the learning module to automatically match the learning content in the step S8 includes the following steps:
s801, constructing a data set comprising learner embedding, learner post information, multiple defect image embedding, different post defect image sharing context embedding and marked click behaviors, and obtaining post embedding of each post specific user by adopting a hierarchical difference attention algorithm;
s802, embedding and fusing the individual embedding of the user with the learner post information of different posts respectively by adopting a double-gate control network responsible for controlling the information inflow;
s803, according to different post defect image embedding, each type of context embedding is assigned with a weight, and the context embedding of the corresponding field is obtained through weighting fusion;
s804, inputting the individual embedding, the context embedding and the defect image embedding of the learners of each post into a shared characteristic cross type double MMoE network, and screening to obtain the recommended learning cases of each post based on the dynamic learning integration strategy. Performing post-crossing recommendation by adopting a multi-task learning model MMoE, wherein the multi-task learning model adopts a double-in and double-out MMoE network architecture, a special MMoE network is arranged at each post, and each post network shares a characteristic cross layer; each MMoE network realizes cross-domain recommendation through two special post Tower networks, wherein one is responsible for the recommendation and learning of a target post, the other is responsible for the migration of post sharing information, and a final prediction result of each post is obtained through each post fusion layer; setting weights according to performances of different MMoE networks at different posts, and adjusting the weights through an optimized loss function; the feature crossing layer in the step S804 adopts a plurality of feature crossing modes, including Concat, Inner-product, and Outer-product, wherein the Inner-product operation is an Inner product operation of vectors, and the Outer-product operation is a pairwise crossing of each dimension of input feature vectors to generate a feature crossing matrix; the Tower network adopts a common full-connection layer, and carries out prediction according to the specific position representation transmitted by the gating network.
Wherein the loss function adopts a two-classification cross entropy calculation method and is an MMoE network loss function at each post and an integrated MMoE network loss function LensembleThe sum of (W) and the regularization term R (θ) is specifically:
wherein k is used for marking the input and output of the MMoE network at different positions, M is the number of the positions, N is the total defect number of learning, and W is the weight of the MMoE network at each position.
The method is further described below by taking as an example a physical quality defect in the cigarette manufacturing process.
In this example, physical quality defects are shown in fig. 9, and the defect is a false count in a cigarette pack, which is classified as a type C quality defect.
When the defective cigarette shown in fig. 9 is conveyed from the conveying belt to the position below the shooting camera, the cigarette is shot by the camera to form a shot image, the shot image is converted into an electronic signal through the image AD, the image is collected by the image collecting module, and then the electronic signal of the collected image is filtered, enhanced, smoothed and sharpened by the image preprocessing module, so that the image is clear and the image quality is ensured.
Then, the image feature extraction module performs self-training and feature extraction on the preprocessed image by using a convolutional neural network to form an image feature signal, and the image feature signal formed by the image feature extraction module is compared with the standard quality storage module by using the feature comparison module to obtain comparison data. The data analysis grading module analyzes and grades the comparison data obtained by the characteristic comparison module, and the data analysis grading module classifies the defects into a C-type defect module.
The defect is finally stored in a defect data memory, when a learner learns, the learner logs in a learning system through a first login verification unit, the learning system can be matched with the learning content through automatic calculation of a recommended learning module, and then a defect example display module, a reason generation module, a previous defect analysis module and a scheme solution module of a first display a defect image, a generation reason and a solution, wherein the main contents are as follows:
and defect numbering: QX00-00-00A-001
Defect name: staggered branch in box
Frequency of occurrence of defects: c type (Low frequency)
And (3) defect description:
cigarettes which do not conform to the brand and specification of the small box appear in the cigarette box.
The standard requires that:
the national standard is as follows: the box is not mistakenly assembled, and the quality of the A class is defective.
Company standards: the box is not misloaded, and the class A quality is defective.
Factory standards are as follows: the box is not misloaded, and the class A quality is defective.
Analysis of main reasons:
the cigarette is abnormal.
The cigarettes of the original production brand are not cleaned up after the brand is changed.
The checking and adjusting method comprises the following steps:
when the cards are changed, all cigarette conveying channels and cigarettes in the waste cigarette tray need to be cleaned.
History recording:
2021/11/1 user: ***
The reason is as follows: ***
Machine number: ***
The description is as follows: ***
2021/11/5 user: ***
The reason is as follows: ***
Machine number: ***
The following steps are described: ***
By learning the defects in the cigarette production, the learner can learn the collected cases with various physical quality and appearance defects, so that the skill of workers and the quality management level of cigarette production enterprises are improved.
Although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the spirit and scope of the invention as defined in the appended claims. The techniques, shapes, and configurations not described in detail in the present invention are all known techniques.
Claims (10)
1. Big data-based cigarette product quality defect prevention and control learning system is characterized in that: the device comprises a data acquisition unit, a data processing unit, a data storage unit and a learning unit;
the data acquisition unit comprises a camera arranged on a cigarette production line;
the data processing unit comprises a first processor, the first processor is provided with an analysis system, and the analysis system comprises an image processing unit and a data analysis unit;
the data storage unit comprises a standard quality feature memory, a shot image feature memory and a defect data memory which are electrically connected with the first processor, and the defect data memory comprises a historical quality accident module, an A-type defect storage module, a B-type defect storage module and a C-type defect storage module;
the learning unit comprises a second processor, the second processor is electrically connected with a first display, a voice playing module and a first control module, and a multimedia interface is arranged on the second processor; the second processor is provided with a learning system, the learning system comprises a first login verification unit and a retrieval unit, the retrieval unit comprises an A-type defect retrieval module, a B-type defect retrieval module, a C-type defect retrieval module and a recommended learning module, and the A-type defect retrieval module, the B-type defect retrieval module, the C-type defect retrieval module and the recommended learning module respectively comprise a defect example display module, a reason generation module, a history defect analysis module and a scheme solution module capable of displaying solutions of troubleshooting and equipment adjustment.
2. The big data-based cigarette product quality defect prevention and control learning system according to claim 1, wherein: the image processing unit comprises an image acquisition module for acquiring images shot by the camera, an image preprocessing module for preprocessing the images and an image characteristic extraction module for extracting the characteristics of the images; the data analysis unit comprises a feature comparison module for comparing the extracted wrapping features with a standard quality storage module, a data statistics module for counting the total detection number and the defect number, and a data analysis grading module for analyzing and classifying the generated quality defects; the data analysis grading module comprises a non-defective module, a type A defective module, a type B defective module and a type C defective module.
3. The big-data-based cigarette product quality defect prevention and control learning system according to claim 2, wherein: the camera is electrically connected with a first communication sending module for transmitting images shot by the camera, the first processor is electrically connected with a first communication receiving module and a second communication sending module, the first communication sending module is in communication connection with the first communication receiving module, the second processor is electrically connected with a second communication receiving module, and the second communication receiving module is in communication connection with the second communication sending module.
4. The big-data-based cigarette product quality defect prevention and control learning system according to claim 3, wherein: the data acquisition unit further comprises a second display, the second display is electrically connected with the first processor, a USB interface and a second control module are arranged on the second display, the second control module is electrically connected with the first processor, and the analysis system further comprises a second login verification module.
5. The big-data-based cigarette product quality defect prevention and control learning system according to claim 4, wherein: first treater still electricity is connected with board display screen, production team display screen, maintenance team display screen, quality testing group display screen and workshop management display screen, all install the buzzer siren on board display screen, production team display screen, maintenance team display screen, the quality testing group display screen and the workshop management display screen.
6. The big-data-based cigarette product quality defect prevention and control learning system according to claim 5, wherein: the learning unit further comprises a support frame, the first display is fixedly mounted on the support frame through screws, the support frame comprises a support plate, a telescopic rod is fixedly connected onto the support plate, a base is arranged at one end, away from the support plate, of the telescopic rod, the telescopic rod comprises a first support rod and a second support rod, the first support rod and the second support rod are both of hollow structures, the first support rod is slidably arranged in the hollow structure of the second support rod, a fixing nut is fixedly connected into the hollow structure of the first support rod, a screw rod is connected into the fixing nut in a threaded manner, a servo motor is fixedly mounted at the hollow bottom of the second support rod, and an output shaft of the servo motor is fixedly connected with the screw rod; the servo motor is electrically connected with the second processor.
7. A cigarette product quality defect prevention and control learning method based on big data is characterized in that: the method comprises the following steps:
s1, the camera shoots the cigarettes on the production line to form a shot image, the image is transmitted to the image processing unit through the first communication transmitting module and the first communication receiving module, and the image is collected by the image collecting module;
s2, manually inputting the cigarette defect problem images found in work into a defect data storage through a USB interface and a second control module, and filling the whole defect storage analysis database;
s3, the image preprocessing module carries out filtering, enhancing, smoothing and sharpening on the collected electronic image signals to make the images clear and ensure the image quality;
s4, the image feature extraction module performs self-training and feature extraction on the preprocessed image by using a convolutional neural network to form an image feature signal;
s5, comparing the image characteristic signal formed by the image characteristic extraction module with the standard quality storage module by using the characteristic comparison module to obtain comparison data;
s6, the data analysis grading module analyzes and grades the comparison data obtained by the characteristic comparison module, and the grading is respectively as follows: the defect-free method comprises the steps of defect-free defect, class A defect-free defect-type B and defect-type C defect-free defect;
s7, the data statistics module counts the total detection number, the A-type defect number, the B-type defect number and the C-type defect number and stores the data in the defect data memory;
s8, the learner logs in the learning system through the first login verification unit, searches through the A-type defect search module, the B-type defect search module, the C-type defect search module or the recommended learning module of the search unit, and displays the defect image, the generation reason and the solution for the learner to learn through the first display by the defect example display module, the reason generation module, the previous defect analysis module and the solution module.
8. The big-data-based cigarette product quality defect prevention and control learning method according to claim 7, characterized in that: the method for recommending the learning module to automatically match the learning content in the step S8 includes the following steps:
s801, constructing a data set comprising learner embedding, learner post information, multiple defect image embedding, different post defect image sharing context embedding and marked click behaviors, and obtaining post embedding of each post specific user by adopting a hierarchical difference attention algorithm;
s802, embedding and fusing the individual embedding of the user with the learner post information of different posts respectively by adopting a double-gate control network responsible for controlling the information inflow;
s803, according to different post defect image embedding, each type of context embedding is assigned with a weight, and the context embedding of the corresponding field is obtained through weighting fusion;
s804, inputting the individual embedding, the context embedding and the defect image embedding of the learners of each post into a shared characteristic cross type double MMoE network, and screening to obtain the recommended learning cases of each post based on the dynamic learning integration strategy.
9. The big-data-based cigarette product quality defect prevention and control learning method according to claim 8, characterized in that: in the step S804, a multitask learning model MMoE is adopted for post-crossing recommendation, the multitask learning model adopts a double-in and double-out MMoE network architecture, a special MMoE network is arranged at each post, and the post networks share a feature cross layer; each MMoE network realizes cross-domain recommendation through two special post Tower networks, wherein one is responsible for the recommendation and learning of a target post, the other is responsible for the migration of post sharing information, and a final prediction result of each post is obtained through each post fusion layer; setting weights according to performances of different MMoE networks at different posts, and adjusting the weights through an optimized loss function; the feature crossing layer in the step S804 adopts a plurality of feature crossing modes, including Concat, Inner-product, and Outer-product, wherein the Inner-product operation is an Inner product operation of vectors, and the Outer-product operation is a pairwise crossing of each dimension of input feature vectors to generate a feature crossing matrix; the Tower network adopts a common full-connection layer, and carries out prediction according to the specific position representation transmitted by the gating network.
10. The big-data-based cigarette product quality defect prevention and control learning method according to claim 9, characterized in that: the loss function in the step S804 adopts a two-classification cross entropy calculation method to perform MMoE network loss function and integrated MMoE network loss function L for each postensembleThe sum of (W) and R (θ) is specifically:
wherein k is used for marking the input and output of the MMoE network at different positions, M is the number of the positions, N is the total defect number of learning, W is the weight of the MMoE network at each position, and R (theta) represents a regularization term.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210333292.5A CN114708584B (en) | 2022-03-31 | 2022-03-31 | Big data-based cigarette product quality defect prevention and control learning system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210333292.5A CN114708584B (en) | 2022-03-31 | 2022-03-31 | Big data-based cigarette product quality defect prevention and control learning system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114708584A true CN114708584A (en) | 2022-07-05 |
| CN114708584B CN114708584B (en) | 2024-07-23 |
Family
ID=82171383
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210333292.5A Active CN114708584B (en) | 2022-03-31 | 2022-03-31 | Big data-based cigarette product quality defect prevention and control learning system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114708584B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114998070A (en) * | 2022-08-03 | 2022-09-02 | 广州远程教育中心有限公司 | Immersive digital learning method and system for enterprise digital learning platform |
| CN115586744A (en) * | 2022-12-12 | 2023-01-10 | 南京专注智能科技股份有限公司 | GD packagine machine cigarette intelligent analysis detecting system based on big data |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112328646A (en) * | 2021-01-04 | 2021-02-05 | 平安科技（深圳）有限公司 | Multitask course recommendation method and device, computer equipment and storage medium |
| CN112905648A (en) * | 2021-02-04 | 2021-06-04 | 北京邮电大学 | Multi-target recommendation method and system based on multi-task learning |
| CN113670931A (en) * | 2021-08-09 | 2021-11-19 | 中冶南方工程技术有限公司 | Steel plate surface defect detection method and system based on neural network |
| CN114022469A (en) * | 2021-11-15 | 2022-02-08 | 云南大学 | Cigarette appearance defect image classification method and system |
-
2022
- 2022-03-31 CN CN202210333292.5A patent/CN114708584B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112328646A (en) * | 2021-01-04 | 2021-02-05 | 平安科技（深圳）有限公司 | Multitask course recommendation method and device, computer equipment and storage medium |
| CN112905648A (en) * | 2021-02-04 | 2021-06-04 | 北京邮电大学 | Multi-target recommendation method and system based on multi-task learning |
| CN113670931A (en) * | 2021-08-09 | 2021-11-19 | 中冶南方工程技术有限公司 | Steel plate surface defect detection method and system based on neural network |
| CN114022469A (en) * | 2021-11-15 | 2022-02-08 | 云南大学 | Cigarette appearance defect image classification method and system |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114998070A (en) * | 2022-08-03 | 2022-09-02 | 广州远程教育中心有限公司 | Immersive digital learning method and system for enterprise digital learning platform |
| CN115586744A (en) * | 2022-12-12 | 2023-01-10 | 南京专注智能科技股份有限公司 | GD packagine machine cigarette intelligent analysis detecting system based on big data |
| CN115586744B (en) * | 2022-12-12 | 2023-03-14 | 南京专注智能科技股份有限公司 | GD packagine machine cigarette intelligent analysis detecting system based on big data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114708584B (en) | 2024-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114708584B (en) | Big data-based cigarette product quality defect prevention and control learning system and method | |
| CN109165582B (en) | Urban street garbage detection and cleanliness assessment method | |
| WO2017084186A1 (en) | System and method for automatic monitoring and intelligent analysis of flexible circuit board manufacturing process | |
| CN101105459A (en) | Empty bottle mouth defect inspection method and device | |
| CN112966772A (en) | Multi-person online image semi-automatic labeling method and system | |
| CN111160432A (en) | Automatic classification method and system for panel production defects | |
| DE112019005951T5 (en) | Centralized analysis of multiple visual inspection devices | |
| CN106645180A (en) | Method for checking defects of substrate glass, field terminal and server | |
| CN114858806A (en) | Cigarette quality inspection data analysis system and method | |
| CN111680738A (en) | Screening device for apple quality detection and detection method thereof | |
| CN117876329A (en) | Real-time road disease detection method based on radar, video and data analysis | |
| CN115830403B (en) | Automatic defect classification system and method based on deep learning | |
| CN112836724A (en) | Object defect recognition model training method and device, electronic equipment and storage medium | |
| CN115309871B (en) | Industrial big data processing method and system based on artificial intelligence algorithm | |
| CN114820461A (en) | Surface flaw quality inspection platform and method based on image recognition algorithm | |
| CN113850773A (en) | Detection method, device, equipment and computer readable storage medium | |
| CN115908272A (en) | Method and system for automatically detecting belt tearing state based on vision technology | |
| CN115266744A (en) | Detection system and method for product in production line | |
| CN113192068A (en) | AI visual inspection system for printed wiring board | |
| CN113642473A (en) | Mining coal machine state identification method based on computer vision | |
| CN113837178A (en) | Deep learning-based automatic positioning and unified segmentation method for meter of transformer substation | |
| CN110816902A (en) | System, real-time detection method and medium for detecting redundant materials in satellite structure platform hole | |
| CN115984833A (en) | Method and system for automatically identifying inclined wire network based on deep learning | |
| CN117434068B (en) | Chip detection method and system | |
| CN117250201B (en) | Swimming pool brick layer defect detection method and system |
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 | ||
| GR01 | Patent grant |