Spacecraft production workshop operation simulation evaluation method and system based on digital twin

Abstract

A digital twin-based spacecraft production workshop operation simulation evaluation method and system includes: collecting historical data on workshop operation status events, determining the probability distribution model that the event occurs under based on the historical data; generating a future event table based on the probability distribution model of the event; generating driving events for workshop operation based on the probability distribution model according to the future event table, and then performing corresponding processing actions according to the processing logic of the driving event to obtain processing results; the driving events include arrival events, processing end events, and departure events; performing statistics on the processing results, updating the system status, and then generating new events, and updating the future event table based on the new events. Calculate the average response time of events in a single cycle. If the average response time in a single cycle is greater than the overall average response time, then the single cycle is a production task peak period. The present invention can realize simulation prediction of workshop operation status and improve spacecraft manufacturing capabilities.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent manufacturing technology, and in particular relates to a spacecraft production workshop operation simulation evaluation method and system based on digital twins. Background Art

[0002] Due to the characteristics of spacecraft structural products such as multiple varieties, high complexity, long life, high reliability, high quality, and high precision, as well as the characteristics of design, production, and assembly during the manufacturing process, variable technical status, interactive influence of parts and components production process, parallel delivery of multiple products, and close cooperation of multiple disciplines, structural product tasks have the characteristics of many varieties, large total amount, small batch, multiple models, and coexistence of research and development and batch production, which in turn affects the product manufacturing process and workshop production, resulting in the manufacturing of spacecraft structural products having difficulties in production organization and management, control and optimization, quality consistency assurance, closed-loop control of key product characteristics, and production process data integrity and traceability. It has become one of the bottlenecks restricting the improvement of the capacity of spacecraft structural product manufacturing workshops.

[0003] In order to ensure the long life, high quality and high reliability requirements of spacecraft structural products, a more complex product manufacturing process route is required, with many process nodes and many cross-iterations between processes. The processes involve many professions, equipment and personnel. In addition, multiple models are put into production in parallel, and each model shares manufacturing resources and production lines, resulting in complex and changeable operation status of the spacecraft structural product production line. Faced with the continuously growing task volume, it is even more difficult to achieve refined management and control of the manufacturing process. At present, the production line production plan decomposition, resource tracking, and plan coordination are usually carried out by increasing the number of scheduling personnel. In order to ensure the urgent production and development nodes, each model is competing for limited manufacturing resources. On the premise of ensuring that the large-scale model task plan is issued, the planning and scheduling personnel must also use various coordination methods to ensure the rational use of manufacturing resources to ensure that the model task can be carried out smoothly. There are generally problems such as low efficiency, unscientific and error-prone in production line configuration and planning and scheduling.

[0004] Some organizations have adopted system-based workshop operation status assessment methods. Traditional approaches primarily rely on a constructed production system operation logic model, using offline simulation to analyze and predict the impact of dynamic variables such as processing time and workpiece arrival on manufacturing system performance (e.g., completion time and delivery delay costs). Simulation-based workshop operation status assessment methods often require manual initial system configuration and settings, and can only output simulation results under specific input conditions. Furthermore, most simulations are offline, resulting in a time lag between the simulation and actual system operation, leading to significant functional lags. Prediction methods using big data technologies have certain data volume requirements, but acquiring large amounts of historical data is difficult for some manufacturing systems. Furthermore, big data-based predictions are purely data-based, with low correlation with the system model, resulting in idealized steady-state predictions. For complex manufacturing systems, in addition to steady-state predictions, real-time transient predictions are also required, given the rapidly changing system state. However, current research has rarely considered both real-time and transient aspects of predictions. Summary of the Invention

[0005] This paper proposes a digital twin-based spacecraft production workshop operation simulation evaluation method and system. This method aims to address issues such as complex production lines and difficulty in refined management and control caused by cross-iteration of process routes for aerospace structural products and resource sharing across multiple models.

[0006] The present invention proposes a spacecraft production workshop operation simulation evaluation method based on digital twin, including:

[0007] Collect historical data of driving events related to the workshop operation status, determine the probability distribution model obeyed by the event occurrence based on the historical data; generate a future event table according to the probability distribution model of the event;

[0008] Generate driving events for workshop operation based on the probability distribution model according to the future event table. The driving events include arrival events, processing end events, and departure events. Then, perform corresponding processing actions according to the processing logic of the driving events to obtain the processing results of the events.

[0009] The processing results are counted, and the time taken for the events is accumulated and included in the simulation timeline. The workshop operation status is updated and new events are generated. The future event table is updated based on the new events. The operation status of the production workshop is evaluated based on the simulation process: the total number of events at a workstation within a time period is counted, and the utilization rate of the workstation is calculated based on the total number of events and the ratio of the total duration of the time taken for each event to the total duration of the simulation process.

[0010] The processing station with the largest queue length in the simulation is regarded as the bottleneck station of the workshop operation; the average response time of all events in the simulation process is statistically calculated as the overall average response time of this simulation, and the average response time of events in a single cycle is calculated. If the average response time in a single cycle is greater than the overall average response time, then the single cycle is the peak period of production tasks.

[0011] Furthermore, the collecting of historical data of driving events related to the workshop operation status and determining the probability distribution model obeyed by the event occurrence based on the historical data specifically includes:

[0012] Based on the historical data of the workshop operation status obtained by statistics, the histogram fitting method is used to identify the distribution clusters of the historical data, and the probability distribution model obeyed by the historical data is estimated based on the data distribution clusters; the arrival event obeys the exponential distribution model, and the processing end event and the departure event obey the normal distribution model.

[0013] Furthermore, according to the processing logic of the driving event, corresponding processing actions are performed to obtain the processing result of the event, specifically including: the arrival event processing logic is,

[0014] If an arrival event occurs at time t, first determine the processing station of the entity involved in the arrival event and judge whether the processing station is busy; if the processing station is busy, then put the entity arrival event into the queue of the processing station; otherwise, the entity enters the station for processing and updates the status of the processing station to busy; then use the normal distribution sample generator to generate a processing time p for the entity * , and at t+p * Generate the processing end event at the moment, and send the processing end event F,t+p * Add to the future events table;

[0015] After that, the next arrival event is planned regardless of whether the processing station is busy or not; an arrival time interval a is set for the next arrival entity using the exponential distribution sample generator * , and at t+a * Generate the next arrival event at time t, and send the arrival event A,t+a * Add to the future events table;

[0016] The states that have changed in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the arrival event is accumulated and counted into the simulation timeline, and the simulation continues.

[0017] Furthermore, according to the processing logic of the driving event, corresponding processing actions are performed to obtain the processing result of the event, specifically including: the processing logic of the leaving event is,

[0018] If a leave event occurs at time t, determine whether there are other entities waiting for inspection in the queue of the inspection station where the entity is located; if there are no other entities waiting in line, the inspection station status is updated to idle; otherwise, the entity with the highest priority in the queue is extracted for inspection; and a normal distribution sample generator is used to plan an inspection time i for the entity. * , and at t+i * Generate a new leaving event at any moment and add the new leaving event to the future event table;

[0019] The states that have changed in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the leave event is accumulated and included in the simulation timeline, and the simulation continues.

[0020] Furthermore, according to the processing logic of the driving event, corresponding processing actions are performed to obtain the processing result of the event, specifically including: the processing logic of the processing end event is,

[0021] At time t, the entity has a processing end event. It is determined whether there is a queued entity in the current processing station buffer. If there is a queued entity, the number of queued entities is reduced by 1, and the next entity in the queue is arranged to enter the processing station. At the same time, a normal distribution sample generator is used to generate a processing time p for the entity entering the station. * , and at t+p * Schedule the processing end event at any time and add the processing end event to the future event table; if there is no queued entity, update the station status to idle; at the same time, determine whether the inspection station of this type of entity is busy. If the inspection station is busy, the entity enters the queue and waits for inspection, and the number of queued entities at the inspection station is increased by 1; if the inspection station is idle, arrange the entity to enter the inspection station and set the inspection station status to busy. At the same time, use the normal distribution sample generator to generate an inspection time i for the entity * , and at t+i * Schedule the departure event at any time and add it to the future event table;

[0022] The changed states in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the processing end event is accumulated and included in the simulation timeline, and the simulation continues.

[0023] Furthermore, the queue length is the cumulative sum of the number of events, and the event response time is the time consumed for part processing and inspection.

[0024] The present invention also proposes a spacecraft production workshop operation simulation evaluation system based on digital twin, including: a historical data collection module, a drive event generation and processing module, a workstation utilization statistics module, and a workshop operation status determination module, wherein:

[0025] The historical data collection module collects historical data of driving events related to the workshop operation status, determines the probability distribution model obeyed by the event occurrence based on the historical data; and generates a future event table according to the probability distribution model of the event;

[0026] The drive event generation and processing module generates drive events for workshop operation based on the probability distribution model according to the future event table. The drive events include arrival events, processing end events, and departure events. Then, according to the processing logic of the drive events, corresponding processing actions are performed to obtain the event processing results.

[0027] The workstation utilization statistics module collects statistics on the processing results, accumulates the time taken by the events and adds them to the simulation timeline, updates the workshop operation status, updates the future event table based on the new events, and evaluates the production workshop operation status based on the simulation process: it counts the total number of events of a workstation within a time period, and calculates the utilization rate of the workstation based on the ratio of the total number of events and the total duration of the time taken by each event to the total duration of the simulation process;

[0028] The workshop operation status determination module takes the processing station with the largest queue length in the simulation as the bottleneck station of the workshop operation; the average value of the response time of all events in the simulation process is statistically calculated as the overall average response time of this simulation, and the average response time of events in a single cycle is calculated. If the average response time in a single cycle is greater than the overall average response time, then the single cycle is the peak period of production tasks.

[0029] Furthermore, the historical data collection module collects historical data of driving events related to the workshop operation status, and determines the probability distribution model obeyed by the event occurrence based on the historical data, specifically including:

[0030] Based on the historical data of the workshop operation status obtained by statistics, the histogram fitting method is used to identify the distribution clusters of the historical data, and the probability distribution model obeyed by the historical data is estimated based on the data distribution clusters; the arrival event obeys the exponential distribution model, and the processing end event and the departure event obey the normal distribution model.

[0031] Furthermore, the driving event generation and processing module performs corresponding processing actions according to the processing logic of the driving event to obtain the event processing result, which specifically includes:

[0032] The arrival event processing logic is as follows: if an arrival event occurs at time t, first determine the processing station of the entity involved in the arrival event and judge whether the processing station is busy; if the processing station is busy, then add the entity arrival event to the queue of the processing station; otherwise, the entity enters the processing station for processing and updates the status of the processing station to busy; then use the normal distribution sample generator to generate a processing time p for the entity * , and at t+p * Generate the processing end event at the moment, and send the processing end event F,t+p * Add to the future events table;

[0033] After that, the next arrival event is planned regardless of whether the processing station is busy or not; an arrival time interval a is set for the next arrival entity using the exponential distribution sample generator * , and at t+a * Generate the next arrival event at time t, and send the arrival event A,t+a * Add to the future events table;

[0034] The processing logic of the leave event is as follows: if a leave event occurs at time t, determine whether there are other entities in the queue of the inspection station where the entity is located waiting for inspection; if there are no other entities in the queue, the inspection station status is updated to idle; otherwise, the entity with the highest priority in the queue is extracted for inspection; and a normal distribution sample generator is used to plan an inspection time i for the entity. * , and at t+i * Generate a new leaving event at any moment and add the new leaving event to the future event table;

[0035] The processing logic of the processing end event is as follows: when the processing end event occurs at time t, it is determined whether there is a queued entity in the current processing station buffer area. If there is a queued entity, the number of queued entities is reduced by 1, and the next entity in the queue is arranged to enter the processing station. At the same time, a normal distribution sample generator is used to generate a processing time p for the entity entering the station. * , and at t+p * Schedule the processing end event at any time and add the processing end event to the future event table; if there is no queued entity, update the station status to idle; at the same time, determine whether the inspection station of this type of entity is busy. If the inspection station is busy, the entity enters the queue and waits for inspection, and the number of queued entities at the inspection station is increased by 1; if the inspection station is idle, arrange the entity to enter the inspection station and set the inspection station status to busy. At the same time, use the normal distribution sample generator to generate an inspection time i for the entity * , and at t+i * Schedule the departure event at any time and add it to the future event table;

[0036] The changed states in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the processing end event is accumulated and included in the simulation timeline, and the simulation continues.

[0037] Furthermore, the queue length is the cumulative sum of the number of events, and the event response time is the time consumed for part processing and inspection.

[0038] The present invention proposes a method and system for simulating and evaluating the operating status of a spacecraft production workshop based on digital twins, which introduces digital twin technology into the simulation and prediction of the workshop operating status. The method uses real-time information to simulate and evaluate the operating status of the production line, and reflects the actual working conditions of the production line from multiple dimensions according to the production line operating status data. In addition to product quality issues and equipment status operation and maintenance, the method evaluates the production line operating status based on discrete event simulation process data. On the basis of dynamic simulation, the method performs statistical analysis of the production process data, which can realize real-time, continuous and transient simulation and prediction of the workshop operating status, identify weak links and potential problems in the production line, quickly locate and warn of anomalies and changes, optimize the configuration of the manufacturing system, give full play to the production capacity and efficiency of existing equipment, and comprehensively improve the manufacturing capability and agile response capability of spacecraft structural products. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a block diagram of the spacecraft production workshop operation status simulation evaluation method of the present invention;

[0040] Figure 2 It is a logic block diagram of the arrival event processing in the driving event of the present invention;

[0041] Figure 3 This is a logic block diagram of processing a leaving event in a driving event of the present invention;

[0042] Figure 4 This is a logic block diagram for processing the processing end event in the driving event of the present invention. DETAILED DESCRIPTION

[0043] The present invention proposes a method and system for simulating and evaluating the operating status of a spacecraft production workshop based on digital twins.

[0044] Figure 1 The block diagram of the intelligent evaluation method for the operation status of aerospace machining workshop based on digital twin of the present invention includes the following steps:

[0045] (1) Constructing simulation operation logic based on event scheduling method

[0046] Event-based simulation is a simulation advancement mechanism that uses a Future Event List (FEL) to continuously add and remove events from the FEL as the simulation clock advances, ensuring that all events occur in the correct order. In event-based simulation, only the initial event is pre-scheduled; the remaining events, such as random arrival and end events, occur naturally during the simulation.

[0047] First, an initial future event table is generated based on the current state. This table is based on events related to the workshop's operational state. Workshop operations include events such as the start and end of processing and inspection, which can be considered driving events for production workshop operations. Driving events can include arrival events, processing end events, and departure events. By cyclically scanning the future event table, the next event to occur is determined. This event includes various events related to production operations in the spacecraft production workshop. When the simulation clock advances from Clock = t to Clock = t1, the system state at time t1 is updated, and the event is removed from the future event table. In other words, once an event occurs, it is removed. New future events are then added to the future event table based on real-time input data, arranged in chronological order.

[0048] The event-based scheduling approach to shop floor digital twin system simulation begins by modeling the system input characteristics of the simulation object, using an abstract model to describe the actual production process. The generated simulation samples are then screened based on the simulation requirements and the actual shop floor conditions to determine the sample input generator. Finally, the simulation event processing logic is described based on the shop floor production process. By triggering different event processing logic, the system simulation process is implemented.

[0049] (2) Modeling of simulation system input characteristics

[0050] When modeling real-world phenomena, achieving completely accurate simulation predictions is difficult due to model distortion. Most real-world processes are random and influenced by numerous factors. Even when the influencing factors are clearly defined, it can be difficult to find quantifiable patterns. These random processes are often the driving force behind simulation models, so abstract models are needed to describe real-world processes. This process is generally referred to as system input characteristic modeling.

[0051] Typically, input feature modeling is performed as follows:

[0052] 1) Production data collection

[0053] Collect and export historical production data for simulation analysis from the existing manufacturing execution system. The production data includes the start and end time of processing, inspection and other activities, the equipment used, and other data.

[0054] 2) Identifying distributions from data

[0055] The data was preprocessed using a histogram, and the quantile-quantile (q-q) plot was used to evaluate the histogram fitting effect, thereby determining the distribution type of the variable. In CNC machining and product inspection units in a machining workshop, the main random processes involved include the random arrival of part processing tasks, the random time used for part processing, and the random time used for part inspection. Processing task arrival events can be considered independent of each other. Independent events such as processing task arrival events follow an exponential distribution model, so an exponential distribution model is used for modeling. Part processing time and inspection time are the cumulative values ​​of the execution time of a series of work steps. Processing end events and exit events follow a normal distribution model, so a normal distribution model is used for modeling.

[0056] 3) Parameter estimation

[0057] After identifying the distribution type, we need to estimate the parameters of the distribution. For example, for an exponential distribution, we need to estimate its rate parameter λ; for a normal distribution, we need to estimate its mean μ and standard deviation σ.

[0058] (3) Simulation sample generation

[0059] Once a distribution is completely determined, it is necessary to continuously generate sample variables that meet the requirements based on the distribution in the simulation.

[0060] 1) Generation of exponentially distributed random variables

[0061] The inverse transform technique can be used to sample from exponential, uniform, Weibull, and triangular distributions, and is the fundamental principle for generating samples from most discrete distributions. A random variable with an exponential distribution has the following probability density function:

[0062]

[0063] Its distribution function (CDF) is:

[0064]

[0065] Find the inverse function of the distribution function for the right half of F(X) (X≥0):

[0066]

[0067] Let F in the inverse function be denoted as R, and let R obey the uniform distribution on [0,1]. Then the sample generator of the exponential distribution is:

[0068]

[0069] 2) Generation of Normally Distributed Random Variables

[0070] The Box-Muller algorithm is used to generate random variables that obey a normal distribution. Assume that the random variables X and Y are independent and both obey a standard normal distribution with the mean, that is:

[0071]

[0072]

[0073] Then the joint probability density of X and Y is:

[0074]

[0075] Transform X and Y into polar coordinates:

[0076]

[0077] Then the joint distribution function of X and Y is:

[0078]

[0079] Therefore, the distribution function of R and θ is:

[0080]

[0081] We know that θ~U(0,2π). Let the distribution functions of R and θ be:

[0082]

[0083]

[0084] Find the inverse of the distribution function:

[0085]

[0086] θ(F)=2πF (15)

[0087] Let F in the inverse function be Z, that is

[0088]

[0089] It can be seen that when Z obeys the uniform distribution on [0,1], the distribution function of R is F(r) and the distribution function of θ is F(φ). Therefore, we select two random variables U1 and U2 that obey the uniform distribution on [0,1] to replace Z in the formula, that is,

[0090]

[0091] θ(F)=2πU2 (18)

[0092] Substituting the above formula into the polar coordinate transformation formula of X and Y, we can get

[0093]

[0094]

[0095] Choose any one of them as the normal distribution sample generator.

[0096] (4) Simulation event processing

[0097] In a manufacturing workshop, the lifecycle of a product to be processed is divided into four stages: entry, processing, inspection, and departure. The events that occur during this lifecycle are: arrival event, processing start event, processing end event, inspection start event, and inspection end event (departure event). Since the entity arrival event is the fundamental driving force behind the operation of the workshop system, the processing end event and departure event are secondary driving forces generated by human operations, and the processing start event and inspection start event are naturally occurring slave events, the arrival event, processing end event, and departure event are used as the driving events of the workshop simulation system, with A (Arrive), F (Finish Processing), and D (Departure) representing the event types, respectively. The following section clarifies the processing logic of the three events.

[0098] 1) Arrival event processing logic

[0099] Arrival event processing logic is as follows Figure 2 As shown. If an arrival event occurs at Clock = t, the entity is scheduled and its processing station is determined to determine whether the processing station is busy (i.e., the processing station is in progress). If the station is busy, the entity is placed in the queue of the processing station; otherwise, the entity enters the station for processing and the status of the processing station is updated to busy. Then, a processing time p is planned for the entity using the normal distribution sample generator. * , and at t+p * Arrange new processing end events at all times, that is, the event (F, t+p * ) is inserted into FEL (Future EventList). After that, the next arrival event is planned regardless of whether the workstation is busy or not. An arrival time interval a is planned for the next arrival entity using an exponential distribution sample generator. * , and at t+a * Arrange new arrival events at all times, that is, the event (A, t+a *) is inserted into the FEL. Next, statistics such as the processing status and number of completed items that have changed during the above process are summarized. Finally, the control flow return time is advanced to the main process, the time consumed by the event is accumulated and added to the simulation timeline, and the simulation continues.

[0100] 2) Leave event processing logic

[0101] Leave event handling logic such as Figure 3 As shown. If a leaving event occurs at Clock = t, determine whether there are still entities waiting for inspection in the queue of the inspection station where the entity is located. If there are no entities in the queue, the inspection station status is updated to idle. Otherwise, the entity with the highest priority in the queue is selected for inspection. Use the normal distribution sample generator to plan an inspection time i for the entity * , and at t+i * Arrange new departure events at all times, that is, the event (D, t+i * ) is inserted into the FEL. The state and statistical data that changed during the above process are then summarized. Finally, the control flow return time is advanced to the main process, the time consumed by the event is accumulated and added to the simulation timeline, and the simulation continues.

[0102] The processing logic of the processing end event is as follows Figure 4 As shown in the figure, the processing end event is an intermediate event that takes over the processing event and the inspection event. Therefore, it can be regarded as a departure event of the processing process and an arrival event of the inspection process at the same time. Combining the processing logic of the arrival event and the departure event is the processing logic of the processing end event, as shown in the figure. At Clock = t, the entity has a processing end event. It is determined whether there is a queued entity in the current processing station buffer. If there is a queued entity, the number of queued entities is reduced by 1, and the next entity in the queue is arranged to enter the processing station. At the same time, a normal distribution sample generator is used to plan a processing time p for the entity entering the station. * , and at t+p * Arrange a new processing end event at any time, that is, insert the event (F, t+p*) into FEL; if there is no queued entity, update the station status to idle. At the same time, determine whether the inspection station of this type of entity is busy. If the inspection station is busy, the entity enters the queue and waits for inspection, and the number of queued entities at the inspection station + 1; if the inspection station is idle, arrange the entity to enter the inspection station and set the inspection station status to busy. At the same time, use the normal distribution sample generator to plan an inspection time i for the entity * , and at t+i * Arrange new departure events at all times, that is, the event (D, t+i *) is inserted into the FEL. The state data and statistical data that have changed during the above process are then summarized, and the control flow return time is advanced to the main process. The time consumed by the event is accumulated and added to the simulation timeline, and the simulation continues.

[0103] (5) Intelligent evaluation of machining workshop operation status by integrating real-time data

[0104] After the simulation is initialized, the current process of each material, the current status of each device, and the current queue information of each buffer area are scanned from the workshop digital twin model. The scanned events are inserted into the future schedule to obtain the initial future event table. In this case, the entity completes the four events of arrival, processing, inspection, and departure. If a part is currently in a processing station or inspection station, the processing action or inspection action is triggered. The sample generator arranges a processing end event (F event) or a departure event (D event) for it and inserts it into the FEL. The processing station or inspection station status is set to busy. If a part is in the processing or inspection buffer area, the part type is counted and added to the corresponding entity queue. The statistical data, system status, and future schedule are updated, and the digital twin workshop status is output.

[0105] The number of processing completion events (i.e., entity completion events) and exit events (i.e., entity completion inspection events) triggered within N hours are counted. The sum of these two events is the total number of production tasks within N hours. The ratio of the total processing or inspection time at each workstation to the overall simulation time is the workstation utilization rate. The workstation with the largest queue length in the simulation is the bottleneck for the processing task. The average response time of all production tasks within the overall simulation time is calculated as the average response time for the simulation. Taking 8 hours (1 day) as a cycle, the average response time of each production task is calculated for each cycle. If the average response time for a single day is greater than the overall average response time, it indicates that the processing task is busy that day. The day with the highest average response time is considered the peak production period. The simulation period is set as needed. For example, to analyze production status within a month, the simulation period can be set to 240 hours. To analyze production status within the next working day, the simulation period can be set to 8 hours. This way, the elapsed time corresponding to the simulation is directly removed from the FEL.

[0106] Correspondingly, the present invention also proposes a spacecraft production workshop operation simulation evaluation system based on digital twins, including: a historical data collection module, a drive event generation and processing module, a workstation utilization statistics module, and a workshop operation status determination module.

[0107] The historical data collection module collects historical data of driving events related to the workshop operation status, determines the probability distribution model obeyed by the event occurrence based on the historical data, and generates a future event list (FEL) according to the probability distribution model of the event;

[0108] The drive event generation and processing module generates drive events for workshop operation based on the probability distribution model according to the future event table. The drive events include arrival events, processing end events, and departure events. Then, according to the processing logic of the drive events, corresponding processing actions are performed to obtain the event processing results.

[0109] The station utilization statistics module collects statistics on the processing results, accumulates the time taken by the events and adds it to the simulation timeline, updates the workshop operation status, updates the future event table based on new events, and evaluates the production workshop operation status based on the simulation process: it counts the total number of events at a station within a time period, and calculates the utilization rate of the station based on the total number of events and the ratio of the total duration of the time taken by each event to the total duration of the simulation process;

[0110] The workshop operation status determination module regards the processing station with the largest queue length in the simulation as the bottleneck station of the workshop operation; the average response time of all events in the simulation process is statistically calculated as the overall average response time of this simulation, and the average response time of events in a single cycle is calculated. If the average response time in a single cycle is greater than the overall average response time, then the single cycle is the peak period of production tasks.

[0111] The historical data collection module collects historical data of driving events related to the workshop operation status and determines the probability distribution model obeyed by the event occurrence based on the historical data, specifically including:

[0112] Based on the historical data of the workshop operation status obtained by statistics, the histogram fitting method is used to identify the distribution clusters of the historical data, and the probability distribution model obeyed by the historical data is estimated based on the data distribution clusters; the arrival event obeys the exponential distribution model, and the processing end event and the departure event obey the normal distribution model.

[0113] The driving event generation and processing module performs corresponding processing actions according to the processing logic of the driving event to obtain the event processing result, which specifically includes:

[0114] The arrival event processing logic is as follows: if an arrival event occurs at the time Clock = tt, first determine the processing station of the entity involved in the arrival event and judge whether the processing station is busy; if the station is busy, then add the entity arrival event to the queue of the processing station; otherwise, the entity enters the station for processing and updates the status of the processing station to busy; then use the normal distribution sample generator to generate a processing time p for the entity * , and at t+p * Generate the processing end event at the moment, and send the processing end event (F, t+p * ) is added to the future event table.

[0115] After that, the next arrival event is planned regardless of whether the processing station is busy or not; an arrival time interval a is set for the next arrival entity using the exponential distribution sample generator * , and at t+a * Generate the next arrival event at time t, and send the arrival event (A, t+a * ) is added to the future event table.

[0116] The processing logic of the leave event is as follows: if a leave event occurs at time t, determine whether there are other entities in the queue of the inspection station where the entity is located waiting for inspection; if there are no other entities in the queue, the inspection station status is updated to idle; otherwise, the entity with the highest priority in the queue is extracted for inspection; and a normal distribution sample generator is used to plan an inspection time i for the entity. * , and at t+i * A new departure event is generated at each moment and added to the future event table.

[0117] The processing logic of the processing end event is as follows: when the processing end event occurs at time t, it is determined whether there is a queued entity in the current processing station buffer area. If there is a queued entity, the number of queued entities is reduced by 1, and the next entity in the queue is arranged to enter the processing station. At the same time, a normal distribution sample generator is used to generate a processing time p for the entity entering the station. * , and at t+p * Schedule the processing end event at any time and add the processing end event to the future event list (FEL); if there is no queued entity, update the station status to idle; at the same time, determine whether the inspection station of this type of entity is busy. If the inspection station is busy, the entity enters the queue and waits for inspection, and the number of queued entities at the inspection station is increased by 1; if the inspection station is idle, arrange the entity to enter the inspection station and set the inspection station status to busy. At the same time, use the normal distribution sample generator to generate an inspection time i for the entity * , and at t+i * Schedule a departure event at any time and add the departure event to the future event table.

[0118] The changed states in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the processing end event is accumulated and included in the simulation timeline, and the simulation continues.

[0119] The queue length is the cumulative total of the number of events, and the event response time is the time consumed for parts processing and inspection.

[0120] Example

[0121] In the method for intelligently evaluating the operating status of an aerospace machining workshop based on digital twins of the present invention, the steps for intelligently evaluating the operating status of an aerospace machining workshop are as follows:

[0122] (1) Data collection

[0123] In the high-speed five-axis machining unit, medium-sized structure machining unit, and inspection unit of this machining workshop, since it is difficult to collect historical workshop data, based on interviews with the workshop director, it is learned that the arrival time of the workshop's part processing tasks varies according to the part type, as shown in Table 1, which shows the arrival rate of each type of part processing tasks in this workshop.

[0124] Table 1 Arrival rate of various types of parts processing tasks in this workshop

[0125]

[0126] (2) Parameter estimation

[0127] In actual production, the six types of parts, A, B, C, D, E, and F, have different processing / inspection states at eight processing stations and six inspection stations. Therefore, parameter estimation is performed for the processing / inspection time of different types of parts at different stations. There are interlaced reachable relationships between machine tools and parts, each of which has an independent processing time, as shown in Table 2, Mean Processing Time (Hours) for Each Type of Part on Each Machine Tool. The inspection time for each type of part is shown in Table 3, Inspection Time (Hours) for Each Type of Part. Furthermore, the processing and inspection processes have different standard deviations, as shown in Table 4, Standard Deviation (Hours) of Processing and Inspection Time. A normal distribution allows for negative values, which are meaningless for time. Therefore, during program development, the generated samples should be judged and screened.

[0128] Table 2 Average processing time of various types of parts on various machine tools (hours)

[0129]

[0130] Table 3 Inspection time for various types of parts (hours)

[0131]

[0132] Table 4 Standard deviation of processing time and inspection time (hours)

[0133]

[0134] (3) Sample generator generation

[0135] According to the event processing logic of the workshop simulation, the time intervals between the random arrivals of parts A, B, C, D, E, and F at processing stations follow exponential distributions with means of 0.8, 0.667, 1.82, 1.33, 2, and 2.22, respectively. The processing times of the six parts at different processing stations follow an approximately normal distribution with the mean shown in Table 2 and a standard deviation of 0.3. The inspection times of various parts at the inspection stations follow an approximately normal distribution with the mean shown in Table 3 and a standard deviation of 0.2.

[0136] (4) Evaluation and analysis

[0137] The digital twin workshop simulation system was set to a total simulation time of 240 hours (30 days) and a frequency of 1 simulation per minute. The current workshop status was scanned. In the Processing (Inspection) Status column, 1 indicates that the current workstation is processing or inspecting, and 0 indicates that the current workstation is idle. In the Processing (Inspection) Queue Status column, the number indicates the number of parts currently queued at that workstation. The figure shows the current workshop status at time t = 0 and the simulation output.

[0138] Table 5 Workshop current workstation status table

[0139]

[0140]

[0141] Count the number of times the processing end event is triggered (i.e., the part processing event is completed) and the number of times the leaving event is triggered (i.e., the part inspection event is completed) within 240 hours (30 days). The sum of the two is the total number of production tasks within 30 days. The ratio of the total processing or inspection time of each workstation to the overall simulation time is the utilization rate of the workstation. The processing station with the largest queue length in the simulation is the bottleneck station of the processing task. The average value of the response time of all production tasks within the overall simulation time is calculated as the average response time of this simulation. Taking 8 hours (1 day) as a cycle, calculate the average response time of the production tasks in a single cycle. If the average response time of a single day is greater than the overall average response time, it means that the processing tasks on that day are busy. The day with the largest average response time is the peak period of production tasks.

[0142] Through simulation calculations, information was obtained on the utilization rate of key equipment, inspection station utilization, peak periods, and bottleneck stations within the twin workshop. Assessments indicated that a total of 746 production tasks, including machining and inspection tasks, would be executed within the workshop over the next 30 days. The longest queue in this simulation was at the DMU125PN station, with a total queue length of 12, making it the bottleneck for machining tasks. The average response time for part production tasks was 12.3472 minutes, with 243 tasks exceeding the average response time, representing 32.5737% of all production tasks. The maximum daily average response time occurred in the 15th cycle, indicating that the peak production period will occur on the 15th working day. The simulation report data was evaluated by experts and is considered reliable. The simulation results are displayed in the digital twin workshop as a 2D dashboard, updated every minute.

[0143] This invention addresses the complex aerospace product manufacturing sector and the operational control needs of aerospace machining workshops. It employs a digital twin-based intelligent assessment method for the operational status of aerospace machining workshops. Based on the simulation principles of discrete event systems, it employs an event scheduling method to establish a statistical model of system input characteristics, develop a simulation sample generation program, and clarify the processing logic for three types of events. Real-time data is integrated into the discrete event system simulation process, achieving real-time simulation through state cyclic scanning, simulation continuity through cyclic execution, and simulation transientness through event scheduling. This method effectively addresses the complex production lines and difficulties in refined control caused by cross-iteration of aerospace structural product process routes and shared resources across multiple models. By constructing a digital twin of the workshop, it enables real-time tracking of workshop manufacturing resources and intelligent assessment of operational status, organically linking the spacecraft structural product manufacturing process with manufacturing resource data and information, and improving spacecraft structural manufacturing capabilities and rapid response capabilities.

Claims

1. A spacecraft production workshop operation simulation evaluation method based on digital twin, characterized by: include: Collect historical data of driving events related to the operation status of the workshop, and determine the probability distribution model obeyed by the event occurrence based on the historical data; generating a future event table based on the probability distribution model of events; Generate driving events for workshop operation based on the probability distribution model according to the future event table, wherein the driving events include arrival events, processing end events, and departure events; Then, according to the processing logic of the driving event, corresponding processing actions are performed to obtain the processing result of the event; The processing results are counted, and the time used for the event is accumulated and recorded in the simulation time axis, the workshop operation status is updated and a new event is generated, and the future event table is updated based on the new event; Evaluate the operating status of the production workshop based on the simulation process: count the total number of events at a workstation within a time period, and calculate the utilization rate of the workstation based on the ratio of the total number of events and the total duration of each event to the total duration of the simulation process; The processing station with the largest queue length in the simulation is regarded as the bottleneck station of the workshop operation; the average response time of all events in the simulation process is calculated as the overall average response time of this simulation, and the average response time of events in a single cycle is calculated. If the average response time in a single cycle is greater than the overall average response time, the single cycle is the peak period of production tasks; According to the processing logic of the driving event, corresponding processing actions are performed to obtain the processing result of the event, specifically including: the processing logic of the arrival event is as follows: if an arrival event occurs at time t, first determine the processing station of the entity involved in the arrival event, and judge whether the processing station is busy; if the processing station is busy, then add the entity arrival event to the queue of the processing station; otherwise, the entity enters the station for processing, and the status of the processing station is updated to busy; then use the normal distribution sample generator to generate a processing time p for the entity * , and at t+p * Generate the processing end event at the moment, and send the processing end event F,t+p * Add to the future events table; After that, the next arrival event is planned regardless of whether the processing station is busy or not; an arrival time interval a is set for the next arrival entity using the exponential distribution sample generator * , and at t+a * Generate the next arrival event at time t, and send the arrival event A,t+a * Add to the future events table; The states that have changed in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the arrival event is accumulated and counted into the simulation timeline, and the simulation continues.

2. The method according to claim 1, characterized in that The collecting of historical data of driving events related to the workshop operation status and determining the probability distribution model obeyed by the event occurrence based on the historical data specifically includes: Based on the historical data of the workshop operation status obtained by statistics, the histogram fitting method is used to identify the distribution clusters of the historical data, and the probability distribution model obeyed by the historical data is estimated based on the data distribution clusters; the arrival event obeys the exponential distribution model, and the processing end event and the departure event obey the normal distribution model.

3. The method according to claim 1, characterized in that According to the processing logic of the driving event, a corresponding processing action is performed to obtain the processing result of the event, specifically including: the processing logic of the leaving event is, If a leave event occurs at time t, determine whether there are other entities waiting for inspection in the queue of the inspection station where the entity is located; if there are no other entities waiting in line, the inspection station status is updated to idle; otherwise, the entity with the highest priority in the queue is extracted for inspection; and a normal distribution sample generator is used to plan an inspection time i for the entity. * , and at t+i * Generate a new leaving event at any moment and add the new leaving event to the future event table; The states that have changed in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the leave event is accumulated and included in the simulation timeline, and the simulation continues.

4. The method according to claim 1, wherein According to the processing logic of the driving event, corresponding processing actions are performed to obtain the processing result of the event, specifically including: the processing logic of the processing end event is, At time t, the entity has a processing end event. It is determined whether there is a queued entity in the current processing station buffer. If there is a queued entity, the number of queued entities is reduced by 1, and the next entity in the queue is arranged to enter the processing station. At the same time, a normal distribution sample generator is used to generate a processing time p for the entity entering the station. * , and at t+p * Schedule the processing end event at any time and add the processing end event to the future event table; if there is no queued entity, update the station status to idle; at the same time, determine whether the inspection station of this type of entity is busy. If the inspection station is busy, the entity enters the queue and waits for inspection, and the number of queued entities at the inspection station is increased by 1; if the inspection station is idle, arrange the entity to enter the inspection station and set the inspection station status to busy. At the same time, use the normal distribution sample generator to generate an inspection time i for the entity * , and at t+i * Schedule the departure event at any time and add it to the future event table; The changed states in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the processing end event is accumulated and included in the simulation timeline, and the simulation continues.

5. The method according to claim 1, wherein The queue length is the cumulative total of the number of events, and the event response time is the time consumed for part processing and inspection.

6. A spacecraft production workshop operation simulation evaluation system based on digital twin, characterized by: include: Historical data collection module, drive event generation and processing module, workstation utilization statistics module, workshop operation status determination module, among which, The historical data collection module collects historical data of driving events related to the workshop operation status, determines the probability distribution model obeyed by the event occurrence based on the historical data; and generates a future event table according to the probability distribution model of the event; The drive event generation and processing module generates drive events for workshop operation based on the probability distribution model according to the future event table. The drive events include arrival events, processing end events, and departure events. Then, according to the processing logic of the drive events, corresponding processing actions are performed to obtain the event processing results. The workstation utilization statistics module collects statistics on the processing results, accumulates the time taken by the events and adds them to the simulation timeline, updates the workshop operation status, updates the future event table based on the new events, and evaluates the production workshop operation status based on the simulation process: it counts the total number of events of a workstation within a time period, and calculates the utilization rate of the workstation based on the ratio of the total number of events and the total duration of the time taken by each event to the total duration of the simulation process; The workshop operation status determination module uses the processing station with the largest queue length in the simulation as the bottleneck station of the workshop operation; calculates the average response time of all events in the simulation process as the overall average response time of this simulation, and calculates the average response time of events in a single cycle. If the average response time in a single cycle is greater than the overall average response time, the single cycle is a production task peak period; The arrival event processing logic is as follows: if an arrival event occurs at time t, first determine the processing station of the entity involved in the arrival event and judge whether the processing station is busy; if the processing station is busy, then add the entity arrival event to the queue of the processing station; otherwise, the entity enters the processing station for processing and updates the status of the processing station to busy; then use the normal distribution sample generator to generate a processing time p for the entity * , and at t+p * Generate the processing end event at the moment, and send the processing end event F,t+p * Add to the future events table; After that, the next arrival event is planned regardless of whether the processing station is busy or not; an arrival time interval a is set for the next arrival entity using the exponential distribution sample generator * , and at t+a * Generate the next arrival event at time t, and send the arrival event A,t+a * Add to future events list.

7. The system according to claim 6, characterized in that The historical data collection module collects historical data of driving events related to the workshop operation status, and determines the probability distribution model obeyed by the event occurrence based on the historical data, specifically including: Based on the historical data of the workshop operation status obtained by statistics, the histogram fitting method is used to identify the distribution clusters of the historical data, and the probability distribution model obeyed by the historical data is estimated based on the data distribution clusters; the arrival event obeys the exponential distribution model, and the processing end event and the departure event obey the normal distribution model.

8. The system according to claim 7, characterized in that The driving event generation and processing module performs corresponding processing actions according to the processing logic of the driving event to obtain the event processing result, which specifically includes: The processing logic of the leave event is as follows: if a leave event occurs at time t, determine whether there are other entities in the queue of the inspection station where the entity is located waiting for inspection; if there are no other entities in the queue, the inspection station status is updated to idle; otherwise, the entity with the highest priority in the queue is extracted for inspection; and a normal distribution sample generator is used to plan an inspection time i for the entity. * , and at t+i * Generate a new leaving event at any moment and add the new leaving event to the future event table; The processing logic of the processing end event is as follows: when the processing end event occurs at time t, it is determined whether there is a queued entity in the current processing station buffer area. If there is a queued entity, the number of queued entities is reduced by 1, and the next entity in the queue is arranged to enter the processing station. At the same time, a normal distribution sample generator is used to generate a processing time p for the entity entering the station. * , and at t+p * Schedule the processing end event at any time and add the processing end event to the future event table; if there is no queued entity, update the station status to idle; at the same time, determine whether the inspection station of this type of entity is busy. If the inspection station is busy, the entity enters the queue and waits for inspection, and the number of queued entities at the inspection station is increased by 1; if the inspection station is idle, arrange the entity to enter the inspection station and set the inspection station status to busy. At the same time, use the normal distribution sample generator to generate an inspection time i for the entity * , and at t+i * Schedule the departure event at any time and add it to the future event table; The changed states in the above process are summarized, and the simulation process is updated based on the time when the control flow returns. The time consumed by the processing end event is accumulated and included in the simulation timeline, and the simulation continues.

9. The system according to claim 8, characterized in that The queue length is the cumulative total of the number of events, and the event response time is the time consumed for part processing and inspection.

Images