A riveting strength prediction method for an automobile relay

Abstract

The application discloses a riveting strength prediction method of an automobile relay, and comprises the following steps: step one, collecting extrusion stroke, expansion degree, riveting pressure, mechanical properties of an armature and mechanical properties of a moving spring; step two, constructing a strength prediction model based on a BP neural network and a genetic algorithm; and step three, inputting the collected extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature and mechanical properties of the moving spring into the strength prediction model to obtain predicted strength. The application has the characteristics of improving prediction accuracy, reducing the waste rate and improving robustness.

Description

Technical Field

[0001] This invention relates to the field of automotive relay technology, and more specifically, to a method for predicting the riveting strength of automotive relays. Background Technology

[0002] High Voltage DC Relay (HVR) is a core safety device in the high-voltage electrical system of new energy vehicles, often referred to as the "safety valve" or "main gate" of the electrical system. It is mainly used to control the on / off state of the high-voltage circuit. During normal vehicle operation, it connects the circuit, and in the event of a fault, it quickly separates the energy storage system from the electrical system, serving as the last line of defense for safety.

[0003] In the structure of a high-voltage relay, the riveting of the armature and the moving spring is one of the key assembly links. As an electromagnetic drive component, the armature forms a mechanical connection with the moving spring through riveting, transmitting electromagnetic force to the moving spring to realize the contact opening and closing action. The connection strength of this riveting structure directly affects the relay's operating characteristics, contact reliability, and service life. If the riveting strength is insufficient, it may lead to unstable relay parameters and large parameter changes at high and low temperatures. In severe cases, it may cause the riveting to loosen or even fall off, reducing the relay's resistance to mechanical vibration and impact.

[0004] Existing research on riveting strength prediction has the following main shortcomings:

[0005] 1. Existing riveting strength prediction methods mainly focus on analyzing the influence of specific process parameters on riveting strength through finite element simulation. A systematic method for predicting the strength of riveted structures has not yet been formed. Although finite element simulation can simulate the riveting forming and peeling process relatively accurately, its calculation cost is high and the cycle is long, making it difficult to meet the needs of rapid prediction in the production field.

[0006] 2. Existing research mainly focuses on the optimization analysis of single process parameters, and lacks systematic research on the prediction of riveting strength under the coupled effect of multiple process parameters. In fact, riveting strength is affected by multiple factors, and the analysis of a single parameter is difficult to fully reflect the changing law of riveting strength.

[0007] 3. Although existing research has attempted to use algorithms to predict the mechanical properties of riveted joints, existing models are mostly established for specific material systems and riveting processes, and their generalization ability is limited. For the specific riveting structure of high-voltage relay armature and moving spring, there is still a lack of strength prediction methods specifically for its structural characteristics and failure modes.

[0008] 4. Current riveting strength evaluation mainly relies on destructive tests (such as peel tests) to obtain critical load values. Although the results of this method are reliable, it is a post-event verification and cannot predict the strength before or during riveting. In addition, the test cost is high and the efficiency is low.

[0009] Therefore, there is an urgent need to develop a strength prediction method for the riveting structure of the armature and moving spring of a high-voltage relay, which can comprehensively consider the coupling effect of multiple process parameters and quickly and accurately predict the strength of the riveted structure before riveting or during production, so as to provide a basis for process parameter optimization and quality control. Summary of the Invention

[0010] The purpose of this invention is to design and develop a method for predicting the riveting strength of automotive relays. By combining multi-factor coupling with a BP neural network optimized by a genetic algorithm, the method can predict the riveting strength without destructive testing, while improving the accuracy of the prediction.

[0011] The technical solution provided by this invention is as follows:

[0012] A method for predicting the riveting strength of an automotive relay includes the following steps:

[0013] Step 1: Collect data on extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring;

[0014] Step 2: Construct an intensity prediction model based on BP neural network and genetic algorithm;

[0015] Step 3: Input the collected extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring into the strength prediction model to obtain the predicted strength.

[0016] Preferably, the mechanical properties of both the armature and the moving spring are yield strength.

[0017] Preferably, step one further includes:

[0018] Normalization pretreatment is performed on the extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring.

[0019] Preferably, step two specifically includes:

[0020] Step a: Construct a BP neural network model;

[0021] Step b: Optimize the initial parameters of the BP neural network using a genetic algorithm;

[0022] Step c: Train the BP neural network model to obtain the intensity prediction model.

[0023] Preferably, the BP neural network model includes an input layer, a hidden layer, and an output layer, wherein the input layer has 5 nodes, the hidden layer has 10 nodes, and the output layer has 1 node.

[0024] Preferably, the fitness function of the genetic algorithm is:

[0025] ;

[0026] In the formula, This represents the prediction error.

[0027] Preferably, step c specifically includes:

[0028] The network connection weights and network node biases are updated based on the network prediction error.

[0029] Preferably, updating the network connection weights based on the network prediction error specifically includes:

[0030] ;

[0031] ;

[0032] In the formula, The updated connection weights for the input and hidden layers. The updated connection weights for the hidden and output layers. This is the learning rate.

[0033] Preferably, updating the network node bias based on the network prediction error specifically includes:

[0034] ;

[0035] ;

[0036] In the formula, For the updated hidden layer bias, This is the bias for the updated output layer.

[0037] The beneficial effects of this invention are as follows:

[0038] This invention presents a method for predicting the riveting strength of automotive relays. It predicts the riveting strength of automotive relays by considering multiple influencing factors, taking into account the coupling effect of multiple factors rather than just one, thus improving prediction accuracy. Furthermore, it can predict riveting strength in real time through multiple factors without destructive experiments, reducing scrap rates and enabling early warning while shortening detection time. Simultaneously, the combination of a BP neural network and a genetic algorithm ensures network stability while improving the accuracy of prediction results, effectively avoiding the problem of traditional BP neural networks easily getting trapped in local optima, and improving robustness. Attached Figure Description

[0039] Figure 1This is a schematic diagram showing the comparison between the riveting strength prediction method for automotive relays described in this invention and the prediction curves of a traditional BP neural network. Detailed Implementation

[0040] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0041] The present invention provides a method for predicting the riveting strength of an automotive relay, comprising:

[0042] Step 1: Collect data on extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring;

[0043] In this process, the rivet between the armature and the moving spring is formed by extrusion. Therefore, the extrusion stroke is a key factor affecting the riveting strength. The head diameter of the rivet after forming, as the expansion degree, is also a key factor affecting the riveting strength. Similarly, the riveting pressure is related to the size of the rivet formed during the riveting process, which is also a key factor affecting the riveting strength. The materials of the armature and the moving spring themselves will inevitably affect the riveting strength. Therefore, the extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring are selected as key factors to predict the riveting strength of the high-voltage relay in new energy vehicles.

[0044] The mechanical properties of both the armature and the movable spring are yield strength;

[0045] Since the data points are of different orders of magnitude, normalization preprocessing is performed to avoid error issues:

[0046] ;

[0047] In the formula, For the first One data point, The first normalized preprocessing step One data point, The mean, Standard deviation;

[0048] Step 2: Construct an intensity prediction model based on BP neural network and genetic algorithm, which includes the following steps:

[0049] Step a: Construct the BP neural network model:

[0050] The BP neural network model includes an input layer, a hidden layer, and an output layer, with the input layer having 5 nodes.

[0051] The output of the hidden layer satisfies:

[0052] ;

[0053] In the formula, The output of the hidden layer, , The transfer function of the hidden layer. The connection weights between the input layer and the hidden layer. For hidden layer bias, The number of nodes in the input layer. This represents the number of nodes in the hidden layer.

[0054] The output of the output layer satisfies:

[0055] ;

[0056] In the formula, For output layer, The connection weights between the hidden layer and the output layer. For the bias of the output layer;

[0057] The output layer has only one node, which represents the riveting strength.

[0058] The hidden layer has 10 nodes;

[0059] Step b: Optimize the initial parameters of the BP neural network using a genetic algorithm.

[0060] Existing BP neural networks have large errors because their initial parameters are all fixed. To solve this problem, a genetic algorithm is used to search for the initial parameters of the BP neural network.

[0061] The specific steps of the genetic algorithm include:

[0062] Step 1: Population initialization;

[0063] In this embodiment, the population size is 50.

[0064] Step 2: Select floating-point numbers for encoding;

[0065] Step 3: Calculate the fitness function:

[0066] ;

[0067] In the formula, This represents the prediction error;

[0068] Step 4: Select individuals based on their fitness values ​​and perform crossover and mutation.

[0069] The probability of an individual being selected satisfies:

[0070] ;

[0071] ;

[0072] In the formula, For the first The probability of an individual being selected For population size, The probability of selecting the best individual. For individual debris removal, the individual ranked 1 maximum;

[0073] The intersections are uniform;

[0074] The mutation is performed using a non-uniform mutation method:

[0075]

[0076] Step 5: Update the population based on the principle of higher fitness until the maximum number of iterations is reached;

[0077] In this embodiment, the maximum number of iterations is 200.

[0078] Step c: Train the BP neural network model. After training, the model becomes the intensity prediction model.

[0079] Calculate the network prediction error based on the predicted and expected values ​​output by the output layer:

[0080] ;

[0081] In the formula, This is the expected value;

[0082] Update network connectivity weights based on network prediction errors:

[0083] ;

[0084] ;

[0085] In the formula, The updated connection weights for the input and hidden layers. The updated connection weights for the hidden and output layers. The learning rate;

[0086] Simultaneously, update the network node biases based on the network prediction error:

[0087] ;

[0088] ;

[0089] In the formula, For the updated hidden layer bias, The bias of the updated output layer;

[0090] Step 3: Input the collected extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring into the strength prediction model to obtain the predicted strength.

[0091] To verify the accuracy of the method described in this invention, it is now compared with a traditional BP neural network. The comparison results are shown in Table 1 and... Figure 1 As shown:

[0092] Table 1 Comparison of Intensity Predictions

[0093]

[0094] Combined with appendix Figure 1 As can be seen, the method described in this invention significantly reduces errors compared to traditional BP neural networks. Therefore, compared to using a single BP neural network model, the method described in this invention improves accuracy while ensuring stability.

[0095] This invention presents a method for predicting the riveting strength of automotive relays. It predicts the riveting strength of automotive relays by considering multiple influencing factors, taking into account the coupling effect of multiple factors rather than just one, thus improving prediction accuracy. Furthermore, it can predict riveting strength in real time through multiple factors without destructive experiments, reducing scrap rates and enabling early warning while shortening detection time. Simultaneously, the combination of a BP neural network and a genetic algorithm ensures network stability while improving the accuracy of prediction results, effectively avoiding the problem of traditional BP neural networks easily getting trapped in local optima, and improving robustness.

[0096] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A method for predicting the riveting strength of an automotive relay, characterized in that, Includes the following steps: Step 1: Collect data on extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring; Step 2: Construct an intensity prediction model based on BP neural network and genetic algorithm; Step 3: Input the collected extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring into the strength prediction model to obtain the predicted strength.

2. The method for predicting the riveting strength of an automotive relay according to claim 1, characterized in that, The mechanical properties of both the armature and the moving spring are yield strength.

3. The method for predicting the riveting strength of an automotive relay according to claim 2, characterized in that, Step one also includes: Normalization pretreatment is performed on the extrusion stroke, expansion degree, riveting pressure, mechanical properties of the armature, and mechanical properties of the moving spring.

4. The method for predicting the riveting strength of an automotive relay according to claim 3, characterized in that, Step two specifically includes: Step a: Construct a BP neural network model; Step b: Optimize the initial parameters of the BP neural network using a genetic algorithm; Step c: Train the BP neural network model to obtain the intensity prediction model.

5. The method for predicting the riveting strength of an automotive relay according to claim 4, characterized in that, The BP neural network model includes an input layer, a hidden layer, and an output layer. The input layer has 5 nodes, the hidden layer has 10 nodes, and the output layer has 1 node.

6. The method for predicting the riveting strength of an automotive relay according to claim 5, characterized in that, The fitness function of the genetic algorithm is: ； In the formula, This represents the prediction error.

7. The method for predicting the riveting strength of an automotive relay according to claim 6, characterized in that, Step c specifically includes: The network connection weights and network node biases are updated based on the network prediction error.

8. The method for predicting the riveting strength of an automotive relay according to claim 7, characterized in that, The step of updating network connection weights based on network prediction errors specifically includes: ； ； In the formula, The updated connection weights for the input and hidden layers. The updated connection weights for the hidden and output layers. This is the learning rate.

9. The method for predicting the riveting strength of an automotive relay according to claim 8, characterized in that, The step of updating the network node bias based on the network prediction error specifically includes: ； ； In the formula, For the updated hidden layer bias, This is the bias for the updated output layer.

Images