CN115600488A

CN115600488A - Method and device for designing molded surface of automobile windshield glass mold based on encoder-decoder model

Info

Publication number: CN115600488A
Application number: CN202211071858.8A
Authority: CN
Inventors: 张儒; 路明标; 孙自飞; 甘雨; 郭震; 李作东; 安旭
Original assignee: Nanjing Tianfu Software Co ltd
Current assignee: Nanjing Tianfu Software Co ltd
Priority date: 2022-09-02
Filing date: 2022-09-02
Publication date: 2023-01-13
Anticipated expiration: 2042-09-02
Also published as: CN115600488B

Abstract

The invention provides a method and a device for designing a molded surface of an automobile windshield mold based on an encoder-decoder model. The method comprises the steps of obtaining a plurality of groups of corresponding glass molded surfaces and mold molded surfaces from historical design schemes, extracting discrete points of each glass molded surface, obtaining geometric characteristic data of each discrete point, and obtaining the normal distance from each discrete point to the corresponding mold molded surface. And establishing an encoder-decoder model according to the geometric characteristic data of all the discrete points and the normal distance from the discrete points to the corresponding mold surface. And extracting discrete points of the required glass profile, and acquiring geometric characteristic data of each discrete point of the required glass profile. And predicting the normal distance from each discrete point on the required glass profile to the required mould profile according to the geometric characteristic data of the discrete point on the required glass profile and the encoder-decoder model, and generating the profile of the required mould by using the geometric characteristic data of the discrete point on the required glass profile and the normal distance from the discrete point on the required glass profile to the required mould profile.

Description

Method and device for designing molded surface of automobile windshield mold based on encoder-decoder model

Technical Field

The invention belongs to the technical field of glass mold design, and particularly relates to a method and a device for designing a molded surface of an automobile windshield mold based on an encoder-decoder model.

Background

The production process of the automobile windshield comprises the following steps: first, a glass sheet is cut, edged, cleaned, dried, heated, and then conveyed to a forming chamber. Then, the molding material is sucked to the mold suction surface of the molding machine in the molding chamber, and is lowered along with the mold suction surface and pressed against the glass mold below. When the glass sheet reaches the softening point, the glass sheet is bent downwards under the influence of gravity and is attached to the glass mold, so that the set molded surface curvature of the glass mold is achieved. Finally, the hot-bending formed glass sheet can be made into the automobile windshield meeting the production requirements after cooling and subsequent steps.

Among them, the glass mold plays an important role in molding a glass original sheet. However, the original glass sheet can rebound in the forming process, so that the geometric characteristics of the molded surface of the glass mold are not completely consistent with the designed curved surface of the automobile windshield.

At present, the design of the molded surface of the automobile windshield glass mold mostly adopts a trial and error method, and the general flow of the method is as follows:

(1) Determining a desired glass profile; (2) preliminarily designing a molded surface of a glass mold according to the molded surface of the glass; (3) trial-producing a glass finished product according to the molded surface of the glass mold; (4) Checking the deviation between the trial-produced glass finished product and the required glass molded surface; (5) And (3) if the deviation is not qualified, returning to the step (2) to adjust the molded surface of the glass mold. However, the trial and error method requires repeated modification of the molded surface of the glass mold, and the method has a long design period, a large amount of work, and high cost, which results in low efficiency of producing the automobile windshield.

With the continuous development of the automobile industry, the styles of automobile windshields are more and more, and the design of the glass mold profile faces new requirements and challenges. The trial and error approach has been difficult to meet the trend in the automotive windshield manufacturing industry due to its high design cost and long design cycle.

Disclosure of Invention

The embodiment of the invention provides a method and a device for designing a molded surface of an automobile windshield mold based on an encoder-decoder model, which aim to solve the problems of higher design cost and longer design period in the prior art.

In order to solve the technical problem, the embodiment of the invention discloses the following technical scheme:

one aspect of the present invention provides a method for designing a mold surface of an automotive windshield mold based on an encoder-decoder model, which is applied to generate a required mold surface according to a required glass surface, and comprises the following steps:

acquiring a plurality of groups of molded surfaces from a historical design scheme, wherein each group of molded surfaces comprises a glass molded surface and a corresponding mold molded surface;

respectively extracting discrete points of each glass profile, and acquiring geometric feature data of each discrete point;

obtaining the normal distance from each discrete point to the molded surface of the corresponding mold;

establishing an encoder-decoder model according to the geometric feature data of all discrete points and the normal distance from the discrete points to the corresponding mould profiles, wherein the input variable of the encoder-decoder model is the geometric feature of the discrete points, and the output variable is the normal distance from the discrete points to the corresponding mould profiles;

extracting discrete points of a required glass profile, and acquiring geometric feature data of each discrete point;

predicting the normal distance from each discrete point on the required glass molded surface to the required mold molded surface according to the geometric characteristic data of the discrete points of the required glass molded surface and the encoder-decoder model;

and generating the profile of the required mould by using the geometric characteristic data of the discrete points of the required glass profile and the normal distance from the discrete points on the required glass profile to the profile of the required mould.

Optionally, the respectively extracting discrete points of each glass profile and acquiring geometric feature data of each discrete point includes:

for each glass profile, discrete points are extracted as follows:

determining a minimum bounding box of the glass profile;

establishing a three-dimensional coordinate system of the glass molded surface by taking the central point of the minimum bounding box as an origin, wherein the x axis of the coordinate system is parallel to the longest edge of the minimum bounding box, and the z axis of the coordinate system is parallel to the shortest edge of the minimum bounding box;

respectively taking two longest boundary lines on the glass molded surface as an upper boundary line and a lower boundary line of the glass molded surface;

respectively carrying out p +1 equal division on the upper boundary line and the lower boundary line, and connecting corresponding equal division points on the upper boundary line and the lower boundary line to obtain p equal division lines;

obtaining p planes which are vertical to the XOY plane and pass through any one of p bisectors, wherein the planes are in one-to-one correspondence with the bisectors;

obtaining p intersecting lines which intersect with p planes on the glass molded surface;

dividing each intersection line into q +1 equal parts; sequentially connecting equant points in corresponding sequence on the p intersecting lines to obtain q connecting lines; the intersections of p said intersecting lines and q said connecting lines are discrete points of said glass profile.

(1) Acquiring the coordinates of each discrete point in the coordinate system of the glass profile;

(2) For each discrete point, obtaining a normal vector S according to the following method:

obtaining a tangent vector S of the discrete point parallel to the x axis _x And, a tangent vector S of the discrete point parallel to the y-axis _y ；

And obtaining a normal vector S by cross multiplication of tangent vectors: s = S _x ×S _y ；

(3) For each discrete point, the average curvature H is obtained as follows:

wherein r is ₁ Is the maximum radius of curvature through the discrete point; r is ₂ Is the minimum radius of curvature through the discrete point;

(4) For each discrete point, a gaussian curvature G is obtained as follows:

G＝r ₁ *r ₂

(5) Aiming at each discrete point, acquiring the arch height AH according to the following method;

the arch height is the distance from the discrete point to the corresponding bisector:

AH＝Dist(P,l _i ),P∈l′ _i ,i＝1,2,…p

wherein the discrete points P are on the intersecting line l' _i To above, l _i Is of l' _i Corresponding bisector.

Optionally, the obtaining a normal distance from each discrete point to the corresponding mold surface includes:

for each discrete point, obtaining the normal distance of the discrete point by adopting the following method:

acquiring an intersection point of a straight line which passes through the discrete point and is parallel to the normal vector of the discrete point and the molded surface of the corresponding mold;

and taking the distance between the discrete point and the intersection point as the normal distance from the discrete point to the corresponding mould surface.

Optionally, the building an encoder-decoder model according to the geometric feature data of all the discrete points and the normal distances from the discrete points to the corresponding mold surface includes:

preprocessing the coordinates of all discrete points by adopting a MinMax (maximum and minimum) normalization mode;

preprocessing normal vectors, average curvatures, gaussian curvatures and arch heights of all discrete points by adopting a Robust Normalization mode;

constructing a data set by utilizing the preprocessed geometric characteristic data of the discrete points and the normal distance from the discrete points to the corresponding mould surface, wherein the coordinates, normal vectors, average curvature, gaussian curvature and arch height of the discrete points are used as input variables of the data set, and the normal distance from the discrete points to the corresponding mould surface is used as an output variable of the data set;

a coder-decoder model between the input variables and the output variables is obtained using all data in the data set.

Optionally, the preprocessing the coordinates of all the discrete points by using a MinMax normalization method includes:

the coordinates of each discrete point are preprocessed according to the following method:

wherein f _ max is the maximum value of all the discrete point coordinates; f _ min is the minimum value of all discrete point coordinates, and f is the original discrete point coordinate; f' is the coordinates of the discrete points after preprocessing.

Optionally, the preprocessing the normal vectors, the average curvatures, the gaussian curvatures, and the arch heights of all the discrete points by using a Robust normalization method includes:

the normal vector, the average curvature, the Gaussian curvature and the arch height of each discrete point are preprocessed according to the following modes:

wherein, f is an original geometric characteristic data value, and the geometric characteristic data value is a data value of one geometric characteristic of a normal vector, an average curvature, a Gaussian curvature and an arch height of the discrete point; f' is a preprocessed geometric characteristic data value corresponding to f; f _ mean is the median of the geometric characteristic data values corresponding to all discrete points and f on the glass molded surface to which the discrete points belong; and IQR is the interval length between the 1 st quartile and the 3 rd quartile in the geometric characteristic data values corresponding to all discrete points and f on the glass molded surface to which the discrete points belong.

Optionally, the obtaining a coder-decoder model between the input variable and the output variable by using all data in the data set includes:

selecting any one of all the glass molded surfaces as a test glass molded surface, and taking the rest glass molded surfaces as training glass molded surfaces;

constructing a test data set by using the geometrical characteristic data of the discrete points preprocessed by the test glass molded surface and the normal distance from the discrete points on the test glass molded surface to the corresponding mold surface, wherein the coordinates, normal vectors, average curvatures, gaussian curvatures and arch heights of the discrete points are used as input variables, and the normal distance from the discrete points to the corresponding mold surface is used as an output variable;

and (3) constructing a training data set by adopting the following method aiming at each training glass profile:

mapping the three-dimensional space information of the training glass molded surface to a two-dimensional plane to obtain a two-dimensional glass molded surface, and simultaneously mapping discrete points on the three-dimensional glass molded surface to the two-dimensional plane;

starting from the upper left corner on the two-dimensional glass profile by using a preset dimension n ₀ ×n ₀ The sliding window obtains all discrete points mapped in the range of the sliding window and the discrete points are used as a sample corresponding to the training glass molded surface in the training data set; taking all newly acquired discrete points as a sample corresponding to the training glass profile in a training data set every time the sliding window slides once until all the discrete points mapped on the two-dimensional glass profile are acquired by the sliding window;

constructing a training data set by using all samples, wherein coordinates, normal vectors, mean curvatures, gaussian curvatures and arch heights of all discrete points in each sample are used as input variables, and the dimensionality of the input variables is 9 multiplied by n ₀ ×n ₀ (ii) a Taking the normal distance from all discrete points in each sample to the corresponding mould surface as an output variable, wherein the dimensionality of the output variable is n ₀ ×n ₀ ；

An encoder-decoder model is built from all training data sets and test data sets.

Optionally, the building an encoder-decoder model according to all training data sets and test data sets includes:

converting the low-dimensional geometric features of discrete points on the training glass profile into high-dimensional geometric features by adopting an encoder;

mapping the high-dimensional geometric features to a low-dimensional output space by adopting a decoder to obtain output data; the encoding and decoding process is represented as:

C＝e(input)

output＝d(C)

wherein e is an encoder model, input is input data of the encoder, C is output data of the encoder and input data of the decoder, d is a decoder model, and output is output data of the decoder; the encoder model and the decoder model are machine learning algorithm models or deep learning algorithm models.

Optionally, the encoder model is composed of a plurality of convolutional layers; the decoder model is composed of a plurality of LSTMs (Long-Short Term Memory networks) and a full connection layer.

Optionally, the building a coder-decoder model according to all training data sets and test data sets further includes:

constructing an initial model of an encoder-decoder model using the plurality of convolutional layers and the plurality of LSTMs and fully-connected layers;

obtaining a final established coder-decoder model using a plurality of loss functions according to the test data set and the initial model, wherein the loss functions include: mean squared error loss, absolute value loss, relative entropy loss, style loss, and correlation loss function; wherein:

the average Square Loss (Mean Square Loss, MSL), the absolute value Loss (L1 Loss, L1) and the relative entropy Loss (Kullback-Leibler Divergence Loss, KLDiv) are all indexes of the evaluation point estimation accuracy, and the specific formula is as follows:

wherein,

representing the real value of the normal distance from the ith discrete point to the corresponding mould surface in the test data set,

representing a predicted value of the normal distance between the ith discrete point and the corresponding mould surface according to the geometric characteristic data of the encoder-decoder model and the ith discrete point, wherein N represents the total number of discrete points in all samples of the test data set;

the style Loss (Gram Loss, gram) and the correlation Loss (Corr) are both indexes for evaluating the two-dimensional plane estimation precision;

the style loss is calculated by using a difference value of vector inner products, and the calculation mode is as follows:

the relevancy loss can represent the relevancy of a horizontal vector and a vertical vector in a sliding window, wherein the horizontal vector is a set of each row in the sliding window, and the vertical vector is a set of each column in the sliding window; wherein, the correlation coefficient between the vectors is calculated by using Pearson correlation coefficient, and the calculation mode is as follows:

where m is the total number of samples in the test data set,

and

respectively representing true values on two-dimensional planes in the k-th sample

And predicted values

Formed matrix, K denotes the matrix

Dimension (d); wherein, the matrix Y ^t And matrix Y ^p The form of (A) is as follows:

the row and column vectors of matrix Y can be represented as

Ycol _i ＝[y _i1 ,y _i. ,…,y _ik ],i＝1,2,…,K；

Yrow _i ＝[y _1i ,y _i2i ,...,y _ki ],i＝1,2,…,K。

Optionally, the obtaining a finally established encoder-decoder model by using multiple loss functions according to the test data set and the initial model includes:

carrying out weighted average on the average square error loss, the absolute value loss, the relative entropy loss, the style loss and the correlation loss function to obtain a total loss function;

and based on the total loss function, determining parameters in the encoder-decoder model again by adopting a gradient descent method to obtain the finally established encoder-decoder model.

Optionally, the predicting a normal distance from each discrete point on the demand glass profile to the demand mold profile according to the geometric feature data of the discrete point on the demand glass profile and the encoder-decoder model comprises:

preprocessing the coordinates of all discrete points on the required glass molded surface by adopting a MinMax normalization mode;

preprocessing normal vectors, average curvatures, gaussian curvatures and arch heights of all discrete points on the required glass profile in a Robust normalization mode;

substituting the geometric characteristic data of all discrete points of the preprocessed required glass molded surface into an encoder-decoder model as input data, and calculating to obtain a predicted value of the normal distance from each discrete point on the required glass molded surface to the required mold molded surface;

judging whether discrete points with a plurality of predicted values exist on the required glass molded surface or not,

if yes, the maximum value in the plurality of predicted values is used as the final predicted value of the corresponding discrete point.

Optionally, the generating the profile of the demand mold by using the geometric characteristic data of the discrete points of the demand glass profile and the normal distance from the discrete points on the demand glass profile to the profile of the demand mold includes:

calculating the coordinates of each discrete point of the required mold surface by using the geometric characteristic data of the discrete points on the required glass mold surface and the normal distance from the discrete points of the required glass mold surface to the required mold surface, wherein the discrete points of the required mold surface correspond to the discrete points on the required glass mold surface one to one and are the discrete points assumed on the required mold surface;

the coordinates of each discrete point of the required mold surface are calculated as follows:

wherein: (x' _i ,y′ _i ,z′ _i ) Coordinates of discrete points of the required mould surface are obtained; (x) _i ,y _i ,z _i ) The coordinates of the corresponding discrete points on the required glass molded surface are obtained; (n) _xi ,n _yi ,n _zi ) The normal vector of the corresponding discrete point on the required glass profile is obtained; d is a radical of _i For the predicted required glass surfaceThe normal distance from the point to the required mould profile is dispersed;

constructing a spline curve network wire frame for connecting discrete points according to the coordinates of the discrete points on the required mould surface;

and carrying out surface reconstruction on the spline curve network wire frame by using a geometric modeling engine to generate the molded surface of the required mold.

Another aspect of the present invention provides an apparatus for designing a mold surface of a windshield glass mold of an automobile based on an encoder-decoder model, the apparatus being applied to generate a required mold surface according to a required glass surface, the apparatus comprising:

the device comprises a glass molded surface and mold molded surface acquisition unit, a storage unit and a control unit, wherein the glass molded surface and mold molded surface acquisition unit is used for acquiring a plurality of groups of molded surfaces from a historical design scheme, and each group of molded surfaces comprises a glass molded surface and a corresponding mold molded surface;

the device comprises a glass profile discrete point data acquisition unit, a data acquisition unit and a data processing unit, wherein the glass profile discrete point data acquisition unit is used for respectively extracting discrete points of each glass profile and acquiring geometric feature data of each discrete point;

the normal distance acquisition unit is used for acquiring the normal distance from each discrete point to the molded surface of the corresponding mold;

the encoder-decoder model establishing unit is used for establishing an encoder-decoder model according to the geometric feature data of all discrete points and the normal distance between each discrete point and the corresponding mold surface, the input variable of the encoder-decoder model is the geometric feature of each discrete point, and the output variable of the encoder-decoder model is the normal distance between each discrete point and the corresponding mold surface;

the glass profile discrete point data acquisition unit is also used for extracting discrete points of a required glass profile and acquiring geometric feature data of each discrete point;

the normal distance prediction unit is used for predicting the normal distance from each discrete point on the required glass molded surface to the required mold molded surface according to the geometric characteristic data of the discrete points of the required glass molded surface and the encoder-decoder model;

and the demand mold profile generation unit is used for generating the profile of the demand mold by utilizing the geometric characteristic data of the discrete points of the demand glass profile and the normal distance from the discrete points on the demand glass profile to the profile of the demand mold.

According to the technical scheme, the method and the device for designing the mold surface of the automobile windshield mold based on the encoder-decoder model provided by the embodiment of the invention have the advantages that firstly, a plurality of groups of corresponding glass mold surfaces and mold surfaces are obtained from a historical design scheme, discrete points of each glass mold surface are respectively extracted, the geometric characteristic data of each discrete point is obtained, and the normal distance from each discrete point to the corresponding mold surface is obtained.

Secondly, establishing an encoder-decoder model according to the geometric characteristic data of all the discrete points and the normal distance from the discrete points to the corresponding mold surface. And thirdly, extracting discrete points of the required glass profile, and acquiring the geometric feature data of each discrete point on the required glass profile.

And finally, according to the geometric characteristic data of the discrete points of the required glass profile and the encoder-decoder model, predicting the normal distance from each discrete point on the required glass profile to the required mold profile, and generating the profile of the required mold by using the geometric characteristic data of the discrete points of the required glass profile and the normal distance from the discrete points on the required glass profile to the required mold profile.

Therefore, the method and the device provided by the embodiment of the invention can greatly improve the design efficiency of the molded surface of the automobile windshield glass mold and effectively reduce the iteration cost in the design process.

Drawings

FIG. 1 is a schematic flow chart of a method for designing a profile of an automotive windshield mold based on an encoder-decoder model according to an embodiment of the present invention;

fig. 2 is a flowchart illustrating an implementation of step S102 in fig. 1 according to an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of a minimum enclosure for a windshield of an automobile according to an embodiment of the invention;

FIG. 4 is a schematic view of discrete points on a glass profile provided in accordance with one embodiment of the present invention;

fig. 5 is a schematic flowchart illustrating an implementation of step S104 in fig. 1 according to an embodiment of the present invention;

fig. 6 is a schematic flowchart illustrating an implementation of step S107 in fig. 1 according to an embodiment of the present invention;

fig. 7 is a schematic structural diagram of an apparatus for designing a profile of an automotive windshield mold based on an encoder-decoder model according to an embodiment of the present invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the present invention will be described in further detail with reference to the accompanying drawings and specific embodiments.

FIG. 1 is a schematic flow chart of a method for designing a mold surface of an automobile windshield glass mold based on an encoder-decoder model, which is applied to generate a required mold surface according to a required glass mold surface and is characterized in that. As shown in fig. 1, the method comprises the steps of:

step S101: a plurality of sets of profiles are obtained from a historical design scheme, and each set of profiles comprises a glass profile and a corresponding mold profile.

The glass molded surface and the die molded surface in each group of molded surfaces correspond to each other, and each glass molded surface is manufactured by the corresponding die molded surface.

Step S102: and respectively extracting discrete points of each glass profile, and acquiring geometric characteristic data of each discrete point.

In one embodiment of the present disclosure, for each glass profile, as shown in fig. 2, discrete points are extracted in the following manner:

step S1021: a minimum bounding box of the glass profile is determined.

The minimum bounding box is the minimum bounding rectangle of the glass profile, which in the disclosed embodiment of the invention can be obtained using existing algorithms.

Step S1022: and establishing a three-dimensional coordinate system of the glass molded surface by taking the central point of the minimum bounding box as an origin.

As shown in FIG. 3, the x-axis of the three-dimensional coordinate system is parallel to the longest side of the minimum bounding box, the z-axis of the coordinate system is parallel to the shortest side of the minimum bounding box, and the y-axis of the coordinate system is parallel to the middle length side of the minimum bounding box.

Step S1023: the two longest boundary lines on the glass profile are respectively used as the upper boundary line and the lower boundary line of the glass profile.

A typical glass profile is an irregular curved surface that is approximately rectangular, and the boundaries of the glass profile are formed by curves that curve outward. When the glass mold surface is upright, the boundary lines are divided into an upper boundary line, a lower boundary line, a left boundary line and a right boundary line. Wherein, two upper and lower borderlines of the glass profile are longest.

Step S1024: and respectively dividing the upper boundary line and the lower boundary line by p +1 equally, and connecting corresponding equally dividing points on the upper boundary line and the lower boundary line to obtain p equally dividing lines.

P +1 equal division is respectively carried out on the upper boundary line and the lower boundary line, thereby obtaining p equal division points up on the upper boundary line _i I =1,2, \8230p, and p bisector points lp on the lower boundary line _i ,i＝1,2,…p。

Connecting corresponding bisectors on upper and lower boundary lines, e.g. up ₁ Connection lp ₁ ，up ₂ Connection lp ₂ . After connecting the corresponding p equal division points on the upper and lower boundary lines, obtaining p equal division lines l _i ,i＝1,2,…p。

Step S1025: p planes are obtained that are perpendicular to the XOY plane and pass through any of the p bisectors.

By a bisector l ₁ For example, a bisector l is obtained ₁ And at the same time, perpendicular to the XOY plane on the three-dimensional coordinate system, which can intersect the glass profile. According to the method, p planes are obtained, and each plane passes through a bisector l _i And perpendicular to the XOY plane. Each plane passing through only one bisector l _i The planes correspond to the bisectors one to one.

Step S1026: p intersecting lines intersecting the p planes on the glass profile are obtained.

The obtained p planes all intersect the glass profile surface, so that p intersecting lines l 'on the glass profile surface can be obtained' _i I =1,2, \8230p, p, the intersection line being a curve.

Step S1027: dividing each intersection line into q +1 equal parts; sequentially connecting equant points in corresponding sequence on the p intersecting lines to obtain q connecting lines; the intersection of p intersecting lines and q connecting lines are discrete points of the glass profile.

To each intersecting line l 'on the glass profile surface' _i I =1,2, \8230pis equally divided by q +1, obtaining each line of intersection l' _i I =1,2, \ 8230, q bisectors on p.

Connecting bisector points of corresponding order on p intersecting lines in turn, e.g. connecting intersecting lines l' ₁ The first bisector point, the intersecting line l' ₂ The first bisector point on, in turn, until connected to the intersecting line l' _p To obtain a connecting line connecting the first bisector point on each intersecting line. After the connection has been completed at the qth bisector point on each intersecting line, q connecting lines on the glass profile are obtained.

As shown in fig. 4, the intersection of p intersecting lines with q connecting lines is taken as a discrete point of the corresponding glass profile.

In the embodiment disclosed in the present invention, the geometric characteristics required to obtain the discrete points are the coordinates of the discrete points, normal vectors, mean curvature, gaussian curvature and arch height.

The reason for extracting the above characteristic feature parameters is as follows:

combining the processes of glass hot bending forming and rebounding, after the glass sheet is heated at high temperature, the glass sheet is lifted and attached to the suction mold through the combined action of vacuum adsorption and blowing-up of the suction film; then the suction mould and the glass descend to extrude the glass together with the hot ring; after being pressed and formed, the glass falls on a cold ring and is quickly supported out to be sent to a quenching air grid area for quenching and tempering.

The glass sheet is pressed and formed by the surface of the suction mould and the hot ring surface, and then is dropped on the cold ring by gravity. The stress distribution conditions at each stage are as follows: when the glass sheet is pressed and formed, the surface of the suction mold firstly contacts the middle area of the glass sheet and is gradually attached to the boundary, and the stress is small and uneven. After the edge of the glass is extruded by the surface of the suction mold and the hot ring surface, the boundary stress is increased and is distributed more uniformly. When the mold is dropped, the glass has a certain speed and falls on the cold ring under the action of the gravity of the glass, so that impact load is generated, and the stress of the edge part of the glass is suddenly increased.

Therefore, stress distribution of the automobile windshield at different positions in the forming process has certain difference, which can cause that the deviation of different positions on the formed glass molded surface and the mold molded surface also has difference. Based on this, position information of discrete points on the glass profile, i.e. discrete point coordinates (x/y/z), is extracted.

During the hot bending forming process of the glass sheet, when the glass sheet is subjected to the action of an external bending moment, the shape of the glass sheet is changed. When the glass profile is bent, the original glass sheet on the inner layer of the deformation area is subjected to compression deformation by tangential compressive stress, the original glass sheet on the outer layer is subjected to tensile deformation by tangential tensile stress, and the tangential direction also influences the deformation of the glass profile, so that a unit normal vector (vector _ x/vector _ y/vector _ z) perpendicular to the tangential direction is extracted for representing the tangential characteristic of the glass profile.

Meanwhile, the curvature radius of the mold surface is a main factor for determining the glass surface, the bending degree of the glass surface depends on the shape of the mold surface, and the rebound degree is related to the curvature radius of the mold surface. This is because the smaller the radius of curvature of the die surface, the deeper the relative bending degree becomes, the greater the proportion of elastic deformation in the total deformation becomes, and the larger the spring back becomes after the external force is removed. Thus, the curvature characteristics of the glass profile, including the mean curvature H and the gaussian curvature K, are extracted.

In addition, to characterize the degree of curvature of the glass profile, the camber AH of the discrete points is also extracted.

Respectively acquiring the data value of each geometrical characteristic of the discrete points according to the following modes:

(1) The coordinates of each discrete point are acquired in the coordinate system of the glass profile.

By passingCross multiplication of tangent vectors to obtain a normal vector S: s = S _x ×S _y ；

(3) For each discrete point, the average curvature H is obtained as follows:

(4) For each discrete point, a gaussian curvature G is obtained as follows:

G＝r ₁ *r ₂

wherein r is ₁ Is the maximum radius of curvature through the discrete point; r is a radical of hydrogen ₂ Is the minimum radius of curvature through the discrete point;

(5) Aiming at each discrete point, the arch height AH from the discrete point P to the corresponding bisector l is obtained according to the following method _i The distance of (c):

AH＝Dist(P,l _i ),P∈l′ _i ,i＝1,2,…p

wherein the discrete points P are on the intersecting line l' _i I =1,2, \ 8230;, p is upper, l _i Is l' _i Corresponding bisector.

Step S103: and acquiring the normal distance from each discrete point to the corresponding mould surface.

In a disclosed embodiment of the invention, the normal distance of each discrete point to the corresponding mold profile is obtained as follows:

for each discrete point, the normal distance of the discrete point is obtained by adopting the following method:

(1) And acquiring the intersection point of a straight line which passes through the discrete point and is parallel to the normal vector of the discrete point and the molded surface of the corresponding mold.

(2) And taking the distance between the discrete point and the intersection point as the normal distance from the discrete point to the corresponding mold surface.

Step S104: and establishing an encoder-decoder model according to the geometric characteristic data of all the discrete points and the normal distance from the discrete points to the corresponding mould surface.

The input variable of the encoder-decoder model is the geometrical characteristic of the discrete point, and the output variable is the normal distance from the discrete point to the corresponding mould surface.

In one embodiment of the present disclosure, as shown in fig. 5, the encoder-decoder model is built in the following way:

the geometric feature data of the discrete points is preprocessed before the encoder-decoder model is built.

Step S1041: and preprocessing the coordinates of all discrete points by adopting a MinMax normalization mode.

In one embodiment of the present disclosure, the coordinates of each discrete point are preprocessed according to the following method:

Step S1042: and preprocessing normal vectors, average curvatures, gaussian curvatures and arch heights of all discrete points by adopting a Robust normalization mode.

In one embodiment of the present disclosure, the normal vector, mean curvature, gaussian curvature and arch height of each discrete point are preprocessed as follows:

wherein, f is an original geometric characteristic data value, and the geometric characteristic data value is a data value of one of the geometric characteristics of a normal vector, an average curvature, a Gaussian curvature and an arch height of a discrete point; f' is a preprocessed geometric characteristic data value corresponding to f; f _ mean is the median of the geometric characteristic data values corresponding to all discrete points and f on the glass molded surface to which the discrete points belong; the IQR is the interval length between the 1 st quartile and the 3 rd quartile in the geometric characteristic data values corresponding to all discrete points and f on the glass molded surface to which the discrete points belong. For example, f is a data value of an original camber of the discrete point, f' is a data value of a pre-processed camber of the discrete point, f _ mean is a median of all data values of the camber of the discrete point on the glass surface to which the discrete point belongs, and IQR is an interval length between a 1 st quartile and a 3 rd quartile of all data values of the camber of the discrete point on the glass surface to which the discrete point belongs.

Step S1043: and constructing a data set by utilizing the preprocessed geometric characteristic data of the discrete points and the normal distance from the discrete points to the corresponding mould surface.

The coordinates of the discrete points, the normal vector, the average curvature, the Gaussian curvature and the arch height are used as input variables in the data set, and the normal distance from the discrete points to the corresponding die profiles is used as an output variable in the data set.

Step S1044: a coder-decoder model between the input variables and the output variables is obtained using all data in the data set.

In one embodiment of the present disclosure, this step may be implemented by the following substeps:

(1) Any one of all the glass profiles is selected as a test glass profile, and all the glass profiles except the test glass profile are used as training glass profiles.

(2) And constructing a test data set by using the geometrical characteristic data of the discrete points after the pretreatment of the test glass molded surface and the normal distance from the discrete points on the test glass molded surface to the corresponding mold molded surface. The coordinates, normal vectors, average curvatures, gaussian curvatures and arch heights of the discrete points are used as input variables, and the normal distances from the discrete points to the corresponding die profiles are used as output variables.

(3) And (3) constructing a training data set by adopting the following method aiming at each training glass profile:

and mapping the three-dimensional space information of the training glass molded surface to a two-dimensional plane to obtain the two-dimensional glass molded surface, and simultaneously mapping discrete points on the three-dimensional glass molded surface to the two-dimensional plane.

Starting with the upper left corner on the two-dimensional glass profile, using a predetermined dimension n ₀ ×n ₀ The sliding window of (a) obtains all discrete points mapped within the range of the sliding window as one sample of the training data set corresponding to the training glass profile, in one embodiment of the disclosure, the preset size of the sliding window is 32 x 32. And taking all newly acquired discrete points as a sample corresponding to the training glass molded surface in the training data set every time the sliding window slides once, and sequentially acquiring the samples according to the mode until the sliding window acquires all the discrete points mapped on the two-dimensional glass molded surface.

Constructing a training data set by using all samples, wherein the coordinates, normal vectors, mean curvatures, gaussian curvatures and arch heights of all discrete points in each sample are used as input variables, and the dimensionality of the input variables is 9 xn ₀ ×n ₀ Wherein 9 means the total dimension of the geometric feature, wherein the coordinates and normal vectors of the discrete points have 3 dimensions respectively, and the mean curvature, gaussian curvature and arch height have 1 dimension respectively, for 9 dimensions.

Taking the normal distance from all discrete points in each sample to the corresponding mould surface as an output variable, wherein the dimensionality of the output variable is n ₀ ×n ₀ 。

And acquiring a training data set corresponding to each training glass profile according to the mode.

(4) An encoder-decoder model is built from all training data sets and test data sets.

In the disclosed embodiment of the present invention, the encoder-decoder model can be built in the following way:

(1) And (3) converting the low-dimensional geometric features of the discrete points on the training glass profile into high-dimensional geometric features by adopting an encoder.

(2) And mapping the high-dimensional geometric features to a low-dimensional output space by adopting a decoder to obtain output data. The encoding and decoding process is represented as:

C＝e(input)

output＝d(C)

wherein e is the encoder model, input is the input data of the encoder, C is the output data of the encoder and the input data of the decoder, and is the high-dimensional geometric feature, d is the decoder model, and output is the output data of the decoder. In the disclosed embodiments, the encoder and decoder models include, but are not limited to, common machine learning or deep learning algorithms such as SVR (support vector regression), GBDT (gradient lifting tree), CNN (convolutional neural network), RNN (recurrent neural network), and the like.

In one embodiment of the present disclosure, the encoder model is made up of a plurality of convolutional layers; the decoder model consists of multiple LSTMs (Long-Short Term Memory networks) and a full connection layer.

In one embodiment of the present disclosure, the encoder-decoder model established according to all training data sets and testing data sets in the foregoing embodiment can be implemented as follows:

(1) An initial model of the encoder-decoder model is constructed using all of the training data sets, the plurality of convolutional layers, and the plurality of LSTM and fully-connected layers.

In one embodiment of the present disclosure, the initial model of the encoder-decoder may be constructed in the following manner:

a convolutional neural network is adopted as an encoder, and an LSTM and a full-connection layer are adopted as a decoder, wherein the convolutional neural network is formed by connecting 2 convolutional blocks, and the dimensionality of two convolutional kernels is 3 multiplied by 3 and 3 multiplied by 3 in sequence. The first convolutional layer expands the input 9 channels to 16 channels, and the second convolutional layer expands the 16 channels to 32.

Each convolution layer comprises a convolution kernel and a batch normalization layer, and a LeakyRelu function is used as an activation function; the LeakyRelu function is defined as:

wherein x is the activation value of a single neuron; λ is the slope where x is less than 0, and in one particular embodiment of the present disclosure, λ =0.2.

The decoder converts the spatial feature information into sequence data by adopting 2-layer LSTM for decoding, the input dimension of the first-layer LSTM is 32 x 1024, wherein 32 is the sequence length and is equal to the number of output channels of the encoder, 1024 is the number of sequences, and each channel (32 x 32) is obtained after expansion. The output dimension is 4 × 1024, and the input and output dimensions of the second layer LSTM are all 4 × 1024.

(2) And obtaining a finally established coder-decoder model by adopting various loss functions according to the test data set and the initial model. Wherein the loss function comprises: mean squared error loss, absolute value loss, relative entropy loss, style loss, and correlation loss function, wherein:

mean Square Loss (MSL), absolute value Loss (L1 Loss, L1) and relative entropy Loss (KLDiv) are all indexes for evaluating point estimation accuracy, which is the average error between the predicted data and the actual data of each discrete point in the test data set.

The specific formulas for the mean squared loss, absolute loss and relative entropy loss are as follows:

wherein,

the style Loss (Gram) and the correlation Loss (Corr) are both indexes for evaluating the two-dimensional plane estimation accuracy, and the two-dimensional plane estimation accuracy refers to the average deviation between the predicted data and the real data of each sample in the test data set.

the relevance loss can represent the relevance degree of a transverse vector and a longitudinal vector in the sliding window, wherein the transverse vector is a set of each row in the sliding window, and the longitudinal vector is a set of each column in the sliding window; wherein, the correlation coefficient between the vectors is calculated by using Pearson correlation coefficient, and the calculation mode is as follows:

where m is the total number of samples in the test data set,

and

respectively representing true values on two-dimensional plane in kth sample

And predicted values

Formed matrix, K represents the matrix

therefore, the row and column vectors of the matrix Y can be represented as

Ycol _i ＝[y _i1 ,y _i. ,…,y _ik ],i＝1,2,…,K；

Yrow _i ＝[y _1i ,y _i2i ,...,y _ki ],i＝1,2,…,K。

In an embodiment of the disclosure, the obtaining a final established encoder-decoder model by using multiple loss functions according to the test data set and the initial model in the foregoing embodiment includes the following steps:

(1) And carrying out weighted average on the average squared error loss, the absolute value loss, the relative entropy loss, the style loss and the correlation loss function calculated in the previous embodiment to obtain a total loss function.

(2) And based on the total loss function, re-determining parameters in the encoder-decoder model by adopting a gradient descent method, and obtaining the finally established encoder-decoder model.

After the final encoder-decoder model is built, the process proceeds to step S105.

Step S105: discrete points of the required glass profile are extracted, and geometric characteristic data of each discrete point are obtained.

The required glass profile is a glass profile required to be designed corresponding to the mould profile and is a pre-designed glass profile.

And constructing a three-dimensional coordinate system of the required glass profile in a manner of extracting discrete points of the glass profile according to the previous embodiment, and extracting the discrete points of the required glass profile, which is not described herein again.

According to the method for acquiring the geometric characteristic data of each discrete point on the glass profile in the foregoing embodiment, the geometric characteristic data of each discrete point on the required glass profile is acquired, which is not described herein again.

Step S106: and predicting the normal distance from each discrete point on the required glass molded surface to the required mold molded surface according to the geometric characteristic data of the discrete points of the required glass molded surface and the encoder-decoder model.

And substituting the geometric characteristic data of the discrete points on the required glass molded surface into the finally established encoder-decoder model, wherein the encoder-decoder model can predict the normal distance from each discrete point to the required molded surface of the mold.

In one embodiment of the present disclosure, this step may be accomplished by the following substeps:

(1) According to the mode in the previous embodiment, all the coordinates of discrete points on the required glass profile are preprocessed in a MinMax normalization mode;

(2) According to the mode in the embodiment, preprocessing the normal vectors, the average curvature, the Gaussian curvatures and the arch heights of all discrete points on the required glass molded surface by a Robust normalization mode;

(3) And substituting the geometric characteristic data of all discrete points of the preprocessed required glass molded surface as input data into the finally established encoder-decoder model, and calculating to obtain a predicted value of the normal distance from each discrete point on the required glass molded surface to the required mold molded surface.

(4) Judging whether discrete points with a plurality of predicted values exist on the required glass molded surface or not,

if yes, the one or more discrete points are explained to have more than two predicted values, and when the situation happens, the maximum value in the plurality of predicted values is taken as the final predicted value of the corresponding discrete point.

If not, the abnormal condition does not exist, and the unique predicted value corresponding to each discrete point is normally determined.

Step S107: and generating the profile of the required mould by using the geometric characteristic data of the discrete points of the required glass profile and the normal distance from the discrete points on the required glass profile to the profile of the required mould.

In the disclosed embodiment of the present invention, as shown in fig. 6, step S107 can be completed by the following sub-steps:

step S1071: and calculating the coordinates of each discrete point of the required mold surface by using the geometric characteristic data of the discrete point on the required glass mold surface and the normal distance from the discrete point of the required glass mold surface to the required mold surface.

The discrete points of the required mold surface correspond to the discrete points on the required glass mold surface one to one, and are assumed discrete points on the required mold surface.

wherein: (x' _i ,y′ _i ,z′ _i ) Coordinates of discrete points of the required mould surface are obtained; (x) _i ,y _i ,z _i ) The coordinates of the corresponding discrete points on the required glass molded surface are obtained; (n) _xi ,n _yi ,n _zi ) The normal vector of the corresponding discrete point on the required glass profile is obtained; d _i The normal distance from the corresponding discrete point on the predicted required glass profile to the required mould profile.

Step S1072: and constructing a spline curve network wire frame for connecting the discrete points according to the coordinates of the discrete points on the required mould surface.

In one embodiment of the present disclosure, step S1072 may be implemented by:

(1) The discrete points of the required mould surface are transposed into a p multiplied by q two-dimensional matrix by utilizing the existing geometric modeling engine.

(2) And randomly selecting a plurality of groups of discrete points with preset quantity from all the discrete points of the required mould surface, and correspondingly generating a plurality of detection mould surfaces of the required mould by respectively utilizing a geometric modeling engine. Wherein each test profile is generated from a predetermined number of discrete points.

(3) According to the coordinates of the discrete points of the required mould surface and the detection molded surfaces of a plurality of groups of required moulds, average errors between the discrete points of the required mould surface and the detection molded surface of each required mould are respectively calculated, namely, the average errors between all the discrete points of the required mould surface and the detection molded surface of each required mould are calculated. The average error may be the average of all discrete points to the normal distance of the detection profile.

(4) And determining the detection profile with the minimum average error, and taking the corresponding discrete point as the selected discrete point.

(5) And connecting the selected discrete points to form a spline curve network wireframe.

Step S1073: and carrying out surface reconstruction on the spline curve network wire frame by using a geometric modeling engine to generate the molded surface of the required mold.

And (4) carrying out surface reconstruction on the spline curve network wire frame by utilizing the existing geometric modeling engine to construct a final required mold surface.

FIG. 7 is a schematic structural diagram of an apparatus for designing a profile of an automotive windshield mold based on an encoder-decoder model, which is applied to generate a required mold profile according to a required glass profile. As shown in fig. 7, the apparatus includes:

a glass profile and mold profile obtaining unit 11 configured to obtain a plurality of sets of profiles from a historical design plan, each set of profiles including one glass profile and a corresponding one mold profile;

a glass profile discrete point data acquisition unit 12 configured to extract discrete points of each glass profile, respectively, and acquire geometric feature data of each discrete point;

a normal distance obtaining unit 13 configured to obtain a normal distance of each discrete point to the corresponding mold surface;

an encoder-decoder model establishing unit 14 configured to establish an encoder-decoder model according to the geometric feature data of all the discrete points and the normal distances from the discrete points to the corresponding mold surfaces, wherein the input variable of the encoder-decoder model is the geometric feature of the discrete points, and the output variable is the normal distance from the discrete points to the corresponding mold surfaces;

the glass profile discrete point data acquisition unit 12 is further configured to extract discrete points of a required glass profile and acquire geometric feature data of each discrete point;

a normal distance prediction unit 15 configured to predict a normal distance from each discrete point on the demand glass profile to the demand mold profile based on the geometric feature data of the discrete points of the demand glass profile and the encoder-decoder model;

and a required mold profile generation unit 16 configured to generate a profile of the required mold by using the geometric characteristic data of the discrete points of the required glass profile and the normal distance from the discrete points on the required glass profile to the required mold profile.

It will be understood that the above embodiments are merely exemplary embodiments taken to illustrate the principles of the present invention, which is not limited thereto. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.

Claims

1. A method for designing a mold surface of an automobile windshield glass mold based on an encoder-decoder model is applied to generating a required mold surface according to a required glass mold surface, and is characterized by comprising the following steps:

and generating the profile of the required mold by using the geometric characteristic data of the discrete points of the required glass profile and the normal distance from the discrete points on the required glass profile to the profile of the required mold.

2. The method according to claim 1, wherein said separately extracting discrete points of each glass profile and obtaining geometric feature data for each of said discrete points comprises:

for each glass profile, discrete points are extracted as follows:

determining a minimum bounding box of the glass profile;

respectively dividing the upper boundary line and the lower boundary line by p +1 equally, and connecting corresponding equally dividing points on the upper boundary line and the lower boundary line to obtain p equally dividing lines;

obtaining p intersecting lines which intersect with the p planes on the glass molded surface;

dividing each intersection line into q +1 equal parts; sequentially connecting equant points in a corresponding sequence on the p intersecting lines to obtain q connecting lines; the intersections of the p intersecting lines and the q connecting lines are discrete points of the glass profile.

3. The method of claim 2, wherein said separately extracting discrete points of each glass profile and obtaining geometric feature data for each of said discrete points further comprises:

(1) Acquiring the coordinates of each discrete point in the coordinate system of the glass molded surface;

(3) For each discrete point, the average curvature H is obtained as follows:

(4) For each discrete point, a gaussian curvature G is obtained as follows:

G＝r ₁ *r ₂

AH＝Dist(P,l _i ),P∈l′ _i ,i＝1,2,…p

wherein the discrete points P are on the intersecting line l' _i Upper, l _i Is of l' _i Corresponding bisector.

4. The method of claim 3, wherein said obtaining a normal distance of each discrete point to a corresponding mold profile comprises:

for each discrete point, obtaining the normal distance of the discrete point by adopting the following mode:

and taking the distance between the discrete point and the intersection point as the normal distance from the discrete point to the corresponding mould profile.

5. A method according to claim 3, wherein said building an encoder-decoder model from the geometric characteristic data of all discrete points and the normal distance of said discrete points to the corresponding mold profile comprises:

constructing a data set by utilizing the preprocessed geometric characteristic data of the discrete points and the normal distance from the discrete points to the corresponding mould surface, wherein the coordinates, normal vectors, average curvatures, gaussian curvatures and arch heights of the discrete points are used as input variables of the data set, and the normal distance from the discrete points to the corresponding mould surface is used as an output variable of the data set;

6. The method of claim 5, wherein the preprocessing of the coordinates of all the discrete points by MinMax normalization comprises:

7. The method according to claim 5, wherein the preprocessing of the normal vector, the mean curvature, the Gaussian curvature and the arch height of all discrete points is performed by using a Robust normalization method, and comprises:

wherein, f is an original geometric characteristic data value, and the geometric characteristic data value is a data value of one of the geometric characteristics of the normal vector, the average curvature, the Gaussian curvature and the arch height of the discrete point; f' is a preprocessed geometric characteristic data value corresponding to f; f _ mean is the median of the geometric characteristic data values corresponding to all the discrete points and f on the glass molded surface to which the discrete points belong; and IQR is the interval length between the 1 st quartile and the 3 rd quartile in the geometric characteristic data values corresponding to all discrete points and f on the glass molded surface to which the discrete points belong.

8. The method of claim 5, wherein obtaining a coder-decoder model between an input variable and an output variable using all data in a data set comprises:

the following method is adopted to construct a training data set for each training glass profile:

starting from the upper left corner on the two-dimensional glass profile by using a preset dimension n ₀ ×n ₀ The sliding window obtains all discrete points mapped in the range of the sliding window as a sample corresponding to the training glass molded surface in the training data set; taking all newly acquired discrete points as a sample corresponding to the training glass molded surface in a training data set every time the sliding window slides once until all the discrete points mapped on the two-dimensional glass molded surface are acquired by the sliding window;

constructing a training data set by using all samples, wherein the coordinates, normal vectors, mean curvatures, gaussian curvatures and arch heights of all discrete points in each sample are used as input variables, and the input variables areDimension of 9 Xn ₀ ×n ₀ (ii) a Taking the normal distance from all discrete points in each sample to the corresponding mould surface as an output variable, wherein the dimensionality of the output variable is n ₀ ×n ₀ ；

9. The method of claim 8, wherein building a coder-decoder model from all training data sets and test data sets comprises:

C＝e(input)

output＝d(C)

10. The method of claim 9,

the encoder model is composed of a plurality of convolution layers; the decoder model is composed of a plurality of LSTMs (Long-Short Term Memory networks) and a full connection layer.

11. The method of claim 10, wherein building a coder-decoder model from all of the training data sets and the test data sets further comprises:

obtaining a final established encoder-decoder model using a plurality of loss functions based on the test data set and the initial model, the loss functions including: mean squared error loss, absolute value loss, relative entropy loss, style loss, and correlation loss functions; wherein:

wherein,

the style loss is calculated by using the difference value of the vector inner product, and the calculation mode is as follows:

where m is the total number of samples in the test data set,

and

And the predicted value

Formed matrix, K denotes the matrix

Dimension (d); wherein, the matrix Y ^t And matrix Y ^p The forms of the two are as follows:

the row and column vectors of matrix Y can be represented as

Ycol _i ＝[y _i1 ,y _i. ,…,y _ik ],i＝1,2,…,K；

Yrow _i ＝[y _1i ,y _i2i ,...,y _ki ],i＝1,2,…,K。

12. The method of claim 11, wherein obtaining a final established encoder-decoder model using a plurality of loss functions based on the test data set and the initial model comprises:

13. The method of claim 12, wherein predicting a normal distance of each discrete point on the desired glass profile to the desired mold surface based on the geometric characteristic data of the discrete points of the desired glass profile and the encoder-decoder model comprises:

preprocessing normal vectors, average curvatures, gaussian curvatures and arch heights of all discrete points on a required glass profile by adopting a Robust normalization mode;

judging whether discrete points with a plurality of predicted values exist on the required glass profile,

and if so, taking the maximum value in the plurality of predicted values as the final predicted value of the corresponding discrete point.

14. The method according to any one of claims 1 to 13, wherein generating the profile of the desired mold using the geometric characteristic data of the discrete points of the desired glass profile and the normal distance of the discrete points on the desired glass profile to the desired mold profile comprises:

calculating the coordinate of each discrete point of the required mold surface by using the geometric characteristic data of the discrete points on the required glass mold surface and the normal distance from the discrete points of the required glass mold surface to the required mold surface, wherein the discrete points of the required mold surface correspond to the discrete points on the required glass mold surface one to one and are the assumed discrete points on the required mold surface;

wherein: (x' _i ,y′ _i ,z′ _i ) Coordinates of discrete points of the required mould surface are obtained; (x) _i ,y _i ,z _i ) For on-demand glass profile correspondenceCoordinates of discrete points; (n) _xi ,n _yi ,n _zi ) The normal vector of the corresponding discrete point on the required glass profile is obtained; d _i The normal distance from the corresponding discrete point on the predicted required glass molded surface to the required mold molded surface;

15. An automobile windshield mold profile designing device based on an encoder-decoder model is applied to generating a demand mold profile according to the demand glass profile, and is characterized by comprising the following steps:

the device comprises a glass molded surface and mold molded surface acquisition unit, a storage unit and a control unit, wherein the glass molded surface and mold molded surface acquisition unit is used for acquiring a plurality of groups of molded surfaces from historical design schemes, and each group of molded surfaces comprises a glass molded surface and a corresponding mold molded surface;

the device comprises a glass profile discrete point data acquisition unit, a data acquisition unit and a data acquisition unit, wherein the glass profile discrete point data acquisition unit is used for respectively extracting discrete points of each glass profile and acquiring geometric characteristic data of each discrete point;

the normal distance obtaining unit is used for obtaining the normal distance from each discrete point to the corresponding mould surface;

the encoder-decoder model establishing unit is used for establishing an encoder-decoder model according to the geometric characteristic data of all discrete points and the normal distance from the discrete points to the corresponding mould surface, wherein the input variable of the encoder-decoder model is the geometric characteristic of the discrete points, and the output variable is the normal distance from the discrete points to the corresponding mould surface;

the glass profile discrete point data acquisition unit is also used for extracting discrete points of a required glass profile and acquiring geometric characteristic data of each discrete point;