A method and system for monitoring process parameters for glass sheet processing

Abstract

The application belongs to the technical field of data processing, and particularly relates to a processing process parameter monitoring method and system for glass plate processing, which comprises the following steps: performing adaptive differential coding on data sequences of a glass plate processing production line; obtaining the correlation of any two data sequences according to the similarity of the coding sequences of the different data sequences and the similarity in the adaptive differential coding process; obtaining the data quality of each data sequence based on the volatility of the coding sequences of the data sequences and the correlation of the different data sequences; and establishing a real-time monitoring model of the glass plate processing production line according to the data quality of the data sequences, the coding sequences of the data sequences and the correlation of any two data sequences. The application improves the accuracy and efficiency of real-time data monitoring of the glass plate processing production line.

Description

Technical Field

[0001] This invention relates to the field of data processing technology. More specifically, this invention relates to a method and system for monitoring processing parameters in glass plate processing. Background Technology

[0002] Glass, as a colorless and transparent amorphous inorganic material, possesses excellent light transmittance, chemical stability, and heat resistance, and is widely used in construction, automotive, and electronics industries. Glass sheets refer to thick, flat glass products with a wide range of applications; for example, in industry, they are used as a base material in the manufacture of glass sheet capacitors, whose stability and insulation properties meet the requirements of electronic components. The processing of glass sheets can be referenced in Chinese patent document CN118125135B, which discloses a glass sheet processing production line. This line discloses the processing of glass sheets using components such as edging equipment, cutting equipment, conveyor rollers, and motors, enabling safe processing and transportation of glass sheets.

[0003] In the glass sheet processing, to ensure the pass rate and the safety and normal operation of the production line, it is necessary to monitor the processing parameters. Related technologies, such as the Chinese patent document with authorization announcement number CN118607267B, disclose a method and system for constructing a digital twin of a sheet production line. This method discloses a way to accurately identify equipment anomalies by integrating image and sensor data, solving the problem that when using a digital twin model for equipment monitoring, it is impossible to specifically understand the abnormal state of the equipment.

[0004] However, in the process of using digital twin models to monitor production lines, the complexity of the workshop environment and the variety of data types that need to be monitored lead to insufficient data quality, resulting in poor data monitoring effects on the glass plate processing production line. Furthermore, different types of data may be correlated, and monitoring different data separately lacks analysis of these correlations, failing to fully utilize the digital twin model. Summary of the Invention

[0005] To address the technical problem of poor monitoring results and reduced glass processing accuracy caused by separately detecting low-quality multimodal data when using digital twin models to monitor multimodal data in a glass plate processing production line, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for monitoring processing parameters in glass plate processing, comprising:

[0007] Several data sequences from a glass plate processing production line are acquired. These data sequences are obtained from different sensors at different stages of the production line, with time differences between these stages, representing the time variations in monitoring the same injection mold. The period of each data sequence is determined based on its variation patterns. Adaptive differential encoding is performed on each data sequence based on its period, periodicity, and numerical uniformity to obtain its encoded sequence. The correlation between any two data sequences is determined based on the similarity of their encoded sequences and the similarity observed during the adaptive differential encoding process. The data quality of each data sequence is determined based on the volatility of its encoded sequences and the correlation between different data sequences. Abnormal performance of each data sequence is identified based on its data quality, encoded sequences, and the correlation between any two data sequences. Finally, a real-time monitoring model for the glass plate processing production line is established based on the abnormal performance of each data sequence.

[0008] This invention utilizes digital twin technology to monitor the entire process data of a glass plate processing production line. It compresses the entire process data and establishes a real-time monitoring model, ensuring data consistency and real-time monitoring. By analyzing the correlation between different data points and combining their monitoring data, this invention fully utilizes the digital twin model of the glass plate processing production line, improving the effectiveness of data processing and the accuracy of data monitoring.

[0009] Preferably, obtaining the period of each data sequence based on the variation pattern of the data sequence includes:

[0010] Denote any data sequence as the target sequence. Set a sliding window. When the sliding window is at a preset position in the target sequence, the data within the sliding window is denoted as the target subsequence. Slide the sliding window in the target sequence with equal steps. During the sliding process, the data within the sliding window is denoted as an equal-length subsequence of the target subsequence, resulting in several equal-length subsequences of the target subsequence. Based on the time distance and data differences between the target subsequence and each equal-length subsequence, obtain the periodicity of the target sequence. Obtain the period parameter of the target sequence when its periodicity is maximized, and denot it as the period of the target sequence.

[0011] The present invention obtains the period of each data sequence, avoiding the problem of insufficient real-time performance caused by excessive running time due to data compression on a single data unit, thus making data compression more efficient.

[0012] Preferably, obtaining the periodicity of the target sequence includes:

[0013] Obtain the Euclidean distance between the target subsequence and each subsequence of equal length, and perform negative correlation normalization to obtain the similarity between the target subsequence and each subsequence of equal length; obtain the temporal distance between the target subsequence and all subsequences of equal length; set the periodicity parameter of the target sequence. ;

[0014] ;

[0015] In the formula, This indicates the periodicity of the target sequence; Indicates the number of equal-length subsequences of the target subsequence; This represents the time distance between the target subsequence and the i-th subsequence of equal length; The periodicity parameter representing the target sequence; This represents the similarity between the target subsequence and the i-th subsequence of equal length; Represents the modulo function; Represents the absolute value function; This represents the normalization function.

[0016] Preferably, the adaptive differential encoding of each data sequence includes: counting the frequency of each value in the target sequence; obtaining the adjustability of the differential operation window of the target sequence and the differential operation window of the target sequence based on the periodicity of the target sequence, the period of the target sequence, and the uniformity of the frequency distribution of the values ​​in the target sequence; taking the second period of the target sequence as the first period to be encoded, within the differential operation window range of the target sequence of the period to be encoded, obtaining the part with the smallest Euclidean distance to the period to be encoded, and recording it as the reference differential window of the period to be encoded; performing differential operation between the period to be encoded and the reference differential window to complete the encoding of the period to be encoded; sequentially obtaining the period to be encoded from the starting point of the unencoded data and completing the encoding, with all data except the first period being encoded only once.

[0017] This invention uses adaptive differential coding for each data sequence to improve the coding applicability, reduce the data research cost caused by the increasing amount of subsequent monitoring data, improve coding flexibility, and avoid data compression misalignment caused by data noise and other factors, thus affecting data compression efficiency.

[0018] Preferably, the adjustable degree of the difference operation window for the target sequence and the difference operation window for the target sequence include:

[0019] The absolute values ​​of the differences between any two different values ​​in the target sequence are summed, then positively correlated and normalized. Finally, the sum is multiplied by the period of the target sequence to determine the adjustability of the difference operation window for the target sequence.

[0020] The range of the difference operation window for the target sequence is: ;

[0021] in, The window representing the difference operation of the target sequence. This indicates the adjustability of the difference operation window for the target sequence. Indicates the period of the target sequence.

[0022] Preferably, obtaining the correlation between any two data sequences based on the similarity of their encoded sequences and their similarity in the adaptive differential coding process includes: obtaining the DTW distance between the encoded sequences of any two data sequences; obtaining the DTW distance between the differential operation window sequences of the two data sequences; multiplying the DTW distance between the encoded sequences of the two data sequences and the DTW distance between the differential operation window sequences of the two data sequences, and performing negative correlation normalization to obtain the correlation between the two data sequences.

[0023] This invention calculates the correlation between different data sequences based on the similarity of data sequences and the similarity of encoding processes, providing a reference for the monitoring of low-quality data and improving the accuracy of low-quality data monitoring.

[0024] Preferably, obtaining the data quality of each data sequence includes:

[0025] ;

[0026] In the formula, This indicates the data quality of the p-th data sequence; This represents the length of the encoded sequence of the p-th data sequence; This represents the value of the c-th data in the encoded sequence of the p-th data sequence; Indicates the number of data sequences; This represents the correlation between the p-th data sequence and the q-th data sequence; , This represents the number of extreme points in the p-th and q-th data sequences; Represents an exponential function with the natural constant as its base; This represents the normalization function.

[0027] Preferably, the step of acquiring the abnormal behavior of each data point in each data sequence includes:

[0028] Let any data in the p-th data sequence be the target data; and let the data in the remaining data sequences that match the time difference of the process with the target data be the relevant data of the target data.

[0029] ;

[0030] In the formula, This indicates abnormal behavior in the target data; This indicates the data quality of the p-th data sequence; This represents the numerical value of the encoded data corresponding to the target data; This represents the number of reference sequences for the p-th data sequence; This indicates the data quality of the w-th reference sequence of the p-th data sequence; This represents the numerical value of the encoded data corresponding to the relevant data of the target data in the w-th reference sequence.

[0031] This invention acquires abnormal data, enabling real-time and accurate monitoring of the production process data of the injection mold production line, and timely identification of abnormal nodes and abnormal injection molds in the injection mold production line.

[0032] Preferably, obtaining the data in the remaining data sequence that matches the time difference of the process with the target data includes:

[0033] Obtain a data sequence with a quality greater than that of the p-th data sequence, and denote it as the reference sequence of the p-th data sequence; shift the p-th data sequence over the reference sequence of the p-th data sequence with a preset step size; when the average distance between the overlapping parts of the p-th data sequence and the reference sequence of the p-th data sequence during the shift process reaches its minimum, the time distance of the p-th data sequence shift is denoteed as the process time difference between the p-th data sequence and the reference sequence of the p-th data sequence; obtain the data in the reference sequence of the p-th data sequence whose time distance from the target data is the process time difference, and denote it as the data in the remaining data sequences that has the same process time difference as the target data.

[0034] Secondly, the present invention provides a process parameter monitoring system for glass plate processing, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned process parameter monitoring method for glass plate processing is implemented.

[0035] By adopting the above technical solution, a computer program is generated from the above-mentioned method for monitoring processing parameters of glass plate processing, and stored in a memory for loading and execution by a processor. This allows for the creation of a terminal device based on the memory and processor, making it convenient to use.

[0036] The beneficial effects of this invention are as follows:

[0037] (1) This invention monitors all the monitoring data of the glass plate processing production line as a whole, making full use of all the monitoring data. The data with strong correlations are referenced to each other, which can improve the overall monitoring effect of the glass plate processing production line.

[0038] (2) In this invention, all monitoring data are compressed using the same encoding method, while different encoding parameters are used for different data, so that all monitoring data can be fully compressed while ensuring the real-time performance of data monitoring;

[0039] (3) The present invention obtains the process time difference of different processes on the glass plate processing production line. When abnormal injection molds or abnormal processes are detected, the monitoring time and data of abnormal glass plates in each process can be quickly extracted, as well as the injection molds affected by the abnormal processes can be quickly extracted, so as to provide timely feedback and adjustment, thereby improving the production efficiency of the glass plate processing production line. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating a method for monitoring processing parameters in glass plate processing according to the present invention;

[0041] Figure 2 This is a schematic diagram illustrating the architecture of a digital twin;

[0042] Figure 3 This is a schematic diagram illustrating the time difference between data sequence I and data sequence II. Detailed Implementation

[0043] This invention discloses a method for monitoring processing parameters in glass plate processing, referring to... Figure 1 This includes steps S1-S4:

[0044] S1: Obtain several data sequences from the glass plate processing production line.

[0045] It should be noted that the production and processing of glass sheets involves several steps, such as cutting, grinding and polishing, and bending. Each step requires monitoring several types of data, such as temperature, cooling rate, and surface images of the glass sheet after processing to detect whether there are defects such as dimensional defects. Therefore, in order to comprehensively monitor the glass sheet processing production line, it is necessary to monitor all data.

[0046] It should be noted that, as Figure 2This diagram illustrates a digital twin architecture, employing a physical layer, edge layer, cloud layer, and application layer to achieve real-time monitoring of a glass processing production line. The physical layer collects data from various sensors along the production line. To meet real-time requirements, the edge layer compresses the sensor data. To improve monitoring effectiveness, the cloud layer analyzes the compressed data, considering data quality and filtering for correlations within the process, and performs anomaly detection. This fully leverages the overall structure of the digital twin model, improving the accuracy of real-time monitoring of the glass processing production line. Finally, the real-time monitored data is visualized, reducing the learning curve for staff and increasing monitoring efficiency.

[0047] Specifically, the data output terminals of sensors at several steps in the glass plate processing production line are connected to a computer to acquire data sequences of several data items in real time. These data sequences include data from several sampling times. It should be noted that the sampling time intervals for different data items vary. For example, when the sampling frequency of the infrared thermal imager is 5Hz, the sampling time interval for glass plate stress and images depends on the production cycle of the glass plates. For instance, if one glass plate is produced per minute, the sampling time interval for glass plate stress and images is one minute.

[0048] At this point, a data sequence of several data items has been obtained.

[0049] S2: Based on the changing patterns of the data sequences, obtain the period of each data sequence; based on the period of each data sequence, the periodicity of the data sequences, and the uniformity of the numerical distribution of the data sequences, perform adaptive differential coding on each data sequence to obtain the coding sequence of each data sequence.

[0050] It should be noted that different data have different dimensions and formats. If different compression methods are used for different types of data, the data processing will become increasingly complex as the number of data types increases, thus deviating from the purpose of real-time monitoring. Glass sheets are produced on a large scale on production lines, therefore, the sensors monitoring the glass sheet processing process exhibit data change cycles. For example, an infrared thermal imager monitoring temperature changes during the glass sheet bending process begins heating when the glass sheet arrives at the bending equipment, then maintains the temperature during bending, and finally cools it after bending. A suitable cooling rate can prevent thermal shock cracks. Therefore, obtaining the change cycle of the data sequence is beneficial for detecting abnormal data changes. Thus, this invention first obtains the cycle of each data sequence. Secondly, the cycle of the data sequence provides a more efficient way to compress data. When the data changes completely according to the cycle, all subsequent cycles except the first cycle can be replaced by the first cycle, which can significantly reduce the storage space occupied by the data and the amount of monitoring computation, thereby ensuring the real-time nature of data monitoring. However, in actual monitoring, due to noise and other reasons, the data in continuous periods of the data sequence are not exactly the same. Differential operation can effectively reflect the changes in such continuous periods of data and reduce redundant storage caused by duplicate data. Therefore, this invention uses the same differential operation to compress each data, and determines the differential operation window for each data based on the period of different data to compress the data to the maximum extent.

[0051] Specifically, an arbitrary data sequence is designated as the target sequence. A sliding window is set, and when the sliding window is at a preset position in the target sequence, the data within the sliding window is designated as the target subsequence. The target subsequence is then slid across the target sequence with equal step sizes to obtain several subsequences of equal length. It should be noted that the size of the sliding window and the step size are set by the implementers according to the actual implementation situation. For example, the size of the sliding window can be set to 10 sampling times, and the step size can be set to 1 sampling time.

[0052] It should be noted that in the equal-length subsequences of the target subsequence, when the distance between the equal-length subsequence and the target subsequence is a multiple of the period of the data in which the target subsequence is located, the equal-length subsequence and the target subsequence are relatively similar. Therefore, the period can be determined based on the similarity between the target subsequence and its equal-length subsequence.

[0053] Preferably, the period of obtaining the target sequence includes:

[0054] The Euclidean distance between the target subsequence and each equal-length subsequence is obtained and negatively correlated and normalized, serving as the similarity between the target subsequence and each equal-length subsequence. It should be noted that if the target sequence is an image sequence, the Euclidean distance between the target subsequence and all equal-length subsequences is the sum of the Euclidean distances of the grayscale values ​​at all corresponding positions in all corresponding images of the target subsequence and each equal-length subsequence.

[0055] Obtain the time distance between the target subsequence and all subsequences of equal length; set the period parameter of the target sequence. .

[0056] It should be noted that the higher the similarity between equal-length subsequences and the target subsequence, the more significant the periodic relationship. In addition, if the time distance between adjacent equal-length subsequences with similarity is a multiple of the period, the periodicity of the target subsequence will be more significant.

[0057] The periodicity of the target sequence satisfies the expression:

[0058] ;

[0059] In the formula, This indicates the periodicity of the target sequence; Indicates the number of equal-length subsequences of the target subsequence; This represents the time distance between the target subsequence and the i-th subsequence of equal length; The periodicity parameter representing the target sequence; This represents the similarity between the target subsequence and the i-th subsequence of equal length; Represents the modulo function; Represents the absolute value function; This represents the normalization function.

[0060] In the formula, This represents the remainder of the ratio of the time distance between the target subsequence and the i-th subsequence of equal length to the period parameter of the target sequence. This represents the difference between the remainder and the period parameter of half the target sequence. This value reflects the probability that the time distance between the target subsequence and the i-th equal-length subsequence is the period of the target sequence. The larger the value, the further the remainder is from the period of half the target sequence, and the closer it is to an integer multiple of the period of the target sequence. Therefore, the greater the probability that the time distance between the target subsequence and the i-th equal-length subsequence is the period of the target sequence. On this basis, if the similarity between the target subsequence and the i-th equal-length subsequence is greater, the stronger the periodicity of the i-th equal-length subsequence of the target subsequence relative to the target sequence. This represents a weighted sum of the probability that the period of the target subsequence is the temporal distance between the target subsequence and all subsequences of equal length, with the similarity between the target subsequence and the i-th subsequence of equal length as the weight. The larger this value is, the stronger the periodicity of the target sequence.

[0061] The period parameter of the target sequence is obtained when the periodicity of the target sequence is maximized, and is denoted as the period of the target sequence.

[0062] At this point, the period of the data sequence for each data item has been obtained.

[0063] It should be noted that since the period of a data sequence is not completely stable, using the period of the data sequence as the window size for differential encoding may cause misalignment and affect the encoding effect. Therefore, the window for differential operation should be adjusted within a certain range to achieve better encoding results and avoid the accumulation of encoding errors. The maximum period of different data sequences varies. In some noisy data or data with insufficient regularity, the maximum period of the data sequence may be relatively small, so the range of the differential operation window can be adjusted to be relatively large.

[0064] Preferably, based on the periodicity of the target sequence, the period of the target sequence, and the uniformity of the frequency distribution of the values ​​in the target sequence, the adjustability of the difference operation window of the target sequence and the difference operation window of the target sequence are obtained, including:

[0065] Count the frequency of each value in the target sequence;

[0066] The adjustability of the difference window for the target sequence satisfies the expression:

[0067] ;

[0068] In the formula, Indicates the adjustability of the difference operation window for the target sequence; This indicates the periodicity of the target sequence; Indicates the type of numerical values ​​in the target sequence; , Indicates the frequency of the u-th and v-th values ​​in the target sequence; Represents the maximum value function; Represents the absolute value function; This represents the normalization function.

[0069] In the formula, This represents the frequency difference between the u-th and v-th values ​​in the target sequence; This represents the summation of frequency differences for all different values ​​in the target sequence. The larger this value, the more uneven the data distribution of the target sequence, the more noise there is, and the more irregular the data is. Therefore, the greater the adjustability of the difference operation window for the target sequence.

[0070] The range of the difference operation window for the target sequence is:

[0071] ;

[0072] ;

[0073] In the formula, The window representing the difference operation of the target sequence; Indicates the adjustability of the difference operation window for the target sequence; Indicates the period of the target sequence; The periodicity parameter representing the target sequence; This indicates the periodicity of the target sequence; Represents the maximum value function; The period parameter represents the period of the target sequence, corresponding to the maximum value of its periodicity. It should be noted that the difference window for the target sequence ranges from 0 to twice the period of the target sequence.

[0074] Preferably, the second period of the target sequence is taken as the first period to be encoded. Within the difference operation window of the target sequence of the period to be encoded, the part with the smallest Euclidean distance to the period to be encoded is obtained and recorded as the reference difference window of the period to be encoded. The period to be encoded and the reference difference window are subjected to difference operation to complete the encoding of the period to be encoded. The period to be encoded is obtained from the starting point of the unencoded data in sequence and the encoding is completed. All data except the first period are encoded only once.

[0075] At this point, the encoded data for each item has been obtained.

[0076] S3: Based on the similarity of the encoded sequences of different data sequences and the similarity in the adaptive differential coding process, obtain the correlation between any two data sequences; based on the volatility of the encoded sequences of the data sequences and the correlation between different data sequences, obtain the data quality of each data sequence; based on the data quality of the data sequences, the encoded sequences of the data sequences, and the correlation between any two data sequences, obtain the abnormal performance of each data in each data sequence.

[0077] It's important to note that the purpose of real-time monitoring of the glass processing production line is to ensure smooth glass processing and promptly identify problematic processes. An anomaly in one glass sheet can lead to anomalies being detected in subsequent processes; therefore, different data points have a temporal correlation. Furthermore, some data points are inherently correlated; for example, insufficient glass sheet thickness can cause faster temperature changes during the bending process. Figure 3 This diagram illustrates the time difference between data sequence I and data sequence II. When different data sequences are strongly correlated, the windows for differential operations during encoding are quite similar. Therefore, this invention analyzes the correlation between different data based on the encoding process and the encoded data.

[0078] It should be further noted that monitoring each data point individually and screening out abnormal data separately not only fails to make full use of the digital twin model, but is also greatly affected by the quality of the data itself, resulting in poor monitoring results. Therefore, for data with poor quality, high-quality data with strong correlation can be referenced for anomaly monitoring, thereby improving the accuracy of real-time monitoring of the glass plate processing production line.

[0079] Specifically, based on the similarity of the encoded sequences of different data sequences and their similarity in the adaptive differential coding process, the correlation between any two data sequences is obtained:

[0080] The correlation between the two data sequences is obtained by: obtaining the DTW distance between the encoded sequences of any two data sequences; obtaining the DTW distance between the difference window sequences of the two data sequences; multiplying the DTW distance between the encoded sequences of the two data sequences and the DTW distance between the difference window sequences of the two data sequences, and performing negative correlation normalization. It should be noted that the DTW distance is the Dynamic Time Warped Distance, which is existing technology.

[0081] It should be noted that the high quality and stability of the data sequence are strongly correlated with the glass plate processing and production. In other words, the changes in the data sequence are only related to the production of glass plates and are not related to the environment. Therefore, the encoding effect of the data sequence is better, and differential calculation can make most of the data become 0. In related processes, referring to high-quality data can monitor the data more accurately.

[0082] Preferably, the data quality of each data sequence is obtained based on the volatility of the encoded sequence and the correlation between different data sequences, including:

[0083] Obtain the number of extreme points in the encoded sequence of each data sequence;

[0084] ;

[0085] In the formula, This indicates the data quality of the p-th data sequence; This represents the length of the encoded sequence of the p-th data sequence; This represents the value of the c-th data in the encoded sequence of the p-th data sequence; Indicates the number of data sequences; This represents the correlation between the p-th data sequence and the q-th data sequence; , This represents the number of extreme points in the p-th and q-th data sequences; Represents an exponential function with the natural constant as its base; This represents the normalization function.

[0086] In the formula, This indicates that all encoded data in the p-th data sequence are accumulated. The larger the value, the worse the encoding effect, and therefore the worse the data quality. This represents the difference in the number of extreme points between the p-th and q-th data sequences. This value reflects the stability of the p-th data sequence relative to the q-th data sequence. The larger this value is, the stronger the volatility of the p-th data sequence and the weaker its stability. At the same time, the stronger the correlation between the p-th and q-th data sequences, the lower the relative quality of the p-th data sequence relative to the q-th data sequence. This represents a weighted summation of the differences in the number of extreme points between the p-th data sequence and all data sequences, with the correlation of the data sequences as the weight. This value reflects the relative quality of the p-th data sequence relative to all data sequences. The larger the value, the lower the data quality of the p-th data sequence. express It is negatively correlated with the data quality of the p-th data sequence.

[0087] At this point, the data quality of each data sequence has been obtained.

[0088] It should be noted that on the glass plate processing production line, data anomalies manifest as sudden numerical changes. Therefore, the code of the data to be tested is compared with the corresponding code data of previous cycles. The greater the difference, the more abnormal the data to be tested is. Based on this, the abnormal behavior of the data to be tested is obtained by referring to data with strong correlation and high quality.

[0089] Preferably, based on the data quality of the data sequence, the encoding sequence of the data sequence, and the correlation between any two data sequences, the abnormal behavior of each data in each data sequence is obtained, including:

[0090] Let any data in the p-th data sequence be the target data.

[0091] Obtain a data sequence with a quality greater than that of the p-th data sequence, and denote it as the reference sequence of the p-th data sequence; shift the p-th data sequence over the reference sequence of the p-th data sequence with a preset step size; when the average distance of the overlapping part of the p-th data sequence and the reference sequence of the p-th data sequence reaches its minimum during the shift process, the time distance of the p-th data sequence shift is denoteed as the process time difference between the p-th data sequence and the reference sequence of the p-th data sequence; obtain the data in the reference sequence of the p-th data sequence whose time distance from the target data is the process time difference, and denote it as the data in the remaining data sequences that have the same process time difference as the target data, and denote it as the relevant data of the target data.

[0092] The abnormal behavior of the target data satisfies the expression:

[0093] ;

[0094] In the formula, This indicates abnormal behavior in the target data; This indicates the data quality of the p-th data sequence; This represents the numerical value of the encoded data corresponding to the target data; This represents the number of reference sequences for the p-th data sequence; This indicates the data quality of the w-th reference sequence of the p-th data sequence; This represents the numerical value of the encoded data corresponding to the relevant data of the target data in the w-th reference sequence.

[0095] In the formula, This means that if the target data is large and the data quality of the data sequence is high, then the anomaly of the target data is more accurate, thus indicating that the anomaly of the target data is stronger. This indicates the abnormal behavior of related data in the data sequence containing data of higher quality than the target data. The larger this value, the more abnormal the reference data of the target data is, and the stronger the abnormal behavior of the target data.

[0096] At this point, the abnormal behavior of each data point in each data sequence has been obtained.

[0097] S4: Based on the abnormal performance of each data in each data sequence, establish a real-time monitoring model for the glass plate processing production line.

[0098] Specifically, the data acquired in real time from each data sequence is used to obtain abnormal performance through the methods in steps S1-S3, and then visualized in real time in the digital twin model to complete the real-time monitoring of the glass plate processing production line. When a significant abnormal performance is detected, the digital twin model provides an audio-visual alert, enabling staff to handle the situation promptly and avoid continuous losses.

[0099] This completes the real-time monitoring of the glass plate processing production line.

[0100] This invention also discloses a process parameter monitoring system for glass plate processing, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a process parameter monitoring method for glass plate processing according to the present invention is implemented.

[0101] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0102] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for monitoring processing parameters in glass plate processing, characterized in that, The application relates to a method for monitoring a glass plate processing production line. The method comprises the following steps: acquiring a plurality of data sequences of the glass plate processing production line; the data sequences are obtained by different sensors in different processes of the glass plate processing production line, and the different processes have process time differences; the process time differences are time differences of monitoring the same injection mold in different processes; acquiring a period of each data sequence according to a change rule of the data sequence; based on the period of each data sequence, a condition that the data sequence meets a period change, and a numerical value distribution uniformity of the data sequence, performing adaptive differential coding on each data sequence to acquire a coded sequence of each data sequence; the adaptive differential coding on each data sequence comprises the following steps: counting a frequency of each numerical value of a target sequence; acquiring an adjustable degree of a differential operation window of the target sequence and the differential operation window of the target sequence according to a period performance of the target sequence, a period of the target sequence and a frequency distribution uniformity of numerical values of the target sequence; taking a second period of the target sequence as a first to-be-coded period; acquiring a part with a minimum Euclidean distance in a reference differential window of the to-be-coded period in a range of the differential operation window of the target sequence in the to-be-coded period, and recording the part as the reference differential window of the to-be-coded period; performing differential operation on the to-be-coded period and the reference differential window to complete coding of the to-be-coded period; sequentially acquiring the to-be-coded period from a starting point of uncoded data and completing coding; all data except the first period are coded and coded only once; ； In the formula, represents the data quality of the pth data sequence; represents the length of the encoded sequence of the pth data sequence; represents the cth data value of the encoded sequence of the pth data sequence; represents the number of data sequences; represents the correlation of the pth data sequence and the qth data sequence; , represents the number of extreme points of the pth data sequence and the qth data sequence; represents an exponential function with a natural constant as the base; represents a normalization function; acquiring a correlation between any two data sequences according to a similarity of coded sequences of different data sequences and a similarity in the adaptive differential coding process; acquiring data quality of each data sequence based on volatility of the coded sequences of the data sequences and the correlation between different data sequences, comprising the following steps: acquiring an abnormal performance of each data of each data sequence according to the data quality of the data sequence, the coded sequence of the data sequence and the correlation between any two data sequences, comprising the following steps: ； In the formula, represents the abnormal performance of the target data; represents the data quality of the pth data sequence; represents the value of the corresponding encoded data of the target data; represents the number of reference sequences of the pth data sequence; represents the data quality of the wth reference sequence of the pth data sequence; represents the value of the corresponding encoded data of the related data of the target data in the wth reference sequence. taking any data of a pth data sequence as target data; acquiring data of other data sequences that meet a process time difference with the target data, and recording the data as related data of the target data; 2. The method of claim 1, wherein establishing a real-time monitoring model of the glass plate processing production line according to the abnormal performance of each data of each data sequence. the method comprises the following steps: taking any data sequence as a target sequence; setting a sliding window; recording data in the sliding window as a target subsequence when the sliding window is at a preset position of the target sequence; sliding the sliding window in the target sequence at equal steps; recording data in the sliding window during the sliding process as an equal-length subsequence of a target subsequence; and obtaining a plurality of equal-length subsequences of the target subsequence; acquiring a period performance of the target sequence according to a time distance and a data difference between the target subsequence and each equal-length subsequence; 3. The method of claim 2, wherein acquiring a period parameter of the target sequence when the period performance of the target sequence is maximum, and recording the period parameter as a period of the target sequence. The Euclidean distance of the target subsequence and each equal-length subsequence is obtained, and negative correlation normalization is performed to serve as the similarity of the target subsequence and each equal-length subsequence; the time distance of the target subsequence and all equal-length subsequences is obtained; and the period parameter of the target sequence is set ; ； wherein, represents the periodicity of the target sequence; represents the number of equal-length subsequences of the target subsequence; represents the temporal distance of the target subsequence from the i-th equal-length subsequence; represents the periodicity parameter of the target sequence; represents the similarity of the target subsequence to the i-th equal-length subsequence; represents the modulo function; represents the absolute value function; represents the normalization function.

4. The method of claim 2, wherein the method comprises the following steps: the method comprises the following steps: The absolute value of the difference between any two different numerical categories of the target sequence is summed, then positively correlated and normalized, and multiplied by the period of the target sequence to obtain the adjustable degree of the difference operation window of the target sequence; The range of the difference operation window of the target sequence is: ; wherein, denotes a differential operation window of the target sequence, denotes an adjustable degree of the differential operation window of the target sequence, denotes a period of the target sequence.

5. The method of claim 1, wherein The correlation of any two data sequences is obtained according to the similarity of the encoding sequences of the different data sequences and the similarity in the adaptive difference encoding process, and the correlation of any two data sequences comprises: The DTW distance of the encoding sequences of the two data sequences is obtained, the DTW distance of the difference operation window sequences of the two data sequences is obtained, the DTW distance of the encoding sequences of the two data sequences is multiplied by the DTW distance of the difference operation window sequences of the two data sequences, and negatively correlated and normalized to obtain the correlation of the two data sequences.

6. The method of claim 1, wherein The data in the remaining data sequences that conforms to the process time difference with the target data comprises: A data sequence with a data quality greater than the pth data sequence is obtained, which is recorded as a reference sequence of the pth data sequence; the pth data sequence is shifted on the reference sequence of the pth data sequence by a preset step size; the time distance of the pth data sequence when the average distance of the overlapping part of the pth data sequence and the reference sequence of the pth data sequence is the minimum value is recorded as the process time difference between the pth data sequence and the reference sequence of the pth data sequence; the data in the reference sequence of the pth data sequence that conforms to the process time difference with the target data is recorded as the data in the remaining data sequences that conforms to the process time difference with the target data.

7. A process parameter monitoring system for glass sheet processing, characterized by, The processor and the memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, a processing parameter monitoring method for glass plate processing according to any one of claims 1-6 is realized. ​

Images