Door sealing model parameter determination method, device, equipment, medium and program product

Abstract

The present invention relates to the technical field of computer-aided design, and provides a method, device, equipment, medium and program product for determining parameters of a door sealing model, the method comprising: obtaining multiple candidate parameter combinations; each candidate parameter combination is obtained after optimizing multiple constitutive parameters of the door sealing model; determining the optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with the best evaluation value; the effective parameter combination is a candidate parameter combination whose functional relationship between constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model. The present invention obtains multiple candidate parameter combinations by using an optimization algorithm multiple times, and then determines the optimal constitutive parameters of the door sealing model by using the parameter combination that satisfies the preset inequality relationship and has the best evaluation value among the multiple candidate parameter combinations, thereby overcoming the problem of reduced benchmarking accuracy due to solving the inequality constraint during the benchmarking of the door sealing model.

Description

Technical Field

[0001] The present invention relates to the field of computer-aided design technology, and in particular to a method, device, equipment, medium and program product for determining parameters of a door sealing model. Background Art

[0002] The door sealing model is a simulation model corresponding to the real product or real physical prototype. The process of determining the constitutive parameters of the door sealing model is also the process of benchmarking the parameters of the door sealing model with the real physical prototype.

[0003] The existing method for determining the parameters of the door sealing model is mainly to use the existing optimization algorithm. First, an initial parameter combination composed of the constitutive parameters of the door sealing model is obtained, and an initial parameter combination set is constructed. Then, the parameters of the optimization algorithm are initialized, and the initial parameter combination set is iteratively optimized using the optimization algorithm. The optimal constitutive parameters of the door sealing model are determined according to the iterative optimization results to achieve benchmarking of the door sealing model.

[0004] However, due to the performance limitation of the optimization algorithm itself that it is difficult to solve constraints with inequality relationships, when the relevant optimization algorithm technology is used for benchmarking optimization of the door sealing model, the benchmarking accuracy will decrease due to solving constraints with inequality relationships. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the related art. To this end, the present invention proposes a method for determining door sealing model parameters to ensure that the calibration accuracy does not decrease when solving the door sealing model parameters with inequality constraints.

[0006] The invention also provides a door sealing model parameter determination device, equipment, medium and program product.

[0007] A method for determining door sealing model parameters according to an embodiment of the first aspect of the present invention includes:

[0008] Obtain multiple candidate parameter combinations; each candidate parameter combination is obtained by optimizing multiple constitutive parameters of the door sealing model;

[0009] An optimal parameter combination is determined from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with the best evaluation value; the effective parameter combination is a candidate parameter combination in which the functional relationship between the constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model.

[0010] According to the method for determining the parameters of a door sealing model in an embodiment of the present invention, a plurality of candidate parameter combinations are obtained by repeatedly utilizing an optimization algorithm, and then the parameter combination that satisfies a preset inequality relationship and has the best evaluation value among the plurality of candidate parameter combinations is used to determine the optimal constitutive parameters of the door sealing model, thereby overcoming the problem of decreased calibration accuracy due to solving inequality constraints during the door sealing model calibration process.

[0011] According to an embodiment of the present invention, the obtaining of multiple candidate parameter combinations includes:

[0012] Obtaining the candidate parameter combination;

[0013] If it is determined that the first current iteration number is less than the first termination condition preset iteration number, the process returns to the step of obtaining the candidate parameter combination until the first current iteration number is equal to the first preset iteration number.

[0014] According to an embodiment of the present invention, obtaining the candidate parameter combination includes:

[0015] Acquire a first combination set; the first combination set includes a preset number of first parameter combinations; the first parameter combination is composed of a plurality of constitutive parameters of the door sealing model;

[0016] Based on the evaluation value of the first parameter combination, the first parameter combination is modified to obtain a modified parameter combination, and a modified combination set is constructed; the modified combination set includes the preset number of modified parameter combinations;

[0017] Constructing a second combination set; the second combination set includes the preset number of second parameter combinations; the second parameter combination is determined based on a parameter combination with a better evaluation value between the first parameter combination and the corresponding modified parameter combination;

[0018] Determine that the second current number of iterations is less than the second preset number of iterations, use the second combination set to update the first combination set, return the evaluation value based on the first parameter combination to correct the first parameter combination, obtain a corrected parameter combination, and construct a corrected combination set until the second current number of iterations is equal to the second preset number of iterations;

[0019] The candidate parameter combination is determined based on the second parameter combination with the best evaluation value in the second combination set.

[0020] According to one embodiment of the present invention, the evaluation value of any of the valid parameter combinations is determined based on the following method:

[0021] Constructing a constitutive parameter vector according to the multiple constitutive parameters of the effective parameter combination;

[0022] Inputting the constitutive parameter vector into a deep order reduction model to obtain the fitness output by the deep order reduction model; the deep order reduction model is trained according to the constitutive parameter vector samples and their corresponding fitness labels;

[0023] Determining an augmented function value according to the preset inequality relationship, slack variables and penalty factors;

[0024] Based on the fitness and the sum of the augmentation function value, an evaluation value of the effective parameter combination is determined.

[0025] According to one embodiment of the present invention, the fitness label corresponding to the constitutive parameter vector sample is obtained according to the following steps:

[0026] Constructing a simulation door sealing model corresponding to the constitutive parameter vector sample;

[0027] Performing a simulated door opening experiment using the simulated door sealing model to obtain a simulated force-angle curve;

[0028] Determining the fitness label corresponding to the constitutive parameter vector sample based on the area value of the closed geometric figure defined by the simulated force-angle curve and the real force-angle curve;

[0029] The real force-angle curve is obtained by performing a door opening experiment using a real physical prototype; the door opening experiment corresponds one to one with the experimental steps of the simulated door opening experiment.

[0030] According to one embodiment of the present invention, the real physical prototype includes a prototype body, an outer door body, a door body hinge part and a sealing part;

[0031] The prototype body and the outer door body are hingedly connected via the door body hinge part;

[0032] When the prototype body and the outer door body are in a locked state, pressure is applied to the sealing portion;

[0033] Accordingly, the door opening experiment includes the following steps:

[0034] Determine that the prototype body and the outer door body are in a locked state, control the outer door body to open at a preset angular velocity, and collect multiple door opening force values ​​and their corresponding door opening angle values ​​according to a preset period until the door opening angle value reaches a preset angle value;

[0035] Based on the multiple door opening force values ​​and their corresponding door opening angle values, the true force-angle curve is constructed.

[0036] According to one embodiment of the present invention, the deep order reduction model is built based on a GoogleNet network including a residual network block;

[0037] The deep order reduction model includes an input layer, a first convolutional layer, a first average pooling layer, at least one deep convolutional layer, a second average pooling layer, a random dropout layer, a fully connected layer and an output layer connected in sequence;

[0038] The deep convolution layer includes a residual network layer and a deep aggregation layer connected in sequence; the residual network layer includes at least two parallel residual network blocks;

[0039] The residual network block is a solid line residual network block and / or a dotted line residual network block; the solid line residual network block includes two second convolutional layers connected in sequence; the dotted line residual network block includes two third convolutional layers connected in sequence; the number of convolution kernels in the third convolutional layer is twice that of the second convolutional layer;

[0040] The deep aggregation layer is used to aggregate output results of the at least two parallel residual network blocks.

[0041] According to one embodiment of the present invention, when the number of the depthwise convolutional layers is more than two, a fourth convolutional layer is connected between any two adjacent depthwise convolutional layers.

[0042] A door sealing model parameter determination device according to an embodiment of the second aspect of the present invention includes:

[0043] A candidate parameter combination acquisition module is used to acquire multiple candidate parameter combinations; each candidate parameter combination is obtained by optimizing multiple constitutive parameters of the door sealing model;

[0044] An optimal parameter combination determination module is used to determine the optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with the best evaluation value; the effective parameter combination is a candidate parameter combination whose functional relationship between constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model.

[0045] According to an embodiment of the third aspect of the present invention, the electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, it implements any one of the door sealing model parameter determination methods described above.

[0046] According to the non-transitory computer-readable storage medium of the fourth aspect of the present invention, a computer program is stored thereon, and when the computer program is executed by a processor, any of the door sealing model parameter determination methods described above is implemented.

[0047] According to the computer program product of the fifth aspect of the present invention, when the computer program is executed by a processor, it implements any of the door sealing model parameter determination methods described above.

[0048] The above one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0049] By repeatedly using the optimization algorithm to obtain multiple candidate parameter combinations, and then using the parameter combination that satisfies the preset inequality relationship and has the best evaluation value among the multiple candidate parameter combinations, the optimal constitutive parameters of the door sealing model are determined, which overcomes the problem of decreased benchmarking accuracy due to solving inequality constraints during the door sealing model benchmarking process.

[0050] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 It is one of the flow charts of the method for determining the parameters of the door sealing model provided by the present invention.

[0053] Figure 2 This is an example diagram of the macroscopic constitutive structure of the door sealing model provided by the present invention.

[0054] Figure 3 It is a schematic diagram of a flow chart of obtaining candidate parameter combinations provided by the present invention.

[0055] Figure 4 It is a schematic diagram of a simulated force-angle curve and a real force-angle curve provided by the present invention.

[0056] Figure 5 It is a structural schematic diagram of the deep order reduction model provided by the present invention.

[0057] Figure 6 It is a structural diagram of the inception module in the prior art.

[0058] Figure 7 It is a schematic diagram of the process of training the deep order reduction model provided by the present invention.

[0059] Figure 8 It is the target error graph of the training deep order reduction model provided by the present invention.

[0060] Fig. 9 This is the second flow chart of the method for determining the parameters of the door sealing model provided by the present invention.

[0061] Fig.10 It is a structural schematic diagram of the door sealing model parameter determination device provided by the present invention.

[0062] Fig.11 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0063] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0064] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0065] Combine the following Figure 1-Figure 11 The present invention describes a method, device, equipment, medium and program product for determining door sealing model parameters.

[0066] Figure 1 FIG. 1 is one of the flow charts of the method for determining the parameters of the door sealing model provided by the present invention, such as Figure 1 As shown, the door sealing model parameter determination method includes but is not limited to the following steps 110 to 120.

[0067] Step 110: Acquire multiple candidate parameter combinations; each candidate parameter combination is obtained by optimizing multiple constitutive parameters of the door sealing model.

[0068] The door sealing model is a simulation model constructed based on four structural systems, including the prototype body system, the outer door system, the door hinge system and the sealing system. The real prototype and real product corresponding to the simulation door sealing model include but are not limited to the above four structural systems, and may also include other structures.

[0069] For example, if the real product corresponding to the door sealing model is a dishwasher, the dishwasher can also include a pull rope spring subsystem, a lock subsystem, a skeleton system, etc. Its core components include but are not limited to springs, ropes, pulleys, hinge walls, outer door bodies, door lock buckles, sealing strips and other elastic sealing parts, inner tanks, etc.

[0070] When constructing the door sealing model corresponding to the dishwasher, parts that do not affect the sealing performance of the dishwasher door, such as locks, springs, ropes, etc., can be omitted, ensuring that the door sealing model includes four structural systems: prototype body system, outer door system, door hinge system and sealing system.

[0071] The constitutive parameters are used to describe the response of the door seal model during the door opening and closing behavior.

[0072] In order to obtain a door sealing model with a high fidelity level, it is necessary to continuously benchmark the constitutive parameters of the door sealing model so that the door sealing model under the optimal constitutive parameter conditions can be as consistent as possible with the mechanical behavior of the actual product.

[0073] Specifically, multiple constitutive parameters of the door sealing model that need to be benchmarked and debugged are determined in advance, and several initial parameter combinations with different values ​​of the constitutive parameters are determined to obtain an initial parameter combination set. Based on any optimization algorithm, the initial parameters of the optimization algorithm are initialized, and the optimization algorithm is used to iteratively optimize the initial parameter combination set until the iterative termination condition preset by the optimization algorithm is met, and at least one candidate parameter combination is obtained. The optimization algorithm is used multiple times to optimize multiple constitutive parameters of the door sealing model, that is, multiple candidate parameter combinations are obtained.

[0074] Optionally, the optimization algorithm may be any one of Jaya Algorithm, Random Search, Bayesian Optimization, Genetic Algorithms, Simulated Annealing, Particle Swarm Optimization (PSO) and the like.

[0075] Figure 2 is an example diagram of the macroscopic constitutive structure of the door sealing model provided by the present invention, such as Figure 2As shown in the figure, a two-segment linear mathematical model is used to describe the macroscopic constitutive model of the sealing system in the door sealing model with the sealing strip as the sealing part, and the sealing strip is divided into several equal parts (such as 10 segments, 34 segments, etc.), and the force of each sealing strip is a pair of action and reaction forces, which act on the outer door body system and the prototype body system respectively. A sealing strip is selected, and the macroscopic constitutive model is used to describe the mechanical behavior of the sealing strip, where the horizontal axis is the distance between the action and reaction forces corresponding to the sealing strip, and the vertical axis is the force of the sealing strip.

[0076] By taking the distance between the action force and reaction force corresponding to the sealing strip as the ordinate of the macroscopic constitutive model, the mechanical behavior of the entire sealing strip can be described according to any segment after the sealing strip is divided into several equal segments, and the macroscopic constitutive model of the entire sealing strip can be established.

[0077] Furthermore, if Figure 2 As shown in the figure, the constitutive parameters of the door seal model are mainly described by three points, which are numbered "1", "2" and "3" from left to right. The coordinates of each point are represented by the constitutive parameters xCoord1, yCoord1, xCoord2, yCoord2, xCoord3 and yCoord3. The values ​​of xCoord1 and yCoord3 are 0, so the constitutive parameters that need to be determined by benchmarking include the key parameters yCoord1, xCoord2, yCoord2 and xCoord3.

[0078] Accordingly, any candidate parameter combination or initial parameter combination is [yCoord1, xCoord2, yCoord2, xCoord3], and the value of each key parameter is within its respective value range.

[0079] Step 120: Determine an optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with an optimal evaluation value; the effective parameter combination is a candidate parameter combination in which the functional relationship between constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model.

[0080] The preset inequality relationship refers to the inequality relationship that the functional relationship between the constitutive parameters of the door sealing model needs to satisfy based on the natural physical laws followed by the real product or the real prototype under mechanical behavior.

[0081] The preset inequality relationship can be obtained by using a real physical prototype test corresponding to the door sealing model.

[0082] The evaluation value is used to evaluate the degree of difference between the door sealing model determined based on the specific values ​​of the constitutive parameters in the candidate parameter combination and the actual prototype in responding to the mechanical behavior.

[0083] like Figure 2 As shown, the macroscopic constitutive structure in the door sealing model is described by three points labeled "1", "2" and "3". The two lines determined by points "1", "2" and "3" are always below the straight line determined by points "1" and "3", which means that the functional relationship between the key parameters of the macroscopic constitutive structure yCoord1, xCoord2, yCoord2 and xCoord3 needs to satisfy the preset inequality relationship.

[0084] At this point, the expression of the functional relationship between the key parameters is as follows:

[0085] (1)

[0086] Further, based on the expression of the functional relationship between the aforementioned key parameters, the preset inequality relationship is set to .

[0087] Specifically, after obtaining a plurality of candidate parameter combinations, it is determined whether the functional relationship between the constitutive parameters in each candidate parameter combination satisfies a preset inequality relationship, and a valid parameter combination satisfying the preset inequality relationship is determined.

[0088] For any valid parameter combination, the specific values ​​of each constitutive parameter in the valid parameter combination are input into the simulation software to simulate the response of the door sealing model corresponding to the valid parameter combination to the mechanical behavior, thereby determining the evaluation value of the effective parameter combination. According to the same method, the evaluation values ​​of all valid parameter combinations are obtained.

[0089] Then, the candidate parameter combination with the best evaluation value is selected from the valid parameter combinations as the optimal parameter combination. The specific value of the constitutive parameter in the optimal parameter combination is the optimal constitutive parameter of the door seal model. The door seal model is constructed in the simulation software with the optimal constitutive parameters, that is, the door seal model is benchmarked.

[0090] The accuracy of solving the constitutive parameters of the door sealing model with inequality constraints is improved by obtaining multiple candidate parameter combinations. In the process of obtaining multiple candidate parameter combinations, the constitutive parameters of the candidate parameter combinations of the door sealing model parameters may not satisfy the preset inequality relationship. Therefore, after obtaining multiple candidate parameter combinations, it is necessary to determine whether the constitutive parameters of the candidate parameter combinations satisfy the preset inequality relationship.

[0091] Based on the natural physical laws followed by real products or real prototypes under mechanical behavior, the functional relationship between the constitutive parameters of the door seal model often needs to satisfy the preset inequality relationship. Due to the performance limitation of the optimization algorithm itself that it is difficult to solve the constraints with inequality relationships, the relevant optimization algorithm technology will experience a decrease in benchmarking accuracy due to solving the constraints with inequality relationships when performing benchmarking optimization of the door seal model.

[0092] The method for determining the parameters of a door sealing model provided by the present invention obtains multiple candidate parameter combinations by repeatedly utilizing an optimization algorithm, and then determines the optimal constitutive parameters of the door sealing model by utilizing the parameter combination that satisfies a preset inequality relationship and has the best evaluation value among the multiple candidate parameter combinations, thereby overcoming the problem of decreased benchmarking accuracy due to solving inequality constraints during the benchmarking of the door sealing model.

[0093] Based on the above embodiment, as an optional embodiment, obtaining multiple candidate parameter combinations includes:

[0094] Obtaining the candidate parameter combination;

[0095] If it is determined that the first current iteration number is less than the first preset iteration number, the process returns to the step of obtaining the candidate parameter combination until the first current iteration number is equal to the first preset iteration number.

[0096] The first current iteration number refers to the iteration number corresponding to the candidate parameter combination obtained in each iteration when multiple candidate parameter combinations are obtained in an iterative manner.

[0097] Specifically, based on any optimization algorithm, the initial parameters of the optimization algorithm are initialized, and the optimization algorithm is used to iteratively optimize the initial parameter combination set until the iterative termination condition preset by the optimization algorithm is met, and at least one parameter combination is optimized as the candidate parameter combination obtained by the first optimization.

[0098] After obtaining the candidate parameter combination, the magnitude relationship between the first current iteration number and the first preset iteration number corresponding to the candidate parameter combination obtained by optimization is determined.

[0099] If it is determined that the first current iteration number is less than the first preset iteration number, the first current iteration number is updated, and the process returns to the step of obtaining the candidate parameter combination, and the new initial parameter combination set of the constitutive parameters is re-determined, the initial parameters of the optimization algorithm are initialized, and the optimization algorithm is used to iteratively optimize the new initial parameter combination set until the preset iteration termination condition of the optimization algorithm is met, and the candidate parameter combination corresponding to the updated first current iteration number is obtained. Each iteration will generate a candidate parameter combination, and multiple candidate parameter combinations will be obtained after multiple iterations until the updated first current iteration number is equal to the first preset iteration number.

[0100] If it is determined that the first current number of iterations is equal to the first preset number of iterations, the step of determining the optimal parameter combination from a plurality of candidate parameter combinations is entered.

[0101] For example, if the first current iteration number is 5, and the first preset iteration number is set to 5, then the first current iteration number is equal to the first preset iteration number, and the optimization algorithm has been used for 5 optimizations to obtain multiple different candidate parameter combinations when the first current iteration number is iteration number 1 to 5. Then, it is determined whether the functional relationship between the constitutive parameters in each candidate parameter combination satisfies the preset inequality relationship, and among the candidate parameter combinations that satisfy the preset inequality relationship, the candidate parameter combination with the best evaluation value is selected as the optimal parameter combination.

[0102] For another example, if the first current iteration number is 5 and the first preset iteration number is set to 10, and the first current iteration number is less than the first preset iteration number, the first current iteration number is updated to iteration number 6, and the process returns to the step of obtaining candidate parameter combinations until the iteration number is equal to 10, then 10 candidate parameter combinations are obtained, and the process proceeds to the step of determining the optimal parameter combination from the multiple candidate parameter combinations obtained 10 times.

[0103] The door sealing model parameter determination method provided by the present invention obtains multiple candidate parameter combinations in an iterative manner, and then determines the optimal parameter combination from the multiple candidate parameter combinations. While overcoming the problem of reduced benchmarking accuracy due to solving inequality constraints during the door sealing model benchmarking process, it can also further improve the accuracy of the door sealing model benchmarking.

[0104] Based on the above embodiment, as an optional embodiment, obtaining the candidate parameter combination includes:

[0105] Acquire a first combination set; the first combination set includes a preset number of first parameter combinations; the first parameter combination is composed of a plurality of constitutive parameters of the door sealing model;

[0106] Based on the evaluation value of the first parameter combination, the first parameter combination is modified to obtain a modified parameter combination, and a modified combination set is constructed; the modified combination set includes the preset number of modified parameter combinations;

[0107] Constructing a second combination set; the second combination set includes the preset number of second parameter combinations; the second parameter combination is determined based on a parameter combination with a better evaluation value between the first parameter combination and the corresponding modified parameter combination;

[0108] Determine that the second current number of iterations is less than the second preset number of iterations, use the second combination set to update the first combination set, return the evaluation value based on the first parameter combination to correct the first parameter combination, obtain a corrected parameter combination, and construct a corrected combination set until the second current number of iterations is equal to the second preset number of iterations;

[0109] The candidate parameter combination is determined based on the second parameter combination with the best evaluation value in the second combination set.

[0110] The second preset number of iterations is determined based on the maximum number of iterations in the optimization algorithm parameters; the preset number is determined based on the population size of each generation of the population in the optimization algorithm parameters.

[0111] Specifically, Figure 3 is a schematic diagram of a process for obtaining a candidate parameter combination provided by the present invention, such as Figure 3 As shown, when obtaining a candidate parameter combination corresponding to a first current iteration number, the first current iteration number is kept unchanged, and based on the optimization algorithm, a first combination set is obtained, the first combination set includes a preset number of first parameter combinations, and each first parameter combination is composed of multiple constitutive parameters of the door sealing model.

[0112] For each first parameter combination in the first combination set, the constitutive parameter vector corresponding to the first parameter combination is input into the deep order reduction model, thereby obtaining the evaluation value of each first parameter combination in the first combination set. Based on the evaluation value of each first parameter combination, self-correction is performed to obtain a corrected parameter combination corresponding to each first parameter combination, and a corrected combination set is constructed using all corrected parameter combinations.

[0113] The constitutive parameter vector corresponding to each correction parameter combination is input into the deep order reduction model respectively, and then the evaluation value of each correction parameter combination is obtained. For each correction parameter combination, the evaluation value of the correction parameter combination is compared with the evaluation value of the corresponding first parameter combination, and the parameter combination with a better evaluation value is used as the second parameter combination, thereby constructing a second combination set consisting of the second parameter combination. It can be understood that the number of correction parameter combinations in the correction combination set, the number of first parameter combinations in the first combination set, and the number of second parameter combinations in the second combination set are all the same.

[0114] After the second parameter combination is obtained, the magnitude relationship between the second current iteration number (ie, the iteration number corresponding to when the aforementioned second parameter combination is obtained) and the second preset iteration number is determined.

[0115] If it is determined that the second current number of iterations is less than the second preset number of iterations, the first combination set is updated using the second combination set (for example, the second combination set is used as the new first combination set), the second current number of iterations is updated, and the process returns to the step of correcting the first parameter combination based on the evaluation value of the first parameter combination to obtain the corrected parameter combination, until the second current number of iterations is equal to the second preset number of iterations.

[0116] If it is determined that the second current iteration number is equal to the second preset iteration number, a second parameter combination with the best evaluation value is determined from the latest second combination set as a candidate parameter combination corresponding to the first current iteration number.

[0117] It should be noted that the methods for obtaining evaluation values ​​of parameter combinations such as candidate parameter combinations, valid parameter combinations, first parameter combinations, and second parameter combinations in the present invention are the same.

[0118] In one embodiment, the step of modifying the first parameter combination based on the evaluation value of the first parameter combination to obtain a modified parameter combination and constructing a modified combination set includes:

[0119] According to the evaluation values ​​of the first parameter combinations, determining an optimal combination having the best evaluation value and a worst combination having the worst evaluation value from the first combination set;

[0120] According to the optimal combination and the worst combination, the first parameter combination is modified to obtain a modified parameter combination corresponding to the first parameter combination;

[0121] A correction combination set consisting of the preset number of correction parameter combinations is constructed.

[0122] Optionally, the first parameter combination is modified according to the optimal combination and the worst combination, and an expression of a modified parameter combination corresponding to the first parameter combination is obtained as follows:

[0123] ; (2)

[0124] in, is the correction parameter combination corresponding to the first parameter combination, is the first parameter combination, For the optimal combination, The worst combination, and A random number between 0 and 1.

[0125] The door sealing model parameter determination method provided by the present invention, in the process of obtaining a candidate parameter combination corresponding to a first current number of iterations, firstly corrects the first parameter combination in the first combination set based on the evaluation value, and determines the corrected parameter combination using the corrected first parameter combination to construct the corrected combination set; then compares the first parameter combination and its corresponding corrected parameter combination one by one, determines the parameter combination with a better evaluation value as the second parameter combination to construct the second combination set, and thereby determines the second combination set as the first combination set after the second current number of iterations is updated. By complicating the method of generating the second combination set using the first combination set, it helps to reduce the number of iterations for obtaining the candidate parameter combination without losing the accuracy of the optimization algorithm, thereby improving the efficiency of obtaining the candidate parameter combination.

[0126] Based on the above embodiment, as an optional embodiment, the evaluation value of any of the effective parameter combinations is determined based on the following method:

[0127] Constructing a constitutive parameter vector according to the multiple constitutive parameters of the effective parameter combination;

[0128] Inputting the constitutive parameter vector into a deep order reduction model to obtain the fitness output by the deep order reduction model; the deep order reduction model is trained according to the constitutive parameter vector samples and their corresponding fitness labels;

[0129] Determining an augmented function value according to the preset inequality relationship, slack variables and penalty factors;

[0130] Based on the fitness and the sum of the augmentation function value, an evaluation value of the effective parameter combination is determined.

[0131] Specifically, the deep reduced-order model is trained in advance using constitutive parameter vector samples and their corresponding fitness labels to obtain a trained deep reduced-order model.

[0132] After obtaining the candidate parameter combination, a constitutive parameter vector is constructed according to the specific values ​​of each constitutive parameter in the candidate parameter combination, and then the constitutive parameter vector is input into the trained deep order reduction model to obtain the fitness corresponding to the candidate parameter combination.

[0133] Furthermore, slack variables and penalty factors are introduced, and the augmented function value is determined in combination with the aforementioned preset inequality relationship, and then the sum of the augmented function value and the fitness is used as the evaluation value of the candidate parameter combination.

[0134] In one embodiment, the expression of the evaluation value is as follows:

[0135] ; (3)

[0136] ; (4)

[0137] in, is the evaluation value, For fitness, is the augmented function value, is the penalty factor, To presuppose an inequality relationship, is the slack variable.

[0138] Furthermore, in the present invention, when determining that the first current number of iterations is less than the first preset number of iterations, returning to the step of obtaining the candidate parameter combination until the first current number of iterations is equal to the first preset number of iterations, can be achieved with the help of Sequential Unconstrained Minimization Technique (SUMT).

[0139] At this time, the penalty factor It is determined based on the initial penalty factor of the SUMT algorithm.

[0140] In one embodiment, the augmentation function value When the above expression (4) is used, the termination condition of the SUMT algorithm is expressed as follows:

[0141] ; (5)

[0142] in, is a smaller preset threshold.

[0143] For example, 1 / 1000, 1 / 10000.

[0144] exist When it is 1 / 1000, if the value on the left side of the inequality in expression (5) is less than 1 / 1000, it indicates that each constitutive parameter in the obtained candidate parameter combination has satisfied the preset accuracy condition. At this time, even if the first current iteration number is less than the first preset iteration number, the iteration process of the SUMT algorithm can be terminated to determine the multiple candidate parameter combinations obtained after the iteration of the SUMT algorithm.

[0145] The method for determining the parameters of the door seal model provided by the present invention, on the one hand, by pre-constructing a deep order reduction model, in the actual application process of the door seal model benchmark, the constitutive parameter vector corresponding to the candidate parameter combination is input into the deep order reduction model, and the fitness corresponding to the candidate parameter combination can be obtained. There is no need to input the constitutive parameters of the candidate parameter combination into the simulation software again and again, and there is no need to use the simulation software to simulate the door seal model to obtain the response under the mechanical behavior and then determine the fitness, which greatly shortens the time for determining the fitness. On the other hand, by introducing slack variables and penalty factors in the evaluation value, the optimization problem with inequality constraints can be directly solved.

[0146] Based on the above embodiment, as an optional embodiment, the fitness label corresponding to the constitutive parameter vector sample is obtained according to the following steps:

[0147] Constructing a simulation door sealing model corresponding to the constitutive parameter vector sample;

[0148] Performing a simulated door opening experiment using the simulated door sealing model to obtain a simulated force-angle curve;

[0149] Determining the fitness label corresponding to the constitutive parameter vector sample based on the area value of the closed geometric figure defined by the simulated force-angle curve and the real force-angle curve;

[0150] The real force-angle curve is obtained by performing a door opening experiment using a real physical prototype; the door opening experiment corresponds one to one with the experimental steps of the simulated door opening experiment.

[0151] Specifically, when starting to train the deep reduced-order model, the door opening experiment is performed in advance using the real physical prototype corresponding to the door sealing model, and the response data of the real physical prototype to the mechanical behavior during the door opening experiment is collected. Then, the real force-angle curve is constructed based on the collected data as the benchmark data for the simulated door sealing model.

[0152] After obtaining several constitutive parameter vector samples for training the deep order reduction model, for each constitutive parameter vector sample, the constitutive parameters corresponding to the constitutive parameter vector sample are input into the simulation software to obtain the corresponding simulation door sealing model.

[0153] Based on the simulation software, the simulated door sealing model is used to perform a simulated door opening experiment in which the experimental steps correspond one to one with the door opening experiment, and a simulated force-angle curve is obtained.

[0154] Finally, the area value of the closed geometric figure defined by the simulated force-angle curve and the real force-angle curve is used as the fitness label corresponding to the constitutive parameter vector sample.

[0155] Optionally, the simulation software is MotionSolve software.

[0156] It can be understood that the fitness label is determined based on the area value of the closed geometric figure defined by the simulated force-angle curve and the real force-angle curve. After the deep reduced order model is trained using the constitutive parameter vector samples and their corresponding fitness labels, when the deep reduced order model is applied, the smaller the fitness, the better the corresponding evaluation value when the augmented function value is the same.

[0157] Figure 4 Schematic diagram of the simulated force-angle curve and the real force-angle curve provided by the present invention, such as Figure 4 As shown, the blue curve is the real force-angle curve obtained by performing a door opening experiment using a real physical prototype, and the red curve is the simulated force-angle curve obtained by performing a simulated door opening experiment using a simulated door sealing model corresponding to a certain constitutive parameter vector sample.

[0158] A vertical line parallel to the y-axis (i.e., perpendicular to the x-axis) is extended from the end of the simulated force-angle curve. The closed geometric figure defined by the vertical line, the y-axis, the simulated force-angle curve and the real force-angle curve, that is, the closed geometric figure defined by the simulated force-angle curve and the real force-angle curve, is calculated. The fitness label corresponding to the constitutive parameter vector sample is obtained.

[0159] In one embodiment, the constitutive parameter vector sample is obtained according to the following steps: determine the value range of each constitutive parameter, and use a computer program to obtain multiple optional values ​​of each constitutive parameter; for each constitutive parameter, sample some parameter values ​​from its multiple optional values; combine the sampled parameter values ​​of each constitutive parameter to generate multiple parameter combinations; and use the parameter values ​​in each parameter combination to construct a constitutive parameter vector sample.

[0160] Optionally, for each constitutive parameter, a Modified Extensible Lattice Sequence (MELS) sampling method is used to sample some parameter values ​​from multiple optional values.

[0161] Since the macroscopic constitutive model of the door seal model needs to satisfy certain inequality constraints in physics, such as Figure 2 For the macroscopic constitutive structure shown, which is composed of two linear segments, the ordinate of the middle point labeled "2" must be below the ordinate corresponding to the abscissa of the middle point of the line connecting the two endpoints labeled "1" and "3".

[0162] The MELS sampling method can adapt to certain unequal constraints that the macroscopic constitutive conditions of the door seal model need to satisfy, and obtain parameter combinations that are physically easier to implement.

[0163] Take the constitutive parameters yCoord1, xCoord2, yCoord2 and xCoord3 as an example.

[0164] Firstly, the value ranges of the four constitutive parameters are determined to be 0.01-12.5 N, 0.01-90.0 mm, 0.01-12.5 N and 0.01-90.0 mm respectively.

[0165] Then, the programming method is used to obtain the optional values ​​of the four constitutive parameters within the value range. For example, for yCoord1, an optional value is determined every 0.1N within the value range, and the optional values ​​are 0.1N, 0.2N, 0.3N, ..., 12.3N, 12.4N, 12.5N. For another example, for xCoord2, an optional value is determined every 1mm within the value range, and the optional values ​​are 1mm, 2mm, 3mm, ..., 88mm, 89mm, 90mm.

[0166] Then, the MELS sampling method is used to sample some parameter values ​​from the optional values ​​of each constitutive parameter, and the sampled parameter values ​​of the four constitutive parameters are combined with each other to generate multiple parameter combinations.

[0167] Finally, the constitutive parameter vector samples are constructed using the parameter values ​​in each parameter combination. For example, if the values ​​of yCoord1, xCoord2, yCoord2, and xCoord3 in a parameter combination are 1.5N, 60mm, 5N, and 34mm, respectively, the constitutive parameter vector samples constructed are [1.5, 60, 5, 34].

[0168] The method for determining the parameters of a door sealing model provided by the present invention performs a door opening experiment on a real physical prototype in advance to obtain a real force-angle curve, and then uses a simulated door sealing model constructed according to a constitutive parameter vector sample to perform a simulated door opening experiment whose steps correspond one-to-one to the door opening experiment to obtain a simulated force-angle curve, and uses the area value of the closed geometric figure jointly defined by the two curves as the fitness label corresponding to the constitutive parameter vector sample to obtain training data for training a deep reduced order model, so that the deep reduced order model after training can directly determine the corresponding fitness according to the candidate parameter combination, thereby greatly shortening the time for determining the fitness using simulation software.

[0169] Based on the above embodiment as an optional embodiment, the real physical prototype includes a prototype body, an outer door body, a door body hinge part and a sealing part;

[0170] The prototype body and the outer door body are hingedly connected via the door body hinge part;

[0171] When the prototype body and the outer door body are in a locked state, pressure is applied to the sealing portion;

[0172] Accordingly, the door opening experiment includes the following steps:

[0173] Determine that the prototype body and the outer door body are in a locked state, control the outer door body to open at a preset angular velocity, and collect multiple door opening force values ​​and their corresponding door opening angle values ​​according to a preset period until the door opening angle value reaches a preset angle value;

[0174] Based on the multiple door opening force values ​​and their corresponding door opening angle values, the true force-angle curve is constructed.

[0175] The door opening force is the amount of resistance that the outer door needs to overcome during the opening process.

[0176] Specifically, the real physical prototype corresponding to the door sealing model is composed of components that affect the sealing performance of the opening and closing door, including a prototype body, an outer door body, a door body hinge and a sealing part, wherein the prototype body and the outer door body are connected through the door body hinge, and when the prototype body and the outer door body are in a locked state, pressure is applied to the sealing part.

[0177] Furthermore, a door opening experiment is performed using a real physical prototype. First, the prototype body and the outer door body are locked. Then, the outer door body is opened manually or by a mechanical arm at a preset angular velocity. A force sensor is used to collect multiple door opening force values ​​when the real physical prototype is opened according to a preset period. An angle sensor is used to collect door opening angle values ​​corresponding to multiple door opening force values ​​according to a preset period until the collected door opening angle value reaches a preset angle value. Based on the collected multiple door opening force values ​​and their corresponding door opening angle values, a real force-angle curve is constructed as the benchmark data for the simulated door sealing model.

[0178] Among them, the preset period can be to collect the door opening force value and the door opening angle value once every time the door is opened 1°; or it can be to collect the door opening force value and the door opening angle value once every preset time threshold (such as 0.1 seconds, 0.5 seconds, etc.).

[0179] Accordingly, the simulation door sealing model consists of four subsystems, namely, a prototype body subsystem, an outer door body subsystem, a door body hinge subsystem and a sealing subsystem. The outer door body subsystem and the prototype body subsystem are hinged through the door body hinge subsystem.

[0180] When using the simulated door sealing model to perform a simulated door opening experiment corresponding to the door opening experiment steps, the outer door body submodel of the simulated door sealing model undergoes the same timing action as the outer door body of the real physical prototype, determines that the outer door body is in a locked state, simulates opening the outer door body at the same preset angular velocity, and collects the simulated door opening force value and its corresponding simulated door opening angle value obtained by the simulated door sealing model according to the preset period, until the simulated door opening angle value between the outer door body submodel and the prototype body subsystem reaches the preset angle value. Based on the simulated door opening force value and its corresponding simulated door opening angle value, a simulated force-angle curve is constructed.

[0181] Optionally, when performing a simulated door opening experiment using the simulated door sealing model, MotionView is used as a pre-processor and MotionSolve is used as a pre- and post-processor.

[0182] Optionally, according to the force characteristics of each subsystem of the simulated door sealing model, the sealing subsystem is simulated using Force, the outer door body subsystem and the prototype body subsystem are simulated using rigid body units, and the door body hinge subsystem is set as a revolute pair.

[0183] Optionally, the position of the outer door subsystem door opening is set, and the movement speed which is always perpendicular to the outer door is set to mm / ms, and the total analysis time is 4000ms.

[0184] Optionally, a transient solver is used for solving.

[0185] The door sealing model parameter determination method provided by the present invention adopts an angle control method when using a real physical prototype to perform a door opening experiment to obtain a real force-angle curve, which can improve the accuracy of the obtained real force-angle curve, and use a simulated door opening experiment with one-to-one corresponding experimental steps to obtain a simulated force-angle curve, thereby improving the accuracy of the fitness obtained using the deep reduced-order model.

[0186] Based on the above embodiment, as an optional embodiment, the deep order reduction model is built based on a GoogleNet network including a residual network block;

[0187] The deep order reduction model includes an input layer, a first convolutional layer, a first average pooling layer, at least one deep convolutional layer, a second average pooling layer, a random dropout layer, a fully connected layer and an output layer connected in sequence;

[0188] The deep convolution layer includes a residual network layer and a deep aggregation layer connected in sequence; the residual network layer includes at least two parallel residual network blocks;

[0189] The residual network block is a solid line residual network block and / or a dotted line residual network block; the solid line residual network block includes two second convolutional layers connected in sequence; the dotted line residual network block includes two third convolutional layers connected in sequence; the number of convolution kernels in the third convolutional layer is twice that of the second convolutional layer;

[0190] The deep aggregation layer is used to aggregate output results of the at least two parallel residual network blocks.

[0191] The GoogleNet network is a deep convolutional neural network architecture that introduces the inception module. It has a multi-scale convolution kernel parallel structure, which can enhance the network's perception of features of different scales and is specifically used for image classification and feature extraction tasks.

[0192] The Residual Network (ResNet) is a deep neural network structure that is widely used in computer vision tasks such as image classification, target detection, and semantic segmentation. It solves the gradient vanishing and gradient exploding problems in the deep network training process by introducing residual connections, and can effectively deepen the depth of the network.

[0193] When using a deep reduced-order model built on a deep convolutional neural network architecture for data regression, the data input into the deep reduced-order model is a vector composed of constitutive parameters. Unlike image data, the feature dimensions of constitutive parameter data are relatively small. In order to make the GoogleNet network containing the residual network block suitable for simulation application scenarios where the constitutive parameters of the door sealing model are determined, the inception module in the GoogleNet network is improved.

[0194] Specifically, the deep order reduction model includes an input layer, a first convolutional layer, a first average pooling layer, at least one deep convolutional layer, a second average pooling layer, a random dropout layer, a fully connected layer, and an output layer connected in sequence. Among them, the deep convolutional layer is obtained by improving the inception module in the GoogleNet network.

[0195] Each deep convolutional layer consists of at least two parallel residual network blocks and a deep aggregation layer for aggregating the output results of the residual network blocks.

[0196] The types of residual network blocks include solid-line residual network blocks and dotted-line residual network blocks. The solid-line residual network block includes two second convolutional layers connected in sequence, and the dotted-line residual network block includes two third convolutional layers connected in sequence. The difference between the solid-line residual network block and the dotted-line residual network block is that the number of convolution kernels in the third convolutional layer is twice that of the second convolutional layer.

[0197] Optionally, the first convolution layer and the second convolution layer use small-sized convolution kernels (the convolution kernel size can be greater than or equal to 1×1), and the number of kernels can be used flexibly.

[0198] Figure 5 is a schematic diagram of the structure of the deep order reduction model provided by the present invention, such as Figure 5 As shown in the figure, the deep order reduction model includes two deep convolutional layers, the first deep convolutional layer includes 4 residual network blocks, specifically including 2 solid residual network blocks and 2 dotted residual network blocks. The second deep convolutional layer also includes 4 residual network blocks, specifically including 3 solid residual network blocks and 1 dotted residual network block.

[0199] It can be understood that, when the deep order reduction model includes multiple deep convolutional layers, the structure of each deep convolutional layer may be the same or different.

[0200] It can be understood that the residual network blocks connected in parallel in each deep convolutional layer can all be solid-line residual network blocks, or all be dotted-line residual network blocks, or some can be solid-line residual network blocks and some can be dotted-line residual network blocks, and the present invention does not impose any limitation on this.

[0201] Figure 6 It is a schematic diagram of the structure of the inception module in the prior art, such as Figure 6 As shown, in the prior art, the inception module includes four branches. The first branch includes a convolution layer with a convolution kernel size of 1×1, the second branch includes a convolution layer with a convolution kernel size of 1×1 and a convolution layer with a convolution kernel size of 3×3, the third branch includes a convolution layer with a convolution kernel size of 1×1 and a convolution layer with a convolution kernel size of 5×5, and the fourth branch includes a maximum pooling layer with a convolution kernel size of 3×3 and a convolution layer with a convolution kernel size of 1×1. Finally, a deep aggregation layer is used to aggregate the output results of the four branches.

[0202] The method for determining the parameters of the door sealing model provided by the present invention improves the structure of the deep order reduction model, introduces the residual network block into the inception module in the GoogleNet network, and obtains an improved deep convolutional layer. First, the deep convolutional layer takes into account that the vector composed of the constitutive parameters is a feature with fewer data feature dimensions, deletes the maximum pooling layer in the original inception module, and does not perform downsampling inside the inception module. Secondly, two types of residual network blocks, solid line residual network blocks and dotted line residual network blocks, are introduced into the deep convolutional layer to ensure the complexity of the network itself. Furthermore, outside the deep convolutional layer, considering the features with fewer data feature dimensions, an average pooling layer is used for downsampling, which ensures the performance accuracy of the deep order reduction model in the application scenario of door sealing model parameter determination with fewer data feature dimensions.

[0203] Combination Figure 5 As shown, when the number of the depth convolutional layers is more than two, a fourth convolutional layer is connected between any two adjacent depth convolutional layers.

[0204] Optionally, the fourth convolutional layer uses a small-size convolution kernel with a size greater than or equal to 1×1, and the number of kernels is used flexibly.

[0205] The door sealing model parameter determination method provided by the present invention further improves the structure of the deep order reduction model by adding a fourth convolutional layer between two deep convolutional layers. While ensuring the network complexity, it avoids the problem of a sharp increase in computational complexity caused by the stacking of deep convolutional layers.

[0206] In one embodiment, taking the constitutive parameters yCoord1, xCoord2, yCoord2 and xCoord3 as an example, it is assumed that the values ​​of the relaxation variables are 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8 and 0.9.

[0207] Table 1 is an evaluation value table of a parameter combination under different relaxation variables provided by the present invention. As shown in Table 1, when the relaxation variable is 0.3, the evaluation value is the smallest.

[0208] Table 1

[0209]

[0210] Table 2 is a table of constitutive parameter values ​​when the relaxation variable provided by the present invention is 0.3. As shown in Table 2, when the relaxation variable is 0.3, the augmented function value is the smallest, and the values ​​of each constitutive parameter obtained by solution are 1.0000, 0.6524, 0.0636 and 1.0000.

[0211] Table 2

[0212]

[0213] Table 3 is a table of constitutive parameter values ​​provided by the present invention when the relaxation variable is 0.1. As shown in Table 3, when the relaxation variable is 0.1, the values ​​of the various constitutive parameters obtained by solving are 0.9898, 0.8166, 0.0904 and 0.9771.

[0214] Table 3

[0215]

[0216] At this time, the constitutive parameters yCoord1 in this parameter combination are determined to be 1.0000, xCoord2 is determined to be 0.6524, yCoord2 is determined to be 0.0636, and xCoord3 is determined to be 1.0000.

[0217] In order to better illustrate the technical effect achieved by the method for determining the door sealing model parameters provided by the present invention, the complete process of the method for determining the door sealing model parameters is described below by taking the constitutive parameters of the door sealing model including yCoord1, xCoord2, yCoord2 and xCoord3 as an example.

[0218] Figure 7 It is a flow chart of the training deep order reduction model provided by the present invention, such as Figure 7 As shown, before determining the constitutive parameters of the door sealing model, the deep reduced-order model is first trained using the constitutive parameter vector samples and their corresponding fitness labels.

[0219] Specifically, the value ranges of the four constitutive parameters are determined, and the optional values ​​of the four constitutive parameters within the value range are obtained by programming. The MELS algorithm is used to randomly extract some parameter values ​​from the optional values ​​of each constitutive parameter, and the sampled parameter values ​​of the four constitutive parameters are combined with each other to generate multiple parameter combinations, and the parameter values ​​in each parameter combination are used to construct the constitutive parameter vector sample.

[0220] The constitutive parameters corresponding to the constitutive parameter vector samples are input into the MotionView simulation software for simulation to obtain a simulated door sealing model. The simulated door sealing model is used to perform a simulated door opening experiment to generate a simulated force-angle curve. The area value of the closed geometric figure defined by the simulated force-angle curve and the real force-angle curve is used as the fitness label corresponding to the constitutive parameter vector sample, thereby generating 200 sets of training sample data for the deep order reduction model.

[0221] A deep order reduction model is built based on the structure of the deep order reduction model provided in any of the above embodiments, and the initial learning rate of the deep order reduction model is initialized to 0.01, the target error is 1e-8, the maximum number of iterations is 800, the miniBatch is 60, the activation function is the Relu function, and the SGDM optimizer is used.

[0222] Input the constitutive parameter vector sample into the deep order reduction model for training, propagate information forward, obtain the fitness of the deep order reduction model output, compare the output fitness with the fitness label corresponding to the constitutive parameter vector sample, calculate the network error, and if the termination condition is met, the training is completed and the trained deep order reduction model is obtained. Otherwise, the error is back-propagated, and the weights and thresholds of the deep order reduction model are updated according to the optimization algorithm, and the information is propagated forward again, and this is repeated until the termination condition is met, the solution is completed, and the trained deep order reduction model is obtained.

[0223] The above framework is implemented using Matlab code.

[0224] Figure 8 is the target error graph of the training deep order reduction model provided by the present invention, such as Figure 8 As shown in the figure, the horizontal axis is the number of iterations (Epochs), and the vertical axis is the target error (Loss). The Loss error of the deep order reduction model reaches 1.6e-3 at the 75th iteration, and the convergence accuracy is very high.

[0225] The deep order reduction model trained by the present invention has the characteristics of both GoogleNet and ResNet, and has high training accuracy. In addition, the deep order reduction model has multiple deep convolutional layers similar to the inception module, which makes the deep order reduction model have parallel characteristics. With hardware support, the algorithm can be parallelized, thereby improving the efficiency of the algorithm.

[0226] Fig. 9 FIG. 2 is a flow chart of the method for determining the parameters of the door sealing model provided by the present invention. Fig. 9 As shown, the SUMT algorithm is used to solve the inequality relationship in the door sealing model parameter determination method, and the first current iteration number is determined to be iteration number 1, the first preset iteration number is iteration number 20, the allowable error is 1.0e-6, and the initial penalty factor μ is 0.02.

[0227] The Jaya algorithm is selected as the optimization algorithm, and the evaluation value function of the Jaya algorithm is defined in advance as the above expression (3). The algorithm parameters of the Jaya algorithm are initialized, and the population size (i.e., the preset number) is set to 10, the maximum number of iterations (i.e., the second preset number of iterations) is set to 10, the population individual length is set to 4, etc. Based on the initialized algorithm parameters, an initial population is generated, i.e., a first combination set is generated.

[0228] When the constitutive parameters include yCoord1, xCoord2, yCoord2, and xCoord3, the length of the population individuals is the number of constitutive parameters in the first parameter combination, that is, 4.

[0229] The constitutive parameter vector corresponding to each first parameter combination in the first combination set is input into the trained deep order reduction model, and the fitness is output.

[0230] Substitute the fitness into the evaluation value function to solve the evaluation value of each first parameter combination. Determine the best combination with the best evaluation value and the worst combination with the worst evaluation value from the first combination set, traverse all the first parameter combinations in the first combination set and make corrections, use the corrected first parameter combination as the corrected parameter combination, and construct a corrected combination set.

[0231] The evaluation values ​​of the first parameter combination and the revised parameter combination corresponding to the first combination set and the revised combination set are determined, and the parameter combination with a better evaluation value is used as the second parameter combination to construct the second combination set.

[0232] Determine whether the current number of iterations is equal to the maximum number of iterations 10. If it is less, take the second combination set as the new first combination set, update the current number of iterations, and return to the step of obtaining the evaluation value of the first parameter combination until the current number of iterations is equal to the maximum number of iterations 10. If it is equal, take the second parameter combination with the best evaluation value in the second combination set as the candidate parameter combination corresponding to the first current number of iterations.

[0233] It is determined that the first current number of iterations (i.e., iteration number 1) is less than the first preset number of iterations (i.e., iteration number 20), the first current number of iterations is updated to iteration number 2, and the process returns to the step of obtaining candidate parameter combinations, based on the Jaya algorithm, obtaining the candidate parameter combinations corresponding to the iteration number 2, until the first current number of iterations is equal to the first preset number of iterations.

[0234] If the first current iteration number is equal to the first preset iteration number, the candidate parameter combination that satisfies the preset inequality relationship and has the best evaluation value is determined from the 20 candidate parameter combinations obtained by the iteration as the optimal parameter combination, and the optimal constitutive parameters yCoord1, xCoord2, yCoord2 and xCoord3 of the door sealing model are determined based on the optimal parameter combination.

[0235] Among them, the above entire framework is implemented using Matlab code.

[0236] After obtaining the four optimal constitutive parameters yCoord1, xCoord2, yCoord2 and xCoord3 based on the Jaya-SUMT algorithm, the four optimal constitutive parameters are substituted into the MotionView calculation, and the area jointly defined by the simulated force-angle curve and the real force-angle curve is 5.455N. The area jointly defined by the simulated force-angle curve and the real force-angle curve obtained by the improved deep order reduction model is 5.186N, with an error of about 4.93%. It can be seen that the present invention can obtain the key parameters of the macroscopic constitutive model of the door seal model that basically meet the requirements of engineering design.

[0237] Fig.10 is a schematic diagram of the structure of the door sealing model parameter determination device provided by the present invention, such as Fig.10 As shown, the door sealing model parameter determination device includes but is not limited to a candidate parameter combination acquisition module 1001 and an optimal parameter combination determination module 1002 .

[0238] The candidate parameter combination acquisition module 1001 is used to acquire multiple candidate parameter combinations; each candidate parameter combination is obtained by optimizing multiple constitutive parameters of the door sealing model;

[0239] The optimal parameter combination determination module 1002 is used to determine the optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with the best evaluation value; the effective parameter combination is a candidate parameter combination in which the functional relationship between the constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model.

[0240] It should be noted that the door sealing model parameter determination device provided by the present invention can execute the door sealing model parameter determination method described in any of the above embodiments during specific operation, which will not be elaborated in this embodiment.

[0241] The door sealing model parameter determination device provided by the present invention obtains multiple candidate parameter combinations by repeatedly utilizing an optimization algorithm, and then determines the optimal constitutive parameters of the door sealing model by utilizing the parameter combination that satisfies a preset inequality relationship and has the best evaluation value among the multiple candidate parameter combinations, thereby overcoming the problem of decreased benchmarking accuracy due to solving inequality relationship constraints during the door sealing model benchmarking process.

[0242] Fig.11 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Fig.11 As shown, the electronic device may include: a processor (Processor) 1110, a communication interface (Communications Interface) 1120, a memory (Memory) 1130 and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 may call the logic instructions in the memory 1130 to execute the door sealing model parameter determination method provided in any of the above embodiments, and the door sealing model parameter determination method includes but is not limited to the following steps: obtaining multiple candidate parameter combinations; each candidate parameter combination is obtained by optimizing multiple constitutive parameters of the door sealing model; determining the optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with the best evaluation value; the effective parameter combination is a candidate parameter combination whose functional relationship between constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model.

[0243] In addition, the logic instructions in the above-mentioned memory 1130 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0244] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the door sealing model parameter determination method provided in any of the above embodiments. The door sealing model parameter determination method includes but is not limited to the following steps: obtaining multiple candidate parameter combinations; each candidate parameter combination is obtained after optimizing multiple constitutive parameters of the door sealing model; determining an optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with the best evaluation value; the effective parameter combination is a candidate parameter combination whose functional relationship between constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model.

[0245] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the door sealing model parameter determination method provided in any of the above-mentioned embodiments, the door sealing model parameter determination method including but not limited to the following steps: obtaining multiple candidate parameter combinations; each candidate parameter combination is obtained by optimizing multiple constitutive parameters of the door sealing model; determining an optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with an optimal evaluation value; the effective parameter combination is a candidate parameter combination whose functional relationship between constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model.

[0246] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0247] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0248] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining parameters of a door sealing model, characterized in that: include: Obtain multiple candidate parameter combinations; each candidate parameter combination is obtained by optimizing multiple constitutive parameters of the door sealing model; Determine an optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with the best evaluation value; the effective parameter combination is a candidate parameter combination whose functional relationship between constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model; The evaluation value of any of the valid parameter combinations is determined based on the following method: According to the multiple constitutive parameters of the effective parameter combination, a constitutive parameter vector is constructed; the constitutive parameter vector is input into a deep order reduction model to obtain the fitness output by the deep order reduction model; the deep order reduction model is trained according to the constitutive parameter vector samples and their corresponding fitness labels; according to the preset inequality relationship, slack variables and penalty factors, an augmented function value is determined; based on the sum of the fitness and the augmented function value, an evaluation value of the effective parameter combination is determined; The fitness label corresponding to the constitutive parameter vector sample is obtained according to the following steps: Constructing a simulated door sealing model corresponding to the constitutive parameter vector sample; performing a simulated door opening experiment using the simulated door sealing model to obtain a simulated force-angle curve; determining the fitness label corresponding to the constitutive parameter vector sample based on the area value of the closed geometric figure defined by the simulated force-angle curve and the real force-angle curve; the real force-angle curve is obtained by performing a door opening experiment using a real physical prototype; the door opening experiment corresponds to the experimental steps of the simulated door opening experiment one by one; The real physical prototype includes a prototype body, an outer door body, a door body hinge part and a sealing part; the prototype body and the outer door body are hinged through the door body hinge part; when the prototype body and the outer door body are in a locked state, pressure is applied to the sealing part; accordingly, the door opening experiment includes the following steps: determine that the prototype body and the outer door body are in a locked state, control the outer door body to open at a preset angular velocity, and collect multiple door opening force values ​​and their corresponding door opening angle values ​​according to a preset period until the door opening angle value reaches a preset angle value; based on the multiple door opening force values ​​and their corresponding door opening angle values, construct the real force-angle curve.

2. The method for determining door sealing model parameters according to claim 1, characterized in that: The obtaining of multiple candidate parameter combinations includes: Obtaining the candidate parameter combination; If it is determined that the first current iteration number is less than the first preset iteration number, the process returns to the step of obtaining the candidate parameter combination until the first current iteration number is equal to the first preset iteration number.

3. The method for determining door sealing model parameters according to claim 2, characterized in that: The obtaining of the candidate parameter combination includes: Acquire a first combination set; the first combination set includes a preset number of first parameter combinations; the first parameter combination is composed of a plurality of constitutive parameters of the door sealing model; Based on the evaluation value of the first parameter combination, the first parameter combination is modified to obtain a modified parameter combination, and a modified combination set is constructed; the modified combination set includes the preset number of modified parameter combinations; Constructing a second combination set; the second combination set includes the preset number of second parameter combinations; the second parameter combination is determined based on a parameter combination with a better evaluation value between the first parameter combination and the corresponding modified parameter combination; Determine that the second current number of iterations is less than the second preset number of iterations, use the second combination set to update the first combination set, return the evaluation value based on the first parameter combination to correct the first parameter combination, obtain a corrected parameter combination, and construct a corrected combination set until the second current number of iterations is equal to the second preset number of iterations; The candidate parameter combination is determined based on the second parameter combination with the best evaluation value in the second combination set.

4. The method for determining door sealing model parameters according to claim 1, characterized in that: The deep order reduction model is built based on the GoogleNet network including the residual network block; The deep order reduction model includes an input layer, a first convolutional layer, a first average pooling layer, at least one deep convolutional layer, a second average pooling layer, a random dropout layer, a fully connected layer and an output layer connected in sequence; The deep convolution layer includes a residual network layer and a deep aggregation layer connected in sequence; the residual network layer includes at least two parallel residual network blocks; The residual network block is a solid line residual network block and / or a dotted line residual network block; the solid line residual network block includes two second convolutional layers connected in sequence; the dotted line residual network block includes two third convolutional layers connected in sequence; the number of convolution kernels in the third convolutional layer is twice that of the second convolutional layer; The deep aggregation layer is used to aggregate output results of the at least two parallel residual network blocks.

5. The method for determining door sealing model parameters according to claim 4, characterized in that: When the number of the depthwise convolutional layers is more than two, a fourth convolutional layer is connected between any two adjacent depthwise convolutional layers.

6. A device for determining parameters of a door sealing model, characterized in that: include: A candidate parameter combination acquisition module is used to acquire multiple candidate parameter combinations; Each candidate parameter combination is obtained by optimizing multiple constitutive parameters of the door sealing model; An optimal parameter combination determination module is used to determine an optimal parameter combination from the multiple candidate parameter combinations; the optimal parameter combination is an effective parameter combination with the best evaluation value; the effective parameter combination is a candidate parameter combination whose functional relationship between constitutive parameters satisfies a preset inequality relationship; the optimal parameter combination includes the optimal constitutive parameters of the door sealing model; The evaluation value of any of the valid parameter combinations is determined based on the following method: According to the multiple constitutive parameters of the effective parameter combination, a constitutive parameter vector is constructed; the constitutive parameter vector is input into a deep order reduction model to obtain the fitness output by the deep order reduction model; the deep order reduction model is trained according to the constitutive parameter vector samples and their corresponding fitness labels; according to the preset inequality relationship, slack variables and penalty factors, an augmented function value is determined; based on the sum of the fitness and the augmented function value, an evaluation value of the effective parameter combination is determined; The fitness label corresponding to the constitutive parameter vector sample is obtained according to the following steps: Constructing a simulated door sealing model corresponding to the constitutive parameter vector sample; performing a simulated door opening experiment using the simulated door sealing model to obtain a simulated force-angle curve; determining the fitness label corresponding to the constitutive parameter vector sample based on the area value of the closed geometric figure defined by the simulated force-angle curve and the real force-angle curve; the real force-angle curve is obtained by performing a door opening experiment using a real physical prototype; the door opening experiment corresponds to the experimental steps of the simulated door opening experiment one by one; The real physical prototype includes a prototype body, an outer door body, a door body hinge part and a sealing part; the prototype body and the outer door body are hinged through the door body hinge part; when the prototype body and the outer door body are in a locked state, pressure is applied to the sealing part; accordingly, the door opening experiment includes the following steps: determine that the prototype body and the outer door body are in a locked state, control the outer door body to open at a preset angular velocity, and collect multiple door opening force values ​​and their corresponding door opening angle values ​​according to a preset period until the door opening angle value reaches a preset angle value; based on the multiple door opening force values ​​and their corresponding door opening angle values, construct the real force-angle curve.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for determining the door sealing model parameters as described in any one of claims 1 to 5 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for determining door sealing model parameters as described in any one of claims 1 to 5 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for determining door sealing model parameters as described in any one of claims 1 to 5 is implemented.

Images