Method and device for predicting bottom surface size error of large thin-walled part cavity

Abstract

The present application relates to the technical field of intelligent manufacturing, and particularly relates to a bottom surface size error prediction method and device for large thin-walled part cavities, wherein the method comprises the following steps: collecting spindle milling power signals of each cavity in a large thin-walled part, and clustering each cavity to obtain at least one type of cavity; using a pre-constructed deep auto-encoder neural network to extract initial power signal features of any type of cavity, and measuring the bottom surface size error of the current type of cavity to construct a correlation model; adjusting the parameters of the deep auto-encoder neural network, and inputting the spindle milling power signals of the remaining other types of cavities into the fine-tuned deep auto-encoder network to extract multiple power signal features; inputting the multiple power signal features into the correlation model to predict the bottom surface size error of the multiple cavity milling. The method realizes efficient and accurate prediction of the bottom surface size error of the cavity milling of the large thin-walled part.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing, and in particular relates to a bottom surface size error prediction method and device for cavity milling of a large thin-walled part. BACKGROUND

[0002] A large thin-walled part is a typical product in the modern aerospace industry, and in order to reduce launch costs, the overall weight is usually reduced through cavity milling processing. The key processing size of cavity milling is the bottom surface size. Due to the weak rigidity of the large thin-walled part and the large geometric size, the bottom surface size processing error usually exceeds the size tolerance. At the same time, due to the large number of processing cavities on the large thin-walled part, batch detection and repair of the size precision pose great challenges.

[0003] In related technologies, the milling deformation of the cavity of the thin-walled part can be calculated based on the coupling relationship between the system deformation and the cutting force and according to the established physical model; or the milling process of the cavity of the thin-walled part is simulated by considering factors such as cutting force, material removal, tool path, fixture and tool stiffness.

[0004] However, in related technologies, due to factors such as large size of the large thin-walled part, uneven clamping stress distribution, existence of machine tool error and internal stress, and complex actual milling processing environment, it is difficult for the physical model to accurately describe the milling process of all cavities of the large thin-walled part. As for the numerical simulation technology, the accuracy of the calculation result needs to be based on accurate prerequisite settings as a prerequisite, and the calculation efficiency is often subject to the computer computing power. Therefore, whether in terms of calculation accuracy or calculation efficiency, it is still difficult to guarantee the processing quality of the cavities of the large thin-walled part by relying on the establishment of a physical model or numerical simulation method. In order to guide the size precision repair work of the cavities of the large thin-walled part, it is necessary to implement a fast and accurate cavity bottom surface size error prediction method. SUMMARY

[0005] The present application provides a bottom surface size error prediction method and device for cavities of a large thin-walled part, to solve the problems in related technologies that due to factors such as large size of the large thin-walled part, uneven clamping stress distribution, existence of machine tool error and internal stress, and complex actual milling processing environment, it is difficult for the physical model to accurately describe the milling process of all cavities of the large thin-walled part, and the numerical simulation technology is difficult to accurately predict the milling processing error to guarantee the processing quality of the cavities of the large thin-walled part in terms of both calculation accuracy and calculation efficiency.

[0006] A first aspect of the present invention provides a method for predicting the bottom surface dimension error of cavities in large thin-walled parts, comprising the following steps: acquiring spindle milling power signals for each cavity in the large thin-walled part; clustering each cavity based on the spindle milling power signals of each cavity to obtain at least one type of cavity; extracting initial power signal features of any type of cavity using a pre-constructed deep autoencoder neural network, and measuring the bottom surface dimension error of the current type of cavity; constructing a correlation model between the initial power signal features and their corresponding actual bottom surface dimension errors; adjusting the parameters of the deep autoencoder neural network, and inputting the spindle milling power signals of the remaining other types of cavities into the fine-tuned deep autoencoder network to extract multiple power signal features; and inputting the multiple power signal features into the correlation model to predict the bottom surface dimension errors of the milling of multiple cavities.

[0007] Optionally, the spindle milling power signal for each cavity is:

[0008]

[0009] in, For the first The spindle milling power signal for each cavity. For the first The first cavity The measuring point is the midpoint, and the tool rotates back and forth around this midpoint. Spindle milling power of rotation , The number of measuring points arranged for each cavity.

[0010] Optionally, the step of clustering each cavity based on the spindle milling power signal of each cavity to obtain at least one type of cavity includes:

[0011] Step 1: Calculate the divergence matrix of the N cavities and set a temporary matrix ;

[0012] Step 2: Determine the temporary matrix The non-zero minimum value;

[0013] When the non-zero minimum value is not unique, any one of them can be selected, and let the selected non-zero minimum value be in the temporary matrix. The line, number Column, will the first The and the first The cavities are grouped into one category and set up ;

[0014] Step 3: Redetermine the temporary matrix The non-zero minimum value;

[0015] When the non-zero minimum value is not unique, any one of them can be selected, and let the selected non-zero minimum value be in the temporary matrix. The line, number List;

[0016] Determine the first The and the first Have the cavities been clustered?

[0017] If neither cavity is clustered, then the first... The and the first The cavities are grouped into one category and set up ;

[0018] If only one cavity has been clustered, let the cavity that has been clustered be the first one. The cavity that is not clustered is the th cavity. The cavity is defined to contain the first cavity. The type of each cavity is Define intermediate variables , ,exist In this case, the two cavities are allocated to ,exist In the case of only the first Each cavity is allocated to and set ;

[0019] With both cavities already clustered, and set ;

[0020] Check if all cavities have been clustered. If not, iterate through step three to continue clustering until all cavities have been clustered.

[0021] Optionally, the divergence matrix of the N cavities can be calculated using the Kullback-Leibler divergence calculation method. The Kullback-Leibler divergence is calculated as follows:

[0022]

[0023] in, For the first The spindle milling power signal of each cavity Compared to the first The spindle milling power signal of each cavity Kullback-Leibler divergence, is the midpoint of the first cavity, is the midpoint of the first cavity, is the spindle milling power of the tool rotating Zrot around the midpoint, is the midpoint of the first cavity, is the midpoint of the first cavity, is the spindle milling power of the tool rotating Zrot around the midpoint, wherein, , is the number of measuring points arranged for each cavity.

[0024] Optionally, in the case that only one cavity has been clustered, the intermediate variable defined and are respectively:

[0025]

[0026]

[0027] wherein, is the class containing the cavity that has been clustered, is the cavity that has not been clustered, is the number of cavities in , and is any one cavity in , and is any one cavity in all cavities that do not belong to the class , and is the spindle milling power signal of any one cavity in , and is the Kullback-Leibler divergence of the spindle milling power signal of the cavity that has not been clustered, is the Kullback-Leibler divergence of the spindle milling power signal of any one cavity in all cavities that do not belong to the class , and is the Kullback-Leibler divergence of the spindle milling power signal of the cavity that has not been clustered.

[0028] Optionally, the use of the pre-constructed deep autoencoder neural network to extract the initial power signal features of any type of cavity includes:

[0029] Based on the pre-constructed deep autoencoder neural network, one of the clustered cavity classes is selected, which contains the first cavity, and then the first A spindle milling power signal of each cavity The frequency domain signal input into the deep auto-encoder neural network, and the initial power signal feature of the first cavity is obtained :

[0030]

[0031] wherein, F i is a matrix, is the number of measuring points arranged for each cavity, is the dimension of the power signal feature, is the first order component of the power signal feature of the first cavity and the first measuring point.

[0032] Optionally, the correlation model is:

[0033]

[0034] wherein, is the first order component of the power signal feature at the first measuring point of the first cavity, is the floor size error at the first measuring point of the first cavity.

[0035] The second aspect embodiment of the present application provides a floor size error prediction device for cavities of large thin-walled parts, comprising: a collection module for collecting a spindle milling power signal of each cavity in a large thin-walled part; a cavity clustering module for clustering each cavity based on the spindle milling power signal of each cavity to obtain at least one type of cavity; a feature extraction module for extracting an initial power signal feature of any type of cavity using a pre-constructed deep auto-encoder neural network, and measuring the floor size error of the current type of cavity; a correlation module construction module for constructing a correlation model between the initial power signal feature and its corresponding actual floor size error; a parameter fine-tuning module for adjusting the parameters of the deep auto-encoder neural network, and inputting the spindle milling power signals of the remaining other types of cavities into the fine-tuned deep auto-encoder network to extract a plurality of power signal features; and an error prediction module for inputting the plurality of power signal features into the correlation model to predict the floor size error of the milling of a plurality of cavities.

[0036] ​​The third aspect of the present application provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for predicting the bottom surface size error of a large thin-walled part cavity as described in the above embodiments.

[0037] The fourth aspect of the present application provides a computer readable storage medium storing a computer program, which is executed by a processor to implement the method for predicting the bottom surface size error of a large thin-walled part cavity as described above.

[0038] The method and device for predicting the bottom surface size error of a large thin-walled part cavity according to the embodiments of the present application can collect spindle milling power signals of the milling of the large thin-walled part cavity, input the spindle milling power signals of each cavity into a clustering method based on Kullback-Leibler divergence, divide the cavities into different types, build a deep auto-encoder neural network, input the spindle milling power of a type of cavity into the deep auto-encoder network, extract the power signal features, test the bottom surface size error of this type of cavity, build a correlation model between the power signal features and the corresponding bottom surface size error of the cavity based on support vector regression, realize the prediction of the bottom surface size error based on the power signal features, fine-tune the parameters of the deep auto-encoder neural network based on the transfer learning technology, input the spindle milling power of other types of cavities into the deep auto-encoder network with fine-tuned parameters, extract the power signal features, input the power signal features of other types of cavities into the established correlation model, and accurately predict the bottom surface size error of the cavity milling.

[0039] Additional aspects and advantages of the present application will be made apparent by the following description and the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0040] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description, taken in conjunction with the accompanying drawings, in which:

[0041] Figure 1 A flowchart of a method for predicting the bottom surface size error of a large thin-walled part cavity according to an embodiment of the present application is shown in the figure;

[0042] Figure 2 A specific process diagram of a method for predicting the bottom surface size error of the milling of a large thin-walled part cavity according to a specific embodiment of the present application is shown in the figure;

[0043] Figure 3 A diagram of the spindle power signal measurement points of the milling of a large thin-walled part cavity according to an embodiment of the present application is shown in the figure;

[0044] Figure 4 A schematic diagram of a principle of a bottom surface size error prediction model for cavity milling of a large thin-walled part according to an embodiment of the present application;

[0045] Figure 5 A block schematic diagram of a bottom surface size error prediction device for cavity of a large thin-walled part according to an embodiment of the present application;

[0046] Figure 6 A structural schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0047] Embodiments of the present application are described in detail below with reference to examples illustrated in the accompanying drawings, in which the same or similar components are denoted by the same or similar reference numerals throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are intended to explain the present application, and should not be understood as limiting the present application.

[0048] A bottom surface size error prediction method and device for cavity of a large thin-walled part according to an embodiment of the present application are described below with reference to the accompanying drawings.

[0049] Figure 1 A flowchart of a bottom surface size error prediction method for cavity of a large thin-walled part according to an embodiment of the present application is provided.

[0050] As shown in Figure 1 , the bottom surface size error prediction method for cavity of a large thin-walled part includes the following steps:

[0051] In step S101, spindle milling power signals of each cavity in a large thin-walled part are collected.

[0052] In some embodiments, the spindle milling power signal of each cavity can be:

[0053]

[0054] wherein, the spindle milling power signal of the i-th cavity, the spindle milling power signal of the i-th cavity, the spindle milling power signal of the i-th cavity, the spindle milling power signal of the i-th cavity, the spindle milling power signal of the i-th cavity, , the number of measuring points arranged for each cavity.

[0055] In actual execution, taking large thin-walled part milling as an example, the NI signal acquisition system can realize the acquisition of the milling power signal, can control the system sampling frequency, and can connect the current sensor and the voltage sensor to the control circuit of the main shaft motor respectively, so as to calculate the milling power of the main shaft through the current signal and the voltage signal.

[0056] In step S102, each cavity is clustered based on the main shaft milling power signal of each cavity, and at least one type of cavity is obtained.

[0057] As a possible implementation manner, the embodiment of the application can effectively meet the batch prediction of the machining size error of the bottom surface of multiple cavities by dividing the cavities into different types, considering establishing a size error prediction model corresponding to different cavity types, and improving the intelligent level.

[0058] Further, in an embodiment of the application, each cavity is clustered based on the main shaft milling power signal of each cavity, and at least one type of cavity is obtained, including:

[0059] Step one, according to the calculation method of Kullback-Leibler divergence, the divergence matrix of N cavities is calculated , and a temporary matrix is set , wherein the calculation method of Kullback-Leibler divergence is:

[0060]

[0061] In the formula, is the Kullback-Leibler divergence of the main shaft milling power signal of the i-th cavity relative to the main shaft milling power signal of the j-th cavity , is the main shaft milling power of the i-th cavity at the i-th measuring point of the i-th cavity as the midpoint, and the tool rotates Z rotations before and after the midpoint, is the main shaft milling power of the i-th cavity at the i-th measuring point of the i-th cavity as the midpoint, and the tool rotates Z rotations before and after the midpoint, wherein , , , is the number of measuring points arranged for each cavity;

[0062] , wherein the expression of the Kullback-Leibler divergence matrix D is:

[0063] ; ​​​​

[0064] Step two, determine the non-zero minimum value of the temporary matrix ;

[0065] In the case that the non-zero minimum value is not unique, optionally one of them, suppose the selected non-zero minimum value is in the th row, the th column of the temporary matrix , the th and the th cavities are clustered together, and set ;

[0066] Step three, re-determine the non-zero minimum value of the temporary matrix ;

[0067] In the case that the non-zero minimum value is not unique, optionally one of them, suppose the selected non-zero minimum value is in the th row, the th column of the temporary matrix ;

[0068] Judge whether the th and the th cavities have been clustered:

[0069] In the case that neither of the two cavities has been clustered, the th and the th cavities are clustered together, and set ;

[0070] In the case that only one cavity has been clustered, suppose the clustered cavity is the th cavity, the un-clustered cavity is the th cavity, define the class containing the th cavity as , define the intermediate variable , , in the case that , assign both cavities to , in the case that , only assign the th cavity to , and set ;

[0071] In the case that both cavities have been clustered, and set ;

[0072] Check whether all cavities have been clustered, in the case that not all cavities have been clustered, iterate step three to continue clustering until all cavities have been clustered.

[0073] Optionally, in the case that only one cavity has been clustered, the defined intermediate variable and are respectively:

[0074]

[0075]

[0076] wherein, is a class containing cavities that have been clustered, is a cavity that has not been clustered, is a cavity that has not been clustered, is the number of cavities in is any one cavity that does not belong to the class

[0077] In step S103, the initial power signal features of any type of cavity are extracted by using the pre-constructed deep auto-encoder neural network, and the bottom surface size error of the current type of cavity is measured.

[0078] In actual implementation, the embodiment of the present application can extract the power signal features of various types of cavities based on the deep auto-encoder neural network, and measure the milling depth of the processing point by the coordinate probe on the spindle of the machine tool, calculate the residual wall thickness according to the workpiece thickness and the actual milling depth, and obtain the machining error of the bottom surface size.

[0079] Specifically, the embodiment of the present application selects one type of cavity after clustering based on the pre-constructed deep auto-encoder neural network, assumes that it contains the first cavity, and inputs the frequency domain signal of the spindle milling power signal of the first cavity into the deep auto-encoder neural network to obtain the initial power signal features of the first cavity.​​​​​​​​​​​​​​

[0080]

[0081] wherein, F i is a matrix, is the number of measuring points arranged for each cavity, is the dimension of the power signal feature, is the k-th component of the power signal feature of the i-th measuring point of the j-th cavity. is the i-th component of the power signal feature of the j-th cavity. is the k-th component of the power signal feature of the i-th measuring point of the j-th cavity.

[0082] In step S104, a correlation model between the initial power signal feature and the actual bottom surface size error corresponding thereto is constructed.

[0083] In actual implementation, according to the characteristics of the cavity power signal feature, the embodiment of the application can fit the correlation between various cavity power signal features and the bottom surface size machining error thereof by using a support vector regression (SVR) algorithm.

[0084] wherein the correlation model between the various cavity power signal features and the bottom surface size machining error thereof is:

[0085]

[0086] wherein, is the k-th component of the power signal feature of the i-th measuring point of the j-th cavity. is the i-th component of the power signal feature of the j-th cavity. is the i-th component of the power signal feature of the j-th cavity. is the bottom surface size error of the i-th measuring point of the j-th cavity.

[0087] In step S105, the parameters of the deep autoencoder neural network are adjusted, and the spindle milling power signals of the remaining cavities of other types are input into the fine-tuned deep autoencoder network to extract multiple power signal features.

[0088] In actual implementation, the embodiment of the application can fine-tune the parameters of the deep autoencoder neural network based on the transfer learning technology for each type of cavity other than the one for which the correlation model has been constructed, and extract multiple power signal features from the fine-tuned deep autoencoder network.

[0089] In step S106, the multiple power signal features are input into the correlation model to predict the bottom surface size errors of the multiple cavities.

[0090] ​​​​​In actual execution, the embodiment of the application can input the power signal features of all types of cavities into the corresponding associated model to predict the bottom surface size error of all cavities.

[0091] The working principle of the bottom surface size error prediction method for large thin-walled part cavities in the embodiment of the application will be described in detail below by taking cavity milling of a large thin-walled part as an example.

[0092] As shown in Figure 2 , the embodiment specifically includes the following steps:

[0093] In step S201, the spindle milling power signal of the cavity milling process of the large thin-walled part is acquired.

[0094] For the large thin-walled part milling in the embodiment of the application, the milling power signal is measured by using a NI signal acquisition system, and the spindle power signal measuring point is as shown in Figure 3 , the current sensor and the voltage sensor are connected to the control circuit of the spindle motor, the current sensor is a Hall current sensor, and the measurement accuracy is 0.01 A; the voltage signal is acquired by a voltage acquisition card PXIe-4310, and the measurement accuracy is 0.1 V; the milling power of the spindle is calculated through the current signal and the voltage signal, and the sampling frequency is 10000 Hz.

[0095] In step S202, the cavities are clustered into different types based on a clustering algorithm.

[0096] Specifically, in the embodiment, a clustering method based on Kullback-Leibler divergence is used to cluster a total of N cavities.

[0097] The principle of using the clustering method to classify cavities in the embodiment of the application can be as shown in Figure 4 .

[0098] According to the calculation method of Kullback-Leibler divergence, the Kullback-Leibler divergence matrix of the N cavities can be calculated , and a temporary matrix is set. ;

[0099] The calculation expression of Kullback-Leibler divergence is:

[0100]

[0101] In the formula, the th cavity is the spindle milling power signal of the th cavity relative to the th cavity.Main spindle milling power signal of a cavity Kullback-Leibler divergence, Kullback-Leibler divergence, Main spindle milling power of a cavity, Main spindle milling power of a cavity, Main spindle milling power of a cavity, Main spindle milling power of a cavity, Main spindle milling power of a cavity, , Number of measuring points arranged for each cavity;

[0102] The expression of the Kullback-Leibler divergence matrix is as follows:

[0103]

[0104] Further, the embodiment of the present application can find the non-zero minimum value of the Kullback-Leibler divergence matrix, if the non-zero minimum value is not unique, then optionally one of them, assuming the selected non-zero minimum value is in the i-th row, the j-th column of the Kullback-Leibler divergence matrix, the i-th and the j-th cavity are clustered together, and the cluster number is set to k; ;

[0105] Further, the embodiment of the present application can re-find the non-zero minimum value of the Kullback-Leibler divergence matrix, if the non-zero minimum value is not unique, then optionally one of them, assuming the selected non-zero minimum value is in the i-th row, the j-th column of the Kullback-Leibler divergence matrix, determine whether the i-th and the j-th cavity have been clustered:

[0106] If neither of the two cavities has been clustered, then the i-th and the j-th cavity are clustered together, and the cluster number is set to k; ;

[0107] If only one cavity has been clustered, then it can be assumed that the clustered cavity is the i-th cavity, and the un-clustered cavity is the j-th cavity, define the class containing the i-th cavity as k, and define the intermediate variable ​​​​​​​​​​​​​​​​​​​​Distribute these two cavities to ;if Only the first Each cavity is allocated to and set Among them, intermediate variables , The expressions are as follows:

[0108]

[0109]

[0110] In the formula, For cavities that have been clustered The class, For cavities that are not clustered, for Number of cavities for Any cavity in the middle, For all those who do not belong Any cavity within the cavity of the class, for Spindle milling power signal for any cavity Compared to unclustered cavities Spindle milling power signal Kullback-Leibler divergence, For all those who do not belong Spindle milling power signal of any cavity Compared to unclustered cavities Spindle milling power signal The Kullback-Leibler divergence.

[0111] If both cavities have been clustered and set ;

[0112] Check if all cavities have been clustered. If not, redetermine the temporary matrix. The non-zero minimum value is used to continue clustering until all cavities have been clustered.

[0113] Step S203: A deep autoencoder neural network is used to extract power signal features from the spindle milling power signal and measure the bottom surface dimension error of the current type of cavity. Then, a correlation model between the power signal features and the bottom surface machining error is established based on the support vector regression (SVM) algorithm.

[0114] like Figure 4As shown, embodiments of the present invention can construct a deep autoencoder neural network, select one type of cavity after clustering, and assume that it contains the first... Each cavity will The frequency domain signal is input into a deep autoencoder neural network to obtain the first... Power signal characteristics of each cavity ;

[0115]

[0116] in, for The matrix, The number of measuring points arranged in each cavity. For the dimensions of power signal characteristics, For the first The first cavity Characteristics of power signal at each measurement point Component order.

[0117] Furthermore, the bottom surface dimensions are measured using three coordinate probes on the machine tool spindle, and then determined by the machine tool. The coordinate probe on the axis measures the milling depth of the machining point, calculates the remaining wall thickness based on the workpiece thickness and the actual milling depth, and obtains the machining error of the bottom surface dimension in the current type of cavity.

[0118] Furthermore, based on the characteristics of the cavity power signal, the correlation between the power signal characteristics of various cavities and the machining error of the bottom surface dimension in the current type of cavity can be fitted using the Support Vector Regression (SVR) algorithm, which can be expressed as:

[0119]

[0120] in, For the first The first cavity Power signal characteristics at each measuring point First-order components, Indicates the first The first cavity The bottom surface dimension error at each measuring point.

[0121] Step S204: Based on the transfer learning method, obtain the parameters of the deep autoencoder neural network corresponding to each cavity type. After extracting the power signal features of each cavity, input them into the support vector regression model to realize the prediction of bottom surface size error.

[0122] like Figure 4As shown, the parameters of the deep auto-encoder neural network can be fine-tuned based on the transfer learning technology, the deep auto-encoder network fine-tuned by the spindle milling power input parameters of other types of cavities can extract multiple power signal features, the power signal features of each type of cavity can be input into the established correlation model, and the accurate prediction of the bottom surface size error of the cavity milling can be realized.

[0123] In summary, according to the bottom surface size error prediction method for cavities of large thin-walled parts proposed in the embodiment of the present application, the spindle milling power signals of the cavities of large thin-walled parts can be collected, the spindle milling power signals of each cavity are input into the clustering method based on Kullback-Leibler divergence, the cavities are divided into different types, the deep auto-encoder neural network is built, the spindle milling power of a type of cavity is input into the deep auto-encoder network, the power signal features are extracted, and the bottom surface size error of this type of cavity is tested, the correlation model between the power signal features and the corresponding bottom surface size error of the cavity is built based on support vector regression, the bottom surface size error is predicted based on the power signal features, the parameters of the deep auto-encoder neural network are fine-tuned based on the transfer learning technology, the deep auto-encoder network fine-tuned by the spindle milling power input parameters of other types of cavities extracts the power signal features, the power signal features of other types of cavities are input into the established correlation model, and the bottom surface size error of the cavity milling is accurately predicted. Thus, the problems in the related art that the physical model is difficult to accurately describe the milling process of all cavities of large thin-walled parts due to the large size of large thin-walled parts, uneven clamping stress distribution, existence of machine tool error and internal stress, and complex actual milling environment, and that the accuracy of the calculation result of the numerical simulation technology needs to be based on accurate prerequisite settings as a prerequisite, and the calculation efficiency is often easily restricted by computer computing power are solved.

[0124] Secondly, the bottom surface size error prediction device for cavities of large thin-walled parts proposed in the embodiment of the present application is described with reference to the accompanying drawings.

[0125] Figure 5 It is a block schematic diagram of the bottom surface size error prediction device for cavities of large thin-walled parts in the embodiment of the present application.

[0126] As shown in the figure, Figure 5 The bottom surface size error prediction device 50 for cavities of large thin-walled parts includes an acquisition module 501, a cavity clustering module 502, a feature extraction module 503, a correlation module construction module 504, a parameter fine-tuning module 505, and an error prediction module 506.

[0127] The main shaft milling power signal of each cavity in the large thin-walled part is collected by the collection module 501. The cavities are clustered based on the main shaft milling power signal of each cavity by the cavity clustering module 502, and at least one type of cavity is obtained. The initial power signal features of any type of cavity are extracted by using a pre-constructed deep auto-encoder neural network, and the bottom surface size error of the current type of cavity is measured by the feature extraction module 503. The association module construction module 504 is used to construct an association model between the initial power signal features and the actual bottom surface size error corresponding thereto. The parameters of the deep auto-encoder neural network are adjusted by the parameter fine-tuning module 505, and the main shaft milling power signals of the remaining other types of cavities are input into the fine-tuned deep auto-encoder network to extract multiple power signal features. The error prediction module 506 is used to input the multiple power signal features into the association model to predict the bottom surface size error of the multiple cavity milling.

[0128] Optionally, the main shaft milling power signal of each cavity is:

[0129]

[0130] wherein, is the main shaft milling power signal of the i-th cavity, is the main shaft milling power of the i-th cavity at the midpoint of the i-th measuring point of the i-th cavity, , is the number of measuring points arranged for each cavity.

[0131] Optionally, clustering each cavity based on the main shaft milling power signal of each cavity to obtain at least one type of cavity comprises:

[0132] Step one: calculating the divergence matrix of the N cavities , and setting a temporary matrix ;

[0133] Step two: determining the non-zero minimum value of the temporary matrix ;

[0134] In the case where the non-zero minimum value is not unique, optionally one of them is selected, and it is assumed that the selected non-zero minimum value is in the i-th row and the j-th column of the temporary matrix The i-th and the j-th cavities are clustered into one class, and ;

[0135] Step three: re-determining the temporary matrix ​​​​​​​The non-zero minimum value;

[0136] When the non-zero minimum value is not unique, any one of them can be chosen. Let the selected non-zero minimum value be in the temporary matrix. The line, number List;

[0137] Judge the first The and the first Have the cavities been clustered?

[0138] If neither cavity is clustered, then the first The and the first The cavities are grouped into one category and set up ;

[0139] If only one cavity has been clustered, let the cavity that has been clustered be the first one. The cavity that is not clustered is the th cavity. The cavity is defined to contain the first cavity. The type of each cavity is Define intermediate variables , ,exist In this case, the two cavities are allocated to ,exist In this case, only the first Each cavity is allocated to and set ;

[0140] With both cavities already clustered, and set ;

[0141] Check if all cavities have been clustered. If not, iterate through step three to continue clustering until all cavities have been clustered.

[0142] Optionally, the divergence matrix of the N cavities can be calculated using the Kullback-Leibler divergence calculation method. The Kullback-Leibler divergence is calculated as follows:

[0143]

[0144] in, For the first The spindle milling power signal of each cavity Compared to the first The spindle milling power signal of each cavity Kullback-Leibler divergence, For the first The first cavity The spindle milling power is measured at the midpoint, with the tool rotating Z revolutions before and after this midpoint. For the first The first cavity The spindle milling power is calculated by rotating the tool Z revolutions around the midpoint at each measuring point. , The number of measuring points arranged for each cavity.

[0145] Optionally, the intermediate variable is defined when only one cavity has been clustered. and They are respectively:

[0146]

[0147]

[0148] in, For cavities that have been clustered The class, For cavities that were not clustered, for Number of middle cavities for Any cavity in the middle, For all those who do not belong Any cavity within the cavity of the class, for Spindle milling power signal for any cavity Compared to unclustered cavities Spindle milling power signal Kullback-Leibler divergence, For all those who do not belong Spindle milling power signal of any cavity Compared to unclustered cavities Spindle milling power signal The Kullback-Leibler divergence.

[0149] Optionally, a pre-built deep autoencoder neural network is used to extract the initial power signal features of any type of cavity, including:

[0150] Based on a pre-built deep autoencoder neural network, one type of cavity is selected from the clustered cavities, assuming it contains the first... The first cavity, then the second The spindle milling power signal of each cavity the frequency domain signal input into the deep auto-encoder neural network to obtain the initial power signal feature of the first cavity :

[0151]

[0152] wherein, F i is a matrix, is the number of measuring points arranged for each cavity, is the dimension of the power signal feature, is the first cavity, is the first measuring point, is the first cavity, is the first measuring point, is the first cavity,

[0153] Optionally, the correlation model is:

[0154]

[0155] wherein, is the first cavity, is the first measuring point, is the first cavity, is the first measuring point, is the first cavity, is the first measuring point, is the floor size error.

[0156] It should be noted that the foregoing explanation and description of the embodiment of the floor size error prediction method for the cavity of the large thin-walled part also applies to the embodiment of the floor size error prediction device for the cavity of the large thin-walled part, which will not be described here again.

[0157] ​The device for predicting bottom surface size error of a large thin-walled part cavity according to the embodiment of the present application can collect spindle milling power signals of the milling of the large thin-walled part cavity, input the spindle milling power signals of each cavity into a clustering method based on Kullback-Leibler divergence, divide the cavities into different types, build a deep auto-encoder neural network, input the spindle milling power of a type of cavity into the deep auto-encoder network, extract power signal features, and test the bottom surface size error of the type of cavity, based on support vector regression, build a correlation model between the power signal features and the corresponding cavity bottom surface size error, realize prediction of the bottom surface size error based on the power signal features, fine-tune the parameters of the deep auto-encoder neural network based on the transfer learning technology, input the spindle milling power of other types of cavities into the deep auto-encoder network with fine-tuned parameters, extract power signal features, input the power signal features of other types of cavities into the correlation model built, and accurately predict the bottom surface size error of the cavity milling. Thus, the problems in the prior art that the size of the large thin-walled part is large, the clamping stress is unevenly distributed, there are machine tool errors and internal stress, and the actual milling processing environment is complex, resulting in that a physical model is difficult to accurately describe the milling process of all cavities of the large thin-walled part, and for the numerical simulation technology, the accuracy of the calculation result needs to be based on the premise of accurate prerequisite setting, and the calculation efficiency is often easily restricted by the computer computing power are solved.

[0158] Figure 6 A structural schematic diagram of an electronic device according to an embodiment of the present application. The electronic device can include:

[0159] The memory 601, the processor 3602, and a computer program stored on the memory 601 and executable on the processor 602.

[0160] The processor 602 implements the method for predicting the bottom surface size error of the cavity of the large thin-walled part provided in the above embodiments when executing the program.

[0161] Further, the electronic device further includes:

[0162] The communication interface 603 is used for communication between the memory 601 and the processor 602.

[0163] The memory 601 is used to store a computer program executable on the processor 602.

[0164] The memory 601 can include a high-speed RAM memory, and can also include a non-volatile memory, such as at least one disk memory.

[0165] If the memory 601, the processor 602 and the communication interface 603 are implemented independently, the communication interface 603, the memory 601 and the processor 602 can be connected with each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 6 Only one thick line is used to represent the bus in the figure, but it does not mean that there is only one bus or only one type of bus.

[0166] Optionally, in a specific implementation, if the memory 601, the processor 602 and the communication interface 603 are integrated on a chip, the memory 601, the processor 602 and the communication interface 603 can complete communication between each other through an internal interface.

[0167] The processor 602 can be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present application.

[0168] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the above-mentioned method for predicting the size error of the bottom surface of a large thin-walled part cavity.

[0169] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, the skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

[0170] Moreover, the terms "first", "second", etc. are used herein only to describe different instances, and do not imply or suggest relative importance or a number of indicated technical features. Thus, features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "N" is at least two, for example, two, three, etc., unless explicitly and specifically limited otherwise.

[0171] Any process or method descriptions or descriptions of the flow diagrams herein, or otherwise described herein, can be understood as representing the steps of a method implemented by a computer or a processor, or otherwise embodied in computer-readable or machine-readable medium that causes a computer or processor to perform the steps. The embodiments of the present application are not limited to the order of the steps described herein, and the embodiments of the present application can be implemented by other steps or in other orders.

[0172] Logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be embodied in computer-readable or machine-readable media, for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be a product of "tangible" non-transitory computer-readable medium. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electronic connection (an electronic device having one or more wires), a portable computer diskette (a magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via the optically scanning of the paper or other suitable medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and stored in a computer memory.

[0173] It should be understood that each part of the present application can be realized by hardware, software, firmware or their combination. In the above-mentioned embodiments, the N steps or methods can be realized by software or firmware stored in the memory and executed by a suitable instruction execution system. If realized by hardware and as in another embodiment, any one or their combination of the following technologies known in the art can be used: discrete logic circuit with logic gate circuit for implementing logic function on data signal, application specific integrated circuit with suitable combination logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA) and the like.

[0174] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment method can be completed by a program instructing the relevant hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0175] In addition, each functional unit in each embodiment of the present application can be integrated in one processing module, or each unit can exist physically separately, or two or more units can be integrated in one module. The above-mentioned integrated module can be realized in the form of hardware or in the form of software functional module. The integrated module, if realized in the form of software functional module and sold or used as an independent product, can also be stored in a computer readable storage medium.

[0176] The above-mentioned storage medium can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for predicting the bottom surface dimension error of cavities in large thin-walled parts, characterized in that, Includes the following steps: Acquire spindle milling power signals for each cavity in a large, thin-walled part; Based on the spindle milling power signal of each cavity, each cavity is clustered to obtain at least one type of cavity, specifically including: Step 1: Calculate the divergence matrix of the N cavities and set a temporary matrix ; Step 2: Determine the temporary matrix The non-zero minimum value; When the non-zero minimum value is not unique, any one of them can be selected, and let the selected non-zero minimum value be in the temporary matrix. The line, number Column, will the first The and the first The cavities are grouped into one category and set up ; Step 3: Redetermine the temporary matrix The non-zero minimum value; When the non-zero minimum value is not unique, any one of them can be selected, and let the selected non-zero minimum value be in the temporary matrix. The line, number List; Determine the first The and the first Have the cavities been clustered? If neither cavity is clustered, then the first... The and the first The cavities are grouped into one category and set up ; If only one cavity has been clustered, let the cavity that has been clustered be the first one. The cavity that is not clustered is the th cavity. The cavity is defined to contain the first cavity. The type of each cavity is Define intermediate variables , ,exist In this case, the two cavities are allocated to ,exist In this case, only the first Each cavity is allocated to and set , wherein the intermediate variable and They are respectively: in, For cavities that have been clustered The class, For cavities that were not clustered, for Number of cavities for Any cavity in the middle, For all those who do not belong Any cavity within the cavity of the class, for Spindle milling power signal for any cavity Compared to unclustered cavities Spindle milling power signal Kullback-Leibler divergence, For all those who do not belong Spindle milling power signal of any cavity Compared to unclustered cavities Spindle milling power signal Kullback-Leibler divergence; With both cavities already clustered, and set ; Check whether all cavities have been clustered. If not, iterate through step three to continue clustering until all cavities have been clustered. Using a pre-built deep autoencoder neural network, the initial power signal features of any type of cavity are extracted, and the bottom surface dimension error of the current type of cavity is measured; A correlation model is constructed between the initial power signal characteristics and their corresponding actual bottom surface size error, wherein the correlation model is as follows: in, For the first The first cavity Power signal characteristics at each measuring point First-order components, For the first The first cavity The bottom surface dimension error at each measuring point; The parameters of the deep autoencoder neural network are adjusted, and the spindle milling power signals of the remaining other types of cavities are input into the fine-tuned deep autoencoder network to extract multiple power signal features; The multiple power signal features are input into the correlation model to predict the bottom surface dimension error of multiple cavity milling.

2. The method for predicting the bottom surface dimension error of a cavity in a large thin-walled part according to claim 1, characterized in that, The spindle milling power signal for each cavity is: in, For the first The spindle milling power signal for each cavity. For the first The first cavity The spindle milling power is measured at the midpoint, with the tool rotating Z revolutions before and after this midpoint. , The number of measuring points arranged for each cavity.

3. The method for predicting the bottom surface dimension error of a cavity in a large thin-walled part according to claim 1, characterized in that, Calculate the divergence matrix of N cavities using the Kullback-Leibler divergence method. The Kullback-Leibler divergence is calculated as follows: in, For the first The spindle milling power signal of each cavity Compared to the first The spindle milling power signal of each cavity Kullback-Leibler divergence, For the first The first cavity The spindle milling power is measured at the midpoint, with the tool rotating Z revolutions before and after this midpoint. For the first The first cavity The spindle milling power is calculated by rotating the tool Z revolutions around the midpoint at each measuring point. , The number of measuring points arranged for each cavity.

4. The method for predicting the bottom surface dimension error of a cavity in a large thin-walled part according to claim 1, characterized in that, The method of extracting initial power signal features of any type of cavity using a pre-built deep autoencoder neural network includes: Based on the pre-built deep autoencoder neural network, one type of cavity after clustering is selected, assuming it contains the first... The first cavity, then the second The spindle milling power signal of each cavity The frequency domain signal is input into the deep autoencoder neural network to obtain the first... Initial power signal characteristics of each cavity : in, F i for The matrix, The number of measuring points arranged in each cavity. For the dimensions of power signal characteristics, For the first The first cavity Characteristics of power signal at each measurement point Component order.

5. A device for predicting the bottom surface dimension error of a cavity in a large, thin-walled part, characterized in that, include: The acquisition module is used to acquire the spindle milling power signal of each cavity in a large thin-walled part; A cavity clustering module is used to cluster each cavity based on the spindle milling power signal of each cavity to obtain at least one type of cavity, specifically including: Step 1: Calculate the divergence matrix of the N cavities and set a temporary matrix ; Step 2: Determine the temporary matrix The non-zero minimum value; When the non-zero minimum value is not unique, any one of them can be selected, and let the selected non-zero minimum value be in the temporary matrix. The line, number Column, will the first The and the first The cavities are grouped into one category and set up ; Step 3: Redetermine the temporary matrix The non-zero minimum value; When the non-zero minimum value is not unique, any one of them can be selected, and let the selected non-zero minimum value be in the temporary matrix. The line, number List; Determine the first The and the first Have the cavities been clustered? If neither cavity is clustered, then the first... The and the first The cavities are grouped into one category and set up ; If only one cavity has been clustered, let the cavity that has been clustered be the first one. The cavity that is not clustered is the th cavity. The cavity is defined to contain the first cavity. The type of each cavity is Define intermediate variables , ,exist In this case, the two cavities are allocated to ,exist In the case of only the first Each cavity is allocated to and set , wherein the intermediate variable and They are respectively: in, For cavities that have been clustered The class, For cavities that were not clustered, for Number of middle cavities for Any cavity in the middle, For all those who do not belong Any cavity within the cavity of the class, for Spindle milling power signal for any cavity Compared to unclustered cavities Spindle milling power signal Kullback-Leibler divergence, For all those who do not belong Spindle milling power signal of any cavity Compared to unclustered cavities Spindle milling power signal Kullback-Leibler divergence; With both cavities already clustered, and set ; Check whether all cavities have been clustered. If not, iterate through step three to continue clustering until all cavities have been clustered. The feature extraction module is used to extract the initial power signal features of any type of cavity using a pre-built deep autoencoder neural network, and to measure the bottom surface dimension error of the current type of cavity; The correlation module construction module is used to construct a correlation model between the initial power signal characteristics and their corresponding actual bottom surface size error, wherein the correlation model is: in, For the first The first cavity Power signal characteristics at each measuring point First-order components, For the first The first cavity The bottom surface dimension error at each measuring point; The parameter fine-tuning module is used to adjust the parameters of the deep autoencoder neural network and input the spindle milling power signals of the remaining other types of cavities into the fine-tuned deep autoencoder network to extract multiple power signal features; An error prediction module is used to input the multiple power signal features into the correlation model to predict the bottom surface dimension error of multiple cavity milling.

6. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, the processor executing the program to implement the method for predicting bottom surface dimension errors of cavities for large thin-walled parts as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the bottom surface dimension error prediction method for cavities of large thin-walled parts as described in any one of claims 1-4.

Images