Digital twin model for nuclear grade additive manufacturing process

Abstract

The invention relates to the technical field of digital twinborn models, and discloses a digital twinborn model for a nuclear-grade additive manufacturing process, and a model construction method comprises the following steps: dividing process state units according to a preset rule and constructing feature vectors based on a standardized manufacturing process data set obtained by preprocessing original manufacturing data; performing backtracking consistency evaluation on the process state unit and calculating a comprehensive consistency evaluation value under the constraint of a process, a material and equipment based on an association probability model constructed by historical data; and screening candidate process state units, and outputting a manufacturing sequence formed by defects and corresponding process state information. According to the method, quantitative tracing of the nuclear-grade additive manufacturing defect forming stage can be achieved, reliable support is provided for quality responsibility definition and process improvement, and nuclear-grade manufacturing quality control requirements are met.

Description

Technical Field

[0001] This invention relates to the field of digital twin model technology, specifically a digital twin model of a nuclear-grade additive manufacturing process. Background Technology

[0002] Nuclear-grade additive manufacturing technology is widely used in the manufacture of critical components such as nuclear fuel element support structures, lattice-type structural components, and internal functional structures of the cladding. These components have extremely high requirements for manufacturing consistency and quality traceability. Nuclear-grade additive manufacturing typically employs a layer-by-layer deposition and curing process. During the manufacturing process, factors such as accumulated heat input, overlapping scan paths, and fluctuations in process parameters can easily lead to defects such as localized incomplete fusion, abnormal density, and abnormal microstructure.

[0003] These types of defects are irreversible and insidious. Once formed, they are covered by subsequent cladding or thermal cycling, making them difficult to identify in a timely manner through online monitoring or intermediate testing. They are often only discovered during the non-destructive testing stage of the finished product or even during in-core operation. In existing technologies, defect detection can only obtain the final result and cannot determine the specific layer sequence of defect formation, scanning strategy, and process parameter range, lacking an effective correlation analysis mechanism.

[0004] When handling quality anomalies, process engineers can only rely on experience to trace back historical records, making it difficult to quantify the cause of defects and the effectiveness of parameter adjustments. This leads to difficulties in defining quality responsibility in nuclear-grade manufacturing scenarios, hinders the formation of a closed loop for process improvement, and prevents the standardized inheritance of manufacturing experience. Summary of the Invention

[0005] The purpose of this invention is to provide a digital twin model of the nuclear-grade additive manufacturing process to solve the problems mentioned in the background art.

[0006] According to one aspect of this application, a digital twin model of a nuclear-grade additive manufacturing process is provided, comprising the following steps: Based on a standardized manufacturing process dataset, the manufacturing process is divided into multiple process state units according to preset rules, and a process state feature vector is constructed for each process state unit; the standardized manufacturing process dataset is obtained by preprocessing the original manufacturing data. Based on the correlation probability output by the correlation probability model, a retrospective consistency assessment is performed on each process state unit under process constraints, material constraints, and equipment constraints, and a comprehensive consistency evaluation value is calculated for each process state unit. The correlation probability model is constructed based on process state feature vectors and corresponding defect type labels from historical manufacturing data, used to characterize the correlation between process states and defect types. At least one candidate process state unit is selected based on the comprehensive consistency evaluation value. Based on the candidate process state units, the manufacturing sequence of defect formation and corresponding process state information are output.

[0007] Preferably, the preset rule is as follows: within the same manufacturing layer, if the standard deviation of process parameters, the change of scanning strategy identifier, and the fluctuation of equipment operating status do not exceed a preset stability threshold, then the layer is divided into a process state unit; between adjacent manufacturing layers, if the overlap of process parameter intervals exceeds a preset ratio, the scanning strategy identifier is consistent, and the difference in equipment operating status statistics does not exceed the equipment state stability threshold, then the adjacent layers are merged into a process state unit.

[0008] Preferably, the process state feature vector includes the following components: a layer sequence feature component, used to identify the manufacturing layer range corresponding to the process state unit; a scanning strategy feature component, used to represent the scanning path type and combination method; a process parameter range feature component, used to represent the range of laser power, scanning speed, and layer thickness parameters; and a device operating status feature component, used to represent the statistical indicators of device operating stability. After normalization, each component is spliced ​​into a feature vector of a unified dimension in a preset order.

[0009] Preferably, constructing the association probability model includes taking process state feature vectors as input and defect type labels as output, and based on the training sample set, calculating the co-occurrence frequency of each process state feature vector and each defect type label; and based on the co-occurrence frequency, calculating the association probability value of each process state feature vector leading to each defect type.

[0010] Preferably, the comprehensive consistency evaluation value is obtained by weighted summation of the following three scores: the constraint satisfaction score of the process unit; the correlation probability score between the process unit and the current defect type output by the correlation probability model; and the average feature difference score between the feature vector of the process unit and the feature vectors of other process units.

[0011] Preferably, the average feature difference score is calculated as follows: calculate the cosine similarity between the feature vector of the process state unit and the feature vector of each of the other process state units, then average all the cosine similarities, and use 1 minus the average value as the average feature difference score.

[0012] Preferably, the comprehensive consistency evaluation values ​​of each process state unit are sorted in descending order, and at least one process state unit with a comprehensive consistency evaluation value higher than a preset threshold or ranked higher is selected as a candidate process state unit.

[0013] In another aspect, this application also provides a digital twin model construction system for a nuclear-grade additive manufacturing process, comprising: The process state feature construction module is used to divide the manufacturing process into multiple process state units based on a standardized manufacturing process dataset and according to preset rules, and to construct a process state feature vector for each process state unit; the standardized manufacturing process dataset is obtained by preprocessing the original manufacturing data; The retrospective consistency assessment module is used to perform retrospective consistency assessment on each process state unit under process constraints, material constraints, and equipment constraints, based on the correlation probability output by the correlation probability model. The module calculates the comprehensive consistency evaluation value for each process state unit. The correlation probability model is constructed based on the process state feature vector and corresponding defect type label in historical manufacturing data, and is used to characterize the correlation between process state and defect type. The candidate process state screening module is used to screen at least one candidate process state unit based on the comprehensive consistency evaluation value. The defect tracing and output module is used to output the manufacturing sequence and corresponding process state information of defect formation based on the candidate process state unit.

[0014] This application also provides a computer device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the digital twin model construction method for the nuclear-level additive manufacturing process as described above.

[0015] In another aspect, this application provides a computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor to implement the digital twin model construction method for the nuclear-level additive manufacturing process as described above.

[0016] Another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements a digital twin model construction method for a core-level additive manufacturing process as described above.

[0017] This application achieves quantitative traceability analysis of the defect formation stage in nuclear-grade additive manufacturing through core innovative steps such as process state unit division and feature vector construction, correlation probability model construction, backtracking consistency assessment, and candidate process state screening, without introducing new detection equipment or changing the existing manufacturing process. This method discretizes the continuous manufacturing process into traceable process state units, establishes the correlation between process states and defect types through statistical analysis, and conducts comprehensive consistency assessment under process, material, and equipment constraints. It effectively solves the technical problems in nuclear-grade additive manufacturing, such as the strong concealment of local irreversible defects, the disconnect between the formation and discovery stages, and the lack of traceable quantitative analysis methods for their causes. The output defect formation manufacturing sequence and corresponding process state information provide reliable data support for defining quality responsibility in nuclear-grade manufacturing, closing the loop of process improvement, and standardizing the inheritance of manufacturing experience. It also avoids the problems of existing online monitoring lacking nuclear-grade sensors and the large computational load and unverifiable nature of physical simulation, demonstrating high engineering practicality and adaptability to nuclear-grade quality systems. Attached Figure Description

[0018] Figure 1 A schematic diagram illustrating a method for constructing a digital twin model of a nuclear-grade additive manufacturing process, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the process state feature vector construction flow provided in the embodiments of this disclosure; Figure 3 A schematic diagram illustrating the construction process of the association model provided in this embodiment of the disclosure; Figure 4 This is a schematic diagram of the association probability model training process provided in the embodiments of this disclosure; Figure 5 This is a schematic diagram of the backtracking consistency assessment process provided in the embodiments of this disclosure; Figure 6 A schematic diagram of a digital twin model construction system for a nuclear-grade additive manufacturing process is provided in this embodiment of the invention. Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] It should be noted that all user information (including but not limited to user device information, user personal information, object information corresponding to device usage data, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, device usage data, etc.) involved in all embodiments of this disclosure are information and data authorized by the user or fully authorized by all parties.

[0021] The digital twin model of the nuclear-grade additive manufacturing process described in this invention is applicable to manufacturing scenarios for nuclear-grade components, such as nuclear fuel element support components, lattice-type structural components, and internal functional structures of cladding, where manufacturing consistency and quality traceability are extremely important. This model is deployed on the manufacturing execution system server, quality analysis server, or equivalent computing platform within the manufacturing enterprise. It typically relies on historical data automatically recorded during the manufacturing process and finished product quality inspection results. Without altering the existing manufacturing hardware system, it enables traceability analysis of the defect formation stages of completed nuclear-grade additive manufacturing components, providing data support for quality responsibility determination and closed-loop process improvement.

[0022] The following detailed description, in conjunction with specific embodiments, illustrates the digital twin model construction process of the nuclear-grade additive manufacturing process of the present invention. It should be noted that these embodiments are merely illustrative of the invention and not intended to limit its scope. Any conventional adjustments or substitutions made by those skilled in the art to the steps without departing from the inventive concept should be included within the scope of protection of the present invention.

[0023] like Figure 1 As shown in the diagram, this invention discloses a method for constructing a digital twin model of a nuclear-grade additive manufacturing process, comprising the following steps: S1, based on a standardized manufacturing process dataset, the manufacturing process is divided into multiple process state units according to preset rules, and a process state feature vector is constructed for each process state unit; the standardized manufacturing process dataset is obtained by preprocessing the original manufacturing data; S2, based on the correlation probability output by the correlation probability model, performs a retrospective consistency assessment on each process state unit under process constraints, material constraints, and equipment constraints, and calculates the comprehensive consistency evaluation value of each process state unit; the correlation probability model is constructed based on the process state feature vector and corresponding defect type label in historical manufacturing data, and is used to characterize the correlation between process state and defect type; S3, at least one candidate process state unit is selected based on the comprehensive consistency evaluation value; S4. Based on the candidate process state unit, output the manufacturing sequence of defect formation and the corresponding process state information.

[0024] In some embodiments, for step S1, firstly, manufacturing process data preprocessing is the basic step in model building. Its core purpose is to transform the raw manufacturing data scattered in different subsystems of the manufacturing execution system into a standardized manufacturing process dataset, ensuring the integrity, consistency and validity of the data, and providing reliable input for subsequent process state unit partitioning and feature vector construction.

[0025] According to an embodiment of this disclosure, the executing entity establishes a data communication connection with the manufacturing execution system of the additive manufacturing equipment to obtain the original manufacturing data corresponding to the unique identifier of the component to be analyzed. The original manufacturing data consists of historical data automatically recorded during the manufacturing process, specifically including four types of data: manufacturing layer sequence data, used to characterize the layer numbers and their sequential relationships of each deposition layer, including, for example, layer sequence number, layer deposition timestamp, and interlayer interval duration, with layer numbers increasing sequentially according to the manufacturing order to ensure the uniqueness of each layer's sequence; scanning strategy data, used to characterize the scanning path type and path combination method corresponding to each layer, including, for example, scanning path identifier, scanning direction code, and path overlap rate annotation; process parameter recording data, used to characterize the value range of parameters such as laser power, scanning speed, and layer thickness during the manufacturing process of each layer, including, for example, recording the real-time acquired values ​​of each parameter in time intervals to form a continuous parameter sequence; and equipment operating status data, used to characterize statistical data related to the stability of equipment operation during the manufacturing process, including, for example, temperature fluctuation values ​​in the temperature zone, position deviation statistics of mechanical actuators, and laser output stability indicators.

[0026] The acquired four types of raw manufacturing data undergo format standardization processing. Based on preset field mapping rules, data fields from different sources are uniformly mapped to a preset set of standard fields. For example, fields such as "laser power output value" and "laser energy intensity" defined by different equipment manufacturers are uniformly mapped to the standard field "laser power parameter." For continuous process parameter data, it is segmented and aggregated according to the manufacturing layer sequence, and the value range of each parameter is statistically analyzed on a layer-by-layer basis to form interval-based data. For discrete scanning strategy data, a unified representation is achieved using identifier encoding. For example, "unidirectional scanning path" is encoded as a specific identifier, and "bidirectional cross-scanning path" is encoded as another specific identifier, ensuring that different types of scanning strategies can be quantified. Through the above processing, raw data with different structures, sampling periods, and recording methods are converted into structured data table format, achieving data format uniformity.

[0027] After standardizing the data format, missing or obvious outliers in the data are processed according to the following rules: If a critical field is missing from the process parameter record corresponding to a certain layer, and the duration of the missing field exceeds a preset proportion of the total manufacturing time of that layer (e.g., 10%), then that layer is marked as unusable and will be directly excluded in subsequent process state construction and analysis. If a process parameter record in a certain layer contains values ​​that exceed the allowable range of the equipment, such as laser power exceeding the upper or lower limit of the equipment's rated output power or scanning speed exceeding the maximum operating speed of the equipment's mechanical structure, then the corresponding process state is marked as an abnormal state, and the specific values ​​of the abnormal parameters and their timestamps are recorded. For instantaneous fluctuations in the equipment operating status data, if the fluctuation amplitude does not exceed a preset stability threshold (e.g., temperature fluctuation in the temperature zone does not exceed ±2℃), then the data is retained and included in the calculation of statistical characteristics. If the fluctuation amplitude exceeds the stability threshold and the duration exceeds a preset duration (e.g., 5 seconds), then the data segment is marked as abnormal and will be considered separately when statistically analyzing the characteristics of the equipment operating status. After processing, a standardized manufacturing process dataset is formed. This dataset is based on the manufacturing layer and contains standardized field information, data validity indicators, and abnormal status annotations for each layer.

[0028] In one embodiment, process state unit partitioning and feature vector construction solve the technical problems of continuous manufacturing processes and difficulty in accurately locating defect formation stages in the prior art. The core principle is that in nuclear-grade additive manufacturing, when process parameters, scanning strategies, and equipment states remain stable in the same or adjacent layers, the manufactured component regions have similar quality characteristics. Defect formation is directly related to the process state of this stable stage. Therefore, by partitioning process state units, the defect formation range can be locked within a specific stable process stage. Furthermore, by constructing standardized feature vectors, the process state units become calculable and comparable, laying the foundation for subsequent correlation modeling and backtracking reasoning.

[0029] Specifically, the process state unit partitioning uses the manufacturing layer sequence as the basic index unit and is executed according to preset rules. These preset rules are specifically two-layer partitioning logics, one for the same manufacturing layer and the other for adjacent manufacturing layers: The process state unit division logic is as follows: For a single manufacturing layer, the real-time sequence of process parameters, scanning strategy identifiers, and equipment operating status data during the manufacturing process of that layer are first collected, and the maximum, minimum, average, and standard deviation of the process parameters are calculated. Preset stability thresholds include process parameter stability thresholds, scanning strategy consistency thresholds, and equipment status stability thresholds. Specifically, the process parameter stability thresholds are used to determine the variation range of parameters such as laser power, scanning speed, and layer thickness. For example, the stability thresholds for laser power are set to ±5%, scanning speed to ±3%, and layer thickness to ±2%. The scanning strategy consistency thresholds are binary thresholds; if the scanning strategy identifiers are consistent, the threshold requirement is met; otherwise, it is not met. The equipment status stability thresholds are used to determine the variation range of equipment operating status parameters such as temperature zone stability indicators and execution deviation statistics. For example, the stability thresholds for temperature zone stability indicators are set to ±1℃, and the stability thresholds for execution deviation statistics are set to ±0.1mm.

[0030] If the standard deviation of the process parameters within a layer does not exceed the preset process parameter stability threshold, and the scanning strategy identifier remains unchanged, and the fluctuation range of the equipment operating status data does not exceed the equipment status stability threshold, then the layer is considered a single process state unit. If the standard deviation of a certain process parameter within the layer exceeds the process parameter stability threshold, or the scanning strategy identifier changes, or the fluctuation range of the equipment operating status data exceeds the equipment status stability threshold, then the layer is divided into multiple sub-layers according to the change node, with each sub-layer corresponding to a process state unit. For example, if a layer uses a "unidirectional scanning path" in the first half of manufacturing and switches to a "bidirectional cross-scanning path" in the second half, then the layer is divided into two process state units based on the scanning strategy switching time point. Each process state unit corresponds to a scanning strategy and its corresponding process parameters and equipment status.

[0031] The logic for dividing process state units between adjacent manufacturing layers is as follows: For two adjacent layers, the process parameter ranges, scanning strategy identifiers, and equipment operating status statistics of the preceding and following layers are compared. A preset overlap ratio for process parameter ranges is used, for example, 90%. If the overlap ratio of process parameter ranges between two adjacent layers exceeds this preset ratio, and the scanning strategy identifiers are consistent and the difference in equipment operating status statistics does not exceed the equipment state stability threshold, then these two layers are merged into one process state unit. If the overlap ratio of process parameter ranges between two adjacent layers is lower than the preset ratio, or the scanning strategy identifiers are inconsistent, or the difference in equipment operating status statistics exceeds the equipment state stability threshold, then the two layers are divided into independent process state units. For example, if the laser power range of the Nth layer and the N+1th layer are both [P1-P2], the scanning strategy is both "bidirectional cross-scanning path", and the difference in the equipment temperature stability index is 0.8℃, which does not exceed the equipment stability threshold of ±1℃, then the Nth layer and the N+1th layer are merged into one process state unit; if the laser power range of the N+1th layer becomes [P3-P4], and the overlap with the Nth layer is only 50%, which is lower than the preset ratio of 90%, then the N+1th layer is divided into an independent process state unit.

[0032] After the process state units are divided, a unique process state identifier is assigned to each process state unit. This identifier includes hierarchical range information, such as "layer N - layer N+1", unit number, and stability identifier. The stability identifier is determined based on whether there are abnormal state markings in the layer corresponding to the process state unit. Simultaneously, auxiliary information such as the manufacturing time range and the corresponding component area location of each process state unit is recorded to provide a reference for matching defect locations with process state units. This division method can accurately characterize the stable process stages in the manufacturing process, discretizing the continuous manufacturing process into multiple independent, traceable process state units, significantly reducing the search range for defect formation and improving the accuracy of subsequent defect tracing.

[0033] In one embodiment, the purpose of constructing the process state feature vector is to transform the unstructured state information of each process state unit into a computable and comparable feature vector form, which includes hierarchical feature components, scanning strategy feature components, process parameter range feature components, and equipment operating state feature components. Each component is normalized and then concatenated in a preset order to form a feature vector of uniform dimension. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a schematic diagram illustrating the process of constructing process-state feature vectors according to embodiments of this disclosure. The specific construction process is as follows: In S201, the sequence feature components are constructed. These components identify the manufacturing layer range corresponding to the process state and are constructed using sequence coding and layer number representation. For process state units corresponding to layers N to N+1, the sequence coding uses a combination of the layer start number and the layer end number, i.e., "N-N+1", and converts it into a numerical code, such as "N(N+1)". The layer number represents the number of manufacturing layers contained in the process state unit, which is "2" in this example. The sequence code values ​​and layer number are then normalized using an interval normalization algorithm, mapping the sequence code values ​​and layer number to the [0,1] interval. The calculation formula is as follows:

[0034] in, These are the normalized eigenvalues. These are the original eigenvalues. This is the minimum value of this characteristic across all process state units. This is the maximum value of the feature across all process state units. The normalized hierarchical encoding value is combined with the number of layers to form the hierarchical feature component.

[0035] In S202, the scanning strategy feature components are constructed. These feature components represent the scanning path types and their combinations, and are constructed using one-hot encoding. First, a scanning strategy encoding dictionary is established, assigning a unique code to all possible scanning path types and combinations. For example, a "unidirectional scanning path" is encoded as [1,0,0,0], a "bidirectional cross-scanning path" as [0,1,0,0], a "spiral scanning path" as [0,0,1,0], and a "mixed scanning path" as [0,0,0,1]. For each scanning strategy corresponding to a process unit, a corresponding one-hot encoded vector is generated based on the encoding dictionary; this vector is the scanning strategy feature component. If a process unit contains multiple scanning strategy combinations, for example, one sub-layer uses a "unidirectional scanning path" and another sub-layer uses a "bidirectional cross-scanning path," then the corresponding one-hot encoded vectors are weighted and summed. The weights are the proportions of the manufacturing time corresponding to each scanning strategy to the total manufacturing time of the process unit, forming the combined scanning strategy feature components. One-hot encoding enables the quantitative representation of discrete scanning strategy data, ensuring that different scanning strategies can participate in subsequent correlation calculations.

[0036] In S203, the process parameter interval feature components are constructed. These feature components represent the range of parameters such as laser power, scanning speed, and layer thickness within the process state, and are constructed using parameter interval normalization and feature concatenation. For continuous process parameters such as laser power, scanning speed, and layer thickness, the maximum values ​​of each parameter within each process state unit are first extracted. and minimum value Forming a parameter range Then, interval normalization is performed on the upper and lower limits of the parameter interval, and the normalization formula is the same as the normalization formula for the hierarchical feature components, where... This parameter is the minimum value among all process state units. The parameter is the maximum value across all process state units. Finally, the normalized lower and upper limits of each parameter are combined in a preset order, such as laser power, scanning speed, and layer thickness, to form the characteristic components of the process parameter interval. For example, if the normalized laser power interval is [0.3, 0.5], the normalized scanning speed interval is [0.4, 0.6], and the normalized layer thickness interval is [0.2, 0.3], then the characteristic components of the process parameter interval are [0.3, 0.5, 0.4, 0.6, 0.2, 0.3].

[0037] In S204, the equipment operating status feature components are constructed. These feature components represent statistical indicators of equipment operating stability and are constructed using statistical calculation and normalization. First, statistical analysis is performed on the equipment operating status data within the process unit to calculate temperature zone stability indicators, such as the standard deviation of temperature fluctuations; execution deviation statistics, such as the average position deviation and maximum deviation; and laser output stability indicators, such as the power fluctuation coefficient. Then, each statistic is normalized within a specific interval using the same formula. Finally, the normalized statistics are concatenated in a preset order to form the equipment operating status feature components. For example, if the normalized temperature zone stability indicator is 0.2, the normalized average execution deviation is 0.3, the normalized maximum deviation is 0.4, and the normalized laser output stability indicator is 0.1, then the equipment operating status feature components are [0.2, 0.3, 0.4, 0.1].

[0038] In S205, the components are concatenated to form a process state feature vector of a unified dimension. After each feature component is constructed, it is concatenated in a preset order, such as hierarchical feature components, scanning strategy feature components, process parameter range feature components, and equipment operating status feature components, to form a process state feature vector of a unified dimension. For example, if the hierarchical feature components are [0.1, 0.02], the scanning strategy feature components are [0,1,0,0], the process parameter range feature components are [0.3,0.5,0.4,0.6,0.2,0.3], and the equipment operating status feature components are [0.2,0.3,0.4,0.1], then the concatenated process state feature vector is [0.1,0.02,0,1,0,0,0.3,0.5,0.4,0.6,0.2,0.3,0.2,0.3,0.4,0.1]. In this way, each process state unit is transformed into a standardized feature vector, ensuring that quantitative comparison and correlation analysis can be performed between different process state units. This provides a unified data foundation for the subsequent construction of correlation probability models and avoids correlation analysis errors caused by inconsistent data formats.

[0039] In some embodiments, for step S2, the correlation probability model addresses the technical problem in the prior art of lacking an effective correlation analysis mechanism between manufacturing records and final quality results, and being unable to quantify the contribution of process states to defect formation. Its core principle lies in the statistical correlation between specific types of defects and specific process states in nuclear-grade additive manufacturing. By analyzing the correspondence between process states and defect types in a large amount of historical manufacturing data, the probability of different process states leading to specific defects can be quantified, thus providing data support for tracing the defect formation stage. Furthermore, this model does not pursue high-precision physical simulation, but focuses on the interpretability and statistical reliability of the correlation, avoiding the problems of excessive computational load and unverifiable models caused by physical simulation.

[0040] Please see Figure 3 , Figure 3This is a schematic diagram illustrating the construction process of the association model provided in this embodiment. In S301, the training sample set is constructed. Specifically, firstly, defect type data is acquired; the executing entity obtains the quality inspection results corresponding to historically manufactured components from the quality inspection system of nuclear-grade additive manufacturing components. These quality inspection results are non-destructive testing results from the finished product stage or subsequent process stages, such as ultrasonic testing results, radiographic testing results, metallographic analysis results, etc. The quality inspection results include at least information determining whether the component has defects and identification information for the defect type. The classification of defect types is based on common defect types in nuclear-grade additive manufacturing, such as local non-fusion, density abnormalities, microstructural abnormalities caused by uneven heat input, cracks, and porosity. Each defect type is assigned a unique defect type label, for example, "D001" represents local non-fusion, "D002" represents density abnormalities, and "D003" represents microstructural abnormalities, etc.

[0041] In addition, quality inspection results may include auxiliary information such as defect location, size, and severity level. Defect location information is used to match the location of the component area corresponding to the process unit, while size and severity levels are used for subsequent sample weight adjustment. For example, defects with higher severity levels are assigned greater weight in the association modeling. For instance, the quality inspection results of a historical component show a "local non-fusion" defect, labeled "D001," located in the central region of the component, with a specific size range and a severity level of "medium."

[0042] Secondly, process status and defect results are matched; a correspondence is established between the unique identifier of historical components and process status units and quality inspection results to ensure that the process status feature vector set of each historical component corresponds one-to-one with its quality inspection result. For example, if the unique identifier of a historical component is "CP001", its corresponding process status feature vector set contains the feature vectors of all process status units divided during the manufacturing process of the component, and the quality inspection result is the defect type label and related auxiliary information of the component.

[0043] Based on the defect location information and the corresponding component area location of the process unit, process units that may be related to the defect are screened. If the quality inspection result contains specific coordinates or area description of the defect location, such as "the middle area of ​​the component, at a specific distance from the upper surface", then the manufacturing layer range corresponding to the defect location is determined according to the manufacturing sequence of the process unit and the component area location, and then all process units within that layer range are screened; if the quality inspection result does not clearly indicate the defect location information, then all process units of the component are considered as candidate units that may be related to the defect.

[0044] A training sample set is formed, with each training sample including input and output terms. The input term is a single process-state feature vector, i.e., the feature vector of the selected process-state unit that may be related to the defect. If a component has no defects, the output term corresponding to all its process-state feature vectors is the "no defect" label, such as "D000". The output term is the component's defect type label, i.e., the defect type identifier in the quality inspection result. For example, if the historical component "CP001" has the defect type label "D001", and the corresponding process-state units selected are "PS005", "PS006", and "PS007", then the training sample set contains three samples, with input terms being the feature vectors of "PS005", "PS006", and "PS007" respectively, and the output term being "D001" for all of them.

[0045] Optionally, to improve the effectiveness of the training sample set, data augmentation can be performed on the samples. For example, for components with multiple defects of the same type, their corresponding process-state feature vectors can be merged and statistically analyzed; for defect-free components, all their process-state feature vectors can be retained as negative samples to balance the class distribution of the training samples. Simultaneously, the training sample set can be deduplicated to remove duplicate sample data, ensuring the uniqueness of each sample.

[0046] In S302, the association probability model is trained. The association probability model takes process-state feature vectors as input and defect type labels as output. Based on the training sample set, it calculates the association probability value of each process-state feature vector leading to each defect type by statistically analyzing the co-occurrence frequency between each process-state feature vector and each defect type label. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 This is a schematic diagram of the training process for the association probability model provided in this embodiment of the disclosure. The specific training process is as follows: In S3021, co-occurrence frequency statistics are performed. This involves iterating through all samples in the training sample set and calculating the feature vector for each process state. Labels for each defect type The number of samples that appear simultaneously That is, the number of co-occurrences; at the same time, the feature vector of each process state is counted. The total number of samples appearing in the training sample set This includes the number of times it appears in combination with all defect type labels and the number of times it appears as a sample without a defect label. For example, process state feature vectors. With defect type label The number of samples appearing simultaneously is 20. If the total number of samples appearing in the training sample set is 100, then , .

[0047] In S3022, the association probability is calculated. Based on the statistically obtained co-occurrence frequency, the feature vector of each process state is calculated. Causes various defect types Association probability value The correlation probability is calculated using a frequency-based statistical method, and a smoothing coefficient is introduced to avoid cases where the probability is 0. The calculation formula is as follows:

[0048] in, For process state feature vectors in the training set With defect type label The number of samples appearing simultaneously. For process state feature vectors in the training set Total number of samples appearing This is a smoothing coefficient, for example, 0.1. This represents the total number of defect types, including those without defect labels. A smoothing coefficient is introduced. This can avoid situations where the correlation probability is 0 due to the lack of co-occurrence of certain process state feature vectors and specific defect type labels, thereby improving the generalization ability of the model.

[0049] In S3023, sample weight adjustment (this step is optional). If the quality inspection results include auxiliary information such as defect severity level, different weights can be assigned to samples based on this information. Weighting factors are introduced into the co-occurrence frequency statistics and correlation probability calculation processes to highlight the correlation between the process state and defect type corresponding to severe defects. For example, the weight of a sample with a severity level of "severe" can be set to... The sample weight for "medium" is... The sample weight for "slight" is ,and In statistics When calculating, a weighted summation method is used, that is... ,in This represents the number of co-occurring samples corresponding to the severity level. The number of severity levels.

[0050] In S3024, model validation and parameter optimization are performed. The training sample set is divided into a training set and a validation set according to a preset ratio, such as 7:3. The training set is used to train the model parameters, and the validation set is used to validate the model performance. The validation set is used to validate the association probability values ​​output by the model, and evaluation metrics such as prediction accuracy and recall are calculated. If the model performance does not meet the preset requirements, such as prediction accuracy being lower than 85%, the smoothing coefficient is adjusted. Parameters such as sample weights are recalculated, and co-occurrence frequency statistics and association probabilities are recalculated until the model performance meets the requirements. After the model training is completed, the model parameters, including the association probability values ​​of each process state feature vector and each defect type label, are saved for subsequent backtracking consistency evaluation.

[0051] The construction process of this correlation probability model is based solely on statistical analysis of historical manufacturing data, without relying on complex physical simulation models. It has low computational cost, strong interpretability, and the model parameters can be continuously updated and optimized by adding new historical data. It can adapt to the needs of different nuclear-level additive manufacturing scenarios, providing quantitative correlation basis for subsequent retrospective consistency assessment. This transforms defect cause analysis from experience-based judgment to data-based objective analysis, improving the accuracy and reliability of defect tracing.

[0052] In one embodiment, for step S3, the backtracking consistency assessment and candidate process state screening address the technical problem in the prior art where the lack of an effective constraint mechanism for defect backtracking leads to low reliability of inference results and an inability to meet the quality requirements of nuclear-grade manufacturing. The core principle is that nuclear-grade additive manufacturing processes are subject to strict process specifications, material properties, and equipment capabilities. Defect formation must occur within these constraints. Therefore, introducing constraints can significantly reduce the inference space and improve the reliability of backtracking results. Simultaneously, by comprehensively considering the correlation between process states and defects, differences in process state characteristics, and constraint fulfillment, the rationality of each process state unit as a defect origin state can be comprehensively evaluated, avoiding inference bias caused by a single factor.

[0053] Among them, the constraint conditions are set to define a reasonable range for backtracking reasoning, ensuring that the reasoning results meet the actual requirements of nuclear-grade additive manufacturing. The constraints include three categories: process constraints, material constraints, and equipment constraints, which are specifically set as follows: Process constraints: Process constraints are used to limit the process parameters to be within the range allowed by nuclear-grade manufacturing specifications. Nuclear-grade manufacturing specifications have clear requirements for the process parameters of additive manufacturing. These requirements are based on material properties, equipment capabilities and component quality requirements, and are key to ensuring component quality.

[0054] The specific method for setting process constraints is as follows: based on the nuclear-grade manufacturing specification documents, extract the allowable ranges for each process parameter. ,in This represents the minimum specified value for the process parameters. This sets the maximum allowed values ​​for process parameters; for scanning strategies, it sets a list of permitted scanning strategy types, such as allowing only validated strategies like "bidirectional cross-scan path" and "spiral scan path"; for manufacturing sequences, it sets constraint parameters such as minimum manufacturing layer thickness and maximum interlayer interval duration. For example, nuclear-grade manufacturing specifications specify a permissible range for laser power of [P].min -P max If the scanning strategy only allows "bidirectional cross-scanning path" and "spiral scanning path" and the interlayer interval time does not exceed a specific time, then the process constraints are the above parameter range and scanning strategy list.

[0055] Material constraints: Material constraints are used to limit the analysis object to the same material batch or equivalent material behavior range. Nuclear-grade additive manufacturing has extremely high requirements for the consistency of material batches. Different material batches may have differences in chemical composition and physical properties. These differences will affect the process response and component quality during the manufacturing process. Therefore, defect backtracking analysis must be limited to the same material batch.

[0056] The specific method for setting material constraints is as follows: Based on the manufacturing records of the component to be analyzed, extract the material batch number. The material batch number of the analyzed object must be consistent with the material batch number of the component to be analyzed. If an equivalent material batch exists—that is, other batches whose material behavior is consistent with the target batch after experimental verification—then the equivalent material batch number is included in the allowable range. Simultaneously, set material behavior consistency indicators, such as the allowable fluctuation range of parameters like melting point, thermal conductivity, and tensile strength, to ensure that the material behavior of the analyzed object is consistent with the target material batch. For example, if the material batch number of the component to be analyzed is "M001," its equivalent material batch number is "M002," and the allowable fluctuation range of the material melting point is ±5℃, then the material constraint is that the material batch number is either "M001" or "M002," and the material melting point is within ±5℃ of the target batch's melting point.

[0057] Equipment constraints: Equipment constraints are used to limit the operating state of the equipment to within the allowable working range. The operating state of additive manufacturing equipment directly affects the execution accuracy of process parameters and the quality of components. When the operating state of the equipment exceeds the allowable working range, the probability of defects in the manufactured components will increase significantly. Therefore, it is necessary to constrain the operating state of the equipment.

[0058] The specific method for setting equipment constraints is as follows: Based on the equipment's technical specifications and nuclear-grade manufacturing standards, extract the allowable ranges for the equipment's operating status parameters, such as the allowable range of temperature zones, the allowable positional deviation of mechanical actuators, and the allowable indicators for laser output stability; set equipment fault record constraints, meaning that during the manufacturing process corresponding to the analysis object, no major faults occurred in the equipment, such as laser generator failures or mechanical structure jamming, or if a fault occurred, it was repaired in a timely manner and the post-repair operating status meets the allowable range requirements. For example, the allowable range of the equipment's temperature zone is [T min -T maxThe position deviation is allowed to be ± a specific value, and the laser output stability index is allowed to fluctuate by no more than 3%. The equipment constraints are the above-mentioned equipment operating status parameter range and no major fault records.

[0059] Once the constraints are set, they are stored in the form of a constraint rule base. Each constraint includes the constraint type, constraint parameters, decision criteria, and weight. The weight is used to indicate the importance of the constraint in backtracking reasoning. For example, the weight of the process constraint is 0.4, the weight of the material constraint is 0.3, and the weight of the equipment constraint is 0.3.

[0060] In one embodiment, the retrospective consistency assessment, under the premise of satisfying the above constraints, calculates the comprehensive consistency evaluation value for each process unit as a defect formation stage. This evaluation value is obtained by weighted summation of constraint satisfaction score, correlation probability score, and average feature difference score. Please refer to [link to relevant documentation]. Figure 5 , Figure 5 This is a schematic diagram of the backtracking consistency assessment process provided in an embodiment of this disclosure. The specific calculation process is as follows: In S401, constraint satisfaction score is calculated. The constraint satisfaction score is used to evaluate the degree to which a process unit satisfies each constraint in the constraint rule base. It employs rule matching to check whether each feature of the process unit satisfies each constraint in the constraint rule base, and then calculates the constraint satisfaction score. The calculation formula is as follows:

[0061] in, This represents the total number of constraints. For the first The weight of each constraint condition, For the first The satisfaction score of each constraint condition; when the constraint condition is satisfied. If not satisfied When partially satisfied To satisfy the ratio. For example, a certain process state unit satisfies the process constraints, , Material constraints , However, it partially meets the equipment constraints, with a satisfaction rate of 0.8. , Then the constraint satisfaction score of the process state unit. .

[0062] In S402, the correlation probability score is calculated. The correlation probability score directly uses the correlation probability value output by the correlation probability model, i.e., the process state feature vector. Defect type label of the component to be analyzed Association probability This score characterizes the correlation between a process unit and the current defect type. A higher correlation probability value indicates a greater likelihood that the process unit is causing the current defect. For example, if the defect type label of the component to be analyzed is "D001", and the feature vector of a certain process unit... Probability of association with “D001” Then the correlation probability score of the process state unit. .

[0063] In S403, the average characteristic difference score is calculated. The average characteristic difference score is used to assess the magnitude of the difference between the process state of this process unit and other process units. The greater the difference, the higher the likelihood that it is a defect origin state, because defects are usually caused by abnormal process states. The average characteristic difference score is calculated as follows: first, the eigenvector of the process unit is calculated. eigenvectors of all other process state units ( The cosine similarity between features is calculated; then, the average of all cosine similarities is used as the feature similarity score; finally, the average feature difference score is calculated. .

[0064] The formula for calculating cosine similarity is:

[0065] in, For feature vectors and dot product, For feature vectors The length of the mold, For feature vectors The modulus. For example, a certain process state unit. The cosine similarities with the other five process state units were 0.9, 0.85, 0.88, 0.92, and 0.86, respectively, with an average of 0.882. Therefore, the feature similarity score is 0.882, and the average feature difference score is [missing information]. .

[0066] In S404, the overall consistency evaluation value is calculated. The overall consistency evaluation value is calculated using a weighted summation method, incorporating the constraint satisfaction scores. Association probability score and average feature difference score The formula for calculating the weighted combination is as follows:

[0067] in, , , These are the weights of the constraint satisfaction score, the correlation probability score, and the average feature difference score, respectively. The weight values ​​can be adjusted according to the actual needs of core-level manufacturing; for example, when emphasizing the degree of correlation, , , When focusing on constraint satisfaction, , , In this embodiment, the following is adopted: , , Weight configuration, such as the weight of a certain process state unit. , , Then the overall consistency evaluation value .

[0068] In one embodiment, the candidate process state screening is based on the comprehensive consistency evaluation value of each process state unit to select the process state unit most likely to cause defects. The specific screening process is as follows: Ranking of overall consistency evaluation values. Overall consistency evaluation values ​​for all process state units. Sort the data in descending order to form a sorted list of process units. For example, if the comprehensive consistency evaluation values ​​of 10 process units are 0.7306, 0.685, 0.652, 0.618, 0.583, 0.547, 0.512, 0.476, 0.441, and 0.405, the sorted list in descending order is [0.7306, 0.685, 0.652, 0.618, 0.583, 0.547, 0.512, 0.476, 0.441, 0.405], and the corresponding process unit identifiers are [PS005, PS006, PS007, PS003, PS004, PS008, PS002, PS009, PS001, PS010].

[0069] Screening threshold setting. A screening threshold is set for candidate process states. There are two methods for setting the screening threshold: absolute threshold and relative threshold. These can be selected or combined according to actual needs. The absolute threshold is the preset minimum comprehensive consistency evaluation value, for example, 0.6. Only process state units with a comprehensive consistency evaluation value greater than or equal to 0.6 are selected as candidate process states. The relative threshold is the top N process state units or the top M% of process state units after sorting. For example, the top 3 process state units or the top 30% of process state units are selected as candidate process states. In this embodiment, a combination of absolute and relative thresholds is used. The absolute threshold is set to 0.6, and the top 3 process state units are selected. If the comprehensive consistency evaluation value of the top 3 process state units is greater than or equal to 0.6, then these 3 units are the candidate process states. If the evaluation value of some units is lower than 0.6, then only units with an evaluation value greater than or equal to 0.6 are selected as candidate process states. For example, in the above sorting list, if the evaluation value of the top 3 process state units is greater than 0.6, then the candidate process states are PS005, PS006, and PS007.

[0070] Secondary verification of candidate process states. Secondary verification of candidate process states is performed to ensure their reliability. This includes: whether the manufacturing time of the candidate process state matches the time window of defect formation (e.g., inferring the approximate time range of defect formation based on the thermal history characteristics of the defect, and verifying whether the manufacturing time of the candidate process state falls within this range); whether the abnormal state markings of the candidate process state are related to the defect type (e.g., local non-fusion defects may be related to abnormal equipment execution deviations, verifying whether such abnormal markings exist in the candidate process state); and whether the differences in process states between candidate process states are reasonable (e.g., whether the changes in process parameters between candidate process states conform to the manufacturing process plan, and whether there are any abnormal adjustments). Through secondary verification, candidate process states that do not conform to the actual situation are eliminated. For example, if the manufacturing time of candidate process state PS007 is not within the inferred time range of defect formation, this candidate process state is eliminated, and PS005 and PS006 are ultimately retained as the final candidate process state units.

[0071] This retrospective consistency assessment and candidate process screening process effectively eliminates process units that do not conform to nuclear-grade manufacturing specifications by introducing multi-dimensional constraints and comprehensive consistency evaluation, narrowing the reasoning space and reducing the probability of misjudgment. At the same time, it comprehensively considers the degree of correlation, constraint satisfaction and feature differences, making the screening of candidate process states more scientific and reasonable, and providing a reliable basis for determining the subsequent defect formation stage.

[0072] In some embodiments, the core purpose of the defect formation information output in step S4 is to determine the manufacturing stage and corresponding process state information at which the defect is most likely to form based on the screening results of the candidate process states, and output it in a standardized form to provide direct support for quality analysis, process improvement and responsibility determination.

[0073] Based on the screening results of candidate process states, and combined with the hierarchical information corresponding to the process state units, the manufacturing time range, and the secondary verification results, the manufacturing hierarchical sequence in which defects are most likely to occur is determined. First, the hierarchical range information corresponding to the candidate process states is extracted. For example, if the hierarchical range corresponding to candidate process state PS005 is layer N to layer N+1, and the hierarchical range corresponding to PS006 is layer N+2, then the manufacturing hierarchical sequence in which defects are most likely to occur is the union of the hierarchical ranges corresponding to these candidate process states, i.e., layer N to layer N+2. If the hierarchical ranges corresponding to the candidate process states overlap or are continuous, they are merged into a unified hierarchical interval; if discrete hierarchical ranges exist, each discrete interval is listed separately, and the comprehensive consistency evaluation value of each interval is labeled.

[0074] By combining the manufacturing time range of candidate process states with the inferred time window for defect formation, the defect formation stage can be further precisely located. For example, if the manufacturing time of candidate process state PS005 is a specific time period 1, the manufacturing time of PS006 is a specific time period 2, and the inferred time window for defect formation is a specific time period 3, then the manufacturing time range in which the defect is most likely to form is specific time period 3, and the corresponding hierarchical range is the middle region between layer N and layer N+2. By comprehensively considering the overall consistency evaluation value of candidate process states, the secondary verification results, and the process status analysis, the process state unit most likely to form the defect is determined. The candidate process state unit with the highest overall consistency evaluation value and passing secondary verification is selected as the most likely process state for defect formation, and its corresponding hierarchical range is the manufacturing hierarchical sequence in which the defect is most likely to form.

[0075] In one embodiment, the output information includes the manufacturing sequence of the defect formation and the corresponding process state information, wherein the corresponding process state information includes scanning strategy information and process parameter range information. The specific output content is as follows: Defect type identifier, clearly indicating the specific type of defect and its corresponding label, such as "Local Incomplete Fusion (D001)"; Manufacturing sequence or sequence range where the defect is most likely to form, such as "Layer N - Layer N+2", and indicating the comprehensive consistency evaluation value corresponding to each sequence range; Scanning strategy information corresponding to the process state, including scanning path type, scanning direction combination method and corresponding code, such as "Bidirectional Cross Scanning Path (Code: 02)". If the process state unit contains multiple scanning strategies, each scanning strategy and its corresponding manufacturing time percentage are listed; Process parameter range information corresponding to the process state, including the specific range of parameters such as laser power, scanning speed, and layer thickness, such as "Laser Power: [P1-P2], Scanning Speed: [V1-V2], Layer Thickness: [H1-H2]", and indicating whether each parameter range meets the process constraints of the nuclear-grade manufacturing specification.

[0076] The output results can be presented in the form of text files, tables, or visual reports, and can be exported to manufacturing execution systems, quality analysis systems, or other relevant management systems. This facilitates subsequent quality analysis meetings, the development of process parameter optimization plans, and the definition of quality responsibilities. For example, the output results can be presented as a visual report, containing detailed information on defect types, formation sequences, scanning strategies, and process parameter ranges. It can also display the ranking of the comprehensive consistency evaluation values ​​of candidate process states and the satisfaction of constraints in chart form, presenting the defect backtracking analysis results intuitively and clearly.

[0077] This invention achieves quantitative traceability analysis of the defect formation stage in nuclear-grade additive manufacturing through core innovative steps such as process state unit partitioning and feature vector construction, correlation probability model construction, backtracking consistency assessment, and candidate process state screening, without introducing new detection equipment or changing the existing manufacturing process. This method discretizes the continuous manufacturing process into traceable process state units, establishes the correlation between process states and defect types through statistical analysis, and conducts comprehensive consistency assessment under process, material, and equipment constraints. It effectively solves the technical problems in nuclear-grade additive manufacturing, such as the strong concealment of local irreversible defects, the disconnect between the formation and discovery stages, and the lack of traceable quantitative analysis methods for their causes. The output defect formation manufacturing sequence and corresponding process state information provide reliable data support for defining quality responsibility in nuclear-grade manufacturing, closing the loop of process improvement, and standardizing the inheritance of manufacturing experience. It also avoids the problems of existing online monitoring lacking nuclear-grade sensors and the large computational load and unverifiable nature of physical simulation, demonstrating high engineering practicality and adaptability to nuclear-grade quality systems.

[0078] Please see Figure 6 , Figure 6A digital twin model construction system 600 for a nuclear-grade additive manufacturing process is provided as an embodiment of the present invention. This system embodiment is related to... Figure 6 Corresponding to the illustrated embodiment of the digital twin model construction method, this system can be specifically applied to various computer devices. The system specifically includes: The process state feature construction module 601 is used to divide the manufacturing process into multiple process state units based on a standardized manufacturing process dataset and according to preset rules, and to construct a process state feature vector for each process state unit; the standardized manufacturing process dataset is obtained by preprocessing the original manufacturing data; The retrospective consistency assessment module 602 is used to perform retrospective consistency assessment on each process state unit under process constraints, material constraints, and equipment constraints based on the correlation probability output by the correlation probability model, and calculate the comprehensive consistency evaluation value of each process state unit; the correlation probability model is constructed based on the process state feature vector and corresponding defect type label in historical manufacturing data, and is used to characterize the correlation between process state and defect type; The candidate process state screening module 603 is used to screen at least one candidate process state unit based on the comprehensive consistency evaluation value. The defect tracing output module 604 is used to output the manufacturing sequence and corresponding process state information of defect formation based on the candidate process state unit.

[0079] Based on the same inventive concept, this application also provides a computer device, the method corresponding to which can be the method in the foregoing embodiments, and the principle of solving the problem is similar to that method. The computer device provided in this application includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the methods and / or technical solutions of the foregoing embodiments of this application.

[0080] The computer device can be a user device, or a device formed by integrating user devices and network devices through a network, or it can be an application running on the aforementioned devices. The user device includes, but is not limited to, various terminal devices such as computers, mobile phones, tablets, smartwatches, and smart bands. The network device includes, but is not limited to, network hosts, single network servers, multiple network server sets, or cloud computing-based computer sets, and can be used to implement some processing functions when setting an alarm clock. Here, the cloud consists of a large number of hosts or network servers based on cloud computing. Cloud computing is a type of distributed computing, consisting of a virtual computer composed of a group of loosely coupled computer sets.

[0081] Figure 7 The diagram illustrates the structure of an apparatus suitable for implementing the methods and / or technical solutions in the embodiments of this application. The apparatus 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes based on a program stored in a read-only memory (ROM) 702 or a program loaded from a storage portion 708 into a random access memory (RAM) 703. The RAM 703 also stores various programs and data required for system operation. The CPU 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0082] The following components are connected to I / O interface 705: an input section 706 including a keyboard, mouse, touchscreen, microphone, infrared sensor, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), LED display, OLED display, etc., and speakers, etc.; a storage section 708 including one or more computer-readable media such as hard disk, optical disk, magnetic disk, semiconductor memory, etc.; and a communication section 709 including a network interface card such as a LAN (local area network) card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet.

[0083] In particular, the methods and / or embodiments in this application can be implemented as computer software programs. For example, the embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowchart. When the computer program is executed by the central processing unit (CPU) 701, it performs the functions defined in the methods of this application.

[0084] Another embodiment of this application provides a computer-readable storage medium having computer program instructions stored thereon, which can be executed by a processor to implement the methods and / or technical solutions of any one or more embodiments of this application described above.

[0085] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0086] The flowcharts or block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-specific system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0087] Furthermore, the inclusion of a single word does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any particular order.

Claims

1. A digital twin model of a nuclear-grade additive manufacturing process, characterized in that, The method for constructing this model includes the following steps: Based on a standardized manufacturing process dataset, the manufacturing process is divided into multiple process state units according to preset rules, and a process state feature vector is constructed for each process state unit; the standardized manufacturing process dataset is obtained by preprocessing the original manufacturing data; Based on the correlation probability output by the correlation probability model, a retrospective consistency assessment is performed on each process state unit under process constraints, material constraints, and equipment constraints, and the comprehensive consistency evaluation value of each process state unit is calculated. The correlation probability model is constructed based on the process state feature vector and corresponding defect type label in historical manufacturing data, and is used to characterize the correlation between process state and defect type. At least one candidate process state unit is selected based on the comprehensive consistency evaluation value; Based on the candidate process state unit, the manufacturing sequence of defect formation and the corresponding process state information are output.

2. The digital twin model of a nuclear-grade additive manufacturing process according to claim 1, characterized in that, The preset rule is as follows: within the same manufacturing layer, if the standard deviation of process parameters, the change of scanning strategy identifier, and the fluctuation of equipment operating status do not exceed the preset stability threshold, then the layer is divided into a process state unit. If the overlap of process parameter ranges between adjacent manufacturing layers exceeds a preset ratio, the scanning strategy identifiers are consistent, and the differences in equipment operating status statistics do not exceed the equipment state stability threshold, then the adjacent layers will be merged into a single process state unit.

3. The digital twin model of a nuclear-grade additive manufacturing process according to claim 1, characterized in that, The process state feature vector It includes the following components: a sequence feature component, used to identify the manufacturing layer range corresponding to the process state unit; and a scan strategy feature component, used to represent the scan path type and combination method. The process parameter range feature component is used to represent the range of laser power, scanning speed, and layer thickness parameters; the equipment operating status feature component is used to represent the statistical indicators of equipment operating stability; after normalization, each component is spliced ​​into a feature vector of a unified dimension in a preset order.

4. A digital twin model of a nuclear-grade additive manufacturing process according to claim 1, characterized in that, Constructing the association probability model includes taking process state feature vectors as input and defect type labels as output, and based on the training sample set, calculating the co-occurrence frequency of each process state feature vector and each defect type label; and based on the co-occurrence frequency, calculating the association probability value of each process state feature vector leading to each defect type.

5. A digital twin model of a nuclear-grade additive manufacturing process according to claim 1, characterized in that, The comprehensive consistency evaluation value is obtained by weighted summation of the following three scores: the constraint satisfaction score of the process unit; the correlation probability score between the process unit and the current defect type output by the correlation probability model; and the average feature difference score between the feature vector of the process unit and the feature vectors of other process units.

6. A digital twin model of a nuclear-grade additive manufacturing process according to claim 5, characterized in that, The average feature difference score is calculated as follows: calculate the cosine similarity between the feature vector of the process state unit and the feature vector of each of the other process state units, then average all the cosine similarities, and subtract the average value from 1 to obtain the average feature difference score.

7. A digital twin model of a nuclear-grade additive manufacturing process according to claim 5, characterized in that, The comprehensive consistency evaluation values ​​of each process unit are sorted in descending order, and at least one process unit with a comprehensive consistency evaluation value higher than a preset threshold or ranked higher is selected as a candidate process unit.

8. A digital twin model construction system for a nuclear-grade additive manufacturing process, characterized in that, include: The process state feature construction module is used to divide the manufacturing process into multiple process state units based on a standardized manufacturing process dataset and according to preset rules, and to construct a process state feature vector for each process state unit; the standardized manufacturing process dataset is obtained by preprocessing the original manufacturing data; The retrospective consistency assessment module is used to perform retrospective consistency assessment on each process state unit under process constraints, material constraints, and equipment constraints, based on the correlation probability output by the correlation probability model. The module calculates the comprehensive consistency evaluation value for each process state unit. The correlation probability model is constructed based on the process state feature vector and corresponding defect type label in historical manufacturing data, and is used to characterize the correlation between process state and defect type. The candidate process state screening module is used to screen at least one candidate process state unit based on the comprehensive consistency evaluation value. The defect tracing and output module is used to output the manufacturing sequence and corresponding process state information of defect formation based on the candidate process state unit.

9. A computer device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the digital twin model construction method for any one of claims 1-7 nuclear-level additive manufacturing processes.

10. A computer-readable medium having computer program instructions stored thereon, characterized in that, The computer program instructions can be executed by a processor to implement the digital twin model construction method for any of the nuclear-grade additive manufacturing processes as claimed in claims 1-7.

Images