Surface nanocrystallization ceramic thread original part machining method based on EBPVD (electron beam physical vapor deposition)
By extracting nanoscale features based on EBPVD and integrating them across scales, the problems of uneven nanostructure and specification adaptability in the machining of ceramic threaded components were solved, and efficient and stable machining of ceramic threaded components was achieved.
Patent Information
- Application Number
- CN202511100510.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-07
- Publication Date
- 2025-11-14
AI Technical Summary
Existing EBPVD processes for machining ceramic threaded components suffer from poor nanostructure uniformity, insufficient bonding force, difficulty in meeting high precision requirements, and lack of adaptive adjustment capabilities for parts of different specifications, resulting in high production costs and long production cycles.
By retrieving and analyzing historical process databases, we can extract nanoscale features and perform cross-scale interactive fusion. We can also use deposition process network layers for multi-scale structural analysis to achieve precise nanoscale processing of ceramic threaded components.
It improves the uniformity of the nanostructure on the surface of ceramic threaded components and the stability of the interface bonding, reduces production preparation time and equipment replacement costs, and enhances the consistency of processing quality and the service life of parts.
Smart Images
Figure CN120954588A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ceramic processing technology, specifically to a method for processing surface nano-sized ceramic threaded components based on EBPVD. Background Technology
[0002] In modern industry, ceramic materials, with their excellent high-temperature resistance, corrosion resistance, and high strength, are widely used in the manufacturing of high-end equipment in aerospace, precision machinery, and energy and chemical industries. Among these, ceramic threaded components, as key parts for connections and transmissions, directly affect the operational stability and service life of the entire equipment due to their surface properties. With the continuous development of industrial technology, higher requirements are being placed on the surface properties of ceramic threaded components, especially the optimization of surface structures at the nanoscale, which has become an important direction for improving their wear resistance, fatigue resistance, and sealing performance.
[0003] Traditional methods for machining ceramic threaded parts mainly include mechanical grinding, electrical discharge machining (EDM), and laser processing. Mechanical grinding achieves machining through friction between a grinding wheel and the ceramic surface. However, ceramic materials are extremely hard and brittle, making them prone to surface cracks and stress concentration during grinding, leading to a decrease in part strength and making it difficult to achieve nanometer-level surface precision control. EDM uses the high temperature generated by pulsed discharge to melt and remove ceramic materials. However, this method easily forms a recast layer on the machined surface, resulting in a loose surface structure and affecting the original properties of the material. While laser processing can achieve high machining accuracy, the equipment is expensive, the processing efficiency is low, and a heat-affected zone is generated during the process, leading to uneven surface properties.
[0004] To achieve surface nanostructuring of ceramic materials, existing technologies include physical vapor deposition (PVD) and chemical vapor deposition (CVD). Among these, electron beam physical vapor deposition (EBPVD) has gradually become an important method for ceramic surface nanostructuring due to its advantages such as fast deposition rate, high coating purity, and strong adhesion to the substrate. However, current EBPVD-based ceramic threaded component processing still faces many challenges. Traditional EBPVD process parameter settings rely heavily on accumulated experience and lack systematic analysis of nanostructuring characteristics, resulting in poor uniformity of the nanostructure on the processed ceramic thread surface, making it difficult to meet the requirements of high-precision applications. Furthermore, the structural complexity of ceramic threaded components makes it difficult to coordinate the deposition rate and grain growth direction in different regions during surface nanostructuring, easily leading to uneven nanocoating thickness and insufficient adhesion.
[0005] Existing technologies for nano-sizing ceramic threaded components often employ single-scale process analysis methods, failing to comprehensively capture the multi-scale characteristics of the deposition process. For example, in the nanoparticle deposition stage, microscale grain growth is closely related to macroscale coating thickness variations, but traditional methods struggle to achieve cross-scale feature fusion, resulting in a lack of effective data support for process optimization. Furthermore, due to the high chemical stability of ceramic materials, the interfacial bonding mechanism between their surface and the nano-coating is complex. Traditional processes struggle to precisely control atomic diffusion and chemical bonding at the interface, thus affecting the adhesion strength and durability of the nano-coating.
[0006] In actual production, ceramic threaded components come in various specifications, and different specifications of parts have different requirements for surface nano-sizing. Existing processes lack the ability to adaptively adjust to different specifications of parts. When changing the processing object, a large number of process experiments need to be carried out again to determine the appropriate parameters, which not only increases production costs but also extends the production cycle. Moreover, the extraction of nano-sizing features mostly relies on manual inspection or single sensor data, making it difficult to achieve real-time monitoring and dynamic adjustment of the entire processing process. This results in large fluctuations in product qualification rate, which cannot meet the needs of large-scale production. Summary of the Invention
[0007] The purpose of this invention is to provide a method for processing surface nano-sized ceramic threaded components based on EBPVD, so as to solve the problems mentioned in the background art.
[0008] To achieve the above objectives, the present invention provides a method for processing surface nano-sized ceramic threaded components based on EBPVD, the method comprising:
[0009] Obtain the target specifications of the target ceramic threaded component, and search the historical process database based on the target specifications to determine the set of historical ceramic processing process records;
[0010] The historical ceramic processing technology record set is classified into different categories to identify multiple sets of historical ceramic processing technology record sets.
[0011] The multiple sets of historical ceramic processing technology records are traversed to extract nanoscale features, thereby obtaining multiple nanoscale feature values and the associated duration of multiple nanoscale features;
[0012] The duration of the associated multiple nano-features is used as multiple working cycles of multiple deposition process network layers. The multiple deposition process network layers are configured to perform multi-scale structural analysis on the deposition process data sequence of the target ceramic threaded component within a preset processing time period to obtain multiple deposition process feature sets.
[0013] Cross-scale interactive fusion of the multiple deposition process feature sets is performed to obtain a target interactive fusion deposition process feature set, which is then used as the processing result of the surface nano-ceramic threaded component.
[0014] Preferably, the multiple sets of historical ceramic processing technology records are traversed to extract nanoscale features, obtaining multiple nanoscale feature set values and multiple nanoscale feature association durations, including:
[0015] By traversing the multiple sets of historical ceramic processing technology records, deposition process data is extracted to obtain multiple sets of historical ceramic deposition process data sequences.
[0016] Instantaneous process data are extracted from the multiple sets of historical ceramic deposition process data sequences to determine multiple sets of historical instantaneous process data. Each set of historical instantaneous process data is the process data at the moment when the maximum parameter change fluctuation occurs in each set of historical ceramic deposition process data sequences.
[0017] Feature extraction is performed on the multiple historical instantaneous process data sets to obtain multiple historical instantaneous nanoscale feature sets, and the multiple historical instantaneous nanoscale feature sets are analyzed in a concentrated manner to determine multiple nanoscale feature set values;
[0018] Using the multiple sets of historical instantaneous nano-sizing features as an index, feature association duration diffusion identification is performed on the multiple sets of historical ceramic deposition process data sequences to determine multiple sets of historical instantaneous nano-sizing feature association durations. Each historical instantaneous nano-sizing feature association duration reflects the duration of data associated with the historical instantaneous nano-sizing features at the nano-sizing start time in a set of historical ceramic deposition process data sequences.
[0019] The mean of the multiple historical instantaneous nanoscale feature association duration sets is calculated to obtain the association duration of the multiple nanoscale features.
[0020] Preferably, instantaneous feature extraction is performed on the multiple historical instantaneous process data sets to obtain multiple historical instantaneous nanoscale feature sets, and the multiple historical instantaneous nanoscale feature sets are centrally analyzed to determine multiple nanoscale feature set values, including:
[0021] The nanoscale feature extraction network layer is used to extract instantaneous features from the multiple historical instantaneous process data sets to obtain the multiple historical instantaneous nanoscale feature sets;
[0022] The average value of the nanoscale features is calculated by traversing multiple historical instantaneous nanoscale feature sets to determine the average value of multiple historical instantaneous nanoscale features;
[0023] According to the preset iteration period, the average value of the multiple historical instantaneous nanoscale features is iterated in the multiple historical instantaneous nanoscale feature sets to obtain multiple iterated historical instantaneous nanoscale features;
[0024] When the aggregation amount of the plurality of iterative historical instantaneous nanoscale features is less than or equal to the aggregation amount of the average value of the plurality of historical instantaneous nanoscale features, the average value of the plurality of historical instantaneous nanoscale features is taken as the set value of the plurality of nanoscale features.
[0025] Preferably, the method includes:
[0026] When the aggregation amount of the multiple iterative historical instantaneous nanoscale features is greater than the aggregation amount of the average of the multiple historical instantaneous nanoscale features, it is determined whether the difference between the aggregation amount of the multiple iterative historical instantaneous nanoscale features and the average of the multiple historical instantaneous nanoscale features is greater than or equal to a preset aggregation amount difference threshold. If so, the iteration continues based on the multiple iterative historical instantaneous nanoscale features until the maximum number of iterations is met, and the multiple iterative historical instantaneous nanoscale features obtained in the last iteration are taken as the set value of the multiple nanoscale features.
[0027] If not, then stop the iteration and use the instantaneous nanoscale features of the multiple iteration history as the set value of the multiple nanoscale features.
[0028] Preferably, using the multiple historical instantaneous nanoscale feature sets as indexes, feature association duration diffusion identification is performed on the multiple sets of historical ceramic deposition process data sequences to determine multiple historical instantaneous nanoscale feature association duration sets, including:
[0029] The nanoscale feature extraction network layer is used to extract nanoscale features from the multiple sets of historical ceramic deposition process data sequences to obtain multiple sets of historical ceramic nanoscale feature sequences.
[0030] One historical instantaneous nano-nano-feature is randomly extracted from the multiple sets of historical instantaneous nano-nano-features as the first historical instantaneous nano-nano-feature, and the corresponding first distinguishing historical ceramic nano-nano-feature sequence is matched from the multiple sets of distinguishing historical ceramic nano-nano-feature sequences.
[0031] According to a preset approximate correlation scale, the first historical instantaneous nano-scale feature is searched for nearest neighbors in the first distinguishing historical ceramic nano-scale feature sequence to obtain the neighborhood of the first historical instantaneous nano-scale feature.
[0032] The duration of the neighborhood of the first historical instantaneous nanoscale feature is statistically analyzed, and the statistical results are used as the associated duration of the first historical instantaneous nanoscale feature.
[0033] According to a preset nearest neighbor association scale, the multiple historical instantaneous nanoscale feature sets are subjected to feature association duration diffusion identification in the corresponding multiple sets of historical ceramic deposition process data sequences to determine multiple sets of historical instantaneous nanoscale feature association durations.
[0034] Preferably, the multiple sedimentation process feature sets are cross-scale interactively fused to obtain a target interactively fused sedimentation process feature set, including:
[0035] Randomly extract a first sedimentation process feature set and a second sedimentation process feature set from the plurality of sedimentation process feature sets;
[0036] Calculate the similarity between the first sedimentation process feature set and the second sedimentation process feature set to determine the first feature similarity set;
[0037] The first feature similarity set is normalized to obtain the first feature similarity normalization value set;
[0038] The first set of feature similarity normalization values and the second set of deposition process features are convolved to obtain the first set of interactively fused deposition process features.
[0039] A third set of sedimentation process features is randomly extracted from the plurality of sedimentation process feature sets, and then cross-scale interactive fusion is performed between the third set and the first interactive fusion sedimentation process feature set to obtain a second interactive fusion sedimentation process feature set.
[0040] After multiple cross-scale interactive fusions, until all sedimentation process features in the multiple sedimentation process feature sets are fused, the target interactive fused sedimentation process feature set is obtained.
[0041] Preferably, the method includes:
[0042] The training data includes a set of normalized similarity values of multiple sample features, a set of deposition process features of multiple samples, and a set of interactively fused deposition process features of multiple samples.
[0043] Supervised training of the network layer built on the convolutional neural network is performed using training data until the training converges, thus obtaining the trained convolutional network layer.
[0044] The first set of feature similarity normalization values and the second set of deposition process features are convolved using the convolutional network layer to obtain the first set of interactively fused deposition process features.
[0045] Preferably, the historical ceramic processing technology record set is further categorized to identify multiple distinct historical ceramic processing technology record sets, including:
[0046] Randomly select multiple historical ceramic processing technology records from the aforementioned historical ceramic processing technology record set;
[0047] The multiple historical ceramic processing technology records are enumerated pairwise to obtain multiple enumeration combinations;
[0048] Determine whether there is an enumeration combination whose record similarity is greater than a preset record similarity threshold. If not, then the multiple historical ceramic processing records are used as multiple distinguishing targets.
[0049] Based on the multiple distinguishing targets, the historical ceramic processing technology record set is distinguished from other categories according to a preset record similarity threshold to obtain the multiple distinguishing historical ceramic processing technology record sets, wherein each distinguishing historical ceramic processing technology record set corresponds to a distinguishing target.
[0050] Preferably, it is determined whether there is an enumeration combination whose record similarity is greater than a preset record similarity threshold among the multiple enumeration combinations. If so, the historical ceramic processing technology records in the multiple enumeration combinations whose record similarity is greater than the preset record similarity threshold are merged into the same distinguishing target. Based on the same distinguishing target, the set of historical ceramic processing technology records is distinguished by different types to obtain the multiple distinguished historical ceramic processing technology record sets.
[0051] Preferably, supervised training of network layers built on convolutional neural networks is performed using training data, including:
[0052] Set the initial learning rate and maximum number of training epochs for the network layers;
[0053] In each round of training, multiple sets of sample feature similarity normalization values and multiple sets of sample deposition process features are input into the network layer to obtain the predicted interactive fusion deposition process feature set output by the network layer.
[0054] Calculate the loss value between the predicted interactive fusion deposition process feature set and the multiple sample interactive fusion deposition process feature sets;
[0055] The parameters of the network layer are adjusted according to the loss value until the loss value is less than or equal to the preset loss threshold or the maximum number of training rounds is reached, thus completing the training.
[0056] Compared with the prior art, the beneficial effects of the present invention are:
[0057] By retrieving and analyzing historical process databases, past processing experience can be fully utilized, avoiding the blind parameter setting caused by reliance on manual experience in traditional processes. The heterogeneous classification of historical ceramic processing records allows for the categorization of process data for different processing conditions and parts of different specifications, making subsequent nanoscale feature extraction more targeted and reducing interference from irrelevant data. This historical data-based analysis method enables the processing to better adapt to different specifications of target ceramic threaded parts, eliminating the need for numerous repetitive trials each time the processing object is changed, thus shortening process preparation time to a certain extent.
[0058] In the nanoscale feature extraction stage, by traversing multiple sets of historical ceramic processing records, the concentrated values and correlation durations of multiple nanoscale features obtained provide a specific basis for setting the action period of the deposition process network layer. Using the correlation duration as the action period of the deposition process network layer makes the network layer analysis more closely reflect the actual nanoscale deposition process, avoiding the analytical biases caused by traditional fixed period settings. Multi-scale structural analysis of the deposition process data sequence using multiple deposition process network layers can capture process features at different time scales, such as short-term grain growth fluctuations and long-term coating thickness variation trends. This multi-dimensional analysis method makes the extraction of deposition process features more comprehensive.
[0059] The cross-scale interactive fusion process integrates multiple deposition process feature sets, breaking the limitations of single-scale analysis and enabling microscopic and macroscopic process features to be correlated and complementary. For example, the fusion of nanoparticle distribution characteristics at the microscopic scale and deposition rate characteristics at the macroscopic scale can more accurately reflect the overall effect of surface nanostructuring. This fused set of target interactive deposition process features, as the processing result, can more comprehensively reflect the nanostructuring state of the ceramic threaded component surface, including the uniformity of the nanocoating, grain size distribution, interface bonding state, and other aspects.
[0060] For ceramic threaded components, this processing method effectively improves the uniformity of their surface nanostructure, reducing localized wear or stress concentration caused by structural inhomogeneity. Through multi-scale analysis and cross-scale fusion, various parameters during the deposition process can be precisely controlled, resulting in a more stable bonding interface between the nano-coating and the ceramic substrate, reducing the risk of coating detachment. Simultaneously, this method can adapt to the processing needs of ceramic threaded components of different specifications, improving production flexibility while ensuring processing quality. It allows the same processing system to handle various types of parts, reducing the cost of equipment replacement and debugging.
[0061] In practical applications, ceramic threaded components processed using this method exhibit a surface nano-sizing effect that better meets design expectations, maintaining excellent performance stability under harsh environments such as high temperature, high pressure, and strong corrosion. Compared to traditional processing methods, the surface nano-coating has higher density, superior wear resistance, and fatigue resistance, thereby extending the service life of parts and reducing the frequency of equipment maintenance. Furthermore, the entire processing is based on data-driven analysis and control, reducing human intervention and minimizing quality problems caused by operational errors. This results in more consistent product quality and better meets the stringent requirements of high-end equipment manufacturing for ceramic threaded components. Attached Figure Description
[0062] Figure 1 This is a schematic diagram illustrating the working principle of the surface nano-sized ceramic threaded component processing method based on EBPVD described in this invention.
[0063] Figure 2 Flowchart for determining the extraction and association time of nanoscale features;
[0064] Figure 3 A flowchart for determining the concentrated values of historical instantaneous nanoscale features;
[0065] Figure 4 A flowchart for feature association time-varying recognition;
[0066] Figure 5 This is a flowchart for cross-scale interactive fusion. Detailed Implementation
[0067] 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.
[0068] Please see Figures 1-5 The present invention provides a method for processing surface nano-sized ceramic threaded components based on EBPVD, the method comprising:
[0069] The target specifications of the target ceramic threaded component are obtained, including parameters such as thread size, ceramic material type, and surface nano-thickness requirements. Based on the target specifications, a search is performed in the historical process database, which stores the specification parameters and corresponding processing records of previously processed ceramic threaded components. By matching the target specifications with the specification parameters in the historical records, a set of historical ceramic processing records is determined.
[0070] The historical ceramic processing technology record set is differentiated into different categories. Records with significant differences in process parameters, processing environment, material properties, etc., are divided into different sets to identify multiple sets of historical ceramic processing technology records.
[0071] The multiple sets of historical ceramic processing records are traversed to extract nanoscale features. Feature parameters related to surface nanoscale and their corresponding time information are extracted from each set to obtain multiple nanoscale feature values and multiple nanoscale feature associated durations.
[0072] The duration of association of the multiple nanoscale features is used as multiple operating periods of multiple deposition process network layers. These deposition process network layers are network models constructed based on EBPVD technology for analyzing the deposition process. Using the configured multiple deposition process network layers, multi-scale structural analysis is performed on the deposition process data sequence of the target ceramic threaded component within a preset processing time period. The preset processing time period is set according to the target specifications and processing requirements. The analysis process covers process data changes at different time and spatial scales, obtaining multiple sets of deposition process features.
[0073] Cross-scale interactive fusion of the multiple deposition process feature sets is performed, and the effective integration of features at different scales is achieved through algorithm processing to obtain a target interactive fusion deposition process feature set, which is then used as the processing result of surface nano-ceramic threaded parts.
[0074] Example 1: When extracting deposition process data by traversing multiple sets of differentiated historical ceramic processing records, a comprehensive analysis of each historical ceramic processing record in each set is required. Each record contains various parameters during the processing, such as electron beam power, deposition chamber vacuum, target evaporation rate, substrate temperature, and deposition time. For different sets of differentiated historical ceramic processing records, these parameters are extracted sequentially according to time order, forming multiple sets of differentiated historical ceramic deposition process data sequences. The data in each set is indexed by a timestamp to ensure the complete preservation of the temporal sequence of parameter changes. For example, a set may contain a complete sequence of deposition rates over time for alumina ceramic threaded components under different electron beam powers.
[0075] When extracting instantaneous process data from multiple sets of historical ceramic deposition process data sequences, each data sequence in each set needs to be processed individually. By continuously monitoring the parameters in the data sequences, the moments with the greatest parameter fluctuations are identified. Parameter fluctuations are determined by calculating the absolute value of the parameter difference between adjacent time points and comparing the magnitude of these differences. When the absolute value of the parameter difference at a certain moment is significantly greater than at other moments, it is determined to be the moment of the greatest parameter fluctuation. All process parameters at that moment are extracted, including electron beam power, vacuum level, substrate temperature, etc., and these parameters are summarized into multiple historical instantaneous process data sets. Each historical instantaneous process data set corresponds to a set of historical ceramic deposition process data sequences, and each data point within the set is marked with corresponding time information and parameter variation amplitude.
[0076] When extracting features from multiple historical instantaneous process data sets, in-depth analysis of the instantaneous process data in each set is required. Feature parameters related to surface nanostructuring, such as nanocrystal size, surface roughness, deposition layer density, and nanophase content, are screened from the instantaneous process data. These feature parameters are then quantified to form multiple historical instantaneous nanostructuring feature sets. When performing centralized analysis on multiple historical instantaneous nanostructuring feature sets, statistical methods are used to process feature parameters of the same type, such as calculating the frequency of occurrence and numerical distribution range of the same feature in different sets. By analyzing these statistical results, the central tendency of each feature parameter is determined, thereby obtaining multiple concentrated values of nanostructuring features. These concentrated values can reflect the typical state of nanostructuring features during historical processing.
[0077] When using multiple sets of historical instantaneous nano-sizing features as indexes to identify the diffusion of feature association durations across multiple sets of historical ceramic deposition process data sequences, each historical instantaneous nano-sizing feature needs to be matched with its corresponding data sequence. Starting from the historical instantaneous nano-sizing feature at the beginning of nano-sizing, the changes in relevant parameters before the feature appeared are traced backward in the data sequence, and the continuation of parameters after the feature disappeared is tracked backward. By determining the moment when the parameter associated with the feature begins to change and the moment when it stops changing, the time interval between these two moments is calculated, which is the association duration of the feature. The association durations of all historical instantaneous nano-sizing features in the same set are summarized to form multiple sets of historical instantaneous nano-sizing feature association durations. The arithmetic mean of all durations in each set of association durations is calculated to obtain multiple nano-sizing feature association durations. These durations can reflect the duration characteristics of nano-sizing feature-related parameters in different historical processing processes.
[0078] Example 2: When using a nanoscale feature extraction network layer to extract instantaneous features from multiple historical instantaneous process data sets, this network layer is a trained neural network model. Its input consists of various parameters from the historical instantaneous process data sets, including electron beam current, deposition rate, gas flow rate, and surface temperature. The network layer processes the input data through multiple layers of neurons, performing feature transformation and filtering to separate feature parameters related to surface nanoscale formation, forming multiple historical instantaneous nanoscale feature sets. Each historical instantaneous nanoscale feature set contains feature parameters such as the average diameter of nanoparticles, the thickness gradient of nanolayers, and the porosity of surface nanostructures. These parameters all originate from the parts of the historical instantaneous process data directly related to the nanoscale formation process.
[0079] When calculating the mean value of a feature across multiple historical instantaneous nanoparticle feature sets, each feature parameter in each set needs to be processed separately. For example, for the feature of the average diameter of nanoparticles, the values of this parameter in all historical instantaneous nanoparticle feature sets need to be collected, and then a comprehensive value is calculated using an arithmetic mean, which serves as the historical instantaneous nanoparticle feature mean for this feature. This process is repeated for each feature parameter, ultimately yielding multiple historical instantaneous nanoparticle feature means. These means reflect the central tendency of different historical instantaneous nanoparticle features in terms of their numerical values.
[0080] When iterating over multiple historical instantaneous nanoscale feature sets according to a preset iteration period, the preset iteration period is determined based on the time span of the historical data and the frequency of feature changes. During the iteration process, the average of the historical instantaneous nanoscale features is used as the initial reference value. In each set of historical instantaneous nanoscale features, the feature parameter closest to this average is searched, and new reference values are generated based on these parameters. The new reference value is adjusted in conjunction with the distribution of other feature parameters in its set, and the adjusted value becomes the starting value for the next iteration. After multiple iterations, multiple iterative historical instantaneous nanoscale features are obtained, and these features gradually converge towards a more concentrated numerical range as the number of iterations increases.
[0081] When the aggregation of multiple iterated instantaneous nanoscale features is less than or equal to the aggregation of the average of multiple historical instantaneous nanoscale features, the aggregation is measured by the distribution density and dispersion of the feature parameters. This indicates that the features after iteration are more numerically concentrated and no further adjustment is needed. The average of the historical instantaneous nanoscale features can be directly used as the concentrated value of multiple nanoscale features.
[0082] When the aggregation of multiple iterative historical instantaneous nanoscale features exceeds the aggregation of the mean of multiple historical instantaneous nanoscale features, the difference in aggregation between the two needs to be calculated. If this difference is greater than or equal to a preset aggregation difference threshold, it indicates that the current iteration result differs significantly from the initial mean, and iteration needs to continue until the maximum number of iterations is reached. At this point, the multiple iterative historical instantaneous nanoscale features obtained from the last iteration are used as the aggregated value of multiple nanoscale features. If the difference is less than the preset aggregation difference threshold, it indicates that the difference between the iteration result and the initial mean is within an acceptable range, and iteration does not need to continue. The multiple iterative historical instantaneous nanoscale features are directly used as the aggregated value of multiple nanoscale features. Throughout the process, the aggregation is calculated based on the standard deviation and variance of the feature parameters. These statistics reflect the degree of dispersion or concentration of the feature parameters in terms of values, ensuring that the final aggregated value of nanoscale features can accurately represent the typical state of nanoscale features in historical data.
[0083] Example 3: When using a nanoscale feature extraction network layer to extract nanoscale features from multiple sets of historical ceramic deposition process data sequences, this network layer can analyze continuous parameter changes in the data sequences. Each set of historical ceramic deposition process data sequences contains deposition process parameters from different time periods, such as electron beam current intensity, deposition chamber pressure, and target consumption rate. By continuously monitoring these parameters, the network layer identifies feature changes related to surface nanoscale formation, such as changes in nanoparticle nucleation density and crystal structure transformation of the deposition layer, thereby forming multiple sets of historical ceramic nanoscale feature sequences. The features in each sequence set are arranged in chronological order, fully presenting the evolution trajectory of various features during the nanoscale formation process.
[0084] One historical instantaneous nanoscale feature is randomly extracted from multiple sets of historical instantaneous nanoscale features as the first historical instantaneous nanoscale feature. This feature can be the average particle size of nanoparticles at a certain moment. When matching the corresponding first historical ceramic nanoscale feature sequence from multiple sets of distinguishing historical ceramic nanoscale feature sequences, it is necessary to compare the consistency between the feature parameters at each moment in the sequence and the first historical instantaneous nanoscale feature to find the sequence that contains the feature and has a coherent temporal sequence. For example, if the first historical instantaneous nanoscale feature is an average particle size of 20 nm at a certain moment, then the sequence containing a 20 nm particle size record and with a coherent change in feature before and after is selected from the set of distinguishing historical ceramic nanoscale feature sequences as the first distinguishing historical ceramic nanoscale feature sequence.
[0085] According to a preset approximate correlation scale, a nearest neighbor search is performed on the first historical instantaneous nanostructuring features in the first historical ceramic nanostructuring feature sequence. The preset approximate correlation scale can be set to the allowable fluctuation range of the feature parameters, such as ±2nm of the particle size. During the search, all feature points in the sequence that are within ±2nm of a 20nm particle size are searched. These points together constitute the neighborhood of the first historical instantaneous nanostructuring features. The feature points in the neighborhood include not only the particle size parameter, but also other correlated features at the corresponding time, such as the uniformity of nanoparticle distribution and the thickness variation of the deposition layer.
[0086] When calculating the duration of the neighborhood of the first historical instantaneous nanoscale feature, it is necessary to determine the timestamps of the first and last feature points within the neighborhood. The difference between the two timestamps is the duration of the neighborhood, which is used as the association duration of the first historical instantaneous nanoscale feature. For example, if the timestamp of the first feature point in the neighborhood is the 10th minute and the timestamp of the last feature point is the 18th minute, then the association duration is 8 minutes.
[0087] Following the above process, multiple sets of historical instantaneous nanoscale features are processed within corresponding sets of historical ceramic deposition process data sequences based on a preset nearest-neighbor association scale. For each historical instantaneous nanoscale feature, a nearest-neighbor search is performed within its corresponding sequence to determine the feature neighborhood and calculate the duration, forming multiple sets of associated durations for historical instantaneous nanoscale features. The duration data in each set reflects the continuity of process data associated with each instantaneous nanoscale feature during different historical processing stages. This data will provide the basis for subsequently determining the associated duration of nanoscale features. Throughout the process, it is essential to ensure that each step is based on actual historical process data, avoiding the introduction of subjectively set parameters, to guarantee that the final associated duration accurately reflects the temporal correlation characteristics between nanoscale features and process data.
[0088] Example 4: A first and a second set of deposition process features were randomly extracted from multiple sets of deposition process features. The first set of features may contain nanoscale features, such as the size distribution of nanocrystals, the volume fraction of nanophases, and the roughness of the surface nanostructures. The second set of features may cover macroscale features, such as the overall thickness of the deposited layer, the geometric accuracy of the thread profile, and the macroscopic smoothness of the surface. These feature sets are derived from the analysis of process data of the target ceramic threaded component at different deposition stages, and the feature parameters in each set are accompanied by corresponding measurement time and measurement location information.
[0089] When calculating the similarity between the first and second sedimentation process feature sets, the feature parameters in both sets need to be compared one by one. For numerical feature parameters, the similarity is measured by calculating the relative proportion of their numerical differences; for descriptive feature parameters, the similarity is determined by the overlap of feature word matching. The similarity values of all feature parameters are then aggregated to form the first feature similarity set, where each value corresponds to the degree of similarity between a pair of feature parameters in the two sets.
[0090] When normalizing the first feature similarity set, each similarity value in the set is transformed to a range of 0 to 1. During the transformation, the maximum and minimum similarity values in the set are first determined, and then each similarity value is mapped to the target range using a linear transformation formula, resulting in the normalized set of first feature similarity values. After normalization, similarity values of different orders of magnitude can be compared on the same scale, facilitating subsequent fusion calculations.
[0091] When performing convolution calculations on the first set of normalized feature similarity values and the second set of deposition process features, a preset convolution kernel is used to process the data from both sets. The size of the convolution kernel is determined based on the dimension of the feature set. During the calculation, the convolution kernel slides across the feature data. At each position it slides, the normalized similarity value within the corresponding region is weighted and summed with the second deposition process feature parameters to generate a new feature value. In this way, the feature information from the two sets is fused to obtain a first interactive fused deposition process feature set. The features in this set retain both the macroscopic information of the second deposition process feature set and incorporate the nanoscale information of the first deposition process feature set.
[0092] A third set of deposition process features is randomly extracted from multiple sets of deposition process features. This set may contain mesoscopic-scale feature parameters, such as micrometer-level grain aggregation state, internal porosity distribution of the deposition layer, and microcrack density of the threaded surface. Using the same method as above, the feature similarity between the third set of deposition process features and the first interactive fused deposition process feature set is first calculated to obtain a second feature similarity set. Then, the second feature similarity set is normalized to obtain a second feature similarity normalized value set. Finally, the second feature similarity normalized value set is convolved with the first interactive fused deposition process feature set to obtain the second interactive fused deposition process feature set.
[0093] The cross-scale interactive fusion process described above is repeated. Each time, a new feature set is randomly extracted from the remaining set of deposition process features and subjected to similarity calculation, normalization, and convolution fusion with the currently obtained interactive fused deposition process feature set. For example, if the subsequently extracted set of deposition process features contains dynamic feature parameters related to deposition rate, such as deposition rate change curves and rate fluctuation frequencies over different time periods, this dynamic information can be integrated into the existing interactive fused feature set through fusion. After multiple fusions, until all feature parameters in all deposition process feature sets are incorporated into the interactive fusion process, the final target interactive fused deposition process feature set will cover various feature information from nanoscale to macroscale, as well as dynamic change features at different deposition stages, fully presenting the comprehensive process characteristics of nanoscale processing of the target ceramic threaded component surface.
[0094] The linear transformation formula for normalization is as follows:
[0095] In the formula, y represents the normalized similarity value, and x represents the original similarity value. min Let x represent the minimum value in the set of similarities for the first feature. max This represents the maximum value in the first feature similarity set.
[0096] Example 5: Multiple sets of sample feature similarity normalized values, multiple sets of sample deposition process features, and multiple sets of sample interactive fusion deposition process features were obtained as training data. These sample data cover the processing of ceramic threaded components made of different materials, including common ceramic materials such as zirconium oxide, silicon nitride, and alumina. The sample feature similarity normalized value set is derived from the similarity calculation and normalization of different sample deposition process feature sets, with each value corresponding to the degree of similarity between two sets of sample features. The sample deposition process feature set includes various process parameters, such as electron beam scanning speed, target material composition ratio, and gas flow rate changes during deposition. These parameters were extracted and organized from actual processing records. The sample interactive fusion deposition process feature set is a comprehensive feature set after multi-scale fusion processing, containing feature information from different scales from nanometer to macroscopic.
[0097] When supervising the training of a network layer built on a convolutional neural network using training data, the network layer structure includes an input layer, multiple convolutional layers, pooling layers, and fully connected layers. The input layer receives data from a set of normalized similarity values of sample features and a set of sample deposition process features. The convolutional layers extract features from the input data using pre-defined convolutional kernels. The pooling layers reduce the dimensionality of the convolutional features, reducing the amount of data while retaining key information. The fully connected layers map the processed features to the output layer, generating a predictive interactive fusion deposition process feature set. The initial learning rate and the maximum number of training epochs are set for the network layer. The initial learning rate is determined based on the complexity of the network, while the maximum number of training epochs is determined based on the amount of sample data and the convergence during training.
[0098] In each training round, multiple sets of sample feature similarity normalization values and multiple sets of sample deposition process features are input into the network layer. The network layer generates a predicted interactive fused deposition process feature set through internal parameter calculations. Then, the predicted result is compared with the sample interactive fused deposition process feature set, and the difference between the two is calculated to obtain the loss value. The magnitude of the loss value reflects the degree of deviation between the predicted result and the actual result. The loss value is fed back to the parameters of each layer of the network layer through the backpropagation algorithm, adjusting parameters such as the weights and biases of the convolutional kernels to reduce the loss value in the next training round. This process is repeated until the loss value is less than or equal to a preset loss threshold or the maximum number of training rounds is reached. At this point, the parameters of the network layer tend to stabilize, training is complete, and a trained convolutional network layer is obtained.
[0099] When performing convolution calculations on the first feature similarity normalization value set and the second deposition process feature set using the trained convolutional network layer, the data from both sets are formatted according to the network input requirements before being input into the convolutional network layer. The network layer performs calculations using the adjusted parameters, automatically extracting the correlation information between features and fusing them to generate a first interactive fused deposition process feature set. The features in this set contain both the original information of the second deposition process feature set and the similarity information reflected by the first feature similarity normalization value set.
[0100] Multiple historical ceramic processing records are randomly selected from a historical ceramic processing technology record set. Each record includes information such as the material and specifications of the processed object, the model of the equipment used, various process parameters during processing, and the final processing result. These selected records are paired to form multiple enumerated combinations, each containing two historical records. The similarity between the two records in each enumerated combination is calculated. The similarity calculation involves multiple indicators in the records, such as the composition of the ceramic material, the size parameters of the thread, the electron beam power range during processing, and the deposition time. By comprehensively comparing the consistency of these indicators, the similarity of each combination of records is obtained.
[0101] The system determines whether any of the enumerated combinations have a similarity greater than a preset similarity threshold. If none exist, it indicates significant differences between the extracted records, and these records can be directly used as multiple distinguishing targets. Based on these distinguishing targets, the historical ceramic processing technology record set is classified according to the preset similarity threshold. Records with a similarity higher than the threshold to a certain distinguishing target are grouped into the same set, forming multiple sets of historical ceramic processing technology records. The records in each set have a high degree of consistency in processing characteristics.
[0102] If an enumerated combination exists where the similarity of records exceeds a preset similarity threshold, it indicates that the two records in the combination are quite similar in terms of process characteristics, and they can be merged into the same distinguishing target. For example, if two records both target silicon nitride ceramic threaded components and have similar key parameters such as electron beam power and deposition time during processing, they can be merged into one distinguishing target. Based on the merged distinguishing target, the historical ceramic processing record set is further differentiated, and records with high similarity are grouped into the same set, ultimately resulting in multiple distinguishing historical ceramic processing record sets.
[0103] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0104] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for machining surface-nanosized ceramic threaded components based on EBPVD, characterized in that, The method includes: Obtain the target specifications of the target ceramic threaded component, and search the historical process database based on the target specifications to determine the set of historical ceramic processing process records; The historical ceramic processing technology record set is classified into different categories to identify multiple sets of historical ceramic processing technology record sets. The multiple sets of historical ceramic processing technology records are traversed to extract nanoscale features, thereby obtaining multiple nanoscale feature values and the associated duration of multiple nanoscale features; The duration of the associated multiple nano-features is used as multiple working cycles of multiple deposition process network layers. The multiple deposition process network layers are configured to perform multi-scale structural analysis on the deposition process data sequence of the target ceramic threaded component within a preset processing time period to obtain multiple deposition process feature sets. Cross-scale interactive fusion of the multiple deposition process feature sets is performed to obtain a target interactive fusion deposition process feature set, which is then used as the processing result of the surface nano-ceramic threaded component.
2. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 1, characterized in that, The multiple sets of historical ceramic processing technology records are traversed to extract nanoscale features, resulting in multiple nanoscale feature set values and multiple nanoscale feature association durations, including: By traversing the multiple sets of historical ceramic processing technology records, deposition process data is extracted to obtain multiple sets of historical ceramic deposition process data sequences. Instantaneous process data are extracted from the multiple sets of historical ceramic deposition process data sequences to determine multiple sets of historical instantaneous process data. Each set of historical instantaneous process data is the process data at the moment when the maximum parameter change fluctuation occurs in each set of historical ceramic deposition process data sequences. Feature extraction is performed on the multiple historical instantaneous process data sets to obtain multiple historical instantaneous nanoscale feature sets, and the multiple historical instantaneous nanoscale feature sets are analyzed in a concentrated manner to determine multiple nanoscale feature set values; Using the multiple sets of historical instantaneous nano-sizing features as an index, feature association duration diffusion identification is performed on the multiple sets of historical ceramic deposition process data sequences to determine multiple sets of historical instantaneous nano-sizing feature association durations. Each historical instantaneous nano-sizing feature association duration reflects the duration of data associated with the historical instantaneous nano-sizing features at the nano-sizing start time in a set of historical ceramic deposition process data sequences. The mean of the multiple historical instantaneous nanoscale feature association duration sets is calculated to obtain the association duration of the multiple nanoscale features.
3. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 2, characterized in that, Instantaneous feature extraction is performed on the multiple historical instantaneous process data sets to obtain multiple historical instantaneous nanoscale feature sets. These multiple historical instantaneous nanoscale feature sets are then analyzed to determine multiple nanoscale feature set values, including: The nanoscale feature extraction network layer is used to extract instantaneous features from the multiple historical instantaneous process data sets to obtain the multiple historical instantaneous nanoscale feature sets; The average value of the nanoscale features is calculated by traversing multiple historical instantaneous nanoscale feature sets to determine the average value of the nanoscale features of multiple historical instantaneous moments. According to the preset iteration period, the average value of the multiple historical instantaneous nanoscale features is iterated in the multiple historical instantaneous nanoscale feature sets to obtain multiple iterated historical instantaneous nanoscale features; When the aggregation amount of the plurality of iterative historical instantaneous nanoscale features is less than or equal to the aggregation amount of the average value of the plurality of historical instantaneous nanoscale features, the average value of the plurality of historical instantaneous nanoscale features is taken as the set value of the plurality of nanoscale features.
4. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 3, characterized in that, include: When the aggregation amount of the multiple iterative historical instantaneous nanoscale features is greater than the aggregation amount of the average of the multiple historical instantaneous nanoscale features, it is determined whether the difference between the aggregation amount of the multiple iterative historical instantaneous nanoscale features and the average of the multiple historical instantaneous nanoscale features is greater than or equal to a preset aggregation amount difference threshold. If so, the iteration continues based on the multiple iterative historical instantaneous nanoscale features until the maximum number of iterations is met, and the multiple iterative historical instantaneous nanoscale features obtained in the last iteration are taken as the set value of the multiple nanoscale features. If not, then stop the iteration and use the instantaneous nanoscale features of the multiple iteration history as the set value of the multiple nanoscale features.
5. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 4, characterized in that, Using the multiple historical instantaneous nanoscale feature sets as indexes, feature association duration diffusion identification is performed on the multiple sets of historical ceramic deposition process data sequences to determine multiple historical instantaneous nanoscale feature association duration sets, including: The nanoscale feature extraction network layer is used to extract nanoscale features from the multiple sets of historical ceramic deposition process data sequences to obtain multiple sets of historical ceramic nanoscale feature sequences. One historical instantaneous nano-nano-feature is randomly extracted from the multiple sets of historical instantaneous nano-nano-features as the first historical instantaneous nano-nano-feature, and the corresponding first distinguishing historical ceramic nano-nano-feature sequence is matched from the multiple sets of distinguishing historical ceramic nano-nano-feature sequences. According to a preset approximate correlation scale, the first historical instantaneous nano-scale feature is searched for nearest neighbors in the first distinguishing historical ceramic nano-scale feature sequence to obtain the neighborhood of the first historical instantaneous nano-scale feature. The duration of the neighborhood of the first historical instantaneous nanoscale feature is statistically analyzed, and the statistical results are used as the associated duration of the first historical instantaneous nanoscale feature. According to a preset nearest neighbor association scale, the multiple historical instantaneous nanoscale feature sets are subjected to feature association duration diffusion identification in the corresponding multiple sets of historical ceramic deposition process data sequences to determine multiple sets of historical instantaneous nanoscale feature association durations.
6. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 1, characterized in that, Cross-scale interactive fusion of the multiple sedimentation process feature sets is performed to obtain a target interactive fused sedimentation process feature set, including: Randomly extract a first sedimentation process feature set and a second sedimentation process feature set from the plurality of sedimentation process feature sets; Calculate the similarity between the first sedimentation process feature set and the second sedimentation process feature set to determine the first feature similarity set; The first feature similarity set is normalized to obtain the first feature similarity normalization value set; The first set of feature similarity normalization values and the second set of deposition process features are convolved to obtain the first set of interactively fused deposition process features. A third set of sedimentation process features is randomly extracted from the plurality of sedimentation process feature sets, and then cross-scale interactive fusion is performed between the third set and the first interactive fusion sedimentation process feature set to obtain a second interactive fusion sedimentation process feature set. After multiple cross-scale interactive fusions, until all sedimentation process features in the multiple sedimentation process feature sets are fused, the target interactive fused sedimentation process feature set is obtained.
7. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 6, characterized in that, include: The training data includes a set of normalized similarity values of multiple sample features, a set of deposition process features of multiple samples, and a set of interactively fused deposition process features of multiple samples. Supervised training of the network layer built on the convolutional neural network is performed using training data until the training converges, thus obtaining the trained convolutional network layer. The first set of feature similarity normalization values and the second set of deposition process features are convolved using the convolutional network layer to obtain the first set of interactively fused deposition process features.
8. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 1, characterized in that, The historical ceramic processing technology record set is then categorized to identify different types, resulting in multiple categorized historical ceramic processing technology record sets, including: Randomly select multiple historical ceramic processing technology records from the aforementioned historical ceramic processing technology record set; The multiple historical ceramic processing technology records are enumerated pairwise to obtain multiple enumeration combinations; Determine whether there is an enumeration combination whose record similarity is greater than a preset record similarity threshold. If not, then the multiple historical ceramic processing records are used as multiple distinguishing targets. Based on the multiple distinguishing targets, the historical ceramic processing technology record set is distinguished from other categories according to a preset record similarity threshold to obtain the multiple distinguishing historical ceramic processing technology record sets, wherein each distinguishing historical ceramic processing technology record set corresponds to a distinguishing target.
9. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 8, characterized in that, Determine whether there is an enumeration combination whose record similarity is greater than a preset record similarity threshold among the multiple enumeration combinations. If so, merge the historical ceramic processing technology records in the multiple enumeration combinations whose record similarity is greater than the preset record similarity threshold into the same distinguishing target. Based on the same distinguishing target, perform heterogeneous distinction on the set of historical ceramic processing technology records to obtain the multiple distinguished historical ceramic processing technology record sets.
10. The method for processing surface nano-sized ceramic threaded components based on EBPVD as described in claim 7, characterized in that, Supervised training of network layers built on convolutional neural networks using training data includes: Set the initial learning rate and maximum number of training epochs for the network layers; In each round of training, multiple sets of sample feature similarity normalization values and multiple sets of sample deposition process features are input into the network layer to obtain the predicted interactive fusion deposition process feature set output by the network layer. Calculate the loss value between the predicted interactive fusion deposition process feature set and the multiple sample interactive fusion deposition process feature sets; The parameters of the network layer are adjusted according to the loss value until the loss value is less than or equal to the preset loss threshold or the maximum number of training rounds is reached, thus completing the training.