Disc cutter cutting specific energy consumption prediction and optimization method based on physical constraint neural network

Abstract

The invention relates to the field of mining machinery, in particular to a method for predicting disc cutter specific energy consumption and selecting an optimal specific energy consumption parameter. Firstly, an SEC mathematical model containing key parameters such as the cutting depth, the cutting speed, the feeding rate, the cutting angle, the cutter radius and the friction coefficient is deduced by analyzing the cutter cutting type, the rock damage form and the cutter load; secondly, generating a data set by utilizing finite element simulation and carrying out feature analysis; then, a PINN model is constructed and trained, and the SEC prediction precision is improved in combination with a physical rule and data driving; and finally, based on a model prediction result, selecting an optimal parameter combination to realize energy consumption optimization. The result shows that the method is superior to a traditional method in the aspects of prediction accuracy and parameter optimization, the energy consumption of the disc cutter is remarkably reduced, the working efficiency is improved, and good economic and social benefits are achieved.

Description

Technical Field

[0001] The invention relates to the field of mining machinery, and in particular to a method for predicting and optimizing the energy consumption of disc cutter cutting ratio based on a physical constraint neural network. Background Art

[0002] As a key cutting tool in the mining process, the working efficiency and energy consumption level of disc cutters directly affect the economy and sustainability of the entire mining process. Specific energy consumption (SEC) is an important indicator to measure the working efficiency of disc cutters, reflecting the energy consumed per unit cutting volume. Accurately predicting and optimizing the specific energy consumption of disc cutters is of great significance to improving mining efficiency and reducing energy consumption costs.

[0003] At present, the prediction of disc cutter specific energy consumption mainly relies on empirical formulas or analysis methods based on a single factor. These methods are usually based on limited experimental data or engineering experience, and predict SEC by establishing linear or nonlinear mathematical models. For example, common empirical formulas may only consider the influence of a single parameter such as uniaxial compressive strength, cutting depth or cutting speed on SEC. However, in actual work, the specific energy consumption of disc cutters is affected by a combination of factors, including but not limited to uniaxial compressive strength, cutting depth, cutting speed, feed rate, cutting angle, tool radius and friction coefficient. There are complex nonlinear relationships between these parameters, and traditional empirical formulas are difficult to fully and accurately reflect their impact on SEC. In addition, the analysis method based on a single factor ignores the interaction between parameters, resulting in the accuracy and applicability of the prediction model being limited. In a complex mining environment, a single factor model is difficult to cope with the changing working conditions and material properties, and it is difficult to achieve high-precision prediction and real-time optimization control of SEC.

[0004] In recent years, with the development of artificial intelligence and machine learning technologies, data-driven prediction methods have gradually been applied to the field of mining machinery. However, pure data-driven methods often require a large amount of high-quality data for training, and lack inherent constraints on physical laws, which can easily lead to overfitting or poor performance on unseen data. In response to the above problems, hybrid models that combine physical laws with data-driven have become a research hotspot. As an emerging deep learning method, the Physics-Informed Neural Network (PINN) can enhance the physical consistency and generalization ability of the model on the basis of data-driven by embedding physical laws into the loss function of the neural network. PINN can not only use existing data for efficient training, but also use physical constraints to improve the prediction accuracy and reliability of the model under complex working conditions. Summary of the invention

[0005] The present invention provides a method for predicting and optimizing the specific energy consumption of disc cutters based on a physical constraint neural network, aiming to achieve high-precision prediction and optimization of the specific energy consumption (SEC) of disc cutters by combining the advantages of physical laws and data-driven. The specific contents are as follows: S1. Establishment of mathematical model of specific energy consumption The present invention firstly derives the mathematical expression of the disc cutter's specific energy consumption by analyzing the disc cutter's cutting type, rock damage form and cutter load. Taking into account key parameters such as cutting depth (d), cutting speed (v), feed rate (f), cutting angle (θ), cutter radius (r) and friction coefficient (μ), the following specific energy consumption formula is established: (1) in, q is the wedge surface compressive stress, R is the tool radius, h is the equivalent thickness of the tool wedge surface and the rock, λ is the angle between the feed direction and the wedge surface, f k is the equivalent friction coefficient between the tool wedge surface and the rock, K c is the repetitive shear coefficient, h y is the tooth edge equivalent height, α is the tool wedge angle, β is the tool attitude inclination angle, d is the cutting depth, w is the cutting width, f is the feed rate, t is the cutting time, v is the cutting speed, T 0 is the shear strength of the rock.

[0006] S2. Dataset establishment and data analysis In order to train and verify the prediction model proposed in the present invention, it is necessary to generate an original data set containing the above key parameters through finite element simulation. The specific steps include: a. Parameter range setting: setting the uniaxial compressive strength (UCS) between 50 and 250 MPa, the cutting depth (d) between 2 and 10 mm, the cutting speed (v) between 100 and 500 mm / s, the feed rate (f) between 0.1 and 1.0 mm / s, the cutting angle (θ) between 15 and 30 degrees, the tool radius (r) between 10 and 50 mm, and the friction coefficient (μ) between 0.05 and 0.2, and designing the experimental scheme through orthogonal experiments. b. Simulation: Using finite element simulation software, simulate the cutting process of the disc cutter under different working conditions to generate a large-scale data set including the above parameters. c. Data preprocessing: Clean the generated data set, remove outliers and noise, and ensure the integrity and accuracy of the data. Subsequently, MinMaxScaler is used to normalize the feature matrix X and the target vector y, and all parameters are scaled to the range of [0,1] to improve the efficiency and stability of model training. d. Data analysis: Perform correlation analysis on the preprocessed data set to analyze the correlation between parameters.

[0007] S3. PINN model construction The construction of a neural network model (PINN) based on physical constraints is the core step of the present invention. The specific steps are as follows: a. Network structure design: Design a neural network structure containing multiple hidden layers, each layer contains 128 neurons, and the activation function uses the Tanh function to capture the complex nonlinear relationship between parameters. The output layer uses the Softplus function to ensure that the predicted specific energy consumption value is positive. b. Physical constraint integration: The mathematical formula of specific energy consumption is used as a physical constraint condition and embedded in the loss function of the neural network. Specifically, the physical loss is achieved by calculating the mean square error (MSE) between the predicted value of the model and the calculated value of the mathematical formula to ensure that the model output conforms to the laws of physics. c. Model initialization: Use the PyTorch framework to implement the PINN model, initialize the network parameters, and set the optimizer to the Adam optimizer, the learning rate is set to 1e-3, and the weight attenuation factor is 1e-4 to prevent overfitting. d. Loss function and optimizer design, in order to integrate physical laws and data-driven methods, the present invention designs a comprehensive loss function. The loss function includes two parts: data loss and physical loss: (2) Loss date For data loss, Loss physics For physical losses, λ is a hyperparameter that controls the weight of the physical constraint loss; the data-driven loss expression is as follows: (3) SEC pred,k is the predicted specific energy consumption output by the model, SEC ture,k is the specific energy consumption obtained by simulation, N is the number of data sets; the physical constraint loss measures the error between the SEC predicted by the model and the SEC calculated by the physical formula, and the expression is: (4) SEC phys,k Calculate the specific energy consumption for the physical formula; the physical constraint loss consists of multiple physical penalty terms, expressed as: (5) The specific penalties are: ,

[0008] In addition, the Adam optimizer is used for parameter optimization, combined with the learning rate scheduling strategy (ReduceLROnPlateau) and weight decay (weight_decay) to prevent overfitting and improve the generalization ability of the model.

[0009] S4. Model training During the model training process, the batch training method is used to optimize the model parameters through multiple epoch iterations. The specific steps are as follows: a. Data input: The normalized training set data is input into the PINN model for forward propagation to obtain the predicted specific energy consumption value. b. Loss calculation: Calculate the total loss function, including data loss and physical loss. c. Back propagation and optimization: The gradient is calculated through the back propagation algorithm, and the model parameters are updated using the Adam optimizer. At the same time, the learning rate scheduling strategy (ReduceLROnPlateau) is adopted. When the validation set loss no longer decreases within 20 consecutive epochs, the learning rate is halved to promote model convergence. d. Early stopping strategy: Set the early stopping patience value to 50. When the validation set loss fails to improve within 50 consecutive epochs, terminate the training in advance to prevent the model from overfitting. e. Model preservation: During the training process, save the model parameters with the smallest validation set loss to ensure that the final model has the best generalization ability.

[0010] S5. Model evaluation After the training is completed, the saved optimal model parameters are loaded and the test set data is predicted. The prediction performance of the model is evaluated by calculating the mean square error (MSE) and the coefficient of determination (R²) on the test set. The experimental results show that the PINN model is superior to the traditional empirical formula and single factor analysis method in terms of prediction accuracy and generalization ability, and can effectively reduce the energy consumption of the disc cutter and improve work efficiency. In order to fully understand the performance of the model and the impact of parameter comparison energy consumption, the present invention adopts a variety of visualization methods: a. Training and verification loss curve: Display the loss change of the model during the training process, verify the convergence of the model and the training effect. b. Prediction result comparison chart: Compare the actual SEC value and the predicted SEC value on the test set through a scatter plot to intuitively display the prediction accuracy of the model. c. Parameter correlation analysis: Draw the correlation of each key parameter to show the degree of influence of parameter changes on energy consumption. d. SEC heat map: Draw a heat map of the combination of specific energy consumption and two key parameters to assist in identifying the optimal working parameter area.

[0011] Beneficial effects of the technical solution: 1. Improve prediction accuracy: By combining physical laws with the data-driven PINN model, the shortcomings of the traditional empirical formula or single factor analysis method in prediction accuracy are overcome, and a high-precision prediction of the specific energy consumption (SEC) of the disc cutter is achieved.

[0012] 2. Comprehensive consideration of the influence of multiple factors: Comprehensive consideration of multiple key parameters such as uniaxial compressive strength, cutting depth, cutting speed, feed rate, cutting angle, tool radius and friction coefficient, fully capturing the complex nonlinear relationship between various parameters, and providing a more comprehensive and detailed prediction model.

[0013] 3. Achieve energy consumption optimization: By using the trained PINN model to predict SEC under different parameter combinations and selecting the optimal working parameter combination, the energy consumption of the disc cutter operation is effectively optimized, which significantly reduces energy consumption costs and improves work efficiency.

[0014] 4. Enhanced model adaptability: This method is suitable for mining environments with different working conditions and material properties. It has good generalization ability and adaptability, and can maintain stable prediction performance in changing practical application scenarios.

[0015] 5. Significant economic and social benefits: By effectively reducing the energy consumption of disc cutter operation and improving work efficiency, it not only reduces energy consumption and operating costs, but also promotes the sustainable development of mining, with significant economic and social benefits.

[0016] 6. Visual analysis to assist decision making: A variety of visualization methods are used to conduct a comprehensive analysis of model performance and parameter impact, helping users to deeply understand the impact mechanism of each parameter on SEC, and providing a scientific basis and data support for parameter adjustment and optimization in actual engineering applications.

[0017] 7. Prevent overfitting and improve generalization ability: By introducing physical constraints, the physical consistency of the model is enhanced, the overfitting problem that may occur in pure data-driven methods is avoided, and the generalization ability of the model on unseen data is improved.

[0018] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following is a detailed description in combination with the preferred embodiments and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the drawings of the embodiment are briefly introduced below.

[0020] Figure 1 A flow chart of the present invention is shown.

[0021] Figure 2 are the training and validation loss curves.

[0022] Figure 3 This is a comparison chart of the prediction results.

[0023] Figure 4 This is parameter correlation analysis.

[0024] Figure 5 The thermodynamic diagram of specific energy consumption, uniaxial compressive strength (UCS) and cutting depth (d).

[0025] Figure 6 The thermal diagram of specific energy consumption versus cutting speed (v) and feed rate (f).

[0026] Figure 7 The thermal diagram of specific energy consumption versus cutting angle (θ) and tool radius (r).

[0027] Figure 8 Schematic diagram of disc cutter cutting rock. DETAILED DESCRIPTION

[0028] Figure 1The overall flow chart of the disc cutter specific energy consumption prediction and optimization method based on physical constraint neural network (PINN) of the present invention is presented. The flow chart covers the following main steps: data generation and preprocessing, PINN model construction, loss function and optimizer design, model training, model evaluation, and visualization and parameter analysis. Through the flow chart, the working principle and operation process of the method are clearly demonstrated, providing intuitive guidance for subsequent specific implementation.

[0029] Figure 2 The figure shows the training and validation loss curves. This figure shows how the loss values ​​(Loss) on the training set and validation set change with the training epochs during the training process. In the initial stage, the loss value is high. As the training progresses, the loss value gradually decreases, indicating that the model is constantly optimizing parameters and improving prediction accuracy. When the validation loss curve tends to be stable and close to the training loss curve, it indicates that the model has reached a good convergence state and has a high generalization ability.

[0030] Figure 3 The figure is a comparison chart of the prediction results. The figure shows the scatter distribution of the actual specific energy consumption (SEC) value on the test set and the SEC value predicted by the model. The red dotted line represents the ideal prediction, that is, the predicted value is equal to the true value. Most of the data points are concentrated near the dotted line, indicating that the model has a high prediction accuracy. A small number of data points that deviate from the dotted line reflect the prediction error of the model on some samples, but the overall trend shows that the model has excellent performance.

[0031] Figure 4 The figure is a parameter correlation analysis diagram. This figure shows the correlation between each key parameter (uniaxial compressive strength, cutting depth, cutting speed, feed rate, cutting angle, tool radius and friction coefficient) and specific energy consumption (SEC). The influence of each parameter on SEC and their mutual relationship are clearly shown in the form of scatter plots and correlation coefficient matrices. For example, cutting speed and feed rate are significantly positively correlated with SEC, while the correlation between cutting depth and SEC is weak.

[0032] Figure 5 The figure is a heat map of specific energy consumption, uniaxial compressive strength (UCS) and cutting depth (d). The heat map represents the distribution of SEC values ​​under different combinations of UCS and cutting depth by color depth. It can be observed from the figure that the SEC value is higher under the combination of higher UCS and cutting depth, while the SEC value is lower under the combination of lower UCS and cutting depth. This figure helps to identify the trend of energy consumption under different material strength and cutting depth conditions, providing an intuitive basis for parameter optimization.

[0033] Figure 6 The heat diagram of specific energy consumption, cutting speed (v) and feed rate (f). Figure 5, the heat map shows the distribution of SEC values ​​under different cutting speed and feed rate combinations. It can be found that with the increase of cutting speed and feed rate, the SEC value shows an upward trend. This shows that under high speed and high feed rate conditions, the energy consumption of the disc cutter increases significantly, suggesting that in practical applications, the relationship between speed and energy consumption needs to be weighed to achieve optimization.

[0034] Figure 7 The figure is a heat map of specific energy consumption versus cutting angle (θ) and tool radius (r). The heat map shows the distribution of SEC values ​​for different cutting angle and tool radius combinations. The results show that an increase in the cutting angle and tool radius will lead to an increase in the SEC value, but the degree of influence varies. This figure helps to identify the energy consumption performance at different cutting angles and tool sizes, and supports the selection of the optimal parameter combination under specific working conditions.

[0035] Figure 8 The diagram is a schematic diagram of the disc cutter cutting rock. R is the radius of the cutter, h is the equivalent thickness between the tool wedge surface and the rock, λ is the angle between the feed direction and the wedge surface, α is the tool wedge surface angle, β is the tool attitude inclination angle, t is the cutting time, and v is the cutting speed.

Claims

1. A method for predicting the specific energy consumption (SEC) of a disc cutter and selecting the optimal specific energy consumption parameter, characterized in that: The following steps are involved: Derivation of specific energy consumption formula: By analyzing the tool cutting type, rock failure form and tool load, the mathematical formula of the disc cutter specific energy consumption is derived. The formula takes into account key parameters such as cutting depth (d), cutting speed (v), feed rate (f), cutting angle (θ), tool radius (r) and friction coefficient (μ); Data generation and feature analysis steps: The finite element simulation method is used to generate the original data set containing the above key parameters, and the generated data set is subjected to feature analysis to determine the influence of each parameter on the specific energy consumption; PINN model construction and training steps: The model is constructed based on the physical constraint neural network (Physics-Informed Neural Network, PINN), combined with the derived specific energy consumption mathematical formula and introduced physical constraints, and the physical constraints are added to the model loss function. The PINN model is trained using a method combining data-driven and physical constraints to improve the prediction accuracy of specific energy consumption; Prediction and parameter optimization steps: The trained PINN model is used to predict the specific energy consumption under different parameter combinations, and the optimal working parameter combination is selected based on the prediction results to achieve energy consumption optimization.

2. According to the method for predicting and optimizing the specific energy consumption of disc cutter cutting based on physical constraint neural network in claim 1, the specific energy consumption expression is: （1） in, q is the wedge surface extrusion stress, R is the tool radius, h is the equivalent thickness of the tool wedge surface and the rock, λ is the angle between the feed direction and the wedge surface, f k is the equivalent friction coefficient between the tool wedge surface and the rock, K c is the repetitive shear coefficient, h y is the tooth edge equivalent height, α is the tool wedge angle, β is the tool attitude inclination angle, d is the cutting depth, w is the cutting width, f is the feed rate, t is the cutting time, v is the cutting speed, T 0 is the shear strength of the rock.

3. The method for predicting and optimizing the energy consumption of disc cutter cutting ratio based on physical constraint neural network according to claim 1 is characterized in that The loss function optimization defines the total loss function as the weighted sum of data loss and physical constraint loss, where the physical constraint loss is achieved by penalizing the gradient of the predicted output relative to each input feature, and the Adam optimizer is used to minimize the total loss; the total loss function consists of two parts: data-driven loss and physical constraint loss, and the expression is: （2） λ is a hyperparameter that controls the weight of the physical constraint loss; the data-driven loss expression is as follows: （3） The physical constraint loss measures the error between the SEC predicted by the model and the SEC calculated by the physical formula, and the expression is: （4） The physical constraint loss measures the error between the SEC predicted by the model and the SEC calculated by the physical formula, and the expression is: （5） The physical constraint loss consists of multiple physical penalty terms, expressed as: （6） The specific penalties are: ， （7）。 4. A method for predicting and optimizing the energy consumption of a disc cutter cutting ratio based on a physical constraint neural network according to any one of claims 1 to 3, comprising: The data generation module is used to obtain the original data set including parameters such as uniaxial compressive strength, cutting depth, cutting speed, feed rate, cutting angle, tool radius and friction coefficient by designing orthogonal experiments and performing finite element simulation; the feature analysis module is used to perform feature analysis on the generated original data set; the PINN model construction and training module is used to construct and train a neural network model based on physical constraints; the prediction and optimization module is used to use the trained PINN model to predict specific energy consumption and select the optimal parameters.

Images