Cigarette packaging machine running health condition early warning method and system based on visual detection

Abstract

The application provides a cigarette packaging machine fault diagnosis method based on visual detection, takes cigarette package defects as a breakthrough point, acquires a defective cigarette package image, classifies the defective cigarette package image by using a defect classification model, obtains a cigarette package defect category, and provides a cigarette packaging machine fault diagnosis method based on visual detection to accurately determine the equipment fault factors by analyzing the contribution degree relationship between the cigarette package defect category and the equipment fault factors within a period of time. The application also provides a cigarette packaging machine operation health condition early warning method and system based on visual detection, which organically connects the data of three time dimensions of real-time defect automatic identification of the cigarette packaging machine, short-period equipment fault factor intelligent investigation, and long-period machine operation health condition monitoring by using the defect detection data of the appearance quality of the cigarette package, forms a complete set of digital solutions, and realizes the intelligentization of the whole process of the operation health condition monitoring of the cigarette packaging machine.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of cigarette packaging machine monitoring and early warning, in particular, relates to a cigarette packaging machine operation health condition early warning method and system based on visual detection. BACKGROUND

[0002] Cigarette packaging is an important link in cigarette production. In order to ensure that the cigarette product meets the quality standard requirements and reduce the number of downtime maintenance, the cigarette production enterprise formulates a daily maintenance strategy for the cigarette packaging machine on the one hand, and carries out daily cleaning and equipment adjustment on the machine. On the other hand, a high-speed camera is used to detect the appearance quality of the cigarette package in real time, and downtime maintenance is carried out in time when a large number of defects occur. Since the mainstream packaging machine type has realized high-speed production at present, reducing downtime and maintaining high-speed and stable production of the machine is an important measure to improve the efficiency of cigarette production.

[0003] The existing patent (application number: CN201911120706.0, publication (announcement) number: CN110816938A) relates to a big data analysis method based on a cigarette packaging machine comprehensive detection platform, which includes the steps of obtaining original data, establishing a database, cleaning data, analyzing data and displaying results. However, this patent needs to collect a large amount of real-time detection data on the front line, such as system running state, alarm information, display information, rejection information, detection method type, detection quantity, defect quantity, defect proportion, detection image and detection object data. In addition, the literature "Multi-working condition process fault monitoring and diagnosis method of super-high-speed small box packaging machine" selects the servo motor current, servo motor temperature, hot melt adhesive temperature, vehicle speed and other sensing indexes of the small box packaging machine as the model input, thereby establishing a cigarette packaging machine fault monitoring and diagnosis method. It can be seen that the existing method mainly uses sensing indexes and system running information to diagnose the faults of the cigarette packaging machine.

[0004] In fact, the cigarette packaging process covers the folding and forming of inner liner paper, inner frame paper and trademark paper, and the wrapping of transparent paper, which involves a large number of mechanical components, making it difficult to monitor the status of each component in real time. In addition, during the cigarette packaging process, multiple cigarette appearance defect categories may occur within a certain period of time. In addition to some defect categories having clear equipment failure factors, most cigarette appearance defect categories often correspond to multiple potential equipment failure factors. How to quickly and clearly identify the most likely equipment failure factor causing the cigarette defect category has great guiding significance for later maintenance.

[0005] In addition, the current cigarette production enterprises lack the means of perception for the potential shutdown hidden danger existing in the production and operation of the cigarette packaging machine. Although the fault category can be identified through fault diagnosis when the fault occurs, and the defect can be quickly removed, due to the complex correlation between the defect category and the equipment fault factor, it is difficult to effectively evaluate the running health status of the machine through the defect category.

[0006] In order to solve the above problems, people have been seeking an ideal technical solution. SUMMARY

[0007] The purpose of the present application is to overcome the shortcomings of the prior art. The first aspect of the present application provides a cigarette packaging machine fault diagnosis method based on visual detection, which identifies the most likely equipment fault factor causing the cigarette package defect category according to the defect cigarette package image within a period of time, and provides guidance for later maintenance.

[0008] The second aspect of the present application provides a cigarette packaging machine running health status early warning method and system based on visual detection. The defect detection data of the appearance quality of the cigarette package are used to automatically identify the real-time defects of the cigarette packaging machine, intelligently investigate the short-period equipment fault factors, and monitor the long-period machine running health status. The data of the three time dimensions are organically connected to form a complete set of digital solutions, and the intelligentization of the whole process of monitoring the running health status of the cigarette packaging machine is realized.

[0009] In order to achieve the above purpose, the first aspect of the present application provides a cigarette packaging machine fault diagnosis method based on visual detection, which includes the following steps:

[0010] Obtain the defect cigarette package image, classify the defect cigarette package image by using the defect classification model, and obtain the cigarette package defect category;

[0011] According to the defect fault knowledge graph, determine the equipment fault factor associated with the cigarette package defect category;

[0012] Statistically analyze the composition of the cigarette package defect category set in the defect analysis time window;

[0013] According to the defect cause analysis model, calculate the traceability score representing the relationship between each equipment fault factor and the cigarette package defect category set;

[0014] Sort the traceability scores of all equipment fault factors, and select the top M equipment fault factors with traceability scores reaching the set threshold and the highest scores as the target equipment fault factors.

[0015] The second aspect of the present application provides a cigarette packaging machine running health status early warning method based on visual detection, which includes the following steps:

[0016] An image of a defective cigarette packet is acquired, a defect classification model is used to classify the image of the defective cigarette packet, and a cigarette packet defect category is obtained;

[0017] According to the defect fault knowledge graph, the device fault factors associated with the cigarette packet defect category are determined;

[0018] The cigarette packet defect categories in a defect analysis time window are counted to form a cigarette packet defect category set;

[0019] According to the defect cause analysis model, a traceability score representing the relationship between each device fault factor and the cigarette packet defect category set is calculated;

[0020] Based on the traceability scores of all device fault factors in the early warning time window, the running health status of the cigarette packaging machine is predicted.

[0021] The application also provides a cigarette packaging machine fault diagnosis system based on visual detection, comprising an image acquisition module, a visual detection module, a device fault factor query module, a device fault factor traceability score calculation module, and a fault diagnosis module;

[0022] The image acquisition module is arranged at the front of the original defect removal device on the cigarette packet runway, and is used to shoot the surface of the cigarette packet from different angles in a multi-position mode to acquire an image of a defective cigarette packet;

[0023] The visual detection module is internally provided with a defect classification model, and is used to identify the cigarette packet defect category in the image of the defective cigarette packet;

[0024] The device fault factor query module is internally provided with a defect fault knowledge graph, and is used to determine the device fault factors associated with the cigarette packet defect category according to the defect fault knowledge graph, wherein the defect fault knowledge graph comprises an association mapping between the cigarette packet defect category and the device fault factor;

[0025] The device fault factor traceability score calculation module is internally provided with a defect cause analysis model, and is used to count the cigarette packet defect categories in a defect analysis time window to form a cigarette packet defect category set, and calculate a traceability score representing the relationship between each device fault factor and the cigarette packet defect category set according to the defect cause analysis model;

[0026] The fault diagnosis module is used to sort the traceability scores of all device fault factors, and select the first M device fault factors with the highest scores and reaching a set threshold value as target device fault factors.

[0027] The application also provides a cigarette packaging machine running health status early warning system based on visual detection, comprising an image acquisition module, a visual detection module, a device fault factor query module, a device fault factor traceability score calculation module, and a running health status early warning module;

[0028] The image acquisition module is arranged at the front of the original defect removing device, and is used for shooting the surface of the cigarette packet from different angles in a multi-station mode to obtain a defective cigarette packet image.

[0029] The visual detection module is internally provided with a defect classification model, and is used for identifying the defect category of the defective cigarette packet image.

[0030] The equipment failure factor query module is internally provided with a defect failure knowledge graph, and is used for determining the equipment failure factor associated with the cigarette defect category according to the defect failure knowledge graph.

[0031] The equipment failure factor traceability score calculation module is internally provided with a defect cause analysis model, and is used for counting the cigarette defect categories in the defect analysis time window to form a cigarette defect category set, and calculating a traceability score representing the relationship between each equipment failure factor and the cigarette defect category set according to the defect cause analysis model.

[0032] The operation health condition early warning module is internally provided with a cigarette packaging machine operation health condition early warning model, and is used for predicting the operation health condition of the cigarette packaging machine based on all the equipment failure factor traceability scores in the early warning time window.

[0033] The application further provides an electronic device, comprising:

[0034] at least one processor; and

[0035] a memory in communication with the at least one processor; wherein

[0036] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the aforementioned equipment failure factor analysis recommendation method, or execute the aforementioned visual detection-based cigarette packaging machine equipment failure factor prediction method, or execute the aforementioned visual detection-based cigarette packaging machine operation health condition early warning method.

[0037] The application further provides a non-transitory computer readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the aforementioned visual detection-based cigarette packaging machine failure diagnosis method, or execute the aforementioned visual detection-based cigarette packaging machine operation health condition early warning method.

[0038] The present application has outstanding substantial features and significant progress compared with the prior art, specifically, the present application takes the cigarette package defect as the breakthrough point, obtains the defect cigarette package image, classifies the defect cigarette package image by using the defect classification model, obtains the cigarette package defect category, and provides a cigarette packaging machine fault diagnosis method based on visual detection by analyzing the contribution degree relationship between the cigarette package defect category and the equipment fault factor in a period of time, to accurately determine the equipment fault factor, and provides a new method and new idea for diagnosing the equipment fault factor of the cigarette packaging machine by using digital means.

[0039] The present application also provides a cigarette packaging machine operation health condition early warning method based on visual detection, after calculating the traceability score representing the relationship between each equipment fault factor and the cigarette package defect category set according to the defect cause analysis model, predicting the cigarette packaging machine operation health condition based on the traceability scores of all equipment fault factors in the early warning time window; so that the data in three time dimensions of real-time cigarette package defect automatic identification, short-period equipment fault factor intelligent investigation and long-period machine operation health condition monitoring are organically linked, forming a complete set of digital solutions, which has important significance for improving the digital utilization level of the cigarette industry enterprise. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of the cigarette packaging machine fault diagnosis method based on visual detection described in embodiment 1.

[0041] Figure 2 is a topological structure diagram of the first equipment fault factor and the cigarette package defect category.

[0042] Figure 3 is a topological structure diagram of the second equipment fault factor and the cigarette package defect category.

[0043] Figure 4 is a topological structure diagram of the third equipment fault factor and the cigarette package defect category.

[0044] Figure 5 is a camera layout diagram of the image acquisition device in embodiment 1.

[0045] Figure 6 is a schematic diagram of the equipment fault factor system in embodiment 1.

[0046] Figure 7 is a part of the equipment defect knowledge graph in embodiment 1.

[0047] Figure 8 is a flowchart of the cigarette packaging machine operation health condition early warning method based on visual detection described in embodiment 2.

[0048] Figure 9is a principle block diagram of a cigarette packer running health condition early warning system of Example 3.

[0049] Figure 10 is a principle block diagram of a cigarette packer running health condition early warning system of Example 4. DETAILED DESCRIPTION

[0050] The technical solutions of the present application are described in further detail below through a specific embodiment.

[0051] Example 1

[0052] As shown in Figure 1 The present embodiment provides a cigarette packer fault diagnosis method based on visual detection, comprising the following steps:

[0053] Step 1, acquire defective cigarette package images, classify the defective cigarette package images using a defect classification model, and obtain cigarette package defect categories.

[0054] It can be understood that if there are available defective cigarette package images, they can be used directly; if there are no available defective cigarette package images, they need to be acquired through collection. Specifically, the steps of acquiring defective cigarette package images include:

[0055] Step 1.1, set up an image collection device to collect suspected defect image data;

[0056] Select a cigarette packer in production as the data collection machine, in order to avoid experimental influence on normal production, set up a multi-camera image collection device in front of the original defect removal device on the cigarette package runway, use the current mainstream method in this field, and shoot the surface of the cigarette package from different angles in a multi-camera mode. Specifically, the image collection device is composed of a high-speed camera, a lens, a light source, image processing software and other components, and also includes a photographing trigger, cigarette package separation and other auxiliary devices.

[0057] It should be noted that the data collection device is parallel to the normal defect removal device of the machine, and this device is only used for image collection and does not remove defects. While collecting data, it does not affect the normal production of cigarettes.

[0058] The defect detection equipment in normal production often needs to balance between precision and recall. In order to obtain more defect images, the detection method is set in a way of reducing precision in exchange for recall, suspected defect images are collected from the data collection machine, so that the deep learning model in the later stage can capture as much defective cigarette package data as possible, and filter most of the qualified cigarette package data.

[0059] The collection method of the suspected defect image mainly uses gray scale detection, detects the frame by dividing the image template, judges whether the gray scale of all pixel points in the detection frame exceeds the threshold value, and if the gray scale in any detection frame exceeds the detection limit to a certain area, the image is determined as a suspected defect cigarette package image.

[0060] The specific steps are as follows:

[0061] Collecting qualified cigarette package images of each camera image collection device as an image template;

[0062] During the data collection process, each to-be-detected image is detected based on the image template, and if the to-be-detected image has an abnormal detection frame, the to-be-detected image is determined as a suspected defect image.

[0063] The detection steps of the abnormal detection frame include:

[0064] Based on the image template, a detection frame is divided in the image area where defects may occur in each to-be-detected image, so that the detection area covers the entire cigarette package area as much as possible; wherein the shape of the detection frame is any closed geometric figure;

[0065] Set the gray scale detection limit of each detection frame, and filter out abnormal pixel points that do not meet the qualified cigarette package pixel range requirement according to the gray scale detection limit. It can be understood that the gray scale detection limit is a gray scale range;

[0066] Set the area detection limit of each detection frame, and filter out abnormal detection frames whose total area of abnormal pixel points exceeds the area detection limit according to the area detection limit. It can be understood that the area detection limit is a pixel area range.

[0067] Further, in order to improve the detection effect of small defects, small detection frames can be appropriately used, and the range of the area detection limit can be reduced.

[0068] Step 1.2, data screening and labeling

[0069] Because the detection rules of suspected defect cigarette package images are relatively broad, there are many qualified cigarette package images in the collected data. In order to provide training data for subsequent deep learning models, data screening and labeling are performed by manual labeling. The specific method is as follows:

[0070] Divide the defects into N sub-cigarette defect categories, and set a typical reference image and division rule for each cigarette defect category, wherein the division rule includes but is not limited to the characteristics and causes based on the defect image;

[0071] Check the suspected defect cigarette package image, and filter out the defect image data;

[0072] According to the division rule and the typical reference image, the defect image data is classified and labeled.

[0073] Further, while performing the classification labeling, the defect image data is also labeled for defect features. The labeling for the defect features includes bounding box labeling and defect feature name labeling. Unlike the cigarette packet defect categories, the defect features include more specific defect descriptions, such as skewed pasting, corner damage, and the like.

[0074] It can be understood that, in order to provide training data for subsequent deep learning models, defect image data of defective cigarette packets needs to be collected extensively first. If there is already available defect image data of defective cigarette packets, the step of “Step 1.2, screening and labeling data” can be directly performed.

[0075] It can be understood that, after the defect image data of defective cigarette packets is collected extensively, the training of the defect classification model can be performed, and the specific steps include:

[0076] 1.3, data set construction

[0077] The labeled defect image data is mixed with a certain amount of qualified cigarette packet data to form a data set composed of N+1 categories;

[0078] The data set is divided into a test set, a training set, and a validation set. The specific steps include: reserving a certain amount of data as the test set, and dividing the remaining data into the training set and the validation set. Optionally, the k-fold method can be used to dynamically divide the training set and the validation set.

[0079] 1.4, model construction

[0080] A defect classification model based on a deep neural network is constructed. Specifically, a convolutional neural network framework such as ResNet, ShuffleNet, MobileNet, etc. can be used as a prediction model, and then a pre-trained deep neural network is used as an initial defect classification model.

[0081] When pre-training the deep neural network, model gain techniques such as Batch normalization, Drop out, etc. can be used. Of course, common operations in this field such as changing the model framework, loss function, and nonlinear unit should also be included in the protection scope of the present application.

[0082] Step 1.5, model optimization

[0083] After the data set construction, model construction, and loss function determination are completed, the mathematical model of the deep learning task is also determined, and then a suitable optimizer (Optimizer) is selected to optimize the deep learning model.

[0084] The initial defect classification model is trained and optimized using the training set and the validation set obtained in step 1.3. The optimization objects include data preprocessing methods, model hyperparameters, model structures, etc. The test set is used to test and evaluate the model effect, and the model with the best effect is selected as the best defect classification model.

[0085] The training and optimization process includes combination testing of model hyperparameters based on search algorithms, and the optimal parameter combination is selected as the final model parameters.

[0086] Step 1.6, model training

[0087] The model weights of the best defect classification model obtained in step 1.5 are trained based on all the collected defect image data to obtain the final defect classification model.

[0088] It should be noted that in the visual detection-based cigarette packaging paper fault diagnosis process, in addition to using image classification technology to classify or predict the defect cigarette package image, target detection technology can also be used to detect and locate the defect target of the defect cigarette package image.

[0089] Therefore, in step 1.4, a cigarette package defect target detection model based on a deep neural network can also be constructed. Specifically, the cigarette package defect target detection model can use a single-stage model framework such as SSD, YOLO, or a two-stage model framework such as Faster R-CNN, including related model training techniques. After the model optimization and model training steps in steps 1.5 and 1.6, the final cigarette package defect target detection model is obtained.

[0090] Further, after obtaining the trained defect classification model, the image acquisition device built in step 1.1 is used to collect real-time cigarette package surface images, and the defect cigarette package images are obtained through the screening and labeling data steps. Then, the defect classification model is used to classify the defect cigarette package images to obtain the cigarette package defect category.

[0091] Step 2, according to the defect fault knowledge graph, determine the device fault factors associated with the cigarette package defect category.

[0092] It can be understood that the execution of step 2 depends on the constructed defect fault knowledge graph, and the construction steps of the defect fault knowledge graph include:

[0093] Step 2.1, based on the cigarette packaging process and mechanical components, according to the principle of independence between equipment failure factors, the cigarette packaging process is divided into multiple potential mechanical parts; for each mechanical part, each failure link is labeled according to the failure form, forming an equipment failure factor system, each equipment failure factor includes a mechanical part and a failure form.

[0094] Further, each equipment failure factor has a corresponding maintenance strategy, so the equipment failure factor system can also include a maintenance strategy corresponding to each equipment failure factor.

[0095] For example, a cigarette packaging machine, based on the cigarette packaging process and mechanical components, according to the principle of independence between equipment failure factors, the cigarette packaging process is divided into X potential mechanical parts, each equipment failure factor has a clear corresponding mechanical part. According to the possible failure forms of each equipment failure factor, each failure link is labeled, such as the transmission belt link may have a failure form of too fast or too slow speed, the clamping link may have a failure form of clamping too tight or too loose. The maintenance strategy corresponds to an equipment failure factor, such as the maintenance strategy for the transmission belt link is to adjust the transmission belt speed. Thus, a cigarette packaging defect factor system is formed.

[0096] It should be noted that in step 2.1, a defect classification model based on a deep neural network or a cigarette package defect target detection model based on a deep neural network is used, regardless of which model, feature extraction is required first, and then image classification or target detection is performed based on the extracted features. Therefore, in step 2.2, in addition to being able to predict the cigarette packaging machine equipment failure factor based on the equipment defect knowledge graph composed of cigarette package defect categories and equipment failure factors, a knowledge graph can also be constructed based on defect features, and finally the cigarette packaging machine equipment failure factor is predicted based on the constructed knowledge graph.

[0097] Step 2.2, an expert team divides the equipment failure factors for each cigarette package defect category, one cigarette package defect category corresponds to one or more equipment failure factors, for each equipment failure factor, according to its possibility of causing cigarette package defects, the equipment failure factor is divided into multiple grades, and a quantitative score is given. For example, the possibility is divided into "high, medium, low" three grades, corresponding to (3, 2, 1) three grades of possibility quantitative score.

[0098] Step 2.3, establish the mapping relationship between the cigarette package defect category, the possibility quantitative score and the equipment failure factor, and form an equipment defect knowledge graph. For example: "cigarette package defect category - possibility quantitative score - equipment failure factor", "equipment failure factor - maintenance strategy" and other relationships, wherein each entity can also serve as a system.

[0099] Step 3: Analyze the defect categories of cigarette packs within the time window of the defect analysis to form a set of cigarette pack defect categories.

[0100] Step 4: Calculate the source tracing score based on the defect cause analysis model to represent the relationship between each equipment failure factor and the set of cigarette pack defect categories.

[0101] Let J be the set of equipment failure factors within the defect analysis time window, I be the set of cigarette pack defect categories related to the set of equipment failure factors J, and K be the set of cigarette pack defect categories within the defect analysis time window, where K∈I.

[0102] Therefore, the defect cause analysis model is D. j *Q j *U j ;

[0103] D j The direct contribution factor to the failure of the j-th device is calculated using the following formula:

[0104] Q j The importance contribution factor for the j-th equipment failure factor is calculated using the following formula:

[0105] U j The necessary contribution factor for the failure of the j-th device is calculated using the following formula:

[0106] In the formula, i represents the i-th defect category of the cigarette pack in the defect category set, k represents the k-th defect category of the cigarette pack within the defect analysis time window, n represents the total number of defect categories of the cigarette pack within the defect analysis time window, and e kj This represents the correlation between the i-th cigarette pack defect category and the j-th equipment failure factor; e kj This represents the correlation between the k-th cigarette pack defect category and the j-th equipment failure factor within the defect analysis time window.

[0107] Furthermore, e ij and e kj To directly quantify the score, in specific implementation, the probability quantification score between the cigarette pack defect category and the equipment failure factor in step 2.2 can be directly used as the direct quantification score.

[0108] To facilitate understanding, the design concept of each contributing factor in the defect cause analysis model is given.

[0109] by Figure 2 Taking the topological structure of equipment failure factors and cigarette pack defect categories as an example, the direct contribution factor D is given. j The design concept.

[0110] Figure 2 In the middle, there are three equipment failure factors, and two defect categories, that is, the equipment failure factor set in the defect analysis time window is J={j=1, j=2, j=3}, the cigarette defect category set related to the equipment failure factor set J is I={i=1, i=2}, and the cigarette defect category set in the defect analysis time window is K={k=1, k=2}; and e 11 =1, e 12 =3, e 13 =1, e 21 =3, e 23 =1.

[0111] The index design principle is that the possibility quantification score of cigarette defect category 1 to equipment failure factor 3 is "1", and the corresponding possibility level is "low". Similarly, the possibility quantification score of cigarette defect category 2 to equipment failure factor 3 is also "1". However, because the equipment failure factors corresponding to cigarette defect category 1 are more than those corresponding to cigarette defect category 2, the possibility of causing equipment failure factor 3 when cigarette defect category 2 appears is greater than that when cigarette defect category 1 appears.

[0112] At the same time, among the appearing cigarette defect categories, the equipment failure factors associated with more cigarette defect categories are more likely to actually appear, so the possibility of equipment failure factor 1 appearing is greater than that of equipment failure factor 3.

[0113] Therefore, the design The dimension of is 0 to 1;

[0114] According to the above formula, it is calculated that:

[0115]

[0116]

[0117]

[0118]

[0119]

[0120]

[0121] It can be seen that E 23 is higher than E 13 , that is, the possibility of causing equipment failure factor 3 when cigarette defect category 2 appears is greater than that when cigarette defect category 1 appears; D (j＝1) is greater than D (j＝3)This means that equipment failure factor 1 is more likely to occur than equipment failure factor 3. Clearly, the designed direct contribution factor D index can reflect this logic well.

[0122] by Figure 3 Taking the topological structure of equipment failure factors and cigarette pack defect categories as an example, the importance contribution factor Q is given. j The design concept.

[0123] Figure 3 In the process, a single defect analysis time window includes two equipment failure factors and two cigarette pack defect categories; and e 11 =1,e 12 =1,e 21 =1,e 22 =1; The other two equipment failure factors also correspond to three other cigarette pack defect categories, namely, the set of equipment failure factors within the defect analysis time window is J = {j = 1, j = 2}, the set of cigarette pack defect categories related to the set of equipment failure factors J is I = {i = 1, i = 2, i = 3, i = 4, i = 5}, and the set of cigarette pack defect categories within the defect analysis time window is K = {k = 1, k = 2}.

[0124] Explanation of the indicator design principle: The probability quantification score for cigarette pack defect category 1 regarding equipment failure factor 1 and equipment failure factor 2 is "1". Similarly, the probability quantification score for cigarette pack defect category 2 regarding equipment failure factor 1 and equipment failure factor 2 is also "1". This might lead to the conclusion that equipment failure factor 1 and equipment failure factor 2 have the same actual probability of occurrence. However, considering that equipment failure factor 2 also corresponds to cigarette pack defect categories 3, 4, and 5, and none of these cigarette pack defect categories occurred in the first preset time, the probability of equipment failure factor 1 occurring is considered higher.

[0125] Therefore, the design importance contribution factor The dimensions of the indicator are from 0 to 1.

[0126] The calculation is obtained based on the above formula:

[0127]

[0128]

[0129] That is, P(j=1|k=1,k=2) and P(j=2|k=1,k=2), but P(j=1|i=1,i=2,i≠3,i≠4,i≠5) is greater than P(j=2|i=1,i=2,i≠3,i≠4,i≠5), therefore Q (j＝1) Greater than Q (j＝2)The probability of equipment failure factor 1 occurring is greater than that of equipment failure factor 2. Obviously, the designed importance contribution factor Q index can better reflect the above logic.

[0130] by Figure 4 Taking the topological structure of equipment failure factors and cigarette pack defect categories as an example, the necessity contribution factor U is given. j The design concept.

[0131] Figure 4 In this context, a defect analysis time window includes two equipment failure factors and two cigarette pack defect categories; that is, the set of equipment failure factors within the defect analysis time window is J = {j = 1, j = 2}, the set of cigarette pack defect categories related to the set of equipment failure factors J is I = {i = 1, i = 2}, and the set of cigarette pack defect categories within the defect analysis time window is K = {k = 1, k = 2}; and e 11 =1e 21 =1,e 22 =1.

[0132] Explanation of the indicator design principle: Since cigarette pack defect category 1 is only associated with equipment failure factor 1, equipment failure factor 1 is a necessary and sufficient condition for cigarette pack defect category. Relatively speaking, if equipment failure factor 1 occurs, it may lead to a situation where cigarette pack defect category 1 and cigarette pack defect category 2 occur simultaneously, while the occurrence of equipment failure factor 2 may only lead to cigarette pack defect category 2. Therefore, in terms of necessity, equipment failure factor 1 should be more likely to occur than equipment failure factor 2.

[0133] Therefore, the design necessity contribution factor The dimensions of the indicator are from 0 to 1.

[0134] The calculation is obtained based on the above formula:

[0135]

[0136]

[0137] It can be seen that U (j＝1) Greater than U (j＝2) That is, the probability of equipment failure factor 1 occurring is greater than that of equipment failure factor 2, and the designed necessity contribution factor U can better reflect the above logic.

[0138] As an explanation of the design scheme, D j Q j U j The probable relationships between cigarette pack defect categories and equipment failure factors are characterized from different levels, and E kj (D j (components) Q j Uj , Q j , U j , respectively, are all 0 to 1, and have certain measurement rationality.

[0139] In use, D j , Q j , and U j are calculated according to the above formula, respectively, and then the traceability score of each potential equipment failure factor in the defect analysis time window is calculated based on the formula D j *Q j *U j .

[0140] In an embodiment, the defect cause analysis model can also be αD j *βQ j *λU j , wherein α, β, and λ respectively represent different weights, and α+β+λ=1.

[0141] In another embodiment, the defect cause analysis model can also be f1(D j )*f2(Q j )*f3(U j ), wherein f1(), f2(), and f3() represent arbitrary functions of D j , Q j , and U j , which can be the same or different.

[0142] It can be understood that, regardless of the form, since the traceability score is related to D j , Q j , and U j , i.e., the topological structure of all possible equipment failure factors and cigarette defect categories is analyzed for the cigarette defect category and the equipment failure factor, the maximum probability equipment failure factor corresponding to the cigarette defect category can be accurately found from the complex mapping relationship network of the cigarette defect category and the equipment failure factor, facilitating later fault screening and equipment maintenance.

[0143] Step 5: Sort the traceability scores of all equipment failure factors, and select the top M equipment failure factors with traceability scores reaching a set threshold and the highest scores as target equipment failure factors.

[0144] For ease of understanding, this embodiment takes a ZB45 cigarette packaging machine group as an example to introduce the implementation process of Embodiment 1 in detail, and the specific steps are as follows:

[0145] Step one: Select the ZB45 cigarette packaging machine in production as the data acquisition machine. Image acquisition equipment is installed in front of the original defect removal device to collect images of the cigarette packets. Preferably, the image acquisition equipment uses four high-speed color cameras with 130 million resolution to capture images of five sides of the cigarette packets. The camera layout is shown in Figure 5 .

[0146] Step two: Collect qualified cigarette packet images from each camera as image templates. Based on the image templates, detection boxes are set in the image area where defects may occur, the gray detection limit and area detection limit of each detection box are set, and a suspected defect image data acquisition model is constructed.

[0147] Step three: Based on the suspected defect image data acquisition model constructed in step three, each image to be detected is detected. If the image has an abnormal detection box, it is determined that the image is a suspected defect image, and the data collection is stopped until the number of collected data meets the later model training.

[0148] Step four: Check the collected suspected defect images and select defect image data. The defects are divided into 23 categories of cigarette packet defects, and typical reference images and division rules are set for each category of cigarette packet defects.

[0149] Accordingly, each defect cigarette packet image is labeled by category to form a defect cigarette packet data set.

[0150] The cigarette packet defect category division rules are shown in the following table:

[0151] Table 1 Cigarette packet defect categories

[0152]

[0153]

[0154] Step five: Mix the defect cigarette packet data and qualified cigarette packet data at a ratio of 1:10 to form a data set consisting of 24 categories. Use 10% of the data as the test set, and use the remaining data to dynamically divide the training set and the validation set in a 5-fold manner.

[0155] Step six: Based on the PyTorch programming language development framework, a ResNet50 convolutional neural network is constructed as a classification model.

[0156] Step seven: Train and optimize the model based on the training set and the validation set. The optimization objects include data preprocessing methods, model hyperparameters, model structures, etc. The test set is used to test and evaluate the model effect, and the parameter combination with the best effect is selected as the final model parameters.

[0157] Step eight: based on all the collected defective cigarette packet data, the model weight is trained to obtain a defect classification model.

[0158] Step nine: according to the specific mechanical part and fault form causing the cigarette packet defect, the device fault factor system is constructed, and the maintenance strategy of each device fault factor is set, as shown in the table. Figure 6

[0159] Step ten: according to the device fault factor system constructed in step nine, the device fault factors of each defect are divided by an expert group. One defect may be divided into several different device fault factors. For each potential device fault factor, the expert group needs to mark it according to its possibility of causing defects, according to the three levels of "high, medium and low", which correspond to the three levels of possibility quantization scores (3, 2, 1).

[0160] Step eleven: a knowledge graph composed of cigarette defect categories, device fault factors, maintenance strategies and the like is constructed, as shown in the table. Figure 7

[0161] Step twelve: device fault factor analysis recommendation. Set the time window to 1 hour, calculate the traceability score D of each device fault factor according to the defect cause analysis model described in the embodiment, and take the top 3 device fault factors with scores exceeding 10 and the highest scores as the target device fault factors. j j j

[0162] Finally, the following device fault factors are output by the model: device fault factor 4 appears in the third defect analysis time window, and device fault factor 1 and device fault factor 2 appear in the sixth time window.

[0163]

[0164] The present application takes cigarette defects as the starting point, obtains defective cigarette packet images, classifies the defective cigarette packet images by using a defect classification model, obtains cigarette defect categories, and analyzes the contribution degree relationship between the cigarette defect categories and the device fault factors within a period of time, to provide a cigarette packaging machine fault diagnosis method based on visual detection, to accurately determine the device fault factors, and to provide a new method and new idea for diagnosing the device fault factors of the cigarette packaging machine by using digital means.

[0165] Example 2

[0166] The present embodiment provides a cigarette packaging machine running health condition early warning method based on visual detection, as shown in the table, which includes the following steps: Figure 8

[0167] ​​​​​​Step 2.1, acquire the defective cigarette package image, classify the defective cigarette package image by using the defect classification model, and obtain the cigarette package defect category;

[0168] Step 2.2, according to the defect fault knowledge graph, determine the device fault factors associated with the cigarette package defect category;

[0169] Step 2.3, statistics of the cigarette defect category in the defect analysis time window form a cigarette defect category set;

[0170] Step 2.4, according to the defect cause analysis model, calculate the traceability score representing the relationship between each device fault factor and the cigarette defect category set;

[0171] Step 2.5, based on the traceability scores of all device fault factors in the early warning time window, predict the running health status of the cigarette packaging machine.

[0172] Among them, the steps of steps 2.1-2.4 are consistent with steps 1.1-1.4 in embodiment 1, which will not be repeated here.

[0173] The specific steps of step 2.5 are as follows:

[0174] Step 2.5.1, determine the device fault factor level of the machine according to the floating range of the device fault factor of the current machine;

[0175] Because the machine conditions of each cigarette packaging machine are different, the occurrence frequency of different device fault factors is also different, and there is a certain floating range of device fault factors for different machines, so the floating range is used to judge the level of device fault factors.

[0176] Step 2.5.2, sequentially statistics the traceability scores of each device fault factor in the early warning time window;

[0177] Set the early warning time window, the early warning time window is a certain time interval, such as a shift, a week, a month; Among them, the early warning time window needs to contain multiple defect analysis time windows, and the traceability scores of each device fault factor in the early warning time window are sequentially counted.

[0178] It should be noted that since the purpose of analyzing the running health status of the machine is to find the long-term trend, a longer time can be used as a early warning time window in actual production, such as using 1 hour as a defect analysis time window and 1 week (about 78 hours of production time) as a early warning time window. When the traceability score exceeds the first fixed threshold range or the dynamic threshold range, it is determined that the early warning requirement is met, and the early warning is performed.

[0179] Step 2.5.3, when the traceability score exceeds the fixed threshold range or the dynamic threshold range, it is determined that the early warning requirement is met, and the early warning is performed.

[0180] In order to determine the normal traceability score fluctuation range, two methods can be adopted, one is a fixed threshold range, and the other is a dynamic threshold range. The fixed threshold range can be set according to experience, or analyzed according to a certain amount of equipment failure factor traceability score historical data to obtain the approximate distribution range under normal circumstances, and the distribution range is exceeded to give a warning. The dynamic threshold range needs to be calculated according to the historical data in the recent period, and the threshold range is obtained, for example, according to the traceability score of the equipment failure factor in the L warning time windows before the current warning time window to form a normal distribution, and the normal distribution range is exceeded to give a warning.

[0181] In actual operation, after obtaining the traceability score, the traceability score fluctuation trend can also be obtained according to the traceability score, and a warning can be given according to the traceability score fluctuation trend.

[0182] The steps of obtaining the traceability score fluctuation trend score according to the traceability score are as follows:

[0183] Divide the L warning time windows before the current warning time window into R fluctuation intervals, and denote the set of fluctuation intervals as S;

[0184] Calculate the traceability score sum of each fluctuation interval;

[0185] Calculate the gradient change value of the traceability score sum between different fluctuation intervals:

[0186]

[0187] The fluctuation trend is evaluated by comparing the gradient change of the average equipment failure factor traceability score between different fluctuation intervals, and the fluctuation trend score is obtained.

[0188] Taking the division into 2 fluctuation intervals as an example, the traceability score sum in and the traceability score sum in are calculated, denoted as s1 and s2, and then the gradient is calculated as the fluctuation trend score.

[0189] It can be understood that within a period of time, if a certain equipment failure factor frequently appears or shows a significant growth trend, this phenomenon may be related to the maintenance of the equipment, or some mechanical parts need to be replaced in time. The present embodiment analyzes the equipment failure factors in time series, analyzes the running health status of the cigarette packaging machine from two aspects of the traceability score value range and the fluctuation trend of the equipment failure factors, so as to give an early warning to the potential fault hidden danger, and further reduce the equipment downtime caused by the fault.

[0190] Further, in the actual application process, when the traceability score exceeds the first fixed threshold range or the dynamic threshold range and the fluctuation trend score exceeds the second fixed threshold range, it is determined that the early warning requirement is met, early warning is performed, and the early warning accuracy is improved.

[0191] For the convenience of understanding, still taking the ZB45 cigarette packaging machine group described in the embodiment as an example, the implementation process of embodiment 2 is introduced in detail, and the traceability score D of each equipment failure factor is obtained by using steps one to twelve in embodiment 1 j *Q j *U j .

[0192] Step thirteen: According to the specific situation of the machine, the machine equipment failure factor benchmark is constructed, such as setting the early warning time window to 1 week, the comprehensive traceability score range threshold of the equipment failure factor “stacking conveying belt roll contamination” is 0 to 15, the fluctuation trend of the comprehensive traceability score is the gradient of the current early warning time window and the previous time window, and the threshold is set to 50%. When the equipment failure factor meets the conditions of exceeding the range of 0 to 15 and exceeding 50% in growth trend at the same time, early warning is performed.

[0193] In this embodiment, every 2 defect analysis time windows are recorded as one early warning time window. According to the set condition that early warning is performed when the equipment failure factor meets the conditions of exceeding the range of 0 to 15 and exceeding 50% in growth trend at the same time, the comprehensive traceability score of the equipment failure factor according to the early warning time window is shown in the following table:

[0194]

[0195]

[0196] The corresponding early warning situation is shown in the following table:

[0197]

[0198] After calculating the traceability score representing the relationship between each equipment failure factor and the set of cigarette defect categories according to the defect cause analysis model, the present application predicts the running health status of the cigarette packaging machine based on the traceability scores of all equipment failure factors within the early warning time window; thereby organically concatenating the data in three time dimensions of real-time cigarette defect automatic discrimination, short-period equipment failure factor intelligent investigation, and long-period machine running health status monitoring, forming a complete set of digital solutions, which has important significance for improving the digital utilization level of cigarette industry enterprises.

[0199] Embodiment 3

[0200] The present embodiment provides a cigarette packaging machine fault diagnosis system based on visual detection, which comprises a visual detection device, a data processing device and a fault diagnosis device. Figure 9As shown, it comprises an image acquisition module, a visual detection module, a device fault factor query module, a device fault factor traceability score calculation module, and a fault diagnosis module.

[0201] The image acquisition module is arranged at the front of the original defect removal device and is used to capture images of the surface of the cigarette packet from different angles in a multi-camera mode to obtain defective cigarette packet images.

[0202] The visual detection module has a built-in defect classification model and is used to identify the defect type of the defective cigarette packet image.

[0203] The device fault factor query module has a built-in defect fault knowledge graph and is used to determine the device fault factors associated with the defect type of the cigarette packet according to the defect fault knowledge graph.

[0204] The device fault factor traceability score calculation module has a built-in defect cause analysis model and is used to count the defective cigarette packet types in the defect analysis time window to form a defective cigarette packet type set, and calculate the traceability score representing the relationship between each device fault factor and the defective cigarette packet type set according to the defect cause analysis model.

[0205] The fault diagnosis module is used to sort the traceability scores of all device fault factors, and select the top M device fault factors with the highest scores that reach a set threshold as the target device fault factors.

[0206] The defect classification model, the defect fault knowledge graph, and the defect cause analysis model can all refer to the technical solutions disclosed in Embodiment 1 or Embodiment 2.

[0207] Embodiment 4

[0208] This embodiment provides a cigarette packaging machine running health condition early warning system based on visual detection, as shown in Figure 10 As shown, it comprises an image acquisition module, a visual detection module, a device fault factor query module, a device fault factor traceability score calculation module, and a running health condition early warning module.

[0209] The image acquisition module is arranged at the front of the original defect removal device and is used to capture images of the surface of the cigarette packet from different angles in a multi-camera mode to obtain defective cigarette packet images.

[0210] The visual detection module has a built-in defect classification model and is used to identify the defect type of the defective cigarette packet image.

[0211] The device failure factor query module has a defect failure knowledge graph, and is configured to determine the device failure factor associated with the cigarette defect category according to the defect failure knowledge graph.

[0212] The device failure factor traceability score calculation module has a defect cause analysis model, and is configured to count the cigarette defect category set composed of the cigarette defect categories in the defect analysis time window, and calculate the traceability score representing the relationship between each device failure factor and the cigarette defect category set according to the defect cause analysis model.

[0213] The operation health condition early warning module has a cigarette packaging machine operation health condition early warning model, and is configured to predict the operation health condition of the cigarette packaging machine based on the traceability scores of all device failure factors in the early warning time window.

[0214] It can be understood that the visual detection module, the device failure factor query module, the device failure factor traceability score calculation module, and the operation health condition early warning module all include corresponding hardware devices.

[0215] Similarly, the defect classification model, the defect failure knowledge graph, and the defect cause analysis model can all refer to the technical solutions disclosed in Embodiment 1 or Embodiment 2.

[0216] Embodiment 5

[0217] The embodiment also provides an electronic device, which includes:

[0218] at least one processor; and

[0219] a memory in communication with the at least one processor; wherein

[0220] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the visual detection-based cigarette packaging machine failure diagnosis method of Embodiment 1 or execute the visual detection-based cigarette packaging machine operation health condition early warning method of Embodiment 2.

[0221] Embodiment 6

[0222] The embodiment provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the visual detection-based cigarette packaging machine failure diagnosis method of Embodiment 1 or execute the visual detection-based cigarette packaging machine operation health condition early warning method of Embodiment 2.

[0223] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not to limit the present application; although the present application has been described in detail with reference to the preferred embodiments, it is understood by those skilled in the art that the specific embodiments of the present application can be modified or some technical features can be replaced by equivalent ones; without departing from the spirit of the technical solutions of the present application, all of which should be covered in the technical solution range of the present application claimed by the present application.

Claims

1. A method for diagnosing a fault of a cigarette packing machine based on visual inspection, characterized by, The method comprises the following steps: obtaining a defective cigarette packet image, classifying the defective cigarette packet image by using a defect classification model, and obtaining a cigarette packet defect category; determining device fault factors associated with the cigarette packet defect category according to a defect fault knowledge graph; counting cigarette packet defect categories in a defect analysis time window to form a cigarette packet defect category set; calculating a traceability score representing the relationship between each device fault factor and the cigarette packet defect category set according to a defect cause analysis model; sorting the traceability scores of all device fault factors, and selecting the top M device fault factors with traceability scores reaching a set threshold and the highest scores as target device fault factors; The defect cause analysis model is D j *Q j *U j ; D j The direct contribution factor for the jth equipment failure factor is calculated as: Q j The importance contribution factor of the jth device failure factor is calculated by the following formula: U j The necessity contribution factor for the jth device failure factor is calculated by the following formula: In the formula, i represents the i th cigarette packet defect category in the set of cigarette packet defect categories, k represents the k th cigarette packet defect category in the defect analysis time window, n represents the total number of cigarette packet defect categories in the defect analysis time window, e ij represents the correlation degree between the i th cigarette packet defect category and the j th equipment failure factor; e kj represents the correlation degree between the k th cigarette packet defect category in the defect analysis time window and the j th equipment failure factor.

2. The method according to claim 1, wherein, The construction steps of the defect fault knowledge graph comprise: based on the cigarette packaging process and the mechanical components, the cigarette packaging process is divided into multiple potential mechanical parts according to the principle of independence between device fault factors; for each mechanical part, each fault link is labeled according to the fault form to form a device fault factor system, and each device fault factor includes a mechanical part and a fault form; an expert group divides device fault factors for each cigarette packet defect category, one cigarette packet defect category corresponds to one or more device fault factors, for each device fault factor, according to its possibility of causing cigarette packet defects, the device fault factors are divided into multiple grades, and a quantitative score is given; a mapping relationship between the cigarette packet defect category, the quantitative score and the device fault factor is established to form a device defect knowledge graph.

3. A cigarette packaging machine running health condition early warning method based on visual detection, characterized in that: obtaining a defective cigarette packet image, classifying the defective cigarette packet image by using a defect classification model, and obtaining a cigarette packet defect category; determining device fault factors associated with the cigarette packet defect category according to a defect fault knowledge graph; counting cigarette packet defect categories in a defect analysis time window to form a cigarette packet defect category set; calculating a traceability score representing the relationship between each device fault factor and the cigarette packet defect category set according to a defect cause analysis model; predicting the running health condition of the cigarette packaging machine based on the traceability scores of all device fault factors in the early warning time window; The defect cause analysis model is D j *Q j *U j ; D j The direct contribution factor for the jth equipment failure factor is calculated as: Q j The importance contribution factor for the jth equipment failure factor is calculated by the formula: U j The necessity contribution factor for the jth device failure factor is calculated by the following formula: In the formula, i represents the i th cigarette packet defect category in the set of cigarette packet defect categories, k represents the k th cigarette packet defect category in the defect analysis time window, n represents the total number of cigarette packet defect categories in the defect analysis time window, e ij represents the correlation degree between the i th cigarette packet defect category and the j th equipment failure factor; e kj represents the correlation degree between the k th cigarette packet defect category in the defect analysis time window and the j th equipment failure factor.

4. The method according to claim 3, wherein, The construction steps of the defect fault knowledge graph comprise: based on the cigarette packaging process and the mechanical components, the cigarette packaging process is divided into multiple potential mechanical parts according to the principle of independence between device fault factors; for each mechanical part, each fault link is labeled according to the fault form to form a device fault factor system, and each device fault factor includes a mechanical part and a fault form; an expert group divides device fault factors for each cigarette packet defect category, one cigarette packet defect category corresponds to one or more device fault factors, for each device fault factor, according to its possibility of causing cigarette packet defects, the device fault factors are divided into multiple grades, and a quantitative score is given; a mapping relationship between the cigarette packet defect category, the quantitative score and the device fault factor is established to form a device defect knowledge graph.

5. The method according to claim 3, wherein the method further comprises: determining the running health status of the cigarette packaging machine based on the obtained image data. The specific steps of predicting the running health condition of the cigarette packaging machine based on the traceability scores of all device fault factors in the early warning time window are as follows: Determine the equipment failure factor level of the machine according to the floating range of the equipment failure factor of the current machine; In sequence, the traceability score of each equipment failure factor in the early warning time window is counted; When the traceability score exceeds the first fixed threshold range or the dynamic threshold range, it is determined that the early warning requirement is met, and early warning is performed; Alternatively, according to the traceability score, a fluctuation trend score is obtained, and when the traceability score exceeds the first fixed threshold range or the dynamic threshold range, it is determined that the early warning requirement is met, and early warning is performed; Or, when the traceability score exceeds the first fixed threshold range or the dynamic threshold range and the fluctuation trend score exceeds the second fixed threshold range, it is determined that the early warning requirement is met, and early warning is performed.

6. The method according to claim 5, wherein, The steps of obtaining the traceability score fluctuation trend score according to the traceability score are as follows: Divide the L early warning time windows before the current early warning time window into R fluctuation intervals, and denote the set of fluctuation intervals as S; Calculate the sum of the traceability scores of each fluctuation interval; Calculate the gradient change value of the sum of the traceability scores between different fluctuation intervals: By comparing the gradient change of the average equipment failure factor traceability score between different fluctuation intervals, the fluctuation trend is evaluated, and the fluctuation trend score is obtained.

7. A visual inspection based fault diagnosis system for a cigarette packer, the system comprising: a camera for capturing images of a cigarette packer; a computer for processing the images; and a database for storing information about the cigarette packer. It comprises an image acquisition module, a visual detection module, an equipment failure factor query module, an equipment failure factor traceability score calculation module, and a fault diagnosis module; The image acquisition module is arranged at the front of the original defect removal device and is used to capture images of the surface of the cigarette packet from different angles in a multi-camera mode to obtain defective cigarette packet images. The visual detection module is internally provided with a defect classification model and is used to identify the defect categories of the defective cigarette packet images. The equipment failure factor query module is internally provided with a defect fault knowledge graph and is used to determine the equipment failure factors associated with the defect categories of the cigarette packets according to the defect fault knowledge graph. The equipment failure factor traceability score calculation module is internally provided with a defect cause analysis model and is used to count the defect categories of the defective cigarette packets in the defect analysis time window to form a defective cigarette packet category set, and calculate the traceability scores representing the relationship between each equipment failure factor and the defective cigarette packet category set according to the defect cause analysis model. The fault diagnosis module is used to sort the traceability scores of all equipment failure factors, and select the top M equipment failure factors with the highest scores that reach a set threshold as target equipment failure factors. The defect cause analysis model is D j *Q j *U j ; D j The direct contribution factor for the jth equipment failure factor is calculated as: Q j The importance contribution factor for the jth device failure factor is calculated by the formula: U j The necessity contribution factor for the jth device failure factor is calculated by the following formula: In the formula, i represents the i th cigarette packet defect category in the set of cigarette packet defect categories, k represents the k th cigarette packet defect category in the defect analysis time window, n represents the total number of cigarette packet defect categories in the defect analysis time window, e ij represents the correlation degree between the i th cigarette packet defect category and the j th equipment failure factor; e kj represents the correlation degree between the k th cigarette packet defect category in the defect analysis time window and the j th equipment failure factor.

8. A visual inspection based pre-warning system for the running health condition of a cigarette packer, characterized in that: It comprises an image acquisition module, a visual detection module, an equipment failure factor query module, an equipment failure factor traceability score calculation module, and a running health status early warning module; The image acquisition module is arranged at the front of the original defect removal device and is used to capture images of the surface of the cigarette packet from different angles in a multi-camera mode to obtain defective cigarette packet images. The visual detection module is internally provided with a defect classification model and is used to identify the defect categories of the defective cigarette packet images. The equipment failure factor query module is internally provided with a defect fault knowledge graph and is used to determine the equipment failure factors associated with the defect categories of the cigarette packets according to the defect fault knowledge graph. The device failure factor traceability score calculation module has a defect cause analysis model built in, is used for counting a cigarette defect category set composed of cigarette defect categories in a defect analysis time window, and calculates traceability scores representing the relationship between each device failure factor and the cigarette defect category set according to the defect cause analysis model; The defect cause analysis model is D j *Q j *U j ; D j The direct contribution factor for the jth equipment failure factor is calculated as: Q j The importance contribution factor for the jth device failure factor is calculated by the formula: U j The necessity contribution factor for the jth device failure factor is calculated by the following formula: wherein i represents the i-th cigarette packet defect category in the set of cigarette packet defect categories, k represents the k-th cigarette packet defect category in the defect analysis time window, n represents the total number of cigarette packet defect categories in the defect analysis time window, e ij represents the correlation degree between the i-th cigarette packet defect category and the j-th equipment failure factor; e kj represents the correlation degree between the k-th cigarette packet defect category in the defect analysis time window and the j-th equipment failure factor; The operation health condition early warning module has a cigarette packaging machine operation health condition early warning model built in, and is used for predicting the operation health condition of the cigarette packaging machine based on all device failure factor traceability scores in a warning time window. 9.An electronic device, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the visual detection-based cigarette packaging machine failure diagnosis method of any one of claims 1 to 2, or execute the visual detection-based cigarette packaging machine operation health condition early warning method of any one of claims 3 to 6.

10. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to execute the visual detection-based cigarette packaging machine failure diagnosis method of any one of claims 1 to 2, or execute the visual detection-based cigarette packaging machine operation health condition early warning method of any one of claims 3 to 6.

Images