Spacecraft physical field prediction model construction method based on maximum evidence logarithm method
The maximum evidence logarithm method measures the transferability of pretrained models and selects the appropriate model for transfer learning, which solves the problem of high sample demand and difficulty in selecting pretrained models in spacecraft physics prediction, achieving more efficient model construction and performance improvement.
Patent Information
- Application Number
- CN202510162950.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-14
- Publication Date
- 2025-06-13
AI Technical Summary
Deep learning models require a large number of samples in spacecraft physics prediction, and lack effective pre-trained model transferability measurement methods, making it difficult to guarantee the transfer learning effect.
The maximum evidence logarithm method is used to calculate the compatibility between the physical field features and the corresponding labels, quantify the transferability of the pre-trained model, and select the pre-trained model with the greatest compatibility for transfer learning to build the final physics pre-trained model.
This reduces the model's demand for samples, improves model performance, and ensures the effectiveness of transfer learning.
Smart Images
Figure CN120145811A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of aerospace engineering, and particularly relates to a method for constructing a physical field prediction model of a spacecraft based on the maximum evidence logarithm method. Background Art
[0002] The physical field is an important feature reflecting the performance and operating state of complex systems such as spacecraft. Achieving fast and accurate physical field prediction can significantly improve the optimization design ability of complex systems and the state perception and control level. For example: in the aerospace field, predicting the aerodynamic heat environment helps optimize the design of the thermal protection system of spacecraft, improving flight efficiency and safety. The physical field prediction technology can support system management and maintenance. By establishing a virtual model to simulate the operating state of the system under different working conditions, it helps engineers foresee the performance of equipment or structures under various working conditions during the design stage, thereby optimizing design parameters and avoiding potential performance bottlenecks or safety hazards.
[0003] In the physical field prediction task, the deep learning method has become the preferred technical approach due to its unique structural characteristics and learning ability. The specific advantages include: High-dimensional data processing ability: Physical field data usually involves multi-variable and high-dimensional information (such as the distribution of temperature, pressure, velocity, etc. in time and space). Deep learning networks (such as convolutional neural network CNN and recurrent neural network RNN) can efficiently process such complex data structures, capture multi-scale features, and thus improve the prediction accuracy. Nonlinear modeling advantage: Physical processes often involve complex nonlinear relationships, which are difficult to accurately capture by traditional methods. Deep learning models can learn these complex nonlinear patterns through multi-layer nonlinear transformations, providing more realistic prediction results, especially performing well in dealing with highly nonlinear phenomena such as turbulence and convection. End-to-end learning and automatic feature extraction: Deep learning models can automatically learn useful features from raw data, reducing the need for manual feature engineering, lowering the complexity of model construction, and at the same time enhancing the generalization ability of the model and its adaptability to new data.
[0004] Although the deep learning method performs well in physical field prediction, it still faces the following challenges in practical applications: Large sample demand: Deep learning models usually require a large number of training samples to achieve high prediction accuracy. However, the acquisition cost of spacecraft physical field data is high and the sample size is limited, making it difficult to meet the needs of deep learning models for a large number of samples. Difficult selection of pre-trained models: To reduce the dependence on a large number of samples, transfer learning technology is introduced, that is, using pre-trained models trained on other tasks to assist in the model training of the current task. However, how to select the most suitable model for the current task from numerous pre-trained models has become the key to the success of transfer learning. The existing technology lacks an effective method for measuring the transferability of pre-trained models, resulting in the difficulty of ensuring the effect of transfer learning. Summary of the Invention
[0005] To solve some or all of the technical problems existing in the above-mentioned prior art, the present invention provides a method for constructing a physical field prediction model of a spacecraft based on the maximum evidence logarithm method.
[0006] The technical solution of the present invention is as follows:
[0007] There is provided a method for constructing a physical field prediction model of a spacecraft based on the maximum evidence logarithm method, the method comprising:
[0008] Obtain a data set for a target physical field prediction task, the data set including physical field features F and corresponding physical field labels y, the physical field features being obtained by feature extraction of input data by a pre-trained model;
[0009] Calculate the compatibility between the physical field features and the corresponding labels using the maximum evidence logarithm method to quantify the transferability of each pre-trained model;
[0010] Select the pre-trained model with the maximum compatibility as the optimal pre-trained model;
[0011] Use the optimal pre-trained model to construct a final physical field pre-trained model through transfer learning, and use the physical field prediction model to predict an actual task.
[0012] In an embodiment of the present invention, calculating the compatibility between the physical field features and the corresponding labels using the maximum evidence logarithm method includes:
[0013] Parameterize the relationship between the physical field features F and the physical field label y using a linear model, and calculate the evidence p(y|F);
[0014] Determine hyperparameters α and β that maximize the evidence p(y|F) through an alternating optimization method;
[0015] Calculate the logarithm of the maximum evidence where n is the number of samples.
[0016] In an embodiment of the present invention, the calculation of the LogME value is accelerated by matrix factorization technology, specifically including:
[0017] Perform singular value decomposition on the physical field feature matrix F to obtain F = UΣV T , where U and V are orthogonal matrices, and Σ is a diagonal matrix;
[0018] Through the decomposition, directly calculating the inverse of the matrix and matrix multiplication is avoided, and the computational complexity is reduced to O(n 2 ).
[0019] In the embodiments of the present invention, the physical field feature F is extracted from the output of the intermediate layer of the pre-trained model.
[0020] In the embodiments of the present invention, the pre-trained model is a deep learning model trained based on other physical field prediction tasks, including a convolutional neural network (CNN) or a recurrent neural network (RNN).
[0021] In the embodiments of the present invention, the input data includes the distribution data of temperature, pressure, and velocity in the spacecraft cabin over time and space.
[0022] In the embodiments of the present invention, the transfer learning technique includes the following steps:
[0023] Add a fully connected layer to the tail of the selected pre-trained model and fine-tune the fully connected layer;
[0024] Use the dataset of the target physical field prediction task to train the model until the model converges.
[0025] In the embodiments of the present invention, an end-to-end optimization technique is adopted during the transfer learning process to reduce the complexity of manual feature engineering.
[0026] In the embodiments of the present invention, the method is applicable to the prediction of the aerodynamic heat environment, vibration field, stress field, and other physical field prediction tasks in the spacecraft cabin.
[0027] The main advantages of the technical solution of the present invention are as follows:
[0028] The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method of the present invention measures the transferability of the pre-trained model by calculating the compatibility between the physical field features of the physical field prediction task and the corresponding physical field labels, selects a suitable pre-trained model based on the measurement results, and uses transfer learning to assist in the training of the physical field prediction model, which can not only reduce the demand for samples of the model but also improve the model performance. Description of the Drawings
[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.
[0030] Figure 1 It is a flowchart of the method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method of the embodiments of the present invention;
[0031] Figure 2It is a directed graph model for calculating evidence in the method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to an embodiment of the present invention. Detailed implementation manners
[0032] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
[0033] The following details the technical solutions provided by the embodiments of the present invention in conjunction with the drawings.
[0034] An embodiment of the present invention provides a method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method. As shown in the attached Figure 1 drawing, it includes:
[0035] S1. Obtain a data set for the target physical field prediction task. The data set includes physical field features F and corresponding physical field labels y. The physical field features are obtained by a pre-trained model extracting features from the input data.
[0036] Constructing a data set including physical field features and corresponding physical field labels is the premise and basis for subsequent calculation of compatibility. The physical field features F are obtained by a pre-trained model extracting features from the input data. For example, the input data can be physical field data such as temperature, pressure, and velocity inside the spacecraft cabin. The pre-trained model maps these data into high-dimensional feature representations F. The physical field label y is the true value corresponding to the input data, representing the physical field quantity to be predicted. For example, in a temperature distribution prediction task, y may be the temperature value at a certain position.
[0037] Extracting the features F through the pre-trained model can effectively capture the high-level information of the input data and reduce the complexity of manual feature engineering. The physical field label y provides a supervision signal for training and evaluating the model.
[0038] S2. Use the maximum evidence logarithm method to calculate the compatibility between the physical field features and the corresponding labels to quantify the transferability of each pre-trained model.
[0039] The maximum evidence logarithm method (LogME) evaluates whether the pre-trained model is suitable for transfer to the current task by quantifying the compatibility between the physical field features F and the label y.
[0040] Specifically, the LogME method calculates the maximum evidence logarithm (LogME value) between F and y. The higher the LogME value, the better the compatibility between the features of the pre-trained model and the labels of the current task.
[0041] The LogME method provides an efficient and reliable way to measure the transferability of pre-trained models, which can help select the most suitable pre-trained model for the current task. By quantifying the compatibility, it avoids the uncertainty of subjectively selecting pre-trained models.
[0042] S3. Select the pre-trained model with the maximum compatibility as the optimal pre-trained model.
[0043] According to the LogME value, evaluate the transferability of multiple pre-trained models. Select the pre-trained model with the highest LogME value as the optimal pre-trained model.
[0044] By selecting the pre-trained model with the highest LogME value, the feature extraction ability of the pre-trained model can be utilized to the greatest extent, improving the prediction accuracy of the target task and avoiding the performance degradation caused by blindly selecting pre-trained models.
[0045] S4. Use the optimal pre-trained model to construct the final physical field pre-trained model through transfer learning, and use the physical field prediction model to predict the actual task.
[0046] Use the selected optimal pre-trained model to construct the final physical field prediction model through transfer learning techniques. The specific way of transfer learning can be to add a new layer (such as a fully connected layer) at the end of the pre-trained model and fine-tune the newly added layer. Use the dataset of the target physical field prediction task to train the model until the model converges. After the physical field prediction model is constructed, it can be used to predict the actual task. For example, in the task of predicting the temperature distribution inside the spacecraft cabin, input new physical field data such as temperature, pressure, and velocity, and the model outputs the predicted temperature distribution.
[0047] Through transfer learning, a physical field prediction model suitable for the target task can be quickly constructed, reducing the need for a large number of training samples. Utilizing the feature extraction ability of the pre-trained model can significantly improve the prediction accuracy and generalization performance of the model.
[0048] In summary, the method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method provided by the embodiments of the present invention measures the transferability of the pre-trained model by calculating the compatibility between the physical field features of the physical field prediction task and the corresponding physical field labels, selects a suitable pre-trained model based on the measurement results, and assists in the training of the physical field prediction model through transfer learning, which can not only reduce the demand for samples of the model but also improve the performance of the model.
[0049] In some alternative embodiments of the present invention, the compatibility between the physical field features and the corresponding labels is calculated using the maximum evidence logarithm method, including:
[0050] S21, parameterize the relationship between the physical field feature F and the physical field label y using a linear model, and calculate the evidence p(y∣F). Specifically, it can be assumed that the label y is a linear combination of the feature F plus a certain amount of noise. This parameterization method of the linear model is simple and efficient, and can capture the basic relationship between the feature and the label.
[0051] Calculate the evidence p(y|F), that is, evaluate the overall compatibility between the feature F and the label y without fixing the model parameters. The calculation of the evidence is based on the Bayesian method, and a comprehensive evaluation index is obtained by integrating over all possible model parameters.
[0052] S22, determine the hyperparameters α and β that maximize the evidence p(y|F) through an alternating optimization method.
[0053] Determine the hyperparameters α and β that maximize the evidence p(y|F) through an alternating optimization method. The process of alternating optimization includes fixing a part of the parameters and optimizing the other part of the parameters, and iterating repeatedly until convergence.
[0054] S23, calculate the logarithm of the maximum evidence where n is the number of samples. Calculate the logarithm of the maximum evidence LogME, that is, take the logarithm of the evidence p(y|F) and divide by the number of samples n. The LogME value can be intuitively interpreted as: the average maximum logarithmic evidence of the label y given the feature F.
[0055] In this embodiment, the evidence is calculated by the Bayesian method, which comprehensively considers all possible model parameters, avoids the overfitting problem, and improves the reliability of model evaluation. The alternating optimization method can efficiently determine the optimal hyperparameters and improve the calculation efficiency. The LogME value provides an intuitive evaluation index that can help select the pre-trained model most suitable for the current task.
[0056] In some alternative embodiments of the present invention, the calculation of the LogME value is accelerated by matrix factorization technology, specifically including:
[0057] Perform singular value decomposition on the physical field feature matrix F to obtain F = UΣV T , where U and V are orthogonal matrices, and Σ is a diagonal matrix;
[0058] Through the decomposition, directly calculating the inverse of the matrix and matrix multiplication are avoided, and the computational complexity is reduced to O(n 2 ).
[0059] To improve the computational efficiency of LogME, matrix factorization technology is used to decompose the physical field feature matrix F. Specifically, the feature matrix F is decomposed into the product of three matrices, thus avoiding direct calculation of matrix inversion and matrix multiplication. Through matrix factorization, the computational complexity of LogME calculation is reduced from O(n 3 ) to O(n 2 ), significantly improving the computational efficiency.
[0060] In this embodiment, matrix factorization technology significantly reduces the computational complexity of LogME calculation and is applicable to large-scale datasets. The computational acceleration enables the LogME method to be more efficiently applied to practical tasks.
[0061] In some alternative embodiments of the present invention, the physical field feature F is extracted from the output of the intermediate layer of a pre-trained model. For example, for a convolutional neural network (CNN), the output of the last convolutional layer can be selected as F. The intermediate layer output can capture high-level features of the input data and provide a richer feature representation. Automatically extracting features through the pre-trained model reduces the complexity of manual feature engineering.
[0062] In some alternative embodiments of the present invention, the pre-trained model is a deep learning model trained for other physical field prediction tasks, including a convolutional neural network CNN or a recurrent neural network RNN.
[0063] The pre-trained model is a deep learning model trained for other physical field prediction tasks, such as a convolutional neural network (CNN) or a recurrent neural network (RNN). These pre-trained models have learned rich feature representations on other tasks and can be used as alternatives for transfer learning in the present invention.
[0064] The selection of the pre-trained model can be determined by those skilled in the art according to the actual situation, and will not be elaborated in the embodiments of the present invention.
[0065] In some alternative embodiments of the present invention, the input data includes the distribution data of temperature, pressure, and velocity in the spacecraft cabin over time and space. The multi-variable and multi-dimensional input data can comprehensively reflect the complex characteristics of the physical field, provide rich input information for the model, and improve the prediction accuracy. The combination of time and space distribution data enhances the model's ability to model dynamic physical fields.
[0066] In some alternative embodiments of the present invention, the transfer learning technology includes the following steps:
[0067] Add a fully connected layer to the tail of the selected pre-trained model and fine-tune the fully connected layer;
[0068] Use the dataset of the target physical field prediction task to train the model until the model converges.
[0069] By fine-tuning the fully connected layer, it is possible to quickly adapt to the target task and reduce the training time. By adding a new fully connected layer for transfer learning, the feature extraction ability of the pre-trained model is retained, and the generalization performance of the model is improved.
[0070] In some alternative embodiments of the present invention, an end-to-end optimization technique is adopted during the transfer learning process to reduce the complexity of manual feature engineering. Through end-to-end optimization, the complexity of manual feature engineering is reduced, the automation degree of the model is improved, and the model parameters can be globally optimized to improve the prediction accuracy.
[0071] In some alternative embodiments of the present invention, the method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method provided by the embodiments of the present invention is applicable to tasks such as aerodynamic heat environment prediction, vibration field prediction, stress field prediction, and other physical field prediction tasks inside the spacecraft cabin. This method can be widely applied to various physical field prediction tasks and has high generality.
[0072] The following describes in detail the method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method of the present invention in combination with a specific embodiment.
[0073] The present invention proposes a method of logarithm of maximum evidence (LogME) to measure the transferability of the physical field prediction model. The transferability of the pre-trained model is measured by calculating the compatibility between the physical field features of the physical field prediction task and the corresponding physical field labels, mainly including evidence calculation, evidence maximization and LogME, calculation acceleration, and physical field prediction tasks. Specifically:
[0074] In the selection of the pre-trained model for physical field prediction task adaptation, M pre-trained models are given and the target dataset This dataset has an evaluation metric (MAE, MSE, etc.) that can measure the transfer performance T of the model φ m by appropriate hyperparameter tuning. m . A practical measurement method should output a measurement score S m for each pre-trained model φ m (ideally, without fine-tuning φ on D m ), and the measurement score should be strongly correlated with so that the best-performing pre-trained model can be selected by simply measuring the score.
[0075] I. Evidence calculation
[0076] First, consider a simple case where the physical field features are The target physical field label is The feature matrix contains all features Represents all labels.
[0077] A direct measure of the compatibility between feature F and label y is the probability density p(y|F), which is usually not available without a parametric model. Since conventional transfer learning usually adds a fully connected layer at the tail of the pre-trained model, a linear model parameterized by w is adopted here to parameterize the features. A naive approach to dealing with the linear model is to find the optimal w through logistic regression or linear regression * , and evaluate the pre-trained model through the likelihood p(y|F, w * ). However, this tends to lead to overfitting. In this embodiment, the evidence (marginal likelihood) p(y|F) = ∫p(w)p(y|F, w)dw is adopted, which integrates over all possible w and is superior to simply using an optimal value of w * . This evidence-based method is an elegant model selection method and has a rigorous theoretical basis.
[0078] For p(w) and p(y|F, w), a graphical model as shown in Figure 2 is adopted here, where α and β are two positive parameters. The prior distribution of the weights is an isotropic multivariate Gaussian distribution and the distribution of each observation is a one-dimensional normal distribution
[0079] According to Figure 2 the causal relationship in and the basic principle of the graphical model, the evidence can be analytically calculated as Equation (1):
[0080]
[0081] Since When A is a positive definite matrix, Equation (1) can be simplified by taking the logarithm. Here, let A = αI + βF T F, m = βA -1 F T y, then there is:
[0082]
[0083] II. Evidence Maximization and LogME
[0084] To determine α and β, the values that maximize the evidence can be selected here, that is, using Since m and A are coupled with each other, maximizing is usually difficult. However, this problem can be solved here by alternately calculating m, γ and maximizing α, β (m, γ fixed), and finally obtaining Equation (3):
[0085]
[0086] Among them, I represents the identity matrix.
[0087] When the iteration converges, the logarithm of the maximum evidence can be used to evaluate the compatibility between features and labels. Since is linearly related to n, it is divided by n here and called LogME (logarithm of the maximum evidence), which can be intuitively interpreted as the average maximum logarithmic evidence of the label given the pre-trained features.
[0088] III. Computational Acceleration
[0089] Although the maximum evidence of the Bayesian method has many advantages, it inherits the disadvantage of high computational complexity of the Bayesian method. The original implementation of the LogME algorithm has complexity, and for typical twin-computation application scenarios, the computational cost is very high, which will lead to the computational time being comparable to that of fine-tuning the pre-trained model.
[0090] To improve the computational efficiency, here we utilize the decomposition of F T F = Vdiag{σ}V T (V is an orthogonal matrix) to avoid matrix inversion and matrix multiplication. Specifically, let Λ = diag{(α + βσ)}, then we have A -1 = VΛ -1 V T . To avoid the matrix multiplication A -1 F T , note that y is a column vector, and the associative law allows for the fast calculation of A -1 F T y = (V(Λ -1 (V T (F T y))))). Thus, in each for loop, only Λ needs to be updated instead of the expensive A -1 , and all matrix multiplications can be simplified to matrix-vector products, avoiding matrix inversion.
[0091] IV. Physical Field Prediction Task
[0092] In this embodiment, the transferability of the physical field prediction pre-trained model is measured through the above-mentioned steps I to III, and based on the measurement results, a suitable pre-trained model can be selected to quickly construct the proxy model for the current physical field prediction task.
[0093] It should be noted that in this text, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. In addition, in this text, "front", "rear", "left", "right", "upper" and "lower" are all referenced with respect to the placement state shown in the drawings.
[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method, characterized in that: include: Obtain a data set for a target physical field prediction task, the data set including a physical field feature F and a corresponding physical field label y, the physical field feature being obtained by extracting features from input data using a pre-trained model; The maximum evidence logarithm method is used to calculate the compatibility between physical field features and corresponding labels to quantify the transferability of each pre-trained model; Selecting the pre-trained model with the greatest compatibility as the optimal pre-trained model; The optimal pre-trained model is used to construct a final physical field pre-trained model through transfer learning, and the physical field prediction model is used to predict the actual task.
2. The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to claim 1, characterized in that: The method of using the maximum evidence logarithm method to calculate the compatibility between the physical field characteristics and the corresponding labels includes: Use a linear model to parameterize the relationship between the physical field feature F and the physical field label y, and calculate the evidence p(y|F); Determine the hyperparameters α and β that maximize the evidence p(y|F) by an alternating optimization method; Calculate the logarithm of the maximum evidence Where n is the number of samples.
3. The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to claim 2, characterized in that: The calculation of the LogME value is accelerated by matrix decomposition technology, specifically including: Perform singular value decomposition on the physical field characteristic matrix F to obtain F = UΣV T , where U and V are orthogonal matrices and Σ is a diagonal matrix; By using the decomposition, the direct calculation of the matrix inverse and matrix multiplication is avoided, and the computational complexity is reduced to O(n 2 ).
4. The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to claim 1, characterized in that: The physical field feature F is extracted through the intermediate layer output of the pre-trained model.
5. The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to claim 1, characterized in that: The pre-trained model is a deep learning model trained based on other physical field prediction tasks, including a convolutional neural network CNN or a recurrent neural network RNN.
6. The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to claim 1, characterized in that: The input data include the distribution data of temperature, pressure and velocity in the spacecraft cabin in time and space.
7. The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to claim 1, characterized in that: The transfer learning technique comprises the following steps: Add a fully connected layer to the end of the selected pre-trained model and fine-tune the fully connected layer; The model is trained using a data set of a target physical field prediction task until the model converges.
8. The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to claim 1, characterized in that: End-to-end optimization technology is used in the transfer learning process to reduce the complexity of manual feature engineering.
9. The method for constructing a spacecraft physical field prediction model based on the maximum evidence logarithm method according to claim 1, characterized in that: The method is applicable to the prediction of aerodynamic thermal environment, vibration field, stress field and other physical field prediction tasks in a spacecraft cabin.