A method for detecting missing bars by dynamic weighing

Abstract

The invention relates to the field of automated detection technology, and in particular to a dynamic weighing missing cigarette strip detection method. First, a hybrid model is designed and trained; a multidimensional data acquisition system is used to collect the total weight of a cigarette box, infrared profiles of cigarette strips, and information on the compactness and neatness of cigarette strips obtained through a vibration sensor in real time; data collected by the multidimensional data acquisition system are preprocessed; feature extraction technology is used to extract key features of the preprocessed data; the extracted features are input into a constructed hybrid model; the hybrid model determines whether the cigarette box state is abnormal based on the input features; if an abnormality occurs, an alarm is triggered and displayed on a user interface, and if there is no abnormality, no operation is performed, thereby solving the technical problems of low detection efficiency, susceptibility to human factors, insufficient accuracy and high risk of missed detection in the prior art cigarette strip quality detection method.

Description

Technical Field

[0001] The invention relates to the technical field of automated detection, and in particular to a dynamic weighing missing strip detection method. Background Art

[0002] In the tobacco processing and logistics industry, cigarette boxes are important carriers for product transportation and storage. The quality of the cigarette rods inside is directly related to the overall quality of the product and consumer satisfaction. Traditional cigarette rod quality inspection methods mainly include manual sampling and simple weight sensor inspection. However, both methods have significant limitations in practical applications.

[0003] Although manual sampling can visually check the arrangement and missing status of cigarette boxes, it is limited by the limited human resources and the differences in subjective judgment. Its detection efficiency is low, and the test results are easily affected by human factors, making it difficult to ensure accuracy and stability.

[0004] On the other hand, although simple weight sensors can measure the total weight of cigarette boxes, they cannot accurately identify the specific number of cigarette sticks or the missing ones. When the weight of cigarette sticks changes slightly due to factors such as packaging materials, production processes or vibrations during transportation, weight sensors often cannot make accurate judgments, increasing the risk of missed detection.

[0005] In summary, the existing cigarette rod quality inspection methods have the problems of low inspection efficiency, susceptibility to human factors, insufficient accuracy and high risk of missed inspections. Summary of the invention

[0006] The purpose of the present invention is to provide a dynamic weighing missing cigarette rod detection method, aiming to solve the technical problems of low detection efficiency, susceptibility to human factors, insufficient accuracy and high risk of missed detection in the existing cigarette rod quality detection methods.

[0007] To achieve the above object, the present invention adopts a dynamic weighing missing strip detection method, comprising the following steps:

[0008] First, a hybrid model consisting of a convolutional neural network (CNN) and a recurrent neural network (RNN) is designed and trained;

[0009] A multi-dimensional data acquisition system is used to collect in real time the total weight of the cigarette box, the infrared profile of the cigarette rods (reflecting the arrangement), and the compactness and arrangement of the cigarette rods obtained through vibration sensors;

[0010] Preprocessing the data collected by the multi-dimensional data acquisition system;

[0011] Use feature extraction technology to extract key features from preprocessed data;

[0012] Inputting the extracted features into the constructed hybrid model;

[0013] The hybrid model determines whether the smoke box status is abnormal based on the input features;

[0014] If an abnormality occurs, an alarm is triggered and displayed on the user interface. If there is no abnormality, no operation is performed.

[0015] Among them, when designing and training a hybrid model that includes a convolutional neural network and a recurrent neural network:

[0016] Use historical data sets for model training, including data from normal cigarette boxes and cigarette boxes with known missing cigarettes, and optimize model parameters through the back propagation algorithm;

[0017] The convolutional neural network part is responsible for processing image data, and the recurrent neural network part is responsible for processing time series data.

[0018] The multi-dimensional data acquisition system is constructed using high-precision sensors and infrared imaging technology, and the collected data is transmitted to the pre-processing module through a transmission module (such as Wi-Fi, Bluetooth, etc.).

[0019] Among them, when preprocessing the data collected by the multidimensional data acquisition system:

[0020] First, the data of all dimensions are time-aligned according to the data collection timestamp to ensure the consistency of the data at the same time point;

[0021] And adopt algorithms such as Kalman filtering or median filtering to smooth weight data and vibration signals and remove noise interference;

[0022] At the same time, the infrared profile image is normalized by pixel value, and the weight and vibration data are range normalized to ensure that the data are comparable in different dimensions.

[0023] Among them, when using feature extraction technology to extract key features from preprocessed data:

[0024] The convolutional layer and pooling layer of the convolutional neural network (CNN) are used to extract features from the infrared profile image, and the image feature vector is output through the fully connected layer;

[0025] Use recurrent neural network (RNN) or its variants (such as LSTM, GRU) to perform time series analysis on weight and vibration signals and extract dynamic feature vectors;

[0026] At the same time, the mean, variance, maximum, minimum and other statistics of the weight data are calculated as supplementary features.

[0027] Among them, the hybrid model determines whether the smoke box state is abnormal based on the input features:

[0028] The convolutional neural network part is responsible for processing image data, and the recurrent neural network part is responsible for processing time series data;

[0029] The extracted comprehensive features are input into the hybrid model, and the model outputs a score indicating the degree of abnormality of the smoke box state. A threshold is set. If the score exceeds the threshold, the smoke box state is judged to be abnormal, otherwise, there is no abnormality.

[0030] Among them, when an abnormality occurs, it is transmitted to the storage module for storage in an encrypted manner.

[0031] Among them, the storage module adopts a distributed storage architecture:

[0032] When the detection system finds an anomaly, the abnormal data will be encrypted and sent to the metadata server through the client; after receiving the data, the metadata server will generate corresponding metadata and store it in its own database; at the same time, the metadata server will instruct the data server to store the abnormal data on the appropriate storage node.

[0033] A dynamic weighing missing cigarette strip detection method of the present invention, when used specifically, first designs and trains a hybrid model including a convolutional neural network and a recurrent neural network; adopts a multidimensional data acquisition system to collect in real time the total weight of a cigarette box, the infrared profile of cigarette strips, and the tightness and arrangement neatness information of cigarette strips obtained by a vibration sensor; pre-processes the data collected by the multi-dimensional data acquisition system; adopts feature extraction technology to extract key features of the pre-processed data; inputs the extracted features into the constructed hybrid model; the hybrid model determines whether the status of the cigarette box is abnormal based on the input features; if an abnormality occurs, an alarm is triggered and displayed on a user interface, and if there is no abnormality, no operation is performed, thereby solving the technical problems of low detection efficiency, susceptibility to human factors, insufficient accuracy and high risk of missed detection in the prior art cigarette strip quality detection method.

[0034] The present invention comprehensively considers multi-dimensional information such as weight, shape, vibration, etc., significantly improves the accuracy of detecting missing and abnormal arrangement of cigarette rods, and uses RNN to capture long-term dependencies in data to improve the accuracy of judging the compactness and neatness of arrangement of cigarette rods.

[0035] At the same time, it realizes real-time monitoring and instant feedback, shortens the detection cycle, reduces the need for manual intervention, and improves the automation and intelligence level of the overall logistics and production line. Through high-precision detection and real-time feedback mechanism, it effectively reduces missed detection and ensures product quality.

[0036] And when an anomaly is detected, the abnormal data is securely transmitted to the storage module using encrypted transmission. The storage module adopts a distributed storage architecture, which not only improves the storage efficiency and reliability of data, but also ensures the recoverability of data in the face of single point failures. This design effectively prevents data loss and damage, and provides a solid data foundation for subsequent anomaly analysis and processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 It is a flow chart of the steps of the dynamic weighing missing strip detection method of the present invention. DETAILED DESCRIPTION

[0039] Embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0040] See also Figure 1 , Figure 1 It is a flow chart of the steps of the dynamic weighing missing strip detection method of the present invention.

[0041] The present invention provides a dynamic weighing missing strip detection method, comprising the following steps:

[0042] S1. First, design and train a hybrid model consisting of a convolutional neural network (CNN) and a recurrent neural network (RNN);

[0043] For this specific implementation, historical data sets are used for model training, including data of normal cigarette boxes and cigarette boxes with known missing cigarettes, and the model parameters are optimized by back propagation algorithm;

[0044] The convolutional neural network part is responsible for processing image data, and the recurrent neural network part is responsible for processing time series data.

[0045] S2. Use a multi-dimensional data acquisition system to collect in real time the total weight of the cigarette box, the infrared profile of the cigarette rods (reflecting the arrangement), and the compactness and arrangement neatness of the cigarette rods obtained through a vibration sensor;

[0046] According to this specific implementation, the multi-dimensional data acquisition system is constructed using high-precision sensors and infrared imaging technology, and the acquired data is transmitted to the pre-processing module through a transmission module (such as Wi-Fi, Bluetooth, etc.).

[0047] The specific data collection methods are as follows:

[0048] Collection of total weight of cigarette boxes:

[0049] Sensor selection: Use a high-precision weight sensor that can monitor the total weight of the cigarette box in real time.

[0050] Installation method: Install the weight sensor at a key position on the cigarette box conveyor line to ensure that the cigarette box can accurately trigger the sensor during the conveying process.

[0051] Data collection: When the cigarette box passes the sensor, the sensor will collect the total weight data of the cigarette box in real time and transmit the data to the data processing unit for storage and analysis.

[0052] Collection of infrared profile of tobacco strips:

[0053] Sensor selection: Use infrared profile sensor, which can capture the profile information of the tobacco rod in real time using infrared light technology.

[0054] Installation method: Install the infrared profile sensor above or on the side of the cigarette rod conveyor line to ensure that the arrangement of the cigarette rods can be clearly captured.

[0055] Data collection: When the tobacco rod passes through the sensor, the sensor will collect the infrared profile data of the tobacco rod in real time, including the width, height, arrangement spacing and other information of the tobacco rod. These data will be transmitted to the data processing unit for further processing and analysis.

[0056] Collection of tobacco rod compactness and arrangement neatness:

[0057] Sensor selection: Use a vibration sensor, which can monitor the vibration of the tobacco rods during transportation in real time, thereby reflecting the compactness and neatness of the tobacco rods.

[0058] Installation method: Install the vibration sensor at the key position of the cigarette rod conveyor line to ensure that the vibration information of the cigarette rod can be accurately captured.

[0059] Data collection: When the cigarette rod passes through the sensor, the sensor will collect the vibration data of the cigarette rod in real time. Through data analysis, the compactness and neatness of the cigarette rod can be evaluated. Specifically, the frequency and amplitude of the vibration data can reflect the compactness of the cigarette rod, while the regularity and consistency of the vibration data can reflect the neatness of the cigarette rod.

[0060] In addition, all sensors are strictly calibrated and verified before use to ensure that their measurement accuracy and stability meet the design requirements.

[0061] Regularly maintain and inspect sensors to promptly detect and address potential faults or errors and ensure the continuity and accuracy of data collection.

[0062] At the same time, measures are taken to reduce the impact of environmental factors (such as temperature, humidity, electromagnetic interference, etc.) on sensor performance to ensure the stability and reliability of data acquisition.

[0063] Set up necessary protective facilities around the sensor installation location, such as heat insulation cover, anti-moisture coating, etc., to improve the sensor's anti-interference ability.

[0064] S3, preprocessing the data collected by the multidimensional data collection system;

[0065] For this specific implementation, firstly, the data of all dimensions are time-aligned according to the data acquisition timestamp to ensure the consistency of the data at the same time point;

[0066] And adopt algorithms such as Kalman filtering or median filtering to smooth weight data and vibration signals and remove noise interference;

[0067] At the same time, the infrared profile image is normalized by pixel value, and the weight and vibration data are range normalized to ensure that the data are comparable in different dimensions.

[0068] S4, using feature extraction technology to extract key features from the preprocessed data;

[0069] For this specific implementation, the convolution layer and pooling layer of the convolutional neural network (CNN) are used to extract features from the infrared profile image, and the image feature vector is output through the fully connected layer;

[0070] Use recurrent neural network (RNN) or its variants (such as LSTM, GRU) to perform time series analysis on weight and vibration signals and extract dynamic feature vectors;

[0071] At the same time, the mean, variance, maximum, minimum and other statistics of the weight data are calculated as supplementary features.

[0072] S5, inputting the extracted features into the constructed hybrid model;

[0073] S6, the hybrid model determines whether the smoke box state is abnormal based on the input features;

[0074] For this specific implementation, for this specific implementation, the convolutional neural network part is responsible for processing image data, and the recurrent neural network part processes time series data;

[0075] The extracted comprehensive features are input into the hybrid model, and the model outputs a score indicating the degree of abnormality of the smoke box state. A threshold is set. If the score exceeds the threshold, the smoke box state is judged to be abnormal, otherwise, there is no abnormality.

[0076] S7. If an abnormality occurs, an alarm is triggered and displayed on the user interface. If no abnormality occurs, no operation is performed.

[0077] According to this specific implementation, when an abnormality occurs, the data is transmitted to the storage module for storage in an encrypted manner.

[0078] When the detection system finds an anomaly, the abnormal data will be encrypted and sent to the metadata server through the client. After receiving the data, the metadata server will generate corresponding metadata and store it in its own database. At the same time, the metadata server will instruct the data server to store the abnormal data on the appropriate storage node.

[0079] When accessing abnormal data, the client sends a data access request to the metadata server. The metadata server finds the storage location of the data on the data server based on the metadata information in the request and returns it to the client. The client then directly accesses the storage node on the data server based on the returned location information to obtain the required data.

[0080] During the transmission process, the encrypted transmission method can use symmetric key encryption or asymmetric key encryption:

[0081] The symmetric key encryption transmission method is as follows:

[0082] Symmetric key encryption is a traditional encryption method that uses the same key to encrypt and decrypt data.

[0083] In the present invention, when the detection system finds an anomaly and is ready to transmit data, it generates a symmetric key and uses the key to encrypt the anomaly data.

[0084] The encrypted data is transmitted to the metadata server of the storage module through a secure communication channel (such as SSL / TLS protocol).

[0085] After receiving the encrypted data, the metadata server uses the same symmetric key to decrypt it to restore the original data.

[0086] Asymmetric key encryption transmission is as follows:

[0087] Asymmetric key encryption uses a pair of keys (public key and private key) to encrypt and decrypt data.

[0088] The detection system uses the private key to sign the abnormal data and uses the public key to encapsulate the encrypted data.

[0089] The encapsulated data is transmitted to the metadata server of the storage module through a secure communication channel.

[0090] After receiving the data, the metadata server uses the corresponding public key to verify the signature of the data and uses the private key to decrypt the data to restore the original data.

[0091] At the same time, data visualization technology is used to display the processed data in the form of charts, curves, etc., so that operators can intuitively understand the operating status and quality of the production line.

[0092] Generate quality reports regularly to summarize quality issues in the production process and the effects of improvement measures, providing support for the company's quality management and decision-making.

[0093] By analyzing historical data and trends, we can identify bottlenecks and problem points in the production process and put forward targeted optimization and improvement suggestions.

[0094] After implementing improvement measures, data is collected again and evaluated to verify the effectiveness of the improvement measures.

[0095] A dynamic weighing missing cigarette strip detection method according to the present embodiment is used. When it is used specifically, a hybrid model including a convolutional neural network and a recurrent neural network is first designed and trained; a multidimensional data acquisition system is used to collect in real time the total weight of the cigarette box, the infrared profile of the cigarette strips, and the information on the compactness and neatness of the cigarette strips obtained by the vibration sensor; the data collected by the multidimensional data acquisition system is preprocessed; feature extraction technology is used to extract key features of the preprocessed data; the extracted features are input into the constructed hybrid model; the hybrid model determines whether the status of the cigarette box is abnormal based on the input features; if an abnormality occurs, an alarm is triggered and displayed on the user interface, and if there is no abnormality, no operation is performed. In this way, the technical problems of low detection efficiency, susceptibility to human factors, insufficient accuracy and high risk of missed detection in the existing cigarette strip quality detection methods are solved.

[0096] The present invention comprehensively considers multi-dimensional information such as weight, shape, vibration, etc., significantly improves the accuracy of detecting missing and abnormal arrangement of cigarette rods, and uses RNN to capture long-term dependencies in data to improve the accuracy of judging the compactness and neatness of arrangement of cigarette rods.

[0097] At the same time, it realizes real-time monitoring and instant feedback, shortens the detection cycle, reduces the need for manual intervention, and improves the automation and intelligence level of the overall logistics and production line. Through high-precision detection and real-time feedback mechanism, it effectively reduces missed detection and ensures product quality.

[0098] And when an anomaly is detected, the abnormal data is securely transmitted to the storage module using encrypted transmission. The storage module adopts a distributed storage architecture, which not only improves the storage efficiency and reliability of data, but also ensures the recoverability of data in the face of single point failures. This design effectively prevents data loss and damage, and provides a solid data foundation for subsequent anomaly analysis and processing.

[0099] Adding sensor calibration and verification, data acquisition frequency adjustment, and implementation of environmental interference control will ensure high accuracy and stability of data acquisition, thereby providing more reliable quality inspection results.

[0100] The establishment of a real-time monitoring and early warning system enables operators to instantly obtain key parameter information in the production process and make adjustments quickly to effectively prevent quality problems. Optimizing multi-dimensional data collection and processing processes will not only significantly improve the quality inspection accuracy and efficiency of the cigarette box production line, but also improve production efficiency and cost-effectiveness.

[0101] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A dynamic weighing missing strip detection method, characterized in that: The steps include: First, a hybrid model consisting of a convolutional neural network and a recurrent neural network is designed and trained; A multi-dimensional data acquisition system is used to collect the total weight of the cigarette box, the infrared profile of the cigarette rods, and the compactness and arrangement of the cigarette rods obtained by the vibration sensor in real time; Preprocessing the data collected by the multi-dimensional data acquisition system; Use feature extraction technology to extract key features from preprocessed data; Inputting the extracted features into the constructed hybrid model; The hybrid model determines whether the smoke box status is abnormal based on the input features; If an abnormality occurs, an alarm is triggered and displayed on the user interface. If there is no abnormality, no operation is performed.

2. The dynamic weighing missing strip detection method according to claim 1, characterized in that: When designing and training a hybrid model that includes a convolutional neural network and a recurrent neural network: Use historical data sets for model training, including data from normal cigarette boxes and cigarette boxes with known missing cigarettes, and optimize model parameters through the back propagation algorithm; The convolutional neural network part is responsible for processing image data, and the recurrent neural network part is responsible for processing time series data.

3. The dynamic weighing missing strip detection method according to claim 2, characterized in that: The multi-dimensional data acquisition system is constructed by using high-precision sensors and infrared imaging technology, and the acquired data is transmitted to the pre-processing module through the transmission module.

4. The dynamic weighing missing strip detection method according to claim 3, characterized in that: When preprocessing the data collected by the multidimensional data acquisition system: First, the data of all dimensions are time-aligned according to the data collection timestamp to ensure the consistency of the data at the same time point; And Kalman filtering or median filtering algorithm is used to smooth the weight data and vibration signals to remove noise interference; At the same time, the infrared profile image is normalized by pixel value, and the weight and vibration data are range normalized to ensure that the data are comparable in different dimensions.

5. The dynamic weighing missing strip detection method according to claim 4, characterized in that: When using feature extraction technology to extract key features from preprocessed data: The convolutional layer and pooling layer of the convolutional neural network are used to extract features from the infrared profile image, and the image feature vector is output through the fully connected layer; Use recurrent neural network or its variants to perform time series analysis on weight and vibration signals and extract dynamic feature vectors; At the same time, the mean, variance, maximum and minimum statistics of the weight data are calculated as supplementary features.

6. The dynamic weighing missing strip detection method according to claim 5, characterized in that: When the hybrid model determines whether the smoke box status is abnormal based on the input features: The convolutional neural network part is responsible for processing image data, and the recurrent neural network part is responsible for processing time series data; The extracted comprehensive features are input into the hybrid model, and the model outputs a score indicating the degree of abnormality of the smoke box state. A threshold is set. If the score exceeds the threshold, the smoke box state is judged to be abnormal, otherwise, there is no abnormality.

7. The dynamic weighing missing strip detection method according to claim 6, characterized in that: When an abnormality occurs, the data is transmitted to the storage module in an encrypted manner for storage.

8. The dynamic weighing missing strip detection method according to claim 7, characterized in that: The storage module adopts a distributed storage architecture: When the detection system finds an anomaly, the abnormal data will be encrypted and sent to the metadata server through the client; after receiving the data, the metadata server will generate corresponding metadata and store it in its own database; at the same time, the metadata server will instruct the data server to store the abnormal data on the appropriate storage node.

Images