Display screen production process monitoring system and method based on Internet of Things

Abstract

The invention discloses a display screen production process monitoring system and method based on the Internet of Things, and relates to the technical field of the Internet of Things. Process data of a plurality of processes in the display panel manufacturing process are obtained, the process data comprise process parameters, equipment operation data and output detection data, and quantitative processing is carried out; generating a process data feature vector of each process; based on the process data feature vector, constructing a state representation vector of the process; on the basis of the state representation vector of each process, a correlation model between the processes is constructed in combination with the correlation degree between the processes; in the process of executing the process, the defect risk index is calculated, the process with the defect is determined and early warning is performed, and the potential risk is identified before the defect occurs, so that the response speed and the processing efficiency of production are improved, and the accuracy and the reliability of defect prediction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet of Things, in particular to a display screen production process monitoring system and method based on Internet of Things. BACKGROUND

[0002] In the display panel manufacturing process, the formation of display defects is often not directly caused by a single process or a single parameter, but is the result of the gradual accumulation, amplification or transformation of process states in multiple processes. The existing display defect monitoring methods mostly take a single process as the monitoring object, focus on independent analysis of process parameters or detection results, and usually use threshold judgment, statistical rules or local models to identify abnormalities. However, these methods cannot reflect the inheritance, evolution of abnormal states in different processes and their cumulative impact on the final display defects, resulting in monitoring results that mostly stay at "finding problems" rather than "explaining problems". At the same time, the defect tracing mechanism usually starts from the defect result and infers the possible abnormal process by checking the historical process data. It usually relies on manual experience or static rules for screening, and it is difficult to distinguish between the process states that truly have causal contribution and a large number of irrelevant process fluctuations. Not only is the tracing path long, but it is also prone to inconsistent root cause judgments, making it difficult to support rapid adjustment and closed-loop optimization of the production process. Therefore, how to break through the limitations of single-process monitoring and post-tracing, identify defect-related causal states from a process level, and realize the continuous transmission and traceable expression of the state across processes, so as to complete early warning and positioning before the formation or early manifestation of defects, has become a technical problem that needs to be solved in current display screen production processes. SUMMARY

[0003] The purpose of the present application is to provide a display screen production process monitoring system and method based on Internet of Things to solve the problems in the prior art.

[0004] To solve the above technical problems, the present application provides the following technical solution: a display screen production process monitoring method based on Internet of Things, the method comprising: Step S100: acquiring process data of multiple processes in the display panel manufacturing process, the process data including process parameters, equipment operation data and output detection data, and performing quantitative processing to generate process data feature vectors of each process; Step S200: constructing a state representation vector of the process based on the process data feature vector; Step S300: constructing an association model between processes based on the state representation vector of each process and the correlation degree between processes; Step S400: during the execution of the process, determining the process with defects and issuing a warning by calculating a defect risk index; Further, the process data feature vector of each process generated in step S100 includes: In the historical production record, the process data of each process is collected and classified and summarized according to process parameters, equipment operation data and output detection data; Feature data is extracted for each type of process data, including statistical features, trend features and fluctuation features, the statistical features include mean, standard deviation and variance, the trend features are the change trend over time, and the fluctuation features include fluctuation amplitude and frequency; The feature data is normalized to obtain normalized feature data, and the normalized feature data is vectorized to generate the process data feature vector of the process; By classifying and summarizing the process data of each process and extracting statistical features, trend features and fluctuation features, key variables and their change rules in the production process can be effectively captured, which helps to understand the concentration trend and dispersion degree of the data, the trend features reveal the change trend of the data over time, and the fluctuation features help to identify the fluctuation amplitude and frequency, which can fully and accurately reflect the state and potential problems of each process in the production process. Through normalization processing, it is ensured that the data of different processes and process parameters are in the same scale range, avoiding the interference of different dimensions and data ranges on subsequent analysis, and the normalized feature data can ensure that the calculation results of the model are more stable, avoiding errors caused by extreme values or unbalanced data.

[0005] Further, the state representation vector of the process constructed in step S200 includes: The running state stage of each process is selected from the production process system, the historical production records of each running state stage are summarized, the process data feature vector corresponding to each historical production record is obtained, the threshold value of each process data feature vector is preset, the deviation value of the process data feature vector is calculated and the absolute value is summarized, and the average deviation value of the process data feature vector in each running state stage is calculated; The absolute deviation threshold value of each process data feature vector is preset, the process data feature vector exceeding the absolute deviation threshold value is set as an abnormal vector, and the abnormal frequency of each process data feature vector as an abnormal vector in each running state stage is calculated; The average deviation value and the abnormal frequency of the process data feature vectors in different running state stages are summarized, the total average deviation value and the total abnormal frequency are calculated, the total average deviation value and the total abnormal frequency are multiplied, the deviation score of the process data feature vector in each process is calculated, the process data feature vectors are sorted according to the deviation score from high to low, the deviation score difference value of the adjacent two process data feature vectors is calculated, the position corresponding to the maximum deviation score difference value is determined, and the number of process data feature vectors before the position is taken as the number of state feature vectors, and the state feature vectors are screened out; By screening the running state stages of each process from the production process system and respectively summarizing the historical production records of the corresponding stages, the data analysis is established on the basis of the specific running state, avoiding the interference caused by the mixing of different running conditions, thereby significantly improving the accuracy and pertinence of the process state recognition; By presetting the process data feature vector threshold value, calculating the absolute deviation value and counting the average deviation value in each running state stage, the quantitative evaluation of the deviation of the feature vector from the normal level is realized, the abnormal vector is screened out combined with the absolute deviation value threshold value, and the abnormal frequency is calculated, from the two dimensions of “deviation degree” and “abnormal occurrence probability”, the process fluctuation is described, and the comprehensiveness and reliability of the abnormal identification are improved; By fusing the average deviation value and the abnormal frequency, the deviation score is calculated to realize the comprehensive evaluation of the process data feature vector, not only considering the fluctuation amplitude, but also considering the frequency of abnormal occurrence, which can more objectively reflect the influence degree of the process parameters on the process state, and improve the scientificity of the state evaluation result; By sorting according to the deviation score from high to low and screening combined with the preset number of state feature vectors, the most representative process data feature vector can be automatically selected as the state feature vector, the redundant or less influential features are removed, the data dimension is reduced, and the subsequent analysis and modeling efficiency is improved.

[0006] Further, the step S300 of constructing the correlation model between processes includes: The state feature vectors of each process are weighted and summed to calculate a comprehensive index, in the same time window, based on the comprehensive index of each process, the correlation degree between any two processes is calculated by using the correlation coefficient method, a correlation degree threshold value is preset, adjacent processes exceeding the threshold value are screened out, and a defect correlation chain is constructed combined with the process sequence; A comprehensive index threshold value range of each process is preset, processes exceeding the comprehensive index threshold value range are set as abnormal processes, and historical production records in which both adjacent processes are abnormal processes are set as historical abnormal records; Collect the time interval between the previous abnormal process and the next abnormal process in the historical abnormal records, and use the time interval as the abnormal propagation time. Calculate the maximum deviation value of the comprehensive index corresponding to each process and set it as the fluctuation value of the process. Multiply the abnormal propagation time and the fluctuation value to calculate the propagation attenuation coefficient of the process. Summarize the propagation attenuation coefficients of the historical abnormal records to calculate the average propagation attenuation coefficient of the process. Using the fluctuation value, correlation degree, and anomaly propagation time of the comprehensive index of the previous process as input and the fluctuation value of the comprehensive index of the next process as output, a correlation model between processes is established based on a linear regression model. A comprehensive index is obtained by weighted summation of the state representation vectors of each process, and the correlation coefficient method is used to calculate the correlation between any two processes within the same time window. Strongly correlated adjacent processes are selected by combining the preset correlation threshold, and a defect correlation chain is constructed according to the process sequence to realize the structured expression of the propagation path of defects in the production process, revealing the influence relationship between processes from the perspective of the overall process. By setting a comprehensive index threshold range, processes that exceed the threshold range are identified as abnormal processes, and historical production records of two adjacent processes that are simultaneously abnormal are set as historical abnormal records. This avoids misjudgment caused by occasional fluctuations in a single process and improves the accuracy and stability of anomaly identification. By statistically analyzing the time interval between the occurrence of anomalies in previous and subsequent processes in historical anomaly records, this time interval is defined as the anomaly propagation time. Combined with the maximum deviation value of the comprehensive index of each process, the average propagation attenuation coefficient is calculated, thereby achieving a quantitative description of the anomaly propagation speed and attenuation characteristics. This can reveal the propagation intensity and attenuation trend of anomalies between different processes, and enhance the understanding of the dynamic behavior of the production system. By constructing a defect-related chain and process association model, when anomalies occur in subsequent processes, it is possible to quickly trace back to the possible preceding processes that may have affected them, thereby locating the source of defects. It is also possible to predict the risk changes of subsequent processes based on the abnormal fluctuations of preceding processes, thus achieving early warning and reducing quality risks.

[0007] Furthermore, step S400 includes calculating the defect risk index, which includes: During the execution of the process, the fluctuation value of the comprehensive index of the process and the estimated processing time are obtained and input into the correlation model between processes to predict the fluctuation value of the comprehensive index of the next process. Based on the defect correlation chain and combined with the correlation model, the fluctuation value of the comprehensive index of each process in the defect correlation chain is predicted; Based on the fluctuation value, correlation degree, and average propagation attenuation coefficient of the comprehensive indicators of the process, the defect risk index of each process is calculated. When the defect risk index exceeds the preset risk threshold, a defect warning signal is issued, prompting the corresponding process to handle the abnormality. By acquiring the fluctuation value of the current process's comprehensive index and the estimated processing time during the process execution, and inputting them into the established process inter-process correlation model, the fluctuation value of the next process's comprehensive index is predicted. This enables the system to predict the status of subsequent processes before actual anomalies occur, thus transforming post-event analysis into pre-event early warning and improving the initiative of production process control. By combining defect-related chains and correlation models, the fluctuation values ​​of comprehensive indicators of each process in the defect-related chain are predicted, enabling the linkage analysis of the abnormal propagation trend in the entire process link, avoiding judgment based on only a single process, and improving the overall control capability of the defect propagation process. By comprehensively calculating the fluctuation value of the process comprehensive index, the degree of correlation between processes, and the average propagation attenuation coefficient, the defect risk index of each process is obtained. This allows the risk assessment to consider the abnormal amplitude, the strength of process correlation, and the propagation characteristics simultaneously, which can more realistically reflect the comprehensive risk level of defects in each process and improve the accuracy and reliability of risk judgment. When the defect risk index exceeds the preset risk threshold, a defect warning signal is issued and the corresponding process is prompted to handle the anomaly. This allows the anomaly to be detected in a timely manner and accurately located to the specific process, which helps maintenance personnel to take targeted measures quickly, shorten response time, and reduce losses caused by the spread of defects. By identifying and intervening in high-risk processes in advance, defects can be effectively suppressed from spreading along the defect-related chain to subsequent processes, reducing the probability of batch quality problems and thus improving product consistency and overall yield.

[0008] To better implement the above methods, an IoT-based display production process monitoring system was also proposed. The system includes a process data feature vector module, a state representation vector module, an association model module, and a defect risk index module. Process data feature vector module: acquires process data of multiple processes in the manufacturing process of display panel, including process parameters, equipment operation data and output detection data, and performs quantification processing to generate process data feature vectors for each process; State representation vector module: Constructs a state representation vector for the process based on the feature vector of the process data; Association Model Module: Based on the state representation vectors of each process and the degree of correlation between processes, an association model between processes is constructed; Defect Risk Indicator Module: During the execution of a process, defect risk indicators are calculated to identify processes with defects and issue early warnings.

[0009] Furthermore, the process data feature vector module includes feature data units and a unit for generating process data feature vectors: Feature data unit: In historical production records, process data for each process is collected and classified and summarized according to process parameters, equipment operation data and output detection data; feature data is extracted for each type of process data, including statistical features, trend features and fluctuation features. The statistical features include mean, standard deviation and variance. The trend features are the trend of change over time. The fluctuation features include fluctuation amplitude and frequency. Generating process data feature vector units: Normalize the feature data to obtain normalized feature data, and then vectorize the normalized feature data to generate process data feature vectors for each process.

[0010] Furthermore, the state representation vector module includes a unit for calculating the anomaly frequency and a unit for filtering state representation vectors: The abnormal frequency calculation unit: filters the operating status stages of each process from the production process system, summarizes the historical production records of each operating status stage, obtains the process data feature vector corresponding to each historical production record, presets a threshold for each process data feature vector, calculates the deviation value of the process data feature vector and summarizes the absolute value, calculates the average deviation value of the process data feature vector in each operating status stage; presets an absolute deviation value threshold for each process data feature vector, sets process data feature vectors that exceed the absolute deviation value threshold as abnormal vectors, and calculates the abnormal frequency of each process data feature vector being an abnormal vector in each operating status stage. Screening State Representation Vector Units: The average deviation value and abnormal frequency of process data feature vectors in different operating state stages are summarized to calculate the total average deviation value and total abnormal frequency. The total average deviation value and total abnormal frequency are multiplied to calculate the deviation score of the process data feature vector in each process. The process data feature vectors are sorted from high to low according to the deviation score. The difference in deviation score between two adjacent process data feature vectors is calculated to determine the position corresponding to the maximum deviation score difference. The number of process data feature vectors before the position is taken as the number of state representation vectors, and the state representation vectors are screened out.

[0011] Furthermore, the correlation model module includes a defect correlation chain unit and a correlation model unit: Constructing a defect-related chain unit: The state representation vectors of each process are weighted and summed to calculate a comprehensive index. Within the same time window, based on the comprehensive index of each process, the correlation coefficient method is used to calculate the correlation between any two processes. A correlation threshold is preset, and adjacent processes that exceed the threshold are filtered out. A defect-related chain is constructed in combination with the process sequence. Establish a correlation model unit: Preset a comprehensive indicator threshold range for each process; designate processes exceeding the threshold range as abnormal processes; designate historical production records where two adjacent processes are both abnormal as historical abnormal records; collect the time interval between the previous and subsequent abnormal processes in the historical abnormal records, using this time interval as the abnormal propagation time; calculate the maximum deviation value of the comprehensive indicator corresponding to each process and set it as the process's fluctuation value; multiply the abnormal propagation time and fluctuation value to calculate the process's propagation attenuation coefficient; summarize the propagation attenuation coefficients of historical abnormal records to calculate the process's average propagation attenuation coefficient; using the fluctuation value, correlation degree, and abnormal propagation time of the previous process's comprehensive indicator as input, and the fluctuation value of the subsequent process's comprehensive indicator as output, establish a correlation model between processes based on a linear regression model.

[0012] Furthermore, the defect risk indicator module includes a unit for predicting volatility values ​​and a unit for calculating defect risk indicators: Predicting fluctuation value unit: During the execution of a process, the fluctuation value of the comprehensive index of the process and the estimated processing time are obtained and input into the correlation model between processes to predict the fluctuation value of the comprehensive index of the next process. Defect risk index calculation unit: Based on the fluctuation value, correlation degree, and average propagation attenuation coefficient of the comprehensive process index, the defect risk index of each process is calculated; When the defect risk index exceeds the preset risk threshold, a defect warning signal is issued, prompting the corresponding process to handle the abnormality.

[0013] Compared with the prior art, the beneficial effects of the present invention are: the present invention is based on Internet of Things technology, and combines process data feature vectors, state representation vectors and correlation models of multiple processes to realize intelligent monitoring of the whole process. By quantifying and analyzing the fluctuation characteristics and trend changes of process data, the production status and potential risks of each process can be grasped in real time, thereby improving the level of intelligent monitoring of the production process. By constructing a correlation model and defect-related chain between processes, it is possible to predict the defect risk of subsequent processes during production. By calculating the defect risk index, early warning signals can be issued in a timely manner, and potential risks can be identified before defects occur, thereby improving the response speed and processing efficiency of production. By analyzing historical production records, extracting process data features of each process, and combining various statistical features, fluctuation features, and trend features, an accurate process data feature vector is generated, which helps to identify the normal and abnormal states of the process and improves the accuracy and reliability of defect prediction. By combining the correlation between processes and constructing defect correlation chains to conduct joint analysis of multiple processes, rather than evaluating each process individually, it can effectively identify the propagation path of anomalies, reduce misjudgments that may be caused by evaluating a single process, and enhance the controllability of the overall production process. By comprehensively analyzing factors such as fluctuation values ​​and propagation attenuation coefficients of each process, the source of defects can be accurately identified. Combined with the abnormal propagation time, the problem can be quickly located, which can greatly shorten the response time for abnormal handling and reduce losses in the production process. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the IoT-based display screen production process monitoring system of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Please see Figure 1 This invention provides a technical solution: a method for monitoring the production process of a display screen based on the Internet of Things, the method comprising: Step S100: Obtain process data for multiple processes in the display panel manufacturing process. The process data includes process parameters, equipment operation data, and output detection data. Perform quantification processing to generate process data feature vectors for each process. Among them, the process data feature vectors generated in step S100 include: In historical production records, process data for each process is collected and categorized and summarized according to process parameters, equipment operation data, and output test data; Feature data is extracted for each type of process data. The feature data includes statistical features, trend features, and fluctuation features. The statistical features include mean, standard deviation, and variance. The trend features are the trends over time. The fluctuation features include fluctuation amplitude and frequency. The feature data is normalized to obtain normalized feature data, and the normalized feature data is vectorized to generate process data feature vectors for the process. For example, during the display screen manufacturing process, process data for each step is collected in real time, including process parameters, equipment operating data, and output test data. The specific collection methods are as follows: Process parameters: including values ​​such as temperature, pressure, humidity, and time, which are key factors affecting the process results; Equipment operation data: including equipment startup time, equipment operating status, equipment load, operator operation logs, etc. Output inspection data: including output quality data for each process, such as size, color consistency, brightness, etc. For each type of process data, feature data is extracted. The extracted features include: Statistical characteristics include: Mean: such as the average value of temperature or pressure over a specific period of time; Standard deviation: The degree of fluctuation of various process parameters, such as the stability of temperature changes; Variance: Measures the breadth of the distribution of data for a process, such as the extreme degree of pressure fluctuations during production; Trend characteristics: Analyze the trend of process data over time, such as the rising or falling trend of temperature in a certain process over time; Fluctuation characteristics include: Fluctuation range: such as the range of fluctuation of equipment operating parameters within a certain time range; Fluctuation frequency: describes the frequency of data fluctuations within a unit of time. For each eigenvalue F, normalization is performed using the maximum-minimum normalization formula: F′=(F-Fmin) / (Fmax-Fmin) where Fmax and Fmin are the maximum and minimum values ​​of the feature in the entire dataset, respectively; The normalized statistical features, trend features, and fluctuation features are vectorized to generate the process data feature vector for each process. The feature vector for each process is V = (A, B, C), where A represents the normalized result of the statistical features, B represents the normalized result of the fluctuation features, and C represents the normalized result of the trend features.

[0017] Step S200: Based on the process data feature vector, construct the state representation vector of the process; The state representation vector of the construction process in step S200 includes: The system filters the operational status stages of each process in the production process system, summarizes the historical production records of each operational status stage, obtains the process data feature vector corresponding to each historical production record, presets the threshold of each process data feature vector, calculates the deviation value of the process data feature vector and summarizes the absolute value, and calculates the average deviation value of the process data feature vector in each operational status stage. A threshold value for the absolute deviation of each process data feature vector is preset. Process data feature vectors that exceed the absolute deviation threshold value are set as abnormal vectors. The abnormal frequency of each process data feature vector being an abnormal vector in each operating state stage is calculated. The average deviation value and abnormal frequency of the process data feature vectors in different operating stages are summarized to calculate the total average deviation value and total abnormal frequency. The total average deviation value and total abnormal frequency are multiplied to calculate the deviation score of the process data feature vectors in each process. The process data feature vectors are sorted from high to low according to the deviation score. The difference in deviation score between two adjacent process data feature vectors is calculated to determine the position corresponding to the maximum deviation score difference. The number of process data feature vectors before the position is taken as the number of state characterization vectors, and the state characterization vectors are selected. For example, in the production process of a display screen, each process is divided into multiple operating stages, including: startup stage, stable production stage, load fluctuation stage, parameter adjustment stage, and pre-shutdown stage. Suppose that a certain process has a total of m historical production records in a certain operating state stage, and the corresponding process data feature vector is: V1, V2, ..., Vm; The baseline reference vector for each feature vector is V0=(A0, B0, C0), and the deviation value Di of the i-th feature vector is Di=|Ai-A0|+|Bi-B0|+|Ci-C0|. Within a certain operating state phase, calculate the average deviation value D=(1 / m)·∑Di, where: m is the number of historical records for that operating state phase; The preset absolute deviation threshold is T. If D>T exists, the corresponding process data feature vector is determined to be an anomaly vector. If the number of statistical anomaly vectors is k, then the anomaly frequency F of this operational phase is F = k / m; By combining the average deviation value and anomaly frequency of each operational stage, the deviation score of the process data feature vector is calculated. Where α and β are the weighting coefficients of the average deviation value and the abnormal frequency, and b represents the total number of operating state stages; The deviation scores of the process data feature vectors are sorted, and the adjacent difference Uh = Sh - Sh + 1 of the deviation scores is calculated to obtain the position of the maximum difference. For example, if S1 is 0.932, S2 is 0.895, S3 is 0.854, S4 is 0.424, S5 is 0.365, and S6 is 0.315, the calculated U1 is 0.037, U2 is 0.041, U3 is 0.430, U4 is 0.059, and U5 is 0.050. The maximum difference is 0.430, which corresponds to U3. Therefore, the number of state characterization vectors is 3.

[0018] Step S300: Based on the state representation vectors of each process and the degree of correlation between processes, construct a correlation model between processes; The process of constructing the relationship model between the processes in step S300 includes: The state representation vectors of each process are weighted and summed to obtain a comprehensive index. Within the same time window, based on the comprehensive index of each process, the correlation coefficient method is used to calculate the correlation between any two processes. A correlation threshold is preset, and adjacent processes that exceed the threshold are screened out. A defect correlation chain is constructed in combination with the process sequence. Preset the comprehensive index threshold range for each process, set the process that exceeds the comprehensive index threshold range as an abnormal process, and set the historical production record of two adjacent processes that are both abnormal processes as a historical abnormal record. Collect the time interval between the previous abnormal process and the next abnormal process in the historical abnormal records, and use the time interval as the abnormal propagation time. Calculate the maximum deviation value of the comprehensive index corresponding to each process and set it as the fluctuation value of the process. Multiply the abnormal propagation time and the fluctuation value to calculate the propagation attenuation coefficient of the process. Summarize the propagation attenuation coefficients of the historical abnormal records to calculate the average propagation attenuation coefficient of the process. Using the fluctuation value, correlation degree, and anomaly propagation time of the comprehensive index of the previous process as input and the fluctuation value of the comprehensive index of the next process as output, a correlation model between processes is established based on a linear regression model. For example, in the display screen manufacturing process, suppose there are several processes, such as process 1, process 2, process 3, and process 4. The state representation vector for each process is as follows: State representation vector of process 1: V1 = (A1, B1, C1); The state representation vector for process 2 is: V2 = (A2, B2, C2). State representation vector for process 3: V3 = (A3, B3, C3); State representation vector for process 4: V4 = (A4, B4, C4); Assuming the weighting coefficients w1, w2, and w3 correspond to the statistical characteristics, fluctuation characteristics, and trend characteristics, respectively, the comprehensive index of the j-th process is CIj = w1·Aj + w2·Bj + w3·Cj, where Aj, Bj, and Cj represent the normalized results of the statistical characteristics, fluctuation characteristics, and trend characteristics in the j-th process, respectively. Within the same time window, based on the comprehensive indicators of each process, the Pearson correlation coefficient can be used. Let the comprehensive indicators of process 1 and process 2 be CI1 and CI2, respectively. The formula for calculating the correlation coefficient is as follows: Where CI1' and CI2' are the average comprehensive indicators of process 1 and process 2, respectively. By setting a threshold pt for the degree of correlation, adjacent processes with a correlation degree higher than the threshold are selected. If the correlation coefficient p between process 1 and process 2 exceeds the threshold pt, there is a strong correlation, and a defect correlation chain is further constructed. For example, in the process of display screen production, if the correlation coefficient between process 1 and process 2 is 0.85 and the set threshold is 0.8, then these two processes will be considered to be related and constitute a defect correlation chain. Let the comprehensive index fluctuation values ​​of process 1 and process 2 be BV1 and BV2, respectively. The correlation model is expressed as BV2=r1·BV1+r2·p+r3·t, where r1, r2, and r3 represent the weights of the correlation model, and t represents the anomaly propagation time.

[0019] Step S400: During the execution of the process, the process with defects is identified and an early warning is issued by calculating the defect risk index; The calculation of defect risk indicators in step S400 includes: During the execution of the process, the fluctuation value of the comprehensive index of the process and the estimated processing time are obtained and input into the correlation model between processes to predict the fluctuation value of the comprehensive index of the next process. Based on the defect correlation chain and combined with the correlation model, the fluctuation value of the comprehensive index of each process in the defect correlation chain is predicted; Based on the fluctuation value, correlation degree, and average propagation attenuation coefficient of the comprehensive indicators of the process, the defect risk index of each process is calculated. When the defect risk index exceeds the preset risk threshold, a defect warning signal is issued, prompting the corresponding process to handle the abnormality. For example, by taking the comprehensive index fluctuation value BVj of the current process j and the estimated processing time tj as input, and based on the pre-established inter-process correlation model, the comprehensive index fluctuation value BVj+1 of the next process j+1 is predicted. For example, the defect risk index for each process is calculated as follows: Defect Risk Index DRIj = i1·BVj + i2·pj + i3·qj, where BVj is the comprehensive index fluctuation value, pj is the correlation degree, qj is the average propagation attenuation coefficient, and i1, i2, and i3 represent the weights of the comprehensive index fluctuation value, the correlation degree, and the average propagation attenuation coefficient.

[0020] To better implement the above methods, an IoT-based display production process monitoring system was also proposed. The system includes a process data feature vector module, a state representation vector module, an association model module, and a defect risk index module. Process data feature vector module: acquires process data of multiple processes in the manufacturing process of display panel, including process parameters, equipment operation data and output detection data, and performs quantification processing to generate process data feature vectors for each process; The process data feature vector module includes a feature data unit and a process data feature vector generation unit: Feature data unit: In historical production records, process data for each process is collected and classified and summarized according to process parameters, equipment operation data and output detection data; feature data is extracted for each type of process data, including statistical features, trend features and fluctuation features. The statistical features include mean, standard deviation and variance. The trend features are the trend of change over time. The fluctuation features include fluctuation amplitude and frequency. Generating process data feature vector units: Normalize the feature data to obtain normalized feature data, and then vectorize the normalized feature data to generate process data feature vectors for each process.

[0021] State representation vector module: Constructs a state representation vector for the process based on the feature vector of the process data; The state representation vector module includes a unit for calculating the anomaly frequency and a unit for filtering state representation vectors. The abnormal frequency calculation unit: filters the operating status stages of each process from the production process system, summarizes the historical production records of each operating status stage, obtains the process data feature vector corresponding to each historical production record, presets a threshold for each process data feature vector, calculates the deviation value of the process data feature vector and summarizes the absolute value, calculates the average deviation value of the process data feature vector in each operating status stage; presets an absolute deviation value threshold for each process data feature vector, sets process data feature vectors that exceed the absolute deviation value threshold as abnormal vectors, and calculates the abnormal frequency of each process data feature vector being an abnormal vector in each operating status stage. Screening State Representation Vector Units: The average deviation value and abnormal frequency of process data feature vectors in different operating state stages are summarized to calculate the total average deviation value and total abnormal frequency. The total average deviation value and total abnormal frequency are multiplied to calculate the deviation score of the process data feature vector in each process. The process data feature vectors are sorted from high to low according to the deviation score. The difference in deviation score between two adjacent process data feature vectors is calculated to determine the position corresponding to the maximum deviation score difference. The number of process data feature vectors before the position is taken as the number of state representation vectors, and the state representation vectors are screened out.

[0022] Association Model Module: Based on the state representation vectors of each process and the degree of correlation between processes, an association model between processes is constructed; The correlation model module includes a defect correlation chain construction unit and a correlation model establishment unit: Constructing a defect-related chain unit: The state representation vectors of each process are weighted and summed to calculate a comprehensive index. Within the same time window, based on the comprehensive index of each process, the correlation coefficient method is used to calculate the correlation between any two processes. A correlation threshold is preset, and adjacent processes that exceed the threshold are filtered out. A defect-related chain is constructed in combination with the process sequence. Establish a correlation model unit: Preset a comprehensive indicator threshold range for each process; designate processes exceeding the threshold range as abnormal processes; designate historical production records where two adjacent processes are both abnormal as historical abnormal records; collect the time interval between the previous and subsequent abnormal processes in the historical abnormal records, using this time interval as the abnormal propagation time; calculate the maximum deviation value of the comprehensive indicator corresponding to each process and set it as the process's fluctuation value; multiply the abnormal propagation time and fluctuation value to calculate the process's propagation attenuation coefficient; summarize the propagation attenuation coefficients of historical abnormal records to calculate the process's average propagation attenuation coefficient; using the fluctuation value, correlation degree, and abnormal propagation time of the previous process's comprehensive indicator as input, and the fluctuation value of the subsequent process's comprehensive indicator as output, establish a correlation model between processes based on a linear regression model.

[0023] Defect Risk Indicator Module: During the execution of a process, defect risk indicators are calculated to identify processes with defects and issue early warnings. The defect risk indicator module includes a unit for predicting volatility and a unit for calculating defect risk indicators. Predicting fluctuation value unit: During the execution of a process, the fluctuation value of the comprehensive index of the process and the estimated processing time are obtained and input into the correlation model between processes to predict the fluctuation value of the comprehensive index of the next process. Defect risk index calculation unit: Based on the fluctuation value, correlation degree, and average propagation attenuation coefficient of the comprehensive process index, the defect risk index of each process is calculated; When the defect risk index exceeds the preset risk threshold, a defect warning signal is issued, prompting the corresponding process to handle the abnormality.

[0024] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for monitoring the production process of a display screen based on the Internet of Things, characterized in that, The methods include: Step S100: Obtain process data for multiple processes in the display panel manufacturing process. The process data includes process parameters, equipment operation data, and output detection data. Perform quantification processing to generate process data feature vectors for each process. Step S200: Based on the process data feature vector, construct the state representation vector of the process; Step S300: Based on the state representation vectors of each process and the degree of correlation between processes, construct a correlation model between processes; Step S400: During the execution of the process, the defect risk index is calculated to identify the process with defects and issue an early warning.

2. The IoT-based display screen production process monitoring method according to claim 1, characterized in that, The step S100, which generates the process data feature vectors for each process step, includes the following steps: In historical production records, process data for each process is collected and categorized and summarized according to process parameters, equipment operation data, and output test data; Feature data is extracted for each type of process data. The feature data includes statistical features, trend features, and fluctuation features. The statistical features include mean, standard deviation, and variance. The trend features are the trends over time. The fluctuation features include fluctuation amplitude and frequency. The feature data is normalized to obtain normalized feature data, and then the normalized feature data is vectorized to generate the process data feature vector of the process.

3. The IoT-based display screen production process monitoring method according to claim 1, characterized in that, The step S200 of constructing the state representation vector of the process includes the following steps: The system filters the operational status stages of each process in the production process system, summarizes the historical production records of each operational status stage, obtains the process data feature vector corresponding to each historical production record, presets the threshold of each process data feature vector, calculates the deviation value of the process data feature vector and summarizes the absolute value, and calculates the average deviation value of the process data feature vector in each operational status stage. A threshold value for the absolute deviation of each process data feature vector is preset. Process data feature vectors that exceed the absolute deviation threshold value are set as abnormal vectors. The abnormal frequency of each process data feature vector being an abnormal vector in each operating state stage is calculated. The average deviation value and anomaly frequency of the process data feature vectors in different operating stages are summarized to calculate the total average deviation value and total anomaly frequency. The total average deviation value and total anomaly frequency are multiplied to calculate the deviation score of the process data feature vectors in each process. The process data feature vectors are sorted from high to low according to the deviation score. The difference in deviation score between two adjacent process data feature vectors is calculated to determine the position corresponding to the maximum deviation score difference. The number of process data feature vectors before the position is taken as the number of state characterization vectors, and the state characterization vectors are selected.

4. The IoT-based display screen production process monitoring method according to claim 1, characterized in that, The step S300 of constructing the relationship model between processes includes the following steps: The state representation vectors of each process are weighted and summed to obtain a comprehensive index. Within the same time window, based on the comprehensive index of each process, the correlation coefficient method is used to calculate the correlation between any two processes. A correlation threshold is preset, and adjacent processes that exceed the threshold are screened out. A defect correlation chain is constructed in combination with the process sequence. Preset the comprehensive index threshold range for each process, set the process that exceeds the comprehensive index threshold range as an abnormal process, and set the historical production record of two adjacent processes that are both abnormal processes as a historical abnormal record. Collect the time interval between the previous abnormal process and the next abnormal process in the historical abnormal records, and use the time interval as the abnormal propagation time. Calculate the maximum deviation value of the comprehensive index corresponding to each process and set it as the fluctuation value of the process. Multiply the abnormal propagation time and the fluctuation value to calculate the propagation attenuation coefficient of the process. Summarize the propagation attenuation coefficients of the historical abnormal records to calculate the average propagation attenuation coefficient of the process. Using the fluctuation value, correlation degree, and anomaly propagation time of the comprehensive index of the previous process as input and the fluctuation value of the comprehensive index of the next process as output, a correlation model between processes is established based on a linear regression model.

5. The IoT-based display screen production process monitoring method according to claim 1, characterized in that, The calculation of the defect risk index in step S400 includes the following steps: During the execution of the process, the fluctuation value of the comprehensive index of the process and the estimated processing time are obtained and input into the correlation model between processes to predict the fluctuation value of the comprehensive index of the next process. Based on the defect correlation chain and combined with the correlation model, the fluctuation value of the comprehensive index of each process in the defect correlation chain is predicted; Based on the fluctuation value, correlation degree, and average propagation attenuation coefficient of the comprehensive indicators of the process, the defect risk index of each process is calculated. When the defect risk index exceeds the preset risk threshold, a defect warning signal is issued, prompting the corresponding process to handle the abnormality.

6. An IoT-based display screen production process monitoring system, used to implement the IoT-based display screen production process monitoring method according to any one of claims 1-5, characterized in that, The system includes a process data feature vector module, a state characterization vector module, an association model module, and a defect risk index module. The process data feature vector module acquires process data for multiple processes during the manufacturing of the display panel. The process data includes process parameters, equipment operation data, and output detection data. It then performs quantification processing to generate process data feature vectors for each process. The state representation vector module: constructs a state representation vector for the process based on the process data feature vector; The association model module: Based on the state representation vectors of each process, and combined with the degree of correlation between processes, constructs an association model between processes; The defect risk index module: During the execution of a process, it calculates the defect risk index to identify processes with defects and issues early warnings.

7. The IoT-based display screen production process monitoring system according to claim 6, characterized in that, The process data feature vector module includes a feature data unit and a process data feature vector generation unit: The feature data unit: collects process data for each process from historical production records, and classifies and summarizes them according to process parameters, equipment operation data, and output detection data; Feature data is extracted for each type of process data. The feature data includes statistical features, trend features, and fluctuation features. The statistical features include mean, standard deviation, and variance. The trend features are the trends over time. The fluctuation features include fluctuation amplitude and frequency. The process data feature vector generation unit: performs normalization calculation on the feature data to obtain normalized feature data, and vectorizes the normalized feature data to generate the process data feature vector of the process.

8. The IoT-based display screen production process monitoring system according to claim 6, characterized in that, The state representation vector module includes a unit for calculating the anomaly frequency and a unit for filtering state representation vectors. The abnormal frequency calculation unit: filters the operating status stages of each process in the production process system, summarizes the historical production records of each operating status stage, obtains the process data feature vector corresponding to each historical production record, presets a threshold for each process data feature vector, calculates the deviation value of the process data feature vector and summarizes the absolute value, calculates the average deviation value of the process data feature vector in each operating status stage; presets an absolute deviation value threshold for each process data feature vector, sets process data feature vectors that exceed the absolute deviation value threshold as abnormal vectors, and calculates the abnormal frequency of each process data feature vector being an abnormal vector in each operating status stage. The state characterization vector screening unit: summarizes the average deviation value and abnormal frequency of process data feature vectors in different operating state stages, calculates the total average deviation value and total abnormal frequency, multiplies the total average deviation value and total abnormal frequency to calculate the deviation score of process data feature vectors in each process, sorts the process data feature vectors from high to low according to the deviation score, calculates the difference in deviation score between two adjacent process data feature vectors, determines the position corresponding to the maximum deviation score difference, and uses the number of process data feature vectors before the position as the number of state characterization vectors to screen out the state characterization vectors.

9. The IoT-based display screen production process monitoring system according to claim 6, characterized in that, The correlation model module includes a defect correlation chain construction unit and a correlation model establishment unit: The defect-related chain construction unit involves: weighted summation of the state representation vectors of each process to calculate a comprehensive index; within the same time window, based on the comprehensive index of each process, the correlation coefficient method is used to calculate the correlation between any two processes; a preset correlation threshold is set; adjacent processes exceeding the threshold are filtered out; and a defect-related chain is constructed in combination with the process sequence. The unit for establishing the correlation model includes: pre-setting a comprehensive indicator threshold range for each process; designating processes exceeding the comprehensive indicator threshold range as abnormal processes; designating historical production records where two adjacent processes are both abnormal processes as historical abnormal records; collecting the time interval between the previous and subsequent abnormal processes in the historical abnormal records, using this time interval as the abnormal propagation time; calculating the maximum deviation value of the comprehensive indicator corresponding to each process and setting it as the process's fluctuation value; multiplying the abnormal propagation time and fluctuation value to calculate the process's propagation attenuation coefficient; summing the propagation attenuation coefficients of historical abnormal records to calculate the process's average propagation attenuation coefficient; and using the fluctuation value, correlation degree, and abnormal propagation time of the comprehensive indicator of the previous process as input and the fluctuation value of the comprehensive indicator of the subsequent process as output, establishing a correlation model between processes based on a linear regression model.

10. The IoT-based display screen production process monitoring system according to claim 6, characterized in that, The defect risk indicator module includes a unit for predicting volatility and a unit for calculating defect risk indicators: The predicted fluctuation value unit: during the execution of the process, it acquires the fluctuation value of the comprehensive index of the process and the estimated processing time, and inputs them into the correlation model between processes to predict the fluctuation value of the comprehensive index of the next process. The defect risk index calculation unit calculates the defect risk index for each process based on the fluctuation value, correlation degree, and average propagation attenuation coefficient of the comprehensive process index. When the defect risk index exceeds the preset risk threshold, a defect warning signal is issued, prompting the corresponding process to handle the abnormality.

Images