A process waiting window compression method and system

Abstract

This invention relates to the field of automation technology, specifically to a method and system for compressing process waiting windows. A process waiting window compression system includes: a process data acquisition module, a workpiece state temporal evolution analysis module, an inverse solution module for the entry allowable state, and a process waiting window compression module. The invention predicts the future state of the workpiece during the waiting phase and, combined with the actual equipment state and output quality requirements of the second process, inversely calculates the entry allowable state domain. The matching result of these two parameters is mapped to the time dimension, filtering out invalid time segments in the original waiting window. This results in a compressed waiting window that adapts to the actual evolution of the workpiece state and the processing conditions of the second process, thereby reducing invalid waiting and improving the rationality of workpiece transition timing.

Description

Technical Field

[0001] This invention relates to the field of automation technology, and specifically to a method and system for compressing process waiting windows. Background Technology

[0002] In intelligent manufacturing production lines, workpieces typically need to go through multiple processes sequentially according to a preset process route. Between adjacent processes, there is usually a certain waiting time due to factors such as equipment occupancy, production cycle time, conveying process, and process requirements. For some production processes, the physical state of the workpiece continues to change over the waiting time after completing the previous process. For example, the workpiece's temperature, humidity, viscosity, moisture content, or surface activity will gradually change due to the waiting environment. Therefore, the time it takes for the workpiece to enter the next process not only affects the production cycle time but also further impacts the processing results and product quality of the subsequent process.

[0003] Existing production control methods typically pre-set waiting windows between adjacent processes based on process specifications or historical production experience. They determine the workpiece transfer time based on the completion time of the previous process, the preset waiting time limit, and the production plan of the next process. Some production scheduling methods also adjust the time for the workpiece to enter the next process by considering information such as whether the equipment in the next process is idle, the priority of production tasks, and the production line load. However, these methods mainly use the waiting window as a pre-given time constraint, focusing on the time the workpiece has been waiting and whether the next process has the conditions to receive it at the scheduling level. They fail to fully reflect the state evolution of the workpiece over time in the actual waiting environment. Therefore, even if the workpiece is still within the preset waiting window at a certain moment, its actual state or the state it may reach in the future may no longer be suitable for entering the next process. Summary of the Invention

[0004] This invention acquires workpiece state data, environmental state data, and equipment state data corresponding to the second process under a unified state sampling time benchmark. This establishes a time correspondence between the workpiece-side state, environmental-side state, and equipment-side state during the workpiece waiting process, providing a real-time data foundation for the dynamic determination of the process waiting window. By combining workpiece state data, environmental state data, and a workpiece state evolution model, the invention performs time-series evolution analysis on the workpiece state at each analysis time point within the original waiting window, determining the future workpiece state domain corresponding to each analysis time point. This transforms the determination of the process waiting time from simply relying on a preset waiting time limit to considering the actual state evolution of the workpiece during the waiting process. Furthermore, by representing the future workpiece state with a state domain, the invention reduces the bias caused by using a single predicted state to determine the process entry timing. By combining the current equipment state data of the second process, the quality mapping model, and the output quality constraint data, the invention reversely determines the entry allowable state domain, converting the quality requirements after the second process into constraints on the entry workpiece state under the current equipment operating state. This allows the workpiece entry conditions to be dynamically adjusted. The process is adjusted according to the actual processing status changes in the second process. By matching the future workpiece state domain corresponding to each analysis time point with the allowed entry state domain, and determining the effective entry time set based on the matching results, a correspondence is established between the range of possible future states that the workpiece to be processed may reach and the range of entry states that the second process can currently accept. This eliminates time positions in the original waiting window where the workpiece state does not match the entry conditions of the second process. By forming a compressed waiting window based on the time continuity relationship in the effective entry time set, the time intervals in the original waiting window that do not meet the state domain matching requirements are eliminated, while the continuous effective time intervals that meet the entry conditions of the second process are retained. The newly determined compressed waiting window is used to control the entry of the workpiece to be processed into the second process, realizing the reverse determination from the workpiece state conditions and the processing quality conditions of the second process to the process entry time. On the basis of ensuring that the workpiece to be processed is compatible with the actual processing conditions of the second process when it enters the second process, the invalid waiting time is compressed, improving the adaptability between the workpiece transfer timing and the actual production state in the intelligent manufacturing production line.

[0005] This invention provides a method for compressing process waiting windows, applied to intelligent manufacturing production lines. Workpieces to be processed sequentially undergo a first process and a second process according to a preset process route. The period between the completion time of the first process and the start time of the second process forms a process waiting phase. The method includes the following steps:

[0006] The original waiting window corresponding to the process waiting stage between the completion time of the first process and the start time of the second process is discretized according to the preset state sampling period to obtain several state sampling times. The completion time of the first process is taken as the first state sampling time. At the current state sampling time, the workpiece state data of the workpiece to be processed, the environmental state data corresponding to the environment of the workpiece to be processed, and the equipment state data corresponding to the second process are obtained. The state sampling times after the current state sampling time and within the original waiting window are recorded as the analysis time points.

[0007] Based on the workpiece state data, environmental state data, and workpiece state evolution model corresponding to the current state sampling time, a time-series evolution analysis of the workpiece state at the analysis time point is performed to determine the future workpiece state domain corresponding to the analysis time point.

[0008] Obtain the output quality constraint data corresponding to the second process. Based on the equipment state data, quality mapping model and output quality constraint data corresponding to the current state sampling time, perform reverse solution on the workpiece state when entering the second process to determine the entry allowable state domain corresponding to the second process.

[0009] The future workpiece state domain corresponding to each analysis time point is matched with the entry allowed state domain. The effective entry time set is determined based on the state domain matching result. The time interval of the original waiting window that is later than the current state sampling time is compressed based on the effective entry time set to obtain the compressed waiting window corresponding to the current state sampling time. The workpiece to be processed is controlled to enter the second process according to the newly determined compressed waiting window.

[0010] As a preferred aspect, based on the workpiece state data, environmental state data, and workpiece state evolution model corresponding to the current state sampling time, a time-series evolution analysis is performed on the workpiece state at the analysis time point to determine the future workpiece state domain corresponding to the analysis time point. This specifically includes the following steps:

[0011] The workpiece state data and environmental state data corresponding to the current state sampling time are concatenated to form an analysis vector. Then, the current analysis vector and the previous N-1 analysis vectors are combined to form a time series analysis sequence.

[0012] The process iterates through all analysis time points. For each selected analysis time point, the prediction span is determined based on the number of state sampling times between the selected analysis time point and the current state sampling time. The workpiece state prediction sub-model corresponding to the prediction span is then obtained. The time series analysis sequence is fed into the workpiece state prediction sub-model corresponding to the prediction span for processing, and the baseline predicted workpiece state data corresponding to the analysis time point is output. This process continues until all analysis time points have been traversed, and the baseline predicted workpiece state data corresponding to each analysis time point is output. The workpiece state evolution model consists of workpiece state prediction sub-models corresponding to different prediction spans.

[0013] Acquire several benchmark predicted workpiece state data corresponding to the current state sampling time. Iterate through the workpiece state parameters corresponding to the workpiece state data. For the selected workpiece state parameter, calculate the difference between the data corresponding to the selected workpiece state parameter in the benchmark predicted workpiece state data and the data corresponding to the selected workpiece state parameter in the workpiece state data corresponding to the current state sampling time. Record this difference as the prediction deviation value. Then, perform Gaussian fitting operation based on all prediction deviation values ​​to obtain the Gaussian fitting parameter vector corresponding to the workpiece state parameter. The Gaussian fitting parameter vector includes the average prediction deviation and the standard deviation of the prediction deviation. Continue until all workpiece state parameters corresponding to the workpiece state data have been traversed. Output the Gaussian fitting parameter vector corresponding to all workpiece state parameters.

[0014] The workpiece state parameters in the baseline predicted workpiece state data corresponding to the analysis time point are processed by the Gaussian fitting parameter vector corresponding to the workpiece state parameters to obtain the predicted value interval of the workpiece state parameters corresponding to the analysis time point. Then, the predicted value intervals of all workpiece state parameters are combined to form the future workpiece state domain corresponding to the analysis time point.

[0015] As a preferred aspect, based on the equipment state data, quality mapping model, and output quality constraint data corresponding to the current state sampling time, the workpiece state upon entering the second process is solved in reverse to determine the entry allowable state domain corresponding to the second process. This specifically includes the following steps:

[0016] The inputs to the quality mapping model include the workpiece state data when entering the second process and the equipment state data corresponding to the second process. The output of the quality mapping model is the predicted output quality data after the second process is completed. The order of the data in the predicted output quality data corresponds to the order of the data in the output quality constraint data.

[0017] According to the preset state discrete interval, the input range corresponding to each workpiece state parameter in the workpiece state data is discretized, and the discrete values ​​corresponding to each workpiece state parameter are combined to obtain several candidate entry workpiece state data.

[0018] Arrange the discrete values ​​corresponding to each workpiece state parameter from largest to smallest. Among all candidate entry workpiece state data, if two candidate entry workpiece state data have only one different workpiece state parameter, and the different data are consecutive in the order of the discrete values ​​of the workpiece state parameters after arrangement, the two candidate entry workpiece state data are determined as adjacent candidate entry workpiece state data.

[0019] The process iterates through the candidate entry workpiece state data. For each selected candidate entry workpiece state data, it is input into the quality mapping model along with the equipment state data corresponding to the current state sampling time. This yields the predicted output quality data corresponding to the selected candidate entry workpiece state data. If all workpiece state parameters in the predicted output quality data meet the output quality constraints, the selected candidate entry workpiece state data is determined as qualified entry workpiece state data. This process continues until all candidate entry workpiece state data has been traversed, and then all qualified entry workpiece state data are combined into a qualified entry workpiece state data set.

[0020] Traverse adjacent candidate entry workpiece state data. If adjacent candidate entry workpiece state data belong to the qualified entry workpiece state data set and not to the qualified entry workpiece state data set respectively, record the workpiece state parameters corresponding to the different data in the adjacent candidate entry workpiece state data as state parameters to be refined. If the difference between the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data is less than the state boundary precision, no operation is performed. If the difference between the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data is not less than the state boundary precision, record the average value of the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data as refined data. Then replace the data corresponding to the state parameters to be refined in the candidate entry workpiece state data with the refined data to form new candidate entry workpiece state data. Continue until the adjacent candidate entry workpiece state data has been traversed. Then continue to analyze whether all new candidate entry workpiece state data are qualified entry workpiece state data and update the qualified entry workpiece state data set.

[0021] Until there are no adjacent candidate entry workpiece state data whose difference between the data corresponding to the state parameters to be refined is not less than the state boundary accuracy, the set of qualified entry workpiece state data is determined as the entry allowable state domain corresponding to the second process.

[0022] As a preferred aspect, the future workpiece state domain corresponding to each analysis time point is matched with the entry allowable state domain. Based on the state domain matching results, a valid entry time set is determined. Then, based on the valid entry time set, the time intervals in the original waiting window that are later than the current state sampling time are compressed to obtain the compressed waiting window corresponding to the current state sampling time. Specifically, the steps include the following:

[0023] For each analysis time point, the future workpiece state domain corresponding to the selected analysis time point is obtained. The predicted value intervals corresponding to each workpiece state parameter in the future workpiece state domain are discretized according to the preset state discretization interval. Then, the discrete values ​​corresponding to each workpiece state parameter are combined to obtain several sets of fitted future workpiece state data corresponding to the selected analysis time point.

[0024] For each analysis time point, all fitted future workpiece state data corresponding to that time point are traversed. For a selected set of fitted future workpiece state data, qualified entry workpiece state data corresponding to the selected set of fitted future workpiece state data are searched in the entry allowed state domain. If the difference between the data corresponding to each workpiece state parameter in the selected set of fitted future workpiece state data and the data corresponding to the same workpiece state parameter in the qualified entry workpiece state data is less than the state boundary accuracy, the selected set of fitted future workpiece state data and the entry allowed state domain satisfy the state domain matching. This process continues until all fitted future workpiece state data corresponding to the analysis time point have been traversed. If all fitted future workpiece state data and the entry allowed state domain satisfy the state domain matching, the analysis time point is determined as a valid entry time. If there are fitted future workpiece state data and the entry allowed state domain that do not satisfy the state domain matching, the analysis time point is not determined as a valid entry time.

[0025] All valid entry times are arranged in chronological order to form a set of valid entry times. If the time difference between two adjacent valid entry times in the set of valid entry times is equal to the preset state sampling period, the two valid entry times are assigned to the same continuous valid time group.

[0026] For consecutive valid time groups, a compressed waiting sub-window is constructed based on the valid entry times on the left and right sides of the consecutive valid time group;

[0027] All compressed waiting sub-windows are combined into a compressed waiting window corresponding to the current state sampling time.

[0028] As a preferred approach, the workpiece state evolution model is constructed through the following steps:

[0029] Several prediction spans are set according to the number of state sampling times within the original waiting window, and a workpiece state prediction sub-model is constructed for each prediction span;

[0030] For each prediction span, select the workpiece state prediction sub-model corresponding to the prediction span, obtain workpiece state prediction training samples. The workpiece state prediction training samples include N analysis vectors arranged in time. The number of state sampling times between the state sampling time corresponding to the last analysis vector in the workpiece state prediction training samples is used as the workpiece state data corresponding to the state sampling time of the prediction span. The workpiece state prediction training samples are labeled, and all labeled workpiece state prediction training samples are combined into a workpiece state prediction training set. The workpiece state prediction sub-model is trained using the workpiece state prediction training set.

[0031] All the workpiece state prediction sub-models corresponding to the prediction span are combined into a workpiece state evolution model.

[0032] As a preferred approach, the quality mapping model is trained in the following manner:

[0033] Obtain several quality mapping training samples, which include workpiece status data when the workpiece is sent into the process and equipment status data corresponding to the process. Label the quality mapping training samples by outputting quality data, and form a quality mapping training set by combining all labeled quality mapping training samples. Train the quality mapping model using the quality mapping training set.

[0034] The present invention also provides a process waiting window compression system, comprising:

[0035] The process data acquisition module is used to discretize the original waiting window corresponding to the process waiting stage between the completion time of the first process and the start time of the second process according to the preset state sampling period to obtain a number of state sampling times. The completion time of the first process is used as the first state sampling time. At the current state sampling time, the workpiece state data of the workpiece to be processed, the environmental state data corresponding to the environment where the workpiece to be processed is located, and the equipment state data corresponding to the second process are acquired. The state sampling times after the current state sampling time and within the original waiting window are recorded as the analysis time points.

[0036] The workpiece state temporal evolution analysis module is used to perform temporal evolution analysis on the workpiece state at the analysis time point based on the workpiece state data, environmental state data and workpiece state evolution model corresponding to the current state sampling time, and to determine the future workpiece state domain corresponding to the analysis time point.

[0037] The inverse solution module for the entry permission state is used to obtain the output quality constraint data corresponding to the second process. Based on the equipment status data, quality mapping model and output quality constraint data corresponding to the current state sampling time, it performs inverse solution on the workpiece state when entering the second process to determine the entry permission state domain corresponding to the second process.

[0038] The process waiting window compression module is used to match the future workpiece state domain corresponding to each analysis time point with the entry allowed state domain. Based on the state domain matching result, the effective entry time set is determined. Based on the effective entry time set, the time interval of the original waiting window that is later than the current state sampling time is compressed to obtain the compressed waiting window corresponding to the current state sampling time. The workpiece to be processed is controlled to enter the second process according to the newly determined compressed waiting window.

[0039] The present invention has the following advantages:

[0040] This invention acquires workpiece state data, environmental state data, and equipment state data corresponding to the second process under a unified state sampling time benchmark. This establishes a time correspondence between the workpiece-side state, environmental-side state, and equipment-side state during the workpiece waiting process, providing a real-time data foundation for the dynamic determination of the process waiting window. By combining workpiece state data, environmental state data, and a workpiece state evolution model, the invention performs time-series evolution analysis on the workpiece state at each analysis time point within the original waiting window, determining the future workpiece state domain corresponding to each analysis time point. This transforms the determination of the process waiting time from simply relying on a preset waiting time limit to considering the actual state evolution of the workpiece during the waiting process. Furthermore, by representing the future workpiece state with a state domain, the invention reduces the bias caused by using a single predicted state to determine the process entry timing. By combining the current equipment state data of the second process, the quality mapping model, and the output quality constraint data, the invention reversely determines the entry allowable state domain, converting the quality requirements after the second process into constraints on the entry workpiece state under the current equipment operating state. This allows the workpiece entry conditions to be dynamically adjusted. The process is adjusted according to the actual processing status changes in the second process. By matching the future workpiece state domain corresponding to each analysis time point with the allowed entry state domain, and determining the effective entry time set based on the matching results, a correspondence is established between the range of possible future states that the workpiece to be processed may reach and the range of entry states that the second process can currently accept. This eliminates time positions in the original waiting window where the workpiece state does not match the entry conditions of the second process. By forming a compressed waiting window based on the time continuity relationship in the effective entry time set, the time intervals in the original waiting window that do not meet the state domain matching requirements are eliminated, while the continuous effective time intervals that meet the entry conditions of the second process are retained. The newly determined compressed waiting window is used to control the entry of the workpiece to be processed into the second process, realizing the reverse determination from the workpiece state conditions and the processing quality conditions of the second process to the process entry time. On the basis of ensuring that the workpiece to be processed is compatible with the actual processing conditions of the second process when it enters the second process, the invalid waiting time is compressed, improving the adaptability between the workpiece transfer timing and the actual production state in the intelligent manufacturing production line. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the process waiting window compression system used in an embodiment of the present invention. Detailed Implementation

[0042] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0043] Example 1: A process waiting window compression method is applied to an intelligent manufacturing production line. The workpiece to be processed sequentially performs a first process and a second process according to a preset process route. The time between the completion of the first process and the start of the second process forms a process waiting phase. The method includes the following steps:

[0044] The original waiting window corresponding to the process waiting stage between the completion time of the first process and the start time of the second process is discretized according to a preset state sampling period to obtain several state sampling times. The preset state sampling period is used to unify the time benchmark for actual state acquisition and future state analysis during the process waiting stage, and the completion time of the first process is used as the first state sampling time. At the current state sampling time, the workpiece state data, the environmental state data corresponding to the environment of the workpiece, and the equipment state data corresponding to the second process are acquired. The state sampling times after the current state sampling time and within the original waiting window are recorded as the analysis time points. The workpiece state data is used to characterize the actual physical state of the workpiece at the corresponding state sampling time. The workpiece state data comes from the detection devices arranged at the exit of the first process, the buffer station, or the workpiece conveying path. The data form is a state vector composed of multiple workpiece state parameters. The workpiece state parameters are determined according to the workpiece characteristics that change between the first and second processes and affect the processing result of the second process. In the production scenario where the workpiece enters the drying process after spraying, the workpiece state parameters include coating surface temperature, coating humidity, and coating viscosity. Environmental state data is used to characterize the environmental conditions that affect the workpiece's state evolution during the process waiting stage. The environmental state data comes from environmental sensors set up in the buffer station or workpiece conveying area, and the data format is an environmental state vector composed of ambient temperature, ambient humidity, and airflow velocity. Equipment state data is used to characterize the actual operating state of the equipment corresponding to the second process at the corresponding state sampling time. The equipment state data comes from the second process equipment controller, PLC, and equipment sensors, and the data format is an equipment state vector composed of equipment temperature, equipment speed, actuator status, and power status, such as the actual temperature, actual wind speed, conveying speed, and cavity pressure of the drying equipment. It should be noted that the original waiting window comes from the process route configuration data or production process database corresponding to the intelligent manufacturing production line, and the data format is a time interval composed of the start time and the end time with the completion time of the first process as the time base.

[0045] Based on the workpiece state data, environmental state data, and workpiece state evolution model corresponding to the current state sampling time, a time-series evolution analysis is performed on the workpiece state at the analysis time point to determine the future workpiece state domain corresponding to the analysis time point. The future workpiece state domain is used to characterize the predicted value range of each workpiece state parameter when evolving to the subsequent state sampling time under the corresponding environmental conditions, taking the actual workpiece state at the current state sampling time as the initial state. The data form is a combination of multiple predicted value ranges of workpiece state parameters. It should be noted that the workpiece state evolution model is used to establish the state change relationship between workpiece state parameters, environmental state parameters, and waiting time.

[0046] The output quality constraint data corresponding to the second process is obtained. This data originates from the product quality standard or production quality database corresponding to the workpiece to be processed. It is used to limit the range of quality values ​​that each output quality parameter of the workpiece to be processed must meet after completing the second process. The output quality parameters characterize the product quality result after the second process. Based on the equipment state data, quality mapping model, and output quality constraint data corresponding to the current state sampling time, the workpiece state at the time of entering the second process is solved in reverse to determine the entry allowable state domain corresponding to the second process. The entry allowable state domain is used to characterize the range of values ​​of the entry workpiece state parameters when the output quality parameters meet the output quality constraint data after processing in the second process, under the processing conditions limited by the set of executable process parameters in the second process. It should be noted that the quality mapping model is used to establish the mapping relationship between the workpiece state parameters at the time of entering the second process, the process parameters of the second process, and the output quality parameters of the second process.

[0047] The future workpiece state domain corresponding to each analysis time point is matched with the entry allowed state domain. Based on the state domain matching results, a valid entry time set is determined. Then, the time intervals in the original waiting window that are later than the current state sampling time are compressed based on the valid entry time set, resulting in a compressed waiting window corresponding to the current state sampling time. The compressed waiting window consists of consecutive valid entry time intervals from the valid entry time set, used to characterize the time range for the workpiece to enter the second process, determined based on the current actual workpiece state, current environmental state, and current second process equipment state. By remapping the matching results of the future workpiece state domain and the entry allowed state domain to the time dimension, the process waiting time is determined in reverse from the workpiece state conditions, making the original waiting time... The time portion of the window where the corresponding workpiece status does not meet the entry requirements of the second process is deleted, thus compressing the process waiting window. The workpiece to be processed is then controlled to enter the second process based on the newly determined compressed waiting window. This newly determined compressed waiting window serves as the time constraint for the intelligent manufacturing production line to control the workpiece's entry into the second process. When the start time of the second process plan allocated by the intelligent manufacturing production line's production scheduling system falls within the newly determined compressed waiting window, the workpiece conveying controller outputs a process transfer instruction to the conveyor line, robotic arm, or automated handling equipment, transferring the workpiece from the process waiting position to the second process. When the start time of the second process plan falls outside the newly determined compressed waiting window, the workpiece is prevented from entering the second process according to the corresponding second process plan start time.

[0048] Based on the workpiece state data, environmental state data, and workpiece state evolution model corresponding to the current state sampling time, a time-series evolution analysis is performed on the workpiece state at the analysis time point to determine the future workpiece state domain corresponding to the analysis time point. This includes the following steps:

[0049] The workpiece state data and environmental state data corresponding to the current state sampling time are concatenated to form an analysis vector. Then, the current analysis vector and the previous N-1 analysis vectors are combined to form a time series analysis sequence.

[0050] The process iterates through all analysis time points. For each selected analysis time point, the prediction span is determined based on the number of state sampling times between the selected analysis time point and the current state sampling time. The workpiece state prediction sub-model corresponding to the prediction span is then obtained. The time series analysis sequence is fed into the workpiece state prediction sub-model corresponding to the prediction span for processing, and the baseline predicted workpiece state data corresponding to the analysis time point is output. This process continues until all analysis time points have been traversed, and the baseline predicted workpiece state data corresponding to each analysis time point is output. The workpiece state evolution model consists of workpiece state prediction sub-models corresponding to different prediction spans, and the workpiece state prediction sub-model adopts the TCN model.

[0051] To obtain several baseline predicted workpiece state data corresponding to the current state sampling time, it's important to note that the current state sampling time has already undergone time-series prediction operations with different prediction spans based on the time-series analysis sequences corresponding to previous state sampling times. Therefore, the current state sampling time will have several baseline predicted workpiece state data. The process iterates through the workpiece state parameters corresponding to these data. For each selected workpiece state parameter, the difference between the data corresponding to the selected workpiece state parameter in the baseline predicted workpiece state data and the data corresponding to the selected workpiece state parameter in the workpiece state data at the current state sampling time is calculated and recorded as the prediction deviation value. Then, a Gaussian fitting operation is performed based on all prediction deviation values ​​to obtain the Gaussian fitting parameter vector corresponding to the workpiece state parameters. The Gaussian fitting parameter vector includes the average prediction deviation and the standard deviation of the prediction deviation. Specifically, the Gaussian fitting operation calculates the arithmetic mean of all prediction deviation values ​​to obtain the average prediction deviation. The standard deviation of the prediction deviation is then calculated based on the dispersion of each prediction deviation value relative to the average prediction deviation. The average prediction deviation is used to characterize the current workpiece's performance in the time-series analysis sequence it has undergone. During the process waiting period, the overall offset direction and degree of the workpiece state evolution model are considered. A prediction deviation average greater than zero indicates that the data corresponding to the current actual workpiece state is higher than the baseline prediction value generated by different prediction spans. A prediction deviation average less than zero indicates that the data corresponding to the current actual workpiece state is lower than the baseline prediction value generated by different prediction spans. The prediction deviation standard deviation is used to characterize the dispersion of the prediction results formed by different prediction spans for the same current state sampling time relative to the current actual workpiece state. The larger the prediction deviation standard deviation, the greater the fluctuation of the actual workpiece state relative to the prediction law of the workpiece state evolution model during the current process waiting period. The smaller the prediction deviation standard deviation, the more consistent the offset characteristics of the prediction results formed by different prediction spans. This process continues until all workpiece state parameters corresponding to the workpiece state data have been traversed, and the Gaussian fitting parameter vector corresponding to all workpiece state parameters is output. It should be noted that the actual prediction deviation formed by different prediction spans for the same current state sampling time can reflect the comprehensive state deviation of the workpiece to be processed during the actual waiting process.

[0052] The workpiece state parameters in the baseline predicted workpiece state data corresponding to the analysis time point are processed using the Gaussian fitting parameter vector corresponding to the workpiece state parameters to obtain the predicted value range for the workpiece state parameters at the analysis time point. Specifically, the baseline predicted value corresponding to the selected state data is added to the average prediction deviation in the corresponding Gaussian fitting parameter vector to obtain the corrected predicted value. The standard deviation of the prediction deviation in the corresponding Gaussian fitting parameter vector is multiplied by a preset coverage coefficient to obtain the prediction deviation expansion. The preset coverage coefficient is determined by the confidence range. For example, when selecting a 95% Gaussian confidence range, the preset coverage coefficient can be approximately 1.96. The lower limit of the predicted value is obtained by subtracting the prediction deviation expansion from the corrected predicted value. The lower limit of the predicted value is obtained by adding the prediction deviation expansion to the corrected predicted value. The upper limit of the predicted value is formed by the lower limit of the predicted value and the upper limit of the predicted value, which together form the predicted value interval of the selected state data at the selected analysis time point. Then, the predicted value intervals corresponding to all workpiece state parameters are combined to form the future workpiece state domain corresponding to the analysis time point. Compared with the single benchmark predicted workpiece state data, the future workpiece state domain adds the prediction deviation information corresponding to the current process waiting process. This allows the subsequent matching of the future workpiece state domain with the entry allowable state domain corresponding to the second process to not only determine whether the predicted center of the workpiece state meets the entry requirements of the second process, but also to consider the deviation and fluctuation of the actual state evolution of the workpiece to be processed relative to the model prediction results. This reduces the situation where analysis time points that are not suitable for entering the second process are retained based solely on single-value prediction results.

[0053] The workpiece state evolution model is constructed through the following steps:

[0054] Several prediction spans are set according to the number of state sampling times within the original waiting window, and a workpiece state prediction sub-model is constructed for each prediction span;

[0055] For each prediction span, a workpiece state prediction sub-model corresponding to the prediction span is selected, and workpiece state prediction training samples are obtained. The workpiece state prediction training samples include N analysis vectors arranged in time. The workpiece state prediction training samples are labeled with the number of state sampling times between the state sampling times corresponding to the last analysis vector in the workpiece state prediction training samples and the state sampling times corresponding to the prediction span. All labeled workpiece state prediction training samples are combined into a workpiece state prediction training set. The workpiece state prediction sub-model is trained using the workpiece state prediction training set. During training, the workpiece state prediction training samples are used as the input of the workpiece state prediction sub-model, and the labeled workpiece state data is used as the target output of the workpiece state prediction sub-model. The workpiece state prediction loss value is constructed based on the difference between the predicted output and the target output of the workpiece state prediction sub-model. The workpiece state prediction loss value can be constructed by MSE, which minimizes the direction of the workpiece state prediction loss value. The parameters of the workpiece state prediction sub-model are adjusted by gradient backpropagation until the training conditions are met. The training conditions can be reaching a certain number of training iterations and outputting a trained workpiece state prediction sub-model.

[0056] All the workpiece state prediction sub-models corresponding to the prediction span are combined into a workpiece state evolution model;

[0057] Based on the equipment state data, quality mapping model, and output quality constraint data corresponding to the current state sampling time, the workpiece state upon entering the second process is solved in reverse to determine the entry allowable state domain corresponding to the second process. This includes the following steps:

[0058] The inputs to the quality mapping model include the workpiece state data when entering the second process and the equipment state data corresponding to the second process. The output of the quality mapping model is the predicted output quality data after the second process is completed. The order of the data in the predicted output quality data corresponds to the order of the data in the output quality constraint data.

[0059] According to the preset state discrete interval, the input range corresponding to each workpiece state parameter in the workpiece state data is discretized. Specifically, several discrete values ​​are selected from the input range corresponding to the workpiece state parameters, and the discrete values ​​corresponding to each workpiece state parameter are combined to obtain several candidate entry workpiece state data.

[0060] Arrange the discrete values ​​corresponding to each workpiece state parameter from largest to smallest. Among all candidate entry workpiece state data, if two candidate entry workpiece state data differ in only one workpiece state parameter, and the different data are consecutive in the order of the discrete values ​​of the workpiece state parameters after arrangement, the two candidate entry workpiece state data are determined as adjacent candidate entry workpiece state data. Since the workpiece state parameters corresponding to the two adjacent candidate entry workpiece state data are consistent except for one workpiece state parameter, when the two adjacent candidate entry workpiece state data obtain qualified and unqualified quality prediction results respectively, it can be determined that the boundary of the entry allowable state domain is located in the direction of the workpiece state parameter where the value changes.

[0061] The process iterates through the candidate entry workpiece state data. For each selected candidate entry workpiece state data, it inputs the selected candidate entry workpiece state data and the equipment state data corresponding to the current state sampling time into the quality mapping model. The quality mapping model uses a multilayer perceptron to obtain the predicted output quality data corresponding to the selected candidate entry workpiece state data. If the data corresponding to each workpiece state parameter in the predicted output quality data all meet the output quality constraints, the selected candidate entry workpiece state data is determined as qualified entry workpiece state data. Qualified entry workpiece state data is used to represent the combination of entry workpiece states that, under the current second process equipment state, the predicted output quality data after processing in the second process meets the product quality requirements. This process continues until all candidate entry workpiece state data has been traversed, and all qualified entry workpiece state data are combined into a qualified entry workpiece state data set. It should be noted that by fixing the equipment state data corresponding to the current state sampling time as part of the input of the quality mapping model, the subsequent allowed entry state domain represents the entry workpiece state that can meet the output quality requirements under the current second process equipment state, rather than the fixed entry range corresponding to the theoretical processing capacity of the second process.

[0062] Traverse adjacent candidate entry workpiece state data. If adjacent candidate entry workpiece state data belong to the qualified entry workpiece state data set and not to the qualified entry workpiece state data set respectively, record the workpiece state parameters corresponding to different data in the adjacent candidate entry workpiece state data as state parameters to be refined. State parameters to be refined are used to characterize the direction of workpiece state parameters where the boundary of the entry allowable state domain is located. The refined data is located between a qualified entry workpiece state data and a non-qualified entry workpiece state data, and is used to further determine which side of the current two adjacent data is the actual boundary of the entry allowable state domain. If the difference between the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data is... If the value is less than the state boundary precision, no operation is performed. If the difference between the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data is not less than the state boundary precision (set by the operator), the average value of the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data is recorded as the refined data. The refined data is then used to replace the data corresponding to the state parameters to be refined in the candidate entry workpiece state data to form new candidate entry workpiece state data. This process continues until the adjacent candidate entry workpiece state data has been traversed. All new candidate entry workpiece state data are then analyzed to determine whether they are qualified entry workpiece state data, and the set of qualified entry workpiece state data is updated.

[0063] Until there are no adjacent candidate entry workpiece state data whose difference between the data corresponding to the state parameters to be refined is not less than the state boundary accuracy, the set of qualified entry workpiece state data is determined as the entry allowable state domain corresponding to the second process.

[0064] The quality mapping model is trained as follows:

[0065] Several quality mapping training samples are obtained, including workpiece state data when the workpiece is sent into the process and equipment state data corresponding to the process. Both workpiece state data and equipment state data are actually collected. The quality mapping training samples are labeled with output quality data, which is measured by the operator based on the actual workpiece. All labeled quality mapping training samples are combined into a quality mapping training set. The quality mapping model is trained using the quality mapping training set. During training, the quality mapping training samples are used as the input of the quality mapping model, and the labeled output quality data is used as the target output of the quality mapping model. A quality mapping loss value is constructed based on the predicted output and target output of the quality mapping model. The quality mapping loss value can be constructed by using MSE (Minimum Sequence of Errors) to minimize the direction of the quality mapping loss value. The parameters of the quality mapping model are adjusted by gradient backpropagation until the training conditions are met. The training conditions can be reaching a certain number of training iterations and outputting a trained quality mapping model.

[0066] The future workpiece state domain corresponding to each analysis time point is matched with the entry allowed state domain. Based on the state domain matching results, the effective entry time set is determined. Then, based on the effective entry time set, the time intervals in the original waiting window that are later than the current state sampling time are compressed to obtain the compressed waiting window corresponding to the current state sampling time. The specific steps include the following:

[0067] For each analysis time point, the future workpiece state domain corresponding to the selected analysis time point is obtained. The predicted value intervals corresponding to each workpiece state parameter in the future workpiece state domain are discretized according to the preset state discretization interval. Then, the discrete values ​​corresponding to each workpiece state parameter are combined to obtain several sets of fitted future workpiece state data corresponding to the selected analysis time point.

[0068] For each analysis time point, all fitted future workpiece state data corresponding to that time point are traversed. For a selected set of fitted future workpiece state data, qualified entry workpiece state data corresponding to the selected set of fitted future workpiece state data are searched in the entry allowed state domain. If the difference between the data corresponding to each workpiece state parameter in the selected set of fitted future workpiece state data and the data corresponding to the same workpiece state parameter in the qualified entry workpiece state data is less than the state boundary accuracy, the selected set of fitted future workpiece state data and the entry allowed state domain satisfy the state domain matching. This process continues until all fitted future workpiece state data corresponding to the analysis time point have been traversed. If all fitted future workpiece state data and the entry allowed state domain satisfy the state domain matching, the analysis time point is determined as a valid entry time. If there are fitted future workpiece state data and the entry allowed state domain that do not satisfy the state domain matching, the analysis time point is not determined as a valid entry time.

[0069] All valid entry times are arranged in chronological order to form a set of valid entry times. If the time difference between two adjacent valid entry times in the set of valid entry times is equal to the preset state sampling period, the two valid entry times are assigned to the same continuous valid time group.

[0070] For consecutive valid time groups, a compressed waiting sub-window is constructed based on the valid entry times on the left and right sides of the consecutive valid time group;

[0071] All compressed waiting sub-windows are combined to form the compressed waiting window corresponding to the current state sampling time;

[0072] This application acquires workpiece state data, environmental state data, and equipment state data corresponding to the second process under a unified state sampling time benchmark. This establishes a time correspondence between the workpiece-side state, environmental-side state, and equipment-side state during the workpiece waiting process, providing a real-time data foundation for the dynamic determination of the process waiting window. By combining workpiece state data, environmental state data, and a workpiece state evolution model, the application performs time-series evolution analysis on the workpiece state at each analysis time point within the original waiting window, determining the future workpiece state domain corresponding to each analysis time point. This transforms the determination of the process waiting time from simply relying on a preset waiting time limit to considering the actual state evolution of the workpiece during the waiting process. Furthermore, by representing the future workpiece state with a state domain, the application reduces the bias caused by using a single predicted state to determine the process entry timing. By combining the current equipment state data of the second process, the quality mapping model, and the output quality constraint data, the application reversely determines the entry allowable state domain, converting the quality requirements after the second process into constraints on the entry workpiece state under the current equipment operating state. This allows the workpiece entry conditions to be dynamically adjusted. The process is adjusted according to the actual processing status changes in the second process. By matching the future workpiece state domain corresponding to each analysis time point with the allowed entry state domain, and determining the effective entry time set based on the matching results, a correspondence is established between the range of possible future states that the workpiece to be processed may reach and the range of entry states that the second process can currently accept. This eliminates time positions in the original waiting window where the workpiece state does not match the entry conditions of the second process. By forming a compressed waiting window based on the time continuity relationship in the effective entry time set, the time intervals in the original waiting window that do not meet the state domain matching requirements are eliminated, while the continuous effective time intervals that meet the entry conditions of the second process are retained. The newly determined compressed waiting window is used to control the entry of the workpiece to be processed into the second process, realizing the reverse determination from the workpiece state conditions and the processing quality conditions of the second process to the process entry time. On the basis of ensuring that the workpiece to be processed is compatible with the actual processing conditions of the second process when it enters the second process, the invalid waiting time is compressed, improving the adaptability between the workpiece transfer timing and the actual production state in the intelligent manufacturing production line.

[0073] Example 2, a process waiting window compression system, such as Figure 1 As shown, it includes:

[0074] The process data acquisition module is used to discretize the original waiting window corresponding to the process waiting stage between the completion time of the first process and the start time of the second process according to a preset state sampling period, resulting in several state sampling times. The preset state sampling period is used to unify the time benchmark for actual state acquisition and future state analysis within the process waiting stage, with the completion time of the first process serving as the first state sampling time. At the current state sampling time, the module acquires the workpiece state data, the environmental state data corresponding to the environment of the workpiece, and the equipment state data corresponding to the second process. The state sampling times after the current state sampling time and within the original waiting window are recorded as analysis time points. The workpiece state data is used to characterize the actual physical state of the workpiece at the corresponding state sampling time. The workpiece state data comes from detection devices arranged at the exit of the first process, the buffer station, or the workpiece conveying path. The data format is a state vector composed of multiple workpiece state parameters. The workpiece state parameters are determined based on the changes between the first and second processes that affect the processing result of the second process. The workpiece characteristics are determined. In the production scenario where the workpiece enters the drying process after spraying, the workpiece state parameters include coating surface temperature, coating humidity, and coating viscosity. Environmental state data is used to characterize the environmental conditions that affect the workpiece's state evolution during the process waiting stage. The environmental state data comes from environmental sensors set up in the buffer station or workpiece conveying area, and the data format is an environmental state vector composed of ambient temperature, ambient humidity, and air velocity. Equipment state data is used to characterize the actual operating state of the equipment corresponding to the second process at the corresponding state sampling time. The equipment state data comes from the second process equipment controller, PLC, and equipment sensors, and the data format is an equipment state vector composed of equipment temperature, equipment speed, actuator status, and power status, such as the actual temperature, actual wind speed, conveying speed, and cavity pressure of the drying equipment. It should be noted that the original waiting window comes from the process route configuration data or production process database corresponding to the intelligent manufacturing production line, and the data format is a time interval composed of the start time and the end time with the completion time of the first process as the time base.

[0075] The workpiece state temporal evolution analysis module is used to perform temporal evolution analysis on the workpiece state at the analysis time point based on the workpiece state data, environmental state data, and workpiece state evolution model corresponding to the current state sampling time. It determines the future workpiece state domain corresponding to the analysis time point. The future workpiece state domain characterizes the predicted value range of each workpiece state parameter when evolving from the actual workpiece state at the current state sampling time as the initial state under corresponding environmental conditions to subsequent state sampling times. The data format is a combination of multiple predicted value ranges for workpiece state parameters. It should be noted that the workpiece state evolution model is used to establish the state change relationship between workpiece state parameters, environmental state parameters, and waiting time.

[0076] The inverse solution module for the entry allowable state is used to obtain the output quality constraint data corresponding to the second process. The output quality constraint data comes from the product quality standard or production quality database corresponding to the workpiece to be processed. It is used to limit the range of quality values ​​that each output quality parameter of the workpiece to be processed must meet after completing the second process. The output quality parameters represent the product quality result after the second process is completed. Based on the equipment status data, quality mapping model and output quality constraint data corresponding to the current state sampling time, the module performs inverse solution on the workpiece state when entering the second process to determine the entry allowable state domain corresponding to the second process. The entry allowable state domain is used to represent the range of values ​​of the corresponding entry workpiece state parameters when the output quality parameters meet the output quality constraint data after the second process, under the processing conditions limited by the set of executable process parameters in the second process. It should be noted that the quality mapping model is used to establish the mapping relationship between the workpiece state parameters when entering the second process, the process parameters of the second process and the output quality parameters of the second process.

[0077] The process waiting window compression module is used to match the future workpiece state domain corresponding to each analysis time point with the entry allowed state domain. Based on the state domain matching results, it determines the effective entry time set and compresses the time intervals in the original waiting window that are later than the current state sampling time based on the effective entry time set. This results in the compressed waiting window corresponding to the current state sampling time. The compressed waiting window consists of consecutive effective entry time intervals from the effective entry time set. It characterizes the time range within which the workpiece to be processed enters the second process, determined based on the current actual workpiece state, the current environmental state, and the current second process equipment state. By remapping the matching results of the future workpiece state domain and the entry allowed state domain to the time dimension, it enables the reverse determination of the process waiting time from the workpiece state conditions. During this period, the time portion of the original waiting window where the corresponding workpiece status does not meet the entry requirements of the second process is deleted, thus compressing the process waiting window. Based on the newly determined compressed waiting window, the workpiece to be processed is controlled to enter the second process. The newly determined compressed waiting window serves as the time constraint for the intelligent manufacturing production line to control the workpiece to be processed to enter the second process. When the start time of the second process plan allocated by the production scheduling system of the intelligent manufacturing production line falls within the newly determined compressed waiting window, the workpiece conveying controller outputs a process transfer instruction to the conveyor line, robotic arm, or automated handling equipment, transferring the workpiece to be processed from the process waiting position to the second process. When the start time of the second process plan falls outside the newly determined compressed waiting window, the workpiece to be processed is prevented from entering the second process according to the corresponding second process plan start time.

[0078] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for compressing process waiting windows, applied to intelligent manufacturing production lines, characterized in that, The workpiece to be processed undergoes the first and second processes sequentially according to a preset process route. The period between the completion of the first process and the start of the second process forms a process waiting phase. The method includes the following steps: The original waiting window corresponding to the process waiting stage between the completion time of the first process and the start time of the second process is discretized according to the preset state sampling period to obtain several state sampling times. The completion time of the first process is taken as the first state sampling time. At the current state sampling time, the workpiece state data of the workpiece to be processed, the environmental state data corresponding to the environment of the workpiece to be processed, and the equipment state data corresponding to the second process are obtained. The state sampling times after the current state sampling time and within the original waiting window are recorded as the analysis time points. Based on the workpiece state data, environmental state data, and workpiece state evolution model corresponding to the current state sampling time, a time-series evolution analysis of the workpiece state at the analysis time point is performed to determine the future workpiece state domain corresponding to the analysis time point. Obtain the output quality constraint data corresponding to the second process. Based on the equipment state data, quality mapping model and output quality constraint data corresponding to the current state sampling time, perform reverse solution on the workpiece state when entering the second process to determine the entry allowable state domain corresponding to the second process. The future workpiece state domain corresponding to each analysis time point is matched with the entry allowed state domain. The effective entry time set is determined based on the state domain matching result. The time interval of the original waiting window that is later than the current state sampling time is compressed based on the effective entry time set to obtain the compressed waiting window corresponding to the current state sampling time. The workpiece to be processed is controlled to enter the second process according to the newly determined compressed waiting window.

2. The method for compressing a process waiting window according to claim 1, characterized in that, Based on the workpiece state data, environmental state data, and workpiece state evolution model corresponding to the current state sampling time, a time-series evolution analysis is performed on the workpiece state at the analysis time point to determine the future workpiece state domain corresponding to the analysis time point. This specifically includes the following steps: The workpiece state data and environmental state data corresponding to the current state sampling time are concatenated to form an analysis vector. Then, the current analysis vector and the previous N-1 analysis vectors are combined to form a time series analysis sequence. The process iterates through all analysis time points. For each selected analysis time point, the prediction span is determined based on the number of state sampling times between the selected analysis time point and the current state sampling time. The workpiece state prediction sub-model corresponding to the prediction span is then obtained. The time series analysis sequence is fed into the workpiece state prediction sub-model corresponding to the prediction span for processing, and the baseline predicted workpiece state data corresponding to the analysis time point is output. This process continues until all analysis time points have been traversed, and the baseline predicted workpiece state data corresponding to each analysis time point is output. The workpiece state evolution model consists of workpiece state prediction sub-models corresponding to different prediction spans. Acquire several benchmark predicted workpiece state data corresponding to the current state sampling time. Iterate through the workpiece state parameters corresponding to the workpiece state data. For the selected workpiece state parameter, calculate the difference between the data corresponding to the selected workpiece state parameter in the benchmark predicted workpiece state data and the data corresponding to the selected workpiece state parameter in the workpiece state data corresponding to the current state sampling time. Record this difference as the prediction deviation value. Then, perform Gaussian fitting operation based on all prediction deviation values ​​to obtain the Gaussian fitting parameter vector corresponding to the workpiece state parameter. The Gaussian fitting parameter vector includes the average prediction deviation and the standard deviation of the prediction deviation. Continue until all workpiece state parameters corresponding to the workpiece state data have been traversed. Output the Gaussian fitting parameter vector corresponding to all workpiece state parameters. The workpiece state parameters in the baseline predicted workpiece state data corresponding to the analysis time point are processed by the Gaussian fitting parameter vector corresponding to the workpiece state parameters to obtain the predicted value interval of the workpiece state parameters corresponding to the analysis time point. Then, the predicted value intervals of all workpiece state parameters are combined to form the future workpiece state domain corresponding to the analysis time point.

3. The method for compressing a process waiting window according to claim 2, characterized in that, Based on the equipment state data, quality mapping model, and output quality constraint data corresponding to the current state sampling time, the workpiece state upon entering the second process is solved in reverse to determine the entry allowable state domain corresponding to the second process. This includes the following steps: The inputs to the quality mapping model include the workpiece state data when entering the second process and the equipment state data corresponding to the second process. The output of the quality mapping model is the predicted output quality data after the second process is completed. The order of the data in the predicted output quality data corresponds to the order of the data in the output quality constraint data. According to the preset state discrete interval, the input range corresponding to each workpiece state parameter in the workpiece state data is discretized, and the discrete values ​​corresponding to each workpiece state parameter are combined to obtain several candidate entry workpiece state data. Arrange the discrete values ​​corresponding to each workpiece state parameter from largest to smallest. Among all candidate entry workpiece state data, if two candidate entry workpiece state data have only one different workpiece state parameter, and the different data are consecutive in the order of the discrete values ​​of the workpiece state parameters after arrangement, the two candidate entry workpiece state data are determined as adjacent candidate entry workpiece state data. Traverse the candidate entry workpiece state data. For the selected candidate entry workpiece state data, input the selected candidate entry workpiece state data and the equipment state data corresponding to the current state sampling time into the quality mapping model to obtain the predicted output quality data corresponding to the selected candidate entry workpiece state data. If the data corresponding to each workpiece state parameter in the predicted output quality data all meet the output quality constraint data, the selected candidate entry workpiece state data is determined as qualified entry workpiece state data. The process continues until all candidate entry workpiece status data have been traversed, and all qualified entry workpiece status data are combined into a qualified entry workpiece status data set. Traverse adjacent candidate entry workpiece state data. If adjacent candidate entry workpiece state data belong to the qualified entry workpiece state data set and not to the qualified entry workpiece state data set respectively, record the workpiece state parameters corresponding to the different data in the adjacent candidate entry workpiece state data as state parameters to be refined. If the difference between the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data is less than the state boundary precision, no operation is performed. If the difference between the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data is not less than the state boundary precision, record the average value of the data corresponding to the state parameters to be refined in the adjacent candidate entry workpiece state data as refined data. Then replace the data corresponding to the state parameters to be refined in the candidate entry workpiece state data with the refined data to form new candidate entry workpiece state data. Continue until the adjacent candidate entry workpiece state data has been traversed. Then continue to analyze whether all new candidate entry workpiece state data are qualified entry workpiece state data and update the qualified entry workpiece state data set. Until there are no adjacent candidate entry workpiece state data whose difference between the data corresponding to the state parameters to be refined is not less than the state boundary accuracy, the set of qualified entry workpiece state data is determined as the entry allowable state domain corresponding to the second process.

4. The method for compressing a process waiting window according to claim 3, characterized in that, The future workpiece state domain corresponding to each analysis time point is matched with the entry allowed state domain. Based on the state domain matching results, the effective entry time set is determined. Then, based on the effective entry time set, the time intervals in the original waiting window that are later than the current state sampling time are compressed to obtain the compressed waiting window corresponding to the current state sampling time. The specific steps include the following: For each analysis time point, the future workpiece state domain corresponding to the selected analysis time point is obtained. The predicted value intervals corresponding to each workpiece state parameter in the future workpiece state domain are discretized according to the preset state discretization interval. Then, the discrete values ​​corresponding to each workpiece state parameter are combined to obtain several sets of fitted future workpiece state data corresponding to the selected analysis time point. For each analysis time point, all fitted future workpiece state data corresponding to that time point are traversed. For a selected set of fitted future workpiece state data, qualified entry workpiece state data corresponding to the selected set of fitted future workpiece state data are searched in the entry allowed state domain. If the difference between the data corresponding to each workpiece state parameter in the selected set of fitted future workpiece state data and the data corresponding to the same workpiece state parameter in the qualified entry workpiece state data is less than the state boundary accuracy, the selected set of fitted future workpiece state data and the entry allowed state domain satisfy the state domain matching. This process continues until all fitted future workpiece state data corresponding to the analysis time point have been traversed. If all fitted future workpiece state data and the entry allowed state domain satisfy the state domain matching, the analysis time point is determined as a valid entry time. If there are fitted future workpiece state data and the entry allowed state domain that do not satisfy the state domain matching, the analysis time point is not determined as a valid entry time. All valid entry times are arranged in chronological order to form a set of valid entry times. If the time difference between two adjacent valid entry times in the set of valid entry times is equal to the preset state sampling period, the two valid entry times are assigned to the same continuous valid time group. For consecutive valid time groups, a compressed waiting sub-window is constructed based on the valid entry times on the left and right sides of the consecutive valid time groups; All compressed waiting sub-windows are combined to form the compressed waiting window corresponding to the current state sampling time.

5. The method for compressing a process waiting window according to claim 4, characterized in that, The workpiece state evolution model is constructed through the following steps: Several prediction spans are set according to the number of state sampling times within the original waiting window, and a workpiece state prediction sub-model is constructed for each prediction span; For each prediction span, select the workpiece state prediction sub-model corresponding to the prediction span, obtain workpiece state prediction training samples. The workpiece state prediction training samples include N analysis vectors arranged in time. The number of state sampling times between the state sampling time corresponding to the last analysis vector in the workpiece state prediction training samples is used as the workpiece state data corresponding to the state sampling time of the prediction span. The workpiece state prediction training samples are labeled, and all labeled workpiece state prediction training samples are combined into a workpiece state prediction training set. The workpiece state prediction sub-model is trained using the workpiece state prediction training set. All the workpiece state prediction sub-models corresponding to the prediction span are combined into a workpiece state evolution model.

6. The method for compressing a process waiting window according to claim 5, characterized in that, The quality mapping model is trained as follows: Obtain several quality mapping training samples, which include workpiece status data when the workpiece is sent into the process and equipment status data corresponding to the process. Label the quality mapping training samples by outputting quality data, and form a quality mapping training set by combining all labeled quality mapping training samples. Train the quality mapping model using the quality mapping training set.

7. A process waiting window compression system, characterized in that, The system employs a process waiting window compression method according to any one of claims 1-6, comprising: The process data acquisition module is used to discretize the original waiting window corresponding to the process waiting stage between the completion time of the first process and the start time of the second process according to the preset state sampling period to obtain a number of state sampling times. The completion time of the first process is used as the first state sampling time. At the current state sampling time, the workpiece state data of the workpiece to be processed, the environmental state data corresponding to the environment where the workpiece to be processed is located, and the equipment state data corresponding to the second process are acquired. The state sampling times after the current state sampling time and within the original waiting window are recorded as the analysis time points. The workpiece state temporal evolution analysis module is used to perform temporal evolution analysis on the workpiece state at the analysis time point based on the workpiece state data, environmental state data and workpiece state evolution model corresponding to the current state sampling time, and to determine the future workpiece state domain corresponding to the analysis time point. The inverse solution module for the entry permission state is used to obtain the output quality constraint data corresponding to the second process. Based on the equipment status data, quality mapping model and output quality constraint data corresponding to the current state sampling time, it performs inverse solution on the workpiece state when entering the second process to determine the entry permission state domain corresponding to the second process. The process waiting window compression module is used to match the future workpiece state domain corresponding to each analysis time point with the entry allowed state domain. Based on the state domain matching result, the effective entry time set is determined. Based on the effective entry time set, the time interval of the original waiting window that is later than the current state sampling time is compressed to obtain the compressed waiting window corresponding to the current state sampling time. The workpiece to be processed is controlled to enter the second process according to the newly determined compressed waiting window.

Images