Multi-dimensional process data collaborative simulation control methods, systems and storage media

Abstract

This invention relates to the field of manufacturing process control technology, and discloses a multi-dimensional process data collaborative simulation control method, system, and storage medium. The method first collects a set of multi-dimensional process parameters from multiple process equipment in a manufacturing production line, performs cross-dimensional feature extraction, and generates a comprehensive feature matrix containing time-series features, spatial distribution features, and energy consumption features. Then, based on the feature correlations of different dimensions in the comprehensive feature matrix, a dynamic coupling model of process parameters is constructed to collaboratively simulate the interactions between process equipment, outputting a process state evolution sequence. Next, abnormal fluctuation features in the sequence are extracted to generate a set of process parameter adjustment instructions. Finally, based on the set of adjustment instructions and real-time process feedback data, the model simulation parameters are dynamically corrected, generating an optimized process control strategy and executing it. This method achieves collaborative analysis and dynamic control of multi-dimensional process parameters, improving the accuracy and adaptability of process control.

Description

Technical Field

[0001] This invention relates to the field of manufacturing process control technology, specifically to a multi-dimensional process data collaborative simulation control method, system, and storage medium. Background Technology

[0002] In modern manufacturing, the complexity of production line processes continues to rise, and the interaction between parameters of various process equipment is becoming increasingly close. Traditional process control methods are often limited to the monitoring and adjustment of single equipment or single-dimensional parameters, making it difficult to cope with the dynamic changes when multiple equipment work together. For example, in the semiconductor wafer manufacturing process, processes such as thin film deposition, photolithography, and etching involve dozens of pieces of equipment, and even slight fluctuations in parameters such as temperature, pressure, and gas flow rate can affect the final product quality through the coupling relationship between equipment.

[0003] In existing technologies, most process control models rely on static parameter association rules, neglecting the evolution of parameters over time and the differences in their spatial distribution. In a certain automotive welding production line using independent temperature closed-loop control, when current fluctuations at adjacent welding stations cause spatial heat conduction interference, the system fails to recognize this cross-equipment parameter association, leading to batch-to-batch variations in welding strength.

[0004] In energy consumption management, traditional methods typically treat energy consumption data as an independent indicator, failing to establish a correlation with the dynamic changes in process parameters. In a precision machining workshop, the CNC machine tool group exhibits a non-linear coupling relationship between cutting speed, feed rate, and spindle motor energy consumption. Due to the lack of multi-dimensional parameter synergistic analysis, the equipment often operates in a state of localized energy efficiency optimization but overall low process efficiency, resulting in energy waste and extended production cycles.

[0005] Existing process anomaly detection methods are mostly based on single-dimensional threshold judgments. When the anomaly manifests as a coordinated shift in multiple parameters, the system struggles to identify it effectively. For example, a lithium-ion battery electrode coating production line experienced anomalies in coating thickness due to minute coordinated fluctuations in coating speed, slurry viscosity, and drying temperature. Because each parameter was within its individually set normal range, the traditional monitoring system failed to issue a timely warning, ultimately leading to the scrapping of a large number of products. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-dimensional process data collaborative simulation control method, system, and storage medium to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a multi-dimensional process data collaborative simulation control method, the method comprising:

[0008] Collect a set of multidimensional process parameters from multiple process equipment in the manufacturing production line, perform cross-dimensional feature extraction on the set of multidimensional process parameters, and generate a comprehensive feature matrix that includes time series features, spatial distribution features, and energy consumption features.

[0009] Based on the feature correlation of different dimensions in the comprehensive feature matrix, a dynamic coupling model of process parameters is constructed. The interaction between process equipment is simulated through the dynamic coupling model of process parameters, and the process state evolution sequence is output.

[0010] Extract abnormal fluctuation features from the process state evolution sequence, and combine them with a preset process stability threshold to generate a set of process parameter adjustment instructions;

[0011] Based on the set of process parameter adjustment instructions and the real-time collected process feedback data, the simulation parameters of the dynamic coupling model of the process parameters are dynamically corrected, and an optimized process control strategy is generated and executed.

[0012] Preferably, the cross-dimensional feature extraction of the multidimensional process parameter set includes:

[0013] The temperature, pressure, and flow rate parameters in the multidimensional process parameter set are spatiotemporally aligned to obtain a synchronized process data stream.

[0014] Calculate the parameter gradient change of adjacent sampling points in the synchronized process data stream, and determine the dynamic density parameter based on the distribution range of the parameter gradient change.

[0015] When the dynamic density parameter exceeds the preset density threshold, identify the high-density interval in the synchronized process data stream and extract the thermodynamic feature vector within the high-density interval.

[0016] The comprehensive feature matrix is ​​generated by fusing the thermodynamic feature vector and the energy consumption feature data.

[0017] Preferably, calculating the parameter gradient change of adjacent sampling points in the synchronized process data stream includes:

[0018] Obtain the absolute values ​​of the temperature parameter difference, pressure parameter difference, and flow velocity parameter difference between adjacent sampling points;

[0019] Thermodynamic fluctuation index is calculated based on the ratio coefficients of the absolute value of the temperature parameter difference to the temperature dynamic threshold and the ratio coefficients of the absolute value of the pressure parameter difference to the pressure dynamic threshold.

[0020] Based on the offset of the absolute value of the flow velocity parameter difference from the flow velocity reference value, and in conjunction with the thermodynamic fluctuation index, a set of parameter gradient changes is generated.

[0021] Preferably, identifying high-density regions in the synchronized process data stream includes:

[0022] The synchronized process data stream is decomposed in the frequency domain to extract low-frequency vibration components and high-frequency noise components;

[0023] The frequency domain eigenvector is calculated based on the amplitude stability of the low-frequency vibration component and the energy attenuation rate of the high-frequency noise component.

[0024] When the frequency domain feature vector satisfies the preset frequency domain conditions, the data interval corresponding to the frequency domain feature vector is marked as a high-density interval.

[0025] Preferably, the simulation parameters of the dynamically coupled model for dynamically correcting the process parameters include:

[0026] The process feedback data is input into the Monte Carlo simulation engine to generate a probability distribution model of process parameters.

[0027] Based on the priority of the instructions in the process parameter adjustment instruction set, the probability distribution model of the process parameters is weighted and sampled.

[0028] The parameter constraint boundary of the dynamic coupling model of the process parameters is updated by the weighted sampling results to generate the corrected simulation parameters.

[0029] Preferably, the extraction of abnormal fluctuation features in the process state evolution sequence includes:

[0030] The process state evolution sequence is decomposed into long-term trend components and short-term fluctuation components.

[0031] Calculate the deviation between the short-term fluctuation component and the process stability threshold. When the deviation exceeds a preset deviation threshold, mark an abnormal timestamp.

[0032] By associating the device sensor data and process control logs corresponding to the abnormal timestamps, an abnormal fluctuation feature vector is generated.

[0033] Preferably, the generation and execution of the optimized process control strategy includes:

[0034] The process parameter adjustment instruction set is mapped to the equipment control protocol to generate an initial control instruction stream;

[0035] The timestamp sequence of the initial control command stream is adjusted based on the response delay parameter of the real-time process feedback data.

[0036] The optimized process control strategy is generated by integrating the timestamp sequence with the equipment status prediction data.

[0037] Preferably, after generating the set of process parameter adjustment instructions, it further includes:

[0038] When the transmission delay of process feedback data exceeds a preset delay threshold, a dynamic compensation mechanism is activated.

[0039] Based on the periodic patterns of historical process state evolution sequences, predictive process parameter sequences are generated.

[0040] The predictive process parameter sequence is weighted and fused with real-time process feedback data to generate compensated process data;

[0041] The compensated process data is input into the dynamic coupling model of the process parameters for real-time simulation calibration.

[0042] Preferably, the present invention further includes a multi-dimensional process data collaborative simulation control system for implementing the multi-dimensional process data collaborative simulation control method described above, the system comprising:

[0043] The parameter acquisition module is used to collect a set of multi-dimensional process parameters from multiple process equipment in the manufacturing production line.

[0044] The feature fusion module is used to perform cross-dimensional feature extraction on the multi-dimensional process parameter set to generate a comprehensive feature matrix.

[0045] The collaborative simulation module is used to construct a dynamic coupling model of process parameters based on the comprehensive feature matrix and output a process state evolution sequence.

[0046] The instruction generation module is used to extract abnormal fluctuation features in the process state evolution sequence and generate a set of process parameter adjustment instructions.

[0047] The strategy optimization module is used to dynamically modify the dynamic coupling model of the process parameters by combining process feedback data, and to generate and execute the optimized process control strategy.

[0048] Preferably, the present invention further includes a computer-readable storage medium, the storage medium including a computer program that, when executed by a processor, implements the multi-dimensional process data co-simulation control method as described above.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] By collecting multidimensional process parameter sets from multiple process devices and performing cross-dimensional feature extraction, a comprehensive feature matrix is ​​generated that includes time series features, spatial distribution features, and energy consumption features. This allows for the comprehensive capture of correlation information between parameters across different dimensions. This comprehensive feature matrix no longer treats each parameter as an isolated entity, but rather constructs a network of connections between parameters from multiple perspectives of time, space, and energy, thus revealing the correlation features that were originally hidden in single-dimensional data.

[0051] A dynamic coupling model of process parameters, constructed based on the feature correlations of different dimensions in the comprehensive feature matrix, enables collaborative simulation of the interactions between process equipment. This simulation does not simply mimic the operating state of a single piece of equipment, but rather reproduces the dynamic coupling process of multiple devices in actual production. It reflects the chain reaction of changes in parameters of one device on other devices through spatial transmission and temporal accumulation. By outputting the process state evolution sequence, it provides an intuitive dynamic perspective for understanding the process change patterns of the entire production line.

[0052] In terms of anomaly detection, this method extracts abnormal fluctuation features from the process state evolution sequence and generates adjustment instructions by combining them with a preset process stability threshold, overcoming the limitations of traditional single-parameter threshold judgment. This method can identify abnormal patterns formed by the coordinated shift of multiple parameters. Even if a single parameter is within the normal range, as long as its correlation with other parameters deviates from the stability threshold, the system can detect it in a timely manner and generate corresponding adjustment instructions, thereby discovering potential process anomalies earlier.

[0053] The simulation parameters of the model are dynamically corrected based on the set of adjustment instructions and real-time feedback data, generating and executing an optimized process control strategy, thus forming a closed-loop dynamic optimization mechanism. This mechanism enables the process control strategy to adaptively adjust as the actual operating status of the production line changes. When new equipment is introduced, raw materials are changed, or production specifications are altered, the model can gradually adapt to the new process environment through a continuous correction process, maintaining a sensitive response to process changes.

[0054] Through comprehensive analysis of multi-dimensional features, the adjustment of process parameters is no longer limited to a single parameter of a local device, but rather takes a global perspective of the entire production line to achieve synergistic optimization of the parameters of each device. Regarding energy consumption, by incorporating energy consumption characteristics and dynamic changes in process parameters into a unified model, a synergistic balance between energy consumption and process efficiency can be achieved while ensuring process quality, avoiding situations where local energy consumption is optimal but overall efficiency is low. Attached Figure Description

[0055] Figure 1 This is a schematic diagram illustrating the working principle of the multi-dimensional process data collaborative simulation control method described in this invention.

[0056] Figure 2A flowchart for cross-dimensional feature extraction;

[0057] Figure 3 A flowchart for calculating the gradient change of parameters;

[0058] Figure 4 Flowchart for high-density region identification;

[0059] Figure 5 A flowchart for optimizing and implementing process control strategies. Detailed Implementation

[0060] 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.

[0061] Please see Figure 1 This invention provides a multi-dimensional process data collaborative simulation control method, the method comprising:

[0062] A multi-dimensional set of process parameters is collected from multiple process equipment in the manufacturing production line. These parameters cover multiple dimensions such as temperature, pressure, flow rate, and energy consumption, and include operational data of different equipment at different time points. Cross-dimensional feature extraction is performed on the collected multi-dimensional set of process parameters. In this process, the temporal variation patterns, spatial distribution, and energy consumption characteristics of each parameter are comprehensively considered. After integrating and processing this information, a comprehensive feature matrix containing time series features, spatial distribution features, and energy consumption features is generated.

[0063] Based on the feature correlations of different dimensions in the comprehensive feature matrix, a dynamic coupling model of process parameters is constructed. This model can simulate the interaction and influence between different process parameters. Through this model, the interaction between process equipment is co-simulated, thereby outputting a process state evolution sequence, which reflects the changes of the process over time.

[0064] Abnormal fluctuation characteristics are extracted from the process state evolution sequence, and these characteristics are compared and analyzed with preset process stability thresholds. Based on the analysis results, a set of process parameter adjustment instructions is generated to regulate the process.

[0065] Based on the set of process parameter adjustment instructions and the real-time collected process feedback data, the simulation parameters of the dynamic coupling model of process parameters are dynamically corrected, so that the model can more accurately reflect the actual process situation, thereby generating and executing an optimized process control strategy to achieve precise control of the manufacturing line process.

[0066] Example 1: See Figure 2 When performing cross-dimensional feature extraction on a multi-dimensional set of process parameters, the first step is to perform spatiotemporal alignment processing on the temperature, pressure, and flow rate parameters within the multi-dimensional process parameter set. Parameter acquisition systems for different process equipment in a manufacturing line may have time synchronization errors; some equipment samples at 10 times per second, while others sample at 5 times per second. Furthermore, the spatial layout of the equipment on the production line is dispersed, and the physical regions corresponding to the parameters collected by sensors at different locations differ. Through a time calibration algorithm, the timestamps of all parameters are unified to the same reference clock. For data from equipment with lower sampling frequencies, interpolation is used to supplement parameter values ​​at intermediate moments, ensuring complete parameter records at the same time points. Simultaneously, a spatial coordinate system is established by combining the three-dimensional coordinate information of each piece of equipment on the production line, associating each parameter value with its corresponding spatial location label. This ensures that the parameters not only contain numerical information but also their spatial distribution attributes on the production line, thus forming a synchronized process data stream. Each record in this data stream contains a unified timestamp, spatial coordinates, and corresponding temperature, pressure, and flow rate parameter values.

[0067] When calculating the parameter gradient changes between adjacent sampling points in a synchronized process data stream, the temperature, pressure, and flow rate parameter values ​​of two adjacent sampling points are extracted sequentially based on the spatiotemporally aligned data stream. For the temperature parameter, the temperature difference between the subsequent sampling point and the previous sampling point is calculated, and its absolute value is taken. Similarly, the absolute values ​​of the pressure parameter difference and the flow rate parameter difference are calculated respectively. These absolute values ​​are combined with the physical characteristics of the parameters themselves to analyze the drasticness of parameter changes. For example, in the high-temperature process section, a small change in the temperature parameter may reflect significant process fluctuations, while in the low-temperature section, the same value change may be within the normal range. In this way, quantitative indicators that can reflect the magnitude of change of each parameter in adjacent time moments are obtained, and these indicators together constitute the basic data of parameter gradient changes.

[0068] Determining the dynamic density parameter based on the distribution range of parameter gradient changes requires statistical analysis of the parameter gradient changes at all adjacent sampling points. During the statistical analysis, the frequency of different change values ​​is recorded, a change distribution curve is plotted, and the peak interval and dispersion of the curve are analyzed. The value of the dynamic density parameter is related to the density of the change distribution; when a large number of parameter gradient changes are concentrated within a certain value range, the dynamic density parameter will increase accordingly, and vice versa. This parameter can dynamically reflect the density of process parameter changes, providing a basis for subsequent identification of key data intervals.

[0069] When the dynamic density parameter exceeds a preset density threshold, it indicates that the process parameters change relatively intensively within this period, potentially containing critical information affecting process stability. It is necessary to identify high-density intervals in the synchronized process data stream. A sliding window technique is used to scan the synchronized process data stream. The window size is set according to the response time of the process equipment, typically covering 10 to 30 sampling points. The dynamic density parameter within each window is calculated. When the dynamic density parameter within a window consistently exceeds a preset threshold, the corresponding time period is marked as a high-density interval. Consecutive high-density windows are merged into a single, complete high-density interval to ensure continuity and integrity, avoiding interval fragmentation caused by window division.

[0070] Extracting thermodynamic eigenvectors within a high-density region requires a comprehensive analysis of the overall trends and interrelationships of temperature and pressure parameters. This involves calculating the average, maximum, minimum, and rate of change of temperature within the region, and similarly calculating the aforementioned statistics for pressure parameters. Simultaneously, the correlation between temperature and pressure must be analyzed, such as the pressure change pattern as temperature increases, and the stability of this correlation within the region. These statistics and correlation indicators are then integrated into a multi-dimensional vector, the thermodynamic eigenvector, which comprehensively reflects the thermodynamic state and characteristics within the high-density region.

[0071] When fusing energy spectrum data with thermodynamic feature vectors and energy consumption characteristics, the energy consumption parameters are first analyzed spectrally. By performing a Fourier transform on the energy consumption data, the energy distribution at different frequencies is obtained, i.e., the energy spectrum data, which reflects the fluctuation frequency and intensity characteristics of energy consumption. Thermodynamic feature vectors and energy spectrum data are then matched in dimension; for vectors of different lengths, data expansion or compression is used to make their dimensions consistent. Subsequently, the two are combined through feature concatenation to form a comprehensive data structure containing both thermodynamic and energy consumption characteristics. This data structure is then standardized to eliminate the influence of differences in parameter magnitudes, ultimately generating a comprehensive feature matrix. Each row in this matrix corresponds to a comprehensive feature of a high-density interval, while the columns cover various dimensions of time series features, spatial distribution features, and energy consumption features, fully preserving the various feature information extracted across dimensions.

[0072] Example 2: See Figure 3When calculating the parameter gradient changes between adjacent sampling points in a synchronized process data stream, it is first necessary to analyze the synchronized process data stream point by point. The synchronized process data stream contains a continuous sequence of sampling points. Each sampling point records the temperature, pressure, and flow rate parameters at the corresponding timestamp, and these parameters have been aligned in both time and space. Adjacent sampling points refer to two consecutive sampling points in the time series, with their time interval determined by the data acquisition frequency, which may be on the order of milliseconds or seconds, depending on the monitoring accuracy requirements of the process equipment.

[0073] When obtaining the absolute value of the temperature parameter difference between adjacent sampling points, it is necessary to extract the temperature values ​​of the previous and subsequent sampling points, calculate the difference between them, and then eliminate the influence of positive and negative signs through absolute value calculation. Similarly, the same operation is performed on the pressure and flow rate parameters to obtain the absolute values ​​of the pressure and flow rate parameter differences. These absolute values ​​can intuitively reflect the magnitude of the changes in each parameter at adjacent times. Whether the parameter is rising or falling, the degree of change is reflected by the magnitude of the absolute value. In this process, it is necessary to ensure that the units of all parameters are consistent to avoid numerical comparison deviations caused by unit differences. For example, temperature parameters should be uniformly expressed in degrees Celsius, pressure parameters in Pascals, and flow rate parameters in cubic meters per second.

[0074] The setting of dynamic temperature and pressure thresholds needs to be considered in conjunction with the specific process type and equipment characteristics. Different manufacturing processes have different sensitivities to temperature and pressure. For example, the dynamic temperature threshold for high-temperature smelting processes may be set to a larger value, while the dynamic temperature threshold for precision electronics manufacturing processes may be smaller. These thresholds are usually determined based on a comprehensive analysis of the equipment's design parameters, the standard operating range of the process, and the normal fluctuation range in historical operating data, and are dynamically adjusted according to different process stages. For example, during the equipment startup phase, when temperature and pressure are changing rapidly, the thresholds can be appropriately relaxed; while during the stable operation phase, the thresholds are set more strictly.

[0075] When calculating the proportionality coefficient between the absolute value of the temperature parameter difference and the temperature dynamic threshold, the absolute value of the temperature parameter difference is divided by the temperature dynamic threshold. The result is the proportionality coefficient, and its magnitude reflects the proportion of the actual temperature change relative to the allowable fluctuation range. Similarly, the proportionality coefficient between the absolute value of the pressure parameter difference and the pressure dynamic threshold is calculated using the same method. The generation of the thermodynamic fluctuation index requires comprehensive consideration of these two proportionality coefficients, achieved through weighted integration. The weight allocation is determined based on the degree of influence of temperature and pressure in a specific process. If temperature has a greater impact on process stability, the temperature proportionality coefficient has a higher weight; conversely, the pressure proportionality coefficient has a higher weight. The integrated thermodynamic fluctuation index can comprehensively reflect the overall fluctuation of temperature and pressure parameters.

[0076] The flow rate reference value is the standard flow rate setting value during normal process operation. This value is predetermined based on process requirements, equipment performance, and product specifications. For example, in liquid transfer processes, the flow rate reference value may be determined based on pipe diameter and transfer efficiency. The deviation of the absolute value of the flow rate parameter difference from the flow rate reference value is obtained by dividing the absolute value of the flow rate parameter difference by the flow rate reference value. This deviation reflects the proportion of the actual flow rate change deviating from the standard value and can reflect the relative severity of flow rate fluctuations.

[0077] When comprehensively analyzing flow velocity offsets and thermodynamic fluctuation indices, the correlation between the two must be considered. In most processes, changes in temperature and pressure affect flow velocity, and conversely, changes in flow velocity can lead to fluctuations in temperature and pressure. Therefore, when generating the set of parameter gradient changes, temperature, pressure, and flow velocity gradient changes should be included as independent elements in the set, while the comprehensive correlation result between the thermodynamic fluctuation indices and flow velocity offsets should also be included as an important element. The comprehensive correlation result can be generated by numerically combining the two, such as superimposing or multiplying the thermodynamic fluctuation indices and flow velocity offsets, depending on the interaction between the two in the process. The final set of parameter gradient changes contains multi-dimensional parameter change information, comprehensively reflecting the gradient change characteristics of process parameters between adjacent sampling points.

[0078] Example 3: See Figure 4 When identifying high-density intervals in a synchronized process data stream, frequency domain decomposition is required. A synchronized process data stream is a continuous sequence of parameters after spatiotemporal alignment, containing signals such as temperature, pressure, and flow rate that change over time. Frequency domain decomposition is achieved through Fourier transform, converting the signal in the time domain into a spectral distribution in the frequency domain, thereby separating different frequency components. Specifically, each parameter sequence in the data stream is processed segment by segment, with each segment containing a preset number of sampling points. The segment length is typically determined based on the process cycle to ensure that each segment reflects a complete local change characteristic. After the Fourier transform, the signal is decomposed into multiple frequency components. Low-frequency vibration components correspond to slowly changing parts of the signal, usually related to the stable operating state of the process equipment, such as parameter fluctuations when the equipment is running at a constant speed. High-frequency noise components correspond to rapidly changing parts, often caused by random factors such as equipment vibration and environmental interference.

[0079] After extracting the low-frequency vibration component and the high-frequency noise component, the amplitude stability of the low-frequency vibration component needs to be calculated. Amplitude stability is described by the range of amplitude variation of the low-frequency component within a set time window. The difference between the maximum and minimum amplitude values ​​within this window is calculated, and then divided by the average value. The smaller the result, the more stable the low-frequency vibration. For the high-frequency noise component, its energy decay rate is calculated, that is, the rate at which high-frequency energy decreases as the frequency increases. The energy distribution of high-frequency noise typically decreases with increasing frequency. Two adjacent frequency points within the high-frequency band are selected, and the ratio of the energy difference between these two points to the frequency difference is calculated to characterize the rate of energy decay.

[0080] A frequency domain feature vector is constructed based on the amplitude stability of the low-frequency vibration component and the energy attenuation rate of the high-frequency noise component. This vector contains two main elements: the quantized value of the low-frequency amplitude stability and the quantized value of the high-frequency energy attenuation rate. In addition, auxiliary features such as the dominant frequency of the low-frequency component and the total energy of the high-frequency component can be incorporated to make the vector more comprehensively reflect the frequency domain characteristics. Preset frequency domain conditions are determined according to the process type. For example, for precision machining processes, the quantized value of the low-frequency amplitude stability is required to be less than a certain value, while the high-frequency energy attenuation rate is required to be greater than a certain value, to ensure that stable components dominate the signal and noise decays rapidly. When all elements of the frequency domain feature vector meet the preset conditions, the corresponding time period is marked as a high-density interval.

[0081] When dynamically adjusting the simulation parameters of the dynamic coupling model of process parameters, the process feedback data is first input into the Monte Carlo simulation engine. This feedback data includes the actual parameter values ​​after the equipment executes adjustment commands, product quality inspection data, etc., reflecting the actual effect of the control commands. The Monte Carlo simulation engine simulates the process results under different combinations of process parameters through extensive random sampling, with each sample generated based on the probability distribution in the feedback data. After a sufficient number of samplings, a probability distribution model of the process parameters is formed. This model describes the probability of each parameter occurring within different value ranges, such as the probability of temperature parameters within a certain interval, and the joint probability distribution of pressure and flow rate parameters.

[0082] Each instruction in the process parameter adjustment instruction set has a corresponding priority, determined by its impact on process stability; the greater the impact, the higher the priority. When performing weighted sampling on the process parameter probability distribution model, weights are assigned to different parameters based on instruction priority. Parameters involved in instructions with higher priority have a correspondingly higher probability of being selected during sampling. For example, if the temperature adjustment instruction has a higher priority than the pressure adjustment instruction, the temperature parameter will have a greater weight during sampling, and its sampling results will be more likely to be biased towards the range suggested by the instruction.

[0083] The weighted sampling results generate a series of parameter combinations that meet the adjustment requirements. These combinations are then used to update the parameter constraint boundaries of the dynamically coupled process parameter model. The parameter constraint boundaries are the value ranges of each parameter in the model. Originally set based on equipment design parameters and historical data, they can be dynamically adjusted using the sampling results. For example, the allowable fluctuation range of a parameter can be expanded or reduced, or the correlation coefficients between parameters can be adjusted. The corrected simulation parameters are generated based on the updated constraint boundaries, making the model more closely match the actual process adjustment direction during simulation, thereby improving simulation accuracy.

[0084] The formula used to calculate the energy attenuation rate of high-frequency noise components is:

[0085]

[0086] in, This represents the energy attenuation rate of the high-frequency noise component. Indicates frequency points within the high-frequency band The energy value at that location, Indicates frequency points within the high-frequency band The energy value at that location, and , and Let be any two adjacent frequency points in the high-frequency band. This formula quantifies the rate attenuation of high-frequency noise energy as the frequency increases by calculating the energy change within a unit frequency interval.

[0087] Example 4: See Figure 5 When extracting abnormal fluctuation features from the process state evolution sequence, the sequence is first decomposed into multiple scales. The process state evolution sequence is continuous state data output from co-simulation, containing the trajectories of multiple process parameters over time, covering the entire stage from equipment startup to stable operation. Multi-scale decomposition employs wavelet transform, selecting appropriate wavelet basis functions to decompose the sequence into components at different time scales. During decomposition, the number of decomposition levels is set according to the length of the process cycle, typically 3-5 levels, each corresponding to a different time resolution. Long-term trend components are extracted from the decomposition results, mainly reflecting the overall direction of change in the process state over a longer period, such as the slow upward or downward trend of temperature parameters with increasing production batches. Short-term fluctuation components contain the high-frequency components decomposed from each level, reflecting subtle oscillations in the process state over a short period, such as pressure parameter fluctuations caused by instantaneous equipment vibration.

[0088] After obtaining the long-term trend component and the short-term fluctuation component, the deviation of the short-term fluctuation component from the process stability threshold is calculated. The process stability threshold is the allowable fluctuation range determined based on process standards and equipment performance. Different parameters have different thresholds; for example, the threshold for flow rate might be set at ±5% of the baseline value, and the threshold for temperature might be set at ±2℃. The deviation calculation requires first removing the portion of the short-term fluctuation component that overlaps with the long-term trend, retaining only the pure fluctuation component, and then comparing this component with the threshold. The calculation uses the following formula:

[0089]

[0090] in, Indicates the degree of deviation. Indicating the first component of short-term fluctuations Fluctuation value at each sampling point This indicates the number of sampling points within the calculation window. This represents the process stability threshold. The formula quantifies the degree of deviation of short-term fluctuations from the threshold by calculating the average fluctuation amplitude at a given unit threshold.

[0091] When the deviation exceeds a preset deviation threshold, the corresponding abnormal timestamp is marked. The preset deviation threshold is set according to the sensitivity of the process to fluctuations. For high-precision manufacturing processes, this threshold may be set to 1.2, meaning that when the average fluctuation amplitude reaches 1.2 times the threshold, it is judged as abnormal; for rough processes, the threshold may be relaxed to 1.5. The abnormal timestamp is accurate to the sampling time, and the corresponding equipment number and parameter type are also recorded to facilitate subsequent identification of the abnormal source. When associating the equipment sensor data corresponding to the abnormal timestamp with the process control log, it is necessary to retrieve the original sensor data for a period of time before and after the timestamp, including auxiliary parameters not involved in the simulation, such as equipment vibration frequency and ambient humidity. At the same time, the operation records within this period in the process control log, such as valve opening adjustments and power setting changes, are extracted. By comparing the temporal correlation between sensor data and operation records, it is identified whether the abnormal fluctuation is caused by operation adjustments or by equipment failure. This correlation information is converted into quantitative indicators, such as the number of operations before the abnormality occurred and the magnitude of the sudden change in sensor data, and integrated into an abnormal fluctuation feature vector.

[0092] When generating and executing the optimized process control strategy, the process parameter adjustment instruction set is first mapped to the equipment control protocol. This instruction set contains multiple adjustment instructions for different equipment, each specifying the parameter type, target value, and adjustment range. Equipment control protocols vary by manufacturer and model, requiring the instructions to be converted to the protocol's defined format. For example, temperature adjustment instructions are converted to Modbus protocol register write commands, and pressure adjustment instructions are converted to OPCUA protocol method calls. During the conversion process, the validity of the instructions must be verified to ensure that the target value is within the equipment's allowed physical range; for instance, temperature instructions must not exceed the equipment's maximum heating temperature to avoid invalid instructions or equipment damage.

[0093] Based on the response delay parameters from real-time process feedback data, the timestamp sequence of the initial control command flow is adjusted. The response delay parameters are obtained by statistically analyzing the transmission times of historical feedback data, including the transmission delay from equipment sensors to the control system and the execution delay of control commands from the system to the equipment, typically measured in milliseconds. During adjustment, the corresponding delay parameter is added to the original timestamp of each command to match the actual execution time of the command with the rhythm of process state changes. For example, if the response delay of a certain piece of equipment is 500ms, the execution time of the command is postponed by 500ms to avoid parameter overshoot caused by premature command execution.

[0094] When fusing timestamp sequences and equipment status prediction data, the equipment status prediction data is generated through a time-series prediction model. Based on the equipment's historical operating data and current status, it predicts parameter change trends over a future period, such as the temperature rise curve and pressure stability value within the next 10 minutes. During the fusion process, each execution moment in the timestamp sequence is matched with the corresponding moment in the prediction data. If the prediction data shows that the equipment parameter will approach a threshold at a certain moment, the command intensity at that moment is adjusted in advance, such as increasing the adjustment range of the cooling water flow rate. The final optimized process control strategy includes the adjusted timestamp, target parameter value, execution equipment number, and emergency termination conditions. It is issued to each piece of equipment for execution through the control system, and execution feedback is received in real time to monitor the strategy's effectiveness.

[0095] Example 5: After generating the set of process parameter adjustment instructions, continuous monitoring of the process feedback data transmission process is required. Process feedback data is collected in real-time by sensors distributed across various nodes of the manufacturing line, including equipment operating parameters, material status data, and environmental monitoring information. This data is transmitted to the control system via an industrial bus or wireless network. Transmission delay refers to the time interval from data acquisition to its reception and parsing by the control system. It is affected by network bandwidth, data volume, and transmission path complexity and may fluctuate. During monitoring, timestamps are embedded in the data frames, and the difference between the acquisition timestamp and the reception timestamp is compared to calculate the transmission delay of each data frame in real time. A preset delay threshold is set according to the real-time requirements of process control. For example, in a fast-response precision assembly process, the threshold may be set to 50 milliseconds, while in a smelting process with a longer reaction cycle, the threshold can be relaxed to 500 milliseconds.

[0096] When a transmission delay exceeds a preset delay threshold, a dynamic compensation mechanism is triggered. This mechanism first marks the delayed data, distinguishing between invalid data whose real-time performance has been lost and valid delayed data that is still partially usable. Simultaneously, direct control operations based on the delayed data are suspended to prevent control inaccuracies caused by data lag. The core of the dynamic compensation mechanism lies in using historical data to predict and fill information gaps during delay periods, ensuring the continuity of process control.

[0097] When generating predictive process parameter sequences based on the periodic patterns of historical process state evolution sequences, in-depth analysis of historical data is necessary. Historical process state evolution sequences store complete records of parameter changes across multiple past production cycles, covering various scenarios such as normal operation, anomaly handling, and shutdown / restart. Time series analysis methods are used to identify recurring patterns in the sequences, such as parameter fluctuation cycles at fixed times each day and stage characteristics during the processing of each batch of materials. Extracting periodic patterns requires eliminating the interference of random outliers. A sliding window is used to statistically analyze the mean and variance of parameters for each time period to determine stable cycle lengths and trends. Based on these patterns, time series prediction methods are used to generate parameter sequences for a future period. The prediction duration matches the currently detected transmission delay; for example, if the delay is 200 milliseconds, a predictive process parameter sequence for the next 200 milliseconds is generated. During the prediction process, the historical patterns need to be adaptively adjusted based on the real-time status of the current process. For example, if there are differences between the current batch of materials and historical batches, the prediction results need to be corrected based on material characteristic parameters.

[0098] When weighted fusion of predictive process parameter sequences and real-time process feedback data, the reliability of both must be evaluated separately. The weight of the predictive sequence changes dynamically over time, with a higher weight initially and gradually decreasing over time as prediction errors accumulate. Even with delays, the accuracy of real-time process feedback data at the time of acquisition is still higher than that of the predictive data; therefore, weights are assigned based on the delay duration, with shorter delays resulting in higher weights. The fusion process uses point-by-point weighted calculation, where the compensated data value at each time point is the weighted sum of the predicted and feedback values ​​at that moment. For example, if the predicted value at a certain moment is A, the delayed feedback value is B, the prediction weight is 0.3, and the feedback weight is 0.7, then the compensated data value is 0.3A + 0.7B. For future periods not yet covered by the feedback data, only the values ​​from the predictive sequence are used as compensation data.

[0099] The compensated process data is input into the dynamic coupling model of process parameters for real-time simulation calibration, requiring re-initialization of the model's input layer parameters. The model adjusts the correlation between its internal parameters based on the compensated data, such as correcting the coupling coefficient between temperature and pressure and updating the parameter values ​​of the energy consumption model. During simulation calibration, the model continuously outputs process state predictions based on the compensated data and compares them with subsequently received real-time feedback data, constantly correcting prediction deviations. In this way, even with data transmission delays, the model can still accurately track the process state, ensuring that the generated process control strategy meets actual operational requirements. The calibrated model parameters are temporarily stored, and once the transmission delay returns to normal, the system automatically switches back to parameter configuration based on real-time data, ensuring stable operation under different network conditions.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for multi-dimensional process data collaborative simulation control, characterized in that, The method comprises the following steps: Collecting a set of multi-dimensional process parameters of a plurality of process equipment in a manufacturing production line, performing cross-dimension feature extraction on the set of multi-dimensional process parameters to generate a comprehensive feature matrix comprising time series features, spatial distribution features and energy consumption features, which comprises: Performing spatio-temporal alignment processing on temperature parameters, pressure parameters and flow rate parameters in the set of multi-dimensional process parameters to obtain a synchronized process data stream; calculating the parameter gradient change amount of adjacent sampling points in the synchronized process data stream, and determining a dynamic density parameter according to the distribution range of the parameter gradient change amount; the value of the dynamic density parameter is related to the distribution density of the parameter gradient change amount; when more parameter gradient change amounts are concentrated in a certain value range, the dynamic density parameter will increase accordingly, and vice versa; the dynamic density parameter dynamically reflects the density of the process parameter change, and provides a basis for identifying the key data interval; when the dynamic density parameter exceeds a preset density threshold, a high-density interval in the synchronized process data stream is identified, and a thermodynamic feature vector in the high-density interval is extracted; the thermodynamic feature vector and the energy spectrum data of the energy consumption feature are fused to generate the comprehensive feature matrix; According to the correlation of features in different dimensions in the comprehensive feature matrix, a process parameter dynamic coupling model is constructed, the interaction between process equipment is simulated through the process parameter dynamic coupling model, and a process state evolution sequence is output; Extracting abnormal fluctuation features in the process state evolution sequence, combining a preset process stability threshold to generate a set of process parameter adjustment instructions; According to the process parameter adjustment instruction set and the real-time collected process feedback data, the simulation parameters of the process parameter dynamic coupling model are dynamically corrected, an optimized process control strategy is generated and executed; The calculation of the parameter gradient change amount of adjacent sampling points in the synchronized process data stream comprises: Obtaining the temperature parameter difference absolute value, the pressure parameter difference absolute value and the flow rate parameter difference absolute value of adjacent sampling points; According to the proportion coefficient of the temperature parameter difference absolute value and the temperature dynamic threshold, and the proportion coefficient of the pressure parameter difference absolute value and the pressure dynamic threshold, a thermodynamic fluctuation index is calculated; According to the offset of the flow rate parameter difference absolute value and the flow rate reference value, and combining the thermodynamic fluctuation index, a set of parameter gradient change amounts is generated; The identification of the high-density interval in the synchronized process data stream comprises: Performing frequency domain decomposition on the synchronized process data stream to extract low-frequency vibration components and high-frequency noise components; According to the amplitude stability of the low-frequency vibration component and the energy attenuation rate of the high-frequency noise component, a frequency domain feature vector is calculated; When the frequency domain feature vector meets a preset frequency domain condition, the data interval corresponding to the frequency domain feature vector is marked as a high-density interval; The extraction of abnormal fluctuation features in the process state evolution sequence comprises: Performing multi-scale decomposition on the process state evolution sequence to obtain long-term trend components and short-term fluctuation components; Calculating the deviation degree of the short-term fluctuation component and the process stability threshold; when the deviation degree exceeds a preset deviation threshold, an abnormal timestamp is marked. Correlate the device sensor data corresponding to the abnormal timestamp with the process control log to generate an abnormal fluctuation feature vector.

2. The multi-dimensional process data co-simulation control method of claim 1, wherein, The dynamic correction of the simulation parameters of the process parameter dynamic coupling model includes: Input the process feedback data into a Monte Carlo simulation engine to generate a process parameter probability distribution model; According to the instruction priority in the process parameter adjustment instruction set, the process parameter probability distribution model is weighted sampled; The parameter constraint boundary of the process parameter dynamic coupling model is updated through the weighted sampling result to generate a corrected simulation parameter.

3. The multi-dimensional process data co-simulation control method of claim 1, wherein, The generation of the optimized process control strategy and the execution include: Map the process parameter adjustment instruction set to the device control protocol to generate an initial control instruction stream; According to the response delay parameter of the real-time process feedback data, the timestamp sequence of the initial control instruction stream is adjusted; Fuse the timestamp sequence with the device state prediction data to generate the optimized process control strategy.

4. The multi-dimensional process data co-simulation control method of claim 1, wherein, The generation of the process parameter adjustment instruction set further includes: When it is detected that the transmission delay of the process feedback data exceeds the preset delay threshold, a dynamic compensation mechanism is started; According to the periodicity of the historical process state evolution sequence, a predictive process parameter sequence is generated; The predictive process parameter sequence is weighted fused with the real-time process feedback data to generate compensated process data; The compensated process data is input into the process parameter dynamic coupling model for real-time simulation calibration.

5. A multi-dimensional process data collaborative simulation control system for implementing the multi-dimensional process data collaborative simulation control method according to any one of claims 1-4, characterized in that, It includes: The parameter acquisition module is used for acquiring a multi-dimensional process parameter set of multiple process equipment in a manufacturing line; The feature fusion module is used for cross-dimension feature extraction on the multi-dimensional process parameter set to generate a comprehensive feature matrix; The cooperative simulation module is used for constructing a process parameter dynamic coupling model according to the comprehensive feature matrix, and outputting a process state evolution sequence; The instruction generation module is used for extracting abnormal fluctuation features in the process state evolution sequence to generate a process parameter adjustment instruction set; The strategy optimization module is used for dynamically correcting the process parameter dynamic coupling model combined with process feedback data to generate and execute an optimized process control strategy.

6. A computer-readable storage medium, characterized in that, The computer program is stored and executed by the processor to realize the multi-dimensional process data cooperative simulation control method in any one of claims 1 to 5. The computer program is stored and executed by the processor to realize the multi-dimensional process data cooperative simulation control method in any one of claims 1 to 5.

Images