Methods and systems for the whole-process control of fiber-faced gypsum board production

Abstract

This invention relates to the field of production process control technology, specifically disclosing a method and system for full-process control of fiber-faced gypsum board production. The method includes acquiring a production flow chart, querying the participating entities in each process of the production flow chart, and determining a monitor based on the participating entities; receiving production data within a preset time range obtained by the monitor, identifying the production data, and determining the anomaly degree of each participating entity; statistically analyzing the anomaly degree of each participating entity to obtain an anomaly array, and locating the anomaly entity based on the anomaly array. This invention inserts intelligent devices into the production line based on the production flow chart. The intelligent devices acquire data in real time during the production process, analyze the data, determine the status of each production device, and thus identify the anomaly entity. This process only requires staff to pre-set some parameters; in actual use, no manual intervention is required, resulting in a high degree of automation and efficiency.

Description

Technical Field

[0001] This invention relates to the field of production process control technology, specifically a method and system for controlling the entire production process of fiber-faced gypsum board. Background Technology

[0002] Currently, the production technology of fiber-faced gypsum board has made some progress, but there are still some bottlenecks in the quality control of the entire production process. For example, the automation level of production equipment and the standards for quality inspection need to be further improved. These problems not only affect the stability of product quality but also restrict the improvement of production efficiency. The key to these problems is actually the data collection and analysis process. Most existing data collection and analysis solutions are independent testing solutions for each piece of equipment, with staff managing each piece of production equipment individually. This approach is feasible and facilitates accountability, but it is inefficient and requires a large number of personnel. Excessive personnel involvement also affects the automation level of the entire production line. Therefore, how to provide a quality control solution for the entire production process based on intelligent equipment is the technical problem that this invention aims to solve. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for controlling the entire production process of fiber-faced gypsum board, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A method for full-process control of fiber-faced gypsum board production, the method comprising:

[0006] Obtain the production flow chart, query the participating entities in each process of the production flow chart, and determine the monitor based on the participating entities; the monitor includes at least sensors and cameras;

[0007] Receive production data within a preset time range obtained by the monitor, identify the production data, and determine the degree of abnormality of each participating entity;

[0008] The abnormality degree of each participating entity is counted to obtain an abnormality array, and the abnormal entity is located based on the abnormality array;

[0009] The anomaly degree of the participating entity is determined by both sensor anomaly degree and visual anomaly degree. The sensor anomaly degree is calculated from the data acquired by the sensor, and the visual anomaly degree is calculated from the data acquired by the camera.

[0010] As a further aspect of the present invention: the steps of obtaining the production process flow chart, querying the participating entities of each process in the production process flow chart, and determining the monitor based on the participating entities include:

[0011] Obtain the production flow chart and query the participating entities in each process of the production flow chart;

[0012] The sensors of each participating entity are queried, and the monitoring data of each sensor during the preset test period is obtained at regular intervals. The information content of the monitor is determined based on the monitoring data.

[0013] Sensors are selected based on the amount of information, and the total amount of information for each participating entity is determined based on the selection process.

[0014] Based on the total amount of information, cameras are inserted between adjacent participating entities to establish a correspondence between cameras and participating entities; wherein, the correspondence is as follows: the camera corresponds to the previous participating entity among the adjacent participating entities;

[0015] The calculation process for information content is as follows: I = ασ; where I is the information content, α is a preset coefficient, and σ is the standard deviation of the monitoring data during the test period.

[0016] Based on the information quantity, sensors are selected. The determination of the total information quantity for each participating entity based on the selection process includes:

[0017] The selection probability of each sensor is determined based on the amount of information, and sensors are randomly selected based on the selection probability until at least one sensor is selected in each participating entity.

[0018] The selected sensors for each participating entity are counted, and the information is summed to obtain the total amount of information.

[0019] The calculation process for the selection probability is as follows:

[0020] In the formula, P i Let T be the probability of selecting the i-th sensor. i Let I be the feature value of the i-th sensor, β be a preset correction coefficient, and I be the feature value of the i-th sensor. i Let I be the information content of the i-th sensor, N be the total number of unselected sensors, M be the total number of selected sensors, and I be the information content of the i-th sensor. j Let Δ represent the information content of the j-th selected sensor, and let Δ represent the difference in the subject number between the j-th selected sensor and the i-th sensor. When the two sensors belong to the same subject, the difference in subject number is zero. The sensor number and the subject number are both known data, which are determined by the production flow chart in the preprocessing stage.

[0021] As a further aspect of the present invention: the step of receiving production data within a preset time range obtained by the monitoring device, identifying the production data, and determining the degree of abnormality of each participating entity includes:

[0022] Receive sensor data containing a first identification tag uploaded by all sensors, and statistically analyze the sensor data according to the identification tag to obtain the data sequence corresponding to each sensor; the first identification tag is used to identify which sensor uploaded the sensor data.

[0023] Read the data sequences from each sensor of the same participating subject, perform a Fourier transform on each data sequence, and obtain the frequency domain diagram;

[0024] The sensing anomaly of the participating subject is determined based on the frequency domain diagram;

[0025] The system receives images with second identity tags uploaded by all cameras, queries the participating entity corresponding to the camera based on the location tag, reads the sensing anomaly degree of the participating entity, and determines the recognition accuracy of the image; the recognition accuracy is used to adjust the amount of recognition resources; the second identity tag is used to characterize which camera uploaded the image.

[0026] Based on the recognition accuracy, the image is identified to determine the degree of visual anomaly of the participating subject;

[0027] The anomaly degree of the participating subject is determined based on the sensor anomaly degree and the visual anomaly degree.

[0028] As a further aspect of the present invention: the step of determining the sensing anomaly degree of the participating subject based on the frequency domain diagram includes:

[0029] By comparing the frequency domain maps of different sensors of the same participating entity, the characteristic frequency domain map is extracted, and the fidelity of the characteristic frequency domain map is determined simultaneously.

[0030] Input the feature frequency domain map into the trained frequency domain recognition model, output the stability, and correct the stability based on the fidelity.

[0031] The sensing anomaly is calculated based on the corrected stability, and the sensing anomaly is inversely proportional to the corrected stability.

[0032] The step of identifying the image based on recognition accuracy and determining the visual anomalousness of the participating subjects includes:

[0033] Convolution kernels are sequentially selected from a preset kernel table based on recognition accuracy, and the image is matched and recognized based on the selected kernels; the kernel table includes kernel entries and outlier entries; each data entry in the kernel table is arranged in descending order based on outliers.

[0034] Query the outliers corresponding to the successfully matched convolutional kernels, and accumulate all matched outliers;

[0035] The degree of visual aberration of the participating subjects is determined based on the cumulative outlier values;

[0036] The steps of comparing the frequency domain maps of different sensors of the same participating subject, extracting the feature frequency domain map, and simultaneously determining the fidelity of the feature frequency domain map include:

[0037] Query all sensors of the same participating entity. For any sensor, compare its frequency domain map with the frequency domain maps of other sensors of the same participating entity to obtain the similarity. Calculate the mean of the similarity to obtain the average similarity. Select the frequency domain map of the sensor with the highest average similarity as the feature frequency domain map. Use the average similarity as the fidelity.

[0038] As a further aspect of the present invention: the step of receiving production data within a preset time range obtained by the monitoring device, identifying the production data, and determining the degree of abnormality of each participating entity further includes:

[0039] Adjust the image acquisition frequency of the corresponding camera based on the degree of anomaly of the participating entity.

[0040] As a further aspect of the present invention: the step of statistically analyzing the anomaly degree of each participating entity to obtain an anomaly array, and locating the anomaly entity based on the anomaly array includes:

[0041] Query the sequence number of the participating entity in the production process;

[0042] Read the anomaly degree of the participating entities, using the sequence number as the independent variable and the anomaly degree as the dependent variable, to determine the anomaly curve and curve function of the production model;

[0043] Calculate the derivative of the curve function, truncate the derivative according to a preset threshold line, and locate the abnormal subject based on the truncation result.

[0044] The present invention also provides a whole-process control system for the production of fiber-faced gypsum board, the system comprising:

[0045] The monitor determination module is used to acquire the production process flow chart, query the participating entities of each process in the production process flow chart, and determine the monitor based on the participating entities; the monitor includes at least sensors and cameras;

[0046] The anomaly calculation module is used to receive production data within a preset time range obtained by the monitor, identify the production data, and determine the anomaly degree of each participating entity.

[0047] An abnormal subject localization module is used to count the abnormality of each participating subject, obtain an abnormality array, and locate the abnormal subject based on the abnormality array;

[0048] The anomaly degree of the participating entity is determined by both sensor anomaly degree and visual anomaly degree. The sensor anomaly degree is calculated from the data acquired by the sensor, and the visual anomaly degree is calculated from the data acquired by the camera.

[0049] As a further aspect of the present invention: the monitor determination module includes:

[0050] The participant query unit is used to obtain the production flow chart and query the participants in each process of the production flow chart;

[0051] The information calculation unit is used to query the sensors of each participating entity, periodically acquire the monitoring data of each sensor within a preset test period, and determine the information content of the monitor based on the monitoring data.

[0052] A sensor selection unit is used to select sensors based on the amount of information and to determine the total amount of information for each participating entity based on the selection process.

[0053] A camera selection unit is used to insert cameras between adjacent participating entities based on the total amount of information, and to establish a correspondence between cameras and participating entities; wherein, the correspondence is that the camera corresponds to the previous participating entity among the adjacent participating entities;

[0054] The calculation process for information content is as follows: I = ασ; where I is the information content, α is a preset coefficient, and σ is the standard deviation of the monitoring data during the test period.

[0055] Based on the information quantity, sensors are selected. The determination of the total information quantity for each participating entity based on the selection process includes:

[0056] The selection probability of each sensor is determined based on the amount of information, and sensors are randomly selected based on the selection probability until at least one sensor is selected in each participating entity.

[0057] The selected sensors for each participating entity are counted, and the information is summed to obtain the total amount of information.

[0058] The calculation process for the selection probability is as follows:

[0059] In the formula, P i Let T be the probability of selecting the i-th sensor. i Let I be the feature value of the i-th sensor, β be a preset correction coefficient, and I be the feature value of the i-th sensor. i Let I be the information content of the i-th sensor, N be the total number of unselected sensors, M be the total number of selected sensors, and I be the information content of the i-th sensor. j Let Δ represent the information content of the j-th selected sensor, and let Δ represent the difference in the subject number between the j-th selected sensor and the i-th sensor. When the two sensors belong to the same subject, the difference in subject number is zero. The sensor number and the subject number are both known data, which are determined by the production flow chart in the preprocessing stage.

[0060] As a further aspect of the present invention: the anomaly calculation module includes:

[0061] The data sequence query unit is used to receive sensor data containing a first identity tag uploaded by all sensors, and to statistically analyze the sensor data according to the identity tag to obtain the data sequence corresponding to each sensor; the first identity tag is used to characterize which sensor uploaded the sensor data.

[0062] The frequency domain graph generation unit is used to read the data sequences of each sensor of the same participating subject, perform Fourier transform on each data sequence, and obtain the frequency domain graph.

[0063] The first calculation unit is used to determine the sensing anomaly degree of the participating subject based on the frequency domain diagram;

[0064] The recognition accuracy determination unit is used to receive images containing second identity tags uploaded by all cameras, query the participating entity corresponding to the camera according to the location tag, read the sensing anomaly degree of the participating entity, and determine the recognition accuracy of the image; the recognition accuracy is used to adjust the amount of recognition resources; the second identity tag is used to characterize which camera uploaded the image;

[0065] The second computing unit is used to identify the image based on the recognition accuracy and determine the visual abnormality of the participating subject.

[0066] The integrated calculation unit is used to determine the anomaly degree of the participating subject based on the sensing anomaly degree and the visual anomaly degree.

[0067] As a further aspect of the present invention: the content of determining the sensing anomaly degree of the participating subject based on the frequency domain diagram includes:

[0068] By comparing the frequency domain maps of different sensors of the same participating entity, the characteristic frequency domain map is extracted, and the fidelity of the characteristic frequency domain map is determined simultaneously.

[0069] Input the feature frequency domain map into the trained frequency domain recognition model, output the stability, and correct the stability based on the fidelity.

[0070] The sensing anomaly is calculated based on the corrected stability, and the sensing anomaly is inversely proportional to the corrected stability.

[0071] The process of identifying images based on recognition accuracy to determine the visual anomalousness of the participating subjects includes:

[0072] Convolution kernels are sequentially selected from a preset kernel table based on recognition accuracy, and the image is matched and recognized based on the selected kernels; the kernel table includes kernel entries and outlier entries; each data entry in the kernel table is arranged in descending order based on outliers.

[0073] Query the outliers corresponding to the successfully matched convolutional kernels, and accumulate all matched outliers;

[0074] The degree of visual aberration of the participating subjects is determined based on the cumulative outlier values;

[0075] The steps of comparing the frequency domain maps of different sensors of the same participating subject, extracting the feature frequency domain map, and simultaneously determining the fidelity of the feature frequency domain map include:

[0076] Query all sensors of the same participating entity. For any sensor, compare its frequency domain map with the frequency domain maps of other sensors of the same participating entity to obtain the similarity. Calculate the mean of the similarity to obtain the average similarity. Select the frequency domain map of the sensor with the highest average similarity as the feature frequency domain map. Use the average similarity as the fidelity.

[0077] Compared with the prior art, the beneficial effects of the present invention are:

[0078] This invention inserts intelligent devices into the production line based on the production flow chart. The intelligent devices acquire data in real time during the production process, analyze the data, determine the status of each production device, and then identify the abnormal entity. This process only requires staff to pre-set some parameters. In actual use, no manual intervention is required, resulting in a high degree of automation and high efficiency.

[0079] This invention establishes a real-time monitoring and data analysis system that monitors and collects data on each stage of the production process in real time. Through data analysis, abnormal situations in the production process can be detected in a timely manner and then uploaded to the central control terminal, where corresponding corrective measures are taken to ensure the stability of the production process and the consistency of product quality. Attached Figure Description

[0080] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0081] Figure 1 A flowchart illustrating the process control method for the entire production process of fiber-faced gypsum board.

[0082] Figure 2 This is the first sub-process flowchart of the whole-process control method for fiber-faced gypsum board production.

[0083] Figure 3 This is the second sub-process flowchart of the whole-process control method for fiber-faced gypsum board production.

[0084] Figure 4 This is the third sub-process flowchart of the whole-process control method for fiber-faced gypsum board production.

[0085] Figure 5 This is a structural diagram of the entire process control system for fiber-faced gypsum board production. Detailed Implementation

[0086] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0087] Figure 1 This invention provides a flowchart illustrating a method for controlling the entire production process of fiber-faced gypsum board. The method, as described in this embodiment, includes:

[0088] Step S100: Obtain the production process flow chart, query the participating entities of each process in the production process flow chart, and determine the monitor based on the participating entities; the monitor includes at least sensors and cameras;

[0089] Regardless of the product being manufactured, the manufacturer will pre-create a production flow chart containing various processes. The manufacturer then queries the participating entities for each process, which may not be unique, but are mostly production equipment. Based on these participating entities, monitoring devices are selected or installed to acquire data during the production process. These monitoring devices include sensors and cameras. Specifically, sensors are typically located within the production equipment, while cameras are usually installed at the equipment's exit point to capture the equipment's performance. Sensors can indirectly indicate the quality of the equipment's operation, while the performance results directly determine its quality. In other words, the operational status of the production equipment is determined by the results obtained from that equipment.

[0090] Step S200: Receive production data within a preset time range obtained by the monitor, identify the production data, and determine the degree of abnormality of each participating entity;

[0091] The data uploaded by the monitoring device is called production data. Identifying this production data allows us to determine the anomaly level of each participating entity. This identification includes recognizing data acquired by sensors to obtain sensor anomaly levels, and recognizing images acquired by cameras to obtain visual anomaly levels. Combining these sensor and visual anomaly levels yields the overall anomaly level of each participating entity. In short, the anomaly level of a participating entity is jointly determined by sensor and visual anomaly levels. Sensor anomaly levels are calculated from data acquired by sensors, and visual anomaly levels are calculated from data acquired by cameras. Analyzing data acquired by sensors is a numerical analysis process, which is one-dimensional, while analyzing data acquired by cameras is an image analysis process, which is two-dimensional.

[0092] Step S300: Calculate the abnormality of each participating entity to obtain an abnormality array, and locate the abnormal entity based on the abnormality array;

[0093] After the anomaly degree of each participating entity is calculated, an anomaly array is obtained. The element number of the anomaly array corresponds to the participating entity, and the element value corresponds to the anomaly degree of the participating entity. By analyzing the anomaly array, the participating entities with problems can be identified, which are called abnormal entities. After the abnormal entities are identified, warning information is reported to achieve the purpose of full-process control.

[0094] It should be noted that the execution process of this method is a timed execution process. Every preset period, such as five minutes, an abnormal subject is identified. Specifically, every preset period, a retrieval command is generated to retrieve production data. The method of retrieving production data is as follows: based on the time when the retrieval command is generated, data within a certain period of time is retrieved. This period is also preset, such as one hour.

[0095] Figure 2 The first sub-process flowchart of the method for full-process control of fiber-faced gypsum board production includes the following steps: obtaining the production flow chart, querying the participating entities in each process of the production flow chart, and determining the monitor based on the participating entities.

[0096] Step S101: Obtain the production flow chart and query the participating entities in each process of the production flow chart;

[0097] Step S102: Query the sensors of each participating entity, periodically acquire the monitoring data of each sensor within a preset test period, and determine the amount of information of the monitor based on the monitoring data;

[0098] Step S103: Select a sensor based on the amount of information, and determine the total amount of information for each participating entity based on the selection process;

[0099] Step S104: Based on the total amount of information, insert cameras between adjacent participating entities and establish a correspondence between cameras and participating entities; wherein, the correspondence is: the camera corresponds to the previous participating entity among the adjacent participating entities.

[0100] In one example of the technical solution of this invention, the process of determining the monitor is described. A production flow chart is obtained, and the participating entities in each process of the production flow chart are queried. In a fully automated production workshop, the participating entities are the production equipment. The sensors of each participating entity are queried, and tests are conducted to determine a test time span. Within the test time span, monitoring data from the sensors is acquired periodically. The amount of information is calculated based on the acquired monitoring data. Sensors are selected based on the amount of information. After selecting the sensors, the amount of information from the selected sensors in each participating entity is queried. The information amounts are summed to obtain the total amount of information. Cameras are inserted between adjacent participating entities based on the total amount of information, and a correspondence between the cameras and the participating entities is established. Each camera corresponds to the previous participating entity among the adjacent participating entities.

[0101] Furthermore, some simple explanations are needed regarding the above content. Under normal circumstances, an intelligent fiber-faced gypsum board production workshop has multiple production equipment. These production equipment are arranged in sequence according to the production process. After the raw materials enter the production equipment, they are processed and then exit the production equipment to enter the next production equipment. There is a transportation process between the production equipment. The transportation distance may be long or short. Cameras are installed during transportation to capture the processing status of the products. The processing status of the products is analyzed to obtain the operating status of the previous production equipment. Based on this, the camera actually corresponds to the semi-finished product, and it corresponds to the source of the semi-finished product (the production equipment).

[0102] The calculation process for information content is as follows: I = ασ; where I is the information content, α is a preset coefficient, and σ is the standard deviation of the monitoring data during the test period.

[0103] Based on the information quantity, sensors are selected. The determination of the total information quantity for each participating entity based on the selection process includes:

[0104] The selection probability of each sensor is determined based on the amount of information, and sensors are randomly selected based on the selection probability until at least one sensor is selected in each participating entity.

[0105] The selected sensors for each participating entity are counted, and the information is summed to obtain the total amount of information.

[0106] The calculation process for the selection probability is as follows:

[0107] In the formula, P iLet T be the probability of selecting the i-th sensor. i Let I be the feature value of the i-th sensor, β be a preset correction coefficient, and I be the feature value of the i-th sensor. i Let I be the information content of the i-th sensor, N be the total number of unselected sensors, M be the total number of selected sensors, and I be the information content of the i-th sensor. j Let Δ represent the information content of the j-th selected sensor, and let Δ represent the difference in the subject number between the j-th selected sensor and the i-th sensor. When the two sensors belong to the same subject, the difference in subject number is zero. The sensor number and the subject number are both known data, which are determined by the production flow chart in the preprocessing stage.

[0108] The above describes the sensor selection process. First, during the testing phase, the information content of the sensor is determined. The greater the fluctuation of the sensor's data, the greater its information content is considered, and therefore, it is directly proportional to the standard deviation. Then, for each sensor, a characteristic value is calculated. The characteristic value is directly proportional to its own information content and inversely proportional to the information content of other sensors. The closer other sensors are to the current sensor, the smaller their influence. Finally, the calculated characteristic values ​​are combined to calculate the selection probability of each sensor.

[0109] It should be noted that α and β are both positive values ​​to ensure the positive and negative relationship between the parameters. In addition, the N value needs to be updated once for each selected sensor, and the selection probability of the sensor needs to be recalculated. This process is repeated until at least one sensor in each participating subject is selected.

[0110] Figure 3 The second sub-process flowchart of the method for full-process control of fiber-faced gypsum board production includes the following steps: receiving production data within a preset time range from the monitor, identifying the production data, and determining the degree of abnormality of each participating entity.

[0111] Step S201: Receive sensor data containing the first identification tag uploaded by all sensors, and statistically analyze the sensor data according to the identification tag to obtain the data sequence corresponding to each sensor; the first identification tag is used to characterize which sensor uploaded the sensor data.

[0112] Step S202: Read the data sequences of each sensor of the same participating subject, perform Fourier transform on each data sequence, and obtain the frequency domain diagram;

[0113] Step S203: Determine the sensing anomaly degree of the participating subject based on the frequency domain diagram;

[0114] Step S204: Receive images containing the second identity tag uploaded by all cameras, query the participating entity corresponding to the camera according to the location tag, read the sensing anomaly degree of the participating entity, and determine the recognition accuracy of the image; the recognition accuracy is used to adjust the amount of recognition resources; the second identity tag is used to characterize which camera uploaded the image;

[0115] Step S205: Based on the recognition accuracy, identify the image and determine the visual anomalousness of the participating subject;

[0116] Step S206: Determine the anomaly degree of the participating subject based on the sensing anomaly degree and the visual anomaly degree.

[0117] In one embodiment of the technical solution of this invention, for each sensor, sensing data needs to be acquired in real time according to a preset frequency. The uploading process of the sensors is not ordered. The execution subject of this method may receive uploaded data from different sensors within the same second. In addition, due to network fluctuations, the sensing data received by the execution subject of this method is very messy. Therefore, when the sensor uploads sensing data, a first identification tag needs to be inserted to indicate which sensor uploaded the sensing data. In addition, the sensing data also needs to have a time tag. After receiving the data, the execution subject of this method arranges the sensing data according to the time tag to obtain a data sequence. Thus, each sensor corresponds to a data sequence.

[0118] Similar to the above, the images uploaded by the cameras also need to identify which camera uploaded the image. Therefore, when the camera uploads an image, a second identity tag is inserted. Similarly, the image itself also contains time information. The execution subject of this method sorts the images from each camera to obtain an image sequence.

[0119] Furthermore, the data sequence of the sensor is essentially a discrete array in the time domain. By performing a frequency domain transformation on it, frequency domain features are obtained. Based on these frequency domain features, the degree of anomaly in the sensor data can be determined, and the sensing anomaly degree of the participating subject can be calculated. On this basis, this application determines the recognition accuracy of the image based on the sensing anomaly degree. The greater the sensing anomaly degree, the higher the recognition accuracy needs to be. Based on the recognition accuracy, the image is recognized to determine the visual anomaly degree of the participating subject. By combining the sensing anomaly degree and the visual anomaly degree, the anomaly degree of the participating subject is obtained.

[0120] Specifically, the sensor anomaly degree is a feature of the participating subject obtained from the data sequence of multiple sensors. It affects the image recognition process of the corresponding camera. That is, the process of determining the recognition accuracy of the image based on the sensor anomaly degree is to determine the recognition accuracy of the corresponding camera based on the sensor anomaly degree of a certain participating subject. After recognizing the image based on the recognition accuracy, the visual anomaly degree is obtained. The visual anomaly degree is the anomaly of the product coming out of the corresponding production equipment, and it also reflects the anomaly of the production equipment. Therefore, both the sensor anomaly degree and the visual anomaly degree are parameters that reflect the anomaly of the same participating subject.

[0121] In one embodiment of the technical solution of the present invention, the step of determining the sensing anomaly degree of the participating subject based on the frequency domain diagram includes:

[0122] By comparing the frequency domain maps of different sensors of the same participating entity, the characteristic frequency domain map is extracted, and the fidelity of the characteristic frequency domain map is determined simultaneously.

[0123] Input the feature frequency domain map into the trained frequency domain recognition model, output the stability, and correct the stability based on the fidelity.

[0124] The sensing anomaly is calculated based on the corrected stability, and the sensing anomaly is inversely proportional to the corrected stability.

[0125] Sensors are internal components of a participating entity. For the same participating entity, there may be multiple sensors. The data sequence of each sensor is converted into a frequency domain image. By comparing the frequency domain images of different sensors of the same participating entity, a standard frequency domain image, called the feature frequency domain image, is extracted. The fidelity of the feature frequency domain image is determined simultaneously. The feature frequency domain image is then input into a trained frequency domain recognition model, which outputs stability. The recognition process of the frequency domain recognition model is more refined. One type of model is the comparison model, which first pre-compiles some standard frequency domain images and compares and analyzes them with the feature frequency domain image. The more similar the comparison results are, the more standardized the working process is and the higher the stability is. The stability is adjusted based on the determined fidelity, which is usually achieved by direct multiplication. Finally, the sensing anomaly is calculated based on the corrected stability, a process that can be achieved using a subtraction function.

[0126] In one embodiment of the technical solution of the present invention, the step of identifying the image based on the recognition accuracy and determining the visual anomalousness of the participating subject includes:

[0127] Convolution kernels are sequentially selected from a preset kernel table based on recognition accuracy, and the image is matched and recognized based on the selected kernels; the kernel table includes kernel entries and outlier entries; each data entry in the kernel table is arranged in descending order based on outliers.

[0128] Query the outliers corresponding to the successfully matched convolutional kernels, and accumulate all matched outliers;

[0129] The degree of visual aberration of the participating subjects is determined based on the cumulative outlier values.

[0130] The above describes the visual recognition process. A convolution kernel table is pre-created by staff. The convolution kernel table includes convolution kernel items and outlier items. The data items in the convolution kernel table are arranged in descending order based on the outliers. The number of data items in the convolution kernel table is large, and staff need to count historical data, extract and count the convolution kernels. This occurs in the preprocessing stage and is known data for this application.

[0131] Furthermore, convolutional kernels are sequentially selected from a preset convolutional kernel table based on the recognition accuracy. The higher the recognition accuracy, the more convolutional kernels are selected. The image is matched and recognized based on the selected convolutional kernels. The matching and recognition process belongs to the convolutional recognition process, which will not be described in detail in this application. Then, the outlier corresponding to the successfully matched convolutional kernel is queried, and all matched outlier values ​​are accumulated. The visual anomalousness of the participating subject is determined based on the accumulated outlier values. The larger the accumulated outlier values, the higher the visual anomalousness.

[0132] The steps of comparing the frequency domain maps of different sensors of the same participating subject, extracting the feature frequency domain map, and simultaneously determining the fidelity of the feature frequency domain map include:

[0133] Query all sensors of the same participating entity. For any sensor, compare its frequency domain map with the frequency domain maps of other sensors of the same participating entity to obtain the similarity. Calculate the mean of the similarity to obtain the average similarity. Select the frequency domain map of the sensor with the highest average similarity as the feature frequency domain map. Use the average similarity as the fidelity.

[0134] It is worth mentioning that since the sensor data is discrete data, the resulting frequency domain map is also discrete data. The comparison process between frequency domain maps is actually an array comparison process, which can be achieved by using existing array similarity calculation schemes.

[0135] As a preferred embodiment of the technical solution of the present invention, the step of receiving production data within a preset time range obtained by the monitoring device, identifying the production data, and determining the degree of abnormality of each participating entity further includes:

[0136] Adjust the image acquisition frequency of the corresponding camera based on the degree of anomaly of the participating entity.

[0137] In one example of the technical solution of the present invention, after calculating the anomaly degree of the participating subject, the image acquisition frequency of the corresponding camera is adjusted according to the anomaly degree of the participating subject. The higher the anomaly degree, the higher the image acquisition frequency, and the longer the image sequence obtained in the above content.

[0138] It is worth mentioning that the participating entities and cameras in this application are not necessarily in one-to-one correspondence. One camera definitely corresponds to one participating entity, but one participating entity may correspond to multiple cameras. For example, when two cameras are installed between adjacent participating entities, both cameras correspond to the same participating entity.

[0139] Figure 4 The third sub-process flowchart of the whole-process control method for fiber-faced gypsum board production includes the following steps: Statistically calculating the anomaly degree of each participating entity to obtain an anomaly array, and locating the anomaly entity based on the anomaly array.

[0140] Step S301: Query the serial number of the participating entity in the production process;

[0141] Step S302: Read the anomaly degree of the participating entities, using the sequence number as the independent variable and the anomaly degree as the dependent variable, to determine the anomaly curve and curve function of the production model;

[0142] Step S303: Calculate the derivative of the curve function, truncate the derivative function according to a preset threshold line, and locate the abnormal subject based on the truncation result.

[0143] In one example of the technical solution of this invention, the process of locating the abnormal subject is described. The sequence number of the participating subject in the production process is queried. The sequence number is determined by the order in which the product passes through each production device. The abnormality degree of the participating subject is read, with the sequence number as the independent variable and the abnormality degree as the dependent variable. The abnormality curve and curve function of the production model are determined. By differentiating the curve function, it can be determined whether there is a sudden change in the abnormality degree. If the change reaches a preset threshold, the participating subject corresponding to the sequence number at the point of change can be considered an abnormal subject. The threshold is treated as a constant function relative to the sequence number to obtain the threshold line. The derivative function is extracted, and the sequence number corresponding to the function line above the derivative function is the sequence number of the abnormal subject.

[0144] It should be noted that the implied meaning of the anomaly assessment process based on changes is that small changes or gradual changes can be ignored. Neither of these will be identified as an anomaly in this application. In reality, this is also the case. If the anomaly of each device increases slowly, it is very likely that the misjudgment is caused by the environment or the duration of operation. For example, as the duration of operation increases, the temperature of the device gradually increases.

[0145] Figure 5 This is a structural block diagram of a fiber-faced gypsum board production process control system. In this embodiment of the invention, a fiber-faced gypsum board production process control system 10 includes:

[0146] The monitor determination module 11 is used to acquire the production process diagram, query the participating entities of each process in the production process diagram, and determine the monitor based on the participating entities; the monitor includes at least sensors and cameras;

[0147] The anomaly calculation module 12 is used to receive production data within a preset time range obtained by the monitor, identify the production data, and determine the anomaly degree of each participating entity.

[0148] The abnormal subject localization module 13 is used to count the abnormality of each participating subject, obtain an abnormality array, and locate the abnormal subject based on the abnormality array;

[0149] The anomaly degree of the participating entity is determined by both sensor anomaly degree and visual anomaly degree. The sensor anomaly degree is calculated from the data acquired by the sensor, and the visual anomaly degree is calculated from the data acquired by the camera.

[0150] Furthermore, the monitor determination module 11 includes:

[0151] The participant query unit is used to obtain the production flow chart and query the participants in each process of the production flow chart;

[0152] The information calculation unit is used to query the sensors of each participating entity, periodically acquire the monitoring data of each sensor within a preset test period, and determine the information content of the monitor based on the monitoring data.

[0153] A sensor selection unit is used to select sensors based on the amount of information and to determine the total amount of information for each participating entity based on the selection process.

[0154] A camera selection unit is used to insert cameras between adjacent participating entities based on the total amount of information, and to establish a correspondence between cameras and participating entities; wherein, the correspondence is that the camera corresponds to the previous participating entity among the adjacent participating entities;

[0155] The calculation process for information content is as follows: I = ασ; where I is the information content, α is a preset coefficient, and σ is the standard deviation of the monitoring data during the test period.

[0156] Based on the information quantity, sensors are selected. The determination of the total information quantity for each participating entity based on the selection process includes:

[0157] The selection probability of each sensor is determined based on the amount of information, and sensors are randomly selected based on the selection probability until at least one sensor is selected in each participating entity.

[0158] The selected sensors for each participating entity are counted, and the information is summed to obtain the total amount of information.

[0159] The calculation process for the selection probability is as follows:

[0160] In the formula, P i Let T be the probability of selecting the i-th sensor. i Let I be the feature value of the i-th sensor, β be a preset correction coefficient, and I be the feature value of the i-th sensor. i Let I be the information content of the i-th sensor, N be the total number of unselected sensors, M be the total number of selected sensors, and I be the information content of the i-th sensor. j Let Δ represent the information content of the j-th selected sensor, and let Δ represent the difference in the subject number between the j-th selected sensor and the i-th sensor. When the two sensors belong to the same subject, the difference in subject number is zero. The sensor number and the subject number are both known data, which are determined by the production flow chart in the preprocessing stage.

[0161] Specifically, the anomaly calculation module 12 includes:

[0162] The data sequence query unit is used to receive sensor data containing a first identity tag uploaded by all sensors, and to statistically analyze the sensor data according to the identity tag to obtain the data sequence corresponding to each sensor; the first identity tag is used to characterize which sensor uploaded the sensor data.

[0163] The frequency domain graph generation unit is used to read the data sequences of each sensor of the same participating subject, perform Fourier transform on each data sequence, and obtain the frequency domain graph.

[0164] The first calculation unit is used to determine the sensing anomaly degree of the participating subject based on the frequency domain diagram;

[0165] The recognition accuracy determination unit is used to receive images containing second identity tags uploaded by all cameras, query the participating entity corresponding to the camera according to the location tag, read the sensing anomaly degree of the participating entity, and determine the recognition accuracy of the image; the recognition accuracy is used to adjust the amount of recognition resources; the second identity tag is used to characterize which camera uploaded the image;

[0166] The second computing unit is used to identify the image based on the recognition accuracy and determine the visual abnormality of the participating subject.

[0167] The integrated calculation unit is used to determine the anomaly degree of the participating subject based on the sensing anomaly degree and the visual anomaly degree.

[0168] Furthermore, the determination of the sensing anomaly degree of the participating subject based on the frequency domain diagram includes:

[0169] By comparing the frequency domain maps of different sensors of the same participating entity, the characteristic frequency domain map is extracted, and the fidelity of the characteristic frequency domain map is determined simultaneously.

[0170] Input the feature frequency domain map into the trained frequency domain recognition model, output the stability, and correct the stability based on the fidelity.

[0171] The sensing anomaly is calculated based on the corrected stability, and the sensing anomaly is inversely proportional to the corrected stability.

[0172] The process of identifying images based on recognition accuracy to determine the visual anomalousness of the participating subjects includes:

[0173] Convolution kernels are sequentially selected from a preset kernel table based on recognition accuracy, and the image is matched and recognized based on the selected kernels; the kernel table includes kernel entries and outlier entries; each data entry in the kernel table is arranged in descending order based on outliers.

[0174] Query the outliers corresponding to the successfully matched convolutional kernels, and accumulate all matched outliers;

[0175] The degree of visual aberration of the participating subjects is determined based on the cumulative outlier values;

[0176] The steps of comparing the frequency domain maps of different sensors of the same participating subject, extracting the feature frequency domain map, and simultaneously determining the fidelity of the feature frequency domain map include:

[0177] Query all sensors of the same participating entity. For any sensor, compare its frequency domain map with the frequency domain maps of other sensors of the same participating entity to obtain the similarity. Calculate the mean of the similarity to obtain the average similarity. Select the frequency domain map of the sensor with the highest average similarity as the feature frequency domain map. Use the average similarity as the fidelity.

[0178] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for managing the entire process of manufacturing a fibered gypsum board, characterized in that, The method comprises: obtaining a production flowchart, querying participating subjects of each process in the production flowchart, and determining a monitor according to the participating subjects; the monitor at least comprises a sensor and a camera; receiving production data in a preset time range obtained by the monitor, identifying the production data, and determining an abnormality degree of each participating subject; counting the abnormality degrees of the participating subjects to obtain an abnormality array, and positioning an abnormal subject according to the abnormality array; wherein the abnormality degree of the participating subject is determined by a sensing abnormality degree and a visual abnormality degree; the sensing abnormality degree is calculated from data obtained by the sensor, and the visual abnormality degree is calculated from data obtained by the camera; the step of obtaining the production flowchart, querying the participating subjects of each process in the production flowchart, and determining the monitor according to the participating subjects comprises: obtaining a production flowchart, querying participating subjects of each process in the production flowchart; querying sensors of each participating subject, obtaining monitoring data of each sensor in a preset test period in a timely manner, and determining an information amount of the monitor according to the monitoring data; selecting the sensor according to the information amount, and determining an information total amount of each participating subject according to the selection process; inserting a camera between adjacent participating subjects according to the information total amount, and establishing a corresponding relationship between the camera and the participating subjects; wherein the corresponding relationship is that the camera corresponds to a previous participating subject among adjacent participating subjects; The calculation process of the information quantity is as follows: ; wherein, is the information quantity, is a preset coefficient, is a standard deviation of the monitoring data in the test period; the content of selecting the sensor according to the information amount and determining the information total amount of each participating subject according to the selection process comprises: determining a selection probability of each sensor according to the information amount, randomly selecting the sensor according to the selection probability, and selecting at least one sensor in each participating subject until the selection is completed; counting the selected sensors of each participating subject, summing the information amount, and obtaining the information total amount; wherein the calculation process of the selection probability is: ; In the formula, For the first The probability of selecting each sensor, For the first The characteristic values ​​of each sensor The preset correction factor. For the first The amount of information from each sensor This represents the total number of sensors that were not selected. This represents the total number of sensors selected. For the first The amount of information from each selected sensor. Indicates the first The selected sensor and the first The difference in the main body number of each sensor is zero when two sensors belong to the same participating entity. The sensor number and the main body number are both known data, which are determined by the production flow chart during the preprocessing stage.

2. The method according to claim 1, wherein the step of receiving the production data in the preset time range obtained by the monitor, identifying the production data, and determining the abnormality degree of each participating subject comprises: receiving sensor data containing a first identity tag uploaded by all sensors, counting the sensor data according to the identity tag, and obtaining a data sequence corresponding to each sensor; the first identity tag is used to represent which sensor uploads the sensor data; reading the data sequence of each sensor of the same participating subject, performing Fourier transform on each data sequence, and obtaining a frequency domain graph; determining the sensing abnormality degree of the participating subject according to the frequency domain graph; receiving images containing a second identity tag uploaded by all cameras, querying the participating subject corresponding to the camera according to the position tag, reading the sensing abnormality degree of the participating subject, and determining the recognition accuracy of the image; the recognition accuracy is used to adjust the recognition resource amount; the second identity tag is used to represent which camera uploads the image; identifying the image based on the recognition accuracy, and determining the visual abnormality degree of the participating subject; determining the abnormality degree of the participating subject according to the sensing abnormality degree and the visual abnormality degree.

3. The method according to claim 2, wherein the method is characterized by, the step of determining the sensing abnormality degree of the participating subject according to the frequency domain graph comprises: The frequency domain graphs of different sensors of the same participant are compared, a characteristic frequency domain graph is extracted, and the fidelity of the characteristic frequency domain graph is determined synchronously; The characteristic frequency domain graph is input into a trained frequency domain recognition model, and a stability is output, and the stability is corrected according to the fidelity; The sensing abnormality degree is calculated according to the corrected stability, and the sensing abnormality degree is inversely proportional to the corrected stability; The step of identifying the image based on the recognition accuracy to determine the visual abnormality degree of the participant includes: According to the recognition accuracy, a convolution kernel is sequentially selected from a preset convolution kernel table, and the image is matched and identified according to the selected convolution kernel; the convolution kernel table includes convolution kernel items and abnormal value items; each data item in the convolution kernel table is arranged in descending order based on the abnormal value; The abnormal value corresponding to the convolution kernel that successfully matches and identifies is queried, and all the abnormal values obtained by matching are accumulated; The visual abnormality degree of the participant is determined according to the accumulated abnormal value; The step of comparing the frequency domain graphs of different sensors of the same participant, extracting a characteristic frequency domain graph, and synchronously determining the fidelity of the characteristic frequency domain graph includes: For any sensor, the frequency domain graph of the sensor is compared with the frequency domain graphs of other sensors of the same participant to obtain a similarity; the average similarity is calculated to obtain an average similarity, and the frequency domain graph of the sensor with the maximum average similarity is selected as the characteristic frequency domain graph, and the average similarity is taken as the fidelity.

4. The method according to claim 2, wherein the method is characterized by, The step of receiving the production data in a preset time range obtained by the monitor, identifying the production data, and determining the abnormality degree of each participant further includes: Adjusting the image acquisition frequency of the corresponding camera according to the abnormality degree of the participant.

5. The method of claim 1, wherein the method further comprises: determining a number of the fibered gypsum board production processes; and determining a number of the fibered gypsum board production processes that are in progress. The step of positioning the abnormal participant according to the abnormality array includes: Querying the serial number of the participant in the production process; Reading the abnormality degree of the participant, taking the serial number as the independent variable, and the abnormality degree as the dependent variable, determining the abnormal curve and curve function of the production model; Calculating the derivative function of the curve function, intercepting the derivative function according to a preset threshold line, and positioning the abnormal participant according to the interception result.

6. A fibered gypsum board production whole-process management system, characterized in that, The system includes: A monitor determination module for obtaining a production process diagram, querying the participants of each process in the production process diagram, and determining a monitor according to the participants; the monitor includes at least a sensor and a camera; An abnormality degree calculation module for receiving production data in a preset time range obtained by the monitor, identifying the production data, and determining the abnormality degree of each participant; An abnormal participant positioning module for calculating the abnormality degree of each participant, obtaining an abnormality array, and positioning the abnormal participant according to the abnormality array; The abnormality degree of the participant is determined by the sensing abnormality degree and the visual abnormality degree, the sensing abnormality degree is calculated from the data obtained by the sensor, and the visual abnormality degree is calculated from the data obtained by the camera; The monitor determination module includes: A participant query unit for obtaining a production process diagram and querying the participants of each process in the production process diagram. An information amount calculation unit is configured to query sensors of each participating subject, acquire monitoring data of each sensor in a preset test period in time, and determine an information amount of the monitor according to the monitoring data; A sensor selection unit is configured to select sensors according to the information amount, and determine a total information amount of each participating subject according to the selection process; A camera selection unit is configured to insert a camera between adjacent participating subjects according to the total information amount, and establish a corresponding relationship between the camera and the participating subjects; wherein the corresponding relationship is that the camera corresponds to a previous participating subject among the adjacent participating subjects; The information amount calculation process is: ; wherein, is the information amount, is a preset coefficient, is the standard deviation of the monitoring data in the test period; The content of selecting sensors according to the information amount and determining a total information amount of each participating subject according to the selection process includes: determining a selection probability of each sensor according to the information amount, and randomly selecting sensors according to the selection probability until at least one sensor in each participating subject is selected; counting the selected sensors of each participating subject, summing the information amount, and obtaining the total information amount; wherein the calculation process of the selection probability is: ; In the formula, For the first The probability of selecting each sensor, For the first The characteristic values ​​of each sensor The preset correction factor. For the first The amount of information from each sensor This represents the total number of sensors that were not selected. This represents the total number of sensors selected. For the first The amount of information from each selected sensor. Indicates the first The selected sensor and the first The difference in the main body number of each sensor is zero when two sensors belong to the same participating entity. The sensor number and the main body number are both known data, which are determined by the production flow chart during the preprocessing stage.

7. The fibered faced gypsum board production full-process management system according to claim 6, characterized in that, The abnormality degree calculation module includes: A data sequence query unit is configured to receive sensor data uploaded by all sensors and containing a first identity tag, count the sensor data according to the identity tag, and obtain a data sequence corresponding to each sensor; the first identity tag is used to represent which sensor uploads the sensor data; A frequency domain graph generation unit is configured to read data sequences of each sensor of the same participating subject, perform Fourier transform on each data sequence, and obtain a frequency domain graph; A first calculation unit is configured to determine a sensor abnormality degree of the participating subject according to the frequency domain graph; An identification accuracy determination unit is configured to receive images uploaded by all cameras and containing a second identity tag, query the participating subject corresponding to the camera according to the position tag, read the sensor abnormality degree of the participating subject, and determine the identification accuracy of the image; the identification accuracy is used to adjust the identification resource amount; the second identity tag is used to represent which camera uploads the image; A second calculation unit is configured to identify the image based on the identification accuracy, and determine a visual abnormality degree of the participating subject; A comprehensive calculation unit is configured to determine the abnormality degree of the participating subject according to the sensor abnormality degree and the visual abnormality degree.

8. The fibered faced gypsum board production full-process management system according to claim 7, characterized in that, The content of determining the sensor abnormality degree of the participating subject according to the frequency domain graph includes: comparing the frequency domain graphs of different sensors of the same participating subject, extracting a characteristic frequency domain graph, and synchronously determining a fidelity rate of the characteristic frequency domain graph; inputting the characteristic frequency domain graph into a trained frequency domain identification model, outputting a stability degree, and correcting the stability degree according to the fidelity rate; calculating the sensor abnormality degree according to the corrected stability degree; the sensor abnormality degree is inversely proportional to the corrected stability degree; The content of identifying the image based on the identification accuracy, and determining the visual abnormality degree of the participating subject includes: sequentially selecting convolution kernels in a preset convolution kernel table according to the identification accuracy, and performing matching identification on the image according to the selected convolution kernels; the convolution kernel table includes convolution kernel items and abnormal value items; each data item in the convolution kernel table is arranged in descending order based on the abnormal value; querying abnormal values corresponding to the convolution kernels that successfully pass the matching identification, and accumulating all obtained abnormal values; According to the accumulated abnormal value, the visual abnormality degree of the subject is determined; The step of comparing the frequency domain graphs of different sensors of the same subject, extracting a characteristic frequency domain graph, and synchronously determining the fidelity of the characteristic frequency domain graph comprises: Inquiring all sensors of the same subject, for any sensor, comparing its frequency domain graph with the frequency domain graphs of other sensors of the same subject to obtain a similarity; calculating a mean value of the similarities to obtain an average similarity; selecting the frequency domain graph of the sensor with the maximum average similarity as the characteristic frequency domain graph; and taking the average similarity as the fidelity.

Images