Mechanism and data driven process index prediction method for sugarcane crushing process

Abstract

The application discloses a kind of based on mechanism and data driven's sugarcane squeezing process process index prediction method, comprising the following steps: (1) multi-sensor field data acquisition and preprocessing, establish original data set;(2) establish sugarcane material elastic-plastic constitutive model;(3) establish sugarcane squeezing process porous medium control equation;(4) sugarcane squeezing process fluid-solid coupling model and its simulation, and establish new data set;(5) the establishment of physical guide neural network model;(6) the training details of physical guide neural network, obtain based on mechanism and data driven's sugarcane squeezing process process index prediction model.The method of the application is mainly used to predict the key process indexes such as sugarcane juice precipitation amount and sugarcane juice precipitation speed in the sugarcane squeezing production process.The cost of data acquisition in the production process is greatly reduced.High-dimensional system is characterized by small data set;The consistency of network can be greatly improved, and the generalization ability is enhanced.

Description

Technical Field

[0001] This invention relates to the field of sugarcane pressing process design and optimization technology, and in particular to a method for predicting process indicators of sugarcane pressing based on mechanism and data-driven approaches. Background Technology

[0002] Sugarcane juice extraction is the first step in sugar production, and its success directly impacts the smooth operation of the entire sugar production process and the economic benefits for the enterprise. Currently, sugarcane sugar mills in my country primarily use a combined percolation method with 4-6 sets of presses to extract juice. The sugarcane juice extraction process is a complex process involving multiple factors, constraints, objectives, strong coupling, large deformation, high nonlinearity, and uncertainty. In 2017, Li Bing et al. used the finite element method to simulate the sugarcane pressing process, obtaining the effects of compression ratio, roller linear speed, and feeding pressure on the pressure and torque of the pressing rollers. In 2018, He Xiao et al. used ABAQUS software to conduct a two-dimensional simulation of the sugarcane pressing process, obtaining the relationship between sugarcane layer thickness and juice discharge speed, pressing roller diameter, pore pressure, and torque.

[0003] However, with the development of big data and artificial intelligence technologies, there is an increasing amount of research on diagnosing and predicting production processes using operational data. Introducing artificial intelligence into the sugarcane sugar production process is a noteworthy trend. For example, in 2012, Song et al. used the PCA method to process production data and employed a generalized dynamic fuzzy neural network to predict the color and pH values ​​of sugarcane juice during carbonization and clarification. In 2017, de Souza Sartori et al. introduced an artificial neural network (ANN) model into the modeling of the sugarcane juice clarification process to predict the effects of different variables on the color value and sucrose content of sugarcane juice. In 2019, Meng et al. proposed a dual support vector machine that uses seven easily measurable variables as input to predict mother liquor supersaturation and mother liquor purity.

[0004] Current research methods mainly focus on two aspects: firstly, the study and exploration of the pressing mechanism; and secondly, using data-driven methods to model and predict the pressing production process. However, both methods have their own drawbacks. The pressing mechanism is the foundation for explaining various problems that arise during the pressing process, but pressing and juice extraction is a complex industrial production process with frequently changing production conditions, such as changes in external sugarcane supply, delivery dates, yield limits for pressing volume, differences in sugarcane variety attributes, as well as fluctuations in processing time of internal processes, changes in equipment status, and logistical bottlenecks. The optimization of its production process is extremely complex, making it difficult to establish a mathematical model for optimizing and controlling workshop production operations through mechanism analysis. While data-driven modeling, as an emerging artificial intelligence technology in recent years, has many advantages, it still has significant drawbacks. Firstly, it has high requirements for the quality and quantity of data; collecting and labeling data is a very resource-intensive task. Secondly, there is the black-box output problem, where the basic relationship between input and output cannot be interpreted and understood, resulting in weak model generalization ability. Finally, there is the enormous computational cost, especially for deep learning models, which often involve tens of thousands of parameters, requiring substantial computational resources for updates and iterations.

[0005] The sugarcane pressing and juice extraction process is a complex process characterized by multiple factors, constraints, objectives, strong coupling, high nonlinearity, and uncertainty. Currently, the design of domestic pressing production systems mainly relies on empirical calculation and analysis methods, with production process control focusing primarily on the local control of a few key process points, resulting in a relatively extensive production flow. Therefore, when the pressing system is fully operational, changes in raw material composition, pressing volume, and other operating conditions lead to significant dynamic "weak link" effects in production balance and stability, causing changes in the operating parameters and states of each process. This results in numerous problems and contradictions in production efficiency, effectiveness, and energy consumption.

[0006] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a mechanism- and data-driven method for predicting process indicators in sugarcane crushing, thereby overcoming the shortcomings of weak generalization ability and high computational resource consumption in sugarcane crushing production system models.

[0008] To achieve the above objectives, this invention provides a mechanism- and data-driven method for predicting process parameters in sugarcane crushing, comprising the following steps:

[0009] (1) Collect big data information from the system from the field equipment and sensors, and combine it with the workshop big data resources that have undergone preprocessing operations including cleaning, noise reduction, integration, and conversion to create raw sample data and establish the original dataset.

[0010] (2) Establish an elastic-plastic constitutive model of sugarcane;

[0011] (3) Establish the porous media control equations for the sugarcane pressing process;

[0012] (4) Based on the elastoplastic constitutive model of sugarcane in step (2) and the porous media control equation of sugarcane pressing process in step (3), establish a fluid-structure interaction model of sugarcane pressing process, simulate the fluid-structure interaction model of sugarcane pressing process, sample the simulation results and combine them with the original data collected in step (1) to establish a new dataset;

[0013] (5) Physical guidance for the establishment of neural network models;

[0014] (6) The dataset from step (4) is sent to the physical guided neural network model in step (5) to train the physical guided neural network and obtain a mechanism- and data-driven prediction model for sugarcane pressing process indicators.

[0015] Preferably, in the above technical solution, the data collected in step (1) includes parameters such as compression ratio, roller speed, roller surface speed, and sugarcane layer thickness, spatial and temporal coordinates of the sugarcane during the pressing process, as well as the amount of sugarcane juice extracted, the sugarcane juice extraction rate, and the true values ​​of the sugarcane material stress prediction index.

[0016] Preferably, in the above technical solution, step (1) establishes the original dataset as follows: assuming the data samples are X = {x1, ... x2}. i ,…x n}, where x i ={f i1 ,…f ij ,…f im ,y i1 ,…y iω ,…y is}, f ij The j-th feature of the i-th sample, y iω Let be the w-th predicted target for the i-th sample, where n, m, and s are the number of samples, input features, and predicted targets, respectively.

[0017] Preferably, in the above technical solution, the elastic characteristics of the sugarcane in step (2) are represented by an isotropic linear elastic body with Young's modulus, bulk modulus, and Poisson's ratio as the shape of the sugarcane changes, as measured by experiments; the plastic characteristics of the sugarcane are described by a modified Cambridge model or a modified DPC model, and an elastic-plastic constitutive model of the sugarcane is established.

[0018] Preferably, in the above technical solution, the yield surface of the modified DPC model mainly consists of two parts, namely the shear yield surface F. S and cap face F C The two are connected by a gradually smoothing curve F. t To connect. The plastic potential surfaces are G. S and G C On the cap surface F C The two above coincide at the shear yield surface F. S and the gradient smooth curve F t The tops do not overlap; among them

[0019] Shear yield surface

[0020] F S =t-ptanβ-d=0

[0021] Cap yield surface

[0022]

[0023] Transition yield surface

[0024]

[0025] Plastic potential surface function on the cap surface

[0026]

[0027] Plastic potential surface on shear yield surface and transition yield surface

[0028]

[0029] In the formula, p is the equivalent pressure, q is the Mises equivalent stress, r is the third invariant of deviatoric stress, S is the stress tensor, I is the identity matrix, t is the deviatoric stress, β is the friction angle, d is the cohesion, R is the cap eccentricity, used to control the geometry of the cap surface, and α is a small value used to control the shape of the transition surface. a It is the p-value corresponding to the intersection of the cap surface and the transition surface. Where t is given by the following formula:

[0030]

[0031] k is the ratio of triaxial tensile strength to triaxial compressive strength, which mainly reflects the relationship between principal stress and yield surface.

[0032] Preferably, in the above technical solution, the method for establishing the porous media control equation for the sugarcane pressing process in step (3) is as follows: the sugarcane material is a porous media material, and the relevant theories and methods of porous media are applied to the sugarcane pressing process for analysis;

[0033] Overall equilibrium equations for porous media materials

[0034]

[0035] According to the generalized Darcy's law, the liquid equilibrium equation for porous media materials

[0036]

[0037] According to the law of conservation of mass, the continuity equation for porous media materials...

[0038]

[0039] The following assumptions also need to be made:

[0040] a) Darcy's Law is valid;

[0041] b) The Terzaghi effective stress principle is effective;

[0042] c) The medium is isotropic;

[0043] d) The medium is completely saturated;

[0044] e) The solid and liquid phases are incompressible;

[0045] f) The influence of fluids is ignored in the quasi-static test of sugarcane fiber;

[0046] g) Ignore the effect of gravity;

[0047] Under the above assumptions, the three major equations for porous media materials can be rewritten as follows:

[0048] Overall equilibrium equation

[0049] L T σ=0

[0050] Liquid equilibrium equation

[0051]

[0052] Continuity equation

[0053]

[0054] L is the differential operator, given by the following formula; T is the transpose operator; σ is the stress tensor; v is the pore fluid velocity; It represents the fluid pressure gradient; K is the permeability coefficient matrix; It is the strain rate; m T It is the transpose of the trace operator, given by the following formula.

[0055]

[0056] m T =(1 1 1 0 0 0)

[0057] The system's governing equations can be solved using the finite element method. The equations are discretized by performing operations such as mesh generation and time step setting, and then the governing equations are solved.

[0058] Preferably, in the above technical solution, the simulation method of step (4) is as follows: the sugarcane is pressed by a three-roller structure device consisting of a front roller, a rear roller and a top roller, and then the seepage stress is coupled according to the elastoplastic constitutive model of the sugarcane in step (2) and the porous media control equation of the sugarcane pressing process in step (3), and the sugarcane pressing process is simulated in 2D using ABAQUS finite element simulation software;

[0059] The model employs a seepage stress coupling element and is solved using the Lagrange formula. The analysis method is transient analysis of soil consolidation. Initial conditions include initial pore pressure and initial void ratio. Boundary conditions include zero surface pore pressure and no-slip solid boundary conditions at the press roll-cane contact surface. The physical properties of the cane are obtained from data measured by the KANNAPIRAN experiment.

[0060] Preferably, in the above technical solution, the method for establishing a new dataset in step (4) is as follows: After the simulation is completed, the Mises stress cloud map and fluid velocity vector map are obtained. The results are subjected to Latin hypercube sampling and combined with the original data collected in step (1) to form a new dataset sample D = {d1…d2}. i …d n},in The ωth output value obtained after the simulation of the i-th sample will be used to train the physical guided neural network in step (5).

[0061] Preferably, in the above technical solution, the method for establishing the physical-guided neural network model in step (5) is as follows: the physical-guided neural network is divided into three layers: an input layer, a hidden layer, and an output layer; the input data of the physical-guided neural network is linearly processed layer by layer in the hidden layer, and transformed into a nonlinear output through an activation function, until the final prediction of the output layer is expressed as follows:

[0062] a l =σ(ω) l a l-1 +b l )

[0063] Among them, a l Let a be the output value of the l-th layer of the neural network. l-1 b is the output of the (l-1)th layer of the neural network. lLet ω be the bias from layer (l-1) to layer l in the neural network, ω be the weights from layer (l-1) to layer l in the neural network, and σ be the activation function;

[0064] The loss function uses the minimum mean square error (MSE) and then performs backpropagation iteratively to solve for b; its loss function expression is:

[0065] Loss = MSE DATA +λ1MSE PDE +λ2MSE BC +λ3MSE IC

[0066]

[0067] MSE DATA The minimum mean square error (MSE) for data matching PDE To minimize the mean square error of the governing equations, MSE BC The minimum mean square error of the boundary conditions, MSE IC Let be the minimum mean square error of the initial conditions, m be the number of samples, λ1, λ2, λ3 be the corresponding weights, and pi be the predicted value. The actual value;

[0068] The initial conditions include initial pore pressure and initial void ratio, specified by data obtained from KANNAPIRAN experiments; the boundary conditions include zero surface pore pressure and no slippage at the press roll-sugar cane contact surface; the governing equations include the overall equilibrium equation, the liquid equilibrium equation, and the continuity equation, as follows:

[0069] Boundary conditions:

[0070] P0 = 0

[0071]

[0072]

[0073] Governing equations:

[0074] L T σ=0

[0075]

[0076]

[0077] Where P0 is the surface pore pressure. Let S be the sugarcane velocity, α be the roller surface tangential velocity, and α be the contact angle between the sugarcane and the roller surface. L is the differential operator, given by the following formula; T is the transpose operator; σ is the stress tensor; and v is the pore fluid velocity. It represents the fluid pressure gradient; K is the permeability coefficient matrix; It is the strain rate; m T The trace operator transpose is given by the following formula:

[0078]

[0079] m T =(1 1 1 0 0 0).

[0080] Preferably, in the above technical solution, the method for physically guiding the training of the neural network in step (6) is as follows:

[0081] Based on the actual sugarcane pressing process, the control equations are selected from the overall equilibrium equation, liquid equilibrium equation and continuity equation of porous media material. The boundary conditions are set to zero surface pore pressure and no slippage on the contact surface between the pressing roller and the sugarcane. The activation function is selected as tanh activation function, and the optimizer is selected as Adam optimizer. After training for 10,000 steps, L-BFGS is used for training until convergence, thus completing the detailed settings of the physical guided neural network.

[0082] The dataset obtained in step (4) is divided into a training set D. train and verification set D test The data is further divided into an input set and an output set, where the input set is... The output set is Y i ={y i1 ,…y iω ,…y is}. The training set D train The data is fed into a physical-guided neural network for training, ultimately resulting in a predictive model for sugarcane pressing process parameters based on mechanism and data-driven principles.

[0083] Compared with existing technologies, this invention has the following advantages: The mechanism- and data-driven method for predicting process indicators in sugarcane pressing is mainly used to predict key process indicators such as juice extraction rate and volume during sugarcane pressing. This model combines finite element simulation and data-driven approaches. Finite element simulation clarifies the relevant mechanisms of the sugarcane pressing process while simultaneously acquiring data, significantly reducing the cost of data collection during production. Simultaneously, the data-driven side utilizes a physics-guided neural network model, incorporating the governing equations of the sugarcane pressing process, such as Darcy's law and continuity equations, into the loss function. This greatly reduces the amount of data required while limiting the range of output target values, enabling the representation of a high-dimensional system with a small dataset. Furthermore, the inclusion of physical laws during network training significantly improves network consistency and enhances generalization ability. Attached Figure Description

[0084] Figure 1This is a flowchart of a mechanism- and data-driven method for predicting process indicators in sugarcane pressing according to the present invention.

[0085] Figure 2 This is a flowchart of the equipment at the sugarcane pressing site according to the method of the present invention;

[0086] Figure 3 This is a flowchart of the elastoplastic constitutive model of sugarcane in the method according to the present invention;

[0087] Figure 4 This is a schematic diagram of the modified DPC model according to the method of the present invention;

[0088] Figure 5 This is a simulation diagram of the fluid-structure interaction in sugarcane pressing according to the method of the present invention;

[0089] Figure 5 In the sugarcane pressing process, area A is the region where the sugarcane is acted upon by the front roller and the top roller, while area B is the region where the sugarcane is acted upon by the rear roller and the top roller.

[0090] Figure 6 This is a flowchart of the physically guided neural network in the method according to the present invention. Detailed Implementation

[0091] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0092] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0093] like Figures 1 to 6 As shown, a method for predicting process indicators in sugarcane crushing based on mechanism and data-driven approaches according to a specific embodiment of the present invention includes the following steps:

[0094] (1) Data acquisition and preprocessing by multiple sensors

[0095] First, based on the required data, big data information from the system's on-site equipment and sensors is collected. This data is then combined with pre-processed workshop big data resources, including cleaning, noise reduction, integration, and conversion, to create raw sample data. The collected data includes process parameters such as compression ratio, roller speed, roller surface speed, and cane layer thickness; spatial and temporal coordinates of the sugarcane during the pressing process; and true values ​​of predicted indicators such as juice extraction rate and cane stress. Assume the data sample is X = {x1,…x}. i ,…x n}, where xi ={f i1 ,…f ij ,…f im ,y i1 ,…y iω ,…y is}, f ij The j-th feature of the i-th sample, y iω Let be the w-th predicted target for the i-th sample, where n, m, and s are the number of samples, input features, and predicted targets, respectively. Field equipment such as... Figure 2 As shown.

[0096] (2) Establishing an elastic-plastic constitutive model of sugarcane.

[0097] The constitutive model of sugarcane must reflect the stress-strain relationship during compression. The elastic characteristics of sugarcane can be represented by experimentally measured values ​​such as Young's modulus, bulk modulus, and Poisson's ratio, which vary isotropically with the shape of the sugarcane, thus representing a linear elastic body. Experimental observations show that the critical state of sugarcane is similar to that of clay, and classical models from soil mechanics, such as the modified Cambridge model or the modified DPC model, can be used to describe the plastic characteristics of sugarcane during pressing. Since the sugarcane fibers are affected by the frictional force on the surface of the pressing rollers during pressing, thus experiencing tension, the modified DPC model, which describes tension behavior, was chosen as the plastic model for sugarcane. The process of establishing the elastic-plastic constitutive model of sugarcane is as follows: Figure 3 As shown.

[0098] The yield surface of the modified DPC model mainly consists of two components: the shear yield surface F. S and cap face F C The two are connected by a gradually smoothing curve F. t To connect. The plastic potential surfaces are G. S and G C On the cap surface F C The two above coincide at the shear yield surface F. S and the gradient smooth curve F t The above do not overlap, such as Figure 4 As shown. Among them.

[0099] Shear yield surface

[0100] F S =t-ptanβ-d=0

[0101] Cap yield surface

[0102]

[0103] Transition yield surface

[0104]

[0105] Plastic potential surface function on the cap surface

[0106]

[0107] Plastic potential surface on shear yield surface and transition yield surface

[0108]

[0109] In the formula, p is the equivalent pressure, q is the Mises equivalent stress, r is the third invariant of deviatoric stress, S is the stress tensor, I is the identity matrix, t is the deviatoric stress, β is the friction angle, d is the cohesion, R is the cap eccentricity, used to control the geometry of the cap surface, and α is a small value used to control the shape of the transition surface. a It is the p-value corresponding to the intersection of the cap surface and the transition surface. Where t is given by the following formula:

[0110]

[0111] k is the ratio of triaxial tensile strength to triaxial compressive strength, which mainly reflects the relationship between principal stress and yield surface.

[0112] (3) Establish the porous media control equation for the sugarcane pressing process.

[0113] Sugarcane is a typical porous media material, and related theories and methods of porous media can be applied to the pressing problem of sugarcane, for example...

[0114] Overall equilibrium equations for porous media materials

[0115]

[0116] According to the generalized Darcy's law, the liquid equilibrium equation for porous media materials

[0117]

[0118] According to the law of conservation of mass, the continuity equation for porous media materials...

[0119]

[0120] To apply the theory of porous media mechanics to the problem of sugarcane pressing, the following assumptions are required: a) Darcy's law is valid.

[0121] b) The Terzaghi effective stress principle is effective.

[0122] c) The medium is isotropic.

[0123] d) The medium is completely saturated.

[0124] e) The solid and liquid phases are incompressible.

[0125] f) The influence of fluids is ignored in the quasi-static test of sugarcane fiber.

[0126] g) Ignore the effect of gravity

[0127] Under the above assumptions, the three major equations for porous media materials can be rewritten as follows:

[0128] Overall equilibrium equation

[0129] L T σ=0

[0130] Liquid equilibrium equation

[0131]

[0132] Continuity equation

[0133]

[0134] L is the differential operator, given by the following formula; T is the transpose operator; σ is the stress tensor; v is the pore fluid velocity; It represents the fluid pressure gradient; K is the permeability coefficient matrix; It is the strain rate; m T It is the transpose of the trace operator, given by the following formula.

[0135]

[0136] m T =(1 1 1 0 0 0)

[0137] The system's governing equations can be solved using the finite element method. The equations are discretized by performing operations such as mesh generation and time step setting, and then the governing equations are solved.

[0138] (4) Fluid-structure interaction model and simulation of sugarcane pressing

[0139] For the three-roll pressing process commonly used in factories, 2D simulation was performed using ABAQUS finite element simulation software, such as... Figure 5 As shown. Specifically, a three-roller structure device consisting of a front roller, a rear roller, and a top roller is used to press sugarcane. Then, based on the elastoplastic constitutive model of the sugarcane material in step (2) and the porous media control equation of the sugarcane pressing process in step (3), the seepage stress coupling is performed, and the ABAQUS finite element simulation software is used to perform 2D simulation of the three-roller sugarcane pressing process.

[0140] The model employs a seepage stress coupling element and is solved using the Lagrange formula. The analysis method is transient analysis of soil consolidation. Initial conditions include initial pore pressure and initial void ratio. Boundary conditions include zero surface pore pressure and no-slip solid boundary conditions at the press roll-cane contact surface. The physical properties of the cane are obtained from data measured by the KANNAPIRAN experiment.

[0141] After the simulation, Mises stress cloud map, fluid velocity vector map, etc. are obtained. Latin hypercube sampling is performed on the results, and they are combined with the original data collected in step (1) to form a new dataset sample D = {d1…d i …d n},in The ωth output value obtained after the simulation of the i-th sample will be used to train the physical guided neural network in step (5).

[0142] (5) Modeling of physical-guided neural networks

[0143] Physically Guided Neural Networks (PINNs) are neural networks trained to solve supervised learning tasks while respecting any given physical laws described by general nonlinear partial differential equations. This involves incorporating known physical laws, such as governing equations, boundary conditions, and initial conditions, into the loss function, adding a structural risk minimization strategy to the empirical risk minimization framework.

[0144] Physically Guided Neural Networks (PINNs) consist of three layers: an input layer, a hidden layer, and an output layer, as follows: Figure 6 As shown. Hidden layers typically have three or more layers. The more hidden layers, the stronger the model's predictive ability, but also the more parameters and computational resources it consumes. The input data of PINN undergoes linear operations layer by layer in the hidden layers, and is transformed into a non-linear output through activation functions until the final prediction of the output layer is expressed mathematically as follows:

[0145] a l =σ(ω) l a l-1 +b l )

[0146] Where a l Let a be the output value of the l-th layer of the neural network. l-1 b is the output of the (l-1)th layer of the neural network. l ω represents the bias from layer (l-1) to layer l in the neural network, ω represents the weights from layer (l-1) to layer l in the neural network, and σ represents the activation function.

[0147] The loss function is typically the minimum mean squared error (MSE), followed by backpropagation iteratively to solve for b. Its expression is:

[0148] Loss = MSE DATA +λ1MSE PDE +λ2MSE BC +λ3MSE IC

[0149]

[0150] MSE DATA The minimum mean square error (MSE) for data matching PDE To minimize the mean square error of the governing equations, MSE BC The minimum mean square error of the boundary conditions, MSE IC Let be the minimum mean square error of the initial conditions, m be the number of samples, λ1, λ2, λ3 be the corresponding weights, and pi be the predicted value. This is the actual value.

[0151] The initial conditions, including initial pore pressure and initial void ratio, are specified using data obtained from the KANNAPIRAN experiments. Boundary conditions include zero surface pore pressure and no slippage at the press roll-sugar cane contact surface. The governing equations include the overall equilibrium equation, the liquid equilibrium equation, and the continuity equation, as shown below:

[0152] Boundary conditions:

[0153] P0 = 0

[0154]

[0155]

[0156] Governing equations:

[0157] L T σ=0

[0158]

[0159]

[0160] Where P0 is the surface pore pressure. Let S be the sugarcane velocity, α be the roller surface tangential velocity, and α be the contact angle between the sugarcane and the roller surface. L is the differential operator, given by the following formula; T is the transpose operator; σ is the stress tensor; and v is the pore fluid velocity. It represents the fluid pressure gradient; K is the permeability coefficient matrix; It is the strain rate; m T It is the transpose of the trace operator, given by the following formula.

[0161]

[0162] m T =(1 1 1 0 0 0)

[0163] (6) Training details of physically guided neural networks

[0164] The dataset from step (4) is fed into the physical-guided neural network model in step (5) for training, resulting in a mechanism- and data-driven prediction model for sugarcane pressing process parameters. For example... Figure 6 As shown.

[0165] Specifically, based on the actual sugarcane pressing process, the governing equations are the overall equilibrium equation, liquid equilibrium equation, and continuity equation for porous media materials. Boundary conditions are set to zero surface pore pressure and no slippage at the contact surface between the pressing roller and the sugarcane. The activation function is tanh, and the optimizer is Adam, trained for 10,000 steps, followed by L-BFGS training until convergence, completing the detailed settings of the PINN neural network.

[0166] The dataset obtained in step four is divided into a training set D. train and verification set D test The data is further divided into an input set and an output set, where the input set is... The output set is Y i ={y i1 ,…y iω ,…y is}. The training set D train The data is fed into the PINN neural network for training, ultimately yielding a predictive model for sugarcane pressing process parameters based on mechanism and data-driven principles.

[0167] The mechanism- and data-driven hybrid model proposed in this invention can effectively solve the problem of online prediction of process indicators in sugarcane crushing production, providing a strong basis for production process adjustment and decision-making. Based on the mechanism and data-driven hybrid model, in terms of mechanism analysis, finite element simulation can clarify the physical laws such as stress changes and juice velocity changes during sugarcane crushing, while also acquiring some data, reducing data acquisition costs. In terms of data-driven aspects, the recently emerging Physics-Guided Neural Network (PINN) is used for modeling, incorporating the physical equations and prior knowledge of the sugarcane crushing process as a loss function into the neural network. This provides a physical explanation for the model while further reducing the amount of data required for deep learning, lowering computational costs, and improving the model's generalization ability. This provides a new direction for the artificial intelligence of the sugarcane crushing process.

[0168] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for predicting process indicators in sugarcane pressing based on mechanism and data-driven approaches, characterized in that, Includes the following steps: (1) Collect big data information from the field equipment and sensors, combine it with the workshop big data resources that have undergone cleaning, noise reduction, integration and conversion preprocessing operations, and make them into raw sample data to establish the raw dataset; (2) Establish an elastic-plastic constitutive model of sugarcane; (3) Establish the porous media control equations for the sugarcane pressing process; (4) Based on the elastoplastic constitutive model of sugarcane in step (2) and the porous media control equation of sugarcane pressing process in step (3), establish a fluid-structure interaction model of sugarcane pressing process, simulate the fluid-structure interaction model of sugarcane pressing process, sample the simulation results and combine them with the original data collected in step (1) to establish a new dataset; (5) Establishment of a physical-guided neural network model; (6) Send the dataset from step (4) to the physical guided neural network model in step (5) to train the physical guided neural network and obtain a predictive model of sugarcane pressing process indicators based on mechanism and data-driven. Step (2) The elastic characteristics of sugarcane are represented by an isotropic linear elastic body with Young's modulus, bulk modulus and Poisson's ratio as the shape of sugarcane changes, as measured by experiments; the plastic characteristics of sugarcane are described by the modified Cambridge model or the modified DPC model, and an elastic-plastic constitutive model of sugarcane is established. Step (5) The method for establishing the physical-guided neural network model is as follows: The physical-guided neural network is divided into three layers: input layer, hidden layer, and output layer; the input data of the physical-guided neural network is linearly processed layer by layer in the hidden layer, and transformed into a nonlinear output through the activation function, until the final prediction of the output layer is expressed as follows: ； in, For the neural network The output value of the layer, For the neural network Layer output, For the neural network layer to the first Layer bias, For the neural network layer to the first Layer weights, For activation functions; The loss function uses the minimum mean square error (MSE) and then backpropagation is performed iteratively to solve for b; its loss function expression is: ； ； in The minimum mean square error for data matching. This represents the minimum mean square error of the governing equations. The minimum mean square error of the boundary conditions. Let m be the minimum mean square error of the initial conditions, and m be the sample size. These are the corresponding weights, and the predicted value of pi. The actual value; The initial conditions include initial pore pressure and initial void ratio, specified by data measured by KANNAPIRAN experiments; the boundary conditions include zero surface pore pressure and no slippage at the press roll-sugar cane contact surface; the governing equations include the overall equilibrium equation, the liquid equilibrium equation, and the continuity equation. Step (6) The method for training the physical-guided neural network is as follows: Based on the actual sugarcane pressing process, the control equations are selected as the overall equilibrium equation, liquid equilibrium equation and continuity equation of porous media material. The boundary conditions are set as the surface pore pressure is 0 and there is no slippage on the contact surface between the pressing roller and the sugarcane. The activation function is selected as the tanh activation function, and the optimizer is selected as the Adam optimizer. After training for 10,000 steps, L-BFGS is used for training until convergence, thus completing the setup of the physical-guided neural network.

2. The method for predicting process indicators of sugarcane crushing based on mechanism and data-driven methods according to claim 1, characterized in that, The data collected in step (1) includes parameters such as compression ratio, roller speed, roller surface speed, and sugarcane layer thickness, spatial and temporal coordinates of the sugarcane during the pressing process, as well as the amount of sugarcane juice extracted, the sugarcane juice extraction rate, and the true values ​​of the predicted stress index of the sugarcane material.

3. The method for predicting process indicators of sugarcane pressing based on mechanism and data-driven methods according to claim 2, characterized in that, Step (1) Establish the original dataset as follows, assuming the data sample is... ,in , The j-th feature of the i-th sample Let be the w-th predicted target for the i-th sample, where n, m, and s are the number of samples, input features, and predicted targets, respectively.

4. The method for predicting process indicators of sugarcane crushing based on mechanism and data-driven methods according to claim 1, characterized in that, The yield surface of the modified DPC model consists of two components: the shear yield surface and the shear yield surface. Hat Cover The two are separated by a gradient smooth curve. To connect; the plastic potential surfaces are respectively and On the cap surface The two overlap at the shear yield surface. and gradient smooth curve The two do not overlap; in Shear yield surface ； Cap yield surface ； Transition yield surface ； Plastic potential surface function on the cap surface ； Plastic potential surface on shear yield surface and transition yield surface ； In the formula, p is the equivalent pressure, q is the Mises equivalent stress, r is the third invariant of deviatoric stress, S is the stress tensor, I is the identity matrix, and t is the deviatoric stress. denoted by φ, where d is the friction angle, d is the cohesive force, and R is the cap eccentricity, used to control the geometry of the cap surface. It is a tiny value used to control the shape of the transition surface. It is the p-value corresponding to the intersection of the cap surface and the transition surface; where t is given by the following formula: ； k is the ratio of triaxial tensile strength to triaxial compressive strength, reflecting the relationship between principal stress and yield surface.

5. The method for predicting process indicators of sugarcane crushing based on mechanism and data-driven methods according to claim 1, characterized in that, Step (3) The method for establishing the porous media control equation for the sugarcane pressing process is as follows: Sugarcane is a porous media material, and the relevant theories and methods of porous media are applied to the sugarcane pressing process for analysis; Overall equilibrium equations for porous media materials ； According to the generalized Darcy's law, the liquid equilibrium equation for porous media materials ； According to the law of conservation of mass, the continuity equation for porous media materials... ； The following assumptions also need to be made: a) Darcy's Law is valid; b) The Terzaghi effective stress principle is effective; c) The medium is isotropic; d) The medium is completely saturated; e) The solid and liquid phases are incompressible; f) The influence of fluids is ignored in the quasi-static test of sugarcane fiber; g) Ignore the effect of gravity; Under the above assumptions, the three major equations for porous media materials can be rewritten as follows: Overall equilibrium equation ； Liquid equilibrium equation ； Continuity equation ； L is the differential operator, given by the following formula; T is the transpose operator; It is the stress tensor; v is the pore fluid velocity; It represents the fluid pressure gradient; K is the permeability coefficient matrix; It is the strain rate; The trace operator transpose is given by the following formula: ； ； The system's governing equations can be solved using the finite element method. The equations are discretized by mesh generation and time step setting, and then the governing equations are solved.

6. The method for predicting process indicators of sugarcane crushing based on mechanism and data-driven methods according to claim 1, characterized in that, The simulation method for step (4) is as follows: use a three-roller structure device consisting of a front roller, a rear roller, and a top roller to press sugarcane, and then perform seepage stress coupling based on the elastoplastic constitutive model of sugarcane material in step (2) and the porous media control equation of the sugarcane pressing process in step (3). Use ABAQUS finite element simulation software to perform 2D simulation of the three-roller sugarcane pressing process. The model employs a seepage stress coupling element and is solved using the Lagrange formula. The analysis method is transient analysis of soil consolidation. Initial conditions include initial pore pressure and initial void ratio. Boundary conditions include zero surface pore pressure and no-slip solid boundary conditions at the press roll-sugar cane contact surface. The physical properties of the sugarcane are obtained from KANNAPIRAN experiments.

7. The method for predicting process indicators of sugarcane crushing based on mechanism and data-driven methods according to claim 1, characterized in that, Step (4) to establish a new dataset is as follows: After the simulation is completed, the Mises stress cloud map and fluid velocity vector map are obtained. The results are subjected to Latin hypercube sampling and combined with the original data collected in step (1) to form a new dataset sample. ,in , The i-th sample is obtained after simulation. The output values, forming new data samples, will be used in step (5) for training the physical-guided neural network.

8. The method for predicting process indicators of sugarcane crushing based on mechanism and data-driven methods according to claim 1, characterized in that, Boundary conditions include zero surface pore pressure and no slippage at the press roll-sugar cane contact surface; governing equations include the overall equilibrium equation, the liquid equilibrium equation, and the continuity equation, as follows: Boundary conditions: ； ； ； Governing equations: ； ； ； Where P0 is the surface pore pressure. S is the sugarcane material speed, and S is the roller cutting speed. The contact angle between the sugarcane and the roller surface; L is the differential operator, given by the following formula; T is the transpose operator; It is the stress tensor; v is the pore fluid velocity; It represents the fluid pressure gradient; K is the permeability coefficient matrix; It is the strain rate; The trace operator transpose is given by the following formula: ； 。 9. The method for predicting process indicators of sugarcane crushing based on mechanism and data-driven methods according to claim 1, characterized in that, The dataset obtained in step (4) is divided into training sets. and verification set The data is further divided into an input set and an output set, where the input set is... The output set is ; training set The data is fed into a physical-guided neural network for training, ultimately resulting in a predictive model for sugarcane pressing process parameters based on mechanism and data-driven principles.

Images