Intermittent industrial computing engine-oriented method based on independent data space

Abstract

The invention relates to an intermittent industrial computing engine-oriented method based on an independent data space, and belongs to the field of industrial automation and process control, and the method comprises the following steps: S1, operating system environment configuration; s2, acquiring data; s3, updating the batch progress; s4, data preprocessing; s5, constructing a PCA model; s6, filling missing values; s7, performing nearest neighbor treatment; s8, performing anomaly detection; s9, data quality analysis and correction; s10, analyzing a pipeline and an envelope state; s11, updating an alarm state and ending judgment; and S12, exporting the data. The method has the advantages that the whole process is completed in a coordinated mode through a series of methods, and the steps can be executed regularly in a continuous operation mode, so that the defects in the prior art are overcome. Meanwhile, a brand-new framework is adopted, a single data space mode is used, a plurality of calculation engines can run in parallel at the same time and do not interfere with one another, and the calculation capacity of the calculation engines is greatly improved.

Description

Technical Field

[0001] The invention relates to a method for an intermittent industrial computing engine based on a separate data space, and belongs to the field of industrial automation and process control. Background Art

[0002] In the field of industrial automation and process control, the computing engine of the monitoring system plays a vital role. It not only needs to process a large amount of data in real time, but also needs to be able to accurately analyze this data in order to make timely decisions and ensure the efficiency and stability of the production process. However, most of the existing monitoring system computing engines are designed for continuous process models. These engines perform well when processing data from continuous production processes because continuous processes are usually more stable and predictable.

[0003] However, when it comes to intermittent processes, the limitations of existing computing engines become apparent. Intermittent processes, also known as batch processes, are characterized by the fact that production activities are not carried out continuously, but in a series of discrete batches. This mode of production is very common in industries such as pharmaceuticals, fine chemicals, and food processing. The complexity of intermittent process models is much higher than that of continuous processes because they involve more variables and parameters, which may change significantly at different stages of production. In addition, intermittent processes also have high requirements for real-time performance, because any delay may lead to a decrease in production efficiency and may even affect product quality.

[0004] The disadvantages of the existing technology are mainly manifested in the following aspects:

[0005] Poor adaptability: The computational engines of continuous process models often have difficulty adapting to the variability of intermittent processes. They may not be able to handle the nonlinear and dynamic changes in intermittent processes, resulting in the monitoring system being unable to accurately predict and respond to changes in the production process.

[0006] Insufficient real-time performance: Intermittent processes require computing engines to be able to quickly respond to changes in the production process. However, existing computing engines may experience delays when processing large amounts of data and cannot meet the needs of real-time monitoring.

[0007] High resource consumption: In order to handle the complexity of intermittent processes, existing computing engines may need to consume more computing resources, which not only increases costs but may also affect the stability of the system.

[0008] Lack of flexibility: Existing computing engines may have certain parameters and models fixed during design, which limits their adaptability to different production batches and conditions.

[0009] Therefore, in order to meet the needs of modern industrial production, it is necessary to develop a new generation of computing engines that should be able to better adapt to the complexity and real-time requirements of intermittent processes while having higher flexibility and integration capabilities. Summary of the invention

[0010] In order to overcome the defects of the prior art, the present invention provides a method for an intermittent industrial computing engine based on a separate data space. The technical solution of the present invention is:

[0011] A method for an intermittent industrial computing engine based on a separate data space comprises the following steps:

[0012] S1. Operating system environment configuration; S2. Acquire data; S3. Update batch progress; S4. Data preprocessing; S5. PCA model construction; S6. Missing value filling; S7. Nearest neighbor processing; S8. Anomaly detection; S9. Data quality analysis and correction; S10. Pipeline and envelope status analysis; S11. Alarm status update and end judgment; S12. Data export.

[0013] The step S1 is specifically as follows:

[0014] 1.1 Use Windows Server operating system;

[0015] 1.2 Configure the database, which includes MySQL5.7 database, Redis database and force control database;

[0016] The MySQL5.7 database is used to obtain intermittent process model information and intermittent batch configuration information created using intermittent process modeling software;

[0017] The Redis database is used to store the results generated by the engine calling the model calculation and provide them to the interface for real-time visual display;

[0018] The force control database is used to realize the reading of original data and the persistence of engine calculation results, and is used for reading and visualizing historical results.

[0019] The step S2 is specifically as follows:

[0020] 2.1 Obtain intermittent model configuration information from the MySQL database, including model parameter configuration, point weights, and data point names;

[0021] 2.2 Format the data point names to make them conform to the force control database name specification;

[0022] 2.3 Obtain real-time production data from the force control database.

[0023] The step S3 is specifically as follows:

[0024] 3.1 Check whether the batch status has entered the semi-online state, that is, the batch has started but the data is not yet stable;

[0025] 3.2 Initialize a new batch, create batch records and prepare data;

[0026] 3.3 Update the current batch and update the batch status according to the new data;

[0027] 3.4 Check whether the intermediate state is reached. If the current batch exceeds 30%-70% of the total steps, the batch reaches the intermediate state and the intermediate state processing flag is turned on;

[0028] 3.5 Check whether the residual calculation state is reached. If the current batch is greater than 80% of the preset total steps, turn on the residual calculation flag.

[0029] The step S4 is specifically as follows:

[0030] 4.1 Call the data preprocessor in the model to perform the same preprocessing on the data as in modeling, including data cleaning, format conversion and standardization; the formula for data standardization is:

[0031]

[0032] Among them: X: represents the original data feature value; μ: represents the mean of the feature, which is the average of all sample values ​​of the feature; σ: represents the standard deviation of the feature, which is the average of the degree to which all sample values ​​of the feature deviate from the mean;

[0033] 4.2 Create a warning information library for the current step, and save the current data warning information according to the point restrictions configured in the model and the warning dictionary.

[0034] The step S5 is specifically as follows:

[0035] 5.1 Check whether the batch status has entered the semi-online state;

[0036] 5.2 Perform PCA projection on the newly acquired preprocessed data and store the first three main components of the PCA projection: Y = X scaled V, where V is the matrix composed of the first k eigenvectors and Y is the projected data;

[0037] 5.3 Calculate the PCA T2 statistic to measure the degree of abnormality of the new data point in the principal component space; calculate the PCA conditional percentage based on the T2 statistic and ensure that it is not less than 0;

[0038] 5.4 Smooth the condition percentages through a first-order low-pass filter, and update the PCA condition status according to the filtered condition percentages;

[0039] 5.5 Determine the quadrant angle of the projection based on the results of the PCA projection and save the results.

[0040] The step S6 is specifically as follows:

[0041] 6.1 Check whether the batch status has entered the semi-online state and determine whether the data needs to be resampled according to the time series;

[0042] 6.2 If the data does not need to be resampled, the NIPALS algorithm is used for data filling to ensure the integrity and consistency of the data during batch processing;

[0043] The NIPALS algorithm is specifically:

[0044]

[0045] Where X represents the initial feature data;

[0046] The step S7 is specifically as follows:

[0047] 7.1 Check whether the batch status has entered the semi-online state;

[0048] 7.2 The closest normal and golden batch neighbors are found through the dynamic time warping algorithm to help predict and monitor quality in a semi-online state, ensuring the accuracy and stability of the batch processing process;

[0049] The dynamic time warping algorithm is specifically:

[0050]

[0051] The step S8 is specifically as follows:

[0052] 8.1 Check whether the batch status has entered the semi-online state;

[0053] 8.2 Call the model to scale and PCA project the original data;

[0054] 8.3 Calculating T via PCA Projection 2 and residual control limits SPE statistics and their contribution values, and determine T through filtering and smoothing methods 2 and the state of the residual control limit SPE; where T 2 % indicates relative to a certain limit value T 2 limit The percentage of T is used to evaluate the abnormality of the data point. 2 The statistical calculation formula is:

[0055]

[0056] The specific calculation formula of the residual control limit is:

[0057]

[0058] 8.4 Through T 2 And SPE status analysis can achieve real-time monitoring of the production process and detect abnormalities to ensure product quality and process stability.

[0059] The step S9 is specifically as follows:

[0060] 9.1 Check whether the batch status has entered the semi-online state;

[0061] 9.2 Call the model to predict the quality of the filling data;

[0062] 9.3 Use the residual model to predict each reference variable;

[0063] 9.4 Quality prediction based on nearest neighbor data;

[0064] The formula for calculating the nearest neighbor data in step 9.4 is as follows:

[0065]

[0066] x i and i : Respectively represent the values ​​of samples x and y on the i-th feature;

[0067] n: represents the number of features and the dimension of the sample;

[0068] 9.5 Take a weighted average of the quality prediction values ​​of the filling data and the neighbor data to obtain the final prediction quality value;

[0069] 9.6 Add a constant bias to the predicted quality value;

[0070] 9.7 The predicted quality value is corrected based on the feedback data and the feedback percentage in the configuration.

[0071] The step S10 is specifically as follows:

[0072] 10.1 Calculate the distance between the pipeline center and the current PCA projection point to find the nearest pipeline center;

[0073] 10.2 Calculate the pipeline condition percentage and smooth it through the filter;

[0074] 10.3 Set pipeline status according to conditional percentage;

[0075] 10.4 Calculate pipeline contribution and generate description information;

[0076] 10.5 Calculate the distance between the envelope center and the current PCA projection point and find the nearest envelope center;

[0077] 10.6 Calculation of T 2 value and its upper limit, and calculate the envelope condition percentage, smoothed by a filter;

[0078] 10.7 Set the envelope state according to the condition percentage;

[0079] 10.8 Calculate the envelope contribution and generate description information.

[0080] The step S11 is specifically as follows:

[0081] 11.1 Identify potential abnormalities by analyzing post-batch and intra-batch alarm status;

[0082] 11.2 Based on PCA projection and T 2 The value determines whether the batch process is nearing the end, ensuring the accuracy and stability of the process.

[0083] The step S12 is specifically as follows:

[0084] 12.1 Output the calculation results to the Redis database for real-time display and call by the model monitoring system.

[0085] 12.2 The calculation results are output to the force control database for reading and visualizing the historical results of the model monitoring system.

[0086] The advantages of the present invention are:

[0087] By reading raw data, performing data preprocessing, using models for principal component analysis (PCA), detecting anomalies, making quality predictions and corrections, and finally exporting the results to the force control database. The entire process is coordinated through a series of methods, and these steps can be executed regularly in a continuous operation mode to address the shortcomings of existing technologies. At the same time, a new architecture is adopted, using a separate data space method, which can run multiple computing engines in parallel without interfering with each other, greatly enhancing the computing power of the computing engine. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0089] The present invention will be further described below in conjunction with specific embodiments, and the advantages and features of the present invention will become clearer as the description proceeds. However, these embodiments are exemplary only and do not constitute any limitation to the scope of the present invention. It should be understood by those skilled in the art that the details and forms of the technical solution of the present invention may be modified or replaced without departing from the spirit and scope of the present invention, but these modifications and replacements all fall within the scope of protection of the present invention.

[0090] See also Figure 1 The present invention relates to a method for an intermittent industrial computing engine based on a separate data space, comprising the following steps:

[0091] S1. Operating system environment configuration; S2. Acquire data; S3. Update batch progress; S4. Data preprocessing; S5. PCA model construction; S6. Missing value filling; S7. Nearest neighbor processing; S8. Anomaly detection; S9. Data quality analysis and correction; S10. Pipeline and envelope status analysis; S11. Alarm status update and end judgment; S12. Data export.

[0092] In the above steps, based on the settings of the above solution, the following are achieved:

[0093] Comprehensive data processing: From operating system environment configuration S1 to data export S12, the entire process covers all stages of data processing, ensuring the integrity and systematicness of data processing.

[0094] Data quality assurance: Through data preprocessing S4, missing value filling S6, anomaly detection (S8) and data quality analysis and correction S9, the quality and accuracy of the data are ensured, providing a reliable data foundation for subsequent analysis.

[0095] PCA model construction: Constructing the principal component analysis (PCA) model in step S5 helps reduce data dimensions, extract key features, and improve data processing efficiency.

[0096] Anomaly detection and processing: Through nearest neighbor processing S7 and anomaly detection S8, outliers in the data can be effectively identified and processed to improve the reliability of the data.

[0097] Dynamic progress update: Updating the batch progress in the S3 step helps monitor the real-time status of data processing and facilitates management and scheduling.

[0098] Real-time feedback and adjustment: Through the alarm status update and end judgment S11, the system status can be fed back in real time, and adjustments and optimizations can be made in time.

[0099] Flexible data export: In step S12, data can be exported for further analysis or integration with other systems.

[0100] The step S1 is specifically as follows:

[0101] 1.1 Use Windows Server operating system;

[0102] 1.2 Configure the database, which includes MySQL5.7 database, Redis database and force control database;

[0103] The MySQL5.7 database is used to obtain intermittent process model information and intermittent batch configuration information created using intermittent process modeling software;

[0104] The Redis database is used to store the results generated by the engine calling the model calculation and provide them to the interface for real-time visual display;

[0105] The force control database is used to realize the reading of original data and the persistence of engine calculation results, and is used for reading and visualizing historical results.

[0106] The step S2 is specifically as follows:

[0107] 2.1 Obtain intermittent model configuration information from the MySQL database, including model parameter configuration, point weights, and data point names;

[0108] 2.2 Format the data point names to make them conform to the force control database name specification;

[0109] 2.3 Obtain real-time production data from the force control database.

[0110] The step S3 is specifically as follows:

[0111] 3.1 Check whether the batch status has entered the semi-online state, that is, the batch has started but the data is not yet stable;

[0112] 3.2 Initialize a new batch, create batch records and prepare data;

[0113] 3.3 Update the current batch and update the batch status according to the new data;

[0114] 3.4 Check whether the intermediate state is reached. If the current batch exceeds 30%-70% of the total steps, the batch reaches the intermediate state and the intermediate state processing flag is turned on.

[0115] 3.5 Check whether the residual calculation state is reached. If the current batch is greater than 80% of the preset total steps, turn on the residual calculation flag.

[0116] The step S4 is specifically as follows:

[0117] 4.1 Call the data preprocessor in the model to perform the same preprocessing on the data as in modeling, including data cleaning, format conversion and standardization; the formula for data standardization is:

[0118]

[0119] in:

[0120] X: represents the original data feature value.

[0121] μ: represents the mean of the feature, which is the average of all sample values ​​of the feature;

[0122] σ: represents the standard deviation of the feature, which is the average of the degree to which all sample values ​​of the feature deviate from the mean;

[0123] 4.2 Create a warning information library for the current step, and save the current data warning information according to the point restrictions configured in the model and the warning dictionary.

[0124] The step S5 is specifically as follows:

[0125] 5.1 Check whether the batch status has entered the semi-online state;

[0126] 5.2 Perform PCA projection on the newly acquired preprocessed data, and store the first three main components of the PCA projection: Y = Xscaled V, where V is the matrix composed of the first k eigenvectors and Y is the projected data;

[0127] 5.3 Calculate the PCA T2 statistic to measure the degree of abnormality of the new data point in the principal component space; calculate the PCA conditional percentage based on the T2 statistic and ensure that it is not less than 0;

[0128] 5.4 Smooth the condition percentages through a first-order low-pass filter, and update the PCA condition status according to the filtered condition percentages;

[0129] 5.5 Determine the quadrant angle of the projection based on the results of the PCA projection and save the results.

[0130] The step S6 is specifically as follows:

[0131] 6.1 Check whether the batch status has entered the semi-online state and determine whether the data needs to be resampled according to the time series;

[0132] 6.2 If the data does not need to be resampled, the NIPALS algorithm is used for data filling to ensure the integrity and consistency of the data during batch processing;

[0133] The NIPALS algorithm is specifically:

[0134]

[0135] Where X represents the initial feature data;

[0136] The step S7 is specifically as follows:

[0137] 7.1 Check whether the batch status has entered the semi-online state;

[0138] 7.2 The closest normal and golden batch neighbors are found through the dynamic time warping algorithm to help predict and monitor quality in a semi-online state, ensuring the accuracy and stability of the batch processing process;

[0139] The dynamic time warping algorithm is specifically:

[0140]

[0141] The step S8 is specifically as follows:

[0142] 8.1 Check whether the batch status has entered the semi-online state;

[0143] 8.2 Call the model to scale and PCA project the original data;

[0144] 8.3 Calculate T2 and residual control limit SPE statistics and their contribution values ​​through PCA projection, and determine the status of T2 and residual control limit SPE through filtering and smoothing methods; where T2% represents the percentage relative to a certain limit value T2limit, which is used to evaluate the abnormality of the data point; the T2 statistic calculation formula is:

[0145]

[0146] The specific calculation formula of the residual control limit is:

[0147]

[0148] 8.4 By analyzing the status of T2 and SPE, real-time monitoring of the production process and detection of abnormalities can be achieved to ensure product quality and process stability.

[0149] The step S9 is specifically as follows:

[0150] 9.1 Check whether the batch status has entered the semi-online state;

[0151] 9.2 Call the model to predict the quality of the filling data;

[0152] 9.3 Use the residual model to predict each reference variable;

[0153] 9.4 Quality prediction based on nearest neighbor data;

[0154] The formula for calculating the nearest neighbor data in step 9.4 is as follows:

[0155]

[0156] xi and yi: represent the values ​​of samples x and y on the i-th feature respectively;

[0157] n: represents the number of features and the dimension of the sample;

[0158] 9.5 Take a weighted average of the quality prediction values ​​of the filling data and the neighbor data to obtain the final prediction quality value;

[0159] 9.6 Add a constant bias to the predicted quality value;

[0160] 9.7 The predicted quality value is corrected based on the feedback data and the feedback percentage in the configuration.

[0161] The step S10 is specifically as follows:

[0162] 10.1 Calculate the distance between the pipeline center and the current PCA projection point to find the nearest pipeline center;

[0163] 10.2 Calculate the pipeline condition percentage and smooth it through the filter;

[0164] 10.3 Set pipeline status (normal, warning, alarm) according to condition percentage;

[0165] 10.4 Calculate pipeline contribution and generate description information;

[0166] 10.5 Calculate the distance between the envelope center and the current PCA projection point and find the nearest envelope center;

[0167] 10.6 Calculate the T2 value and its upper limit, and calculate the envelope condition percentage, and smooth it through the filter;

[0168] 10.7 Set the envelope status (normal, warning, alarm) according to the condition percentage;

[0169] 10.8 Calculate the envelope contribution and generate description information.

[0170] The step S11 is specifically as follows:

[0171] 11.1 Identify potential abnormalities by analyzing post-batch and intra-batch alarm status;

[0172] 11.2 Based on PCA projection and T2 value, determine whether the batch process is close to completion to ensure the accuracy and stability of the process.

[0173] The step S12 is specifically as follows:

[0174] 12.1 Output the calculation results to the Redis database for real-time display and call by the model monitoring system.

[0175] 12.2 The calculation results are output to the force control database for reading and visualizing the historical results of the model monitoring system.

[0176] In the scheme of the present invention, by the following:

[0177] Operating system environment configuration:

[0178] Use Windows Server operating system;

[0179] Configure three databases: MySQL 5.7 is used to store model and batch configuration information, Redis is used to store real-time calculation results, and the force control database is used to read raw data and persist results.

[0180] Data acquisition:

[0181] Get model configuration information from the MySQL database.

[0182] Format data point names to conform to force control database specifications.

[0183] Get real-time production data from the force control database.

[0184] Batch progress update:

[0185] Check batch status, initialize new batch, update batch status.

[0186] Determine whether the batch has reached the intermediate state or residual calculation state based on the percentage set by the process expert.

[0187] Data preprocessing:

[0188] The data were processed using the same preprocessing methods as those used for modeling.

[0189] Create a warning information library to save data warning information.

[0190] PCA model construction:

[0191] The data were projected with PCA and T2 statistics and conditional percentages were calculated.

[0192] The condition percentages were smoothed using a low-pass filter.

[0193] Missing value filling: Use NIPALS algorithm to fill in the missing values.

[0194] Nearest Neighbor Processing: Find the closest normal and golden batch neighbors through the dynamic time warping algorithm.

[0195] Anomaly Detection: Calculating T 2 and SPE statistics, determine the status through filtering and smoothing methods. Real-time monitoring and detection of anomalies.

[0196] Data quality analysis and correction:

[0197] Make quality predictions for infill data; use residual models for predictions; make quality predictions based on nearest neighbor data and perform weighted averaging.

[0198] Pipe and envelope condition analysis:

[0199] Calculate the distance between the pipe and the center of the envelope and set the status. Calculate the condition percentage and set warnings and alarms.

[0200] Alarm status update and end judgment: Analyze the alarm status and determine whether the batch process is nearing the end.

[0201] Data export: Output the calculation results to Redis and Likong database.

[0202] Advantages of the present invention:

[0203] Comprehensive data management: From data acquisition to export, the entire process covers all stages of data processing.

[0204] Real-time data processing: Real-time data display is achieved through the Redis database to improve response speed.

[0205] Data quality assurance: Ensure data accuracy and reliability through preprocessing, anomaly detection, and data quality analysis.

[0206] Flexible data processing: PCA and dynamic time warping algorithms are used to improve the flexibility and accuracy of data processing.

[0207] Efficient anomaly detection: Through T 2 and SPE statistics to monitor and detect anomalies in real time and improve the stability of the production process.

[0208] Persistence of historical data: The persistent storage of historical data is achieved through the power control database, which facilitates historical data analysis and backtracking.

[0209] Automation and intelligence: Automated data processing reduces manual intervention and improves efficiency and accuracy.

[0210] Scalability: The various steps of this method can be adjusted or expanded according to actual needs to adapt to different industrial scenarios.

[0211] Data visualization: Real-time display of data and visualization of historical results are achieved through a modeled monitoring system to facilitate monitoring and management.

[0212] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for intermittent industrial computing engines based on a separate data space, characterized in that: The following steps are involved: S1, operating system environment configuration; S2, data acquisition; S3, update batch progress; S4, data preprocessing; S5, PCA model construction; S6, missing value filling; S7, nearest neighbor processing; S8, anomaly detection; S9, data quality analysis and correction; S10, pipeline and envelope status analysis; S11, alarm status update and end judgment; S12, data export.

2. The method for intermittent industrial computing engine based on separate data space according to claim 1, characterized in that: The step S1 is specifically as follows: 1.1 Use Windows Server operating system; 1.2 Configure the database, which includes MySQL5.7 database, Redis database and force control database; The MySQL5.7 database is used to obtain intermittent process model information and intermittent batch configuration information created using intermittent process modeling software; The Redis database is used to store the results generated by the engine calling the model calculation and provide them to the interface for real-time visual display; The force control database is used to realize the reading of original data and the persistence of engine calculation results, and is used for reading and visualizing historical results. The step S2 is specifically as follows: 2.1 Obtain intermittent model configuration information from the MySQL database, including model parameter configuration, point weights, and data point names; 2.2 Format the data point names to make them conform to the force control database name specification; 2.3 Obtain real-time production data from the force control database.

3. The method for intermittent industrial computing engine based on separate data space according to claim 2, characterized in that the step S3 is specifically: 3.1 Check whether the batch status has entered the semi-online state, that is, the batch has started but the data is not yet stable; 3.2 Initialize a new batch, create batch records and prepare data; 3.3 Update the current batch and update the batch status according to the new data; 3.4 Check whether the intermediate state is reached. If the current batch exceeds 30%-70% of the total steps, the batch reaches the intermediate state and the intermediate state processing flag is turned on; 3.5 Check whether the residual calculation state is reached. If the current batch is greater than 80% of the preset total steps, turn on the residual calculation flag.

4. The method for intermittent industrial computing engine based on separate data space according to claim 3, characterized in that: The step S4 is specifically as follows: 4.1 Call the data preprocessor in the model to perform the same preprocessing on the data as in modeling, including data cleaning, format conversion and standardization; the formula for data standardization is: in: X: represents the original data feature value; μ: represents the mean of the feature, which is the average of all sample values ​​of the feature; σ: represents the standard deviation of the feature, which is the average of the degree to which all sample values ​​of the feature deviate from the mean; 4.2 Create a warning information library for the current step, and save the current data warning information according to the point restrictions configured in the model and the warning dictionary.

5. The method for intermittent industrial computing engine based on separate data space according to claim 4, characterized in that: The step S5 is specifically as follows: 5.1 Check whether the batch status has entered the semi-online state; 5.2 Perform PCA projection on the newly acquired preprocessed data and store the first three main components of the PCA projection: Y = X scaled V, where V is the matrix composed of the first k eigenvectors and Y is the projected data; 5.3 Calculating PCA T 2 Statistics are used to measure the abnormality of new data points in the principal component space; according to T 2 The statistic calculates the PCA conditional percentage and ensures that it is not less than 0; 5.4 Smooth the condition percentages through a first-order low-pass filter, and update the PCA condition status according to the filtered condition percentages; 5.5 Determine the quadrant angle of the projection based on the results of the PCA projection and save the results.

6. The method for intermittent industrial computing engine based on separate data space according to claim 5, characterized in that: The step S6 is specifically as follows: 6.1 Check whether the batch status has entered the semi-online state and determine whether the data needs to be resampled according to the time series; 6.2 If the data does not need to be resampled, the NIPALS algorithm is used for data filling to ensure the integrity and consistency of the data during batch processing; The NIPALS algorithm is specifically: in, X represents the initial feature data; The step S7 is specifically as follows: 7.1 Check whether the batch status has entered the semi-online state; 7.2 The closest normal and golden batch neighbors are found through the dynamic time warping algorithm to help predict and monitor quality in a semi-online state, ensuring the accuracy and stability of the batch processing process; The dynamic time warping algorithm is specifically: The step S8 is specifically as follows: 8.1 Check whether the batch status has entered the semi-online state; 8.2 Call the model to scale and PCA project the original data; 8.3 Calculating T via PCA Projection 2 and residual control limits SPE statistics and their contribution values, and determine T through filtering and smoothing methods 2 and the state of the residual control limit SPE; where T 2 % indicates relative to a certain limit value T 2 limit The percentage of T is used to evaluate the abnormality of the data point. 2 The statistical calculation formula is: The specific calculation formula of the residual control limit is: 8.4 Through T 2 And SPE status analysis can achieve real-time monitoring of the production process and detect abnormalities to ensure product quality and process stability.

7. The method for intermittent industrial computing engine based on separate data space according to claim 6, characterized in that: The step S9 is specifically as follows: 9.1 Check whether the batch status has entered the semi-online state; 9.2 Call the model to predict the quality of the filling data; 9.3 Use the residual model to predict each reference variable; 9.4 Quality prediction based on nearest neighbor data; The formula for calculating the nearest neighbor data in step 9.4 is as follows: x i and i : Respectively represent the values ​​of samples x and y on the i-th feature; n: represents the number of features and the dimension of the sample; 9.5 Take a weighted average of the quality prediction values ​​of the filling data and the neighbor data to obtain the final prediction quality value; 9.6 Add a constant bias to the predicted quality value; 9.7 The predicted quality value is corrected based on the feedback data and the feedback percentage in the configuration.

8. The method for intermittent industrial computing engine based on separate data space according to claim 7, characterized in that: The step S10 is specifically as follows: 10.1 Calculate the distance between the pipeline center and the current PCA projection point to find the nearest pipeline center; 10.2 Calculate the pipeline condition percentage and smooth it through the filter; 10.3 Set pipeline status according to conditional percentage; 10.4 Calculate pipeline contribution and generate description information; 10.5 Calculate the distance between the envelope center and the current PCA projection point and find the nearest envelope center; 10.6 Calculation of T 2 value and its upper limit, and calculate the envelope condition percentage, smoothed by a filter; 10.7 Set the envelope state according to the condition percentage; 10.8 Calculate the envelope contribution and generate description information.

9. The method for intermittent industrial computing engine based on separate data space according to claim 8, characterized in that: The step S11 is specifically as follows: 11.1 Identify potential abnormalities by analyzing post-batch and intra-batch alarm status; 11.2 Based on PCA projection and T 2 The value determines whether the batch process is nearing the end, ensuring the accuracy and stability of the process.

10. The method for intermittent industrial computing engine based on separate data space according to claim 8, characterized in that: The step S12 is specifically as follows: 12.1 Output the calculation results to the Redis database for real-time display and call by the model monitoring system. 12.2 The calculation results are output to the force control database for reading and visualizing the historical results of the model monitoring system.

Images