Sheet riveting assembly deviation prediction method and equipment based on stacked auto-encoders

Abstract

The invention relates to a sheet riveting assembly deviation prediction method and device based on stacked auto-encoders. The method comprises the following steps: obtaining assembly deviation data of key measuring points in each process of a sheet riveting assembly process; constructing a typical variable analysis model, and obtaining assembly deviation characteristics of each process of the thin plate; and fusing the assembly deviation characteristics of each process and inputting the assembly deviation characteristics into the stacked auto-encoder network to obtain deep characteristics of the assembly deviation, and designing a regression model based on the stacked auto-encoder network to carry out assembly deviation prediction. Compared with the prior art, the method has the advantages that the assembly deviation prediction is more accurate, the improvement of the assembly process and the improvement of the product quality are facilitated, and the like.

Description

Technical Field

[0001] The invention relates to the field of thin plate assembly deviation prediction in a manufacturing process, and in particular to a thin plate riveting assembly deviation prediction method and device based on a stacked autoencoder. Background Art

[0002] Large flexible sheets are widely used in the assembly of parts such as ship hulls and aircraft fuselages. These flexible sheets are usually large in size and relatively thin. Due to the characteristics of "large size" and "low stiffness", they are prone to local deformation under the action of gravity. In the manufacturing process, riveting of flexible sheets is a common assembly method, which usually includes multiple processes and integration from multiple components. In the multi-process sheet assembly process, due to the flexibility of the flexible sheet and the cumulative errors between the processes, the final product often has large geometric deviations. In order to improve assembly accuracy and reduce rework and scrap rates, it is particularly important to predict these deviations. To this end, it is usually necessary to combine technical means such as statistical analysis, finite element analysis and machine learning to establish an error prediction model for complex multi-process assembly processes. This prediction can not only optimize the assembly process, but also guide error compensation and quality control in the actual assembly process.

[0003] The traditional product structure assembly deviation model is based on the finite element analysis method, which constructs the elastic mathematical relationship between the force and deformation of the parts by extracting the super-element stiffness matrix. However, this method does not fully consider the influencing factors of assembly deviation and the transmission relationship of assembly deviation, so it is difficult to be fully applied to actual production. In particular, the latest breakthrough technologies and methods in the field of artificial intelligence are needed to realize the learning of the mutual influence relationship between deviations in each stage of the multi-process assembly process of flexible boards. Therefore, it is of great theoretical significance to establish a deviation prediction method for the multi-process riveting assembly process of thin plates based on canonical variable analysis cascade stacked convolutional autoencoders to realize the prediction of multi-process thin plate riveting assembly deviation.

[0004] After searching, Chinese invention patent publication number CN116306246A discloses a method for predicting and optimizing riveting deformation of large wall panels, which includes the following steps: Step 1, using multiple sensors to sense multimodal signals such as pressure, sound, vibration, etc. during the riveting process, and training a prediction model based on a neural network; based on the trained prediction model, signals from multiple sensors are collected during the drilling and riveting process to predict in real time the stress and strain distribution of the wall panels around the rivets after each rivet is riveted; Step 2, based on the stress distribution prediction results of each rivet, a "local-global" mapping method is used to build a prediction model based on the prediction model. The local deformation field generated by a single riveting is sequentially loaded on the independent finite element analysis model of the wall panel to predict the overall deformation of the wall panel; step 3, based on the prediction result of the overall deformation of the wall panel after riveting, the optimization algorithm is used to optimize the riveting sequence of subsequent rivets, so that the overall deformation of the wall panel after riveting is minimized; step 4, rivet according to the riveting sequence obtained by the optimization algorithm, and repeat the above steps of predicting the wall panel deformation according to the sensor and then optimizing the riveting sequence during the riveting process; optimize the riveting process online, and perform the next riveting according to the optimization result; repeat the above process until all riveting is completed. The existing patent uses finite element analysis prediction, but does not fully consider the influencing factors of assembly deviation and the transmission relationship of assembly deviation, which leads to inaccurate prediction.

[0005] How to achieve accurate prediction of thin plate riveting assembly deviation has become a technical problem that needs to be solved. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and device for predicting the deviation of thin plate riveting assembly based on stacked autoencoders.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] According to one aspect of the present invention, a method for predicting assembly deviation of thin plate riveting based on a stacked autoencoder is provided, the method comprising the following steps:

[0009] Step S1, obtaining assembly deviation data of key measuring points in each process of thin plate riveting assembly;

[0010] Step S2, constructing a typical variable analysis model to obtain assembly deviation characteristics of each process of the thin plate;

[0011] In step S3, the assembly deviation features of each process obtained in step S2 are input into the stacked autoencoder network to obtain the deep features of the assembly deviation, and a regression model based on the stacked autoencoder network is designed to predict the assembly deviation.

[0012] Preferably, the step S2 comprises:

[0013] Step S2-1, expanding the assembly deviation data obtained in step S1 to form a complete data set including historical data, current measurement data and future trends;

[0014] Step S2-2, constructing a typical variable analysis model to obtain the typical variable subspace and residual subspace of the assembly process data;

[0015] Step S2-3, by analyzing the canonical variable subspace and the residual subspace, the key features of the deviation in the riveting assembly process are identified.

[0016] More preferably, the step S2-1 is specifically as follows:

[0017] Assume that the assembly deviation data x of the key measuring points of each process is x = {x1, x2, x3, ..., x m}, where the assembly deviation data includes the three-dimensional spatial coordinates of the key measuring points of the thin plate during the actual riveting process.

[0018] The collected sample data is expanded using historical and future data. The expanded historical vectors and future vectors are expressed as:

[0019] x p(n) =[x (n-1) ,x (n -2),...,x (n-l) ]

[0020] x f(n) =[x (n) ,x (n+1) ,...,x (n+h) ]

[0021] where x p(n) and x f(n) They represent the expanded historical vector and future vector respectively, l and h represent the lag orders of the past vector and future vector respectively.

[0022] More preferably, the step S2-2 comprises:

[0023] Get the covariance matrix:

[0024]

[0025] where ∑ pp ,∑ ff and∑ pf denote the covariance matrix of the historical vector, the covariance matrix of the future vector, and the cross-covariance matrix of the historical vector and the future vector, respectively. K = Nh-l+1, where N denotes the number of sampling points, l and h denote the lag order of the past vector and the future vector, respectively.

[0026] Performing singular value decomposition on the above formula, we can get:

[0027] (∑ pp ) -1 / 2 ∑ pf (∑ ff ) -1 / 2 =UΣV

[0028] J=U T (∑ pp ) -1 / 2 ,T=V T (∑ ff ) -1 / 2

[0029] Where U and V represent the projection coefficient matrices corresponding to the typical variable subspace Z and the residual subspace E respectively; therefore, the typical variable subspace Z and the residual subspace E are expressed as:

[0030]

[0031] Where k represents the model order, x (p) and x (f) Represent the historical vector and the future vector respectively.

[0032] More preferably, the canonical variable subspace contains the variation patterns and features in the riveting assembly process, and the residual subspace reflects the unexplained parts of the variation patterns, namely, noise and abnormal data.

[0033] Preferably, the step S3 comprises:

[0034] Step S3-1, constructing a stacked autoencoder network, and further learning the fused assembly deviation features to extract deep features in the data;

[0035] Step S3-2, based on the deep features of step S3-1, a regression model based on stacked autoencoders is constructed, and the prediction of assembly deviation is realized based on the trained regression model.

[0036] More preferably, the stacked autoencoder network includes multiple layers of encoders and decoders for extracting deep features, and the input of the stacked autoencoder network is the fused assembly deviation feature vector of the key measurement points of each process.

[0037] The hidden layer s of the stacked autoencoder network i It is expressed as:

[0038]

[0039] Where d represents the size of the hidden layer, represents the dth hidden layer;

[0040] The encoder transforms the input vector through a nonlinear activation function Mapped into hidden features, expressed as:

[0041]

[0042] where f ae represents the activation function of the encoder, W ae and b ae Represent the weight and bias of the encoder respectively;

[0043] The decoder transforms the hidden layer variables s i Reconstructed to the original input space, expressed as:

[0044]

[0045] in represents the activation function of the decoder, and are the weight and bias of the decoder respectively;

[0046] The reconstruction loss of the stacked autoencoder network is expressed as:

[0047]

[0048] Where I represents the total number of assembly processes, Represents the feature representation after reconstruction.

[0049] More preferably, the stacked autoencoder-based regression model uses unsupervised layer-by-layer training to achieve assembly deviation prediction and calculate the mean square error and R 2 The R 2 The indicator is calculated as follows:

[0050]

[0051] Among them, y i and Represent the true value of the input and the predicted value of the network output, respectively. represents the mean of the true value, and I represents the total number of assembly processes.

[0052] Preferably, in step S1, obtaining assembly deviation data of key measuring points in each process of the flexible thin plate riveting assembly process includes: setting key measuring points on the surface of the thin plate, and using a robotic arm to control a laser rangefinder to collect assembly deviation data of the flexible thin plate in each process.

[0053] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the method described above is implemented when the processor executes the program.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1) The present invention uses a typical variable analysis model to extract the assembly deviation characteristics of each stage of the riveting assembly process, and can learn the characteristic variables that best represent the deviation coupling transmission of the multi-process riveting assembly process; the deviation characteristics of each process are fused to form a comprehensive feature set, and the stacked autoencoder is used to extract the fused multi-process deviation characteristics to obtain a more abstract and high-level feature representation. Finally, a regression model based on the stacked autoencoder is designed to use the extracted comprehensive features to achieve accurate prediction of thin plate riveting assembly deviation.

[0056] 2) The present invention expands the measured deviation data of the riveting assembly process and links the current measurement data with the data points before and after it, thereby forming a complete data set containing historical information and future trends to more comprehensively reflect the dynamic change law of the deviation in the riveting assembly process. Therefore, the prediction based on the complete data set is more accurate.

[0057] 3) The present invention uses stacked autoencoders for regression modeling, which can automatically learn and extract complex features in the multi-process riveting assembly process, adapt to the riveting deviation prediction under different processes, different materials and different assembly conditions, and has a wide range of applications.

[0058] 4) By integrating the assembly deviation characteristics of each stage, the present invention allows engineers to adjust the riveting process parameters according to these characteristics, improve the assembly process, and reduce deviations, thereby achieving global optimization of the entire assembly process, improving the accuracy and reliability of the overall prediction, and thus improving product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a schematic diagram of the process of the assembly deviation prediction method in the present invention;

[0060] Figure 2 It is a structural schematic diagram of the assembly deviation prediction network model in the present invention. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0062] This embodiment relates to a thin plate riveting assembly deviation prediction method based on stacked autoencoders, which is used in the assembly process of ship hulls and aircraft fuselages, multi-process riveting assembly deviation prediction, and riveting process parameters are adjusted according to the deviation prediction to improve the assembly process, guide assembly error compensation and quality control, reduce deviations, and thus improve product quality. The thickness of the thin plate in the application scenario does not exceed 1mm.

[0063] like Figure 1 , the method comprises the following steps:

[0064] Step S1, building a thin plate riveting assembly test bench, conducting a thin plate riveting assembly experiment, and obtaining assembly deviation data of key measuring points in each process of the assembly process;

[0065] Step S2, for each assembly process of the thin plate, a typical variable analysis model is constructed to learn the assembly deviation characteristics of each stage respectively;

[0066] Step S3: Based on the assembly deviation features after fusion of each process, a regression model based on stacked autoencoders is designed to predict assembly deviation, and the mean square error and R are calculated. 2 To evaluate the accuracy of the model's prediction results.

[0067] In step S1, the thin plate riveting assembly test bench mainly includes a base plate, a positioning rod, a positioning fixture and a clamping chuck, etc., wherein the positioning rod is designed to be slidable to realize multi-process assembly. Key measuring points are set on the surface of the thin plate, and a mechanical arm is used to control a laser rangefinder to collect the assembly deviation data of the thin plate in each process.

[0068] In step S2, the measured deviation data of the riveting assembly process is first expanded to obtain historical vectors and future vectors, and the current measurement data is linked to the data points before and after it, thereby forming a complete data set containing historical information and future trends, so as to more comprehensively reflect the dynamic change law of the deviation in the riveting assembly process.

[0069] Deviation measurement data x = {x1, x2, x3, ..., x m}, y=[y1,y2,y3,…,y m ] In order to extract the error coupling characteristics of the thin plate riveting assembly process, the collected sample data is first expanded using historical and future data. The expanded historical vector and future vector can be expressed as:

[0070]

[0071] where x p(n) and x f(n)They represent the expanded historical vector and future vector respectively, l and h represent the lag orders of the past vector and future vector respectively.

[0072] Then, a canonical variate analysis (CVA) model is constructed to obtain the typical variable subspace and residual subspace of the assembly process data. The typical variable subspace contains the main change patterns and characteristics of the riveting assembly process, while the residual subspace reflects the unexplained parts of these change patterns, namely noise and abnormal data.

[0073] According to formula (1), the covariance matrix is ​​obtained:

[0074]

[0075] where ∑ pp ,∑ ff and∑ pf They represent the covariance matrix of the historical vector, the covariance matrix of the future vector, and the cross-covariance matrix of the historical vector and the future vector respectively, K=Nh-l+1, where N represents the number of sampling points, l and h represent the lag orders of the past vector and the future vector respectively.

[0076] Performing singular value decomposition on equation (2), we can obtain:

[0077] (∑ pp ) -12 ∑ pf (∑ ff ) -12 =UΣV (3)

[0078] J=U T (∑ pp ) -1 / 2 ,T=V T (∑ ff ) -1 / 2 (4)

[0079] Where U and V represent the projection coefficient matrices corresponding to the canonical variable subspace Z and the residual subspace E, respectively. The canonical variable subspace Z and the residual subspace E can be expressed as follows using formulas (3) and (4):

[0080]

[0081] Where k represents the model order, x (p) and x (f) Represent the historical vector and the future vector respectively;

[0082] Finally, by analyzing these two subspaces, the key features of the deviation in the riveting assembly process can be identified.

[0083] In step S3, the assembly deviation features of each process obtained in step S2 are integrated, and these assembly deviation features are further learned based on the stacked autoencoder. Through the architecture of the multi-layer encoder and decoder in the stacked autoencoder, the deep features in the data can be extracted and the high-level feature representation can be obtained. Based on the learned deep features, a regression model is constructed to capture the relationship between the assembly deviation features and the final assembly deviation. Finally, based on the trained regression model, the data is predicted, which can effectively realize the prediction of assembly deviation.

[0084] Step S3 includes the following steps:

[0085] Step S3-1, construct a stacked autoencoder (AE: AutoEncoder) network, such as Figure 2 As shown in Figure 1, the stacked autoencoder network consists of multiple layers of encoding and decoding networks, which can extract deep features and express complex input-output relationships in layers. It is the assembly deviation feature of each process after fusion, which can be expressed as:

[0086]

[0087] in, is the assembly deviation feature of the i-th key measuring point in the P-th process.

[0088] The hidden layers of the network i It can be expressed as:

[0089]

[0090] Where d represents the size of the hidden layer, represents the dth hidden layer. The encoder maps the input vector to the hidden features through a nonlinear activation function, which can be expressed as:

[0091]

[0092] where f ae represents the tanh activation function, W ae and b ae denote the weight and bias of the encoder respectively. The decoder transforms the hidden variable s i Reconstructing to the original input space can be expressed as:

[0093]

[0094] in represents the activation function of the decoder, and are the weights and biases of the decoder. The reconstruction loss can be expressed as:

[0095]

[0096] Where I represents the total number of assembly processes, Represents the feature representation after reconstruction.

[0097] Step S3-2, build a regression model based on stacked autoencoders, use the fused deviation features, adopt unsupervised layer-by-layer training to realize assembly deviation prediction, and calculate the mean square error to verify the prediction effect of the model, which can be expressed as:

[0098]

[0099] where y i and Represent the true value of the input and the predicted value of the network output respectively. The smaller the value of MSE, the better the prediction effect of the model. In addition, calculate R 2 Indicators are used to evaluate the prediction effect of the model, R 2 The calculation method is as follows:

[0100]

[0101] in Represents the mean of the true values. R 2 It can be used to evaluate the reliability of the model, R 2 The closer the value is to 1, the higher the reliability of the model.

[0102] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0103] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0104] The processing unit performs the various methods and processes described above, such as methods S1 to S3. For example, in some embodiments, methods S1 to S3 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S3 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S3 in any other appropriate manner (e.g., by means of firmware).

[0105] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0106] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0107] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for predicting assembly deviation of thin plate riveting based on stacked autoencoders, characterized in that: The method comprises the following steps: Step S1, obtaining assembly deviation data of key measuring points in each process of thin plate riveting assembly; Step S2, constructing a typical variable analysis model to obtain assembly deviation characteristics of each process of the thin plate; In step S3, the assembly deviation features of each process obtained in step S2 are input into the stacked autoencoder network to obtain the deep features of the assembly deviation, and a regression model based on the stacked autoencoder network is designed to predict the assembly deviation.

2. The method for predicting assembly deviation of thin plate riveting based on stacked autoencoders according to claim 1, characterized in that: The step S2 comprises: Step S2-1, expanding the assembly deviation data obtained in step S1 to form a complete data set including historical data, current measurement data and future trends; Step S2-2, constructing a typical variable analysis model to obtain the typical variable subspace and residual subspace of the assembly process data; In step S2-3, the key features of the deviation in the riveting assembly process are identified by analyzing the canonical variable subspace and the residual subspace.

3. The method for predicting assembly deviation of thin plate riveting based on stacked autoencoders according to claim 2, characterized in that: The step S2-1 is specifically as follows: Assume that the assembly deviation data x of the key measuring points of each process is x = {x1, x2, x3, ..., x m }; The collected sample data is expanded using historical and future data. The expanded historical vectors and future vectors are expressed as: x p(n) =[x (n-1) ,x (n-2) ,...,x (n-l) ] x f(n) =[x (n) ,x (n+1) ,...,x (n+h) ] where x p(n) and x f(n) They represent the expanded historical vector and future vector respectively, l and h represent the lag orders of the past vector and future vector respectively.

4. The method for predicting assembly deviation of thin plate riveting based on stacked autoencoders according to claim 3 is characterized in that: The step S2-2 comprises: Get the covariance matrix: where ∑ pp ,∑ ff and∑ pf denote the covariance matrix of the historical vector, the covariance matrix of the future vector, and the cross-covariance matrix of the historical vector and the future vector, respectively. K = Nh-l+1, where N denotes the number of sampling points, l and h denote the lag order of the past vector and the future vector, respectively. Performing singular value decomposition on the above formula, we can get: (∑ pp ) -12 ∑ pf (∑ ff ) -12 =UΣV J=U T (∑ pp ) -1 / 2 ,T=V T (∑ ff ) -1 / 2 Where U and V represent the projection coefficient matrices corresponding to the typical variable subspace Z and the residual subspace E respectively; therefore, the typical variable subspace Z and the residual subspace E are expressed as: Where k represents the model order, x (p) and x (f) Represent the historical vector and the future vector respectively.

5. The method for predicting assembly deviation of thin plate riveting based on stacked autoencoders according to claim 2, characterized in that: The typical variable subspace contains the variation patterns and characteristics in the riveting assembly process, and the residual subspace reflects the unexplained parts of the variation patterns, namely, noise and abnormal data.

6. The method for predicting assembly deviation of thin plate riveting based on stacked autoencoders according to claim 1, characterized in that: The step S3 comprises: Step S3-1, constructing a stacked autoencoder network, and further learning the fused assembly deviation features to extract deep features in the data; Step S3-2, based on the deep features of step S3-1, a regression model based on stacked autoencoders is constructed, and the prediction of assembly deviation is realized based on the trained regression model.

7. The method for predicting assembly deviation of thin plate riveting based on stacked autoencoders according to claim 6, characterized in that: The stacked autoencoder network includes multiple layers of encoders and decoders for extracting deep features. The input of the stacked autoencoder network is the assembly deviation feature vector of the key measurement points of each process after fusion. The hidden layer s of the stacked autoencoder network i It is expressed as: Where d represents the size of the hidden layer, represents the dth hidden layer; The encoder transforms the input vector through a nonlinear activation function Mapped into hidden features, expressed as: where f ae represents the activation function of the encoder, W ae and b ae Represent the weight and bias of the encoder respectively; The decoder transforms the hidden layer variables s i Reconstructed to the original input space, expressed as: in represents the activation function of the decoder, and are the weight and bias of the decoder respectively; The reconstruction loss of the stacked autoencoder network is expressed as: Where I represents the total number of assembly processes, Represents the feature representation after reconstruction.

8. The method for predicting assembly deviation of thin plate riveting based on stacked autoencoders according to claim 6, characterized in that: The stacked autoencoder-based regression model uses unsupervised layer-by-layer training to predict assembly deviations and calculates the mean square error and R 2 Indicators are used to evaluate the accuracy of the model's prediction results. The R 2 The indicator is calculated as follows: Among them, y i and Represent the true value of the input and the predicted value of the network output, respectively. represents the mean of the true value, and I represents the total number of assembly processes.

9. The method for predicting assembly deviation of thin plate riveting based on stacked autoencoders according to claim 1, characterized in that: In the step S1, obtaining assembly deviation data of key measuring points in each process of the flexible thin plate riveting assembly process includes: setting key measuring points on the surface of the thin plate, and using a mechanical arm to control a laser rangefinder to collect assembly deviation data of the flexible thin plate in each process.

10. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 9 is implemented.

Images