A method for intelligent composite strengthening treatment of cutting tooth surface

Abstract

This invention discloses an intelligent composite strengthening treatment method for the surface of a cutting tooth. Its key feature is the use of existing surface strengthening methods to strengthen the tooth head, while simultaneously employing laser local quenching for secondary strengthening of weak areas. The advantages are that the secondary strengthening of weak areas can be completed simultaneously during the tooth head surface strengthening process, effectively improving the fatigue strength and hardness of these areas and thus extending the overall service life of the cutting tooth. Furthermore, laser local quenching is a local heat treatment process with a small heat-affected zone, which does not negatively impact the overall quality of the cutting tooth. It allows for precise control of the local quenching quality, has a high degree of automation, and achieves high-quality and high-efficiency processing of the cutting tooth.

Description

Technical Field

[0001] This invention relates to the field of surface strengthening technology for workpieces, and in particular to an intelligent composite strengthening treatment method for the surface of cutting teeth. Background Technology

[0002] Cutting picks are mainly used in mining and tunneling machinery, and are divided into coal mining machine cutting picks and tunneling machine cutting picks. During use, cutting picks are subjected to compression and friction with the rock mass, leading to wear on the pick head. The high temperatures generated during operation and the corrosiveness of some rock masses also accelerate the wear of the pick head. Design and heat treatment issues can cause uneven stress distribution or stress exceeding allowable stress. Furthermore, encountering high-hardness rock during operation can subject the pick to significant impact loads, which, under cyclic impact loads, can easily lead to fatigue fracture at the connection area between the pick head and the shank. Therefore, the reliability and service life of cutting picks are often improved by selecting higher-quality materials, adjusting heat treatment processes, and strengthening the surface of the pick.

[0003] Currently, commonly used surface strengthening methods for cutting tools include wire welding, thermal spraying, plasma cladding, and laser cladding. However, these surface strengthening methods alone cannot achieve the desired fatigue strength and service life of the cutting tools. Therefore, shot blasting is sometimes used during the cutting tool production process for secondary cleaning and strengthening of the tooth surface. The main function of shot blasting is to remove the surface oxide layer and impurities, improving the surface hardness, wear resistance, and fatigue strength of the cutting tools. However, when the cutting tool shape is complex, shot blasting can create blind spots, where fatigue strength and hardness are lower, increasing the risk of fatigue fracture. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent composite strengthening treatment method for the surface of a cutting tooth. During the strengthening of the surface of the cutting tooth head, the method can simultaneously complete the secondary strengthening of the weak parts of the cutting tooth, effectively improving the fatigue strength and hardness of the weak parts of the cutting tooth, thereby improving the overall service life of the cutting tooth, and also achieving high-quality and high-efficiency treatment of the cutting tooth.

[0005] The technical solution adopted by this invention to solve the above-mentioned technical problems is as follows: an intelligent composite strengthening treatment method for the surface of a cutting tooth, which strengthens the cutting tooth head using existing surface strengthening methods, and simultaneously strengthens the weak parts of the cutting tooth by laser local quenching, specifically including the following steps:

[0006] (1) Establish an initial surface strengthening database for each type of cutting tooth after applying different surface strengthening methods. The initial surface strengthening database includes different processing parameters for each surface strengthening method and the corresponding strengthening results.

[0007] (2) Establish an initial laser strengthening database for each type of cutting tooth after laser local quenching. The initial laser strengthening database includes different processing parameters used during laser local quenching and the corresponding strengthening results.

[0008] (3) For tooth surface strengthening treatment, a surface strengthening intelligent database is established using the PSO-BP-ANN neural network model, and the initial optimal weights and thresholds of the BP-ANN neural network in the PSO-BP-ANN neural network model are determined using the initial surface strengthening database; For laser local quenching treatment, a laser strengthening intelligent database is established using the GA-BP-ANN neural network model, and the initial optimal weights and thresholds of the BP-ANN neural network in the GA-BP-ANN neural network model are determined using the initial laser strengthening database.

[0009] (4) During processing, the processing system automatically identifies the type of cutting tooth to be strengthened, and retrieves the processing parameters and results of tooth surface strengthening treatment that match the type of cutting tooth through the surface strengthening intelligent database, and retrieves the processing parameters and results of laser local quenching treatment that match the type of cutting tooth through the laser strengthening intelligent database. Then, the processing system uses the two retrieved processing parameters as the initial processing parameters for surface strengthening and the initial processing parameters for quenching, respectively, to perform surface strengthening treatment and laser local quenching treatment on the cutting tooth to be strengthened at the same time.

[0010] (5) The PSO-BP-ANN neural network model and the GA-BP-ANN neural network model predict the surface strengthening result and the quenching strengthening result according to the corresponding initial surface strengthening processing parameters and initial quenching processing parameters, respectively. The predicted surface strengthening result and the quenching strengthening result are called the predicted result, the actual surface strengthening result and the quenching strengthening result after actual processing are called the actual result, and the retrieved tooth surface strengthening result and the quenching strengthening result are called the theoretical result. The three results are compared one by one. If the three are consistent, the corresponding strengthening process is continued according to the initial surface strengthening processing parameters and the initial quenching processing parameters until the cutting tooth processing is completed. If at least one of the three is inconsistent, step (6) is performed.

[0011] (6) Based on the theoretical results, the initial processing parameters for surface strengthening and quenching are corrected by the PSO-BP-ANN neural network model and the GA-BP-ANN neural network model respectively until the predicted results, actual results and theoretical results are consistent. The corrected surface strengthening and quenching processing parameters are used as the optimal processing parameters to strengthen the cutting teeth until the cutting teeth are processed. At the same time, the two optimal processing parameters are imported into the corresponding surface strengthening intelligent database and laser strengthening intelligent database for corresponding data correction.

[0012] Furthermore, after the cutting teeth are processed, 5-10% of the finished cutting teeth are sampled for testing. The actual surface hardening results and actual laser local quenching hardening results of the cutting teeth samples, as well as the corresponding processing parameters, are recorded. These samples are used as the validation datasets for the PSO-BP-ANN neural network model and the GA-BP-ANN neural network model, and are imported into the surface hardening intelligent database and the laser hardening intelligent database for data correction and optimization learning of the neural networks.

[0013] Furthermore, the types of cutting teeth are divided into coal mining machine cutting teeth, tunneling machine cutting teeth, and rotary drilling machine cutting teeth.

[0014] Furthermore, the surface strengthening method is wire welding, thermal spraying, surface penetration, plasma cladding, or laser cladding.

[0015] Furthermore, the processing parameters in the surface strengthening treatment in step (1) are surface strengthening power, scanning speed and spot size, and the strengthening result parameters are strengthening layer height, dilution rate, temperature distribution and strengthening layer width; the processing parameters in the laser local quenching in step (2) are scanning speed, laser power and spot diameter, and the strengthening result parameters are quenching layer hardness and quenching layer depth.

[0016] Furthermore, the initial surface strengthening database and the initial laser strengthening database are obtained through experimental processing or numerical simulation.

[0017] Further, in step (3), the PSO-BP-ANN neural network model combines an ANN neural network, a feedforward backpropagation learning algorithm (BP), and a particle swarm optimization algorithm (PSO). It has one input layer, two hidden layers, and one output layer. The input layer parameters are surface enhancement power, scanning speed, and spot size. The Sigmoid function is used as the activation function from the input layer to the hidden layer and from the hidden layer to the hidden layer. The output layer parameters are enhancement layer height, enhancement layer width, temperature distribution, and dilution rate. The hidden layer has 9 neurons, and the activation function from the hidden layer to the output layer is the purelin linear function. The update algorithm of the PSO-BP-ANN neural network model is expressed as:

[0018] ,

[0019] ,

[0020] in: y’ n The predicted results for surface strengthening include the strengthening layer height, dilution rate, temperature distribution, and strengthening layer width. x1 represents the surface strengthening power, x2 represents the scanning speed, and x3 represents the spot size. vFor particle update speed, j This represents the current iteration number; r 1 and r 2 The random numbers are distributed in (0,1). c 1 and c 2 For learning constants or acceleration factors, pop It is a particle. pbest This represents the optimal individual extreme value of the previous generation. gbest It is the global optimal extreme value. ω This is the weighting factor.

[0021] Furthermore, in step (3), the GA-BP-ANN neural network model combines an ANN neural network, a feedforward backpropagation learning algorithm (BP), and a genetic algorithm (GA). It has an input layer, a hidden layer, and an output layer. The input layer parameters are scanning speed, laser power, and spot diameter. The output layer parameters are quenching layer depth and quenching layer hardness. The hidden layer has 8 neurons. In the GA-BP-ANN neural network, the input layer parameters are grouped, and different parameter values ​​correspond to different groups. Each group is considered an individual, and each individual corresponds to a set of weights and thresholds. The fitness calculation formula for each individual is as follows:

[0022] ,

[0023] in: R i The hardness of the quenched layer is measured at any measurement point i. r i It is the predicted hardness of the quenched layer at any measurement point i obtained from ANN; H i It is the depth of the quenched layer measured at any measurement point i. h i The predicted quenching layer depth is obtained from any measurement point i using ANN, where n is the total number of measurement points. fit 1 To accommodate the hardness of the quenched layer. fit 2 The fitness of the quenching layer depth is determined by selecting the individual with the smallest fitness to determine the initial weights and thresholds. The smaller the fitness, the closer the predicted value of the data set is to the measured value.

[0024] Further, in step (3), the process of determining the initial optimal weights and thresholds of the BP-ANN neural network in the PSO-BP-ANN neural network model is as follows: Construct a basic BP-ANN neural network structure, import the processing parameters from the initial surface strengthening database into the BP-ANN neural network structure, import the processing parameters from the initial surface strengthening database and their corresponding strengthening results into the PSO optimization algorithm module, and perform normalization processing; then, use the PSO optimization algorithm module embedded in the BP-ANN neural network structure to obtain the initial optimal weights and thresholds of the BP-ANN neural network. The threshold and PSO optimization algorithm module process is as follows: First, 100 particles with different positions and velocities are added to the PSO optimization algorithm. Each particle corresponds to a set of weights and thresholds. The position represents the weights and thresholds of the processing parameters carried by the particle, and the velocity represents the direction and distance of the particle's movement in the next iteration. During the PSO optimization process, the position and velocity of the particles are continuously updated by comparing the individual optimal solution and the global optimal solution. After obtaining the optimal fitness after a set number of iterations, the weights and thresholds carried by the particles are used as the initial optimal weights and thresholds and substituted into the BP-ANN neural network.

[0025] The process of determining the initial optimal weights and thresholds of the BP-ANN neural network in the GA-BP-ANN neural network model is as follows: First, a basic BP-ANN neural network structure is constructed. The processing parameters from the initial laser enhancement database are imported into the BP-ANN neural network structure. The processing parameters and corresponding enhancement results from the initial laser enhancement database are then imported into the GA algorithm module and normalized. Then, the initial optimal weights and thresholds of the BP-ANN neural network are obtained using the GA algorithm module embedded in the BP-ANN neural network structure. The GA algorithm module process is as follows: The input layer parameters are treated as groups, with different parameter values ​​corresponding to different groups. Each group is treated as an individual, and each individual corresponds to a set of weights and thresholds. Thus, 100 individuals are added for neural network training. During the GA optimization process, the predicted results of each individual are compared with the actual results. The smaller the difference between the predicted and actual results, the more suitable the set of weights and thresholds corresponding to that individual is. After a preset number of iterations, the weights and thresholds of the individual with the smallest difference are used as the initial optimal weights and thresholds and substituted into the BP-ANN neural network.

[0026] Furthermore, in step (6), the process of correcting the initial surface strengthening processing parameters using the PSO-BP-ANN neural network model is as follows:

[0027] (1) After the BP-ANN neural network operation, when the actual result of surface strengthening is inconsistent with the predicted result, the initial processing parameters of surface strengthening and the actual result are added as training data to the initial particle swarm of the PSO optimization algorithm module, and the PSO optimization algorithm module processing in step (3) is repeated to calculate the initial optimal weight and threshold again until the predicted result is consistent with the actual result.

[0028] (2) When the predicted surface enhancement result matches the actual result, compare the predicted surface enhancement result with the theoretical result. When the predicted result and the theoretical result are inconsistent, activate the parameter optimization function of the feedforward backpropagation learning algorithm BP in the BP-ANN neural network. Specifically, based on the theoretical result, calculate the error between the predicted result and the theoretical result to obtain the loss function. The loss function is the mean squared error function MSE, i.e.:

[0029] ,

[0030] in: y n This is a theoretical result of surface strengthening. y’ n The predicted results are for surface strengthening. n This refers to the number of neurons in the hidden layer of the PSO-BP-ANN neural network model. n Taking the values ​​1, 2, ..., 9; by taking the partial derivative of the loss function with the corresponding input parameters before weighted summation, we can obtain the loss gradient between the predicted and theoretical results of surface enhancement, i.e.:

[0031] ,

[0032] in: The loss gradient between the input layer and the first hidden layer. The loss gradient between the second hidden layer and the first hidden layer. The loss gradient between the output layer and the second hidden layer. These are the initial processing parameters for surface strengthening, i.e., the input parameters before the weighted summation in the first hidden layer. These are the input parameters before the weighted summation in the second hidden layer. The input parameters are those before the weighted summation of the input to the output layer;

[0033] , The calculation process is as follows:

[0034] ,

[0035] ,

[0036] in: x 1. x 2. x 3 represents the initial processing parameters for surface strengthening, i.e. ; , , for x 1. x 2. x The weight corresponding to 3; The input to the first hidden layer. , ... The input parameters before the weighted summation in the second hidden layer, i.e. ; , ... for , ... The corresponding weights; This is the input quantity to the second hidden layer;

[0037] (3) After obtaining the loss gradient, use the gradient descent formula to evaluate the input parameters of each layer. , and The input parameters are updated and optimized to obtain the updated and optimized parameters. , and Specifically:

[0038] ,

[0039] in: η For learning rate, The initial processing parameters for optimized surface strengthening. k Selecting 1, 2, and 3 represents the surface strengthening power, scanning speed, and spot size in the surface strengthening processing parameters for the cutting tooth head, thus enabling the adjustment of input parameters. The optimal surface strengthening processing parameters can be output after the predicted, actual, and theoretical results are consistent through correction and optimization.

[0040] Furthermore, in step (6), the process of correcting the initial quenching parameters using the GA-BP-ANN neural network model is as follows:

[0041] (1) After the BP-ANN neural network operation, when the actual result of laser quenching is inconsistent with the predicted result, the initial processing parameters of quenching and the actual result are added as training data to the initial individual group of the GA algorithm module, and the GA algorithm module processing process in step (3) is repeated to calculate the initial optimal weight and threshold again until the predicted result is consistent with the actual result.

[0042] (2) When the predicted result of laser quenching is consistent with the actual result, compare the predicted result of laser quenching with the theoretical result. When the predicted result is inconsistent with the theoretical result, activate the parameter optimization function of the feedforward backpropagation learning algorithm BP in the BP-ANN neural network. Specifically, based on the theoretical result, calculate the error between the predicted result and the theoretical result to obtain the loss function. In the GA-BP-ANN neural network, the loss function is the mean square error function. ,Right now:

[0043] ,

[0044] in: z n This is the theoretical result of laser hardening. z’ n The predicted results for laser quenching. n This refers to the number of neurons in the hidden layer of the GA-BP-ANN neural network model. n Taking the values ​​1, 2, ..., 8; by taking the partial derivative of the loss function with the corresponding input parameters before weighted summation, we can obtain the loss gradient between the predicted and theoretical results of laser quenching, i.e.:

[0045] ,

[0046] in: The loss gradient between the input layer and the hidden layer. The loss gradient between the output layer and the hidden layer. The laser power and scanning speed are the initial processing parameters for quenching, i.e., the input parameters before the weighted summation in the hidden layer. The input parameters are those before the weighted summation of the input to the output layer;

[0047] The calculation process is as follows:

[0048] ,

[0049] in: λ 1. λ 2 represents the laser power and scanning speed in the initial quenching processing parameters, i.e. ; , for λ 1. λ The weight corresponding to 2; The input quantity to the hidden layer;

[0050] (3) After obtaining the loss gradient, use the gradient descent formula to evaluate the input parameters of each layer. , The input parameters are updated and optimized to obtain the updated and optimized parameters. , Specifically:

[0051] ,

[0052] in: For learning rate, To optimize the laser power and scanning speed in the initial quenching processing parameters, the input parameters are adjusted. The optimal quenching parameters can be output after the predicted, actual, and theoretical results are consistent through correction and optimization.

[0053] Compared with the prior art, the advantages of the present invention are:

[0054] (1) This method can simultaneously strengthen the weak parts of the cutting tooth (such as the area where the tooth head and the tooth shank are connected) during the strengthening process of the cutting tooth head surface, effectively improving the fatigue strength and hardness of the weak parts of the cutting tooth, thereby improving the overall service life of the cutting tooth. Moreover, laser local quenching is a local heat treatment process with a small heat-affected zone, which will not have a negative impact on the overall quality of the cutting tooth. It can achieve precise control of the local quenching quality, has a high degree of automation, and achieves high-quality and high-efficiency processing of the cutting tooth.

[0055] (2) This intelligent composite strengthening method improves the surface strengthening quality stability of cutting teeth while reducing labor costs;

[0056] (3) This method simultaneously corrects and optimizes the intelligent database during the surface strengthening process of the cutting teeth, further improving the stability of the strengthening quality. Attached Figure Description

[0057] Figure 1 This is a structural diagram of the PSO-BP-ANN neural network model of the present invention;

[0058] Figure 2 This is a structural diagram of the GA-BP-ANN neural network model of the present invention;

[0059] Figure 3 This is a microscopic schematic diagram of the laser-hardened layer on the surface of the cutting tooth after composite strengthening treatment.

[0060] Figure 4 The image shows the hardness test results of the cutting tool substrate before composite strengthening treatment.

[0061] Figure 5 This is a diagram showing the hardness test results of the cutting tooth head after surface strengthening treatment.

[0062] Figure 6 This image shows the hardness test results of the weak parts of the cutting teeth after laser local quenching. Detailed Implementation

[0063] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0064] As shown in the figure, an intelligent composite strengthening method for the surface of a cutting tooth is proposed. This method strengthens the tooth tip using existing surface strengthening methods and simultaneously strengthens the weak areas of the cutting tooth using laser local quenching for secondary strengthening. The specific steps include:

[0065] (1) Establish an initial surface strengthening database for each type of cutting tooth after using different surface strengthening methods. The initial surface strengthening database includes different processing parameters and corresponding strengthening results for each surface strengthening method. Among them, the processing parameters are surface strengthening power, scanning speed and spot size, and the strengthening results are strengthening layer height, dilution rate, temperature distribution and strengthening layer width.

[0066] (2) Establish an initial laser strengthening database for each type of cutting tooth after laser local quenching. The initial laser strengthening database includes different processing parameters used during laser local quenching and the corresponding strengthening results. Among them, the processing parameters are scanning speed, laser power and spot diameter, and the strengthening results are hardness of quenched layer and depth of quenched layer.

[0067] (3) For tooth surface strengthening treatment, a surface strengthening intelligent database is established using a PSO-BP-ANN neural network model, and the initial optimal weights and thresholds of the BP-ANN neural network in the PSO-BP-ANN neural network model are determined using the initial surface strengthening database; for laser local quenching treatment, a laser strengthening intelligent database is established using a GA-BP-ANN neural network model, and the initial optimal weights and thresholds of the BP-ANN neural network in the GA-BP-ANN neural network model are determined using the initial laser strengthening database; specifically:

[0068] The PSO-BP-ANN neural network model combines an ANN neural network, the feedforward backpropagation (BP) learning algorithm, and the particle swarm optimization (PSO) algorithm. It has one input layer, two hidden layers, and one output layer. The input layer parameters are surface enhancement power, scanning speed, and spot size. The sigmoid function is used as the activation function from the input layer to the hidden layer and from the hidden layer to the hidden layer. The output layer parameters are enhancement layer height, enhancement layer width, temperature distribution, and dilution rate. The hidden layer has 9 neurons, and the activation function from the hidden layer to the output layer is the purelin linear function. Figure 1 As shown, the update algorithm for the PSO-BP-ANN neural network model is expressed as follows:

[0069] ,

[0070] ,

[0071] in: y’ n The predicted results for surface strengthening include the strengthening layer height, dilution rate, temperature distribution, and strengthening layer width. x1 represents the surface strengthening power, x2 represents the scanning speed, and x3 represents the spot size. v For particle update speed, j This represents the current iteration number; r 1 and r 2 The random numbers are distributed in (0,1). c 1 and c 2 For learning constants or acceleration factors, pop It is a particle. pbest This represents the optimal individual extreme value of the previous generation. gbest It is the global optimal extreme value. ω For weighting factors;

[0072] The process for determining the initial optimal weights and thresholds of the BP-ANN neural network in the PSO-BP-ANN neural network model is as follows: A basic BP-ANN neural network structure is constructed. The processing parameters from the initial surface strengthening database are imported into the BP-ANN neural network structure. The processing parameters and corresponding strengthening results from the initial surface strengthening database are imported into the PSO optimization algorithm module and normalized. Then, the initial optimal weights and thresholds of the BP-ANN neural network are obtained using the PSO optimization algorithm module embedded in the BP-ANN neural network structure. The PSO optimization algorithm module process is as follows: First, 100 particles with different positions and velocities are added to the PSO optimization algorithm. Each particle corresponds to a set of weight and threshold solutions, where: position represents the weight and threshold of the processing parameters carried by the particle, and velocity represents the direction and distance of the particle's movement in the next iteration. During the PSO optimization process, the position and velocity of the particles are continuously updated by comparing the individual optimal solution and the global optimal solution. After obtaining the optimal fitness after a set number of iterations, the weights and thresholds carried by the particles are used as the initial optimal weights and thresholds and substituted into the BP-ANN neural network.

[0073] The GA-BP-ANN neural network model combines an ANN neural network, the feedforward backpropagation (BP) learning algorithm, and the genetic algorithm (GA). It has one input layer, one hidden layer, and one output layer. The input layer parameters are scanning speed, laser power, and spot diameter; the output layer parameters are quenching layer depth and quenching layer hardness; and the hidden layer has 8 neurons. Figure 2 As shown, in the GA-BP-ANN neural network, the parameters in the input layer are treated as groups, with different parameter values ​​corresponding to different groups. Each group is considered an individual, and each individual corresponds to a solution with a set of weights and thresholds. The formula for calculating the fitness of each individual is as follows:

[0074] ,

[0075] in: R i The hardness of the quenched layer is measured at any measurement point i. r i It is the predicted hardness of the quenched layer at any measurement point i obtained from ANN; H i It is the depth of the quenched layer measured at any measurement point i. h i The predicted quenching layer depth is obtained from any measurement point i using ANN, where n is the total number of measurement points. fit 1 To accommodate the hardness of the quenched layer. fit 2The fitness of the quenching layer depth is determined by selecting the individual with the smallest fitness to determine the initial weights and thresholds. The smaller the fitness, the closer the predicted value of the data set is to the measured value.

[0076] The process for determining the initial optimal weights and thresholds of the BP-ANN neural network in the GA-BP-ANN neural network model is as follows: First, a basic BP-ANN neural network structure is constructed. The processing parameters from the initial laser enhancement database are imported into the BP-ANN neural network structure. The processing parameters and corresponding enhancement results from the initial laser enhancement database are imported into the GA algorithm module and normalized. Then, the initial optimal weights and thresholds of the BP-ANN neural network are obtained using the GA algorithm module embedded in the BP-ANN neural network structure. The GA algorithm module process is as follows: The input layer parameters are grouped, and different parameter values ​​correspond to different groups. Each group is treated as an individual, and each individual corresponds to a set of weights and thresholds. Thus, 100 individuals are added for training the neural network. During the GA optimization process, the predicted results of each individual are compared with the actual results. The smaller the difference between the predicted results and the actual results, the more suitable the set of weights and thresholds corresponding to that individual are. After a preset number of iterations, the weights and thresholds of the individual with the smallest difference are used as the initial optimal weights and thresholds and substituted into the BP-ANN neural network.

[0077] (4) During processing, the processing system automatically identifies the type of cutting tooth by taking pictures of the cutting tooth to be strengthened through the high-speed camera, and retrieves the processing parameters and results of tooth surface strengthening treatment that match the type of cutting tooth through the surface strengthening intelligent database, and retrieves the processing parameters and results of laser local quenching treatment that match the type of cutting tooth through the laser strengthening intelligent database. Then, the processing system uses the above two processing parameters as the initial processing parameters for surface strengthening and the initial processing parameters for quenching, respectively, to perform surface strengthening treatment and laser local quenching treatment on the cutting tooth to be strengthened at the same time.

[0078] (5) The PSO-BP-ANN neural network model and the GA-BP-ANN neural network model predict the surface strengthening result and the quenching strengthening result according to the corresponding initial surface strengthening processing parameters and initial quenching processing parameters, respectively. The predicted surface strengthening result and the quenching strengthening result are called the predicted result, the actual surface strengthening result and the quenching strengthening result after actual processing are called the actual result, and the retrieved tooth surface strengthening result and the quenching strengthening result are called the theoretical result. The three results are compared one by one. If the three are consistent, the corresponding strengthening process is continued according to the initial surface strengthening processing parameters and the initial quenching processing parameters until the cutting tooth processing is completed. If at least one of the three is inconsistent, step (6) is performed.

[0079] (6) Based on the theoretical results, the initial processing parameters for surface strengthening and quenching are corrected by the PSO-BP-ANN neural network model and the GA-BP-ANN neural network model respectively until the predicted results, actual results and theoretical results are consistent. The corrected surface strengthening and quenching processing parameters are used as the optimal processing parameters to strengthen the cutting teeth until the cutting teeth are processed. At the same time, the two optimal processing parameters are imported into the corresponding surface strengthening intelligent database and laser strengthening intelligent database for corresponding data correction.

[0080] The process of correcting the initial processing parameters for surface strengthening using the PSO-BP-ANN neural network model is as follows:

[0081] (1) After the BP-ANN neural network operation, when the actual result of surface strengthening is inconsistent with the predicted result, the initial processing parameters of surface strengthening and the actual result are added as training data to the initial particle swarm of the PSO optimization algorithm module, and the PSO optimization algorithm module processing in step (3) is repeated to calculate the initial optimal weight and threshold again until the predicted result is consistent with the actual result.

[0082] (2) When the predicted surface enhancement result matches the actual result, compare the predicted surface enhancement result with the theoretical result. When the predicted result and the theoretical result are inconsistent, activate the parameter optimization function of the feedforward backpropagation learning algorithm BP in the BP-ANN neural network. Specifically, based on the theoretical result, calculate the error between the predicted result and the theoretical result to obtain the loss function. The loss function is the mean squared error function MSE, i.e.:

[0083] ,

[0084] in: y n This is a theoretical result of surface strengthening. y’ n The predicted results are for surface strengthening. n This refers to the number of neurons in the hidden layer of the PSO-BP-ANN neural network model. n Taking the values ​​1, 2, ..., 9; by taking the partial derivative of the loss function with the corresponding input parameters before weighted summation, we can obtain the loss gradient between the predicted and theoretical results of surface enhancement, i.e.:

[0085] ,

[0086] in: The loss gradient between the input layer and the first hidden layer. The loss gradient between the second hidden layer and the first hidden layer. The loss gradient between the output layer and the second hidden layer. These are the initial processing parameters for surface strengthening, i.e., the input parameters before the weighted summation in the first hidden layer. These are the input parameters before the weighted summation in the second hidden layer. The input parameters are those before the weighted summation of the input to the output layer;

[0087] , The calculation process is as follows:

[0088] ,

[0089] ,

[0090] in: x 1. x 2. x 3 represents the initial processing parameters for surface strengthening, i.e. ; , , for x 1. x 2. x The weight corresponding to 3; The input to the first hidden layer. , ... The input parameters before the weighted summation in the second hidden layer, i.e. ; , ... for , ... The corresponding weights; This is the input quantity to the second hidden layer;

[0091] (3) After obtaining the loss gradient, use the gradient descent formula to evaluate the input parameters of each layer. , and The input parameters are updated and optimized to obtain the updated and optimized parameters. , and Specifically:

[0092] ,

[0093] in: η For learning rate, The initial processing parameters for optimized surface strengthening. kSelecting 1, 2, and 3 represents the surface strengthening power, scanning speed, and spot size in the surface strengthening processing parameters for the cutting tooth head, thus enabling the adjustment of input parameters. The optimal surface strengthening processing parameters can be output after the predicted results, actual results and theoretical results are consistent through correction and optimization.

[0094] The process of correcting the initial quenching parameters using the GA-BP-ANN neural network model is as follows:

[0095] (1) After the BP-ANN neural network operation, when the actual result of laser quenching is inconsistent with the predicted result, the initial processing parameters of quenching and the actual result are added as training data to the initial individual group of the GA algorithm module, and the GA algorithm module processing process in step (3) is repeated to calculate the initial optimal weight and threshold again until the predicted result is consistent with the actual result.

[0096] (2) When the predicted result of laser quenching is consistent with the actual result, compare the predicted result of laser quenching with the theoretical result. When the predicted result is inconsistent with the theoretical result, activate the parameter optimization function of the feedforward backpropagation learning algorithm BP in the BP-ANN neural network. Specifically, based on the theoretical result, calculate the error between the predicted result and the theoretical result to obtain the loss function. In the GA-BP-ANN neural network, the loss function is the mean square error function. ,Right now:

[0097] ,

[0098] in: z n This is the theoretical result of laser hardening. z’ n The predicted results for laser quenching. n This refers to the number of neurons in the hidden layer of the GA-BP-ANN neural network model. n Taking the values ​​1, 2, ..., 8; by taking the partial derivative of the loss function with the corresponding input parameters before weighted summation, we can obtain the loss gradient between the predicted and theoretical results of laser quenching, i.e.:

[0099] ,

[0100] in: The loss gradient between the input layer and the hidden layer. The loss gradient between the output layer and the hidden layer. The laser power and scanning speed are the initial processing parameters for quenching, i.e., the input parameters before the weighted summation in the hidden layer. The input parameters are those before the weighted summation of the input to the output layer;

[0101] The calculation process is as follows:

[0102] ,

[0103] in: λ 1. λ 2 represents the laser power and scanning speed in the initial quenching processing parameters, i.e. ; , for λ 1. λ The weight corresponding to 2; The input quantity to the hidden layer;

[0104] (3) After obtaining the loss gradient, use the gradient descent formula to evaluate the input parameters of each layer. , The input parameters are updated and optimized to obtain the updated and optimized parameters. , Specifically:

[0105] ,

[0106] in: For learning rate, To optimize the laser power and scanning speed in the initial quenching processing parameters, the input parameters are adjusted. The optimal quenching parameters can be output after the predicted, actual, and theoretical results are consistent through correction and optimization.

[0107] To further improve the accuracy of the intelligent database, after the cutting teeth are processed, 5-10% of the finished cutting teeth are sampled for testing. The actual surface strengthening results and actual laser local quenching strengthening results of the cutting teeth samples, as well as the corresponding processing parameters, are recorded. These samples are then used as the validation dataset for the PSO-BP-ANN neural network model and the GA-BP-ANN neural network model, and imported into the surface strengthening intelligent database and the laser strengthening intelligent database for data correction and optimization learning of the neural networks.

[0108] The types of cutting teeth processed by this invention can be divided into coal mining machine cutting teeth, tunneling machine cutting teeth and rotary drilling machine cutting teeth. The existing surface strengthening methods used for the cutting tooth heads are welding wire surfacing technology, thermal spraying technology, surface penetration technology, plasma cladding technology or laser cladding technology.

[0109] Furthermore, the initial surface strengthening database and the initial laser strengthening database in steps (1) and (2) of the above embodiments can be obtained through experimental processing or numerical simulation, such as by numerical simulation using finite element software such as Ansys or Abaqus.

[0110] The cutting teeth with a substrate material of 42CrMo were subjected to the intelligent composite strengthening treatment of the present invention, and then hardness testing was performed. Figure 3 The image shows the laser-hardened layer obtained by the intelligent composite strengthening method on the surface of the cutting tooth. The dark area at the top is the cutting tooth substrate, and the white area at the bottom is the laser-hardened layer. There is a good transition interface between the cutting tooth substrate and the laser-hardened layer. Figure 4 For the substrate area of ​​the cutting tooth, two points were taken for hardness testing before its composite strengthening treatment. The microhardnesses were 555.5 HV and 552 HV, respectively (Rockwell hardnesses were 52.7 HRC and 52.5 HRC, respectively). Figure 5 The Rockwell hardness of the cutting tooth head after surface strengthening treatment is increased to about 68 HRC. Figure 6 For the laser-hardened areas of the weak points in the cutting tooth, hardness tests were performed on two sampling points. The results showed that the microhardness of the laser-hardened areas increased to 577.2 HV and 701.3 HV (Rockwell hardness of 59.1 HRC and 60.1 HRC, respectively). This demonstrates that after intelligent composite strengthening treatment, the hardness of the cutting tooth surface and the hardness of the weak points are significantly improved compared to the overall base material of the cutting tooth.

[0111] The scope of protection of this invention includes, but is not limited to, the above embodiments. The scope of protection is defined by the claims. Any substitutions, modifications, or improvements to this technology that are easily conceived by those skilled in the art fall within the scope of protection of this invention.

Claims

1. A pick surface intelligent composite strengthening treatment method, characterized in that The tooth head of the cutting tooth is strengthened by using the existing surface strengthening method, and the weak part of the cutting tooth is simultaneously strengthened by using laser local quenching, specifically including the following steps: (1) Establish an initial surface strengthening database of each type of cutting tooth after using different surface strengthening methods, wherein the initial surface strengthening database includes different processing parameters of each surface strengthening method and the corresponding strengthening results; (2) Establish an initial laser strengthening database of each type of cutting tooth after laser local quenching, wherein the initial laser strengthening database includes different processing parameters of laser local quenching and the corresponding strengthening results; (3) Establish a surface strengthening intelligent database for tooth head surface strengthening treatment by a PSO-BP-ANN neural network model, and determine the initial optimal weight and threshold value of the BP-ANN neural network in the PSO-BP-ANN neural network model through the initial surface strengthening database; establish a laser strengthening intelligent database for laser local quenching treatment by a GA-BP-ANN neural network model, and determine the initial optimal weight and threshold value of the BP-ANN neural network in the GA-BP-ANN neural network model through the initial laser strengthening database; (4) During processing, the processing system automatically identifies the type of the cutting tooth to be strengthened, and calls out the processing parameters and the tooth head surface strengthening result of the tooth head surface strengthening treatment matched with the type of the cutting tooth through the surface strengthening intelligent database, and calls out the processing parameters and the quenching strengthening result of the laser local quenching treatment matched with the type of the cutting tooth through the laser strengthening intelligent database, and then the processing system simultaneously performs surface strengthening treatment and laser local quenching treatment on the cutting tooth to be strengthened by using the above two called-out processing parameters as the surface strengthening initial processing parameters and the quenching initial processing parameters respectively; (5) The PSO-BP-ANN neural network model and the GA-BP-ANN neural network model respectively predict the surface strengthening result and the quenching strengthening result according to the corresponding surface strengthening initial processing parameters and the quenching initial processing parameters, and compare the predicted surface strengthening result and the quenching strengthening result (referred to as predicted result), the actual surface strengthening result and the quenching strengthening result after actual processing (referred to as actual result), and the called-out tooth head surface strengthening result and the quenching strengthening result (referred to as theoretical result) one by one, if the three are consistent, the corresponding strengthening treatment is continued according to the surface strengthening initial processing parameters and the quenching initial processing parameters until the cutting tooth processing is completed; if at least one of the three is inconsistent, step (6) is performed; (6) Adjust the surface strengthening initial processing parameters and the quenching initial processing parameters, and then return to step (4) to continue the processing. (6), taking the theoretical result as the benchmark, the surface strengthening initial processing parameters and the quenching initial processing parameters are respectively corrected by the PSO-BP-ANN neural network model and the GA-BP-ANN neural network model until the predicted result, the actual result and the theoretical result are consistent, and the corrected surface strengthening processing parameters and quenching processing parameters are taken as the optimal processing parameters for the corresponding strengthening treatment of the cutting tooth until the cutting tooth processing is completed, and the two optimal processing parameters are imported into the corresponding surface strengthening intelligent database and laser strengthening intelligent database for corresponding data correction.

2. The method of claim 1, wherein the method is characterized by: After the cutting tooth processing is completed, 5-10% of the cutting tooth products are taken as samples for detection, and the real tooth head surface strengthening result and the real laser local quenching strengthening result of the cutting tooth samples and the corresponding processing parameters are recorded, which are taken as the verification data set of the PSO-BP-ANN neural network model and the GA-BP-ANN neural network model, and are imported into the surface strengthening intelligent database and the laser strengthening intelligent database for neural network data correction and optimization learning.

3. The method of claim 1, wherein the method is characterized by: The cutting tooth type is divided into a coal cutter cutting tooth, a heading machine cutting tooth and a rotary excavator cutting tooth.

4. The method of claim 1, wherein the method is characterized by: The surface strengthening method is a welding wire surfacing technology, a thermal spraying technology, a surface infiltration technology, a plasma cladding technology or a laser cladding technology.

5. The method of claim 1, wherein the method is characterized by: The processing parameters in the surface strengthening treatment in the step (1) are surface strengthening power, scanning speed and spot size, and the strengthening result parameters are strengthening layer height, dilution rate, temperature distribution and strengthening layer width; the processing parameters during the laser local quenching in the step (2) are scanning speed, laser power and spot diameter, and the strengthening result parameters are quenching layer hardness and quenching layer depth.

6. The method of claim 1, wherein the method is characterized by: The initial surface strengthening database and the initial laser strengthening database are obtained by experimental processing or numerical simulation.

7. The method of claim 1, wherein the method is characterized by: In the step (3), the PSO-BP-ANN neural network model combines the ANN neural network, the back propagation learning algorithm BP and the particle swarm optimization algorithm PSO, has an input layer, two hidden layers and an output layer, the input layer parameters are surface strengthening power, scanning speed and spot size, the Sigmoid function is used as the activation function from the input layer to the hidden layer and from the hidden layer to the hidden layer, the output layer parameters are strengthening layer height, strengthening layer width, temperature distribution and dilution rate, the number of hidden layer neurons is 9, and the activation function from the hidden layer to the output layer is a purelin linear function; the updating algorithm of the PSO-BP-ANN neural network model is represented as: 、 ， wherein: y’ n is a prediction of the surface reinforcement, including reinforcement layer height, dilution rate, temperature distribution, reinforcement layer width, xi is the surface reinforcement power, x2 is the scan speed and x3 is the spot size, v is the particle update velocity, j is the current iteration number; r 1 and r 2 is a random number distributed in (0, 1), c 1 and c 2 is a learning constant or acceleration factor, pop is a particle, pbest is the optimal individual extreme value of the last generation, gbest is the global optimal extreme value, ω is the weight factor.

8. The method of claim 1, wherein the method is characterized by: In step (3), the GA-BP-ANN neural network model combines ANN neural network, back propagation learning algorithm BP and genetic algorithm GA, has an input layer, a hidden layer and an output layer, wherein the input layer parameters are scanning speed, laser power and spot diameter, the output layer parameters are quenching layer depth and quenching layer hardness, the number of hidden layer neurons is 8, the input layer parameters are taken as groups in the GA-BP-ANN neural network, different parameter values correspond to different groups, each group is taken as an individual, each individual corresponds to a set of weight and threshold solutions, and the fitness calculation formula of each individual is: ， wherein: R i Hs(i) is the measured quenching layer hardness at any measurement point i, r i Hs(i) is the predicted quenching layer hardness at any measurement point i obtained from the ANN; H i Hd(i) is the measured quenching layer depth at any measurement point i, h i Hd(i) is the predicted quenching layer depth at any measurement point i obtained from the ANN, n is the total number of measurement points, fit 1 fitness for quenching layer hardness, In step (3), the determination process of the initial optimal weight and threshold of the BP-ANN neural network in the PSO-BP-ANN neural network model is as follows: a basic BP-ANN neural network structure is constructed, the processing parameters in the initial surface strengthening database are imported into the BP-ANN neural network structure, the processing parameters in the initial surface strengthening database and the corresponding strengthening results are imported into the PSO optimization algorithm module, and normalization processing is performed; then the initial optimal weight and threshold of the BP-ANN neural network are obtained by using the PSO optimization algorithm module embedded in the BP-ANN neural network structure, and the PSO optimization algorithm module processing process is as follows: first, 100 particles with different positions and speeds are added in the PSO optimization algorithm, each particle corresponds to a set of weight and threshold solutions, wherein: the position represents the weight and threshold of the processing parameters carried by the particle, and the speed represents the direction and distance of the movement of the particle in the next iteration; in the PSO optimization process, the position and speed of the particle are constantly updated by comparing the individual optimal solution and the global optimal solution of the particle, and after a set number of iterations, the weight and threshold carried by the particle are taken as the initial optimal weight and threshold and substituted into the BP-ANN neural network; 2 fitness for quenching layer depth, the smaller the fitness, the closer the predicted value of the data set to the measured value, the initial weight and threshold value are determined by selecting the individual with the smallest fitness.

9. The method of claim 1, wherein the method is characterized by: ​ The determination process of the initial optimal weight value and threshold value of the BP-ANN neural network in the GA-BP-ANN neural network model is as follows: first, the basic BP-ANN neural network structure is constructed, the machining parameters in the initial laser strengthening database are introduced into the BP-ANN neural network structure, the machining parameters in the initial laser strengthening database and the corresponding strengthening results are introduced into the GA algorithm module and are normalized, and then the initial optimal weight value and threshold value of the BP-ANN neural network are obtained by using the GA algorithm module embedded in the BP-ANN neural network structure. The GA algorithm module processing process is as follows: the input layer parameters are taken as groups, different parameter values correspond to different groups, each group is taken as an individual, and each individual corresponds to a set of weight value and threshold value solution. Thus, 100 individuals are added for neural network training. In the GA optimization process, the predicted result of each individual carrying the machining parameters is compared with the actual result. The smaller the difference between the predicted result and the actual result, the more suitable the set of weight value and threshold value solution corresponding to the individual. After a preset number of iteration operations, the weight value and threshold value carried by the individual with the smallest difference are taken as the initial optimal weight value and threshold value and are substituted into the BP-ANN neural network.

10. The method of claim 9, wherein the method further comprises: In step (6), the correction process of the surface strengthening initial machining parameters by the PSO-BP-ANN neural network model is as follows: ​ (1) After the BP-ANN neural network operation, when the actual result of the surface strengthening is inconsistent with the predicted result, the surface strengthening initial machining parameters and the actual result are added to the initial particle group of the PSO optimization algorithm module as training data, and the initial optimal weight value and threshold value are calculated again by repeating the PSO optimization algorithm module processing process in step (3) until the predicted result is consistent with the actual result; (2) When the predicted result of the surface strengthening is consistent with the actual result, the predicted result and the theoretical result of the surface strengthening are compared. When the predicted result is inconsistent with the theoretical result, the parameter optimization function of the back propagation learning algorithm BP in the BP-ANN neural network is activated. Specifically, the error between the predicted result and the theoretical result is calculated to obtain a loss function, and the loss function is a mean square error function MSE, that is: ， wherein: y n the theoretical result of surface strengthening, y’ n the predicted result of surface strengthening, n the number of neurons in the hidden layer of the PSO-BP-ANN neural network model, n take 1, 2, …, 9; the loss gradient between the predicted result and the theoretical result of surface strengthening can be obtained after taking the partial derivative of the loss function with respect to the input parameters before the corresponding weighted sum, that is: ， wherein: is the loss gradient between the input layer and the first hidden layer, is the loss gradient between the second hidden layer and the first hidden layer, is the loss gradient between the output layer and the second hidden layer, is the surface enhancement initial process parameter, i.e. the input parameter before the weighted summation into the first hidden layer, is the input parameter before the weighted summation into the second hidden layer, is the input parameter before the weighted summation into the output layer; ,The calculation process for is: ， ， wherein: x 1, x 2, x 3 is the surface strengthening initial processing parameter, i.e. ; , , is x 1, x 2, x 3 corresponding weight value; is the input amount input to the first hidden layer, , … is the input parameter before the weighted sum input to the second hidden layer, i.e. ; , … is , … corresponding weight value; is the input amount input to the second hidden layer; (3) After obtaining the loss gradient, the input parameters of each layer are updated and optimized using the gradient descent formula to obtain updated and optimized input parameters 、 and , 、 and , specifically: ， wherein: η is the learning rate, is the optimized surface-strengthening initial machining parameter, k Take 1, 2, 3, that is, the surface-strengthening power, the scanning speed and the spot size in the surface-strengthening machining parameter of the pick head, to realize the correction and optimization of the input parameter , until the predicted result, the actual result and the theoretical result are consistent, that is, the optimal surface-strengthening machining parameter can be output.

11. The method of claim 9, wherein the method further comprises: In step (6), the correction process of the quenching initial machining parameters by the GA-BP-ANN neural network model is as follows: ​ (1) After the BP-ANN neural network operation, when the actual result of the laser quenching is inconsistent with the predicted result, the quenching initial machining parameters and the actual result are added to the initial individual group of the GA algorithm module as training data, and the initial optimal weight value and threshold value are calculated again by repeating the GA algorithm module processing process in step (3) until the predicted result is consistent with the actual result; (2), when the prediction result of laser quenching is consistent with the actual result, comparing the prediction result of laser quenching with the theoretical result, when the prediction result is inconsistent with the theoretical result, the parameter optimization function of the back propagation learning algorithm BP in the BP-ANN neural network is activated, specifically: taking the theoretical result as the benchmark, the error between the prediction result and the theoretical result is calculated to obtain the loss function, and the loss function in the GA-BP-ANN neural network is the mean square error function That is: ， wherein: z n is the theoretical result of laser quenching, z’ n is the predicted result of laser quenching, n is the number of neurons in the hidden layer of the GA-BP-ANN neural network model, n take 1, 2, …, 8; the loss gradient between the predicted result and the theoretical result of laser quenching can be obtained after taking the partial derivative of the loss function and the input parameters before the corresponding weighted summation, that is: ， wherein: is the loss gradient between the input layer and the hidden layer, is the loss gradient between the output layer and the hidden layer, is the laser power and the scanning speed in the quenching initial process parameters, i.e. the input parameters before the weighted sum into the hidden layer, is the input parameter before the weighted sum into the output layer; The calculation process is as follows: ， wherein: λ 1, λ 2 is the laser power and the scanning speed in the quenching initial processing parameter, i.e. ; , is λ 1, λ 2 corresponding weight; is the input amount input to the hidden layer; (3) After obtaining the loss gradient, the input parameters of each layer are updated and optimized by using the gradient descent formula to obtain the updated and optimized input parameters , , , Specifically:​ ， wherein: is a learning rate, is the laser power and scanning speed in the optimized quenching initial processing parameters, and the correction and optimization of the input parameters are realized until the predicted results, actual results and theoretical results are consistent, and the optimal quenching processing parameters can be output.

Images