A SPR2 invalid card expansion method based on neural network self-learning

Abstract

The invention discloses an SPR2 failure card expansion method based on neural network self-learning, comprising the following operation methods: trial-producing FDS or SPR joints; establishing a database, and conducting targeted database establishment according to the input and output settings of subsequent machine learning; establishing an artificial neural network path; building an improved machine learning architecture, first determining the failure form of a data set to be predicted, and outputting it, and then using a corresponding failure data subset for subsequent prediction and calculation; reading data and outputting SPR2 parameters, and adjusting parameter values ​​according to a pulling and shearing condition; automatically adjusting individual parameters; using an overall mode method, directly inputting a new material combination of the FDS to be tested, and finally directly outputting the failure form, predicted force displacement curve, and new material card parameter values. The invention relates to the field of hot-melt self-tapping connection technology simulation, and is an efficient and high-quality SPR2 failure card expansion method based on neural network self-learning, suitable for hot-melt self-tapping and self-piercing riveting.

Description

Technical Field

[0001] The present invention belongs to the technical field of simulation of connection of dissimilar materials, and relates to an SPR2 failure card expansion method based on neural network self-learning, and specifically to an SPR2 failure card expansion method based on neural network self-learning which is efficient and high-quality and suitable for hot-melt self-tapping and self-piercing riveting. Background Art

[0002] As the application of different metal materials on automobile bodies gradually increases, the most representative of which is the large-scale application of high-strength steel and aluminum alloy materials. In order to enable high-strength structural parts to achieve efficient and high-quality connection, traditional spot welding technology can no longer meet the new application scenarios. Therefore, the emergence of technologies such as self-pierce riveting (SPR) and hot-melt self-tapping (FDS) has effectively solved the problem of connecting dissimilar materials between steel and aluminum.

[0003] In traditional automobile collision simulations, joints between structural parts generally use rigid connections that will not fail. Accurately simulating joint failure behavior becomes a prerequisite for accurate collision simulation models.

[0004] The existing simulation methods for self-piercing rivets and hot-melt self-tapping joints mainly include the common node method, rigid unit connection method, beam unit connection method and solid unit modeling method. Among them, the simulation accuracy of the common node method, rigid unit connection method and beam unit connection method is poor, which is quite different from the actual test results; while the model unit size of the solid unit modeling method is too small, the calculation efficiency is low, and the model calculation stability is poor. Among them, the constraint-based SPR2 material card has been favored by most OEMs due to its simulation accuracy and wide applicability to working conditions.

[0005] However, several parameters that control the simulation accuracy of SPR2 material cards require a large number of experiments and fine model simulations to obtain, and after changing a material combination, all previous steps need to be started over again. Therefore, in view of the complicated steps of the parameter calibration method of traditional SPR2 material cards, the large amount of experiments and the high consumption of computing resources, an expansion method of efficient and high-quality SPR2 material cards with failures suitable for hot-melt self-tapping and self-piercing rivets has become an urgent need in the vehicle design industry, in order to reduce the steps, time, computing power and money consumption of spr2 failure cards for the development of new material combinations, and at the same time, it is conducive to improving the accuracy of subsequent vehicle collisions. Summary of the invention

[0006] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an efficient and high-quality SPR2 failure card expansion method based on neural network self-learning that is suitable for hot-melt self-tapping and self-piercing riveting.

[0007] In order to achieve the above object, the present invention adopts the following technical solution: a SPR2 invalid card expansion method based on neural network self-learning, including the following operation methods:

[0008] S1: Trial production of FDS or SPR connector;

[0009] S2: Establish a database, and conduct targeted database establishment according to the input and output settings of subsequent machine learning;

[0010] S3: Establishing artificial neural network pathways;

[0011] S4: The improved machine learning architecture is built to first determine the failure form of the data set to be predicted and output it, and then use the corresponding failure data subset for subsequent prediction and calculation;

[0012] S5: Read data and output SPR2 parameters, and adjust parameter values ​​according to the tension-shear working conditions;

[0013] S6: Automatically adjust individual parameters;

[0014] S7: How to use the overall mode: directly input the new material combination of FDS or SPR to be tested, and finally directly output the failure form, predicted force displacement curve, and new material card parameter values.

[0015] As a further description of the above technical solution: in step S2, the following process is included:

[0016] SS1: The establishment of the independent variable database includes the measurement and recording of the thickness and tensile strength of the parent material, as well as cutting the parent material and performing a quasi-static tensile test on a universal testing machine at a tensile speed of 2 mm / min, and finally recording the tensile strength of the parent material;

[0017] SS2: Perform straight shear and cross tensile tests on FDS connectors to obtain force-displacement curves. The tensile test speed is 5 mm / min, and a gasket is set in the clamping area during the shear test. The tensile process is stopped manually when the force value reaches 0.

[0018] SS3: Data processing, 5 specimens were prepared for each material combination, and the force-displacement curves of the lap shear and cross tensile tests were measured. The average of these curves was taken to form a new curve. The load data of every 0.10 mm in the load-displacement curve was used as the output data at the displacement of 0 ~ 15 mm.

[0019] As a further description of the above technical solution: in step S3, the thickness of the upper and lower plates and the tensile strength of the upper and lower plates are input values, and the load data of every 0.10 mm in the load-displacement curve is used as the output data at the displacement of 0 ~ 15 mm, of which 80% of the data set is used for learning, and the remaining 20% ​​is used for testing the trained artificial neural network, each mathematically defined neuron receives input from the previous layer of neurons, each input is multiplied by a separate weight value, and the weighted input is summed and passed to the output layer through a transfer function.

[0020] As a further description of the above technical solution: in step S3, the original database is classified according to the failure modes of the hot melt self-tapping screws. The failure modes of the hot melt self-tapping screws are divided into three types: base material fracture (BMF), screw pull-out (SOP) and screw shearing (SSF).

[0021] As a further description of the above technical solution: in step S5, for the tension-shear working condition, four parameter values ​​are adjusted, which are:

[0022] &ft—peak tensile shear force;

[0023] &dt—the corresponding value of displacement at the end;

[0024] &xit—displacement value corresponding to peak shear force;

[0025] &thick—total thickness of upper and lower plates;

[0026] &diameter—Rivet diameter.

[0027] As a further description of the above technical solution: in step S5, for the cross-stretching condition, four parameter values ​​are adjusted, which are:

[0028] &fn—Peak pull-out force;

[0029] &dn—the corresponding value of displacement at the end;

[0030] &xin—displacement value corresponding to peak pull-out force;

[0031] &thick—total thickness of upper and lower plates;

[0032] &diameter—Rivet diameter.

[0033] As a further description of the above technical solution: in step S6, when the experimental component is subjected to a load at a special angle, an automatic adjustment module is added to achieve high-precision fitting through an automatic adjustment method.

[0034] As a further description of the above technical solution: in step S6, a tensile test is carried out on the FDS or SPR connection of a specific angle load condition to obtain a standard force-displacement curve. The subsequent parameter calibration and curve calibration are based on this standard. CAE and SPR2 material cards are used to simulate the FDS or SPR material combination to be predicted. Since the &thick and &diameter parameters have been determined, they are not adjusted. The six major parameters &ft, &dt, &xit, &fn, &dn, &xin have been accurately deduced from the previous machine learning model, so they are not adjusted. The three angle-related parameters α1, α2, and α3 of the SPR2 failure card have a numerical range of 0 to 1, so they are all adjusted downward from 1, and R is introduced for the curve fitting degree. 2 Conduct quantitative assessment.

[0035] As a further description of the above technical solution: In step S6, the automatic adjustment module is specifically implemented as follows:

[0036] First, make a threshold judgment;

[0037] Then optimize and adjust the original parameters x0, y0, z0, and the optimized parameters are x1, y1, z1. The fitting degree of the original parameters is R0. 2 , the parameter fitting degree after optimization is R1 2 , the contribution of parameter x1 to the optimization of the fit is φ1, the contribution of parameter y1 to the optimization of the fit is φ2, and the contribution of parameter z1 to the optimization of the fit is φ3;

[0038] After multiple iterations, the threshold judgment module performs numerical comparison to obtain the best fitting parameter value and finally outputs the values ​​of X1, Y1, and Z1.

[0039] Due to the adoption of the above technical scheme, the beneficial effect achieved by the present invention is that it can realize high-efficiency and high-quality self-learning of neural networks in the expansion of SPR2 failure cards of hot-melt self-tapping and self-piercing rivets, so as to reduce the steps, time, computing power and money consumption for developing new material combinations for SPR2 failure cards, and finally integrate them into the whole vehicle model for simulation through the connection points, which is beneficial to improve the accuracy of subsequent whole vehicle collisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a machine learning schematic diagram of a SPR2 invalid card expansion method based on neural network self-learning proposed by the present invention;

[0041] Figure 2 It is an overall flow chart of an SPR2 invalid card expansion method based on neural network self-learning proposed by the present invention. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Reference Figure 1-2 The present invention provides the following embodiments to describe the present invention in detail:

[0044] Embodiment 1: A method for expanding SPR2 invalid cards based on neural network self-learning, including the following operation methods:

[0045] S1: Trial production of FDS or SPR connector;

[0046] S2: Establish a database, and conduct targeted database establishment according to the input and output settings of subsequent machine learning;

[0047] S3: Establishing artificial neural network pathways;

[0048] S4: The improved machine learning architecture is built to first determine the failure form of the data set to be predicted and output it, and then use the corresponding failure data subset for subsequent prediction and calculation;

[0049] S5: Read data and output SPR2 parameters, and adjust parameter values ​​according to the tension-shear working conditions;

[0050] S6: Automatically adjust individual parameters;

[0051] S7: How to use the overall mode: directly input the new material combination of FDS or SPR to be tested, and finally directly output the failure form, predicted force displacement curve, and new material card parameter values.

[0052] As a further description of the above technical solution: in step S2, the following process is included:

[0053] SS1: The establishment of the independent variable database includes the measurement and recording of the thickness and tensile strength of the parent material, as well as cutting the parent material and performing a quasi-static tensile test on a universal testing machine at a tensile speed of 2 mm / min, and finally recording the tensile strength of the parent material;

[0054] SS2: Perform straight shear and cross tensile tests on FDS connectors to obtain force-displacement curves. The tensile test speed is 5 mm / min, and a gasket is set in the clamping area during the shear test. The tensile process is stopped manually when the force value reaches 0.

[0055] SS3: Data processing, 5 specimens were prepared for each material combination, and the force-displacement curves of the lap shear and cross tensile tests were measured. The average of these curves was taken to form a new curve. The load data of every 0.10 mm in the load-displacement curve was used as the output data at the displacement of 0 ~ 15 mm.

[0056] As a further description of the above technical solution: in step S3, the thickness of the upper and lower plates and the tensile strength of the upper and lower plates are input values, and the load data of every 0.10 mm in the load-displacement curve is used as the output data at the displacement of 0 ~ 15 mm, of which 80% of the data set is used for learning, and the remaining 20% ​​is used for testing the trained artificial neural network, each mathematically defined neuron receives input from the previous layer of neurons, each input is multiplied by a separate weight value, and the weighted input is summed and passed to the output layer through a transfer function.

[0057] As a further description of the above technical solution: in step S3, the original database is classified according to the failure modes of the hot melt self-tapping screws. The failure modes of the hot melt self-tapping screws are divided into three types: base material fracture (BMF), screw pull-out (SOP) and screw shearing (SSF).

[0058] As a further description of the above technical solution: in step S5, for the tension-shear working condition, four parameter values ​​are adjusted, which are:

[0059] &ft—peak tensile shear force;

[0060] &dt—the corresponding value of displacement at the end;

[0061] &xit—displacement value corresponding to peak shear force;

[0062] &thick—total thickness of upper and lower plates;

[0063] &diameter—Rivet diameter.

[0064] As a further description of the above technical solution: in step S5, for the cross-stretching condition, four parameter values ​​are adjusted, which are:

[0065] &fn—Peak pull-out force;

[0066] &dn—the corresponding value of displacement at the end;

[0067] &xin—displacement value corresponding to peak pull-out force;

[0068] &thick—total thickness of upper and lower plates;

[0069] &diameter—Rivet diameter.

[0070] As a further description of the above technical solution: in step S6, when the experimental component is subjected to a load at a special angle, an automatic adjustment module is added to achieve high-precision fitting through an automatic adjustment method.

[0071] As a further description of the above technical solution: in step S6, a tensile test is carried out on the FDS or SPR connection of a specific angle load condition to obtain a standard force-displacement curve. The subsequent parameter calibration and curve calibration are based on this standard. CAE and SPR2 material cards are used to simulate the FDS or SPR material combination to be predicted. Since the &thick and &diameter parameters have been determined, they are not adjusted. The six major parameters &ft, &dt, &xit, &fn, &dn, &xin have been accurately deduced from the previous machine learning model, so they are not adjusted. The three angle-related parameters α1, α2, and α3 of the SPR2 failure card have a numerical range of 0 to 1, so they can all be adjusted from 1 downward, and R is introduced for the curve fitting degree. 2 Conduct quantitative assessment.

[0072] As a further description of the above technical solution: In step S6, the automatic adjustment module is specifically implemented as follows:

[0073] First, make a threshold judgment;

[0074] Then optimize and adjust the original parameters x0, y0, z0, and the optimized parameters are x1, y1, z1. The fitting degree of the original parameters is R0. 2 , the parameter fitting degree after optimization is R1 2 , the contribution of parameter x1 to the optimization of the fit is φ1, the contribution of parameter y1 to the optimization of the fit is φ2, and the contribution of parameter z1 to the optimization of the fit is φ3;

[0075] After multiple iterations, the threshold judgment module performs numerical comparison to obtain the best fitting parameter value and finally outputs the values ​​of X1, Y1, and Z1.

[0076] Embodiment 2: In a neural network self-learning, efficient and high-quality SPR2 failure card expansion method suitable for hot melt self-tapping and self-piercing riveting, the following seven main stages are mainly included:

[0077] Trial production of FDS joint (SPR joint): The upper and lower plates are made of materials and thicknesses commonly used in car bodies, including but not limited to rolled steel plates: DP590 DP780 HC220 SX780CF steel MS1180 Q345B DC03 ST280 Mn22B5 SAPH440HC380

[0078] Extruded aluminum plate: 5754 5182 5052 5083 6082 6016 6061 Cast aluminum plate: AlSi10MnMg, all adopt the best production process for FDS (SPR) connection. Dimensions are based on GB / T 3250-2007,

[0079] Establish a database and establish a targeted database based on the input and output settings of subsequent machine learning:

[0080] ① Establishment of independent variable database (measurement of base material thickness and base material tensile strength), cutting of base material according to GT 228.1-2021, such as Figure 1 , and carried out quasi-static tensile testing on a universal testing machine with a tensile speed of 2 mm / min, and finally recorded the tensile strength of the base material.

[0081] ② Perform straight shear and cross tensile tests on the FDS (SPR) connector to obtain the force-displacement curve. The tensile test conditions are as follows: tensile speed: 5 mm / min. To reduce the influence of bending moment, a gasket of corresponding thickness is set in the clamping area in the shear test experiment. To obtain a complete tensile curve, all tensile processes need to pull the force value to 0 and then stop manually.

[0082] ③Data processing: Prepare 5 specimens for each material combination. Use lap shear and cross tensile tests to measure the force-displacement curves of the lap shear and cross tensile specimens. Take the average of these curves to form a new curve. The load data of every 0.10 mm in the load-displacement curve is used as the output data at the displacement of 0 ~ 15 mm.

[0083] An artificial neural network pathway was established, with the thickness of the upper and lower plates and the tensile strength of the upper and lower plates as input values. The load data of every 0.10 mm in the load-displacement curve was used as the output data at the displacement of 0 to 15 mm. 80% of the data set was used for learning, and the remaining 20% ​​was used to test the trained artificial neural network. Each mathematically defined neuron received input from the neurons in the previous layer, and each input was multiplied by a separate weight value. The weighted inputs were summed and passed to the output layer through the transfer function.

[0084] Through experiments, it is found that the failure modes of hot-melt self-tapping screws can be divided into three types: base material fracture (BMF), screw pull-out (SOP) and screw shearing (SSF). Considering the completely different trends of force-displacement curves shown by different failure modes, the original database needs to be classified according to the failure mode, otherwise the accuracy of the machine learning results will be greatly reduced.

[0085] The improved machine learning architecture is built to first determine the failure form of the data set to be predicted and output it, and then use the corresponding failure data subset for subsequent predictions and calculations.

[0086] Read the graph (table) and output SPR2 parameters through:

[0087] For the tension-shear working condition, four parameter values ​​are mainly adjusted, namely:

[0088] &ft—peak tensile shear force;

[0089] &dt—the corresponding value of displacement at the end;

[0090] &xit—displacement value corresponding to peak shear force;

[0091] &thick—total thickness of upper and lower plates (can be directly obtained);

[0092] &diameter—Rivet diameter (fixed value).

[0093] These parameter values ​​can be realized by reading the (predicted) force-displacement curve obtained by machine learning, because the force-displacement curve is actually the isolated force points corresponding to every 0.1mm, and &ft (peak force) is the largest value of these force points, &dt (corresponding value of displacement at the end) is the displacement value corresponding to the force value of 0, and &xit (displacement value corresponding to peak force) is the displacement value corresponding to the maximum force value, which can all be obtained from reading the graph of the force-displacement curve.

[0094] For the cross-stretching condition, four parameter values ​​are mainly adjusted, namely:

[0095] &fn—Peak pull-out force;

[0096] &dn—the corresponding value of displacement at the end;

[0097] &xin—displacement value corresponding to peak pull-out force;

[0098] &thick—total thickness of upper and lower plates (can be directly obtained);

[0099] &diameter—Rivet diameter (fixed value).

[0100] Similarly, &fn (peak force) is the largest value among these force points, &dn (corresponding value of displacement at the end) is the displacement value corresponding to when the force value is 0, and &xin (displacement value corresponding to peak force) is the displacement value corresponding to when the force value is maximum, all of which can be obtained from reading the force-displacement curve.

[0101] Automatically adjust individual parameters. Considering that the SPR2 failed card still has three angle-related parameters, α1, α2, and α3, which do not affect the prediction accuracy in the cross-tension and shear tests, they can be ignored. However, when the experimental components are subjected to loads at some special angles, this system adds an automatic adjustment module to achieve high-precision fitting through automatic adjustment methods.

[0102] For the FDS (SPR) connection under specific angle load conditions, a tensile test is performed to obtain a standard force-displacement curve, which is used as the standard for subsequent parameter calibration and curve comparison.

[0103] CAE and SPR2 material cards are used to simulate the FDS (SPR) material combination to be predicted. Since the &thick and &diameter parameters have been determined, they are not adjusted. The six major parameters &ft, &dt, &xit, &fn, &dn, &xin have been accurately derived from the previous machine learning model, so they are not adjusted. The value range of α1, α2, and α3 is 0 to 1, so they can all be adjusted from 1 downward. Among them, R is introduced for the curve fitting degree. 2 Conduct quantitative assessment.

[0104] The specific implementation of the automatic closed-loop module is as follows: first perform threshold judgment, then optimize and adjust. The original parameters are x0, y0, z0, and the optimized parameters are x1, y1, z1. The fitting degree of the original parameters is R0. 2 , the parameter fitting degree after optimization is R1 2 , the contribution of parameter x1 to the optimization of fit is φ1, the contribution of parameter y1 to the optimization of fit is φ2, and the contribution of parameter z1 to the optimization of fit is φ3.

[0105] After multiple iterations, the threshold judgment module performs numerical comparison to obtain the best fitting parameter value and finally outputs the values ​​of X1, Y1, and Z1.

[0106] 7. How to use the overall mode: directly input the new material combination of the FDS to be tested, and finally directly output the failure form, predicted force displacement curve, and new material card parameter value.

[0107] The present invention uses the above embodiments and methods.

[0108] Able to realize high efficiency and high quality of neural network self-learning in the expansion of SPR2 failure cards for hot melt self-tapping and self-piercing rivets;

[0109] Reduce the steps, time, computing power and money consumption of developing new material combinations for spr2 invalid cards;

[0110] The connection points ultimately need to be integrated into the entire vehicle model for simulation, which will help improve the accuracy of subsequent vehicle collisions.

[0111] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for expanding SPR2 invalid cards based on neural network self-learning, characterized in that: The following methods are included: S1: Trial production of FDS or SPR connector; S2: Establish a database, and conduct targeted database establishment according to the input and output settings of subsequent machine learning; S3: Establishing artificial neural network pathways; S4: The improved machine learning architecture is built to first determine the failure form of the data set to be predicted and output it, and then use the corresponding failure data subset for subsequent prediction and calculation; S5: Read data and output SPR2 parameters, and adjust parameter values ​​according to the tension-shear working conditions; S6: Automatically adjust individual parameters; S7: How to use the overall mode: directly input the new material combination of FDS or SPR to be tested, and finally directly output the failure form, predicted force displacement curve, and new material card parameter values.

2. According to the SPR2 invalid card expansion method based on neural network self-learning according to claim 1, in step S2, The process includes: SS1: The establishment of the independent variable database includes the measurement and recording of the thickness and tensile strength of the parent material, as well as cutting the parent material and performing a quasi-static tensile test on a universal testing machine at a tensile speed of 2 mm / min, and finally recording the tensile strength of the parent material; SS2: Perform straight shear and cross tensile tests on FDS connectors to obtain force-displacement curves. The tensile test speed is 5 mm / min, and a gasket is set in the clamping area during the shear test. The tensile process is stopped manually when the force value reaches 0. SS3: Data processing, 5 specimens were prepared for each material combination, and the force-displacement curves of the lap shear and cross tensile tests were measured. The average of these curves was taken to form a new curve. The load data of every 0.10 mm in the load-displacement curve was used as the output data at the displacement of 0 ~ 15 mm.

3. According to the SPR2 failure card expansion method based on neural network self-learning described in claim 1, in step S3, the thickness of the upper and lower plates and the tensile strength of the upper and lower plates are input values, and the load data of every 0.10 mm in the load-displacement curve is used as the output data at the displacement of 0 ~ 15 mm, wherein 80% of the data set is used for learning, and the remaining 20% ​​is used for testing the trained artificial neural network, each mathematically defined neuron receives input from the previous layer of neurons, each input is multiplied by a separate weight value, and the weighted input is summed and passed to the output layer through a transfer function.

4. According to the SPR2 failure card expansion method based on neural network self-learning in claim 3, in step S3, the original database is classified according to the failure forms of hot-melt self-tapping screws. The failure forms of hot-melt self-tapping screws are divided into three types: base material breakage, screw pull-out and screw shearing.

5. According to the SPR2 failure card expansion method based on neural network self-learning according to claim 1, in step S5, four parameter values ​​are adjusted for the pulling and shearing working conditions, which are: &ft—peak tensile shear force; &dt—the corresponding value of displacement at the end; &xit—displacement value corresponding to peak shear force; &thick—total thickness of upper and lower plates; &diameter—Rivet diameter.

6. According to the SPR2 failure card expansion method based on neural network self-learning according to claim 1, in step S5, four parameter values ​​are adjusted for the cross-stretching condition, which are: &fn—Peak pull-out force; &dn—the corresponding value of displacement at the end; &xin—displacement value corresponding to peak pull-out force; &thick—total thickness of upper and lower plates; &diameter—Rivet diameter.

7. According to the SPR2 failure card expansion method based on neural network self-learning described in claim 1, in step S6, when the experimental component is subjected to a load at a special angle, an automatic adjustment module is added to achieve high-precision fitting through an automatic adjustment method.

8. According to the SPR2 failure card expansion method based on neural network self-learning described in claim 7, in step S6, a tensile test is carried out for the FDS or SPR connection of a specific angle load condition to obtain a standard force-displacement curve, and subsequent parameter calibration and curve calibration are based on this standard. CAE and SPR2 material cards are used to simulate the FDS or SPR material combination to be predicted. Because the &thick and &diameter parameters have been determined, no adjustment is made, and the six major parameters &ft, &dt, &xit, &fn, &dn, &xin have been accurately derived from the previous machine learning model, so no adjustment is made. The three angle-related parameters α1, α2, and α3 of the SPR2 failure card have a numerical range of 0 to 1, so they are all adjusted downward from 1, wherein R is introduced for the curve fit. 2 Conduct quantitative assessment.

9. According to the SPR2 invalid card expansion method based on neural network self-learning of claim 8, in step S6, the automatic adjustment module is specifically implemented as follows: First, make a threshold judgment; Then optimize and adjust the original parameters x0, y0, z0, and the optimized parameters are x1, y1, z1. The fitting degree of the original parameters is R0. 2 , the parameter fitting degree after optimization is R1 2 , the contribution of parameter x1 to the optimization of the fit is φ1, the contribution of parameter y1 to the optimization of the fit is φ2, and the contribution of parameter z1 to the optimization of the fit is φ3; After multiple iterations, the threshold judgment module performs numerical comparison to obtain the best fitting parameter value and finally outputs the values ​​of X1, Y1, and Z1.

Images