Particle concentration field prediction method
Patent Information
- Application Number
- CN202610863763.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-15
- Publication Date
- 2026-10-30
AI Technical Summary
[0003]离散采样方式依赖粒子计数器对若干固定点位进行离散采样，其受限于粒子计数器数量、安装位置、采样频率及工程布设条件，通常只能得到少量离散点数据，难以恢复装配区域内粒子浓度场的连续时空分布；
[0027]通过构建平衡型物理信息神经网络模型，能够在有限监测点条件下实现粒子浓度场的连续重建与预测，相较于传统离散采样方法，不仅能够给出指定监测点的粒子浓度结果，还能输出整个洁净装配区域内粒子浓度场的连续时空分布，并且利用联合损失函数为约束，能够使平衡型物理信息神经网络模型的预测结果更加符合真实物理规律，避免小样本条件下的失真风险。
Smart Images
Figure CN123012689A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of superconducting cavity environment detection technology, specifically a method for predicting particle concentration fields. Background Technology
[0002] The superconducting cavity is a key component of a superconducting linear accelerator, and its clean assembly quality directly affects the cavity's radio frequency performance and operational reliability. During assembly, if tiny particles enter the cavity, they may induce field emission, secondary electron multiplication, and other problems, potentially leading to performance degradation or even failure. Therefore, effective monitoring of particle contamination within the clean assembly area of the superconducting cavity is necessary. Existing particle contamination monitoring methods mainly include discrete sampling and numerical simulation. These methods have the following limitations:
[0003] Discrete sampling relies on particle counters to sample a number of fixed points. It is limited by the number of particle counters, installation location, sampling frequency and engineering layout conditions. It can usually only obtain a small amount of discrete point data and it is difficult to recover the continuous spatiotemporal distribution of particle concentration field in the assembly area.
[0004] Numerical simulation uses computational fluid dynamics models to numerically solve the flow field and particle transport process in cleanrooms. It usually relies on relatively complete initial conditions, boundary conditions and fine mesh discretization, which makes modeling complex and computationally expensive. In actual cleanroom assembly conditions, particle source terms often have locality, transientity and uncertainty, resulting in problems such as insufficient adaptability, low reconstruction efficiency and difficulty in updating traditional numerical simulation methods in engineering applications. Summary of the Invention
[0005] In view of the deficiencies in the existing technology, the technical problem to be solved by this application is: how to reconstruct the spatiotemporal continuous distribution of particle concentration in a clean assembly area.
[0006] To achieve the above objectives, embodiments of this application provide a method for predicting particle concentration fields, the method comprising the following steps:
[0007] A planar coordinate system is constructed based on the two-dimensional cross-section of the clean assembly environment of the superconducting cavity;
[0008] Monitoring points were set up in the clean assembly area of the superconducting cavity to collect wind speed and particle concentration data at each monitoring point, and the collection time and spatial coordinates of the collection points were obtained to generate a dataset.
[0009] Using spatial coordinates and acquisition time as inputs, and predicted wind speed data and predicted particle concentration as outputs, a balanced physical information neural network model is constructed, and the balanced physical information neural network model is trained with the dataset.
[0010] The spatial coordinates of the time or time interval to be predicted and the acquisition time are input into the trained balanced physical information neural network model to obtain predicted wind speed data and predicted particle concentration.
[0011] In one embodiment, before generating the dataset, the wind speed data, particle concentration data, acquisition time and spatial coordinates of the monitoring points are preprocessed. The preprocessing method includes time alignment, format unification, outlier removal, background value correction and normalization.
[0012] In one embodiment, the balanced physical information neural network model is constrained by a joint loss function, which is obtained by weighted summation of data loss term, physical loss term, and boundary loss term.
[0013] In one embodiment, the data loss term is used to constrain the error between the predicted wind speed data and the predicted particle concentration and the corresponding wind speed data and particle concentration in the dataset.
[0014] The physical loss term is used to constrain the predicted wind speed data and predicted particle concentration to satisfy the background flow field model and particle transport model.
[0015] The boundary loss term is used to impose boundary constraints on the predicted wind speed data and the predicted particle concentration.
[0016] In one embodiment, the background flow field model is described based on incompressible flow control equations;
[0017] The particle transport model is described based on the convection-diffusion control equations.
[0018] In one embodiment, the boundary constraints include supply air boundary, return air boundary, and wall boundary.
[0019] In one embodiment, the training process of the balanced physical information neural network model includes:
[0020] The dataset is divided into a training set and a validation set according to a preset ratio;
[0021] Train a balanced physical information neural network model using a training set;
[0022] After the joint loss function converges, the balanced physical information neural network model is validated using a validation set.
[0023] The training of the balanced physical information neural network model is completed when the output of the model meets the preset standard.
[0024] In one embodiment, the judgment logic for whether the output result of the balanced physical information neural network model meets the preset standard is as follows:
[0025] Based on the predicted wind speed data and predicted particle concentration output by the balanced physical information neural network model, and the corresponding wind speed data and particle concentration in the validation set, the root mean square error, mean absolute error, and coefficient of determination are calculated. When the root mean square error and mean absolute error are both less than the preset threshold, and the coefficient of determination is greater than the preset threshold, the output result of the balanced physical information neural network model is determined to meet the preset standard.
[0026] Compared with the prior art, the advantages of this application are:
[0027] By constructing a balanced physical information neural network model, continuous reconstruction and prediction of particle concentration field can be achieved under limited monitoring point conditions. Compared with traditional discrete sampling methods, it can not only provide particle concentration results for specified monitoring points, but also output the continuous spatiotemporal distribution of particle concentration field in the entire clean assembly area. Furthermore, by using the joint loss function as a constraint, the prediction results of the balanced physical information neural network model can be made more consistent with the real physical laws, avoiding the risk of distortion under small sample conditions. Attached Figure Description
[0028] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0029] Figure 1 This is a flowchart of the particle concentration field prediction method in the embodiments of this application;
[0030] Figure 2 This is a schematic diagram showing the spatial layout, air supply outlet location, and return air zone location of the clean assembly area of the superconducting cavity in this embodiment of the application.
[0031] Figure 3 This is a schematic diagram of the planar coordinate system constructed in the embodiments of this application;
[0032] Figure 4 This is a schematic diagram of the layout of monitoring points in Embodiment 1 of this application;
[0033] Figure 5 This is a schematic diagram of the predicted particle concentration field in Embodiment 1 of this application;
[0034] Figure 6 This is a schematic diagram of the layout of monitoring points in Embodiment 2 of this application;
[0035] Figure 7 This is a schematic diagram of the predicted particle concentration field in Embodiment 2 of this application;
[0036] Figure 8This is a schematic diagram of the weight coordination mechanism of the joint loss function in an embodiment of this application. Detailed Implementation
[0037] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0038] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.
[0039] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.
[0040] This application provides a method for predicting particle concentration fields, referring to... Figure 1 The method includes the following steps:
[0041] S1. Construct a planar coordinate system based on the two-dimensional cross-section of the clean assembly environment of the superconducting cavity; refer to... Figure 2 It demonstrates the spatial layout of the clean assembly area of the superconducting cavity, the location of the air supply outlets, and the distribution of the return air zone. Based on this, a planar coordinate system is constructed as follows: Figure 3 As shown, it uses the clean assembly area of the superconducting cavity as the calculation area and treats the equipment in the clean assembly area of the superconducting cavity as obstacles.
[0042] S2. Set up monitoring points in the clean assembly area of the superconducting cavity. The monitoring points cover the area near the pollution source, the main particle propagation path area and the far-field attenuation area. Use wind speed sensors and particle counters to collect wind speed data (x-direction velocity component and y-direction velocity component) and particle concentration data (particle number concentration value at the corresponding time) at each monitoring point, and obtain the collection time and spatial coordinates of the collection points to generate a dataset.
[0043] Before generating the dataset, the wind speed data, particle concentration data, acquisition time and spatial coordinates of the monitoring points are preprocessed. The preprocessing methods include time alignment, format unification, outlier removal, background value correction and normalization.
[0044] S3. Using spatial coordinates and acquisition time as inputs, and predicted wind speed data and predicted particle concentration as outputs, construct a locally balanced physical information neural network (LbPINN). Train the LbPINN model with the dataset. The LbPINN model is constrained by the joint loss function.
[0045] S4. Input the spatial coordinates of the time to be predicted or the time interval to be predicted and the acquisition time into the trained balanced physical information neural network model to obtain the predicted wind speed data and the predicted particle concentration.
[0046] By constructing a balanced physical information neural network model, continuous reconstruction and prediction of particle concentration field can be achieved under limited monitoring point conditions. Compared with traditional discrete sampling methods, it can not only provide particle concentration results for specified monitoring points, but also output the continuous spatiotemporal distribution of particle concentration field in the entire clean assembly area.
[0047] In one embodiment, the joint loss function is obtained by weighted summation of data loss term, physical loss term, and boundary loss term, wherein the data loss term is used to constrain the error between the predicted wind speed data and the predicted particle concentration and the corresponding wind speed data and particle concentration in the dataset.
[0048] The physical loss term is used to constrain the predicted wind speed data and predicted particle concentration to satisfy the background flow field model (based on the incompressible flow control equations) and the particle transport model (based on the convection-diffusion control equations).
[0049] The boundary loss term is used to impose boundary constraints on the predicted wind speed data and predicted particle concentration. These boundary constraints include the supply air boundary, return air boundary, and wall boundary.
[0050] Therefore, by using the joint loss function as a constraint, the prediction results of the balanced physical information neural network model can be made more consistent with the real physical laws, avoiding the risk of distortion under small sample conditions.
[0051] In one embodiment, the training process of the above-mentioned balanced physical information neural network model includes:
[0052] The dataset is divided into a training set and a validation set according to a preset ratio (7:3);
[0053] A balanced physical information neural network model was trained using a training set. Spatial coordinates and acquisition time from the training set were input into the balanced physical information neural network model to obtain predicted wind speed data and predicted particle concentration. During the process, the Adam optimizer was used for iterative training, with 30,000 iterations.
[0054] When the joint loss function converges (≤1×10) -6 After that, the balanced physical information neural network model is validated using a validation set. The spatial coordinates and acquisition time in the validation set are input into the balanced physical information neural network model to obtain predicted wind speed data and predicted particle concentration.
[0055] Based on the predicted wind speed data and predicted particle concentration output by the balanced physical information neural network model, and the corresponding wind speed data and particle concentration in the validation set, the root mean square error, mean absolute error, and coefficient of determination are calculated.
[0056] Determine whether the root mean square error, mean absolute error, and coefficient of determination meet the preset standards. If the root mean square error is less than the preset threshold (≤0.05), the mean absolute error is less than the preset threshold (≤5%), and the coefficient of determination is greater than the preset threshold (≥0.95), the output of the balanced physical information neural network model is determined to meet the preset standards, and the training of the balanced physical information neural network model is completed; otherwise, continue training the balanced physical information neural network model.
[0057] Furthermore, to mitigate the imbalance between data items and physical items caused by noisy data, learnable weight parameters are introduced into the joint loss function to obtain a balanced physical information neural network, referring to... Figure 8 It demonstrates a mechanism for coordinating the relative weights of data loss terms, physical loss terms, and boundary loss terms. This enables the balanced physical information neural network model to dynamically coordinate the relative weights of data loss terms, physical loss terms, and boundary loss terms during training, thereby improving training stability and noise resistance.
[0058] The following describes the prediction of particle concentration fields using two examples.
[0059] Example 1
[0060] In the clean assembly process of superconducting cavities, the purging process is taken as a typical contamination scenario. The rapid removal of particles from the surface of flanges or related parts under the action of an ion gun is equated to a localized, short-duration pulse-type contamination source. Based on this, the steps for predicting particle concentration include:
[0061] Several monitoring points were deployed in the vicinity of the pollution source, the main transmission path area, and the far-field attenuation area. The monitoring points were deployed as follows: Figure 4 As shown, wind speed data and particle concentration are collected, and combined with the collection time and spatial coordinates of the monitoring points, a dataset is generated after preprocessing.
[0062] Train a balanced physical information neural network model based on the dataset (see the training process of the balanced physical information neural network model above).
[0063] By inputting the spatial location and temporal variables to be predicted into a trained balanced physical information neural network model, a continuous distribution of the particle concentration field during the purging process as a function of time is obtained, such as... Figure 5 The predicted particle concentration field shows that the particle concentration rapidly increases in the vicinity of the source term within a short period of time, and then diffuses and decays along the mainstream direction. This demonstrates that the prediction results of the balanced physical information neural network model can reflect the rapid release-diffusion migration-gradual decay process of the purging process.
[0064] Example 2
[0065] In the clean assembly process of superconducting cavities, the swivel cap process is taken as a typical contamination scenario. The particle release process generated by the contact friction between the swivel cap sleeve and the flange is equated to a continuous strip-shaped contamination source distributed along the contact area. Based on this, the steps for predicting particle concentration include:
[0066] Several monitoring points were deployed in the vicinity of the pollution source, the main transmission path area, and the far-field attenuation area. The monitoring points were deployed as follows: Figure 6 As shown, wind speed data and particle concentration are collected, and combined with the collection time and spatial coordinates of the monitoring points, a dataset is generated after preprocessing.
[0067] Train a balanced physical information neural network model based on the dataset (see the training process of the balanced physical information neural network model above).
[0068] By inputting the spatial location and temporal variables to be predicted into a trained balanced physical information neural network model, a continuous distribution of the particle concentration field of the swirl cap process over time is obtained, such as... Figure 7 The predicted particle concentration field shown indicates that the particle concentration gradually accumulates near the source region and diffuses downstream along the main propagation direction, exhibiting characteristics of continuous release, local accumulation, directional diffusion, and slow decay.
[0069] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.
[0070] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.
[0071] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.
[0072] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.
[0073] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.
[0074] The above are merely specific embodiments of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the scope of the claims.
Claims
1. A method for predicting particle concentration fields, characterized in that, The method includes the following steps: A planar coordinate system is constructed based on the two-dimensional cross-section of the clean assembly environment of the superconducting cavity; Monitoring points were set up in the clean assembly area of the superconducting cavity to collect wind speed and particle concentration data at each monitoring point, and the collection time and spatial coordinates of the collection points were obtained to generate a dataset. Using spatial coordinates and acquisition time as inputs, and predicted wind speed data and predicted particle concentration as outputs, a balanced physical information neural network model is constructed, and the balanced physical information neural network model is trained with the dataset. The spatial coordinates of the time or time interval to be predicted and the acquisition time are input into the trained balanced physical information neural network model to obtain predicted wind speed data and predicted particle concentration.
2. The particle concentration field prediction method according to claim 1, characterized in that, Before generating the dataset, the wind speed data, particle concentration data, acquisition time and spatial coordinates of the monitoring points are preprocessed. The preprocessing methods include time alignment, format unification, outlier removal, background value correction and normalization.
3. The particle concentration field prediction method according to claim 1, characterized in that, The balanced physical information neural network model is constrained by a joint loss function, which is obtained by weighted summation of data loss term, physical loss term, and boundary loss term.
4. The particle concentration field prediction method according to claim 3, characterized in that, The data loss term is used to constrain the error between the predicted wind speed data and the predicted particle concentration and the corresponding wind speed data and particle concentration in the dataset. The physical loss term is used to constrain the predicted wind speed data and predicted particle concentration to satisfy the background flow field model and particle transport model. The boundary loss term is used to impose boundary constraints on the predicted wind speed data and the predicted particle concentration.
5. The particle concentration field prediction method according to claim 4, characterized in that, The background flow field model is described based on incompressible flow control equations; The particle transport model is described based on the convection-diffusion control equations.
6. The particle concentration field prediction method according to claim 4, characterized in that, The boundary constraints include the supply air boundary, the return air boundary, and the wall boundary.
7. The particle concentration field prediction method according to claim 3, characterized in that, The training process of the balanced physical information neural network model includes: The dataset is divided into a training set and a validation set according to a preset ratio; A balanced physical information neural network model was trained using a training set. After the joint loss function converges, the balanced physical information neural network model is validated using a validation set. The training of the balanced physical information neural network model is completed when the output of the model meets the preset standard.
8. The particle concentration field prediction method according to claim 7, characterized in that, The logic for determining whether the output of the balanced physical information neural network model meets the preset standard is as follows: Based on the predicted wind speed data and predicted particle concentration output by the balanced physical information neural network model, and the corresponding wind speed data and particle concentration in the validation set, the root mean square error, mean absolute error, and coefficient of determination are calculated. When the root mean square error and mean absolute error are both less than the preset threshold, and the coefficient of determination is greater than the preset threshold, the output result of the balanced physical information neural network model is determined to meet the preset standard.