Industrial signal dynamic weighting method based on process log feedback control

Abstract

The invention relates to the technical field of industrial automation control, in particular to an industrial signal dynamic weighting method based on process log feedback control, and aims to solve the problems that in the prior art, an existing system cannot dynamically adjust data weights according to real-time events in the production process; and the key time and stage in the production process cannot be accurately reflected. By analyzing the process event log in real time, dynamically adjusting the signal weight factor, and combining the closed-loop adjustment instruction and the accumulative error analysis mechanism, the problem that the existing system cannot dynamically optimize production parameters and process accumulative errors is solved; the method has the advantages that accurate closed-loop control is achieved by dynamically adjusting the signal weight, the multi-device cooperation efficiency is improved, and accumulated errors are effectively processed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial automation control, and particularly relates to an industrial signal dynamic weighting method based on process log feedback control. BACKGROUND

[0002] In existing manufacturing execution systems (MES) and supervisory control and data acquisition (SCADA) systems, there is a fault between the information layer (IT) and the control layer (OT), making it difficult to achieve dynamic optimization based on real-time events. Specifically, the existing system cannot dynamically adjust the data weight according to real-time events in the production process (such as heating completion, injection start, etc.), resulting in the inability to accurately reflect the key moments and stages in the production process. At the same time, there is a lack of effective feedback mechanism, making it difficult to adjust production parameters in a timely manner according to abnormal situations. These problems lead to low production efficiency and the inability to achieve precise real-time optimization and control.

[0003] Especially in multi-device collaborative scenarios, such as multi-robot collaborative control of automobile welding lines or temperature and humidity closed-loop control of food fermentation processes, the existing system cannot achieve dynamic signal coordination between devices. In addition, for cumulative errors in the process, traditional methods lack effective analysis mechanisms and cannot achieve predictive maintenance and adaptive calibration, which is particularly prominent in precision manufacturing fields such as semiconductor lithography processes. SUMMARY

[0004] (I) Technical problems to be solved

[0005] To solve the above problems, the present application proposes an industrial signal dynamic weighting method based on process log feedback control, aiming to solve the problem that the existing system cannot dynamically adjust the data weight according to real-time events in the production process, resulting in the inability to accurately reflect the key moments and stages in the production process.

[0006] (II) Technical solutions

[0007] The industrial signal dynamic weighting method based on process log feedback control of the present application comprises:

[0008] S100, real-time analysis of process event logs of a manufacturing execution system (MES) or a supervisory control and data acquisition system (SCADA), extraction of event types and associated signal segments;

[0009] S200, dynamically adjusting the signal weight factor of the corresponding process stage according to the event type, wherein the key event triggering stage is given a weighting coefficient higher than the baseline weight;

[0010] S300, based on the weighted signal similarity analysis result, triggering the production parameter closed-loop adjustment instruction and feeding back to the control system.

[0011] (III) Beneficial Effects

[0012] Compared with the prior art, the beneficial effects of the present application are:

[0013] (1) In the present application, by real-time analyzing the process event log and dynamically adjusting the signal weight factor, combined with the closed-loop adjustment instruction and the cumulative error analysis mechanism, the problem that the existing system cannot dynamically optimize the production parameters and handle the cumulative error is solved, which has the advantages of realizing precise closed-loop control by dynamically adjusting the signal weight, improving the collaborative efficiency of multiple devices and effectively handling the cumulative error. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0015] Fig. 1 The schematic diagram of the logic structure of the industrial signal dynamic weighting method based on process log feedback control;

[0016] Fig. 2 The schematic diagram of the framework structure of the overall system. DETAILED DESCRIPTION

[0017] There is a long-standing fault problem between the information layer and the control layer of the manufacturing execution system and the monitoring data acquisition system. The traditional method relies on a fixed weight mode to process sensor signals, and cannot adjust the data attention according to real-time process events, resulting in a lag in abnormal response in critical production stages. For example, in an automobile welding production line, if the sudden fluctuation of the welding current is not identified and processed in time when the welding robot is working cooperatively, it may cause batch quality defects. Due to the lack of an event-driven dynamic weight mechanism, the existing system cannot prioritize the processing of the current signal at the beginning of welding, and the adjustment instruction cannot be fed back to the device control end in real time.

[0018] In order to solve the above problems, the inventors found that the pre-defined process event markers in the production log have the value of indicating the key stage. By analyzing the spatio-temporal correlation between event types and sensor signals, it is proposed to use event triggering as the basis for dynamically adjusting the signal weight. Further, a closed-loop feedback path based on the weighted results is designed, so that the analysis results can be directly converted into control instructions. For example, when an abnormal fluctuation of the welding current is detected within the event window, the system automatically increases the calculation weight of the signal segment, and triggers the welding parameter calibration according to the weighted similarity result.

[0019] Embodiment 1

[0020] As Figs. 1-2 shown in a process log feedback control-based industrial signal dynamic weighting method, comprising the following steps:

[0021] S100, real-time analysis of process event log of manufacturing execution system (MES) or supervisory control and data acquisition system (SCADA), extracting event type and associated signal segment;

[0022] S200, dynamically adjusting signal weight factor of corresponding process stage according to event type, wherein key event trigger stage is given higher weighting coefficient than baseline weight;

[0023] S300, based on the weighted signal similarity analysis result, triggering production parameter closed-loop adjustment instruction and feeding back to the control system.

[0024] Specifically, the process event log refers to the production process record data containing pre-defined stage markers, which can be implemented by using structured log files in XML or JSON format, each event is associated with a time stamp and corresponding sensor signal segment. The log analysis provides semantic event trigger basis for subsequent dynamic weighting.

[0025] Dynamic adjustment of signal weight factor refers to changing the contribution of sensor signal in the calculation model in a specific process stage, which can be implemented by using sliding time window mechanism, and the weight coefficient is improved in the window period after the key event trigger. This mechanism makes the signal change in the key stage be able to be identified and processed preferentially.

[0026] Production parameter closed-loop adjustment instruction refers to the device control command generated according to the weighted analysis result, which can be implemented by encapsulating the instruction content using OPC UA protocol and transmitting it to PLC controller. The instruction directly acts on the physical device execution end.

[0027] Specifically, when the monitoring system of the injection molding machine detects the "injection start" event, the pressure sensor signal segment within 60 seconds before and after the event timestamp is automatically extracted. The system increases the weight factor of the signal segment to the preset value, and the deviation of the signal segment will be amplified when calculating the pressure curve similarity of the injection stage. If the weighted similarity is lower than the set threshold, injection pressure compensation instruction is generated immediately, which is transmitted to the hydraulic control system of the injection molding machine through industrial bus, and the plunger advancing speed is adjusted in real time.

[0028] Compared with the prior art, the traditional method adopts a global fixed weight to process the sensor signal, and cannot distinguish the data value of the normal stage and the key process stage. The scheme can strengthen the monitoring sensitivity of the key signal segment in a specific time window through an event-driven dynamic weighting mechanism, and can solve the root problem of the coordination lag of the information layer and the control layer by establishing a direct link from data analysis to equipment control.

[0029] Through the above technical scheme, the application realizes dynamic allocation of signal weight based on real-time process events, so that abnormal fluctuations in key production stages can be identified first. By directly converting the weighted analysis result into a control instruction, a seamless connection between data perception and device execution is established, effectively improving the timeliness and accuracy of production parameter adjustment.

[0030] The application further proposes that the process event log includes pre-defined stage marker events, and the stage marker events at least include one of heating completion, injection start, and welding start, and each event is associated with a time stamp and a corresponding sensor signal segment.

[0031] The stage marker event refers to an event type with a clear process stage identification significance, which can be realized by device state triggering or process parameter threshold judgment, for example, triggering a heating completion event marker when the temperature of the injection molding machine barrel reaches a set value. This feature provides a reliable stage identification basis for dynamic weight adjustment by predefining process semantic labels.

[0032] The time stamp refers to the time information of the event occurrence, which can be realized by a millisecond-level time synchronization protocol, for example, the clock synchronization of PLC and sensor is realized by IEEE 1588 protocol. This feature ensures that the time sequence of the event and the production data strictly corresponds. The sensor signal segment refers to the original data set in a specific time range before and after the event trigger, which can be realized by a sliding window interception method, for example, the current waveform data of 5 seconds before and 10 seconds after the welding start event trigger is automatically extracted. This feature realizes the spatiotemporal binding of events and physical signals.

[0033] Specifically, in the injection molding production scene, when the injection screw advances to the preset position, the system automatically generates an injection start event marker, records the millisecond-level time stamp at that moment, and extracts the data segment of 30 seconds before and after the event trigger as the associated signal. On the automobile welding line, the welding robot contacts the workpiece at the moment to generate a welding start event, and synchronously collects the current and voltage curves 15 seconds before and after this moment. By predefining three types of typical process events, both the temperature control scene of process industry and the mechanical action monitoring demand of discrete manufacturing are met.

[0034] The binding mechanism of events and signal segments enables the subsequent dynamic weighting process to accurately lock the data range of the key process stage, for example, assigning a higher weight factor to the 60-second pressure data after the injection starts.

[0035] The traditional manufacturing execution system only records the device operation log, such as the basic state change of valve opening, motor starting, etc., and lacks event marking with process stage semantics. Although the existing data acquisition system stores sensor historical data, it does not establish the association between event trigger time and signal segment, which leads to inaccurate positioning of the data interval of the key process stage during analysis. The present scheme overcomes the dual defects of ambiguous event semantics and data positioning difficulty by constructing a process event system with space-time tags.

[0036] The present application realizes the accurate division of the process stage boundary, ensuring that the dynamic weight adjustment mechanism only runs within the effective process window. In the injection molding process, the injection start event accurately identifies the starting point of the molten plastic filling the mold, enabling the subsequent high-weight processing of injection pressure data to focus on the core period that affects product quality. In the automobile welding scene, the binding of the welding start event and the corresponding current waveform avoids misjudging abnormal signals as data interference in the non-welding stage.

[0037] The present application further proposes a technical scheme in which the weight factor β of the associated signal segment is increased to 1.5-2.5 within a preset time window after the triggering of a key event, and a baseline weight factor β=1.0 is used in the non-event stage.

[0038] Among them, the preset time window after the triggering of a key event refers to the signal intensification monitoring period set in advance according to the process type, which can be realized by the cooperative work of the event trigger module and the timer. This feature ensures that the weight increase operation matches the time characteristics of the key process period. The weight factor β refers to the contribution coefficient of the signal segment in the similarity calculation, which can be realized by using a dynamic weight distribution algorithm. This feature balances the signal sensitivity and system stability requirements through a gradient numerical range.

[0039] Specifically, when a predefined key event is detected, the system automatically starts the timer and applies a higher weight factor to the associated signal segment within the preset time window. For example, after detecting the "injection start" event in the injection molding process, the system increases the weight of the pressure sensor signal in the subsequent 60 seconds to 2.5 times. In the regular production stage without event triggering, the system maintains the baseline weight factor to avoid non-key signal interference with the analysis results. This mechanism realizes precise signal intensification in the process sensitive period through the cooperative action of event triggering and time constraints.

[0040] The traditional system adopts a global fixed weight factor, which cannot distinguish the importance of signals in key process stages and regular stages. Although the existing static segmented weight scheme divides different time periods, it lacks dynamic response capability driven by events. The scheme realizes real-time dynamic adjustment of signal weight through the combination of event triggering mechanism and adaptive time window, which improves the monitoring sensitivity of key stages and avoids the calculation burden caused by global high weight.

[0041] Through the above technical scheme, the application effectively improves the detection sensitivity of signal changes in key process stages, for example, finer current fluctuations can be identified within a 60-second window period after the start of welding. At the same time, signal interference in non-key stages is reduced, and false positives caused by irrelevant signals such as equipment idle vibration are prevented. The scheme also optimizes the allocation of computing resources, enabling high-weight calculation mode only during necessary periods, reducing the overall computational load of the system.

[0042] The application further proposes that the preset time window is adaptively set according to the process type. For the injection start event of injection molding process, the time window is 60 seconds; for the fermentation start event of food fermentation process, the time window is 4-6 hours.

[0043] Among them, the process type adaptive setting means automatically matching the time window parameters according to the physical property differences of the production process, which can be specifically realized by the linkage of the process parameter database and the real-time event type recognition module, and the corresponding window configuration is triggered by analyzing the process event marker. The preset time window refers to the high-weight signal acquisition period set for a specific process stage, which is specifically realized by an event-triggered timer module. The module calls the pre-stored time parameters to start the countdown according to the process type.

[0044] When the system identifies the injection start event in the injection molding process, a 60-second timing window is automatically activated, which covers the key physical change process of the polymer material filling stage and the initial stage of pressure maintenance. In the food fermentation scenario, the fermentation start event triggers a 4-6 hour monitoring period, which matches the key stage of the transition of microbial metabolic activity from the logarithmic growth phase to the stable phase. The adaptive switching of the time window is completed by the process type recognition module, which matches the event type with the physical property data in the process knowledge base and dynamically loads the corresponding time parameter configuration.

[0045] The application further proposes generating an MES work order adjustment instruction when the weighted signal similarity is lower than a threshold value, the adjustment instruction including at least one of temperature deviation compensation, mechanical parameter calibration, or production rhythm adjustment.

[0046] The weighted signal similarity being lower than a threshold value means that the matching degree threshold value of the dynamically set process parameters and the standard model is matched, which can be specifically realized by a process adaptive algorithm for quantitative judgment of abnormal production state.

[0047] MES work order adjustment instruction refers to the standardized control command generated by the manufacturing execution system, which can be implemented in XML format data packet, and is used to transmit parameter correction information to the device control system.

[0048] Temperature deviation compensation refers to the proportional adjustment of the output power of the thermal system, which can be implemented by PID control algorithm, and is used to eliminate the difference between the temperature monitoring value and the set value.

[0049] Mechanical parameter calibration refers to the correction of the motion trajectory or force of the actuator, which can be implemented by servo motor position feedback mechanism, and is used to eliminate the device execution error.

[0050] Production rhythm adjustment refers to the dynamic optimization of process time interval, which can be implemented by beat controller, and is used to balance the device load and process demand.

[0051] Specifically, when the real-time monitoring of the weighted signal similarity is lower than the preset threshold, the system automatically triggers the instruction generation module. The module selects the corresponding adjustment strategy according to the abnormal type: for temperature related abnormality, generate an instruction containing heater power adjustment parameter, for mechanical positioning deviation, generate an instruction containing servo motor compensation parameter, for process connection, generate an instruction containing beat optimization parameter. The instruction is transmitted to the device control unit through the MES system to execute parameter update, forming a complete closed loop from abnormality detection to parameter correction.

[0052] The traditional method relies on manual experience to judge the abnormal type and manually input the adjustment parameter, which has response lag and operation error. This scheme realizes automatic identification of abnormality through quantitative threshold triggering mechanism, combines with standardized instruction format to ensure the accuracy of parameter adjustment, supports parallel execution of multiple types of adjustment strategies, and can cope with complex production abnormalities.

[0053] The application further applies the dynamic weighting method to multi-device collaborative scenarios, including multi-robot collaborative control of automobile welding line or temperature and humidity closed-loop control of food fermentation process.

[0054] Multi-device collaborative scenario refers to a production environment in which multiple industrial devices perform associated operations in a unified production process, which can be implemented by multi-robot collaborative work of multiple welding robots in automobile welding line or linkage of temperature and humidity control devices in food fermentation process. This feature solves the cross-device signal interference problem by establishing a dynamic weight distribution relationship between devices.

[0055] Dynamic weighting method refers to the operation mechanism of adjusting the signal weight according to the process event, which can be implemented by analyzing the stage marker event in the manufacturing execution system and associating the corresponding device signal segment. This feature enhances the signal influence of key devices and improves the collaborative control accuracy.

[0056] Specifically, in the multi-robot collaborative control of the automobile welding line, when the welding start event is detected, the system automatically raises the signal weight factor of the current welding robot, while reducing the weight factor of other auxiliary equipment. By focusing on the signal analysis of key equipment, the precise matching of the action timing of multiple robots is realized. In the temperature and humidity closed-loop control of the food fermentation process, when the fermentation start event is triggered, the system continuously strengthens the signal segment of the temperature and humidity sensor, and generates a collaborative adjustment instruction by jointly analyzing the weighted data, so that the temperature and humidity parameters remain in dynamic balance during the active peak period.

[0057] The application further proposes the generation logic of the MES work order adjustment instruction, which includes: if a temperature negative deviation is detected, the instruction contains proportionally increasing the heater power output; if abnormal welding current fluctuation is detected, the instruction contains adjusting the welding time or suspending the production line.

[0058] Wherein, the temperature negative deviation refers to the difference between the actual temperature and the set value, which can be specifically collected by a thermocouple or an infrared sensor in real time, and the deviation amount is determined by a difference calculation module. This feature enables the system to quantify the temperature deviation, providing a calculation benchmark for power compensation.

[0059] Abnormal welding current fluctuation refers to the case where the current value exceeds the preset stable interval, which can be specifically monitored in real time by a Hall sensor combined with a sliding window variance algorithm. This feature can identify different levels of current fluctuation events, providing a basis for hierarchical response. Proportionally increasing the heater power output refers to dynamically adjusting the power compensation amplitude according to the temperature deviation, which can be specifically converted into a power adjustment coefficient by a linear proportional algorithm.

[0060] This feature avoids the temperature overshoot phenomenon caused by traditional on-off control. Adjusting the welding time refers to modifying the action timing of the welding robot according to the current fluctuation amplitude, which can be specifically implemented by dynamically rewriting the welding process parameters through a PLC program. This feature realizes online optimization of process parameters, reducing the number of shutdowns. Suspending the production line refers to triggering a safety interlock mechanism to cut off the power source of the equipment, which can be specifically implemented by executing an emergency stop operation through a safety relay module. This feature forcibly interrupts production in the event of a serious anomaly, preventing equipment damage.

[0061] Specifically, when the temperature monitoring module identifies that the actual temperature continuously falls below the set value, the deviation calculation unit converts the negative temperature difference into a power compensation coefficient, which is linked to the heater control module to linearly increase the power output with the amount of deviation. For example, in the injection molding process, when the mold temperature has a negative deviation, the system automatically generates a work order instruction containing the proportion of heater power increase, which directly acts on the temperature control unit. For welding current abnormalities, the signal analysis unit first determines whether the fluctuation amplitude exceeds the first threshold value. If it is in the range of slight fluctuation, a welding time adjustment instruction is generated to compensate for the current change by extending or shortening the welding cycle. If the fluctuation amplitude exceeds the second threshold value, a production line suspension instruction is generated to trigger the safety protection mechanism. For example, on an automobile welding line, when the spot welding gun current fluctuates violently, the system adjusts the welding robot dwell time first. If the adjustment still cannot be stabilized, the production at the station is immediately suspended.

[0062] The traditional scheme only triggers the start-stop operation of the heater when the temperature is abnormal, resulting in repeated temperature oscillation. The present scheme achieves smooth compensation through proportional adjustment, eliminating the control lag effect. The existing technology adopts a unified alarm mechanism for welding current abnormalities, which cannot distinguish between slight fluctuations and serious faults. The present scheme establishes a hierarchical response strategy, preferentially performing parameter fine-tuning and executing production line suspension only when necessary, significantly reducing unplanned downtime. The existing technology relies on human experience to determine the adjustment amplitude. The present scheme converts process knowledge into automated instruction generation rules, improving response speed and consistency.

[0063] Through the above technical scheme, the present application can automatically match the heating power compensation amount according to the temperature deviation degree, avoiding the problem of temperature overshoot or insufficient adjustment caused by traditional on-off control. For welding current abnormalities, the system automatically distinguishes between slight fluctuations and serious faults, and takes parameter adjustment or emergency shutdown measures respectively, maximizing the continuity of production while ensuring the safety of equipment. The present scheme realizes a fully automatic closed loop from abnormal detection to control execution, eliminating the delay of manual intervention and ensuring the timeliness and accuracy of process parameter adjustment.

[0064] The present application further proposes a cumulative error analysis mechanism, including time decay weighting of continuous process signals, in which the weight of recent signals is higher than that of historical signals, and the cumulative error is calculated through a weighted moving average algorithm, which triggers predictive maintenance when the error exceeds the threshold.

[0065] Time decay weighting refers to a data processing method that assigns different weight coefficients to continuous process signals according to their time proximity. Specifically, an exponential decay function can be used to achieve this, for example, setting the weight coefficient to decrease exponentially with the collection time interval. This way, the trend of the current production state dominates in error analysis, effectively reducing the interference of historical data.

[0066] The weighted moving average algorithm refers to a method of dynamically accumulating errors of signals weighted by time attenuation, and can be specifically realized by weighted summation in a sliding window. The algorithm accurately reflects the overall trend and local fluctuation characteristics of error accumulation by dynamically adjusting the contribution of each time node.

[0067] The predictive maintenance triggering mechanism refers to a control logic that starts the maintenance process according to the comparison result of the accumulated error calculation result and the preset threshold, and can be specifically realized by the linkage of a threshold comparator and an instruction generation module. The mechanism converts the error quantization result into a maintenance action instruction, forming a closed-loop control from monitoring to execution.

[0068] Specifically, in a continuous production process, process signals are collected in real time by sensors and input into the error analysis module. After the signals enter the processing flow, they are first weighted by time attenuation according to their generation time, for example, the weight of the current time signal is set to the highest, and the weight of the previous time node signal is decreased by a preset attenuation coefficient.

[0069] The weighted signal sequence is input into the weighted moving average algorithm to calculate a dynamic index reflecting the degree of error accumulation. When the index exceeds the preset safety threshold, the system automatically generates a maintenance instruction and transmits it to the execution mechanism, such as triggering the device calibration program or sending a maintenance work order. Through the synergistic effect of time attenuation weighting and moving average algorithm, the system can identify small errors with a tendency to accumulate continuously and start maintenance measures before they reach the critical value.

[0070] The present application further proposes that predictive maintenance includes adaptive calibration: if the accumulated error is caused by thermal deformation, adjust the mechanical compensation parameters; if the accumulated error is caused by light source stability, dynamically correct the exposure time or intensity.

[0071] Thermal deformation refers to the structural deformation of a device due to temperature changes, which can be specifically calculated by a thermal expansion coefficient model to generate a mechanical parameter adjustment amount through a reverse compensation algorithm, thereby offsetting the impact of temperature changes on device accuracy. Light source stability refers to the fluctuation degree of the output light intensity of an optical device, which can be specifically monitored by a light intensity sensor in real time to fit the attenuation trend by an exponential function and dynamically adjust the exposure parameters, thereby maintaining the energy consistency of the lithography process.

[0072] Specifically, when detecting that the accumulated error mainly comes from thermal deformation, the system automatically calls the pre-stored thermal expansion coefficient model, calculates the corresponding mechanical deformation according to the real-time temperature change, generates compensation parameters and sends them to the actuator for reverse displacement compensation. For example, in a semiconductor lithography machine, when the cavity temperature rises and causes the lens group to expand, the compensation algorithm drives the fine adjustment mechanism to correct the focus plane position in real time. When detecting that the error comes from light source attenuation, the system obtains the current light source output intensity through the light intensity sensor, combines the historical attenuation curve to predict the next stage light intensity value, dynamically increases the exposure time or improves the driving current to maintain the total exposure energy constant.

[0073] In some embodiments, thermal deformation compensation can be achieved through embedded temperature sensors and piezoelectric ceramic actuators. Each degree Celsius of temperature change triggers a corresponding micron-level displacement compensation. Light source correction can use a light intensity closed-loop control system to collect light source intensity data every five minutes and update the exposure parameter table.

[0074] Traditional maintenance strategies use a unified calibration method for all error sources, which can easily cause the compensation measures to be mismatched with the causes of failure. For example, adjusting the exposure parameters for thermal deformation errors will exacerbate wafer distortion, and stopping to adjust mechanical parameters when the light source attenuates will cause production loss. The present scheme distinguishes the physical causes and matches the corresponding compensation mechanism to eliminate the error generation path from the root.

[0075] The present application realizes accurate compensation for different error sources, avoids secondary errors caused by misoperation, and reduces unnecessary downtime maintenance. In the semiconductor lithography scene, this scheme can simultaneously maintain mechanical precision and exposure energy stability, ensuring the consistency of wafer alignment accuracy and pattern transfer quality.

[0076] The present application further proposes that the accumulated error analysis mechanism be applied to semiconductor lithography processes, real-time wafer production calibration instructions are generated through continuous monitoring of alignment errors and exposure intensity, and calibration effect data is fed back to the learning model to optimize the calibration algorithm.

[0077] The alignment error refers to the positional deviation between the mechanical system of the lithography machine and the wafer positioning reference, which can be measured in real time by a nanometer-level laser interferometer to identify the mechanical deviation caused by thermal deformation or vibration. The exposure intensity refers to the energy density output by the light source of the lithography machine during the exposure process, which can be continuously sampled by a photoelectric sensor array to detect the optical performance changes caused by light source attenuation or fluctuation. The calibration instruction generation refers to the operation instruction for dynamically adjusting the process parameters according to the monitoring data, which can be implemented by using a fuzzy control algorithm combined with a preset compensation rule to eliminate the real-time detected process deviation. The learning model optimization refers to the process of updating the algorithm parameters based on historical calibration effect data, which can be implemented by using a recursive least squares method for online parameter estimation to improve the adaptability and accuracy of the calibration strategy.

[0078] Specifically, during the execution of the semiconductor lithography process, the alignment error and exposure intensity data are synchronously collected by high-precision sensors to capture the process fluctuations caused by mechanical deformation and light source attenuation in real time. When the detected alignment error exceeds the preset tolerance range, a calibration instruction containing mechanical compensation parameters is immediately generated, for example, adjusting the fine adjustment mechanism of the lithography machine platform; when the monitored exposure intensity deviates from the target value, a light source power correction instruction is generated. After each calibration operation is completed, the actual calibration effect data is input to the learning model, and the compensation coefficients are dynamically updated by an online parameter optimization algorithm, so that the subsequent calibration instructions can adapt to the gradual changes in device performance.

[0079] Compared with the prior art, the traditional semiconductor lithography calibration method usually only compensates for a single error source independently, for example, only adjusts the mechanical parameters according to the temperature change, without considering the coupling effect of optical system attenuation on process precision. The fixed compensation coefficients used in the prior art cannot adapt to the performance degradation of the device during long-term operation, and lack a closed-loop verification mechanism for calibration effect, resulting in hysteresis and incompleteness of error compensation.

[0080] Through the above technical solutions, the present application effectively solves the problem of insufficient calibration accuracy caused by multi-source error coupling in the lithography process, and realizes the collaborative compensation of mechanical deformation and optical attenuation. Through real-time feedback of calibration effect data and learning model optimization, a dynamically adaptive calibration strategy is formed, which significantly reduces the error transmission risk of cross-batch production and ensures the sustained stability of wafer processing precision.

[0081] The above-described embodiments are merely descriptions of the preferred embodiments of the present application and do not limit the concept and scope of the present application. Various modifications and improvements to the technical solutions of the present application made by those of ordinary skill in the art without departing from the design concept of the present application shall fall within the protection scope of the present application. The technical content claimed by the present application has been fully recorded in the claims.

Claims

1. A dynamic weighting method for industrial signals based on process log feedback control, characterized in that: The industrial signal dynamic weighting method based on process log feedback control includes: S100, parsing the process event log of the manufacturing execution system (MES) or the supervisory control and data acquisition system (SCADA) in real time to extract the event type and associated signal segments; S200, dynamically adjusting the signal weight factor of the corresponding process stage according to the event type, wherein the key event triggering stage is assigned a weighting coefficient higher than the baseline weight; S300 , based on the weighted signal similarity analysis result, triggering a closed-loop adjustment instruction for production parameters and feeding it back to the control system.

2. The industrial signal dynamic weighting method based on process log feedback control according to claim 1 is characterized in that: The process event log includes predefined stage marking events, which include at least one of heating completion, injection start, and welding start, and each event is associated with a timestamp and a corresponding sensor signal segment.

3. The industrial signal dynamic weighting method based on process log feedback control according to claim 1 is characterized in that: The dynamic adjustment of the signal weight factor specifically includes: In the preset time window after the key event is triggered, the weight factor β of the associated signal segment is increased to 1.5-2.5; During the event-free period, a baseline weighting factor β = 1.0 was used.

4. The industrial signal dynamic weighting method based on process log feedback control according to claim 3 is characterized in that: The preset time window is adaptively set according to the process type: For the injection start event of the injection molding process, the time window is 60 seconds; For the fermentation start event of food fermentation process, the time window is 4-6 hours.

5. The industrial signal dynamic weighting method based on process log feedback control according to claim 1 is characterized in that: The triggering of the closed-loop adjustment instruction of the production parameters includes: When the weighted signal similarity is lower than the threshold, an MES work order adjustment instruction is generated; The adjustment instruction includes at least one of temperature deviation compensation, mechanical parameter calibration or production rhythm adjustment.

6. The industrial signal dynamic weighting method based on process log feedback control according to claim 1 is characterized in that: The method is applied to multi-device collaboration scenarios, including multi-robot collaborative control of automobile welding lines or temperature and humidity closed-loop control of food fermentation processes.

7. The industrial signal dynamic weighting method based on process log feedback control according to claim 5 is characterized in that: The generation logic of the MES work order adjustment instruction includes: If a negative temperature deviation is detected, the command includes proportionally increasing the heater power output; If abnormal fluctuations in welding current are detected, the instructions include adjusting the welding time or pausing the production line.

8. The industrial signal dynamic weighting method based on process log feedback control according to claim 1 is characterized in that: Also includes a cumulative error analysis mechanism: Perform time-attenuation weighting on continuous process signals, where the weight of recent signals is higher than that of historical signals; The accumulated error is calculated using a weighted moving average algorithm, and predictive maintenance is triggered when the error exceeds a threshold.

9. The industrial signal dynamic weighting method based on process log feedback control according to claim 8, characterized in that: The predictive maintenance includes adaptive calibration: If the cumulative error is caused by thermal deformation, adjust the mechanical compensation parameters; If the accumulated error is due to light source stability, the exposure time or intensity is dynamically corrected.

10. The industrial signal dynamic weighting method based on process log feedback control according to claim 8, characterized in that: The cumulative error analysis mechanism is applied to semiconductor lithography process: Generate wafer production calibration instructions in real time through continuous monitoring of alignment error and exposure intensity; The calibration effect data is fed back to the learning model to optimize the calibration algorithm.

Images