Predictions of characteristic target data

Abstract

Device (100; 1900) for predicting characteristic data, the device comprising: a processor (2000); and one or more computer-readable media (2020, 2040, 2060, 2095) which collectively contain instructions which, when executed by the processor, cause the processor to: obtain (S410) a plurality of physical structure data and a plurality of characteristic data, wherein each physical structure corresponds to a characteristic data from the plurality of characteristic data and the characteristic data contains a plurality of characteristic values, wherein each characteristic value relates to a physical structure corresponding to a physical structure data from the plurality of physical structure data, wherein each characteristic data contains at least one characteristic value from the plurality of characteristic values ​​that represents a temporal change of a characteristic.and includes at least one characteristic value from the plurality of characteristic values, representing a change in a characteristic with respect to an impact on the corresponding physical structure or representing a deformation of the physical structure; Determining (S430) a similarity function for estimating a new structural similarity, wherein the determination is based on at least one characteristic similarity between characteristic data corresponding to the at least two physical structures; Estimating (S440), based on the similarity function, at least one structural similarity between at least two physical structures,the physical structure data from the multitude of physical structure data; generating (S450) an estimation model for estimating a characteristic target datum from a physical target structure datum using characteristic data and according to structural similarities between the physical target structure datum and each from the multitude of physical structure data, specifying a weighting of an estimated characteristic value to reduce a difference between the estimated characteristic value and a characteristic target value of a characteristic target datum, the estimated characteristic value being based on a composition of a multitude of structural similarities of the physical target structure datum with each physical structure datum from the multitude of physical structure data; and estimating (S470) a characteristic datum of a physical target structure using the estimation model,where the characteristic datum includes an acceleration / time curve, and the acceleration / time curve is to be used for the design of the physical target structure instead of a physical crash test.

Description

TECHNICAL AREA

[0001] The present invention relates to predictions of target characteristic values. BACKGROUND

[0002] Computer-aided engineering (CAE) has been used for a variety of manufacturing industries, such as vehicles and electronic devices.

[0003] For example, the design of vehicle hoods is one application of CAE. Strict regulations govern hood designs to ensure compliance with safety standards (e.g., head-hood impact injury assessments). These assessments are calculated from acceleration-time curves during the impact. Acceleration-time curves can be measured using costly physical crash tests or estimated from relatively inexpensive computer simulations of collisions. However, due to the significant memory and processing power required to simulate a collision, estimating an acceleration-time curve from a computer simulation is currently very time-consuming.

[0004] Therefore, there is a need in the field of technology to address the aforementioned problem. QUICK OVERVIEW

[0005] Starting from a first aspect, the present invention provides a device for predicting characteristic data, the device comprising: a processor; and one or more computer-readable media which collectively contain instructions which, when executed by the processor, cause the processor to: obtain a plurality of data of a physical structure and a plurality of characteristic data, wherein each physical structure corresponds to characteristic data from the plurality of characteristic data and the characteristic data contain a plurality of characteristic values, each characteristic value relating to a physical structure which corresponds to a data value of the physical structure from the plurality of data of the physical structure; estimate at least one structural similarity between at least two physical structures which correspond to data of the physical structure from the plurality of data of the physical structure;and generating an estimation model for estimating target characteristics from target data of the physical structure using at least one characteristic value and according to at least one structural similarity between the target data of the physical structure and all data values ​​of the physical structure from the plurality of data of the physical structure.

[0006] Starting from a further aspect, the present invention provides a computer-implemented method for predicting characteristic data, wherein the method comprises: obtaining a plurality of data of a physical structure and a plurality of characteristic data, wherein each physical structure corresponds to characteristic data from the plurality of characteristic data and the characteristic data contain a plurality of characteristic values, wherein each characteristic value relates to a physical structure, the data of a physical structure corresponds to the plurality of data of the physical structure;Estimating at least one structural similarity between at least two physical structures, corresponding to data of the physical structure from the plurality of data of the physical structure, and generating an estimation model for estimating a target data value from a target data value of the physical structure by using at least one characteristic value and according to at least one structural similarity between the target data of the physical structure and each data value of the plurality of data of the physical structure.

[0007] Starting from another aspect, the present invention provides a computer program product for predicting characteristic data, wherein the computer program product has a computer-readable storage medium that can be read by a processing circuit and stores instructions for execution by the processing circuit in order to execute a method for carrying out the steps of the invention.

[0008] From another perspective, the present invention provides a computer program that is stored in a computer-readable medium and can be loaded into the internal memory of a digital computer, and includes software code sections for performing the steps of the invention when the program is run in a computer.

[0009] Therefore, one aspect of the innovations mentioned here is to predict target characteristics in such a way as to overcome the aforementioned disadvantages associated with the prior art. The aforementioned and further objectives can be achieved through combinations described in the claims. A first aspect of the innovations may be a device comprising a processor and one or more computer-readable storage media that collectively contain instructions. When executed by the computer, the instructions can cause the processor to receive a plurality of physical structure data and a plurality of characteristic values, wherein each physical structure corresponds to a characteristic value from the plurality of characteristic values, and the characteristic values ​​comprise a plurality of characteristic values, each characteristic value relating to a physical structure.which corresponds to a data value of the physical structure from the plurality of data of the physical structure, estimate at least one structural similarity between at least two physical structures, which correspond to data of the physical structure from the plurality of data of the physical structure, and generate an estimation model for estimating a target characteristic value from a target data value of the physical structure, wherein at least one characteristic value is used and according to at least one structural similarity between the target data of the physical structure and each data value from the plurality of data of the physical structure. According to a first aspect of the innovations, the device can directly estimate characteristic data such as an acceleration / time curve from data of the physical structure without simulating the collision.

[0010] The first aspect may also be a method implemented by a computer that performs the operations of the device, or a computer program product that has a computer-readable storage medium containing program instructions, wherein the program instructions are executable to perform the operations of the device.

[0011] The summary section does not necessarily describe all features of the embodiments of the present invention. The present invention may also be a partial combination of the features described above. The above and further features and advantages of the present invention will become clearer from the following description of embodiments, which is given in conjunction with the attached drawings. List of characters

[0012] The present invention will now be described only by way of example with reference to preferred embodiments, which are illustrated in the following figures: Fig. Figure 1 shows an application for predicting target characteristic data according to an embodiment of the present invention. Fig. Figure 2 shows an acceleration / time curve estimated by predictions of target parameters according to an embodiment of the present invention. Fig. Figure 3 shows an exemplary configuration of a device 100 according to an embodiment of the present invention. Fig. Figure 4 shows a functional sequence of a device according to an embodiment of the present invention. Fig. 5 shows an example of data X i a physical structure according to an embodiment of the present invention. Fig. Figure 6 shows an example of key data Y iaccording to one embodiment of the present invention. Fig. Figure 7 shows an example of first learning according to an embodiment of the present invention. Fig. Figure 8 shows an example of φ(X) i ), according to an embodiment of the present invention. Fig. 9 shows an example of V(X i , X n ) according to an embodiment of the present invention. Fig. Figure 10 shows an example of a response surface of an acceleration / time curve according to an embodiment of the present invention. Fig. Figure 11 shows a computer according to an embodiment of the present invention. DETAILED DESCRIPTION

[0013] Some embodiments of the present invention are described below. These embodiments do not limit the invention to the claims, and the combinations of features described in the embodiments are not necessarily essential for means provided by aspects of the invention.

[0014] Fig. Figure 1 shows an application for predicting target characteristics according to an embodiment of the present invention. A device for predicting target characteristics can generate an estimation model for estimating target characteristics from target data of a physical structure. The estimation model can be generated using training data of physical structures that have measured characteristics. Each physical structure in the training data can represent a part of the body of a moving object, such as a vehicle hood. 12 a vehicle 10 , which in Fig. 1 is shown.

[0015] The vehicle hood 12 has in the embodiment of Fig. 1. 20 points, wherein data on the physical structure of these 20 points can be input into a device for predicting target characteristics, and it can output target characteristics for these 20 points. One of the 20 points is in Fig. 1 as point P i specified. In one embodiment, the device can store data X i the physical structure of point P i use and key data Y i of point P i spend.

[0016] Fig. Figure 2 shows an acceleration / time curve used to predict target characteristics according to an embodiment of the present invention. A device for predicting target characteristics can estimate an acceleration / time curve by using training data such as that shown in Figure 2. Fig. The acceleration / time curve shown is shown in Figure 2. The device can, for example, generate the acceleration / time curves during a collision for the 20 points of Fig. Output 1 based on the estimation model without performing the physical crash test or computer simulation.

[0017] Fig. Figure 3 shows a block diagram of a device 100 according to one embodiment of the present invention. The device 100 The device can generate an estimation model and predict target parameters based on this model. 100 It can comprise a processor and one or more computer-readable media that collectively contain instructions. When executed by the processor, these instructions can cause the processor to operate as a plurality of operating sections. Therefore, the device can 100 to be considered as if they have a section. 110Calculate a section 130 , create a section 150 and a section on predictions 170 exhibits.

[0018] In the section Receive 110 can be a plurality of data of the physical structure, such as the data of the physical structure of the vehicle hood 12, which is in Fig. 1 is shown, and a number of key data such as those in Fig. The acceleration / time curve shown in section 2 can be obtained as training data. 110 can create the section 150 the majority of data from the physical structure and the Calculate section 130 The majority of key data will be provided as training data.

[0019] In the section Receive 110 Furthermore, new data on the physical structure can be obtained and added to the predictions section. 170The new target data of the physical structure will be provided for the prediction of the estimation model.

[0020] In the Calculate section 130 A characteristic similarity between a first characteristic value and a second characteristic value can be calculated from the majority of characteristic values.

[0021] In the Calculate section 130 One or more of the characteristic similarities can be calculated from a plurality of pairs of characteristics. (See the Calculation section.) 130 can create the section 150 The calculated characteristic similarities are provided.

[0022] In the section Create 150 Can an estimation model be generated to estimate a target data value from a target data value of the physical structure? The section "Generation" 150 can determine a section first 152and a section on second determination 154 exhibit.

[0023] In the section "First Determination" 152 An initial learning process can be performed to determine a similarity function that estimates the similarity between two physical structures based on the two data values ​​of the physical structure.

[0024] In the section "First Determination" 152 Furthermore, at least one structural similarity between at least two physical structures can be estimated, corresponding to the data of the physical structure from the majority of data of the physical structure, based on the similarity function determined by the initial learning. (See the section "Initial Determination") 152 can determine the second section 154 One or more estimated structural similarities will be provided.

[0025] In the section on second determination 154A second learning process can be performed to generate the estimation model. The estimation model may include weighting and sensitivity, and this is discussed in the second determination section. 154 The weighting and sensitivity can be optimized during the second learning phase. (See the section on second assessment.) 154 can be found in the predictions section 170 The generated estimation model will be provided.

[0026] In the section on predictions 170 A target characteristic value of a new physical target structure can be estimated using the estimation model. In one embodiment, the predictions section... 170Estimation based on the estimation model is performed by using at least one structural similarity between the target data of the physical structure and all data from the majority of the physical structure's data in the training data. In one embodiment, predictions can be made in the section 170 The new target data of the physical structure are entered, and the characteristic data corresponding to the new target data of the physical structure are estimated based on the estimation model.

[0027] As described above, the device can 100 Key performance indicators are estimated based on data from the physical structure by performing first and second learning without actually generating a computer simulation of the physical structure data, thereby saving costs and time in manufacturing products such as vehicles.

[0028] Fig. Figure 4 shows a functional sequence of a device according to an embodiment of the present invention. The present invention describes an example in which a device such as the device 100 Operations S410 to S470 are executed, which are in Fig. 4 are shown. Fig. Figure 4 shows an example of the operating sequence of the device. 100 , which in Fig. 3 is shown, which is in Fig. 3 device shown 100 However, it is not limited to one use of this functional sequence.

[0029] First, in S410, you can find a section for getting, such as the section for getting. 110 Training data can be obtained from a memory within the device or from a database outside the device. The training data can contain data from multiple points in a physical structure, such as a vehicle hood.

[0030] In the "Receive" section, the majority of the physical structure data can be provided to the "Generate" section, and the majority of the characteristic data can be provided to the "Calculate" section.

[0031] Subsequently, in S420, you can calculate in a section such as the Calculate section. 130A characteristic similarity between a first characteristic value and a second characteristic value can be calculated from the plurality of characteristic values. In the Calculate section, all possible pairs or some of all possible pairs from the plurality of characteristic values ​​can be generated, and the characteristic similarity of the two characteristic values ​​in each pair can be calculated. In one embodiment, the Calculate section can calculate a characteristic similarity between the characteristic values ​​Y1 and Y2 (which can be denoted as S(Y1, Y2)), a characteristic similarity S(Y1, Y3), a characteristic similarity S(Y1, Y4),..., a characteristic similarity S(Y N-2 , Y N ), a characteristic similarity S(Y N-1 , Y N ) are calculated, where a variable N represents a number of training data points, such as the number of points on the vehicle hood(s).

[0032] In the Calculation section, the calculation can be performed based on at least one difference between corresponding characteristic values ​​of the first characteristic value and the second characteristic value. In one embodiment, the characteristic similarity S(Y) can be calculated in the Calculation section. i , Y j ) can be calculated by finding the Euclidean distance between the vectors Y i and Y j The calculated characteristic similarities are then provided to the "Generate" section in the "Calculate" section.

[0033] Subsequently, in S430, a section for initial determination can be performed, such as the section for initial determination. 152 An initial learning process is performed to determine the similarity function for estimating a new structural similarity between two physical structures from the data of the physical structure.

[0034] Subsequently, in the S440 "First Determination" section, a structural similarity of the physical structures corresponding to the data of the physical structure from the majority of data of the physical structure can be estimated based on the learned similarity function. This allows the estimation in the "First Determination" section to be performed based on at least one characteristic similarity between characteristic data corresponding to the at least two physical structures.

[0035] In the first determination section, the structural similarities of the pairs of vectors of the physical structure (e.g., a pair (Xs1, Xs2), a pair (Xs1, Xs3), a pair (Xs1, Xs4),..., a pair (Xs N-2 , Xs N ), a pair (Xs N-1 , Xs N )) by calculating output values ​​of the similarity function L1(Xs1, Xs2), L1(Xs1, Xs3), L1(Xs1, Xs4),..., L1(Xs N-2 , Xs N ), L1(XS N-1 , Xs N) can be estimated. In the first determination section, the estimated structural similarities can be provided to the second determination section.

[0036] Subsequently, in section S450, the second determination can be performed, such as the second determination section. 154 A second learning process is performed to determine a predictive function for estimating the characteristics of the physical structure. During the second learning process, the section "second determination" can be used. 154 Optimize an objective function as shown in formula (1): Argmin ∑ t T ∑ i N L 2 ( y it − φ ( X i ) T ω ) 2 + λ | ω | 2 where φ(·) is a function representing a type of Gaussian kernel, ω is a weighting vector containing a plurality of weighting variables, λ is a regularization term (e.g., an L2 regularization term), y it is a target characteristic value of a target characteristic value V i .

[0037] As explained above, the second determination section can be used to determine the prediction function, which outputs multiple values, such as acceleration values ​​in the acceleration / time curve of an object impacting each point. In other words, the second determination section performs a multi-label linear regression. The estimation model can then be provided to the prediction section within this second determination section.

[0038] Subsequently, in the S640 section, a new target data value of the physical structure can be obtained to predict a target characteristic value. In one embodiment, a new target data value X can be obtained in the "Get" section. i' The physical structure of a new target vehicle hood is obtained. In the "Obtain" section, the new target data of the physical structure can be provided to the "Predictions" section.

[0039] Subsequently, in S470, in the predictions section, such as the predictions section, you can 170 Key characteristics of the new physical structure are estimated using the estimation model. In one embodiment, the key characteristic value V can be predicted in the Predictions section. i' the new target data X i' the physical structure can be estimated by X i' and values ​​of the time variable t into the estimation model φ(X i ) T ω can be entered. The estimated characteristic value V i' may contain characteristic values, each corresponding to an acceleration value as a function of time t in an acceleration / time curve.

[0040] Fig. 5 shows an example of data X iof the physical structure according to an embodiment of the present invention. Each data value of the physical structure from the plurality of data of the physical structure obtained by a section obtained, such as the section obtained 110 , can include a feature that represents the physical structure of a point of a physical structure, a feature that represents the location (e.g., absolute position or relative position) of the point of the physical structure, etc., and time.

[0041] In the embodiment of Fig. 5 can the data X i of the physical structure point P i on the hood of the vehicle Fig. 1 correspond and can be represented by a vector that has scalar values ​​s i1 , s i2 ,...,s iS , p i1 , p i2 , . , p iP , and t contains. The scalar values ​​s i1 , s i2 ,...,s iScan form a shape around point P i on the vehicle hood. In one embodiment, the scalar values ​​s can be i1 , s i2 ,...,s iS These are values ​​that represent a relative height of S points (e.g., 10 points) on the vehicle hood around point P. i represent. The scalar values ​​s i1 , s i2 ,...,s iS can create a vector Xs i form.

[0042] For the scalar values ​​p i1 , p i2 ,...,p iP These could be values ​​that define the location of point P. i represent. The value of P can, in one embodiment, 3 amount to, and p i1 can be a relative location in a first dimension (x-axis) of point P i correspond to i2 can be a relative location in a second dimension (y-axis) of point P i correspond and p i3can be a relative location in a third dimension (z-axis) of point P i are equivalent to.

[0043] The scalar values ​​p i1 , p i2 , p i3 and t can have a position vector Xp i form. In addition to or instead of Xs i and / or XP i In the "Receive" section, data on the physical structure can be obtained, including values ​​that define other characteristics of point P. i represent, for example, the thickness at one or more points in the vehicle hood around point P i The scalar value t can correspond to time. Most of the time during which acceleration values ​​are obtained on the acceleration / time curve can be assigned to a variable t in Xp. i correspond. In other words, in the initial determination section, a plurality of values ​​Xp can be used. iwith different time values ​​t for each acceleration / time curve, and each value of t corresponds to the time at which an acceleration value is obtained in the acceleration / time curve.

[0044] Fig. Figure 6 shows an example of the key data Y. i Each characteristic value from the plurality of characteristics can correspond to a physical structure from the plurality of physical structures. Each characteristic value can contain at least one characteristic value from the plurality of characteristics that represents a change over time of a characteristic property. The change over time of a characteristic property can be a characteristic property relating to an impact on the corresponding physical structure (e.g., the acceleration of an object impacting a point on the physical structure) or it can represent a deformation of the corresponding physical structure.

[0045] In one embodiment, the characteristic data Y correspond i the data X i The physical structure, for example, can be any characteristic value Y i represent a vector containing a plurality of characteristic values ​​y i1 , y i2 ,..., and y iT It contains. Each characteristic value y it the key data Y i can represent a characteristic property of a physical structure that contains data X i the physical structure corresponds to the majority of data of the physical structure.

[0046] In one embodiment, the characteristic Y i the acceleration values ​​in the acceleration / time curve of point P i correspond to those that, as in Fig. 2 shows the data X i the physical structure. In this embodiment, each of the characteristic values ​​y corresponds to i1 , y i2 ,...,y iT an acceleration value of point Pi at the respective time t1, t2,...,t T .

[0047] Fig. Figure 7 shows an example of initial learning. In the initial determination section, a similarity function L can be used. {1,d} (·,·) are learned. In one embodiment, the similarity function L estimates {i,d} (Xs i , Xs j ) the structural similarity between a vector Xs i the data X i the physical structure and a vector Xs j the data X j of the physical structure. In the initial determination section, the similarity function L can be used. {1,d} (·,·) are learned so that output values ​​of the similarity function L {1,d} (Xs i , Xs j ) with the similarity of the characteristic data (Y a , Y b ) match the data (Xs a , Xs b ) correspond to the physical structure.

[0048] For initial learning, in the "First Determine" section, a plurality of vector pairs from Xs can first be generated from the plurality of data points of the physical structure in the training data, corresponding to the pairs of characteristic data generated in S420. In one embodiment, in the "First Determine" section, a pair (Xs1, Xs2), a pair (Xs1, Xs3), a pair (Xs1, Xs4),..., a pair (Xs N-2 , Xs N ), a pair (Xs N-1 , Xs N ) are generated.

[0049] Subsequently, in the first determination section, the similarity function for estimating a new structural similarity can be determined based on the at least one structural similarity. In one embodiment, the similarity function can be determined in the first determination section based on the at least one characteristic similarity between characteristic data corresponding to the at least two physical structures. In this embodiment, in the first determination section, D types of similarity function L can be determined. {1,1} (Xs i , Xs j ), L {1,2} (Xs i , Xs j ),..., L {1,D} (Xs i , Xs j ) can be determined using formula (2), which is shown below: L { 1, d } ( Xsi ,Xsj ) = ∑ cgc ,d ( Xsi ,Xsj ) ⋅ fc ,d ( Ytrain ) where c is a variable representing a perspective of similarity, g c,d(·,·) is a function that generates a division rule of the tree model based on input vectors, and f c,d (Y train ) is a function that produces two characteristic values, where these two characteristic values ​​are Xs i and Xs j correspond and are produced with regard to the perspective of similarity c.

[0050] In one embodiment, the function g evaluates c,d (Xs i , Xs j ) the similarity of Xs i and Xs j from a perspective of similarity, represented by a value of the variable c. Therefore, the function g evaluates c,d (Xs i , Xs j ) the similarity of Xs i and Xs j in different perspectives for each value of the variable c.

[0051] The function g c,d (Xs i , Xs j) can determine the similarity of a subset of variables from scalar variables in the vectors Xs i and Xs j evaluate. In one embodiment, the function g 1,d (Xs i , Xs j ) the similarity of the first three variables in the vectors Xs i and Xs j (i.e., s i1 , s i2 , s i3 in XS i and s j1 , s j2 , s j3 in XS j ) evaluate. The function g 2,d (Xs i , Xs j ) can determine the similarity of the next three variables in the vectors Xs i and Xs j (i.e., s i4 , s i5 , s i6 in XS i and s j4 , s j5 , s j6 in XS j ) evaluate. This is how the functions g evaluate 1,d (Xs i , Xs j ), g 2,d (Xs i , Xs j ),..., g C,d (Xs i , Xs j ) the vectors Xsi , Xs j in C different perspectives.

[0052] If it is determined in the embodiment that s i1 , s i2 , s i3 in XS i and s j1 , s j2 , s j3 in XS j similar, the function g 1,d (Xs i , Xs j ) Output 1 and other functions g 2,d (Xs i , Xs j ), g 3,d (Xs i , Xs j ),..., g C,d (Xs i , Xs j ) can output 0. And if it is determined based on the tree model that s i4 , s i5 , s i6 in XS i and s j4 , s j5 , s j6 in XS j similar, the function g 2,d (Xs i , Xs j ) Output 1 and other functions g 1,d (Xs i , Xs j ), g 3.d (Xs i , Xs j ),..., g C,d (Xsi , Xs j ) can output 0.

[0053] In the first determination section, functions g can be used. c,d (·,·) can be determined during the initial learning process. In the initial determination section, functions g can be found. c,d The similarity function L1(·,·) can be determined using a nonlinear model (e.g., a tree model). In the initial determination section, a neural network model can be used instead of the tree model.

[0054] The function f c,d (Y train ) first extracts Y from a plurality of characteristic values ​​of the entered training data. train (e.g. all characteristics Y1, Y2,...,Y N ) two key data values ​​(e.g. Y a and Y b ) of which two corresponding data values ​​of the physical structure (e.g. X) a and X b ) by the corresponding function g c,d (X a , X b) are considered similar, and then assesses the similarity of these two characteristic values ​​Y a and Y b .

[0055] Fig. Figure 8 shows an example of φ(X) i ). As in Fig. As shown in 8, φ(X) is i ) to create a function where the data X i The physical structure is input and a vector is output containing a plurality of values ​​based on X. i be appreciated. As in Fig. As shown in Figure 8, the vector can provide output values ​​of the similarity functions K(·,·) of an input vector V(X). i , X n ) of the data X i the physical structure and all data values ​​of the data X1,...,X N the physical structure contained in the training data. In the initial determination section, D types of functions g can be identified. c,d (·,·) and functions f c,d (Y train) generated and thereby D types of similarity functions are determined.

[0056] Fig. 9 shows an example of V(X i , X n ) according to one embodiment of the present invention. As in Fig. 9 can be shown in V(X i , X n ) X i and X n Input and output a vector containing elements of output values ​​of the functions θ1×L {1,1} (X i ,X n ), θ2×L {1,2} (X i ,X n ),..., θ D1 ×L {1,D1} (X i ,X n ) and output values ​​of the functions θ {D1+1} ×L {2,1} (X i ,X n ), θ {D1+2} ×L {2,2} (X i ,X n ),..., θ {D1+D2} ×L {2,D2} (X i ,X n ) contains.

[0057] In one embodiment, the d1-th function L {1,d1} (X i ,X n ) in V(X i , X n ) L {1,d1}(Xs i , Xs j ) correspond to formula (2).

[0058] For the {D1+d2}-th function L {2,d2} (X i ,X n ) in V(X i , X n ) it can be a function of the mean squared error or a function of the absolute error of values ​​of the position vectors Xp i and XP n act, and it can be represented as shown below: L { 2 ,d2 } ( X i ,X n ) = ( Xp i − Xp n ) 2 <?page 9=""?>

[0059] Alternatively, the function L can be used {2,d2} (X i ,X n ) are: L { 2 ,d2 } ( X i ,X n ) = | Xp i − Xp n |

[0060] Further implementations of the D2 types of functions L {2,d2} are also possible. The variables θ1, θ2,...,θ D1 , θ D1+1 ,...,θ D1+d2 represent the sensitivity of all elements in the vector V(X) i , X nIn the "Creating" section, the D1 + D2 variables θ can also be learned. As described, in the "Creating" section, D1 types of functions L can be created. {1,d1} and D2 types of functions L {2,d2} be provided.

[0061] The vector φ(X i ) can be represented by: φ ( X i ) = θ 0 exp [ − 1 / 2 × ∑ { j ,k ,d } ( θ d × L { j ,k } ( X i ,X j ) ) ]

[0062] The weighting vector ω contains weighting variables that correspond to each of the functions K(V(X). i , X n )) correspond. The weighting variables w1, w2,...,w N In one embodiment, K(V(X) correspond to i , X1)), K(V(X i , X2)),..., K(V(X i , X N )).

[0063] In the second determination section, the weighting ω and the sensitivity θ of an estimated characteristic value can be determined in order to calculate a difference between the estimated characteristic value φ(X) i ) T ω and a target value y ita target characteristic value Y i to reduce using formula (1). The function L2(·) in the objective function in formula (1) can be a loss function, such as a mean squared error function or a function of absolute error.

[0064] During the second learning phase, values ​​of the weighting vector ω and values ​​of the parameters θ0, θ1,...,θ can be determined in the section of the second determination. D1 ,..., θ {D1+D2} in φ(X i ) can be learned. In the second determination section, the weighting vector ω and values ​​of the parameters θ0, θ1,...,θ can be determined. D1 ,.., θ {D1+D2}The weighting can be determined in an alternative way. In the second determination section, the weighting can be determined using a kernel method (such as ARD kernel) and ridge regression. In other embodiments, the weighting vector can contain any number of weights.

[0065] In the second determination section, φ(X) can be determined. i ) T ω in the optimized result of formula (1) can be used as an estimation model. Since φ(X i ) T If ω represents a synthesis of a plurality of structural similarities between the target data of the physical structure and all data of the physical structure, and a synthesis of the plurality of relative positions between the target structure and all physical structures, the characteristic value can be estimated on the basis of these syntheses by the estimation model.

[0066] As above regarding the functional sequence of the Fig. 4 to Fig. As explained in section 8, the device can estimate a higher-order feature (such as the acceleration / time curve) from a lower-order feature (such as the shape of the vehicle hood) using a hierarchical model (such as first-order and second-order learning). For example, the device can predict the acceleration curve of a head-hood collision for a new vehicle hood design without actually performing the physical test or crash simulation.

[0067] Fig. Figure 10 shows an example of the action surface of the acceleration / time curve. One axis in the graphical representation of Fig. 10 corresponds to the acceleration value, another axis in the graphical representation of Fig. 10 corresponds to time, and the other axis in the graphical representation of Fig. 10 corresponds to higher parameters derived from the simulation based on the data of the physical structure. The action surface can represent the estimation model generated by the second determination section, and a curve derived from the action surface by intersecting the action surface in a plane of the time and acceleration axes represents an acceleration / time curve.

[0068] The description in relation to the Fig. 1 to Fig. Figure 10 primarily deals with a vehicle hood as a physical structure and an acceleration / time curve as characteristic data. However, other implementations are also possible.

[0069] Fig. Figure 11 shows an example configuration of a computer. 1900 according to one embodiment of the present invention. The computer 1900 According to the present invention, a CPU contains 2000 , a RAM 2020, a graphics controller 2075 and a display device 2080 , which are interconnected via a host controller 2082 are connected. The computer 1900 It also includes input / output units such as a 2030 data transmission interface and a hard disk drive. 2040 and a DVD-ROM drive 2060 , which use an input / output controller 2084 with the host controller 2082 are connected. The computer also contains common input / output units such as a ROM. 2010 and a keyboard 2050 , which are driven by an input / output chip 2070 with the input / output controller 2084 are connected.

[0070] The host controller 2082 connects the RAM 2020 with the CPU 200 and the graphics controller 2075 , which is based on the RAM 2020 accesses the data at a high transfer rate. The CPU 2000works according to programs that are in the ROM 2010 and the RAM 2020 are stored, which is how all units are controlled. The 2075 graphics controller receives image data processed by the CPU. 200 generated in a frame buffer or the like, which is located in the RAM 2020 is provided, and causes the image data to be displayed on the display device. 2080 will be displayed. Alternatively, the graphics controller can be used. 2075 a frame buffer or similar to store image data processed by the CPU 2000 be generated.

[0071] The input / output controller 2084 connects the host controller 2082 with the data transmission interface 2030 , the hard disk drive 2040 and the DVD-ROM drive 2060 , which are relatively fast input / output units. Data transfers between the data transfer interface 2030and other electronic units are accessed via a network. The hard disk drive 2040 stores programs and data that are processed by the CPU 2000 on the computer 1900 can be used. The DVD-ROM drive 2060 reads the programs or data from the DVD-ROM 2095 and provides the hard disk drive 2040 the programs or data via the RAM 2020.

[0072] The ROM 2010 and the keyboard 2050 and the input / output chip 2070 , which are relatively slow input / output units, are connected to the input / output controller 2084 connected. The ROM 2010 stores a boot program or similar that is used by the computer 1900 When activated, a program is executed that is dependent on the computer's hardware. 1900 depends. Via the keyboard 2050Text data or commands are entered by a user, and the text data or commands can be sent via RAM 2020 to the hard disk drive. 2040 be provided. The input / output chip 2070 connects a keyboard 2050 with an input / output controller 2084 and can connect various input / output units to the input / output controller via a parallel port, a serial port, a keyboard port, a mouse port, and the like. 2084 connect.

[0073] A program that is installed on the hard disk drive 2040 about the RAM 2020 The data to be stored is provided by a recording medium such as the DVD-ROM 2095 and an IC card. The program is read from the recording medium via RAM. 2020 in the computer 1900 in the hard disk drive 2040 installed and in the CPU 2000 executed.

[0074] A program that runs in the computer 1900 If installed, it can cause the computer to 1900 as a device, such as device 100 from Fig. 3. The program or module runs in the CPU 2000 , to cause the computer 1900 as a section, component or element, such as each element of the device 100 from Fig. 3 (e.g. the section Receive) 110 , the section Calculate 130 , the section Generate 150 , the section on predictions 170 and the like).

[0075] The data processing described in these programs is read into the 1900 computer, such as the device. 100 from Fig.3. To function as a section for obtaining, which is the result of the interaction between the program or module and the various types of hardware resources mentioned above. Furthermore, the device is used by performing the operation or processing data according to the computer's usage. 1900 educated.

[0076] In response to the data exchange between the computer 1900 and an external unit can, for example, the CPU 2000 run a data processing program that is in RAM 2030 was loaded to perform data processing of a data transmission interface based on the processing described in the data processing program. 2030 to instruct.

[0077] The data transmission interface 2030 reads under the control of the CPU 200the transmission data stored in the transmission buffer area provided in the recording medium, such as RAM 2020 , a hard disk drive 2040 or a DVD-ROM 2095 and transmits the read transmission data to a network or writes received data received from a network into a receive buffer area or the like provided in the recording medium. In this way, the data transmission interface 2030 Exchange transmission / reception data with the recording medium using a DMA (Direct Memory Access) method or through a configuration where the CPU 2000 the data from the recording medium or data transmission interface 2030 a transfer target reads the data into the data transfer interface. 2030or to write to the recording medium of the transfer destination in order to transfer the transmission / reception data.

[0078] Furthermore, the CPU can 2000 cause the entire file or a required part of the file to be loaded into the database's RAM. 2020 for example, it is read via DMA transfer, where the file or database has been stored on an external recording medium such as a hard disk drive. 2040 , the DVD-ROM drive 2060 (DVD-ROM 2095 ), to perform different types of processing on the data in the RAM 2020 to execute. The CPU 2000 The processed data can then be written back to the external recording medium using a DMA transfer method or similar. During this processing, the RAM can be used. 2020 It can be considered as temporarily storing the contents of the external recording medium, and therefore the RAM 2020, the external recording device and the like are collectively referred to as a storage device, a storage section, a recording medium, a computer-readable medium, etc.

[0079] Various types of information, such as different types of programs, data, tables, and databases, can be stored in the recording device for data processing. It should be noted that the CPU 2000 also part of the RAM 2020 used to perform read / write operations on the cache memory. In such an embodiment, the cache memory is considered to be located in the RAM. 2020 , located in the memory and / or in the recording medium, unless otherwise specified, as the cache memory performs part of the function of RAM 2020 executes.

[0080] The CPU 2000It can perform various types of processing on the data read from a memory, such as RAM. 2020 , including various types of operations, information processing, condition evaluation, information search / replacement, etc., as described in the present embodiment and designated by a sequence of program instructions, and writes the results back to memory such as RAM. 2020 For example, when an assessment of conditions is performed, the CPU may 2000 The process decides whether each type of variable shown in the present embodiment is greater than, less than, not less than, not greater than or equal to the other variable or constant, and if the results of judging conditions are affirmative (or negative), the process branches off to another sequence of instructions or calls a subroutine.

[0081] Furthermore, the CPU can200 search for data in a file, database, etc., on the recording medium. For example, if a plurality of entries, each with an attribute value of a first attribute that corresponds to an attribute value of a second attribute, are stored on a recording device, the CPU can 2000 Search among the plurality of entries stored in the recording medium for an entry matching conditions, whose attribute value is designated as the first attribute, and read the attribute value of the second attribute stored in the entry, thereby obtaining the attribute value of the second attribute belonging to the first attribute that satisfies the pre-defined condition.

[0082] The program or module described above can be stored on an external recording medium. Examples of such recording media include a DVD-ROM. 2095as well as an optical recording medium such as a Blu-ray disc or a CD, a magneto-optical recording medium such as an MO, a tape medium, and a semiconductor memory such as an IC card. Additionally, a recording medium such as a hard drive or RAM provided in a server system connected to a dedicated data transmission network or the internet can be used as a recording medium, thereby transferring the program to the computer. 1900 is provided via the network.

[0083] The present invention may be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium (or media) on which computer-readable program instructions are stored to induce a processor to execute aspects of the present invention.

[0084] The computer-readable storage medium can be a physical unit capable of retaining and storing instructions for use by a system for executing instructions. For example, the computer-readable storage medium can be an electronic storage unit, a magnetic storage unit, an optical storage unit, an electromagnetic storage unit, a semiconductor storage unit, or any suitable combination thereof, without limitation.

[0085] A non-exhaustive list of more specific examples of computer-readable storage media includes the following: a portable computer floppy disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact storage disk read-only memory (CD-ROM), a DVD (digital versatile disc), a memory stick, a floppy disk, a mechanically coded unit such as punched cards or raised structures in a groove on which instructions are stored, and any suitable combination thereof.A computer-readable storage medium shall not be considered, in the use herein, as volatile signals in themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., light pulses guided through a fiber optic cable), or electrical signals transmitted through a wire.

[0086] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to individual data processing units or, via a network such as the internet, a local area network, a wide area network, and / or a wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission lines, wireless transmission, routing computers, firewalls, switching units, gateway computers, and / or edge servers. A network adapter card or network interface in each data processing unit receives computer-readable program instructions from the network and forwards them for storage on a computer-readable storage medium within the respective data processing unit.

[0087] Computer-readable program instructions for executing the steps of the present invention can be assembly instructions, ISA (Instruction Set Architecture) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., as well as conventional procedural programming languages ​​such as C or similar languages. The computer-readable program instructions can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server.

[0088] In the latter case, the remotely located computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be established with an external computer (for example, via the internet using an internet service provider). In some embodiments, electronic circuits, including, for example, programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), can execute computer-readable program instructions by using state information from the computer-readable program instructions to personalize the electronic circuits to perform aspects of the present invention.

[0089] Aspects of the present invention are described herein with reference to flowcharts and / or block diagrams or diagrams of methods, devices (systems), and computer program products according to embodiments of the invention. It is pointed out that each block of the flowcharts and / or block diagrams or diagrams, as well as combinations of blocks in the flowcharts and / or block diagrams or diagrams, can be executed by computer-readable program instructions.

[0090] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a specialized computer, or another programmable data processing device to create a machine, such that the instructions executed via the processor of the computer or other programmable data processing device generate a means of implementing the functions / steps specified in the block(s) of the flowcharts and / or block diagrams or charts.

[0091] These computer-readable program instructions may also be stored on a computer-readable storage medium capable of controlling a computer, programmable data processing device, and / or other units to function in a particular manner, such that the computer-readable storage medium on which instructions are stored has a manufactured product, including instructions that implement aspects of the function / step specified in the block(s) of the flowchart and / or block diagrams or charts.

[0092] The computer-readable program instructions can also be loaded onto a computer, other programmable data processing device, or other unit to cause the execution of a series of process steps on the computer or other programmable device or other unit in order to generate a process executed on a computer, such that the instructions executed on the computer, other programmable device, or other unit implement the functions / steps specified in the block(s) of the flowcharts and / or block diagrams or charts.

[0093] The flowcharts and block diagrams or charts in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this context, each block in the flowcharts or block diagrams or charts can represent a module, a segment, or a part of instructions that includes one or more executable instructions for performing the specific logical function(s).

[0094] In some alternative implementations, the functions specified in the block may occur in a different order than shown in the figures. For example, two blocks shown consecutively may in reality be executed essentially simultaneously, or the blocks may sometimes be executed in reverse order depending on their respective functionality. It should also be noted that each block in the block diagrams or flowcharts, as well as combinations of blocks in the block diagrams or flowcharts, may be implemented by special hardware-based systems that perform the specified functions or steps, or by combinations of special hardware and computer instructions.

[0095] Although the embodiment(s) of the present invention have been described, the technical scope of the invention is not limited to the embodiment(s) described above. It is clear to a person skilled in the art that various modifications and improvements to the embodiment(s) described above can be added. Furthermore, it is evident from the scope of the claims that embodiments incorporating such modifications and improvements can be included within the technical scope of the invention.

[0096] The operations, procedures, steps, and stages of any process performed by a device, system, program, or method shown in the claims, embodiments, or representations may be performed in any order, as long as the order is not specified by "before," "before," or the like, and as long as the output of a preceding process is not used in a subsequent process. Even if the process flow is described using expressions such as "first" or "then" in the claims, embodiments, or representations, this does not necessarily mean that the process must be performed in that order.

[0097] It has become clear from the above that the embodiments of the present invention can be used to realize a device, a method and a computer program product for predicting a target characteristic value.

Claims

[1] An apparatus for predicting characteristics, the apparatus comprising: a processor; and one or more computer-readable media collectively containing instructions that, when executed by the processor, cause the processor to: Obtaining a plurality of physical structure data and a plurality of characteristics, each physical structure corresponding to characteristics from the plurality of characteristics and the characteristic including a plurality of characteristics, each characteristic relating to a physical structure that a data value of the physical structure corresponds to the physical structure from the plurality of data; estimating at least one structural similarity between at least two physical structures corresponding to physical structure data of the plurality of physical structure data; and generating an estimation model for estimating a target characteristic value from a target physical structure data value by using at least one characteristic value and according to at least one structural similarity between the target physical structure data and each physical structure data value of the plurality of physical structure data Structure. [2] The apparatus of claim 1, wherein estimating the at least one structural similarity is based on at least one characteristic similarity between characteristics corresponding to the at least two physical structures. The apparatus of claim 2, wherein the instructions further cause the processor to calculate a characteristic similarity between a first characteristic value and a second characteristic value from the plurality of characteristics, the calculating being based on at least one difference between corresponding characteristic values ​​of the first Characteristic data value and the second characteristic data value takes place. [4] Device according to one of the preceding claims, wherein each characteristic data value contains at least one characteristic value from the plurality of characteristic values, which represents a change in a characteristic property over time. [5] The apparatus of claim 4, wherein each characteristic data value includes at least one characteristic value from the plurality of characteristic values ​​that represents a change in a characteristic property related to a collision with the corresponding physical structure or represents a deformation of the physical structure. [6] The apparatus of claim 5, wherein each physical structure is a part of a body of a moving object. [7] The apparatus of any one of claims 4 to 6, wherein each physical structure data value includes an attribute representing physical structure location and time. The apparatus of any preceding claim, wherein the instructions further cause the processor to determine a similarity function for estimating a new structural similarity, the determining being based on at least one characteristic similarity between characteristics associated with the at least two physical structures correspond to. [9] The apparatus of any preceding claim, wherein the instructions further cause the processor to determine the similarity function using a tree structure model or a neural network model. [10] The apparatus of any preceding claim, wherein the instructions further cause the processor to set a weight of a determined characteristic to reduce a difference between the estimated characteristic and a target characteristic of target characteristics, the estimated characteristic being a composition of a plurality of structural similarities of the target physical structure data with all physical structure data from the plurality of physical structure data. The apparatus of claim 10, wherein the instructions further cause the processor to set a sensitivity of the plurality of structural similarities of the target physical structure data to each physical structure data of the plurality of physical structure data by a difference between the estimated characteristic and the target characteristic of the target characteristic, the estimated characteristic being further based on a composite of the plurality of relative positions between the target structure and each physical structure. [12] The apparatus of claim 11, wherein the instructions further cause the processor to determine the weight using a kernel method. [13] The apparatus of any preceding claim, wherein the instructions further cause the processor to estimate characteristics of the target physical structure using the estimation model. [14] A computer-implemented method for predicting characteristics, the method comprising: Obtaining a plurality of physical structure data and a plurality of characteristics, each physical structure corresponding to characteristics from the plurality of characteristics and the characteristic including a plurality of characteristics, each characteristic relating to a physical structure that a data value of the physical structure corresponds to the physical structure from the plurality of data; estimating at least one structural similarity between at least two physical structures corresponding to physical structure data of the plurality of physical structure data, and generating an estimation model for estimating target characteristics from a target physical structure data value using at least one characteristic value and according to at least one structural similarity between the target physical structure data and each data value of the plurality of physical structure data. The computer-implemented method of claim 14, further comprising calculating a characteristic similarity between a first characteristic value and a second characteristic value from the plurality of characteristics, the calculating being based on at least one difference between corresponding characteristic values ​​of the first characteristic value and the second characteristic value based. [16] A computer-implemented method according to any one of claims 14 to 15, wherein each characteristic data value includes at least one characteristic value from the plurality of characteristic values ​​which represents a change in a characteristic property over time. [17] A computer-implemented method according to claim 14, wherein each characteristic value includes at least one characteristic value from the plurality of characteristic values ​​that represents a change in a characteristic property related to a collision with the corresponding physical structure or a deformation of the corresponding physical structure . [18] Computer program product for predicting characteristics, the computer program product comprising: a computer-readable storage medium readable by processing circuitry and storing instructions for execution by the processing circuitry to carry out a method according to any one of claims 14 to 17. [19] A computer program, stored on a computer-readable medium and loadable into the internal memory of a digital computer, comprising portions of software code for performing the method of any one of claims 14 to 17 when the program is run on a computer.