Intelligent control method and system for workpiece heat treatment equipment

Abstract

The application relates to the technical field of industrial automation control, and discloses an intelligent control method and system for a workpiece heat treatment device, the method comprising the following steps: acquiring and preprocessing device data to generate a clean data set; performing information fusion and threshold filtering on the clean data set to output a fusion state vector; generating a preliminary allocation scheme through iterative calculation on the fusion state vector and performing iterative adjustment to output an optimized work sequence; if the optimized work sequence has resource waste, re-allocating the inefficient part, applying log recording and performance index evaluation, and outputting a final scheduling sequence; integrating the initial device state as a benchmark based on the final scheduling sequence, and if there is a dynamic deviation, outputting a final optimization sequence through deviation correction adjustment; issuing the final optimization sequence instruction, continuously monitoring the execution feedback, realizing closed-loop control, and being used for cyclic optimization control. The method can realize efficient cooperation of multi-device parallel processing.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control technology, and in particular to an intelligent control method and system for workpiece heat treatment equipment. Background Technology

[0002] Currently, heat treatment equipment is a core piece of equipment in the manufacturing industry, and its control precision directly affects workpiece performance and production efficiency. In current industrial scenarios, it is necessary to coordinate the status of multiple devices and process parameters in real time to ensure heat treatment quality while optimizing resource allocation.

[0003] In one existing technology, a static scheduling scheme based on a rule engine allocates tasks through preset thresholds. The system judges the equipment status according to preset rules and generates an initial scheduling sequence, then executes control operations according to a fixed strategy. For example, when a temperature sensor of a certain device detects an anomaly, the system only triggers an alarm without dynamically adjusting the task sequence in conjunction with process parameters, resulting in idle equipment resources or task conflicts.

[0004] Existing technologies suffer from a lack of dynamic adjustment capabilities due to static scheduling and excessive reliance on manual experience to configure rules, resulting in insufficient coordination between equipment status and process data, and low efficiency in task allocation for workpiece heat treatment equipment. Therefore, existing technologies suffer from inefficient work sequence allocation due to the fragmented management of equipment status and process data. Summary of the Invention

[0005] This invention provides an intelligent control method and system for workpiece heat treatment equipment to solve the problem of inefficient work sequence allocation caused by the fragmented management of equipment status and process data.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides an intelligent control method and system for workpiece heat treatment equipment, comprising: Acquire initial equipment status and process data, and preprocess them to obtain a cleaning dataset; If the equipment status index of the cleaning dataset exceeds the preset status threshold, the process data is fused and filtered to obtain a fused status vector. The fusion state vector is corrected based on real-time feedback data, task allocation is calculated, a preliminary working sequence is generated and iteratively adjusted to obtain an optimized working sequence. If the optimized work sequence contains wasted resources, the inefficient parts are reallocated, and the final scheduling sequence is output. Based on the final scheduling sequence and the real-time device status, deviation is judged. If there is a dynamic deviation, deviation correction and adjustment are performed, and the final optimized sequence is output. The final optimized sequence is sent to the heat treatment equipment, and the equipment execution feedback data is collected in real time. The control parameters are updated based on the feedback results.

[0007] In one optional implementation, the step of acquiring initial equipment status and process data, and preprocessing them to obtain a cleaning dataset, includes: The initial equipment status and process data of multiple workpiece heat treatment equipment are obtained, and the timestamps are aligned to obtain the original dataset; The original dataset is filtered for noise using median filtering to output a pre-processed dataset. The initially processed dataset is subjected to outlier detection and outlier removal using the Isolation Forest algorithm, and a filtered dataset is output. The filtered dataset is filled with missing values ​​to generate the clean dataset.

[0008] In one optional implementation, if the equipment status indicators of the cleaning dataset exceed a preset status threshold, then information fusion and filtering are performed on the process data to obtain a fused status vector, including: When the equipment status indicators of the cleaning dataset exceed the preset status threshold, a weighted average method is used to integrate the process data to generate an integrated dataset. Threshold filtering is applied to the integrated dataset to remove outliers, and multi-source data is fused using DS evidence theory to generate an initial fused state vector. If the initial fusion state vector is detected to exceed the preset fusion state threshold, a weighted adjustment method is used to correct it through real-time feedback data, and the corrected fusion state vector is output.

[0009] In one optional implementation, the step of correcting the fused state vector based on real-time feedback data, calculating task allocation, generating a preliminary working sequence, and iteratively adjusting it to obtain an optimized working sequence includes: The fusion state vector is corrected based on real-time feedback data, the task allocation scheme is evaluated through a fitness function, and a preliminary working sequence is output. The preliminary work sequence is evaluated using performance indicators to screen for abnormal states. If the performance indicators exceed the preset performance indicator threshold, the task allocation is optimized by a dynamic weight adjustment method, and a preliminary allocation scheme is generated in combination with the preset operation log records. The initial allocation scheme is iteratively adjusted, and particle swarm optimization is used to calculate task priorities and conflict probabilities to generate a first optimization sequence; If the conflict probability of the first optimized sequence exceeds the preset conflict threshold, an abnormal alarm is triggered. The task is then redistributed through feedback fusion, and the resource allocation order is adjusted by applying the control loop optimization method to obtain the optimized work sequence.

[0010] In one optional implementation, if the optimized work sequence contains resource waste, the inefficient portion is reallocated, and a final scheduling sequence is output, including: The resource consumption of the simulation calculation task is calculated in parallel. If the resource utilization rate of the optimized work sequence is lower than the preset resource utilization threshold, the inefficient part is marked. The task priorities of the inefficient parts are adjusted to correct for deviations, and a redistribution scheme is generated. The adjustment process of the redistribution scheme is recorded, and the execution efficiency index is calculated in combination with performance tracking data. When the execution efficiency reaches the preset execution efficiency threshold, the final scheduling sequence is output.

[0011] In one optional implementation, the deviation judgment based on the final scheduling sequence and real-time device status, and the deviation correction and adjustment if dynamic deviation exists, to output the final optimized sequence, includes: Based on the final scheduling sequence and the real-time device status, deviation is judged. If the deviation exceeds the preset resource utilization threshold, a correction parameter is generated by Kalman filtering. Based on the correction parameters, logistic regression is used to adjust task priorities and generate an optimized scheduling sequence. Bottleneck identification and verification are performed on the performance indicators of the optimized scheduling sequence, and the final optimized sequence is output by integrating equipment status data.

[0012] In one optional implementation, the step of sending the final optimized sequence to the heat treatment equipment and collecting equipment execution feedback data in real time, and updating control parameters based on the feedback results, includes: The final optimized sequence is sent to the heat treatment equipment. If the equipment response time difference exceeds the preset response time difference threshold, the distribution frequency is dynamically adjusted to obtain consistent equipment response. Based on the device response consistency, data stream processing is used to collect feedback data. If the feedback data loss rate exceeds the preset feedback threshold, a retransmission mechanism is triggered to generate a complete feedback set. An isolated forest detection is performed on the abnormal states in the complete feedback set. If an anomaly is found, the control loop is interrupted and an alarm is generated to obtain the anomaly alarm trigger result. Based on the abnormal alarm triggering result and the complete feedback set, multi-source feedback data are integrated, and the control parameters are iteratively updated based on the integrated data to obtain an optimized closed-loop control result.

[0013] In a second aspect, the present invention provides an intelligent control system for a workpiece heat treatment equipment, comprising: The data processing module is used to acquire initial equipment status and process data, and perform preprocessing to obtain a cleaning dataset; The state fusion analysis module is used to perform information fusion and filtering on the process data to obtain a fused state vector if the equipment state index of the cleaning dataset exceeds a preset state threshold. The dynamic optimization module is used to correct the fusion state vector based on real-time feedback data, calculate task allocation, generate a preliminary working sequence and perform iterative adjustments to obtain an optimized working sequence. The resource reallocation module is used to reallocate inefficient parts and output the final scheduling sequence if there is resource waste in the optimized work sequence. The deviation correction module is used to determine the deviation based on the final scheduling sequence and the real-time device status. If there is a dynamic deviation, the deviation is corrected and adjusted, and the final optimized sequence is output. The control execution module is used to send the final optimized sequence to the heat treatment equipment, collect the equipment execution feedback data in real time, and update the control parameters based on the feedback results.

[0014] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement an intelligent control method and system for a workpiece heat treatment device as described in any one of the above.

[0015] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute any one of the above-described intelligent control methods and systems for workpiece heat treatment equipment.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention collects the status and process data of multiple devices in real time through sensor networks, uses median filtering and isolated forest algorithm for noise filtering and outlier detection, and combines timestamp alignment and missing value filling to generate a highly consistent clean dataset, effectively eliminating data distortion caused by device heterogeneity and environmental interference, and providing a reliable data foundation for subsequent status fusion and dynamic scheduling.

[0017] (2) This invention integrates multi-source process data through information fusion to generate a fusion state vector, and introduces real-time feedback for deviation correction. It combines genetic algorithm and particle swarm optimization to iteratively optimize the task allocation sequence, significantly improving the task scheduling efficiency of multi-device parallel processing, reducing resource conflicts and equipment idleness, and realizing dynamic coordination and efficient resource utilization in the heat treatment process.

[0018] (3) This invention identifies resource waste points through parallel operation simulation and deviation correction mechanism, generates the final scheduling sequence by combining log recording and performance index evaluation, and realizes accurate instruction issuance and closed-loop feedback control through instruction distribution protocol, forming a closed-loop intelligent control of the whole process from data acquisition, status evaluation, task allocation to execution feedback, which significantly improves the stability and controllability of heat treatment production.

[0019] (4) This invention integrates data cleaning, state fusion, optimization algorithm and closed-loop control to build an intelligent control system suitable for multi-workpiece heat treatment equipment. It can be widely used in high-precision heat treatment scenarios such as aerospace and automobile manufacturing, and has significant technical versatility and engineering practical value. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an intelligent control method for a workpiece heat treatment equipment according to the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent control system for a workpiece heat treatment equipment provided in the second embodiment of the present invention. Detailed Implementation

[0021] 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.

[0022] Reference Figure 1 The first embodiment of the present invention provides an intelligent control method for a workpiece heat treatment equipment, comprising the following steps: S11: Obtain initial equipment status and process data, and perform preprocessing to obtain a cleaning dataset; S12, if the equipment status index of the cleaning dataset exceeds the preset status threshold, then information fusion and filtering are performed on the process data to obtain a fused status vector; S13, Based on real-time feedback data, the fusion state vector is corrected, task allocation is calculated, a preliminary working sequence is generated and iteratively adjusted to obtain an optimized working sequence; S14, If the optimized work sequence has resource waste, the inefficient part is reallocated and the final scheduling sequence is output; S15, Based on the final scheduling sequence and the real-time device status, deviation is judged. If there is a dynamic deviation, deviation correction and adjustment are performed, and the final optimized sequence is output. S16, the final optimized sequence is sent to the heat treatment equipment, and the equipment execution feedback data is collected in real time, and the control parameters are updated according to the feedback results.

[0023] In step S11, initial equipment status and process data are acquired and preprocessed to obtain a cleaning dataset, including: S111, Obtain the initial equipment status and process data of multiple workpiece heat treatment equipment, and align the timestamps to obtain the original dataset; S112, Median filtering is used to filter noise from the original dataset, and a preliminary processed dataset is output; S113, The isolated forest algorithm is used to detect outliers in the pre-processed dataset and remove outliers, and the filtered dataset is output. S114, Fill in missing values ​​in the filtered dataset to generate the clean dataset.

[0024] In step S111, the initial equipment status and process data of multiple workpiece heat treatment equipment are obtained, and the timestamps are aligned to obtain the original dataset.

[0025] It should be noted that initial equipment status data and process parameter data are collected in real time through a sensor network deployed on multiple workpiece heat treatment devices. Equipment status data includes real-time monitored operating parameters such as temperature, pressure, vibration, and current, while process parameter data includes process requirements such as temperature curves, holding times, and cooling rates set for the heat treatment process. Due to differences in sampling frequencies and clock references among different sensors, a network time protocol (NTP) is used to uniformly align all data timestamps, ensuring that all data points have a consistent time reference and forming the original dataset. Timestamp alignment is fundamental to ensuring collaborative analysis of data from multiple devices. In industrial scenarios, different sensors may experience time recording discrepancies due to network latency or hardware clock drift. The NTP protocol can unify heterogeneous data sources to the same time reference. This step effectively solves the misalignment problem of multi-source asynchronous data in time series analysis, providing structurally consistent data input for subsequent processing.

[0026] In one implementation, a heat treatment workshop is equipped with temperature, pressure, and vibration sensors. The temperature sensor records the time as 10:00:00.123, and the pressure sensor records as 10:00:00.356. The data are uniformly aligned to the timestamp of 10:00:00.000 via the NTP protocol to form an original dataset containing the values ​​of each parameter at that time.

[0027] In step S112, median filtering is used to filter noise from the original dataset, and a preliminary processed dataset is output.

[0028] It should be noted that the noise filtering of the original dataset uses a median filtering method. This involves traversing the dataset using a sliding window, calculating the median of each data point's neighborhood, and replacing any data point with the median if the deviation from the median exceeds a preset threshold. This method effectively suppresses impulse noise and transient interference while preserving the true signal characteristics.

[0029] In one implementation, the temperature sensor continuously collects data sequences of 25.0℃, 24.8℃, 30.2℃, 25.1℃, and 24.9℃, with 30.2℃ showing a significant anomaly. The median of five neighboring points, 25.0℃, is selected, and a threshold of ±2.0℃ is set. Since the deviation of this point is 5.2℃, exceeding the threshold, it is replaced with the median of 25.0℃, effectively eliminating noise caused by electromagnetic interference.

[0030] In step S113, outlier detection and removal are performed on the preliminary processed dataset using the isolated forest algorithm, and a filtered dataset is output.

[0031] It should be noted that the Isolation Forest algorithm calculates outlier scores for data points by randomly partitioning the feature space, identifying and removing outliers that deviate from the main distribution. This effectively identifies outliers caused by sensor malfunctions or process anomalies. The algorithm's default settings are 256 samples and 100 random trees. These parameters can be adjusted based on the data size to balance detection efficiency and accuracy. For example, when dealing with large datasets (e.g., over 100,000 data points), the number of random trees can be increased to 200 to improve detection accuracy; conversely, if the data size is small (e.g., less than 10,000 data points) and real-time performance is critical, the number of random trees can be reduced to 50 to improve computational efficiency.

[0032] In one implementation, vibration sensor data is mostly distributed in the range of 0.1-0.3 mm / s. If a data point shows a speed of 1.2 mm / s, it is considered an outlier. The Isolation Forest algorithm isolates the outlier by constructing a random tree, marks and removes it, and outputs a filtered dataset to reduce the risk of misjudgment and ensure the accuracy of the state assessment.

[0033] In step S114, missing values ​​are filled into the filtered dataset to generate the clean dataset.

[0034] It should be noted that missing value imputation uses the K-nearest neighbor algorithm to calculate the imputation value based on a weighted average of neighboring data points, ensuring data continuity. The weights are set according to the reciprocal of the Euclidean distance between data points; the closer the distance, the higher the weight. The specific calculation formula is as follows:

[0035] in This represents the weight of the i-th data point. Represents the i-th data point Missing points corresponding to missing values The Euclidean distance.

[0036] The formula for calculating the fill value is:

[0037] in, Indicates the fill value. The weight corresponding to the i-th data point. This represents the data of the i-th data point.

[0038] For example, if temperature data is missing at a certain moment, the three nearest points in time, such as 24.8, 25.0, and 24.9℃, are selected. After standardization using Z-score normalization, their Euclidean distances to the missing point are calculated to be 0.2, 0.1, and 0.3, respectively. The weighted calculation results are: w1 = 1 / 0.2 = 5, w2 = 1 / 0.1 = 10, w3 = 1 / 0.3 ≈ 3.33. The weighted average calculation for the imputed value is: (5*24.8 + 10*25.0 + 3.33*24.9) / (5 + 10 + 3.33) ≈ (124 + 250 + 82.917) / 18.33 ≈ 456.917 / 18.33 ≈ 24.92℃. This data is used to fill in the missing temperature data, generating a complete and clean dataset, providing a reliable foundation for subsequent analysis.

[0039] In step S12, if the equipment status indicators of the cleaning dataset exceed a preset status threshold, the process data is fused and filtered to obtain a fused status vector, including: S121, when the equipment status index of the cleaning dataset exceeds the preset status threshold, the weighted average method is used to integrate the process data to generate an integrated dataset. S122, apply threshold filtering to the integrated dataset to remove abnormal indicators, and generate an initial fusion state vector by fusing multi-source data through DS evidence theory; S123, if the initial fusion state vector is detected to exceed the preset fusion state threshold, then the weighted adjustment method is used to correct it through real-time feedback data, and the corrected fusion state vector is output.

[0040] In step S121, when the equipment status index of the cleaning dataset exceeds a preset status threshold, a weighted average method is used to integrate the process data to generate an integrated dataset.

[0041] It should be noted that the weighted average method first compares the measured values ​​of each process parameter with their corresponding preset state thresholds, converting them into state index values ​​with unified dimensions. The criteria for classifying state index values ​​are based on the degree to which the parameter deviates from the threshold: if the parameter value is within the threshold range, the state index value is recorded as 0; if the parameter value exceeds the threshold range but is within 5% of the threshold, it is considered a slight deviation, and the state index value is recorded as ±1 (below the threshold is recorded as -1, above the threshold as +1); if the parameter value exceeds 5% of the threshold range, it is considered a serious anomaly, and the state index value is recorded as ±2. Then, corresponding weights are assigned to the state indices of different parameters for weighted averaging, resulting in a comprehensive calculation of an index value that better reflects the overall operating status of the equipment. The weighted average method can balance the importance of different process parameters, avoiding excessive influence of a single parameter anomaly on the overall state judgment. The weight allocation is based on the degree of influence of the parameter on equipment operating safety and process quality, pre-set through expert evaluation or historical data statistical analysis; parameters with greater influence have higher weights. The preset state threshold is based on the rated operating range provided by the equipment manufacturer (e.g., the allowable temperature fluctuation range of a heat treatment furnace is ±5°C) and combined with historical operating data. The mean (μ) and standard deviation (σ) of this parameter under normal operating conditions are calculated, and it is usually set as μ±2σ or μ±3σ as the threshold boundary. For example, if the furnace body temperature is stable at 250°C for a long time and the standard deviation is 2°C, then its state threshold can be set as (244°C, 256°C) (μ±3σ).

[0042] In one implementation, the preset state threshold ranges for temperature, pressure, and flow rate data of a heat treatment device are [245°C, 255°C], [0.4MPa, 0.6MPa], and [25L / min, 29L / min], respectively. At a certain moment, the measured values ​​are 26.0°C (within the threshold, state index value recorded as 0), 0.5MPa (within the threshold, state index value recorded as 0), and 27.0L / min (within the threshold, state index value recorded as 0). At this time, no single parameter exceeds the limit, but for preventative state fusion, the system still calculates the weighted average state index according to the predetermined weights (temperature importance weight 0.6, pressure 0.2, flow rate 0.2): (0*0.6) + (0*0.2) + (0*0.2) = 0. This result of 0 indicates that the overall state of the equipment is normal, and an integrated dataset is generated.

[0043] In another implementation, the measured values ​​become 253.0℃ (within the threshold, the status index value is recorded as 0), 0.35MPa (below the lower threshold of 0.4MPa by 12.5%, which is a serious anomaly, and the status index value is recorded as -2), and 30.0L / min (above the upper threshold of 29L / min by 3.4%, which is a slight deviation, and the status index value is recorded as +1). The system assigns weights to each parameter according to their importance (temperature 0.6, pressure 0.2, flow rate 0.2), and calculates the weighted average status index value as: (0*0.6)+(-2*0.2)+(1*0.2)=0+(-0.4)+0.2=-0.2. This comprehensive status index value is negative, indicating that the overall state of the equipment tends to be abnormal. The integrated dataset will include this weighted average result and the values ​​of each individual status index, providing a richer and more quantitative basis for subsequent fusion analysis and effectively reducing the impact of fluctuations in a single data point.

[0044] In step S122, threshold filtering is applied to the integrated dataset to remove abnormal indicators, and multi-source data are fused using DS evidence theory to generate an initial fused state vector.

[0045] It should be noted that threshold filtering requires setting reasonable threshold boundaries to filter out data points that do not conform to actual operating conditions, ensuring dataset quality. Dempster-Shafer Evidence Theory (DS) is a mathematical theory for handling uncertain information. It describes and quantifies uncertainty by introducing a belief function and a likelihood function, effectively handling situations with incomplete, conflicting, or ambiguous information. The core of this theory lies in fusing information from multiple evidence sources through Dempster's synthesis rules to derive a comprehensive belief assignment, thereby supporting the decision-making process. The Basic Belief Assignment Function (BPA) in DS Evidence Theory is set based on sensor accuracy, historical false alarm rate, and expert experience. For example, the belief of a temperature sensor in normal state can be initially set to 0.6-0.8, 0.2-0.4 in abnormal state, and 0.1 for the uncertain part.

[0046] In one implementation, if the vibration data at a certain moment is 0.20 mm / s and the preset threshold is 0.18 mm / s, the system marks it as abnormal and discards it, retaining data that falls within the threshold range to provide a reliable foundation for subsequent data fusion. For example, three sensors provide temperature, vibration, and flow data as evidence sources. First, a Basic Trust Assignment Function (BPA) is defined for each sensor's data. For example, for the proposition "equipment is normal": the BPA provided by the temperature sensor is m1(normal) = 0.6, m1(abnormal) = 0.3, m1(uncertain) = 0.1; the BPA provided by the vibration sensor is m2(normal) = 0.5, m2(abnormal) = 0.4, m2(uncertain) = 0.1; and the BPA provided by the flow sensor is m3(normal) = 0.7, m3(abnormal) = 0.2, m3(uncertain) = 0.1.

[0047] Next, Dempster's rules of composition are applied for fusion. First, the conflict coefficient K is calculated, which measures the degree of conflict between pieces of evidence: K = m1(normal)m2(abnormal)m3(abnormal) + m1(normal)m2(abnormal)m3(uncertain) + ... (sum of all conflict combinations). For simplicity, assume K = 0.15 in this example. Then, the intersection of the propositions is assigned a trust and normalized (divided by 1-K). For example, for the proposition "equipment is normal", the overall trust is calculated as: [m1(normal)m2(normal)m3(normal)] / (1-K) = (0.6*0.5*0.7) / (1-0.15) = 0.21 / 0.85 ≈ 0.247. The overall trust of other propositions (abnormal, uncertain) is calculated similarly. Assume that after final fusion, the overall trust of "equipment is normal" is 0.84, the overall trust of "abnormal" is 0.12, and the uncertain part is 0.04. This result indicates a high degree of confidence that the device is in a normal state, with significantly reduced uncertainty. Finally, an initial fused state vector (0.84, 0.12, 0.04) is generated based on the overall trust level. This vector quantifies the trust level of the device state in numerical form, providing accurate and reliable data support for subsequent state assessment and decision-making.

[0048] In step S123, if the initial fusion state vector is detected to exceed the preset fusion state threshold, the weighted adjustment method is used to correct it through real-time feedback data, and the corrected fusion state vector is output.

[0049] It should be noted that the weighted adjustment method refers to correcting the fused state vector based on the weights of real-time feedback data to ensure data accuracy. The feedback weight allocation is determined quantitatively based on the sensor's calibration accuracy, recent failure rate, and data volatility: sensors with higher accuracy, lower failure rate, and less volatility are assigned higher feedback weights. The specific calculation formula is as follows:

[0050] Where raw represents the feedback weight, Se represents the accuracy coefficient, which is a dimensionless quantitative value of the sensor calibration accuracy, usually provided by the sensor manufacturer or obtained through calibration experiments; the higher the value, the higher the accuracy. Ar represents the average error, which is the average absolute error between the sensor's recent measurements and the true values, used to reflect its measurement accuracy. Ne represents the failure rate, which is the frequency of failures of the sensor in a recent period, ranging from 0 to 1. Bo represents the volatility, which is the standard deviation of the sensor's output data, used to measure the stability of its output.

[0051] Next, the raw feedback weights of all sensors are normalized, typically by dividing the feedback weight of a single sensor by the sum of the feedback weights of all sensors, ensuring that the sum of all weights equals 1. The preset fusion state threshold is set based on a large amount of historical fusion data through statistical analysis: historical fusion state vectors of the system under known normal and abnormal states are collected, and their magnitude or confidence score distributions are calculated. The high quantile of the normal state distribution (e.g., the 95th percentile) is taken as the threshold boundary; if the fusion vector exceeds this boundary, it is determined that correction is required.

[0052] In one implementation, the temperature component of the fused state vector at a certain moment is 26.5℃, while the feedback data from the actual high-precision temperature sensor (with a calculated feedback weight of 0.7, and the corresponding calculated weight of the existing temperature component set to 1-0.7=0.3) is 26.0℃, exceeding the threshold of 0.3℃. Therefore, the corrected temperature component is: (26.5*0.3) + (26.0*0.7) = 26.15℃. Based on this, the system adjusts the temperature component to 26.15℃, generates a corrected fusion state vector, and improves the reliability of subsequent analysis.

[0053] In step S13, the fusion state vector is corrected based on real-time feedback data, task allocation is calculated, a preliminary working sequence is generated and iteratively adjusted to obtain an optimized working sequence, including: S131, The fusion state vector is corrected based on real-time feedback data, the task allocation scheme is evaluated through the fitness function, and a preliminary working sequence is output; S132, Apply performance indicators to the preliminary work sequence to evaluate and screen for abnormal states. If the performance indicators exceed the preset performance indicator threshold, optimize the task allocation through a dynamic weight adjustment method and generate a preliminary allocation scheme in combination with the preset operation log records. S133, the preliminary allocation scheme is iteratively adjusted, and the task priority and conflict probability are calculated by particle swarm optimization to generate the first optimization sequence; S134, if the conflict probability of the first optimized sequence exceeds the preset conflict threshold, an abnormal alarm is triggered. The task is redistributed through feedback fusion, and the resource allocation order is adjusted by applying the control loop optimization method to obtain the optimized work sequence.

[0054] In step S131, the fusion state vector is corrected based on real-time feedback data, the task allocation scheme is evaluated through the fitness function, and a preliminary working sequence is output.

[0055] It should be noted that real-time feedback data refers to the operational status information collected in real time from the equipment's sensors, such as real-time monitoring data on temperature, pressure, vibration, current, and power. This data is used to dynamically correct the fused state vector. Its dynamism is reflected in the continuous updating of the state vector using a sliding window method based on the timeliness and reliability of the sensor data. The correction method adopts a weighted average method, setting the weight of both the real-time feedback data and the fused state vector to 0.5 to ensure that they accurately reflect the current actual operating conditions of the equipment. The fitness function is a quantitative evaluation function used to select the optimal solution from multiple candidate task allocation schemes. Its evaluation indicators include equipment load balance, expected task completion time, and resource utilization. This function obtains a comprehensive score by normalizing each indicator and then weighting and summing them. Finally, the scheme with the highest score is output as the initial working sequence.

[0056] In one implementation, a fitness function is used to calculate the scores of three candidate allocation schemes (Scheme A: load balancing score 0.9, completion time 100 min, resource utilization 85%; Scheme B: 0.95, 90 min, 90%; Scheme C: 0.85, 110 min, 88%). To balance system stability and efficiency, the weights for load balancing score are set as w1=0.4, completion time as w2=0.3, and resource utilization as w3=0.3. Min-Max linear normalization is used to normalize the three evaluation indicators. The maximum and minimum values ​​of each indicator in historical tasks are collected first. For example, historical data shows that the maximum load balancing score is 0.95 and the minimum is 0.6, the maximum completion time is 120 min and the minimum is 90 min, and the maximum resource utilization is... With a minimum value of 70%, the evaluation metrics for Scheme A are normalized. The load balancing is normalized to (0.9-0.6) ÷ (0.95-0.6) = 0.86 (rounded to two decimal places), the completion time is normalized to (120-100) ÷ (120-90) = 0.66 (rounded to two decimal places; shorter completion time is better, so reverse mapping is needed), and the resource utilization is normalized to (85%-70%) ÷ (95%-70%) = 0.6. Using the normalized evaluation metrics, the score for Scheme A is calculated as: (0.86×0.4+0.66×0.3+0.6×0.3)×100 = 72 (rounded to the nearest integer). Similarly, the scores for the other two schemes are calculated as follows: Scheme B: 94, Scheme C: 60. Finally, Scheme B, with the highest score, is selected as the initial working sequence output.

[0057] In step S132, the preliminary work sequence is evaluated using performance indicators to screen for abnormal states. If the performance indicators exceed the preset performance indicator threshold, the task allocation is optimized using a dynamic weight adjustment method, and a preliminary allocation scheme is generated in combination with the preset operation log records.

[0058] It should be noted that performance indicator evaluation screens work sequences using preset key performance indicator (KPI) thresholds. Key performance indicators include core production metrics such as equipment utilization, thermal efficiency, and energy efficiency ratio. The identification logic compares the actual values ​​of each indicator with the thresholds; when the deviation exceeds the allowable range, it is judged as an abnormal state. Dynamic weight adjustment methods refer to dynamically adjusting the weight proportion of optimization objectives (such as efficiency and load balancing) in the objective function based on real-time performance feedback. For example, using gradient descent-based optimization methods, the optimization weights are updated by calculating the partial derivatives of the objective function with respect to the weights. The calculation formula is as follows:

[0059] in To optimize the weights, the weight coefficients represent the weights of the current optimization objective (such as efficiency or load balancing) in the objective function. The initial values ​​are set according to process requirements or experience. η is the learning rate, a positive real number hyperparameter used to control the step size of each weight update. It is usually selected based on experience or through cross-validation, and its value range is generally from 0.01 to 0.3. This represents the partial derivative of the objective function with respect to the current optimization weights, reflecting the direction and strength of the impact of the weight change on the overall objective function value. This gradient descent update process iteratively adjusts the optimization weights, causing the objective function value to converge towards its minimum, thereby achieving dynamic optimization of the task allocation scheme. The preset performance index thresholds are set based on process requirements and economic objectives, or through statistical analysis of the equipment's historical best performance. For example, 85% of the equipment's average utilization rate over the past year might be used as the threshold to incentivize the system to optimize to a higher level.

[0060] In one implementation, performance evaluation reveals that the expected thermal efficiency of the tempering furnace in the initial working sequence is 80%, lower than the preset threshold of 85%, which the system classifies as an abnormal state. The objective function is set to minimize the thermal efficiency deviation, and the objective function value is the absolute value of the difference between the actual thermal efficiency and the threshold thermal efficiency. The system calculates the partial derivative of the objective function with respect to the optimization weight of "improving thermal efficiency" to determine the adjustment direction. Assuming the current objective function value is 5% (i.e., 85%-80%), the partial derivative calculation indicates that this weight needs to be increased to improve thermal efficiency. Taking the learning rate η=0.1, the weight adjustment amount calculated according to the update formula is:

[0061] Therefore, the weight of the "improving thermal efficiency" objective was increased from 0.3 to 0.4. The optimization algorithm was rerun to generate a new scheme, and the triggering reason for this adjustment, the weight change value, and a summary of the new scheme were recorded in the operation log to form a preliminary allocation scheme.

[0062] In step S133, the preliminary allocation scheme is iteratively adjusted, and the task priority and conflict probability are calculated using particle swarm optimization to generate the first optimization sequence.

[0063] It should be noted that Particle Swarm Optimization (PSO) is a swarm intelligence optimization algorithm that simulates the foraging behavior of bird flocks. It guides the search direction through individual and group extreme values, and is used here to perform multi-round iterative optimization of task allocation schemes. Its goal is to calculate a set of task priority assignments, with priority values ​​ranging from 0 to 1, where higher values ​​indicate higher priority. It also estimates the probability of conflict between tasks due to competition for equipment resources under these priorities, ultimately generating a task sequence with lower conflict and higher efficiency. In the PSO algorithm, the number of particles is set to 20-50, the number of iterations is 50-100, the inertia weight ω decreases linearly from 0.9 to 0.4, and the cognitive factor c1 and social factor c2 are both set to 2.0.

[0064] In one implementation, the system aims to minimize the total completion time and equipment idle time. It uses particle swarm optimization to iterate the initial allocation scheme for 50 rounds. The initial conflict probability of a certain quenching task is 0.35. After adjusting its priority from 0.8 to 0.6 and migrating it from the original equipment to another idle furnace, its conflict probability gradually decreases to 0.12 in the iterations. Finally, the first optimized sequence with significantly reduced conflicts is output.

[0065] In step S134, if the conflict probability of the first optimized sequence exceeds a preset conflict threshold, an abnormal alarm is triggered. The task is redistributed through feedback fusion, and the resource allocation order is adjusted by applying a control loop optimization method to obtain an optimized working sequence.

[0066] It should be noted that the preset conflict threshold probability is generally calculated by dividing the number of historical conflicts by the total number of task executions and then multiplying by a safety factor (the default setting is 0.8, which can be adjusted in practice according to the permissible conflict level of the task; for example, for task types with low conflict tolerance, it can be set to 0.6). Feedback fusion refers to integrating and analyzing real-time equipment status data with task execution feedback information, specifically by fusing multi-source data using an average weighting method. Control loop optimization is a closed-loop method that continuously fine-tunes resource allocation strategies based on feedback results, aiming to dynamically eliminate resource competition and ensure the smooth operation of the production process.

[0067] In one implementation, the task conflict probability of the device used for nitriding in the first optimization sequence is calculated to be 0.28, which exceeds the preset conflict threshold of 0.15, triggering an abnormal alarm in the system. The control center uses an average weighting method to integrate the device's recent temperature fluctuation feedback and current queue status, and through a control loop optimization method, decides to postpone two non-urgent preheating tasks and insert them into another compatible device for execution. After adjustment, the conflict probability of the device is reduced to 0.09, and finally an executable optimized work sequence is output.

[0068] In step S14, if the optimized work sequence contains wasted resources, the inefficient portion is reallocated, and the final scheduling sequence is output, including: S141, simulate the resource consumption of the task by running in parallel. If the resource utilization rate of the optimized work sequence is lower than the preset resource utilization threshold, the inefficient part is marked. S142, perform deviation correction and adjustment on the task priority of the inefficient part, and generate a redistribution scheme; S143, record the adjustment process of the redistribution scheme, calculate the execution efficiency index in combination with performance tracking data, and output the final scheduling sequence when the execution efficiency reaches the preset execution efficiency threshold.

[0069] In step S141, the resource consumption of the simulation calculation task is calculated in parallel. If the resource utilization rate of the optimized work sequence is lower than the preset resource utilization threshold, the inefficient part is marked.

[0070] It should be noted that parallel operation simulation is achieved by establishing a virtual production environment and simultaneously simulating the execution process of multiple tasks to accurately predict their resource consumption, such as CPU, memory, energy consumption, and equipment occupancy time. Resource utilization rate measures the percentage of the above resources that are effectively utilized. The preset threshold is set based on the statistical analysis of historical equipment operation data and resource optimization configuration targets. It is usually taken as 70%-80% of the average utilization rate of the equipment over a certain period of time as the benchmark. If it is lower than this threshold, it indicates that there are idle resources, and the corresponding equipment or task segment will be marked as an inefficient part.

[0071] In one implementation, the system performs parallel simulation of the optimized work sequence. The results show that an atmosphere furnace only performs the heat preservation task during the midday period, with an average CPU utilization of only 30% and a thermal energy utilization of only 40%, both far below the preset resource utilization threshold of 60%. Therefore, the equipment and its heat preservation task segment are marked as inefficient parts.

[0072] In step S142, the task priorities of the inefficient parts are adjusted by deviation correction to generate a redistribution scheme.

[0073] It should be noted that the deviation correction adjustment is for the identified inefficient units. It analyzes the gap between the current task priority and the ideal resource utilization rate. The deviation is calculated by dividing the absolute value of the difference between the current utilization rate and the target utilization rate by the target utilization rate. Then, the priority value is adjusted proportionally according to the deviation. The new priority is 1 minus the difference of the deviation and then multiplied by the original priority. The aim is to guide tasks to more suitable equipment or time periods for execution, so as to eliminate resource idleness and form a new task redistribution scheme with higher resource utilization efficiency.

[0074] In one implementation, for a marked inefficient atmosphere furnace, system analysis reveals that its previous long-duration annealing task with a priority of 0.6 has a high actual thermal energy requirement but a low computational requirement, which is mismatched with the characteristics of the current equipment. By calculating the deviation = |40%-60%| / 60% = 0.33, the priority is adjusted from 0.6 to 0.6×(1-0.33)≈0.4, and it is swapped with a short-duration high-heat task on another high-computation-load equipment, thereby generating a new redistribution scheme, which is expected to increase the thermal energy utilization rate of the furnace to 65%.

[0075] In step S143, the adjustment process of the redistribution scheme is recorded, and the execution efficiency index is calculated in combination with performance tracking data. When the execution efficiency reaches the preset execution efficiency threshold, the final scheduling sequence is output.

[0076] It should be noted that the adjustment process log must record in detail the logic of the priority adjustment, the task IDs and device IDs involved, and the expected benefits; performance tracking data comes from the system's historical operation logs or simulation prediction data, and is used to calculate the execution efficiency indicators under the new scheme. For example, the average task completion time is obtained by weighted averaging of the expected execution times of each task, with the weights set according to the urgency of the tasks, with an urgent task weight of 0.6 and a normal task weight of 0.4; the overall device utilization rate is calculated by the arithmetic mean of the resource utilization rates of each device; the preset execution efficiency threshold is determined based on the historical best level, for example, the average task completion time threshold is taken as 110% of the historical best value, and the overall device utilization rate threshold is taken as 90% of the historical average level; only when all execution efficiency indicators reach the preset execution efficiency threshold will the scheme be determined as the final scheduling sequence.

[0077] In one implementation, the system records in the log the reason for adjusting the annealing task T-102 from furnace L-5 to furnace L-7 and the expected increase in thermal efficiency. Then, combined with the historical performance data of the past three months, the system calculates that the overall average task completion time after the reallocation is (120min×0.6+110min×0.4)=116min, and the average comprehensive utilization rate of the equipment reaches 88%, which is better than the preset execution efficiency threshold (120 minutes and 85%). Therefore, the scheme is approved as the final scheduling sequence and issued for execution.

[0078] In step S15, deviation judgment is performed based on the final scheduling sequence and the real-time device status. If dynamic deviation exists, deviation correction and adjustment are performed, and the final optimized sequence is output, including: S151, Based on the final scheduling sequence and the real-time device status, deviation is judged. If the deviation exceeds the preset resource utilization threshold, a correction parameter is generated by Kalman filtering. S152, Adjust task priorities using logistic regression calculation based on the correction parameters to generate an optimized scheduling sequence; S153, bottleneck identification and verification are performed on the performance indicators of the optimized scheduling sequence, and the final optimized sequence is output by integrating equipment status data.

[0079] In step S151, deviation is judged based on the final scheduling sequence and the real-time device status. If the deviation exceeds the preset resource utilization threshold, a correction parameter is generated by Kalman filtering.

[0080] It should be noted that dynamic deviation refers to the difference between the actual resource utilization rate of the equipment and the expected value of the final scheduling sequence. The resource utilization threshold is a preset critical value used to determine whether resources are being used effectively. Kalman filtering is a filtering method that optimally estimates the system state through a recursive algorithm. Here, it is used to generate optimal correction parameters based on the deviation between real-time equipment state data and expected values ​​to reduce system uncertainty. In specific applications, the Kalman filter takes historical equipment utilization and current real-time monitoring data as input, performs recursive calculations through prediction and update steps, and finally outputs the optimal state estimate as the correction parameter.

[0081] In one implementation, the final scheduling sequence expects a certain device's CPU utilization rate to be 75%, but real-time monitoring data shows that its actual utilization rate is only 50%, significantly lower than the preset resource utilization threshold of 65%. The system determines that there is a dynamic deviation. Subsequently, the Kalman filter algorithm is applied, using the device's historical utilization rate data and the current real-time fluctuation as input, and the correction parameters are obtained through recursive calculation: state estimate x_est=70%, error covariance P_est=0.1. The system uses these parameters to correct the expected utilization rate from 75% to 70%, and provides a quantitative basis for subsequent priority adjustment.

[0082] In step S152, the task priority is adjusted by logistic regression calculation based on the correction parameters to generate an optimized scheduling sequence.

[0083] It should be noted that logistic regression is a statistical model used to calculate the probability of events occurring. Here, it is used to analyze the relationship between correction parameters and task attributes (such as resource requirements and urgency), calculate and adjust the probability of tasks being prioritized, thereby generating a new, more optimized scheduling sequence that better reflects the actual equipment resource situation. The model uses the sigmoid function as the activation function, and its functional form is:

[0084] in The regularization parameter λ is set to 0.01 to prevent overfitting. The model parameters θ are solved using the maximum likelihood estimation method, and its likelihood function is: .

[0085] In one implementation, the system uses a logistic regression model to analyze all tasks on the device based on the correction parameters generated by Kalman filtering (e.g., the expected CPU utilization of a device is corrected to 70%). Using task resource requirements, urgency, and current device load as characteristics, the system calculates that the allocation weight of task A is reduced from 0.3 to 0.2, and the weight of task B is increased from 0.4 to 0.5. The task allocation order is then reordered according to the new weights to generate an optimized scheduling sequence: task B (weight 0.5) -> task C (weight 0.3) -> task A (weight 0.2), generating an optimized scheduling sequence that better matches the current load capacity of the device.

[0086] In step S153, bottleneck identification and verification are performed on the performance indicators of the optimized scheduling sequence, and the final optimized sequence is output by fusing device status data.

[0087] It should be noted that bottleneck identification and verification identifies potential performance limitations in the system by analyzing key performance indicators (such as task completion time, device load rate, queue length, etc.) of the optimized scheduling sequence. The sequence is then validated by integrating real-time device status data (such as current CPU temperature and memory availability). If any indicator exceeds 90% of a preset threshold for three consecutive cycles, the device is considered to have a bottleneck risk. The final optimized sequence is only output after ensuring efficient execution in a real-world environment. The system uses a weighted deviation integral method to calculate the risk score, comprehensively considering the deviation of predicted and real-time status indicators from their respective benchmark values. This score is calculated using the following formula:

[0088] in, Indicates risk score, This indicates the deviation of the prediction indicator, obtained by subtracting a preset threshold from the predicted value. This indicates the deviation of the real-time indicator, which is obtained by subtracting the indicator benchmark value from the real-time value.

[0089] In one implementation, the optimized scheduling sequence is simulated. Performance indicators show that the average task waiting time has met the target. However, trend analysis using the moving average method reveals that the memory utilization rate of device 3 is predicted to reach 89% during the evening peak hours, which is close to the critical value. The system then integrates the real-time status data of the device (showing that the current memory utilization rate is 65% and the heat dissipation is good). Assuming that the preset memory utilization rate threshold is 85% and the memory utilization rate baseline is 60%, the calculated risk score is 4.5, which is lower than the preset memory risk score threshold of 5. Therefore, the risk is considered controllable, and the sequence is approved as the final optimized sequence and output. At the same time, device 3 is recorded as an object requiring special attention.

[0090] In step S16, the final optimized sequence is sent to the heat treatment equipment, and the equipment execution feedback data is collected in real time. The control parameters are updated according to the feedback results, including: S161, the final optimized sequence is sent to the heat treatment equipment. If the equipment response time difference exceeds the preset response time difference threshold, the distribution frequency is dynamically adjusted to obtain equipment response consistency. S162, based on the device response consistency, data stream processing is used to collect feedback data. If the feedback data loss rate exceeds the preset feedback threshold, a retransmission mechanism is triggered to generate a complete feedback set. S163, perform isolated forest detection on the abnormal state in the complete feedback set. If an abnormality is found, interrupt the control loop and generate an alarm to obtain the abnormal alarm trigger result. S164, Based on the abnormal alarm triggering result and the complete feedback set, integrate the multi-source feedback data, iteratively update the control parameters based on the integrated data, and obtain the optimized closed-loop control result.

[0091] In step S161, the final optimized sequence is sent to the heat treatment equipment. If the equipment response time difference exceeds the preset response time difference threshold, the distribution frequency is dynamically adjusted to obtain consistent equipment response.

[0092] It should be noted that device response time difference refers to the difference between the maximum and minimum response times of different devices under the same command. It is a key indicator for measuring the consistency of multi-device collaboration. If this difference exceeds the preset response time difference threshold, it indicates a delay in inter-device collaboration. The load pressure needs to be reduced by dynamically decreasing the command distribution frequency. The adjustment strategy is as follows:

[0093] in The adjusted instruction dispatch frequency, The original instruction distribution frequency, To preset the response time difference threshold, This represents the actual response time difference.

[0094] The preset response time difference threshold is set based on network communication delay statistics and device controller response characteristics. It is usually taken as 95% of the historical response time difference of the device or one-tenth of the process cycle time as the benchmark value.

[0095] In one implementation, the system synchronously sends heating commands to five heat treatment devices, with a preset response time difference threshold of 500 milliseconds. Actual monitoring revealed that device 4 experienced a response delay, resulting in a time difference of 600 milliseconds, exceeding the preset time difference threshold. The system then automatically calculated a new frequency of 1 × (500 / 600) ≈ 0.83 times / second based on the adjustment strategy. Therefore, the command distribution frequency was dynamically adjusted from once per second to 0.8 times per second. After slowing down the distribution pace, the response time difference between the devices fell back to within 400 milliseconds, achieving good device response consistency.

[0096] In step S162, based on the consistency of the device response, data stream processing is used to collect feedback data. If the feedback data loss rate exceeds a preset feedback threshold, a retransmission mechanism is triggered to generate a complete feedback set.

[0097] It should be noted that data stream processing is a technology for real-time processing of continuous data streams, used for efficiently acquiring continuous feedback information from devices; the feedback data loss rate is an indicator of the integrity of data acquisition, and its calculation formula is as follows:

[0098] in, Indicates the feedback data loss rate. This indicates the expected number of data packets, referring to the total number of data packets that should be received from the target device within a specific time window, based on the system's preset sampling frequency and data reporting rules. This indicates the actual number of data packets received, which refers to the number of data packets that the system successfully received and verified within the same time window.

[0099] The feedback data loss rate is an indicator that measures the degree of missing data in data collection; the higher the value, the worse the data integrity. A formula quantifies the proportion of unsuccessfully received data packets, allowing for real-time monitoring of data stream integrity and providing a basis for triggering retransmission mechanisms or handling anomalies. If the loss rate exceeds a preset threshold (e.g., 5%), it indicates that a network or device anomaly caused data packet loss, requiring a retransmission mechanism to request resending of data from the source of the loss. To ensure data integrity, the system sets a maximum of 3 retransmissions. If the maximum number of retransmissions is reached and complete data still cannot be obtained, the device is marked as having a data anomaly, and a backup data source switching process is initiated to ensure that a complete feedback dataset without missing data is ultimately obtained for subsequent analysis. The loss rate threshold is set based on network reliability test results, such as a historical average loss rate of 3%. Based on network fluctuation tolerance and system reliability requirements, a margin is appropriately increased (e.g., by 2 percentage points) on top of this historical average loss rate; therefore, the anomaly threshold is set at 5%.

[0100] In one implementation, during periods when device response is consistent, the system collects temperature and power feedback data from each device in real time through data stream processing. If the monitoring finds that the data stream loss rate from device 2 reaches 7%, exceeding the preset threshold of 5%, the system immediately triggers a retransmission mechanism. Based on the log identifier, the system re-requests data packets for the lost time period from device 2. After retransmission, a complete feedback set containing all expected data points is successfully formed.

[0101] In step S163, an isolated forest detection is performed on the abnormal states in the complete feedback set. If an anomaly is found, the control loop is interrupted and an alarm is generated, resulting in an anomaly alarm triggering result.

[0102] It should be noted that Isolation Forest detection is suitable for rapid anomaly identification in high-dimensional data. With a sample size of 256 and a random tree count of 100, it identifies anomalies by differentiating the path length between isolated anomaly points and normal points. The control loop is a closed-loop process of continuously issuing commands, collecting feedback, and making adjustments. Once a serious anomaly that may endanger equipment or production (such as a sudden temperature rise) is detected, the current loop must be immediately interrupted and an alarm triggered to prevent the accident from escalating. The anomaly alarm trigger result includes the interruption decision and alarm details. Safety thresholds are set based on safety specifications provided by the equipment manufacturer and historical accident data.

[0103] In one implementation, the system applies the Isolation Forest algorithm to the complete feedback set for detection. It finds that a temperature feedback value of 55°C from device 3 has an extremely short isolation path compared to the 45°C values ​​of other devices and its own historical data, and is quickly identified as an anomaly. Since this value far exceeds the safety threshold of 50°C, the system immediately interrupts the currently executing control loop, stops issuing new instructions to the device, and generates an alarm containing the device ID, abnormal parameters, and suggested inspection content, thus obtaining the anomaly alarm trigger result.

[0104] In step S164, based on the abnormal alarm triggering result and the complete feedback set, multi-source feedback data is integrated, and the control parameters are iteratively updated based on the integrated data to obtain an optimized closed-loop control result.

[0105] It should be noted that multi-source feedback data integration involves aligning and unifying the format of heterogeneous data from different devices and sensors to form a global view. Based on this, and considering the problems revealed by anomaly alarms, control parameters (such as PID controller parameters and scheduling weights) are iteratively updated and optimized. This allows the entire closed-loop control system to learn from and adjust to anomalies, ultimately outputting a more robust and optimized control result that better adapts to the actual production environment. The iterative update of the control parameters uses the gradient descent method, with the objective function being to minimize the mean square error between the actual and expected resource utilization rates. The initial learning rate is set to 0.03, a value determined based on historical convergence speed and empirical values, achieving a balance between convergence speed and stability. The dynamic adjustment range of the learning rate is 0.01-0.05, and the iteration termination condition is an error of less than 5% or reaching the maximum number of iterations (10).

[0106] In one implementation, the system integrates the abnormal alarm result (abnormal temperature of device 3) and the temperature and power data of all devices in the complete feedback set; analysis revealed that the overall system load was too high when the abnormality occurred, and the current control parameters were too aggressive; therefore, iterative optimization was adopted, automatically reducing the heating rate parameter of device 3 by 20% and slightly increasing the load balancing weight of other devices; after two rounds of iterative calculation, a new set of control parameters was obtained, and after application, the system ran smoothly under similar loads, thus obtaining optimized closed-loop control results.

[0107] In summary, this invention discloses an intelligent control method for workpiece heat treatment equipment, comprising: acquiring initial equipment status and process data, and preprocessing them to obtain a cleaning dataset; if the equipment status index of the cleaning dataset exceeds a preset status threshold, then performing information fusion and filtering on the process data to obtain a fused status vector; correcting the fused status vector based on real-time feedback data, calculating task allocation, generating a preliminary work sequence and iteratively adjusting it to obtain an optimized work sequence; if the optimized work sequence has resource waste, then reallocating inefficient parts and outputting a final scheduling sequence; judging deviations based on the final scheduling sequence and real-time equipment status, and if dynamic deviations exist, then correcting and adjusting the deviations and outputting a final optimized sequence; sending the final optimized sequence to the heat treatment equipment, and collecting equipment execution feedback data in real time, and updating control parameters according to the feedback results.

[0108] Reference Figure 2 The second embodiment of the present invention provides an intelligent control system for a workpiece heat treatment equipment, comprising: The data processing module is used to acquire initial equipment status and process data, and perform preprocessing to obtain a cleaning dataset; The state fusion analysis module is used to perform information fusion and filtering on the process data to obtain a fused state vector if the equipment state index of the cleaning dataset exceeds a preset state threshold. The dynamic optimization module is used to correct the fusion state vector based on real-time feedback data, calculate task allocation, generate a preliminary working sequence and perform iterative adjustments to obtain an optimized working sequence. The resource reallocation module is used to reallocate inefficient parts and output the final scheduling sequence if there is resource waste in the optimized work sequence. The deviation correction module is used to determine the deviation based on the final scheduling sequence and the real-time device status. If there is a dynamic deviation, the deviation is corrected and adjusted, and the final optimized sequence is output. The control execution module is used to send the final optimized sequence to the heat treatment equipment, collect the equipment execution feedback data in real time, and update the control parameters based on the feedback results.

[0109] It should be noted that the intelligent control system for a workpiece heat treatment equipment provided in this embodiment of the invention is used to execute all the process steps of the intelligent control method for a workpiece heat treatment equipment in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0110] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as an intelligent control system for a workpiece heat treatment device. When the processor executes the computer program, it implements the steps in the aforementioned intelligent control method embodiments for various workpiece heat treatment devices, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data processing module.

[0111] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0112] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0113] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0114] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0115] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0116] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0117] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. An intelligent control method for workpiece heat treatment equipment, characterized in that, include: Acquire initial equipment status and process data, and preprocess them to obtain a cleaning dataset; If the equipment status indicators of the cleaning dataset exceed a preset status threshold, then the process data is fused and filtered to obtain a fused status vector, including: When the equipment status indicators of the cleaning dataset exceed the preset status threshold, a weighted average method is used to integrate the process data to generate an integrated dataset. The weighted average method involves first comparing the measured values ​​of each process parameter with their corresponding preset status thresholds, converting them into status indicator values ​​with unified dimensions. The criteria for classifying the status indicator values ​​are based on the degree to which the parameter deviates from the threshold: if the parameter value is within the threshold range, the status indicator value is recorded as 0; if the parameter value exceeds the threshold range but is within 5% of the threshold, it is considered a slight deviation, and the status indicator value is recorded as ±1; if the parameter value exceeds the threshold range by more than 5%, it is considered a serious anomaly, and the status indicator value is recorded as ±2. Then, corresponding weights are assigned to the status indicators of different parameters for weighted averaging. The weight allocation is based on a pre-set value for the parameter's impact on equipment operating safety and process quality. The preset status threshold is calculated based on the rated operating range provided by the equipment manufacturer and combined with historical operating data, calculating the mean μ and standard deviation σ of the parameter under normal operating conditions, and setting it as μ±2σ or μ±3σ as the threshold boundary. The integrated dataset is filtered by threshold to remove outliers, and multi-source data is fused using the DS evidence theory to generate an initial fused state vector; wherein the basic trust assignment function in the DS evidence theory is set based on sensor accuracy, historical false alarm rate and expert experience. If the initial fusion state vector is detected to exceed a preset fusion state threshold, it is corrected using a weighted adjustment method based on real-time feedback data, and the corrected fusion state vector is output. The feedback weight allocation is determined quantitatively based on the sensor's calibration accuracy, recent failure rate, and data volatility, using the following formula: Where Se is the sensor calibration accuracy coefficient, Ar is the average error, Ne is the failure rate, and Bo is the volatility. The feedback weights of all sensors are normalized so that the sum of all weights is 1. The preset fusion state threshold is set based on a large amount of historical fusion data through statistical analysis, and the 95th percentile of the fusion state vector distribution under known normal conditions is taken as the threshold boundary. The fusion state vector is corrected based on real-time feedback data, task allocation is calculated, a preliminary working sequence is generated and iteratively adjusted to obtain an optimized working sequence. If the optimized work sequence contains wasted resources, the inefficient parts are reallocated, and the final scheduling sequence is output. Based on the final scheduling sequence and the real-time device status, deviation is judged. If there is a dynamic deviation, deviation correction and adjustment are performed, and the final optimized sequence is output. The final optimized sequence is sent to the heat treatment equipment, and the equipment execution feedback data is collected in real time. The control parameters are updated based on the feedback results.

2. The intelligent control method for a workpiece heat treatment equipment according to claim 1, characterized in that, The process of acquiring initial equipment status and process data, and preprocessing it to obtain a cleaning dataset includes: The initial equipment status and process data of multiple workpiece heat treatment equipment are obtained, and the timestamps are aligned to obtain the original dataset; The original dataset is filtered for noise using median filtering to output a pre-processed dataset. The initially processed dataset is subjected to outlier detection and outlier removal using the Isolation Forest algorithm, and a filtered dataset is output. The filtered dataset is filled with missing values ​​to generate the clean dataset.

3. The intelligent control method for a workpiece heat treatment equipment according to claim 1, characterized in that, The process of correcting the fused state vector based on real-time feedback data, calculating task allocation, generating a preliminary working sequence, and iteratively adjusting it to obtain an optimized working sequence includes: The fusion state vector is corrected based on real-time feedback data, the task allocation scheme is evaluated through a fitness function, and a preliminary working sequence is output. The preliminary work sequence is evaluated using performance indicators to screen for abnormal states. If the performance indicators exceed the preset performance indicator threshold, the task allocation is optimized by a dynamic weight adjustment method, and a preliminary allocation scheme is generated in combination with the preset operation log records. The initial allocation scheme is iteratively adjusted, and particle swarm optimization is used to calculate task priorities and conflict probabilities to generate a first optimization sequence; If the conflict probability of the first optimized sequence exceeds the preset conflict threshold, an abnormal alarm is triggered. The task is then redistributed through feedback fusion, and the resource allocation order is adjusted by applying the control loop optimization method to obtain the optimized work sequence.

4. The intelligent control method for a workpiece heat treatment equipment according to claim 1, characterized in that, If the optimized work sequence contains wasted resources, the inefficient portion is reallocated, and the final scheduling sequence is output, including: The resource consumption of the simulation calculation task is calculated in parallel. If the resource utilization rate of the optimized work sequence is lower than the preset resource utilization threshold, the inefficient part is marked. The task priorities of the inefficient parts are adjusted to correct for deviations, and a redistribution scheme is generated. The adjustment process of the redistribution scheme is recorded, and the execution efficiency index is calculated in combination with performance tracking data. When the execution efficiency reaches the preset execution efficiency threshold, the final scheduling sequence is output.

5. The intelligent control method for a workpiece heat treatment equipment according to claim 1, characterized in that, The step of determining deviations based on the final scheduling sequence and real-time device status, and correcting and adjusting for dynamic deviations if they exist, to output the final optimized sequence, includes: Based on the final scheduling sequence and the real-time device status, deviation is judged. If the deviation exceeds the preset resource utilization threshold, a correction parameter is generated by Kalman filtering. Based on the correction parameters, logistic regression is used to adjust task priorities and generate an optimized scheduling sequence. Bottleneck identification and verification are performed on the performance indicators of the optimized scheduling sequence, and the final optimized sequence is output by integrating equipment status data.

6. The intelligent control method for a workpiece heat treatment equipment according to claim 1, characterized in that, The process of sending the final optimized sequence to the heat treatment equipment and collecting real-time equipment execution feedback data, and updating control parameters based on the feedback results, includes: The final optimized sequence is sent to the heat treatment equipment. If the equipment response time difference exceeds the preset response time difference threshold, the distribution frequency is dynamically adjusted to achieve consistent equipment response. Based on the device response consistency, data stream processing is used to collect feedback data. If the feedback data loss rate exceeds the preset feedback threshold, a retransmission mechanism is triggered to generate a complete feedback set. An isolated forest detection is performed on the abnormal states in the complete feedback set. If an anomaly is found, the control loop is interrupted and an alarm is generated to obtain the anomaly alarm trigger result. Based on the abnormal alarm triggering result and the complete feedback set, multi-source feedback data are integrated, and the control parameters are iteratively updated based on the integrated data to obtain an optimized closed-loop control result.

7. An intelligent control system for a workpiece heat treatment equipment, characterized in that, A method for implementing the intelligent control of a workpiece heat treatment equipment as described in any one of claims 1 to 6, comprising: The data processing module is used to acquire initial equipment status and process data, and perform preprocessing to obtain a cleaning dataset; The state fusion analysis module is used to perform information fusion and filtering on the process data to obtain a fused state vector if the equipment state index of the cleaning dataset exceeds a preset state threshold. The dynamic optimization module is used to correct the fusion state vector based on real-time feedback data, calculate task allocation, generate a preliminary working sequence and perform iterative adjustments to obtain an optimized working sequence. The resource reallocation module is used to reallocate inefficient parts and output the final scheduling sequence if there is resource waste in the optimized work sequence. The deviation correction module is used to determine the deviation based on the final scheduling sequence and the real-time device status. If there is a dynamic deviation, the deviation is corrected and adjusted, and the final optimized sequence is output. The control execution module is used to send the final optimized sequence to the heat treatment equipment, collect the equipment execution feedback data in real time, and update the control parameters based on the feedback results.

Images