Non-Extremum Evaluation and Prediction Methods

Abstract

This invention provides a non-extreme value evaluation and prediction method for accurately evaluating and predicting product performance, comprising the following steps: S1: Establishing a model based on product performance requirements and performing at least one numerical simulation, collecting simulation result sample data; S2: Reconstructing or marking the extreme value region data of the sample in three dimensions or two dimensions to determine the breadth evaluation data, wherein the calculation method of the breadth evaluation includes at least one of shape factor and gradient mean; S3: Evaluating the performance comprehensively based on the height and breadth of the extreme value region; S4: Training the sample data with a machine learning algorithm and predicting the product performance under different conditions in real time. This invention solves the problem in the prior art of accurately evaluating and predicting numerical simulation data.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing, and in particular to a non-extreme value evaluation and prediction method. Background Technology

[0002] Numerical simulation, as a supplement to experimentation, is of great significance in the manufacturing field, especially when experiments are costly and time-consuming. However, numerical simulations often focus on extreme values, neglecting the surrounding correlated influence regions. This increases the risk of errors and hinders the accurate evaluation of product performance. Therefore, a comprehensive evaluation method is urgently needed to address the problem of extreme value evaluation in existing technologies.

[0003] In addition to evaluating real-world performance, predicting the performance of unknown products is also crucial. Current technologies largely rely on training AI models under experimental conditions, which significantly increases the amount of experimental sample data, leading to higher costs and longer cycles. In contrast, numerical simulations offer a wider range of samples and are therefore more favored. Therefore, based on the proposed comprehensive evaluation method, training with numerical simulation-based sample data is extremely important, providing a novel approach for developing new products and predicting superior performance.

[0004] In view of this, it is urgent and necessary to study methods for comprehensive evaluation and prediction of non-extreme values. Summary of the Invention

[0005] This invention provides a non-extreme value evaluation and prediction method to solve the problem that prior art cannot accurately evaluate and predict numerical simulation data.

[0006] To solve the above-mentioned technical problems, the present invention is implemented as follows:

[0007] This invention provides a non-extreme value evaluation and prediction method for realistically evaluating and predicting product performance, comprising the following steps:

[0008] S1: Establish a model based on product performance requirements and conduct at least one numerical simulation, and collect sample data of the simulation results;

[0009] S2: Reconstruct or label the extreme value region data of the sample in three dimensions to determine the breadth evaluation data. The calculation method of the breadth evaluation includes at least one of shape factor and gradient mean.

[0010] S3: Evaluate the performance by combining the height and the breadth of the extreme region;

[0011] S4: Train the sample data with a machine learning algorithm and predict the product performance under different conditions in real time.

[0012] Optionally, the at least one numerical simulation includes the same operating conditions or different operating conditions; the data acquisition method includes the same operating conditions or different operating conditions; the data acquisition method includes the same operating conditions or different operating conditions.

[0013] The acquisition method requires collecting sample data from different angles under the same color band conditions.

[0014] Optionally, the sample data includes at least one of images, text, or videos; the marking includes acquiring the area within a specific color range boundary of the image.

[0015] Optionally, the height is the maximum product performance requirement obtained through the numerical simulation, and the requirement includes at least one of stress, displacement, temperature, pressure, velocity, and concentration.

[0016] Optionally, the breadth evaluation refers to evaluating the extreme value region data of the marked samples under the same working conditions;

[0017] The breadth evaluation method includes the product of the shape factor and the gradient mean.

[0018] Optionally, the gradient mean calculation method includes the gradient mean of the equivalent circle of the extreme value region or the mean of the height value and the boundary value.

[0019] Optionally, the shape factor calculation method includes 4*π*area of ​​extreme region / (perimeter of extreme region*perimeter of extreme region).

[0020] Optionally, the machine learning includes at least one of deep learning, reinforcement learning, and swarm intelligence algorithms.

[0021] Alternatively, the method for comprehensively evaluating the performance is height value - shape factor * gradient mean.

[0022] Optionally, the three-dimensional reconstruction refers to reconstructing the surface image obtained from the numerical simulation image from different angles in order to accurately obtain the extreme value region data of the sample.

[0023] In this embodiment of the invention, the performance is comprehensively evaluated by considering both the height and breadth of the extreme value region, avoiding the one-sided influence caused by only the extreme value; the influence of shape in the breadth is considered by setting a shape factor, and the closer it is to a circle, the safer it is; the gradient mean is set to accurately measure the gradient change in the extreme value region; and the performance is predicted under different conditions by using machine learning algorithms. This method can estimate and predict extreme value regions in two-dimensional images and three-dimensional models, and solves the problem that prior art cannot accurately evaluate and predict numerical simulation data. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This describes the steps of the non-extreme value evaluation and prediction method provided in the embodiments of the present invention;

[0026] Figure 2 The training and prediction graphs of the non-extreme value evaluation and prediction method provided in the embodiments of the present invention are shown.

[0027] Figure 3 This represents the calculation of the equivalent circular gradient of the non-extreme value evaluation and prediction method provided in this embodiment of the invention; Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0030] See Figures 1 to 3 This invention provides a non-extreme value evaluation and prediction method, characterized in that it is used to realistically evaluate and predict product performance, and includes the following steps:

[0031] S1: Establish a model based on product performance requirements and conduct at least one numerical simulation, and collect sample data of the simulation results;

[0032] In the embodiments of the present invention, the numerical simulation includes multiple scales, from microscopic and mesoscopic to macroscopic, and the present invention is preferably a macroscopic finite element simulation;

[0033] The finite element simulation (FEM) specifically includes simulations in different directions such as structure, fluid, and dynamics; it also includes simulations using different methods such as finite volume.

[0034] The product performance requirements include at least one of the following: stress, displacement, temperature, pressure, velocity, and concentration. These multiple conditions are generally coupled; for example, stress and temperature coupling can yield thermal stress, which is the mechanical change in the product due to heat.

[0035] The commonly used finite element software includes, but is not limited to, ANSYS, HyperWorks, etc.

[0036] The preferred method for establishing a model is to establish a finite element / finite volume model. Establishing a finite element model requires meshing, setting loading conditions or working conditions, and debugging calculations. The data obtained is often image data.

[0037] When performing a finite element simulation, if there is a small amount of sample data, a sample database can be established by changing different angles and dimensions.

[0038] Performing multiple finite element simulations requires changing different operating conditions; when creating samples with different operating conditions, the corresponding extreme value regions can still be accurately identified.

[0039] S2: Reconstruct or label the extreme value region data of the sample in three dimensions to determine the breadth evaluation data. The calculation method of the breadth evaluation includes at least one of shape factor and gradient mean.

[0040] In this embodiment of the invention, the three-dimensional reconstruction is based on images from different angles, and the corresponding transformation matrix is ​​used to obtain the corresponding three-dimensional model.

[0041] It should be noted that, in this embodiment of the invention, two-dimensional marking refers to marking image data using existing marking software;

[0042] Existing tagging software includes labelme, labelimg, etc.

[0043] The breadth evaluation data refers to the sample data that is labeled and identified in the calculation results to indicate the required extreme value range; the height evaluation data refers to the extreme values ​​calculated in the finite element software, such as the maximum principal stress and the maximum equivalent stress.

[0044] The breadth evaluation calculation method may include shape factor or gradient mean, or both.

[0045] The shape factor is a factor that takes into account the influence of the boundary shape within the extreme value region (the desired target region);

[0046] The gradient mean refers to the average value of the performance change within the extreme value region. For example, the marked area is the maximum principal stress range, the color is red, the principal stress range is 30-40MPa, the maximum principal stress is 40MPa, then the gradient is the change from 40MPa to 30MPa. The gradient mean refers to the average value of this range.

[0047] S3: Evaluate the performance by combining the height and the breadth of the extreme region;

[0048] In this embodiment of the invention, it is inaccurate to evaluate the performance of the entire model solely by its height; a comprehensive evaluation combining both height and breadth is necessary to determine its effectiveness.

[0049] S4: Train the sample data with a machine learning algorithm and predict the product performance under different conditions in real time.

[0050] Based on the samples from the comprehensive evaluation above, a machine learning algorithm was used to train the model;

[0051] The machine learning algorithms mentioned include existing deep learning, reinforcement learning, swarm intelligence algorithms, and also YOLO, R-CNN, etc.

[0052] In this embodiment of the invention, the performance is comprehensively evaluated by considering both the height and breadth of the extreme value region, avoiding the one-sided influence caused by only the extreme value; the influence of shape in the breadth is considered by setting a shape factor, and the closer it is to a circle, the safer it is; the gradient mean is set to accurately measure the gradient change in the extreme value region; and the performance is predicted under different conditions by using machine learning algorithms. This method can estimate and predict extreme value regions in two-dimensional images and three-dimensional models, and solves the problem that prior art cannot accurately evaluate and predict numerical simulation data.

[0053] Optionally, the at least one numerical simulation includes the same working conditions or different working conditions; the acquisition method includes different angles under the same working conditions or different angles under the same color band under different working conditions.

[0054] In this embodiment of the invention, the color band refers to the color bar in the finite element simulation result image.

[0055] Optionally, the sample data includes at least one of images, text, or videos; the marking includes acquiring the area within a specific color range boundary of the image.

[0056] It should be noted that among existing labeling software, segmentation software, such as LabelMe, is preferred.

[0057] Optionally, the height is the maximum product performance requirement obtained through the numerical simulation, and the requirement includes at least one of stress, displacement, temperature, pressure, velocity, and concentration.

[0058] Preferably, the breadth evaluation refers to evaluating the extreme value region data of the marked samples under the same working conditions;

[0059] The breadth evaluation method includes the product of the shape factor and the gradient mean.

[0060] Preferably, the gradient mean calculation method includes the equivalent circular gradient mean of the extreme value region or the uniform gradient mean of the extreme value region.

[0061] It should be noted that the method for calculating the mean of the equivalent circular gradient can be used for uniform or non-uniform gradients; the non-uniform gradient is based on the gradient assumption that it is a non-uniform change, and the uniform gradient is based on the gradient assumption that it is a uniform gradient.

[0062] The equivalent circular gradient mean is calculated as the difference between the height value and the region mean. The region mean includes the sum of the area ratios of the region corresponding to the mean inside the inscribed circle and the mean outside the inscribed circle within the extreme value region. The mean inside the inscribed circle is the gradient average of the area ratio of the inscribed circle. The mean outside the inscribed circle is equivalent to the boundary value.

[0063] Assuming the gradient is non-uniform, the specific solution method for the equivalent circular gradient is as follows:

[0064] Assume the extreme region (area S, perimeter L), with an inscribed circle radius r, and an amplitude of: Fz = Max - Min.

[0065] The mean value of the non-uniform stress in the inscribed circle is Td, the area ratio of the inscribed circle to the extreme value region is B=π*r*r / S, and the mean gradient value is Ave=Max-(B*Td+(1-B)*Min).

[0066] The solution for Td can be obtained through equivalent or integral methods.

[0067] Under uniform gradient conditions, the solution method is the same as the above formula, except that the mean stress Td of the inscribed circle is the average of the sum of the height value and the boundary value of the inscribed circle.

[0068] It should be noted that the uniform gradient mean is assumed to be under the condition that the extreme value region is a uniform gradient, and it can also be used to solve non-uniform gradients.

[0069] The uniform gradient mean is calculated as the difference between the height value and the regional mean, where the regional uniform mean is the average of the sum of the height value and the boundary value.

[0070] The calculation method is as follows: Assume the height value is Max, the boundary value is Min, and the gradient mean is Ave = Max - (Max + Min) / 2;

[0071] Preferably, the method for calculating the shape factor (Xz) includes 4*π*area of ​​the extreme region S / (perimeter of the extreme region L*perimeter of the extreme region L).

[0072] Optionally, the machine learning includes at least one of deep learning, reinforcement learning, and swarm intelligence algorithms.

[0073] Preferably, the machine learning is deep learning.

[0074] Alternatively, the method for comprehensively evaluating the performance is height value - shape factor * gradient mean.

[0075] It should be noted that the comprehensive evaluation P = Max - Xz * Ave;

[0076] Optionally, the three-dimensional reconstruction refers to reconstructing the surface image obtained from the numerical simulation image from different angles in order to accurately obtain the extreme value region data of the sample.

[0077] Example 1 Comprehensive Evaluation

[0078] Assuming the extreme region is an irregular image with a non-uniform gradient, the maximum principal stress is 320 MPa, the boundary of the extreme region is 280 MPa, the ratio of the inscribed circle to the overall area is 4 / 5, and the non-uniform gradient of the inscribed circle is 15, then the gradient mean is: Ave = 320 - (0.8 * 15 + 0.2 * 280) = 12 MPa, the shape factor is 0.92, the breadth evaluation is G = 11.04, and the overall evaluation is: P = 320 - 11.04.

[0079] Example 2 Simulation Prediction

[0080] The YOLOv10 deep learning algorithm was used to identify and predict the thermal stress influence patterns of the aforementioned model. LabelMe was used for labeling, and the maximum thermally affected area was used as the recognition area for training. During model training, this invention identified and calculated the number of image pixel values ​​in the thermal stress-affected region under different temperature conditions. After training, the effectiveness of the model was monitored and verified at 200 degrees Celsius. The detection showed that the pixel value at 200℃ was 10448, the predicted value was 10530, and the accuracy rate was 99.2%.

[0081] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0082] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0083] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A non-extreme value evaluation prediction method, characterized in that, The application relates to a method for accurately evaluating and predicting numerical simulation data to truly evaluate and predict product performance, wherein the product is obtained through numerical simulation and at least includes at least one performance parameter such as stress, displacement, temperature, pressure, velocity and concentration, and the method comprises the following steps: S1: a model is established according to product performance requirements, at least one numerical simulation is carried out, and simulation results are collected as sample data; S2: three-dimensional reconstruction or two-dimensional marking is carried out on the sample extreme value region data, and a breadth evaluation data is determined, wherein the breadth evaluation includes at least one of a shape factor and a gradient mean value; S3: the performance is comprehensively evaluated according to the height of the extreme value region and the breadth; S4: the sample data is trained by using a machine learning algorithm, and the performance of the product under different conditions is predicted in real time; the height is the maximum product performance requirement obtained through the numerical simulation; the breadth evaluation refers to evaluation of the marked sample extreme value region data under the same working condition; the breadth evaluation method includes the product of the shape factor and the gradient mean value; the comprehensive evaluation method of the performance is height value-shape factor*gradient mean value.

2. The non-extremum evaluating prediction method according to claim 1, characterized in that, The at least one numerical simulation includes the same working condition or different working conditions; and the collection includes the same working condition or different working conditions; wherein the collection needs to collect sample data at different angles under the same color band condition.

3. The non-extremum evaluating prediction method according to claim 1, characterized in that, The sample data includes at least one of an image, text or video; and the marking includes acquisition of a region within a color range boundary of the image.

4. The non-extremum evaluating prediction method of claim 1, wherein, The gradient mean value calculation method includes an equivalent circle gradient of the extreme value region or a uniform gradient of the extreme value region; the calculation method of the equivalent circle gradient mean value is the difference between the height value and a region mean value, the region mean value includes the sum of the area proportion of a corresponding region of an inscribed circle mean value in the extreme value region and an inscribed circle outside mean value, the inscribed circle mean value is the gradient average of the area proportion of the inscribed circle, and the inscribed circle outside mean value is equivalent to a boundary value; the calculation method of the uniform gradient mean value is the difference between the height value and a region mean value, and the region uniform mean value is the average of the sum of the height value and a boundary value.

5. The non-extremum evaluating prediction method of claim 1, wherein, The shape factor calculation method includes 4*pi*extreme value region area / (extreme value region perimeter*extreme value region perimeter).

6. The non-extremum evaluating prediction method of claim 1, wherein, The machine learning at least includes one of deep learning, reinforcement learning and swarm intelligence algorithm.

7. The non-extremum evaluating prediction method of claim 1, wherein, The three-dimensional reconstruction refers to reconstruction of curved surface images obtained from different angles of a numerical simulation image, so as to accurately obtain the sample extreme value region data.

Images