A shell shell production line collaborative scheduling method based on digital twinning

Abstract

The present application relates to the technical field of production scheduling, in particular to a shell production line collaborative scheduling method based on digital twinning, the present application constructs metallurgical dimension and geometric dimension reachable set and controllable window by taking batch, mold section, furnace position and machining unit as pedigree constraint propagation through shell and each process digital twinning model; based on metallurgical dimension, furnace grouping and furnace position optimization are completed, based on geometric dimension, limited material removal grading is completed, and the grade gap is fed back to the furnace production plan; further combining pedigree representative damage sampling, non-destructive proxy migration and residual batch redistribution, closed-loop correction of production scheduling, quality evaluation and grade optimization is realized. The present application can improve production consistency, resource utilization rate and quality control accuracy, reduce test cost, and is suitable for collaborative scheduling and quality management of intelligent shell production process.

Description

Technical Field

[0001] This invention relates to the field of production scheduling technology, specifically to a collaborative scheduling method for a shell casing production line based on digital twins. Background Technology

[0002] The production of artillery shell casings involves multiple processes, including material preparation, forming, heat treatment, machining, and inspection and grading. Equipment, materials, and production capacity are interdependent among these processes, necessitating coordinated scheduling of production commencement and operations across all processes. Existing production scheduling methods (including those incorporating digital twins) typically assume that the product type and specifications are determined at the time of production commencement. They treat the workpiece as a deterministic operation that can be freely disassembled and transferred, aiming to shorten the production cycle, improve equipment utilization, and meet delivery deadlines. The methods then solve for the spatial and temporal allocation of operations on the equipment.

[0003] However, the final delivery grade (weight zone, ballistic consistency grade) of shell casings is unknown at the time of production. Instead, it is determined by the cumulative tolerances of each piece during forming, heat treatment, and machining, which are influenced by batch size and equipment conditions. The resulting batch of blanks produces a random distribution across grades. Furthermore, shell casings must be delivered in batches of the same material, from the same furnace, and within the same process window, according to grade. The dispersion of mass, centroid, and dynamic imbalance within a batch directly determines the dispersion of projectiles fired from that batch. Mixed batches lead to dispersion exceeding tolerances. Batch acceptance is further determined by destructive sampling and testing, and by batch as a whole. Therefore, existing scheduling methods that determine the quantity and type of production for a given product cannot predict the graded output distribution, nor can they balance batching from the same source with graded demand matching. This results in a mismatch between graded output and demand, overproduction to fill gaps, fragmented deliverable batches leading to projectile dispersion exceeding tolerances, and stockpiling of work-in-process and scrap.

[0004] To address this, a collaborative scheduling method for shell casing production lines based on digital twins is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a collaborative scheduling method for artillery shell production lines based on digital twins. By constructing a digital twin model and combining spectral constraints, reachability set analysis, furnace optimization, material removal and classification, and representative sampling, a collaborative closed loop of production scheduling, classification optimization, and quality control is achieved, thereby improving production efficiency and product consistency.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A collaborative scheduling method for a shell casing production line based on digital twins includes: A digital twin model of the shell and the process is established using measured data from the shell production line process; The material batch, mold section, furnace position and machining unit of the shell are used as spectral constraints to propagate along the process. The unreachable level is eliminated by each component and the intersection is taken to obtain the reachable set and controllable window of metallurgical dimension and geometric dimension. Based on metallurgical dimensional reachability sets and controllable windows, shells of the same batch and with compatible grades are grouped into the same heat in the digital twin model, and ranked according to the performance gradient of the same type of furnace under the same loading condition as verified by actual furnace exit to limit the dispersion within the heat and obtain the identity of the delivery batch. Based on the geometric dimension reachable set and controllable window, in the digital twin model, for the shell that can monotonically descend into the adjacent level with the level gap and the metallurgical dimension reachable set covers the level, the material removal position is assigned and merged into the level to obtain the grading result, and the remaining level gap is fed back into the furnace output. Based on the delivery batch identity and grading results, spectral representative destruction sampling is performed on the batch shells. The correspondence between non-destructive proxies and measured ballistic performance is migrated within the compact cluster, and residual batches are redistributed within the qualified grade.

[0007] Preferably, the process for obtaining the reachable sets of the metallurgical and geometric dimensions is as follows: For the material batch, read the physicochemical indicators; for the mold section, read the cumulative number of molded parts; for the furnace position, read the furnace position thermal history identifier; for the machining unit, read the measured dimensions and coaxiality of each shell piece. These four items are used as the spectral data for the material batch component, mold component, furnace batch component, and machining component, respectively. In the digital twin model, each component, based on its spectral data and historical grading records, eliminates levels that are no longer achievable, obtaining a set of permissible levels for the component. The intersection of the permissible level sets of the material batch component and the furnace batch component yields the metallurgical dimension reachable set; the intersection of the permissible level sets of the mold component and the machining component yields the geometric dimension reachable set. Each time the shell flows through a process, the spectral data adds one component, and the digital twin model takes another intersection of the reachable sets, causing the reachable set to narrow as the spectral data accumulates.

[0008] Preferably, the process of acquiring the controllable windows for the metallurgical and geometric dimensions is as follows: For the metallurgical dimension, the digital twin model uses whether the shell has not yet entered heat treatment as a criterion. When the shell has not yet entered heat treatment, the controllable window for the metallurgical dimension is set to open; once the shell has been tempered, the controllable window for the metallurgical dimension is set to close. For the geometric dimension, the digital twin model reads the current measured mass of the shell, the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, and the mass increment that can be removed by axisymmetric finishing within the tolerance. When the removable mass increment is not less than the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, the controllable window for the geometric dimension is set to open; when the removable mass increment is less than the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, the controllable window for the geometric dimension is set to close. The digital twin model only sends the reachable set corresponding to the dimension with the controllable window open to the furnace grouping or material removal classification corresponding to the dimension.

[0009] Preferably, the batch identification process is as follows: Within a shell where the metallurgical dimension controllable window is open, the digital twin model encodes shells that are compatible with the metallurgical dimension and belong to the same batch into the same heat. The digital twin model aggregates the measured performance of samples taken from the furnace under similar loading conditions by furnace position, obtaining the furnace position performance gradient, and retaining only the portion of the furnace position performance gradient whose value exceeds the upper limit of measurement noise. The batch hardenability and forming state of each shell before entering the furnace are read as performance proxies before entering the furnace, and shells with lower performance proxies before entering the furnace are assigned to the appropriate batch. The furnace positions are arranged such that furnace positions with high performance gradient values ​​are assigned to furnace positions with low performance gradient values, and shells with high performance proxy values ​​before entering the furnace are assigned to furnace positions with low performance gradient values. After the furnace is discharged, the digital twin model calculates the discharge performance dispersion within the furnace based on the measured discharge performance of the furnace samples in the same furnace. For furnaces with discharge performance dispersion not exceeding the set bandwidth limit, all members within the furnace are marked as belonging to the same delivery batch. For furnaces with discharge performance dispersion exceeding the set bandwidth limit, all members within the furnace are placed in the state of pending encrypted sampling.

[0010] Preferably, the grading result acquisition process is as follows: read the current gap of each grade, select shells that simultaneously meet three conditions: the shell is located at the upper adjacent weight boundary, the controllable window of the shell's geometric dimension is open, and the reachable set of the shell's metallurgical dimension covers the adjacent lighter grade; for the selected shells, the digital twin model selects a material removal level from several material removal levels where only axisymmetric finishing is performed and the removal amount does not exceed the tolerance limit, and monotonically reduces the shell mass to fall into the adjacent gap grade. Material removal only changes the axial mass and axial centroid of the shell, without changing the shell's wall thickness consistency and ballistic coefficient; the shells that have completed material removal are assigned to the adjacent gap grade, and the shells that have not undergone material removal are assigned to their respective grades according to actual measurements, and the grading results are merged; the current gap of each grade is updated according to the number of shells assigned to the adjacent gap grade, and the updated current gap of each grade is fed back into the production output of the next furnace cycle.

[0011] Preferably, the process of destructive sampling for representative shells in a batch is as follows: Shells of the same delivery batch identity are grouped into several compact clusters based on consistency in three aspects: same material batch, same furnace position, and same material discharge position; when the dispersion of the non-destructive proxy of the shells within a compact cluster is not greater than a set threshold, a digital twin model extracts a representative shell from the compact cluster for destructive sampling test firing to obtain the measured ballistic performance of the representative shell; the non-destructive proxy recorded before the test firing of the representative shell is retained, and the non-destructive proxy is paired with the measured ballistic performance of the representative shell to obtain the correspondence between the non-destructive proxy and the measured ballistic performance within the compact cluster; the result of the destructive sampling test firing is used as the criterion for whether the entire delivery batch is qualified or not, the number of samples in the destructive sampling test firing is not less than the minimum sample size specified in the acceptance procedure, and the number of destructive sampling samples of the delivery batch in the pending encrypted sampling state is higher than that of the delivery batch not in the pending encrypted sampling state.

[0012] Preferably, the process of redistributing residual batches within the already qualified grade is as follows: For the shells that have not been extracted within the same compact cluster, the digital twin model uses the correspondence between the non-destructive proxy within the compact cluster and the measured ballistic performance to convert the non-destructive proxy remaining in the unextracted shells into the corresponding measured ballistic performance level of each unextracted shell; within the deliverable grade range of the unextracted shells that have already qualified, the unextracted shells are classified into a finer grade according to the converted measured ballistic performance level, thus obtaining the residual batch redistribution result; the correspondence within the compact cluster, together with the deviation obtained by comparing the measured grade of each compact cluster with the spectral data, is converted into the component permissible grade set correction amount; the difference between the performance dispersion of each furnace output and the expected furnace position ranking is converted into the furnace position performance gradient correction amount; and the difference between the actual output of each grade after the residual batch redistribution and the grade-specific demand is converted into the furnace output correction amount.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention establishes a digital twin model covering the entire production process of artillery shells. It continuously propagates spectral information from material batches, mold sections, furnace positions, and machining units throughout each process, and constructs reachable sets and controllable windows for both metallurgical and geometric dimensions, achieving collaborative constraints and dynamic convergence between different manufacturing dimensions. This invention can narrow the reachable level range in real time during production, ensuring that each shell remains traceable, predictable, and controllable in its manufacturing state. It can identify subsequent reachable levels and adjustable spaces before entering critical processes such as heat treatment and finishing, reducing cross-batch performance fluctuations from the source, improving the consistency of the entire batch and the stability of production scheduling, while enhancing the mapping accuracy of the digital twin model to the actual production state.

[0014] 2. This invention utilizes metallurgical dimensional reachability sets, geometric dimensional reachability sets, and their respective controllable windows to optimize furnace grouping and perform finite material removal and grading. It also real-time feeds back grade gaps to subsequent furnace production, achieving a closed-loop linkage between grade demand and production planning. This invention can achieve product grade conversion for adjacent grades with grade gaps through finite axisymmetric material removal, improving product availability without altering key ballistic performance and reducing downgrading or scrapping due to grade mismatch. Simultaneously, furnace production is continuously adjusted based on actual gaps, maintaining a dynamic balance between production capacity, market demand, and product grade, thereby reducing inventory backlog, improving raw material utilization, and enhancing the overall production efficiency of the entire production line.

[0015] 3. This invention clusters shells with the same batch, furnace position, and discharge level into compact clusters. While ensuring representativeness, it performs spectral representative destructive sampling, establishing a correspondence between pre-sampling non-destructive proxy parameters and measured ballistic performance. This correspondence is then transferred to unsampled shells within the same cluster, achieving a refined redistribution of remaining product grades. Simultaneously, the sampling quantity is dynamically adjusted based on furnace dispersion, automatically increasing sampling density for abnormal batches. This invention fully utilizes spectral consistency and digital twin historical data, enabling limited destructive testing to cover more product information, reducing the number of destructive tests and testing costs. It also improves the reliability of overall batch quality evaluation and the accuracy of grade classification, continuously refining the permitted grade set, furnace position performance gradient, and production strategy, creating a continuous optimization closed loop for quality control. Attached Figure Description

[0016] Figure 1 A schematic diagram of a collaborative scheduling method for a shell casing production line based on digital twins provided by the present invention; Figure 2 A schematic diagram of the delivery batch identity acquisition process provided by the present invention; Figure 3 This is a schematic diagram of the residual batch redistribution process provided by the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0018] Example 1: The shell production line is equipped with corresponding data acquisition devices along four processes: batch inspection, mold forming, heat treatment (furnace batch), and machining. The physicochemical properties of the batch are measured and recorded by chemical composition analysis equipment and hardness testing equipment before furnace entry. The cumulative number of molded parts in the mold section is accumulated and recorded by an online mold counting device for each forming action. The furnace position's thermal history identifier is generated by a furnace temperature recorder recording the heating parameters (such as heating curves and holding times) for each furnace loading at that position for each furnace batch. The actual measured dimensions and coaxiality of each shell in the machining unit are measured piece by piece by online measuring equipment (such as laser diameter gauges and roundness meters). All the above measurement data are transmitted in real-time or near real-time to the computing platform where the digital twin model resides via an industrial communication network.

[0019] Please see Figure 1This invention provides a collaborative scheduling method for a shell casing production line based on digital twins. The technical solution is as follows: A digital twin model of the shell casing and its processes is established using measured data from the shell casing production line processes; the material batch, mold section, furnace position, and machining unit of the shell casing are used as spectral constraints propagating along the processes; unreachable levels are eliminated from each component, and the intersection is taken to obtain the reachable sets and controllable windows in the metallurgical and geometric dimensions; based on the reachable sets and controllable windows in the metallurgical dimension, shell casings from the same material batch and with compatible levels are grouped into the same furnace in the digital twin model, and the same type of load is verified according to the measured conditions after furnace exit. The furnace position performance gradient ranking is used to limit the dispersion within the furnace batch to obtain the delivery batch identity. Based on the geometric dimension reachable set and controllable window, in the digital twin model, for shells that can monotonically descend into adjacent grades with grade gaps and whose metallurgical dimension reachable set covers the grade only by bounded material removal, the material removal position is assigned and merged into the grade to obtain the grading result. The remaining grade gap is then fed back into the furnace batch production output. Based on the delivery batch identity and grading result, the batch shells are subjected to spectral representative destruction sampling. The correspondence between the lossless proxy and the measured ballistic performance is transferred within the compact cluster, and the residual batches are redistributed within the qualified grades.

[0020] The digital twin model is built in the form of a database on the production line's supporting computing platform. It consists of a phylogenetic data storage unit, and subsequently, reachability set operation units, controllable window determination units, furnace batch grouping units, material removal and grading units, destructive sampling units, and residual batch redistribution units that process the data in the phylogenetic data storage unit. Each unit is implemented as a software module, and the modules do not directly call each other. Instead, data transfer is completed uniformly through reading and writing agreed-upon fields in the phylogenetic data storage unit.

[0021] The construction of a digital twin model is completed in the following steps: Firstly, physical modeling is performed, assigning a unique shell identifier to each shell on the production line, which serves as the primary key for that shell in the digital twin model. At the same time, entity records are established for four types of physical entities: material batch, mold section, furnace batch and furnace position, and machining unit. The material batch record uses the material batch number as the primary key, the mold section record uses the mold number as the primary key, the furnace batch and furnace position record uses the combination of the furnace batch number and furnace position number as the primary key, and the machining unit record uses the equipment number as the primary key.

[0022] Secondly, relationship modeling is used to establish a relationship table between the shell identifier and the batch number, mold number, furnace number, furnace position number, and machining equipment number it passes through. This allows any shell record in the digital twin model to be traced back to all the physical entities it flows through on the production line along this relationship table, and also allows the entire shell that has passed through a certain physical entity to be retrieved in reverse.

[0023] Third, data field mapping: For the four types of physical entities, fields corresponding to the measured data are defined in the digital twin model: the batch record field corresponds to each measurement item of the batch's physicochemical indicators; the mold segment record field corresponds to the cumulative number of molded parts; the furnace position record field corresponds to the heating regime parameters included in the furnace position's thermal history identifier; and the machining unit record field corresponds to each measurement item of the shell's measured dimensions and coaxiality. A one-to-one mapping is established between the data output interfaces of the field measuring equipment (chemical composition analysis equipment, hardness testing equipment, online mold counting device, furnace temperature recorder, laser diameter gauge, and roundness meter) and the above fields, so that the measured data generated by the measuring equipment can be automatically written into the corresponding fields in a fixed format.

[0024] Fourth, the initialization and gradual filling of the shell record: When the shell first enters the first process of the production line, the digital twin model establishes a blank record for the shell containing only the shell identification number; thereafter, each time the shell flows through a process, the measuring device corresponding to that process writes the measured data into the corresponding component field of the shell record through the established field mapping, so that the spectral data of the shell record is gradually filled in as the shell progresses through the actual process on the physical production line. This filling rhythm is synchronized with the actual process flow of the shell.

[0025] The judgment logic of the digital twin model is not obtained through parameter iteration training of machine learning models such as neural networks, but is based on the statistical comparison of historical measured data and historical test results to establish the correspondence between shell lineage data and judgment criteria such as reachability level, performance gradient, and dispersion threshold. The model calibration mentioned in this embodiment refers to this correspondence establishment process based on historical data. The historical data used for calibration comes from the records of previously produced shells in the digital twin model, including its lineage data, intermediate judgment results at each stage, and the delivery level and measured ballistic performance that have been finally confirmed after inspection. This data has been continuously written into the lineage data storage unit during the previous batch production process as each unit is operating normally.

[0026] The calibration process specifically includes: First, based on historical grading records, establishing the correspondence between parameter ranges and achievable levels for batch components, mold components, furnace batch components, and machining components, forming a set of permissible levels for the achievable set calculation unit to call during operation; Second, based on the measured performance data of samples from furnaces that passed historical furnace batches, collecting data by furnace position and eliminating measurement noise, establishing performance gradient benchmarks for each furnace position for the furnace batch grouping unit to call during operation; Third, based on the pairing data of non-destructive proxies within historical compact clusters and measured ballistic performance, establishing judgment thresholds such as non-destructive proxies dispersion threshold, furnace performance dispersion bandwidth upper limit, and removable mass increment reference value, for the controllable window judgment unit, material removal grading unit, and destructive sampling unit to call during operation. The calibration results are stored in the spectral data storage unit in the form of parameter tables or correspondence tables, which can be directly read and called by each unit when processing new shell data during the operation phase. As a new batch of shells is inspected and new grading results and measured data are generated, these calibration results are continuously updated according to the correction mechanism, so that the judgment criteria of the digital twin model can continuously approach the actual state of the production line as production data accumulates.

[0027] Furthermore, the process of obtaining the reachable sets for the metallurgical and geometric dimensions is as follows: The digital twin model reads the physical and chemical indicators of the material batch, the cumulative number of molded parts for the mold section, the thermal history identifier of the furnace position, and the measured dimensions and coaxiality of each shell piece for the machining unit. These four items are respectively used as the spectral data of the material batch component, mold component, furnace component, and machining component. In the digital twin model, each component is classified according to the component spectral data and historical grading records, and the levels that the component can no longer reach are eliminated to obtain the set of permissible levels for the component. The intersection of the permissible level sets of the material batch component and the furnace component is taken to obtain the metallurgical dimension reachable set, and the intersection of the permissible level sets of the mold component and the machining component is taken to obtain the geometric dimension reachable set. Each time the shell flows through a process, the spectral data adds one component, and the digital twin model takes the intersection of the reachable sets again, so that the reachable set narrows as the spectral data accumulates.

[0028] Specifically, the genealogy data storage unit pre-stores historical grading records, which include the physical and chemical indicators of the batch of shells previously put into production, the cumulative number of molded parts, furnace position thermal history identifiers, machining dimensions and coaxiality data, as well as the final delivery grade determined after these shells have been inspected.

[0029] The reachability set calculation unit establishes a correspondence between parameter ranges and historical delivery levels for batch, mold, furnace, and machining components. Specifically, it uses historical spectral data of a component as a sample, divides the parameter values ​​of that component into several ranges according to their numerical range, and statistically analyzes which levels shells falling into each range were historically classified as. If the number of historical samples within a parameter range reaches a preset minimum sample size, and the empirical confidence probability of the shells within that range reaching the target level is not lower than a preset permissible threshold, then that level is retained in the permissible level set for that component. Otherwise, it is considered that the component status represented by that parameter range lacks reliable statistical reachability, and that level is marked as pending review or directly removed from the permissible level set for that component. Following this method, batch, mold, furnace, and machining components each obtain their respective permissible level sets.

[0030] The reachability set operation unit takes the intersection of the batch quantity allowable level set and the furnace quantity allowable level set (i.e., only retains the levels that exist in both sets simultaneously) to obtain the metallurgical dimension reachability set; and takes the intersection of the mold quantity allowable level set and the machining quantity allowable level set to obtain the geometric dimension reachability set.

[0031] As the shell passes through each process on the production line, the spectral data of its corresponding component is collected once and written into the spectral data storage unit. The reachability set operation unit then re-executes the above elimination and intersection operations for the newly added spectral data, so that the metallurgical dimension reachability set and the geometric dimension reachability set gradually narrow as the number of processes the shell goes through increases, and will not expand due to the addition of data, thereby ensuring that the reachability set always accurately reflects the actual range of levels that the shell can reach at the current process node.

[0032] Furthermore, the process of acquiring the controllable windows for the metallurgical and geometric dimensions is as follows: For the metallurgical dimension, the digital twin model uses whether the shell has not yet entered heat treatment as a criterion. When the shell has not yet entered heat treatment, the controllable window for the metallurgical dimension is set to open; once the shell has been tempered, the controllable window for the metallurgical dimension is set to close. For the geometric dimension, the digital twin model reads the current measured mass of the shell, the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, and the mass increment that can be removed by axisymmetric finishing within the tolerance. When the removable mass increment is not less than the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, the controllable window for the geometric dimension is set to open; when the removable mass increment is less than the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, the controllable window for the geometric dimension is set to close. The digital twin model only sends the reachable set corresponding to the dimension with the controllable window open to the corresponding furnace grouping or material removal classification.

[0033] Specifically, for the metallurgical dimension, the controllable window determination unit reads the process flow timestamp of the shell recorded in the spectrum data storage unit and determines whether the shell has recorded the start timestamp of the heat treatment process (tempering process). If the timestamp has not been recorded, that is, the shell has not yet entered the heat treatment, the controllable window status of its metallurgical dimension is set to open. Once the start timestamp of the tempering process corresponding to the shell is detected to be written into the spectrum data storage unit, the controllable window determination unit immediately sets the controllable window status of the shell's metallurgical dimension to closed. Once this status is closed, it will not be reset to open due to subsequent processes, thereby ensuring that the scheduling decision of the metallurgical dimension is only made during the stage when the metallurgical performance of the shell can still be adjusted through furnace grouping.

[0034] The controllable window determination unit, focusing on the geometric dimension, reads the current measured mass of the shell as measured by the machining unit, reads the boundary mass value between the shell's current grade and the adjacent lighter grade, calculates the difference between the two, and obtains the difference between the shell's current measured mass and the boundary of the adjacent lighter grade. The controllable window determination unit also reads a reference value for the mass increment that can be removed from the shell model within the tolerance range through axisymmetric finishing. This reference value is obtained in advance in one of the following ways: First, based on the tolerance range indicated on the shell model's drawings, combined with the shell material density, the theoretically maximum removable material mass is calculated while ensuring that the axisymmetric finishing part does not exceed the lower tolerance limit, and this is used as the reference value for the removable mass increment; Second, the records of actual mass removal in the historical finishing processes of the shell model are statistically analyzed, and the smallest removal amount is taken as a conservative reference value for the removable mass increment to avoid the shell exceeding the lower tolerance limit due to overestimation of the removal amount.

[0035] The controllable window determination unit compares the removable mass increment reference value with the difference between the current measured mass of the shell and the boundary value of the adjacent lighter grade: if the removable mass increment reference value is not less than the difference, it means that the shell mass can be reduced to the adjacent lighter grade after finishing, and the geometric controllable window state of the shell is set to open; if the removable mass increment reference value is less than the difference, it means that the existing finishing capability cannot reduce the shell mass to the adjacent grade, and the geometric controllable window state is set to closed.

[0036] The furnace grouping unit and the material removal and classification unit only read the metallurgical dimension reachable set and the geometric dimension reachable set corresponding to the shell with the metallurgical dimension controllable window in the open state, respectively, as the processing objects for subsequent furnace grouping or material removal and classification; for the shell with the controllable window in the closed state, no scheduling operation is performed on the corresponding dimension.

[0037] Furthermore, referring to Figure 2The process of obtaining the batch identity is as follows: In a shell with the metallurgical dimension controllable window open, the digital twin model encodes shells that are compatible with the metallurgical dimension reachable set and belong to the same batch into the same heat. The digital twin model aggregates the measured performance of samples taken from the furnace under similar loading conditions by furnace position to obtain the furnace position performance gradient, retaining only the portion of the furnace position performance gradient whose value exceeds the upper limit of measurement noise. The batch hardenability and forming state of each shell before entering the furnace are read as performance proxies before entering the furnace. Shells with lower performance proxies before entering the furnace are assigned to furnace positions with higher performance gradient values. Shells with high pre-furnace performance proxy values ​​are assigned to furnace positions with low furnace position performance gradient values ​​to obtain furnace position ranking. The furnace position ranking makes the difference in pre-furnace performance proxy and the difference in furnace position performance gradient opposite in direction within the same furnace batch. After the furnace batch exits, the digital twin model calculates the exit performance dispersion within the furnace batch using the measured exit performance of the furnace samples from the same furnace batch. For furnace batches with exit performance dispersion not exceeding the set bandwidth upper limit, all members within the furnace batch are marked as belonging to the same delivery batch. For furnace batches with exit performance dispersion exceeding the set bandwidth upper limit, all members within the furnace batch are placed in a state of pending encryption sampling.

[0038] Specifically, within the shell area where the metallurgical dimensional reachability window is open, the furnace grouping unit checks one by one whether there is a common level in the metallurgical dimensional reachability sets between any two shells, i.e., whether the intersection of the metallurgical dimensional reachability sets of two shells is non-empty; if a common level exists, the two shells are determined to be compatible in terms of metallurgical dimensional reachability sets. The furnace grouping unit further determines whether the shells belong to the same batch, and includes shells that simultaneously meet the two conditions of compatible metallurgical dimensional reachability sets and belonging to the same batch into the same furnace.

[0039] The furnace batch grouping unit reads the measured performance data of the samples taken from each furnace position under similar loading conditions, groups them by furnace position, and calculates the statistical quantity (such as the average value) of the measured performance of the samples taken from each furnace position to obtain the preliminary performance gradient value for each furnace position. To avoid misinterpreting measurement errors as true performance differences between furnace positions, the furnace batch grouping unit pre-sets an upper limit for measurement noise. This upper limit is obtained by repeatedly performing performance measurements on the samples taken from the same furnace position and the same batch, calculating the standard deviation of the repeated measurements, and setting the upper limit for measurement noise as a preset multiple of this standard deviation. This multiple is determined according to the required measurement confidence level; or the upper limit is set based on the measurement uncertainty index provided by the testing equipment manufacturer. The furnace batch grouping unit retains the portion of the preliminary performance gradient value for each furnace position that exceeds the upper limit for measurement noise as the valid performance gradient for that furnace position, while the portion that does not exceed the upper limit for measurement noise is not retained as a valid performance difference between furnace positions.

[0040] The furnace batch grouping unit reads the hardenability data and forming state data of each shell batch before it enters the furnace. After normalization, the data is weighted according to their respective historical influence weight coefficients, and the two are used together as the pre-furnace performance proxy for that shell. The furnace batch grouping unit sorts the shells to be put into the furnace in this furnace batch according to the pre-furnace performance proxy value from low to high, and sorts the furnace positions involved in this furnace batch according to the furnace position performance gradient value from high to low. According to the sorting results, shells with lower pre-furnace performance proxy values ​​are assigned to furnace positions with higher furnace position performance gradient values, and shells with higher pre-furnace performance proxy values ​​are assigned to furnace positions with lower furnace position performance gradient values, forming a furnace position ranking result. This ensures that the difference in pre-furnace performance proxy and the difference in furnace position performance gradient within the same furnace batch are in opposite directions, and the furnace position performance gradient compensates for the difference in pre-furnace performance of the shells.

[0041] After a furnace batch exits the furnace, the furnace batch grouping unit reads the measured performance data of the samples taken from that furnace batch and calculates the dispersion of the exit performance within that furnace batch (such as range or standard deviation) to obtain the exit performance dispersion. The furnace batch grouping unit pre-sets a bandwidth upper limit, which is obtained by: determining it based on the specific index requirements for the performance consistency of products in the same delivery batch as specified in the acceptance procedure; or statistically analyzing the distribution of exit performance dispersion values ​​of furnace batches that have passed inspection in the past, and taking the value corresponding to the preset confidence level in the distribution as the bandwidth upper limit. If the exit performance dispersion of a furnace batch is not greater than the set bandwidth upper limit, the furnace batch grouping unit marks all shells in that furnace batch as belonging to the same delivery batch; if the exit performance dispersion of a furnace batch is greater than the set bandwidth upper limit, the furnace batch grouping unit places all shells in that furnace batch into a state awaiting encrypted sampling and proceeds to the subsequent encrypted sampling process.

[0042] Furthermore, the grading result acquisition process is as follows: Read the current gaps for each grade, and select shells that simultaneously meet three conditions: the shell is located at the upper adjacent weight boundary, the shell's geometric dimension controllable window is open, and the shell's metallurgical dimension reachable set covers the adjacent lighter grade. For the selected shells, the digital twin model selects a material removal level from several material removal levels where only axisymmetric finishing is performed and the removal amount does not exceed the tolerance limit. This level reduces the shell's mass to fall into the adjacent gap grade. Material removal only changes the shell's axial mass and axial centroid, without changing the shell's wall thickness consistency or ballistic coefficient. Shells that have completed material removal are assigned to the adjacent gap grade, and shells that have not undergone material removal are assigned to their respective grades according to actual measurements. These are then combined to obtain the grading result. The current gaps for each grade are updated according to the number of shells assigned to the adjacent gap grades, and the updated current gaps for each grade are then fed back into the next round of furnace output.

[0043] Specifically, the material removal and grading unit reads the current shortage quantity for each grade. Within the shells with the geometrically controllable window open, the material removal and grading unit filters shells that simultaneously meet the following three conditions: the current measured mass of the shell is near the boundary between its grade and the adjacent lighter grade, i.e., it is at the upper adjacent weight boundary; the geometrically controllable window of the shell is open; and the metallurgical reachable set of the shell contains the adjacent lighter grade.

[0044] For the selected shells, the material removal and grading unit pre-establishes a material removal level table. This table records several sets of material removal levels and the corresponding removal mass for each level. The removal mass for each level does not exceed the upper limit of the allowable tolerance for the axisymmetric finishing process of that shell model. This upper limit of tolerance is obtained in the same way as the aforementioned reference value for removable mass increment, calculated based on the tolerance range indicated on the drawings and the material density, or determined based on the statistical results of measured removal amounts from historical finishing processes. The material removal and grading unit selects the level from the material removal level table that corresponds to the shell's mass falling within the adjacent notch level's mass boundary range after removal, and has the smallest removal mass value. This level is then used as the material removal level for that shell.

[0045] The material removal operation performed by the material removal and grading unit only removes material axially at the axially symmetrical parts of the shell (such as the bottom end face of the shell), without involving cutting in the radial wall thickness direction of the shell. Although material removal will cause slight changes in the axial mass and axial center of gravity of the shell, this material removal operation is strictly limited to the range allowed by the process drawings. After material removal, the digital twin model will be checked or triggered for sampling based on historical measured data to ensure that the mass, axial center of gravity position, dynamic imbalance, and external dimensions of the shell after material removal are all within the range allowed by the acceptance procedure for this model. This ensures that the change in ballistic performance caused by material removal does not exceed the preset acceptance tolerance threshold, thereby ensuring that the product meets the ballistic consistency index requirements of the target level.

[0046] The material removal and grading unit assigns shells that have completed material removal to their adjacent gap levels, and assigns shells not selected for material removal to their respective levels based on their measured mass. The two are combined to obtain the grading result for this round. The material removal and grading unit counts the number of shells assigned to each adjacent gap level, updates the current gap quantity for each level accordingly, and uses the updated current gap quantity for each level as the basis for the production output of the next furnace commissioning plan. This information is then transmitted to the furnace grouping unit to determine the commissioning ratio of shells of each level in the next round.

[0047] Furthermore, the process of performing representative destructive sampling of batch shells is as follows: Shells of the same delivery batch identity are clustered into several compact clusters based on consistency in three aspects: same material batch, same furnace position, and same material discharge position; when the dispersion of the non-destructive proxy of the shells within a compact cluster is not greater than a set threshold, a digital twin model extracts a representative shell from the compact cluster for destructive sampling test firing to obtain the measured ballistic performance of the representative shell; retain the non-destructive proxy recorded before the test firing of the representative shell, and pair the non-destructive proxy with the measured ballistic performance of the representative shell to obtain the correspondence between the non-destructive proxy and the measured ballistic performance within the compact cluster; the result of the destructive sampling test firing serves as the criterion for whether the entire delivery batch is qualified or not, and the number of samples in the destructive sampling test firing is not less than the minimum sample size specified in the acceptance procedure, and the number of destructive sampling samples of the delivery batch in the pending encrypted sampling state is higher than that of the delivery batch not in the pending encrypted sampling state.

[0048] Specifically, the destructive sampling unit clusters shells that have obtained the same delivery batch identity into several compact clusters based on the principle that the three conditions of the same material batch, the same furnace position, and the same material discharge position are all consistent.

[0049] The destructive sampling unit reads the non-destructive proxies of each shell within the compact cluster. These proxies are characteristic quantities (such as ultrasonic attenuation parameters and sound velocity) obtained through non-destructive testing methods (such as ultrasonic testing) that characterize the internal quality state of the shell. The unit then calculates the dispersion (such as range or standard deviation) of the non-destructive proxies within the cluster. The destructive sampling unit pre-sets a threshold, which is obtained by statistically analyzing the distribution of dispersion of non-destructive proxies in historical compact clusters that meet the ballistic performance requirements, based on established correlation data between non-destructive proxies and measured ballistic performance. The upper limit of this dispersion, which ensures consistency in ballistic performance within the cluster, is taken as the set threshold. This threshold can also be corrected by incorporating the measurement repeatability error of the non-destructive testing equipment itself to avoid misinterpreting detection errors as genuine performance differences between shells.

[0050] If the dispersion of the non-destructive proxy within a compact cluster is not greater than a set threshold, the destructive sampling unit selects a representative shell from the compact cluster. After selection and before test firing, it records the non-destructive proxy measurement value of the representative shell. Subsequently, a destructive sampling test firing is performed on the representative shell to obtain the measured ballistic performance of the representative shell. The destructive sampling unit pairs the non-destructive proxy measurement value recorded before the test firing of the representative shell with the measured ballistic performance to form a data pair, which serves as the correspondence between the non-destructive proxy within the compact cluster and the measured ballistic performance, for use by the subsequent batch reassignment unit. In particular, the correspondence between the non-destructive proxy and the measured ballistic performance is a conversion model and must be verified for error using an independent verification sample set before use. Only when the model's prediction conversion error for the verification sample is lower than a preset error upper limit is the correspondence considered statistically reliable and allowed for use in subsequent batches. If the error exceeds the limit, the transfer mapping is not performed, and instead, encrypted live-fire sampling is performed on the compact cluster.

[0051] The results of destructive sampling tests serve as the basis for determining the overall quality of the same delivery batch. The sample size for destructive sampling is no less than the minimum sample size specified in the acceptance procedure for that model of casing. This minimum sample size is determined directly based on the sampling plan specified in the current acceptance procedure or relevant inspection standards. For delivery batches in the pending encryption sampling state, due to the greater uncertainty in the consistency of their batch-out-of-batch performance, the destructive sampling unit arranges a higher number of destructive sampling samples for them than for delivery batches not marked as pending encryption sampling, in order to obtain more sufficient measured data to support the quality determination.

[0052] Furthermore, referring to Figure 3 The process of redistributing residual batches within the already qualified grade is as follows: For the shells that have not been extracted within the same compact cluster, the digital twin model uses the correspondence between the non-destructive proxy within the compact cluster and the measured ballistic performance to convert the non-destructive proxy remaining in the unextracted shells into the corresponding measured ballistic performance level of each unextracted shell; within the deliverable grade range of the unextracted shells that have already passed the qualification, the unextracted shells are classified into a finer grade according to the converted measured ballistic performance level, thus obtaining the residual batch redistribution result; the correspondence within the compact cluster, together with the deviation obtained by comparing the measured grade of each compact cluster with the spectral data, is converted into the component permissible grade set correction amount; the difference between the performance dispersion of each furnace output and the expected furnace position ranking is converted into the furnace position performance gradient correction amount; and the difference between the actual output of each grade after the residual batch redistribution and the grade-specific demand is converted into the furnace output correction amount.

[0053] Specifically, the batch redistribution unit reads the residual non-destructive proxy measurement values ​​of the shells that were not extracted for destructive testing within the same compact cluster. Using the correspondence between the non-destructive proxy and the measured ballistic performance established by the destructive sampling unit for the compact cluster, the non-destructive proxy measurement values ​​of the shells that were not extracted are compared and converted to obtain the corresponding measured ballistic performance level of the shell.

[0054] Within the deliverable grade range where the unextracted casings have been deemed qualified through destructive sampling test firing results, the reassignment unit, based on the measured ballistic performance levels obtained from the above conversion, classifies the unextracted casings into a grade that is more subdivided than the original grade, thus obtaining the reassignment result.

[0055] The batch redistribution unit compares the established non-destructive proxy to the actual ballistic performance correspondence within each compact cluster with the actual classification results obtained from the spectral data of the compact cluster shell, calculates the deviation between the two, and converts the deviation into a correction amount for the component permissible grade set. This correction is used to update the permissible grade sets of the batch component, mold component, furnace batch component, and machining component in the reachability set calculation unit, so that the subsequent reachability set determination of the shell more accurately reflects the actual classification capability of each component.

[0056] The residual batch redistribution unit converts the difference between the actual output performance dispersion of each furnace batch and the expected performance gradient of the furnace position arrangement stage into a furnace position performance gradient correction amount, which is used to update the furnace position performance gradient value of each furnace position in the furnace batch grouping unit, so that the furnace position arrangement of subsequent furnace batches is more in line with the actual heat treatment performance of each furnace position.

[0057] The residual batch redistribution unit converts the difference between the actual output quantity of each grade and the graded demand quantity after the residual batch redistribution into the furnace start-up production correction amount, which is used to update the next round of furnace start-up production obtained by the material removal and grading unit. This allows the production line start-up plan to be dynamically adjusted according to the actual grading and redistribution results, gradually approaching the actual demand of each grade.

[0058] This invention is based on the following process: after the shell is put into production, the digital twin model completes a round of data flow in the following order: the measured data generated by the field measurement equipment is written to the corresponding field of the shell record in the spectrum data storage unit via the industrial communication network; each time the spectrum data storage unit receives a new write, it sends a trigger signal to the reachability set operation unit; the reachability set operation unit reads all the existing spectrum data of the shell and the corresponding table of component permissible level sets established during the calibration stage, performs elimination and intersection operations, and writes the metallurgical dimension reachability set and geometric dimension reachability set obtained by the operation back to the reachability set field of the shell record; the controllable window determination unit reads the process flow timestamp and geometric dimension measured data of the shell, combines them with the reachability set field, determines the controllable window status of the metallurgical dimension and geometric dimension, and writes the status bit back to the shell record; the furnace grouping unit and the material removal and grading unit respectively read the controllable window status in the spectrum data storage unit. For shell records with an open state and their corresponding reachable sets, furnace grouping or material removal and grading calculations are performed. The calculated delivery batch identity and grading results are written back to the shell record. The material removal and grading unit simultaneously writes the current grade gap data into the next round of furnace production output parameter field. The destructive sampling unit reads the shell records with obtained delivery batch identity and grading results, performs compact clustering and destructive sampling, and writes the calculated non-destructive proxy to the correspondence of measured ballistic performance into the spectral data storage unit for subsequent use by the compact cluster. The residual batch redistribution unit reads the correspondence and the shell records not extracted within the compact cluster, performs subdivision grading, and converts the grading deviation, performance gradient deviation, and production output deviation into correction amounts, which are written back to the calibration parameter table called by the reachable set calculation unit, furnace grouping unit, and material removal and grading unit, completing one closed loop, so that the updated calibration parameters can be called when the next batch of shells is put into production. During the aforementioned data flow process, data is transferred between units only by reading and writing predetermined fields in the spectral data storage unit. There is no direct data exchange between units that bypasses the spectral data storage unit. This ensures that the spectral data, reachability set, controllable window status, and grading results recorded by any shell at any time can be completely traced in the spectral data storage unit.

[0059] Example 2: Example 1 has already given a detailed implementation of the complete process of reachable set acquisition, controllable window determination, delivery batch identity determination, classification result acquisition, destructive sampling and residual batch redistribution. Based on this, this example focuses on further explaining the real-time early warning of the heating process and the material removal capacity constraint sorting. The rest of the non-repeated parts follow the description of Example 1.

[0060] Furthermore, during the heating process after the furnace is loaded, the digital twin model collects the current furnace temperature curve of the furnace position in real time, compares the current furnace temperature curve with the heating curve of the historical qualified furnace recorded in the thermal history identifier of the furnace position, and obtains the real-time deviation. The digital twin model presets the upper limit of the process deviation. When the real-time deviation exceeds the upper limit of the process deviation, the furnace is judged to have an abnormal thermal history risk, and the metallurgical controllable window status of all shells in the furnace is set to the pending review status in advance.

[0061] Specifically, after the furnace is loaded, the furnace temperature recorder continuously collects real-time furnace temperature values ​​at the furnace location at preset time intervals. This time interval is determined based on the furnace temperature recording frequency specified in the heat treatment process procedure for that type of shell, or set according to the maximum sampling capacity of the furnace temperature recorder itself. The collection frequency is not lower than the minimum frequency required for furnace temperature fluctuation detection allowed by the heat treatment process for that type of shell. The digital twin model arranges the continuously collected real-time furnace temperature values ​​in chronological order of collection time, forming a real-time furnace temperature curve for the current heating process of that furnace.

[0062] The digital twin model retrieves heating curve records from the furnace location's thermal history identifiers of multiple furnace runs that have been previously deemed to have passed performance testing. It calculates the average value of these qualified furnace runs' heating curves at the same heating time point to obtain the historical qualified heating curve for the furnace location, serving as a comparison benchmark. The digital twin model then compares the current furnace temperature curve with the historical qualified heating curve at the same heating time point point by point, obtaining the deviation value at each time point. The largest deviation is taken as the current real-time deviation for that furnace run. The digital twin model presets a process deviation upper limit, which is obtained by: statistically analyzing the distribution of furnace temperature curve deviations relative to each other during the heating process of multiple furnace runs that have been deemed to have passed performance testing in the past, and taking the upper limit of the deviation in this distribution as the process deviation upper limit; or by comprehensively determining the upper limit based on the measurement accuracy indicators of the furnace temperature recorder and the specific requirements for heating uniformity in the heat treatment process specifications for this type of shell.

[0063] When the real-time deviation exceeds the upper limit of the process deviation, the digital twin model determines that the furnace has an abnormal thermal history risk and sets the metallurgical dimension controllable window status of all shells in the furnace to a pending review state in advance; for furnaces where the real-time deviation does not exceed the upper limit of the process deviation, the metallurgical dimension controllable window status remains in normal judgment.

[0064] Furthermore, for furnace batches placed in a pending review state, the number of destructive sampling samples arranged by the destructive sampling unit within the corresponding compact cluster of that furnace batch is higher than the number of destructive sampling samples of the corresponding compact cluster of furnace batches not placed in a pending review state; the delivery batch identity determination of the inner shell of that furnace batch remains in a pending review state and is not confirmed until the destructive sampling unit completes the destructive sampling test firing of the corresponding compact cluster and confirms that the furnace performance dispersion is not greater than the set bandwidth limit.

[0065] Specifically, for furnaces placed in a pending review status, after the furnaces are produced, they still form compact clusters and the performance dispersion is calculated. However, when the destructive sampling unit arranges the number of destructive sampling samples for the corresponding compact clusters of the furnace, it adds samples to the minimum sample size specified in the acceptance procedure. The additional sample size is determined based on the sampling plan table corresponding to the increase in confidence level in the acceptance procedure, or based on the proportion of furnaces in the pending review status that actually failed to meet the standards in the history of the furnace, and the additional sample size is determined according to the value in the sampling plan table corresponding to that proportion, so that the number of additional samples is higher than the number of samples for the corresponding compact clusters of furnaces that were not placed in a pending review status.

[0066] Before the sampling unit completes the encrypted sampling test of the corresponding compact cluster and verifies that the performance dispersion of the furnace output is not greater than the set bandwidth limit, the digital twin model keeps the delivery batch identity of the inner shell of the furnace in a pending verification state and does not write it into the delivery batch identity confirmation record; after the verification is passed, it is officially marked as the same delivery batch identity.

[0067] Furthermore, the material removal and grading unit obtains the upper limit of available capacity of the finishing equipment in the current production cycle. When the number of candidate shells that simultaneously meet the material removal conditions exceeds the upper limit of available capacity, the material removal and grading unit uses the difference between the reference value of its removable mass increment and the difference between the current measured mass of the shell and the boundary value of the adjacent lighter grade for each candidate shell as the margin value of the shell. It also reads the current number of gaps in the adjacent gap grade to which the shell is to be downgraded and the remaining number of production cycles for that grade. The current number of gaps divided by the remaining number of production cycles is used as the gap urgency of that grade. The material removal and grading unit sorts the candidate shells in order of margin value from small to large and gap urgency from large to small. Within the available capacity range of the current production cycle, the material removal operation is performed on the candidate shells ranked higher according to the sorting results. Candidate shells that do not obtain capacity in the current cycle are retained to participate in the sorting again in the next production cycle.

[0068] Specifically, at the beginning of each production cycle, the material removal and grading unit reads the upper limit of the available capacity of the finishing equipment in that cycle. The method of obtaining this capacity is to calculate the upper limit of the number of shells that can be removed in that cycle based on the rated processing capacity of the equipment and the production time that can be scheduled in that cycle, according to the standard time required for the finishing of a single shell in the process specification.

[0069] When the number of candidate shells selected based on the three criteria exceeds the production capacity limit, the material removal and grading unit uses the difference between the reference value of the removable mass increment of each candidate shell and the difference between the shell's current measured mass and the boundary value of the adjacent lighter grade as the margin value for that shell. The smaller the margin value, the closer the geometric controllable window of the shell is to the closing boundary. The material removal and grading unit also reads the current number of gaps in the adjacent gap grade to which the shell is to be downgraded and the remaining number of production cycles for that grade. The ratio of the current number of gaps to the remaining number of production cycles is used as the gap urgency for that grade.

[0070] The material removal and grading units are sorted by margin value from smallest to largest as the first sorting criterion. If the margin values ​​are the same, they are sorted by the urgency of the gap from largest to smallest as the second sorting criterion, thus obtaining the processing priority order of the candidate shells. Within the available capacity range of the current cycle, the candidate shells with the highest ranking are arranged for material removal processing in sequence. Candidate shells that exceed the capacity range of the current cycle are not processed in the current cycle and are retained to participate in the screening and sorting again in the next production cycle.

[0071] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A collaborative scheduling method for a shell casing production line based on digital twins, characterized in that, include: A digital twin model of the shell and the process is established using measured data from the shell production line process; The material batch, mold section, furnace position and machining unit of the shell are used as spectral constraints to propagate along the process. The unreachable level is eliminated by each component and the intersection is taken to obtain the reachable set and controllable window of metallurgical dimension and geometric dimension. The process of acquiring controllable windows for the metallurgical and geometric dimensions is as follows: For the metallurgical dimension, the digital twin model uses whether the shell has not yet entered heat treatment as a criterion. When the shell has not yet entered heat treatment, the controllable window for the metallurgical dimension is set to open; once the shell has been tempered, the controllable window for the metallurgical dimension is set to close. For the geometric dimension, the digital twin model reads the current measured mass of the shell, the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, and the mass increment that can be removed by axisymmetric finishing within the tolerance. When the removable mass increment is not less than the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, the controllable window for the geometric dimension is set to open; when the removable mass increment is less than the difference between the current measured mass of the shell and the boundary of the adjacent lighter grade, the controllable window for the geometric dimension is set to close. The digital twin model only sends the reachable set corresponding to the dimension with the controllable window open to the corresponding furnace grouping or material removal and grading. Based on metallurgical dimensional reachability sets and controllable windows, shells of the same batch and with compatible grades are grouped into the same heat in the digital twin model, and ranked according to the performance gradient of the same type of furnace under the same loading condition as verified by actual furnace exit to limit the dispersion within the heat and obtain the identity of the delivery batch. Based on the geometric dimension reachable set and controllable window, in the digital twin model, for the shell that can monotonically descend into the adjacent level with the level gap and the metallurgical dimension reachable set covers the level, the material removal position is assigned and merged into the level to obtain the grading result, and the remaining level gap is fed back into the furnace output. Shells from the same delivery batch are grouped into compact clusters based on three criteria: same material batch, same furnace position, and same material removal position. When the dispersion of the non-destructive proxy of the shells within a compact cluster is not greater than a set threshold, the digital twin model takes out a representative shell from the compact cluster for destructive sampling test firing to obtain the measured ballistic performance of the representative shell. Based on the delivery batch identity and grading results, spectral representative destruction sampling is performed on the batch shells. The correspondence between non-destructive proxies and measured ballistic performance is migrated within the compact cluster, and residual batches are redistributed within the qualified grade.

2. The collaborative scheduling method for a shell casing production line based on digital twins according to claim 1, characterized in that, The process of obtaining the reachable sets for metallurgical and geometric dimensions is as follows: for the material batch, read the physical and chemical indicators of the material batch; for the mold section, read the cumulative number of molded parts; for the furnace position, read the furnace position thermal history identifier; for the machining unit, read the measured dimensions and coaxiality of each shell piece. These four items are used as the spectral data for the material batch component, mold component, furnace component, and machining component, respectively. In the digital twin model, each component is processed according to the component spectral data and historical grading records. The levels that the component can no longer reach are eliminated to obtain the set of permissible levels for the component. The intersection of the permissible level sets of the material batch component and the furnace component is taken to obtain the metallurgical dimension reachable set. The intersection of the permissible level sets of the mold component and the machining component is taken to obtain the geometric dimension reachable set. For each process through which the shell flows, the spectral data adds one component. The digital twin model takes the intersection of the reachable sets again, so that the reachable set narrows as the spectral data accumulates.

3. The collaborative scheduling method for a shell casing production line based on digital twins according to claim 1, characterized in that, The process of acquiring the batch identity is as follows: Within a shell where the metallurgical dimensional controllable window is open, the digital twin model encodes shells that are compatible with the metallurgical dimensional reach and belong to the same batch into the same heat. The digital twin model then aggregates the measured performance of samples taken from the furnace under similar loading conditions by furnace position, obtaining the furnace position performance gradient, and retaining only the portion of the gradient exceeding the upper limit of measurement noise. The batch hardenability and forming state of each shell before entering the furnace are read as performance proxies before entering the furnace, and shells with lower performance proxies before entering the furnace are assigned to the furnace. For furnaces with high performance gradient values, shells with high performance proxy values ​​before entering the furnace are assigned to furnaces with low performance gradient values ​​to obtain furnace position ranking. After the furnace exits, the digital twin model calculates the exit performance dispersion within the furnace based on the measured exit performance of the furnace samples from the same furnace. For furnaces with exit performance dispersion not exceeding the set bandwidth limit, all members within the furnace are marked as belonging to the same delivery batch. For furnaces with exit performance dispersion exceeding the set bandwidth limit, all members within the furnace are placed in a state of pending encryption sampling.

4. The collaborative scheduling method for a shell casing production line based on digital twins according to claim 1, characterized in that, The process of obtaining the grading results is as follows: read the current gap of each grade, and select the shell that meets the three conditions at the same time. The three conditions are that the shell is located on the upper adjacent weight boundary, the controllable window of the shell's geometric dimension is open, and the reachable set of the shell's metallurgical dimension covers the adjacent lighter grade. For the selected shells, the digital twin model selects a material removal level from several material removal positions where only axisymmetric finishing is performed and the removal amount does not exceed the upper tolerance limit. The material removal level is selected so that the shell mass is monotonically reduced to fall into the adjacent gap level. Material removal only changes the axial mass and axial centroid of the shell, without changing the shell wall thickness consistency and ballistic coefficient. The shells that have completed material removal are assigned to the adjacent gap level, and the shells that have not been removed are assigned to their respective levels according to actual measurements. The grading results are then merged. The current gap of each level is updated according to the number of shells assigned to the adjacent gap level, and the updated current gap of each level is fed back into the production output of the next furnace cycle.

5. The collaborative scheduling method for a shell casing production line based on digital twins according to claim 1, characterized in that, The process of sampling representative damage of batch shells is as follows: retain the non-destructive proxy recorded before the test firing of the representative shell, pair the non-destructive proxy with the measured ballistic performance of the representative shell, and obtain the correspondence between the non-destructive proxy and the measured ballistic performance within the compact cluster; The results of destructive sampling tests are used as the criterion for determining whether the entire delivery batch is qualified. The number of samples in destructive sampling tests shall not be less than the minimum sample size specified in the acceptance procedure. The number of destructive sampling samples in the delivery batch that is in the state of pending encryption sampling is higher than that in the delivery batch that is not in the state of pending encryption sampling.

6. The collaborative scheduling method for a shell casing production line based on digital twins according to claim 5, characterized in that, The process of redistributing residual batches within the already qualified grade is as follows: For the shells that have not been extracted within the same compact cluster, the digital twin model uses the correspondence between the non-destructive proxy within the compact cluster and the measured ballistic performance to convert the non-destructive proxy remaining in the unextracted shells into the corresponding measured ballistic performance level of each unextracted shell; within the deliverable grade range of the unextracted shells that have already qualified, the unextracted shells are classified into a finer grade according to the converted measured ballistic performance level, thus obtaining the residual batch redistribution result; the correspondence within the compact cluster, together with the deviation obtained by comparing the measured grade of each compact cluster with the spectral data, is converted into the component permissible grade set correction amount; the difference between the performance dispersion of each furnace output and the expected furnace position ranking is converted into the furnace position performance gradient correction amount; and the difference between the actual output of each grade after the residual batch redistribution and the grade-specific demand is converted into the furnace output correction amount.

Images