A method and related device for controlling the assembly accuracy of precision equipment with human in the loop

Abstract

The present invention belongs to the field of precision equipment assembly technology, and discloses a precision control method for precision equipment assembly with a human in the loop and related devices, the method comprising: obtaining parameter input and precision output according to data of actual measurement and adjustment of precision equipment, combining geometric parameters and physical parameters to establish a high-precision sample set database; obtaining empirical data, constructing empirical models and preference learning models based on the empirical data; performing Bayesian optimization based on the high-precision sample set database, establishing a digital twin model in combination with a human in the loop, optimizing through the digital twin model, combining the empirical model with the preference learning model, and selecting an assembly adjustment scheme that meets assembly accuracy and performance. The present invention introduces empirical data into the optimization and control of complex precision equipment assembly, provides an effective method for realizing rapid and accurate control of complex precision equipment, and is of great significance to significantly improving the assembly accuracy of complex precision equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of precision equipment assembly, and relates to a method for controlling the assembly accuracy of precision equipment with a human in the loop and a related device. Background Art

[0002] For complex precision equipment assembly objects, their assembly accuracy directly affects their service performance. Precision mechanisms have extremely high requirements for assembly accuracy, such as rotation accuracy, surface accuracy, pointing accuracy and other indicators. Generally speaking, assembly accuracy will directly determine the service performance of the assembly object. The higher the accuracy, the better the service performance. Traditional assembly theories and methods based on geometric error transmission and geometric quantity control can no longer thoroughly analyze the current problems, nor can they further guide the improvement direction of structural design and assembly processes. Ensuring the accuracy of complex precision assembly by simply improving the processing accuracy or assembly accuracy of the equipment has little effect, but it brings a sharp increase in processing costs. Therefore, it is difficult to meet the permanent accuracy requirements of complex precision equipment through a one-time assembly in the initial assembly link. The key to improving assembly accuracy and assembly performance lies in how to optimize and control quickly and efficiently.

[0003] Taking the optimization and control of large-scale satellite-borne planar antennas as an example, large-scale satellite-borne antennas are essential equipment in the aerospace field and are widely used in various fields such as communications, navigation, deep space exploration and anti-missile warning. For a planar antenna, it is generally composed of an unfolding mechanism, an array frame, a locking hinge, a free hinge, a drive motor, etc. The assembly accuracy of its array has a decisive influence on its electrical performance, and with the development of high power, high gain and high frequency, its accuracy requirements are further increased. Therefore, in actual engineering, it is necessary to adjust its supporting structure and the gaskets at the hinges to adjust its assembly accuracy to meet the standards. However, a large satellite-borne planar antenna is a complex multi-loop closed-chain mechanism with various error factors that affect each other. It is extremely complex and difficult to control.

[0004] At present, the control methods of actual engineering can be generally divided into two categories: the first category is the experience control method based on certain actual installation and adjustment and the accumulation of process knowledge; the second category is the precision control method under the guidance of optimization ideas, but the existing optimization control model is only a single theory drive.

[0005] The first type belongs to the method of measuring and adjusting based on experience, which adopts the local trial and assembly mode of "measurement-adjustment-remeasurement-readjustment". Specifically, in a gravity unloading environment, technicians assemble the deployable structure according to the process regulations, and then, with the assistance of a lifting vehicle, use camera photogrammetry systems, theodolites and other measurement methods to obtain the assembly accuracy, and compare the measurement results with the design accuracy. The length of the rod system is adjusted through local repeated trials to ensure that the assembly accuracy meets the standard. However, this local trial and assembly mode mainly relies on the experience and "feeling" of the process personnel, lacks directionality and pertinence, and has prominent common problems such as low assembly accuracy and uncontrollable assembly cycle. It is difficult to further improve the assembly accuracy while meeting the assembly stress requirements, and it is impossible to take into account both geometric and physical assembly performance.

[0006] The second type is the optimization and control method based on the theoretical simulation analysis model. In this type of method, a theoretical simulation analysis model is established, and then the optimization and adjustment calculation is performed in combination with existing optimization algorithms, such as genetic algorithm, particle swarm algorithm, gradient optimization algorithm, etc., and then direct control is performed based on the optimal adjustment amount solved. That is, the control is essentially abstracted into an optimization problem, so as to be solved from a mathematical perspective. The advantages of this method are reflected in two aspects: (1) It can directly optimize and control the assembly from the essence; (2) The cost of this control is extremely low, and the entire calculation process is executed in the computer, without any actual measurement and control costs. However, this type of method has the following disadvantages: On the one hand, due to the extremely complex antenna equipment, this theoretical model has a certain deviation from the actual prediction result to a large extent, and as the complexity of the antenna structure increases, this deviation will become larger and larger, which makes it difficult to achieve the solution of the global optimal adjustment amount by using theoretical model optimization, and there is a bottleneck in the control accuracy. On the other hand, in this optimization and control process, the measured data and the experience summarized by the process personnel in previous assembly are discarded, which not only makes this control process inconsistent with the actual engineering, but also cannot be combined with the above useful information, making it difficult to improve the control efficiency and control accuracy. Summary of the invention

[0007] The purpose of the present invention is to provide a method and related device for controlling the assembly accuracy of precision equipment with humans in the loop, so as to solve the problem of low control efficiency and control accuracy in the prior art.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A method for controlling the assembly accuracy of precision equipment with a human in the loop, comprising:

[0010] Based on the actual measurement and adjustment data of precision equipment, parameter input and accuracy output are obtained, and a high-precision sample set database is established by combining geometric parameters and physical parameters;

[0011] Obtain experience data and build experience models and preference learning models based on the experience data;

[0012] Bayesian optimization is performed based on the high-precision sample set database. The empirical model and preference learning model are combined to establish a digital twin model based on the human-in-the-loop to optimize assembly parameters and select an assembly adjustment plan that meets assembly accuracy and performance.

[0013] Furthermore, the geometric parameters and physical parameters include geometric shape, size, tolerance, assembly fit relationship, elastic modulus, stiffness, Poisson's ratio and density.

[0014] Furthermore, the method for constructing the empirical model and the preference learning model is:

[0015] Introduce empirical data into the optimization process, mathematically characterize and model the empirical data through quantitative methods, and establish an empirical model suitable for optimization judgment;

[0016] Based on preference learning, the range conditions of the optimal judgment provided by the empirical model are determined, the likelihood model is used to construct the preference learning function, and the preference learning model is improved through iterative optimization.

[0017] Furthermore, the method for establishing the digital twin model is:

[0018] On the basis of traditional Bayesian optimization, empirical models and preference learning models are introduced, and the human-in-the-loop acquisition function and Gaussian regression process standard acquisition function are used to establish a digital twin model of human-in-the-loop.

[0019] Furthermore, the process of optimizing assembly parameters through the digital twin model includes:

[0020] Determine the assembly parameters that can be optimized and adjusted according to the actual working conditions;

[0021] Build an optimization control model based on the standard acquisition function to solve the assembly parameter adjustment plan;

[0022] Determine whether the empirical model participates in and affects the current optimization, build an optimization control model based on the human-in-the-loop acquisition function, solve the assembly parameter adjustment plan, and select the assembly parameter adjustment plan that is more suitable for the actual working conditions.

[0023] Furthermore, whether the empirical model participates in and affects the current optimization is based on the current optimization process:

[0024] In the early stage of optimization, the optimal solution selected by the empirical model is mainly used to speed up the process of searching for the optimal range based on the empirical model;

[0025] In the later stage of optimization, as the optimization progresses and the database is updated, the influence of the empirical model gradually weakens until it returns to the standard Bayesian optimization or meets the assembly accuracy requirements of complex precision equipment.

[0026] Furthermore, the method for selecting the assembly adjustment scheme includes:

[0027] For the optimized assembly adjustment plan, a three-coordinate measuring machine, an infrared laser tracker and an industrial photography system are used to achieve precision measurement of complex precision equipment. The measurement data is analyzed and processed to obtain the assembly accuracy. The assembly accuracy obtained by calculation is compared to see if it meets the working condition requirements. If not, the high-precision sample set database is iterated and updated for further optimization. If so, the assembly adjustment plan is the one that ultimately meets the assembly accuracy and performance requirements.

[0028] A precision control system for assembly of precision equipment with a human in the loop, comprising:

[0029] The sample module is used to obtain parameter input and accuracy output based on the actual measurement and adjustment data of precision equipment, and to establish a high-precision sample set database by combining geometric parameters and physical parameters;

[0030] A modeling module, wherein the modeling module is used to obtain empirical data and construct an empirical model and a preference learning model based on the empirical data;

[0031] The optimization module is used to perform Bayesian optimization based on a high-precision sample set database, combine the empirical model with the preference learning model, establish a digital twin model based on human-in-the-loop to optimize assembly parameters, and select an assembly adjustment plan that meets assembly accuracy and performance.

[0032] A terminal device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the method are implemented when the processor executes the computer program.

[0033] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method are implemented.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention provides a method for precision control of assembly of precision equipment with a human in the loop, by establishing a high-precision sample set database, and constructing an empirical model and a preference learning model based on the empirical data according to the empirical data, based on the Bayesian optimization method, with the preference learning model as the connection, the empirical data is connected with the Bayesian optimization process, and a digital twin model is established based on the human in the loop to optimize the assembly parameters, and finally an assembly adjustment scheme that meets the assembly accuracy and performance is selected. The present invention uses a dynamic idea to judge the influence of empirical data on the actual optimization assembly and adjustment, and introduces empirical data into the optimization control of the assembly of complex precision equipment. The judgment of whether the empirical data participates in the optimization is based on the current optimization process, and a mathematical characterization model based on empirical data is constructed, which achieves the best accuracy while improving the control efficiency, and provides an effective method for realizing the rapid and accurate control of complex precision equipment, which greatly improves the assembly accuracy and assembly efficiency of complex precision equipment.

[0036] Furthermore, in the process of optimization through the digital twin model, the present invention proposes the idea that the influence of empirical data on the optimization process has algebraic limitations. In the early stage, the optimal solution is selected mainly based on empirical data. As the optimization proceeds and the database is updated, the empirical influence is gradually weakened until it regresses to the standard Bayesian optimization or meets the assembly accuracy requirements of complex precision equipment, effectively improving the assembly efficiency and assembly accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 This is a flowchart of the method for controlling the assembly accuracy of precision equipment for humans in the loop of the present invention.

[0039] Figure 2 This is a flow chart of the method for controlling the assembly accuracy of precision equipment with humans in the loop in Example 1 of the present invention.

[0040] Figure 3 This is a schematic diagram of the structure of a large space-borne SAR antenna in Example 1 of the present invention.

[0041] Figure 4 This is a schematic diagram of the antenna board structure of a large space-borne SAR antenna in Example 1 of the present invention.

[0042] Figure 5 This is a front view of a large space-borne SAR antenna in Example 1 of the present invention.

[0043] Figure 6It is a schematic diagram of the structure of the precision equipment assembly accuracy control system of the human-in-the-loop in the preferred embodiment of the present invention.

[0044] Figure 7 It is a schematic diagram of the structure of an electronic device according to a preferred embodiment of the present invention.

[0045] Among them, 1-star body, 2-star body connecting rod, 3-outer support rod, 4-middle support rod, 5-inner support rod, 6-outer antenna plate, 7-inner antenna plate, 8-180° hinge, 9-90° hinge. DETAILED DESCRIPTION

[0046] The following is a description of exemplary embodiments of the present application in conjunction with the accompanying drawings, including various details of the embodiments of the present application to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0047] Obviously, the described embodiments are only part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0048] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0049] See also Figure 1 The present invention provides a method for controlling the assembly accuracy of precision equipment with a human in the loop, which specifically comprises the following steps:

[0050] Step 1: Through actual measurement and adjustment of complex precision equipment, obtain parameter input and accuracy output based on the adjustment data results, and establish a high-precision sample set database.

[0051] Conduct test experiments based on complex precision equipment assembly test benches or actual assembly models, collect a number of data results, obtain parameter input and accuracy output, and establish a high-precision sample set database by combining geometric parameters and physical parameters.

[0052] Geometric parameters and physical parameters include the geometric shape, size, tolerance, assembly fit, elastic modulus, stiffness, Poisson's ratio and density of components.

[0053] Step 2: Construct a mathematical representation experience model and preference learning model based on empirical data.

[0054] Introduce the previous experience data into the optimization installation process, mathematically represent the experience data through quantitative methods, and establish an experience model suitable for optimization judgment.

[0055] Based on preference learning, the range conditions of the optimal judgment provided by the empirical model are determined, the likelihood model is used to construct the preference learning function, and the preference learning model is improved through iterative optimization.

[0056] Step 3: Perform Bayesian optimization based on the high-precision sample set database, combine the empirical model with the preference learning model, and establish a digital twin model of complex precision equipment based on humans in the loop.

[0057] On the basis of Bayesian optimization, empirical models and preference learning models are introduced, and a digital twin model of human-in-the-loop is established based on the acquisition function of human-in-the-loop and the standard acquisition function of Gaussian regression process.

[0058] Step 4: Optimize assembly parameters through the digital twin model, obtain multiple optimization and control schemes, and select and formulate schemes by combining the empirical model and preference learning model.

[0059] Determine the assembly parameters that can be optimized and adjusted according to the actual working conditions, build an optimization control model based on the standard acquisition function, solve the assembly parameter adjustment plan, judge whether the empirical model participates in and affects the current optimization, build an optimization control model based on the human-in-the-loop acquisition function to solve the assembly parameter adjustment plan, and select an assembly adjustment plan that is more suitable for the actual working conditions in combination with the empirical model.

[0060] Step 5: Analyze and judge the experimental test results, iteratively update the high-precision sample set database and re-optimize and adjust it so that the accuracy of each module meets the requirements after adjustment, and ultimately ensure that the assembly accuracy of complex precision equipment meets the requirements.

[0061] The assembly adjustment plan selected in the above steps is adopted, and the precision measurement of complex precision equipment is realized by using rapid measurement methods such as three-dimensional coordinate measuring machine, infrared laser tracker and industrial photography system. The measurement data is analyzed and processed to obtain the assembly accuracy, and the accuracy is compared and calculated to see whether it meets the working condition requirements. If not, the high-precision sample set database is entered for iterative update and re-optimization. If it meets the requirements, the assembly adjustment plan is the assembly adjustment plan that ultimately meets the assembly accuracy and performance.

[0062] The iterative update and re-optimization of the high-precision sample set database includes: collecting the geometric information and physical information of the measurement data in real time and inputting them into the digital twin model, updating and optimizing the high-precision sample set database and digital twin model, and re-optimizing and assembling and testing based on the updated digital twin model.

[0063] The present invention is further described in detail below through specific embodiments:

[0064] Embodiment 1:

[0065] like Figure 2As shown, in this embodiment, the method for controlling the assembly precision of complex precision equipment in the human-in-the-loop takes a large-scale spaceborne synthetic aperture radar antenna (Large-Scale Spaceborne Synthetic Aperture Radar (SAR) Antenna) as a specific case, and the specific steps include:

[0066] (1) Through actual measurement and adjustment of a large-scale satellite-borne SAR antenna model, the parameter error groups of each support rod and the corresponding assembly accuracy are obtained, and a high-precision sample set database is established.

[0067] Based on the actual model of large-scale spaceborne SAR antenna, the experiment was carried out. In this embodiment, the rod length error of the large-scale spaceborne SAR antenna support rod assembly was As input, a database is established with the assembly accuracy of large spaceborne SAR antennas as output. The optimization process uses the rod length error of the support rod assembly as the optimization process design variable, and the value range under this condition is set to:

[0068]

[0069] The optimization goal is to optimize the assembly accuracy of large spaceborne SAR antennas, with the flatness error The photogrammetry system can be used to obtain the coordinate information of several points on the actual large-scale satellite-borne SAR antenna surface after deformation, and the theoretical plane can be fitted using the least squares method, with the maximum offset of the actual coordinate point from the theoretical fitting surface. and minimum offset Flatness error :

[0070]

[0071] According to the actual working conditions of large-scale spaceborne SAR antennas, several rod length error groups are obtained and the initial flatness error is measured. , as a high-precision sample set to establish a database.

[0072] (2) Based on the large-scale satellite-borne SAR antenna shape precision adjustment method, a mathematical representation modeling method and preference learning model based on empirical data enhancement are constructed.

[0073] Based on the process engineering experience data summarized from actual working conditions, mathematical quantitative characterization modeling is carried out. The specific steps are as follows:

[0074] (2-1) Acquisition function based on empirical data enhancement

[0075] For the expensive function optimization problem in Bayesian optimization, the prior probability of the optimal position specified by the empirical data is defined as:

[0076]

[0077] in, represents the independent variable, represents the independent variable corresponding to the optimal value, represents the prior probability function, represents the target variable, Indicates the optimal value.

[0078] In the acquisition function based on empirical data enhancement, in order to better explore the trust region of empirical data, the recognized trust region can be assigned a higher value. Without loss of generality, it is required It is positive in the entire decision space, that is, it does not change its sign in the entire decision space, and any point in the decision space may be the optimal point.

[0079] Empirical data can be used to establish a function of the optimal position in the form of any probability distribution ,at the same time, It is convenient to combine with the acquisition function to ensure that the optimal location of the empirical data can be explored more frequently. Function of the optimal location of the empirical data Combined with the acquisition function, it can be expressed as:

[0080]

[0081] in, represents the acquisition function based on experience enhancement, represents the independent variable, represents the data set of the tth iteration, represents the prior probability function.

[0082] The acquisition function incorporates the influence of the empirical data throughout the optimization, and the area with the best position of the empirical data can be explored more frequently. For accurate areas, the convergence speed will be significantly accelerated, but for inaccurate areas, the convergence speed will drop significantly. This is because, for such problems, the influence of the prior needs to decay over time.

[0083] As the optimization proceeds, we should trust the acquisition function of standard Bayesian optimization more and more. To solve the above problems, we can set is the decay exponent, where It reflects the degree of trust in empirical data. is the number of iterations. By setting the decay index, the empirical data acquisition function is made closer and closer to the standard acquisition function. The expression of the designed acquisition function is:

[0084]

[0085] At the beginning of the iteration, the influence of the empirical data will be relatively large. As the iteration proceeds, the exponent of the empirical data will tend to zero, and the role of the empirical data will become smaller and smaller. On the other hand, the judgment of the optimal position by the empirical data is often in a certain area. The specific location of the optimal point requires the optimization algorithm to be continuously iterated to find, which is also consistent with the expression of the acquisition function. In this way, correct empirical data will speed up the convergence speed, on the contrary, incorrect empirical data will slow down the convergence speed to a certain extent.

[0086] (2-2) Knowledge modeling based on preference learning

[0087] The acquisition function constructed in step (2-1) requires empirical data to clearly provide an approximate range of the optimal value. Sometimes the range may be difficult to give. Therefore, this condition can be relaxed and preference learning can be used to obtain the optimal value through χ→ To estimate ,in, represents the estimate of the prior probability function, represents the set of non-negative real numbers, represents the prior probability function. The core of preference learning is to model the sequential relationship between candidate points χ. For two candidate points, the sample set of the preference learning model can be expressed as , among which hour, The value of is 1; when hour, The value of is taken as 0. For any candidate point and , using the likelihood model to construct a binary preference function:

[0088]

[0089] in, is the Bernoulli likelihood, express Relative to The preference degree of To model, in In the iteration, the preference model can be used to calculate The global preference on

[0090]

[0091] right By integrating, we can obtain the following empirical preference estimation form:

[0092]

[0093] in, represents a normal distribution, represents the expectation of the prior probability function estimate, Represents the variance of the prior probability function estimate.

[0094] As the optimization progresses, more binary preference data can be obtained and updated through Bayesian theorem The posterior of , and updates the preference model at each iteration .

[0095] (2-3) Optimization method based on empirical data using large spaceborne SAR antenna as an example

[0096] Large spaceborne SAR antennas generally include deployable support structures, array frames, locking hinges, rotating hinges, and drive motors. Figure 3 As shown, the deployable support structure of such large satellite-borne SAR antenna is a typical multi-loop closed-chain complex rod system structure, wherein the outer support rod 3 is connected to the outer antenna plate 6, the middle support rod 4 and the inner support rod 5 are connected to the inner antenna plate 7, and the top ends of the outer support rod 3, the middle support rod 4 and the inner support rod 5 are all connected to the satellite 1 through the satellite connecting rod 2. The outer antenna plate 6 is connected to the inner antenna plate 7 through a 180° hinge 8, and the inner antenna plate 7 is connected to the satellite 1 through a 90° hinge 9.

[0097] For large spaceborne SAR antennas, the accuracy of their array assembly has a decisive influence on their electrical performance, and with the development of high power, high gain and high frequency, the assembly accuracy requirements are further increased. Therefore, in actual engineering, it is necessary to adjust the rod length or hinge installation height to ensure that the assembly accuracy meets the standard. The assembly accuracy of large spaceborne SAR antennas is mainly evaluated by flatness error, with the rotation error of the large spaceborne SAR antenna plane as an auxiliary reference. Figure 4 and Figure 5 For reference, after the assembly is completed, the engineering department adjusts the antenna board by adding or removing gaskets. The adjustment order is first the inner antenna board surface and then the outer antenna board surface. To this end, the inner antenna board 7 is first quickly tested for its shape and surface. The industrial photography system is used to Side and The front side of the inner antenna plate 7 and the antenna mounting carrier Effectively shoot all target points on the reference surface, and calculate the flatness error of the inner antenna plate 7 through the three-dimensional coordinates of the target points and plane winding Axis rotation error , this rotation error has a theoretical standard and The assembly accuracy of the inner antenna plate 7 is unqualified (i.e. > or > ), you need to adjust the support rod assembly according to the actual measurement results. The specific method is as follows: The inner side antenna plate 7, if Greater than before adjustment , then in Add a spacer to the support rod of the side antenna installation carrier to increase the rod length; if Less than , then reduce the spacers to reduce the rod length. The inner side antenna plate 7, if Greater than before adjustment , then in The side antenna is installed on the carrier support pole to reduce the length of the pole; if Less than , then add a gasket to increase the rod length. Repeated adjustments are performed until the assembly accuracy of the inner antenna plate 7 meets the working condition requirements.

[0098] The test and adjustment of the outer antenna plate 6 is the same as that of the inner antenna plate 7. The target point coordinates are obtained through the industrial photogrammetry system to calculate the flatness error. and rotation error If the assembly accuracy of the outer antenna plate 6 does not meet the requirements (i.e. > or > ), the assembly accuracy is optimized by adjusting the rod length error until the requirements are met. After each adjustment, the outer antenna plate 6 needs to be retracted and extended, and all antenna plates need to be confirmed to be unfolded and locked in place. Then repeat the shape surface quick test, result analysis and judgment, and adjustment steps until all rotation angles are , , , All are within the specified qualified range (i.e. ≤ , ≤ , ≤ , ≤ , ≤ , ≤ ).

[0099] (3) Based on the Bayesian optimization method, a digital twin model of a large-scale satellite-borne SAR antenna is established in combination with human-in-the-loop. The digital twin model is used to optimize the assembly and adjustment of various support rod component error parameter optimization control schemes, and the scheme is selected and formulated based on empirical data.

[0100] The traditional empirical method adopts a local trial-and-error adjustment mode of "measurement-adjustment-remeasurement-readjustment" for the adjustment of large-scale spaceborne SAR antennas, and achieves accuracy optimization through repeated adjustments. However, the large-scale spaceborne SAR antenna is a complex multi-loop closed-chain mechanism with various error factors affecting each other, which is extremely complex and difficult to control. The multiple adjustment and multiple measurement method based on actual engineering is time-consuming and labor-intensive, and the damage and impact on the structure of large-scale spaceborne SAR antennas, such as coated parts, are unpredictable. The standard optimization method cannot take into account both the guarantee of assembly accuracy and the reduction of time cost of large-scale spaceborne SAR antennas. Therefore, empirical data is introduced to establish a Bayesian optimization adjustment method based on human-in-the-loop.

[0101] (3-1) Bayesian optimization method

[0102] With the help of the prior-posterior criterion of Bayesian theorem, Bayesian optimization searches for the optimal solution by continuously updating the posterior distribution when the prior distribution and data are known. This method has the characteristics of self-adaptation and self-learning, and is particularly suitable for situations where the objective function is difficult to calculate or the computational cost is high. For the single-objective optimization problem of optimizing the assembly accuracy of large-scale spaceborne SAR antennas, Bayesian optimization is used in the decision space. Optimize function , whose expression is:

[0103]

[0104] in, represents the independent variable, represents the d-dimensional search space.

[0105] In practical applications, complex systems or models have high computational costs, or have complexity and uncertainty, which makes it difficult to optimize or analyze directly. Proxy models can effectively reduce computational costs and provide approximate prediction capabilities for the original system or model. Through proxy models, optimization searches can be performed in a shorter time without repeatedly executing the original complex model.

[0106] After building the proxy model, it is necessary to continuously obtain sampling points to iteratively improve the fitting accuracy of the proxy model. At this time, the proxy model needs an acquisition function to obtain new sampling points. The characteristic of the acquisition function is that it predicts the possible optimal value by combining the existing sample data and prior knowledge, so as to select the next optimal sampling point. The search direction selected by the acquisition function is the location with a large predicted value or the location with large uncertainty, so that it is possible to search for the optimal solution of the objective function. Sampling at locations with large predicted values ​​is to promote local development, and sampling at locations with large uncertainty is to strengthen global exploration. Therefore, the acquisition function comprehensively considers the predicted value and uncertainty, that is, it is essentially a balanced relationship between exploration and development, and uses known information to find the next sampling point, that is,

[0107]

[0108] in, represents the maximization acquisition function, Represents a dataset.

[0109] The expected improvement acquisition function EI (Excepted Improvement) is mainly used:

[0110]

[0111] in, represents the mean function, represents the currently known optimal value, represents the standard deviation function, represents the cumulative distribution function of the standard normal distribution, represents the probability density function of the standard normal distribution, Represents the independent variable corresponding to the currently known optimal value.

[0112] It mainly consists of two parts. One part considers the predicted mean, that is, points near the optimal value may be selected as the next evaluation point; the other part considers the variance or covariance, that is, points with large uncertainty may be selected as the next evaluation point, thereby integrating the balance between exploration and development, and taking into account the probability and amount of improvement.

[0113] Bayesian optimization aims to optimize the objective function by continuously updating the surrogate model. In practical applications, the objective function is often affected by factors such as noise interference, and the Bayesian optimization method can improve the stability and robustness through the surrogate model, so that the optimization process can converge quickly and stably.

[0114] (3-2) Iterative optimization under the influence of empirical data

[0115] Since empirical data often deviates slightly from the true optimal value, or even misleads, the Bayesian optimization based on human-in-the-loop is mainly divided into two parts: Generation and later Generation. Among them, and The value of is related to the empirical data affecting algebra. In the next generation, the empirical data can be used to evaluate the next evaluation point of the standard Bayesian optimization The fitness of the model can be used to verify or correct the hypothesis, and it can also increase the degree of trust in Bayesian optimization. In addition, given that empirical data is often fuzzy, it cannot effectively assist the proxy model in optimizing within the range close to the optimal value. In the next generation, we choose to let the standard Bayesian optimization complete the final optimization process, but the sample set is based on the sample set of Bayesian optimization in the human loop. .

[0116] In the front of the human-in-the-loop Bayesian optimization algorithm In the iteration, the participation of empirical data is not only in the acquisition function, but also in the decision of whether to choose Specifically, after training the Gaussian process regression proxy model using the sample set, the next evaluation points of Bayesian optimization and human-in-the-loop Bayesian optimization are obtained respectively. and .here This is the point required for standard Bayesian optimization and the next point where fitness needs to be evaluated. At this point, the known information includes the current optimal decision variables, and Based on this information, we can roughly determine the possible range of the optimal value and decide whether to evaluate If you decide to evaluate The fitness of And its fitness is added to the sample set of standard Bayesian optimization , and will and And the fitness of the two is added to the human-in-the-loop Bayesian optimization sample set If it is decided that no assessment is required The fitness of and their fitness are put into the sample set and In. After In the iteration, empirical data is no longer involved, and the final iteration step is completed by standard Bayesian optimization.

[0117] In the above design, the standard Bayesian optimization plays a reference role and enhances the algorithm's global search capability to a certain extent. According to theoretical analysis, the design can significantly accelerate the convergence process under the condition of correct experience, and can also correct wrong experience in time and find the optimal solution, which has a certain stability.

[0118] (4) The experimental test results are analyzed and judged, and the sample set database is iteratively updated and re-optimized to ensure that the accuracy of each module of the large-scale spaceborne SAR antenna meets the requirements after adjustment, ultimately ensuring that the assembly accuracy of the large-scale spaceborne SAR antenna meets the requirements.

[0119] The large-scale satellite-borne SAR antenna measurement and adjustment experiment includes an antenna model, a photogrammetry system, and a control optimization terminal. Real-time measurement control and man-in-the-loop optimization are achieved based on the control optimization terminal. The experiment uses a two-way screw adjustment to simulate the rod length error adjustment process of each support rod. The binocular photogrammetry system is used to obtain the coordinate information of the target point on the large-scale satellite-borne SAR antenna plane through 3D reconstruction and fitting the theoretical plane to calculate the flatness error and achieve rapid measurement of assembly accuracy.

[0120] The optimal support rod assembly and adjustment scheme is obtained through the Bayesian optimization method with humans in the loop. The corresponding assembly accuracy is obtained by experimental measurement to determine whether it meets the actual working conditions. If so, the optimization and adjustment process is terminated. If not, the group of rod length error parameters and assembly accuracy are used as new high-precision samples for iterative optimization until the assembly accuracy is qualified.

[0121] The assembly precision control method of human-in-the-loop precision equipment of this embodiment is based on a case study of a large satellite-borne SAR antenna, with the minimum assembly precision of the large satellite-borne SAR antenna as the objective function, the rod length adjustment size as the optimization variable, and the industrial photography system for measurement. The empirical data is integrated for assembly precision control, so that the whole process takes 3 hours, which is 1 / 3 of the traditional optimization method, 6 hours less than the traditional Bayesian optimization assembly time, and 5 hours less than the pure process engineering experience assembly time. The results show that the control method of the present invention can significantly accelerate the optimization convergence speed and greatly improve the assembly precision and control efficiency.

[0122] The human-in-the-loop precision equipment assembly precision control method of the present invention has high test adjustment accuracy and fast optimization and adjustment speed. It introduces empirical data into the assembly precision control of complex precision equipment for the first time, organically combines the guided judgment of empirical data and the intelligent calculation of the Bayesian optimization algorithm, reduces the optimization time cost on the basis of ensuring the optimization accuracy, can effectively meet the actual production needs, and has an excellent guiding role in the assembly precision control of complex precision equipment, thereby improving the assembly efficiency of complex precision equipment.

[0123] Embodiment 2:

[0124] Embodiment 2 provided by the present invention is an embodiment of the precision control system for assembling precision equipment of a person in the loop provided by the present invention, such as Figure 6 As shown, an embodiment of the system includes: a sample module, a modeling module and an optimization module.

[0125] The sample module is used to obtain parameter input and accuracy output based on the actual measurement and adjustment data of precision equipment, and to establish a high-precision sample set database by combining geometric parameters and physical parameters;

[0126] A modeling module, wherein the modeling module is used to obtain empirical data and construct an empirical model and a preference learning model based on the empirical data;

[0127] The optimization module is used to perform Bayesian optimization based on a high-precision sample set database, combine the empirical model with the preference learning model, establish a digital twin model based on human-in-the-loop to optimize assembly parameters, and select an assembly adjustment plan that meets assembly accuracy and performance.

[0128] It can be understood that the precision control system for precision equipment assembly with humans in the loop provided by the present invention corresponds to the precision control method for precision equipment assembly with humans in the loop provided by the aforementioned embodiments, and the relevant technical features of the precision control system for precision equipment assembly with humans in the loop can refer to the relevant technical features of the precision control method for precision equipment assembly with humans in the loop, which will not be repeated here.

[0129] Another object of the present invention is to provide an electronic device, such as Figure 7 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the steps of the method for precision control of precision equipment assembly of humans in the loop.

[0130] The method for controlling the assembly accuracy of precision equipment with human-in-the-loop comprises the following steps:

[0131] Based on the actual measurement and adjustment data of precision equipment, parameter input and accuracy output are obtained, and a high-precision sample set database is established by combining geometric parameters and physical parameters;

[0132] Obtain experience data and build experience models and preference learning models based on the experience data;

[0133] Bayesian optimization is performed based on the high-precision sample set database. The empirical model and preference learning model are combined to establish a digital twin model based on the human-in-the-loop to optimize assembly parameters and select an assembly adjustment plan that meets assembly accuracy and performance.

[0134] A fourth object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for controlling the assembly accuracy of precision equipment with humans in the loop.

[0135] The method for controlling the assembly accuracy of precision equipment with human-in-the-loop comprises the following steps:

[0136] Based on the actual measurement and adjustment data of precision equipment, parameter input and accuracy output are obtained, and a high-precision sample set database is established by combining geometric parameters and physical parameters;

[0137] Obtain experience data and build experience models and preference learning models based on the experience data;

[0138] Bayesian optimization is performed based on the high-precision sample set database. The empirical model and preference learning model are combined to establish a digital twin model based on the human-in-the-loop to optimize assembly parameters and select an assembly adjustment plan that meets assembly accuracy and performance.

[0139] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program code.

[0140] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for controlling the assembly accuracy of precision equipment with humans in the loop, characterized in that: include: Based on the actual measurement and adjustment data of precision equipment, parameter input and accuracy output are obtained, and a high-precision sample set database is established by combining geometric parameters and physical parameters; Obtain experience data and build experience models and preference learning models based on the experience data; Bayesian optimization is performed based on a high-precision sample set database. The empirical model and preference learning model are combined to establish a digital twin model based on human-in-the-loop to optimize assembly parameters and select an assembly adjustment solution that meets assembly accuracy and performance. The construction method of the experience model and the preference learning model is: Introduce empirical data into the optimization process, mathematically characterize and model the empirical data through quantitative methods, and establish an empirical model suitable for optimization judgment; Based on preference learning, determine the range of optimal judgment provided by the empirical model, use the likelihood model to build a preference learning function, and establish a preference learning model through iterative optimization and improvement; The method for establishing the digital twin model is: On the basis of Bayesian optimization, the empirical model and preference learning model are introduced, and the human-in-the-loop acquisition function and the Gaussian regression process standard acquisition function are used to establish a digital twin model of human-in-the-loop. The process of optimizing assembly parameters through digital twin models includes: Determine the optimized and adjusted assembly parameters according to the actual working conditions; Build an optimization control model based on the standard acquisition function to solve the assembly parameter adjustment plan; Determine whether the empirical model participates in and affects the current optimization, build an optimization control model based on the human-in-the-loop acquisition function, solve the assembly parameter adjustment plan, and select the assembly parameter adjustment plan suitable for the actual working conditions; Whether the empirical model participates in and affects the current optimization is based on the current optimization process: The optimal solution selected by the empirical model is mainly used to speed up the process of searching for the optimal range based on the empirical model; As the optimization proceeds and the database is updated, the influence of the empirical model gradually weakens until it returns to standard Bayesian optimization or meets the assembly accuracy requirements of complex precision equipment.

2. A method for controlling the assembly accuracy of precision equipment with humans in the loop according to claim 1, characterized in that: The geometric parameters and physical parameters include geometric shape, size, tolerance, assembly fit relationship, elastic modulus, stiffness, Poisson's ratio and density.

3. The method for controlling the assembly accuracy of precision equipment with humans in the loop according to claim 1, characterized in that: The selection method of assembly adjustment plan includes: For the optimized assembly adjustment plan, a three-coordinate measuring machine, an infrared laser tracker and an industrial photography system are used to achieve precision measurement of complex precision equipment. The measurement data is analyzed and processed to obtain the assembly accuracy. The assembly accuracy obtained by calculation is compared to see if it meets the working condition requirements. If not, the high-precision sample set database is iterated and updated for further optimization. If so, the assembly adjustment plan is the one that ultimately meets the assembly accuracy and performance requirements.

4. A precision control system for precision equipment assembly with humans in the loop, characterized in that: The steps for implementing the method of claim 1 include: The sample module is used to obtain parameter input and accuracy output based on the actual measurement and adjustment data of precision equipment, and to establish a high-precision sample set database by combining geometric parameters and physical parameters; A modeling module, wherein the modeling module is used to obtain empirical data and construct an empirical model and a preference learning model based on the empirical data; The optimization module is used to perform Bayesian optimization based on a high-precision sample set database, combine the empirical model with the preference learning model, establish a digital twin model based on human-in-the-loop to optimize assembly parameters, and select an assembly adjustment plan that meets assembly accuracy and performance.

5. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images