A shield machine cutter fatigue life prediction method based on laser cladding technology

Abstract

The application discloses a shield machine cutter fatigue life prediction method based on a laser cladding technology, a plurality of coatings are processed on the surface of a preheated shield machine cutter by adopting a laser cladding process; factors and levels of laser cladding technical parameters are selected by adopting a response surface method; laser cladding coating metallographic samples of the shield machine cutter are prepared; a quality evaluation coefficient of the shield machine cutter after laser cladding is calculated; microscopic hardness data of the shield machine cutter after laser cladding are obtained; mechanical property parameter data of the laser cladding coating sample, fatigue life parameters of the laser cladding coating sample are counted, and an S-N curve is drawn; corresponding performance parameters are taken as inputs, and a laser cladding shield machine cutter fatigue life prediction model based on a deep neural network method distributed by a stacked auto-encoder is constructed. The application can significantly improve the sensitivity of the fatigue life prediction model, thereby realizing laser cladding shield machine cutter fatigue life prediction based on a small data set.

Description

Technical Field

[0001] This invention relates to the field of laser cladding technology, and more specifically to a method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology. Background Technology

[0002] The cutterhead of a tunnel boring machine (TBM) is a critical component. Currently, heat treatment is the primary method used to improve the hardness and fatigue life of the cutterhead. However, due to uneven distribution of the quenched layer after heat treatment, the improvement in fatigue life is not significant. Therefore, laser cladding, with its significant advantages, is considered an effective solution to this technical problem. Laser cladding is an advanced surface modification technology. It boasts advantages such as concentrated energy, a small heat-affected zone, minimal damage to the substrate, and high processing precision, enabling efficient repair of complex parts and thus possessing unique advantages among various additive manufacturing technologies. The working principle of laser cladding involves adding cladding material to the surface of a substrate. A high-energy laser beam acts on the cladding material, causing metallurgical fusion, which allows for rapid cooling and solidification, forming a high-performance and dense cladding layer. Therefore, laser cladding technology is currently widely used in aerospace, automotive, and tool manufacturing industries, particularly in conductor damage tolerance design. It enables green remanufacturing, significantly improves product fatigue life, and has high socio-economic benefits and application value.

[0003] In the field of laser cladding technology, the selection of laser cladding technical parameters directly determines the quality of the cladding layer, thus affecting the fatigue life of the tunnel boring machine (TBM) cutterhead. Although the fatigue life of TBM cutterheads with laser cladding coatings can be predicted using theoretical models, traditional model prediction methods typically only allow predictions under a single working condition. When studying the impact of different technical parameters on fatigue life, traditional methods have many shortcomings. Therefore, it is necessary to propose another method for predicting the fatigue life of TBM cutterheads based on laser cladding technology. In recent years, machine learning methods have been applied in fields such as laser cladding material performance prediction, showing significant advantages in prediction accuracy and green economic development. Professor Li Hua theoretically proposed a continuous damage mechanics model with additive manufacturing effects and collected over 500 sets of data to train the machine learning model, employing two commonly used machine learning models—artificial neural networks and random forests—for fatigue life prediction. However, machine learning typically requires a large dataset as input, while fatigue test data consists of small sample data, resulting in low accuracy for data-driven fatigue life prediction methods. For these reasons, it is currently necessary to develop a high-precision fatigue life prediction method for small sample datasets. In addition to fatigue test data, which can be used as the dataset for fatigue life prediction models, laser cladding technology parameters and the quality evaluation coefficients of the cladding coating (such as dilution rate, effective area per unit area, etc.) also greatly affect the fatigue life of the tunnel boring machine cutterhead. If these are also used as the dataset for the model, the accuracy of the prediction model can be improved.

[0004] Therefore, this invention proposes a method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology. The method uses laser cladding technology parameters, quality evaluation coefficients, microhardness and mechanical property parameters as inputs, and fatigue life as output. A deep neural network method with stacked autoencoders is used to construct a fatigue life prediction model for the tunnel boring machine cutterhead based on laser cladding technology, thereby achieving high-precision fatigue life prediction of the tunnel boring machine cutterhead after laser cladding. Summary of the Invention

[0005] To address the shortcomings of traditional model prediction methods and machine learning methods in predicting the fatigue life of tunnel boring machine cutterheads after laser cladding, the present invention aims to propose a method for predicting the fatigue life of tunnel boring machine cutterheads based on laser cladding technology, which can predict the fatigue life of tunnel boring machine cutterheads after laser cladding according to the established model.

[0006] To achieve the above objectives, the present invention adopts the following design scheme:

[0007] Multiple coatings were applied to the surface of a tunnel boring machine (TBM) cutterhead using laser cladding technology. The laser-clad TBM cutterhead was then cut to prepare metallographic specimens of the laser-clad coating. The quality evaluation coefficient of the laser-clad TBM cutterhead was calculated. The microhardness of the clad coating cross-section was measured using a hardness tester to obtain microhardness data. Tensile tests were conducted on the laser-clad specimens, and the mechanical property parameters of the laser-clad specimens were statistically analyzed. Fatigue tests were performed on the laser-clad specimens, and the fatigue life parameters and SN curves of the laser-clad specimens were statistically analyzed. Based on the aforementioned laser cladding technology parameters, quality evaluation coefficient, microhardness, mechanical property parameters, and fatigue life, a fatigue life prediction model for the TBM cutterhead based on laser cladding technology was constructed using a deep neural network method with a stacked autoencoder.

[0008] Furthermore, the shield machine cutterhead needs to be polished, cleaned, and preheated before cladding, and the processing of various coatings is carried out under different laser cladding technical parameters.

[0009] Furthermore, the different laser cladding technical parameters refer to the factors and levels of the technical parameters selected based on the response surface methodology.

[0010] Furthermore, the sampling direction of the laser cladding coating is perpendicular to the laser scanning direction. After sampling, the cross-section of the cladding layer needs to be polished and etched.

[0011] Furthermore, the quality evaluation coefficient includes the dilution rate and the effective area per unit area.

[0012] Furthermore, the calculation of the quality evaluation coefficient relies on sufficient sample parameter data, such as the geometry of the molten pool (melt width, melt height, and melt depth).

[0013] Furthermore, the model comprises an input layer, a hidden layer, and an output layer. The four types of parameters—laser cladding technology parameters, quality evaluation coefficients, microhardness, and mechanical property parameters—serve as the input layer of the model, while fatigue life serves as the output layer.

[0014] Furthermore, the stacked autoencoders are distributed among the layers of the deep neural network for pre-training the input data. After pre-training on the data from all the stacked autoencoders, the initial weights and biases obtained from each layer are assigned to the corresponding layers in the deep neural network, and modeling begins. After obtaining the neural network structure with the highest accuracy, a corresponding model function for the optimal network, composed of weight and bias values, is generated. The fatigue life prediction is achieved using the obtained generative model function.

[0015] The present invention can achieve the following beneficial effects:

[0016] 1. The method for predicting the fatigue life of the cutterhead of a tunnel boring machine after laser cladding based on a deep neural network with stacked autoencoder allocation can simultaneously consider multiple factors affecting fatigue life, including laser cladding technical parameters, quality evaluation coefficient, hardness, and mechanical property parameters.

[0017] 2. By establishing a suitable deep neural network model, fatigue life prediction of small datasets can be achieved, reducing material and time costs while quickly and accurately predicting the fatigue life of the tunnel boring machine cutterhead after laser cladding.

[0018] 3. Factors such as laser cladding technology parameters and quality evaluation coefficients, when included as part of the model dataset, can improve the sensitivity of deep neural network prediction models assigned by stacked autoencoders.

[0019] 4. Compared with machine learning, the deep neural network method based on stacked autoencoder assignment predicts the mean absolute percentage error of the fatigue life of the tunnel boring machine cutterhead after laser cladding. Attached Figure Description

[0020] Figure 1 Flowchart of a method for predicting the fatigue life of tunnel boring machine cutterheads based on laser cladding technology;

[0021] Figure 2 Schematic diagram of center-feeding laser cladding process;

[0022] Figure 3 Schematic diagram of the geometric dimensions of the laser cladding molten pool;

[0023] Figure 4 Input and output diagrams of the fatigue life prediction model;

[0024] Figure 5 A schematic diagram of a deep neural network model for a five-layer stacked autoencoder. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are descriptive and not limiting, and should not be construed as limiting the scope of protection of the present invention.

[0026] Figure 1 This is a flowchart of a fatigue life prediction method for a laser cladding 42CrMo cutterhead according to the present invention. The following is in conjunction with... Figure 1 The method of the present invention will be described.

[0027] This invention predicts the fatigue life of laser-clad 42CrMo cutterheads based on a deep neural network assigned by a stacked autoencoder.

[0028] In this invention, the process includes polishing and cleaning the cutterhead of the tunnel boring machine.

[0029] In this invention, the preheating temperature of the tunnel boring machine cutterhead is 300-350℃.

[0030] In this invention, preheating the substrate surface can reduce the temperature gradient of the cladding process, thereby effectively improving the cladding quality.

[0031] In this invention, a center-feeding method is used to laser cladize the surface of the preheated tunnel boring machine cutterhead. The powder in the center-feeding method flows in a single stream, thus eliminating collisions and scattering between powder streams from different directions. This effectively improves powder utilization and meets the needs of green economic development. Figure 2 As shown.

[0032] In this invention, the technical parameters of laser cladding include: laser power, scanning speed, powder feeding rate, and spot radius.

[0033] In this invention, the response surface methodology is used to select the factors and levels of technical parameters, and various processing schemes for laser cladding coatings are designed.

[0034] In this invention, the designed laser cladding technology parameters are used as inputs to the fatigue life prediction model.

[0035] In this invention, the cutterhead of the tunnel boring machine after laser cladding is cut along a direction perpendicular to the laser scanning direction, and then the cross-section of the cladding layer is polished and etched to prepare a metallographic sample of the cutterhead of the tunnel boring machine after laser cladding.

[0036] In this invention, a metallographic microscope is used to observe the cross-section of the cladding layer after polishing and etching, and the geometric dimensions of the molten pool are measured under different laser cladding technical parameters. The dilution rate and the effective area per unit area are used as quality evaluation coefficients, such as... Figure 3 As shown, the quality evaluation coefficient is determined by equations (1) and (2). Using the quality evaluation coefficient as input to the fatigue prediction model can effectively improve the prediction accuracy of the model.

[0037]

[0038]

[0039] In this invention, the microhardness of the cladding section is measured using a Vickers hardness tester, wherein the section must be polished and etched. The microhardness data of the tunnel boring machine cutterhead after laser cladding is obtained and used as input for the prediction model.

[0040] In this invention, a standard tensile specimen with a laser cladding coating is prepared in accordance with the national standard GB / T 41477-2022 Test Method for Mechanical Properties of Laser Cladding Repaired Metal Parts, and a tensile test is performed on the laser cladding coated specimen. The mechanical property parameters (yield strength, tensile strength, elongation, and reduction of area) of the laser cladding coated specimen are statistically analyzed and used as input for the prediction model.

[0041] In this invention, standard fatigue specimens with laser cladding coatings are prepared according to the national standard GB / T 41477-2022 Test Method for Mechanical Properties of Laser Cladding Repaired Metal Parts. Tensile-compression fatigue tests are then conducted on the laser cladding coated specimens with a stress ratio of R = -1. The fatigue life of the laser cladding coated specimens is statistically analyzed and SN curves are plotted. The stress amplitude is used as the input to the prediction model, and the fatigue life is used as the output of the prediction model.

[0042] In this invention, based on the aforementioned laser cladding technology parameters, quality evaluation coefficients, microhardness, mechanical property parameters, and fatigue life, a fatigue life prediction model for the cutterhead of a tunnel boring machine (TBM) based on laser cladding technology is constructed using a deep neural network method with stacked autoencoders. After pre-training the data from all autoencoders, the initial weights and biases obtained from each layer are assigned to the corresponding layers in the deep neural network, and modeling begins. After obtaining the neural network structure with the highest accuracy, the corresponding model function of the optimal network composed of weights and biases is generated. The fatigue life is predicted using the obtained generated model function.

[0043] In this invention, the inputs and outputs considered are as follows: Figure 4 As shown. Stacked autoencoders are distributed between layers of a deep neural network for pre-training on the input data, such as... Figure 5 As shown. The mean absolute percentage error (MASE) refers to the average deviation between the predicted value and the experimental value. It can normalize the error of each prediction result, and the calculation method is shown in Equation (3). Therefore, in order to evaluate the fatigue life prediction model under different methods, it is used as an indicator to quantify the accuracy of the prediction model. The smaller the MASE, the higher the prediction accuracy of the model. When it is greater than 20%, the parameters of the prediction model are adjusted.

[0044]

[0045] The technical solution of this invention will be clearly and completely analyzed and described below with reference to Implementation Example 1. The specific implementation steps of this method include:

[0046] Case 1

[0047] Step 1: Polish the tunnel boring machine (TBM) cutterhead with silicon carbide (SiC) sandpaper, then clean the polished cutterhead with alcohol and acetone; subsequently, preheat the surface of the TBM cutterhead to 300–350°C; finally, design the laser cladding technology parameters (as the A input of the fatigue life prediction model) based on the response surface methodology, wherein the laser power is 1800W–2800W, the scanning speed is 120mm / min–240mm / min, the powder feeding rate is 5g / min–15g / min, the spot radius is 3–5mm, the protective gas is argon, and the gas flow rate is 10L / min. Under these parameters, a laser cladding coating is processed on the surface of the TBM cutterhead. According to a specific embodiment of the present invention, laser cladding using a center-feeding method is employed. The cutterhead material is 42CrMo, and the laser cladding powder is Ni60A-25wt.%WC, with the composition shown in Table 1.

[0048] Table 1

[0049]

[0050] Step 2: Cut the cutterhead of the tunnel boring machine after laser cladding along a direction perpendicular to the laser scanning direction, and then polish and etch the cross-section of the cladding layer to prepare a metallographic sample of the cutterhead of the tunnel boring machine after laser cladding.

[0051] Step 3: Observe the cross-section of the cladding layer after polishing and etching using a metallographic microscope, measure the geometric dimensions of the molten pool under different laser cladding technical parameters, and calculate the dilution rate and effective area per unit area of ​​the cladding layer according to equations (1) and (2) and use them as quality evaluation coefficients. Use the quality evaluation coefficients as the B input of the fatigue prediction model.

[0052] Step 4: Use a Vickers hardness tester to measure the microhardness of the polished and etched cladding section, and use the microhardness data as the C input to the fatigue prediction model.

[0053] Step 5: Prepare standard tensile specimens with laser cladding coatings according to the national standard GB / T 41477-2022 Test Method for Mechanical Properties of Laser Cladding Repaired Metal Parts and conduct tensile tests to obtain the mechanical property parameters (yield strength, tensile strength, elongation, and reduction of area) of the laser cladding coating specimens under different technical parameters, and use them as the D input of the fatigue prediction model.

[0054] Step Six: Prepare standard fatigue specimens with laser cladding coatings according to the national standard GB / T 41477-2022 Test Method for Mechanical Properties of Laser Cladding Repaired Metal Parts. Conduct tensile-compressive fatigue tests on the laser cladding coated specimens with a stress ratio of R = -1. Statistically analyze the fatigue life of the laser cladding coated specimens and plot the SN curve. Use the stress amplitude as the D input of the prediction model and the fatigue life as the output of the prediction model.

[0055] Step 7: Establish a fatigue life prediction model based on laser cladding technology parameters, quality evaluation coefficients, microhardness, and mechanical property parameters according to the deep neural network method assigned by the stacked autoencoder. In this case, when developing a 5-layer fully interconnected stacked autoencoder with an input layer, 3 hidden layers, and an output layer, 4 stacked autoencoders are required. The number of neurons used in each stacked autoencoder is equal to the number used in the corresponding deep neural network layer. Therefore, in terms of the number of layers and neurons, the deep neural network assigned to the 5-layer stacked autoencoder with a 12+(24+12+9)+1 structure and the encoder with a 12+(24)+12, 24+(12)+24, 12+(9)+12, 9+(1)+9 structure are assigned. After feeding the data back to the deep neural network, the autoencoder captures the input as its own input and output data. Then, the data is processed, and the output of the first hidden layer is transmitted as a new input to the next encoder, until the last encoder. After pre-training the data using all the autoencoders, the β layer of each layer is... n The initial weights and biases implemented in (0) are assigned to the corresponding layers in the deep neural network, and modeling begins. After obtaining the neural network structure with the highest accuracy, the corresponding model function of the optimal network, consisting of weights and biases, is generated for further sensitivity and parameter analysis. The model function of the deep neural network assigned to the 5-layer stacked autoencoder in this case can be generated as follows:

[0056] α1=f1(β1k+b1) (4.1)

[0057] α2=f2(β2k1+b2) (4.2)

[0058] α3=f3(β3k2+b3) (4.3)

[0059] α4=f4(β4k3+b4) (4.4)

[0060] α5=f5(β5k4+b5) (4.5)

[0061] α1, α2, α3 and α4 are the output parameters of layers 1 to 4, respectively.

[0062] Step 8: The N function maps the 12 inputs to the fatigue life n(1) as the output, as shown in Equation (5). This achieves fatigue life prediction of the shield machine cutterhead after laser cladding based on a small dataset. Furthermore, sensitivity analysis is performed using the obtained network weight matrix via the Garson equation, as shown in Equation (6).

[0063] N(n(1))=f5(β5k4(β4k3(β3k2(β2k1(β1k+b1)+b2)+b3)+b4)+b5) (5)

[0064]

[0065] Where L j M represents the relative importance of the j-th input in relation to the output. i M is the number of input neurons. h is the number of hidden neurons, T is the connection weight, and the superscripts i, h, and o represent the input, hidden, and output neurons, respectively.

[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology, characterized in that: Step 1: Apply multiple coatings to the surface of the tunnel boring machine cutterhead using laser cladding technology; Step 2: Cut the laser-clad shield machine cutterhead to prepare a metallographic sample of the laser-clad coating on the shield machine cutterhead; Step 3: Calculate the quality evaluation coefficient of the tunnel boring machine cutterhead after laser cladding; Step 4: Use a hardness tester to measure the microhardness of the cross section of the cladding coating on the shield machine cutterhead to obtain the microhardness data of the shield machine cutterhead after laser cladding. Step 5: Perform tensile tests on the samples with laser cladding coatings and collect data on the mechanical property parameters of the laser cladding coating samples. Step 6: Conduct fatigue tests on the samples with laser cladding coatings, statistically analyze the fatigue life parameters of the laser cladding coating samples, and plot the SN curve; Step 7: Based on the aforementioned laser cladding coating metallographic specimens, quality evaluation coefficients, microhardness, mechanical property parameters, and fatigue life, a fatigue life prediction model for the tunnel boring machine cutterhead after laser cladding is constructed using a deep neural network method with stacked autoencoders. Step 8: Prediction of fatigue life of the tunnel boring machine cutterhead after laser cladding.

2. The method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology according to claim 1, characterized in that: In step one, the tunnel boring machine cutterhead needs to be polished, cleaned, and preheated before laser cladding. The processing of various coatings is carried out under different laser cladding technical parameters.

3. The method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology according to claim 2, characterized in that: The factors and levels of laser cladding technology parameters were selected using the response surface methodology, and various processing schemes for laser cladding coatings were designed.

4. The method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology according to claim 1, characterized in that: In step two, the sampling direction of the laser cladding coating is perpendicular to the laser scanning direction. After sampling, the cross-section of the cladding layer is polished and etched.

5. The method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology according to claim 1, characterized in that: The quality evaluation coefficient mentioned in step three includes the dilution rate and the effective area per unit area.

6. The method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology according to claim 5, characterized in that: The calculation of the quality evaluation coefficient depends on the sample parameter data.

7. The method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology according to claim 1, characterized in that: The model constructed in step seven includes an input layer, a hidden layer, and an output layer. The four types of parameters—laser cladding technology parameters, quality evaluation coefficients, microhardness, and mechanical property parameters—serve as the input layer of the model, while fatigue life serves as the output layer.

8. The method for predicting the fatigue life of a tunnel boring machine cutterhead based on laser cladding technology according to claim 7, characterized in that: After pre-training on all the autoencoder data, the initial weights and biases obtained from each layer are assigned to the corresponding layers in the deep neural network. After obtaining the neural network structure with the highest accuracy, the corresponding model function of the deep neural network, consisting of weights and bias values, is generated.

Images