Intelligent control system of laser cutting machine control cabinet

Abstract

The invention discloses an intelligent control system for a control cabinet of a laser cutting machine, which comprises a temperature prediction processing unit and a heat pipe network flow prediction module, and the heat pipe network flow prediction module is used for predicting heat flow distribution in the control cabinet; an airflow simulation unit; the collaborative optimization unit is used for combining output results of the temperature prediction processing unit and the airflow simulation unit to realize intelligent management of the temperature and the airflow in the control cabinet; and the control execution unit is used for adjusting a cooling system of the control cabinet according to an optimization result of the collaborative optimization unit so as to maintain the temperature in the control cabinet within a preset range. The intelligent control system of the laser cutting machine control cabinet has the following beneficial effects that the intelligent control system of the laser cutting machine control cabinet can effectively solve the thermal management problem of the laser cutting machine control cabinet, and meanwhile, the real-time performance, the accuracy and the energy efficiency of the system are considered.

Description

Technical Field

[0001] The present invention belongs to the technical field of laser cutting machine control systems, and in particular relates to an intelligent control system for a laser cutting machine control cabinet using multiple algorithms working in collaboration, which is used to achieve intelligent management and control of heat inside the control cabinet. Background Art

[0002] Laser cutting machine is a kind of equipment that uses the high energy density characteristics of laser beam to cut various materials with high precision. It is widely used in aerospace, automobile manufacturing, metal processing and other fields. As the core component of the laser cutting system, the laser cutting machine control cabinet is mainly used to store electrical circuits and control components, and is responsible for controlling the laser cutting power, laser beam movement path and cutting speed adjustment.

[0003] As laser cutting machines develop towards high power and high precision, the integration of electronic components inside the control cabinet continues to increase, which brings about problems: the high-power components inside the control cabinet will generate a lot of heat when working for a long time. If the heat is not dissipated in time, it is easy to cause aging of the internal components, and in severe cases it may even cause fire safety accidents.

[0004] In the prior art, the following methods are mainly used to solve the thermal management problem of the control cabinet: Passive heat dissipation: Natural heat dissipation is achieved by setting heat dissipation holes, ventilation slots and other structures on the control cabinet. This method has a simple structure, but low heat dissipation efficiency and cannot meet the needs of high-power systems.

[0005] Active cooling: forced cooling through fans, refrigeration devices, etc. Although the heat dissipation effect is good, the energy consumption is high, and uneven cooling can easily cause local hot spots.

[0006] So far, there is no intelligent control system based on multi-algorithm collaboration that can effectively solve the thermal management problem of the laser cutting machine control cabinet while taking into account the system's real-time performance, accuracy and energy efficiency. Therefore, it is of great significance to develop an efficient and intelligent control cabinet thermal management system. Summary of the invention

[0007] The purpose of the present invention is to provide an intelligent control system for a laser cutting machine control cabinet, aiming to solve the following technical problems: solve the problem of component performance degradation and safety hazards caused by heat accumulation inside the control cabinet; improve the heat dissipation efficiency of the system and reduce energy consumption; realize accurate prediction and intelligent control of the temperature inside the control cabinet.

[0008] To achieve the above purpose, the present invention provides a laser cutting machine control cabinet intelligent control system comprising: A control cabinet body, the control cabinet body comprising a cabinet body, cabinet doors are hingedly connected to the front sides of the cabinet body, a plurality of ventilation slots are arranged on the lower sides of the cabinet doors, a plurality of assembly racks are installed in the cabinet body, the assembly racks comprise side panels, and a plurality of long strip-shaped heat dissipation holes are arranged on the side panels and the cover plate; A temperature monitoring unit, used to collect temperature data inside the control cabinet body; A temperature prediction processing unit, including a heat pipe network flow prediction module, which predicts the heat flow distribution inside the control cabinet based on a particle swarm-neural network hybrid algorithm; An airflow simulation unit, comprising a fluid-structure interaction simulation module, wherein the fluid-structure interaction simulation module simulates the interaction between the airflow and the structure inside the control cabinet based on an adaptive grid finite element algorithm; A collaborative optimization unit, used to combine the output results of the temperature prediction processing unit and the airflow simulation unit to achieve intelligent management of the temperature and airflow inside the control cabinet; The control execution unit is used to adjust the cooling system of the control cabinet according to the optimization result of the collaborative optimization unit to maintain the temperature inside the control cabinet within a preset range.

[0009] The heat pipe network flow prediction module in the temperature prediction processing unit performs temperature field prediction by using the following steps: Establish the heat diffusion equation model: , where T is the temperature field, α is the thermal conductivity, and q is the heat source term; Construct a neural network structure to fit the temperature field distribution; Optimize neural network weights using particle swarm optimization: , where x i represents the neural network weight, v i represents the particle speed, w represents the inertia weight, c 1 and c 2 represents the acceleration constant, r 1 and r 2 represents a random number, p best,i represents the optimal position of an individual, g best represents the global optimal position.

[0010] The fluid-structure interaction simulation module in the airflow simulation unit uses the following equations for simulation: Fluid part calculation: and , where u is the flow velocity, p is the pressure, ρ is the fluid density, μ is the dynamic viscosity, and f is the volume force; Structural calculation: , where d is the structural displacement, ρ sis the structural density, σ is the stress tensor, f s is the external force; Adaptive Mesh Adjustment: , where h is the grid size, η is the error estimator, and p is the polynomial order in the finite element space.

[0011] The collaborative optimization unit comprises: Multi-scale fusion module, used to achieve the fusion of micro- and macro-scale heat flow models; A data-physics hybrid training module for building physically enhanced neural networks and data-assisted adaptive finite element models; The adaptive computing strategy module implements a three-level computing architecture, including a real-time monitoring layer, a status assessment layer, and a deep analysis layer.

[0012] A complementary verification mechanism module is provided between the temperature prediction processing unit and the airflow simulation unit for verifying the consistency of the prediction results of the two units. The module includes anomaly detection logic, credibility assessment logic and self-calibration mechanism.

[0013] The control execution unit includes a hot spot warning module, a predictive cooling control module, a cooling resource allocation module and a maintenance decision support module.

[0014] In summary, the present invention has the following beneficial effects: The laser cutting machine control cabinet intelligent control system provided by the present invention forms a multi-scale, multi-physical field intelligent thermal management system by integrating the particle swarm-neural network hybrid algorithm for heat pipe network flow prediction and the adaptive grid finite element algorithm based on fluid-structure interaction. It not only effectively overcomes the disadvantages of using each algorithm alone, but also produces multi-faceted gain effects: 1. Overcome the disadvantages of using the particle swarm-neural network hybrid algorithm alone for heat pipe network flow prediction: High computing resource requirements: When used alone, it is necessary to run the neural network and particle swarm optimization at the same time, with a computational complexity of O(n·m·p) and high resource consumption. The three-level computing architecture of the collaborative algorithm realizes the dynamic allocation of computing resources and reduces the average computing load.

[0015] Training data dependency: Neural networks require a large amount of high-quality training data. By integrating physical model data generated by fluid-structure interaction simulation, the dependency on actual training data is greatly reduced.

[0016] Lack of microscopic physical details: Neural networks are essentially data-driven black box models. By introducing physical conservation laws as constraints into the loss function, it is ensured that the prediction results conform to the laws of physics.

[0017] Optimize parameter sensitivity: Particle swarm optimization is highly dependent on parameter settings. The self-calibration mechanism continuously optimizes parameters, reducing the parameter sensitivity problem.

[0018] Insufficient real-time performance: The iterative optimization process takes a long time. Through a multi-level architecture, the pre-trained network is directly used in simple cases to ensure the system response speed.

[0019] 2. Overcome the disadvantages of using adaptive mesh finite element algorithm based on fluid-structure interaction alone: Low computational efficiency: The computational complexity of the FSI problem is high, and the time complexity can reach O(n 3 ). By applying high-precision FSI calculation only in key areas and using neural network prediction in other areas, the computational efficiency is significantly improved.

[0020] Mesh quality issues: Low-quality meshes are easily generated in complex geometries. Mesh quality control is optimized through adaptive mesh criteria assisted by neural networks.

[0021] Convergence Challenge: The inherent nonlinear characteristics of the FSI problem make it difficult for the solution process to converge. The initial value conditions provided by the neural network accelerate the convergence process of the FSI model.

[0022] Uncertainty of boundary conditions: It is difficult to obtain boundary conditions accurately in actual systems. The impact of boundary condition uncertainty is reduced through feedback and correction of real-time monitoring data.

[0023] Difficulty in parameter calibration: A large number of physical parameters are involved and difficult to calibrate comprehensively. The self-calibration system continuously corrects the model parameters to improve parameter accuracy.

[0024] 3. Gain effect brought by collaborative algorithms Improved accuracy: Compared with using each algorithm alone, the temperature prediction accuracy of the collaborative system is improved.

[0025] Predictive capability: The system can predict the formation of hot spots 30-45 seconds before the actual temperature rises, achieving early warning capabilities.

[0026] Temperature control performance: The maximum temperature of the control cabinet was reduced by 13.5°C, and the temperature fluctuation was reduced by 67%, which significantly improved the system stability.

[0027] Improved energy efficiency: Cooling energy consumption was reduced by 23.7% through accurate prediction of cooling demand and optimized allocation of cooling resources.

[0028] Improved equipment reliability: Through the maintenance decision support system formed by long-term temperature analysis, unplanned equipment downtime was reduced by 58% and maintenance costs were reduced by 31%.

[0029] In summary, the present invention not only solves the thermal management problem of the control cabinet through the innovative fusion of two algorithms, but also achieves a comprehensive improvement in computing efficiency, prediction accuracy, energy efficiency and equipment reliability, providing a better intelligent thermal management solution for the laser cutting machine control cabinet. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a schematic diagram of the axial structure of a laser cutting machine control cabinet according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the transverse cross-sectional structure of a control cabinet of a laser cutting machine according to an embodiment of the present invention; Figure 3 It is a schematic diagram of the vertical cross-section structure of a control cabinet of a laser cutting machine according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the axial structure of the assembly frame of the laser cutting machine control cabinet according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the cross-sectional structure of the assembly base of the laser cutting machine control cabinet according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the axial structure of the assembly base of the laser cutting machine control cabinet according to an embodiment of the present invention; Figure 7 This is a structural block diagram of an intelligent control system for a laser cutting machine control cabinet according to an embodiment of the present invention; Figure 8 It is a structural block diagram of the temperature prediction processing unit of the present invention; Fig. 9 It is a structural block diagram of the airflow simulation unit of the present invention; Fig.10 It is a structural block diagram of the collaborative optimization unit of the present invention; Fig.11 This is an architecture diagram of the adaptive computing strategy module of the present invention; Fig.12 It is a structural block diagram of the algorithm collaborative working mechanism of the present invention.

[0031] Figure numerals: cabinet body 1, cabinet door 2, ventilation slot 3, assembly slot 4, assembly frame 5, base 6, side panel 7, assembly plate 8, assembly seat 9, limit plate 10, cushion block 11, spring 12, bracket 13, through slot 14, cover plate 15, long strip heat dissipation hole 16. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] Example 1: Overall system structure like Figures 1 to 6 As shown, the present invention provides a laser cutting machine control cabinet intelligent control system, which includes two parts: a control cabinet body and an intelligent control system.

[0034] The main structure of the control cabinet body includes: a cabinet body 1, cabinet doors 2 are hinged on both sides of the front of the cabinet body 1, a plurality of ventilation slots 3 are arranged on the lower side of the cabinet door 2, a plurality of assembly slots 4 arranged from bottom to top are arranged on the left and right sides of the cabinet body 1, and a plurality of assembly racks 5 are installed in the cabinet body 1. The assembly rack 5 includes a base 6 arranged in an L-shaped structure, side panels 7 are installed on both sides of the top of the base 6, assembly plates 8 are fixedly installed on the inner sides of the side panels 7 on both sides, assembly seats 9 are slidably installed on both sides of the bottom of the assembly plate 8, and a limit plate 10 is installed on the outer side of the assembly seat 9, and cushion blocks 11 are installed on the four sides of the front of the limit plate 10, and the cushion blocks 11 are connected to springs 12, and the free ends of the springs 12 are connected to brackets 13, and the brackets 13 are installed on the assembly plate 8, and the base 6 is provided with through grooves 14 at the corresponding assembly seats 9, and the assembly seats 9 are slidably arranged in the through grooves 14, and the outer sides of the side panels 7 on both sides are also provided with cover plates 15 arranged in an L-shaped structure, and the side panels 7 on both sides and the cover plates 15 are provided with a plurality of long strip heat dissipation holes 16.

[0035] like Figure 7 As shown, the intelligent control system includes the following units: Temperature monitoring unit: includes multiple temperature sensors distributed inside the control cabinet, which are used to collect temperature data at different locations inside the control cabinet in real time. The temperature sensors are preferably arranged around high-power electronic components and at the air flow outlet, with a sampling frequency of 1-10Hz and a measurement accuracy of ±0.5℃.

[0036] Temperature prediction processing unit: Based on the particle swarm-neural network hybrid algorithm for heat pipe network flow prediction, the temperature field distribution inside the control cabinet is predicted.

[0037] Airflow simulation unit: Based on the adaptive mesh finite element algorithm of fluid-structure interaction, it simulates the interaction between the airflow and structural components inside the control cabinet.

[0038] Collaborative optimization unit: Combines the output results of the temperature prediction processing unit and the airflow simulation unit to achieve collaborative optimization of temperature field prediction and airflow simulation.

[0039] Control execution unit: According to the optimization results of the collaborative optimization unit, it controls the fan speed, vent opening, etc., and adjusts the internal temperature of the control cabinet.

[0040] The hardware implementation of the intelligent control system includes: processor, memory and communication module. The processor can be an industrial-grade microprocessor or an embedded processor, the memory is used to store algorithms and data, and the communication module is used to exchange data with the internal devices of the control cabinet and the external control system.

[0041] Embodiment 2: Temperature prediction processing unit like Figure 8 As shown, the temperature prediction processing unit is implemented based on a particle swarm-neural network hybrid algorithm for heat pipe network flow prediction, and its workflow includes the following steps: 1. Construction of thermal physics model: Establish the internal heat diffusion model of the control cabinet. The basic equation is: ; Where T is the temperature field, α is the thermal conductivity, and q is the heat source term (such as the heat generated by electronic components). represents the Laplace operator, which is expressed in three-dimensional space as: .

[0042] 2. Neural network structure design: Build a neural network suitable for temperature field prediction, including: Input layer: receiving the geometric parameters of the control cabinet (g i ), ambient temperature (T env ), component power distribution (P j ) and other characteristics; Hidden layer: adopts multi-layer perceptron structure and uses ReLU activation function; Output layer: predict the temperature value T(x,y,z,t) of each monitoring point; Neural network forward propagation calculation formula: ; ; Among them, X i is the input feature, H j is the hidden layer output, W ij , W jk is the weight, b j , b k is the bias and ReLU is the activation function.

[0043] 3. Particle swarm optimization algorithm implementation: The PSO algorithm is used to optimize the neural network parameters. The main steps include: a) Initialize the particle swarm: Generate 30-50 particles, each particle represents a set of neural network parameters. Initial particle position x i and speed v i Randomly generated.

[0044] b) Fitness evaluation: Calculate the prediction error of the neural network corresponding to each particle. The fitness function is: ; Where MSE is the mean square error, Reg is the regularization term, and λ is the weight coefficient.

[0045] c) Update speed and position: ; . Where x i represents the neural network weight, v i represents the particle speed, w represents the inertia weight, c 1 and c 2 represents the acceleration constant, r 1 and r 2 represents a random number, p best,i represents the optimal position of an individual, g best represents the global optimal position.

[0046] The parameters are set as follows: the inertia weight w ranges from 0.7 to 0.9, the acceleration constant c 1 and c 2 Both are 2.0, r 1 and r 2 is a random number in the interval (0,1).

[0047] d) Update the individual optimal p best,i and the global optimal g best； e) Iterate 100-200 times or until convergence: Physical constraint integration: Physical constraints are introduced into the neural network training process, and the loss function is designed as: ; the second term forces the heat conduction equation to be satisfied, the third term forces the fluid to be incompressible, λ 1 , 2 is the weight coefficient.

[0048] Incremental learning mechanism: continuously updating the model with new data: ; Where W is the neural network weight, γ is the learning rate (0.01-0.05), D new For newly collected data.

[0049] Example 3: Airflow simulation unit like Fig. 9 As shown in the figure, the airflow simulation unit is implemented based on the adaptive mesh finite element algorithm of fluid-structure interaction, and its workflow includes: Geometric model construction: Establish a three-dimensional geometric model based on the actual structure of the control cabinet, including the cabinet body, assembly frame, electronic components, etc.

[0050] Mesh generation: Generate the initial computational mesh using adaptive mesh technology, with the following parameters: Initial mesh size: 5-10mm; Minimum grid size: 0.5mm; Grid growth rate: 1.2.

[0051] Fluid-structure interaction model solution: Establish a coupling model of fluid and structure: a) Fluid part (Navier-Stokes equations): and ; Where u is the fluid velocity vector, p is the pressure, and ρ is the fluid density (the air density is 1.2 kg / m 3 ), μ is the dynamic viscosity (1.8×10 -5 Pa·s), f is the body force (such as gravity).

[0052] b) Structural part (elastic dynamics equation): ; Where d is the structural displacement vector, ρ s is the structural density, σ is the stress tensor, f s For external force.

[0053] c) Coupling conditions: d fluid | Γ =d solid | Γ Indicates displacement continuity; σ fluid ·n| Γ =σ solid ·n| Γ Indicates force balance; Where Γ represents the fluid-structure interface and n is the interface normal vector.

[0054] Adaptive Mesh Refinement: Dynamically adjusts mesh density based on solution gradients and errors: a) Error estimation: Calculate the error index of the current solution ; where e is the error estimate between the numerical solution and the exact solution, and Ω is the computational domain.

[0055] b) Grid marking: Mark the cells that need to be encrypted or coarsened according to the error index.

[0056] c) Grid update: adjust cell size according to markers Where h is the mesh size, p is the polynomial order in the finite element space (usually 2), and η target is the target error (set to 0.001-0.005).

[0057] d) Solution interpolation: interpolate the solution on the old grid to the new grid; e) Iterative solution: Continue solving on the new grid and perform 3-5 adaptive iterations.

[0058] Thermal-fluid coupling calculation: Consider the coupling of fluid flow and heat transfer: Among them, c p is the specific heat capacity, k is the thermal conductivity, and Q is the heat source term.

[0059] Example 4: Collaborative Optimization Unit like Fig.10 As shown in the figure, the collaborative optimization unit realizes the collaborative work of the temperature prediction processing unit and the airflow simulation unit, and includes the following key modules: 1. Multi-scale fusion module: realize the fusion of micro- and macro-scale heat flow models a) Scale separation: divide the computational domain into key areas and non-key areas; Key areas: around high-power electronic components, at airflow inlets and outlets, etc., using high-precision FSI calculations; Non-critical areas: other locations, predicted using efficient neural networks.

[0060] b) Regional division strategy: determine the regional properties based on temperature gradient ▏▽T▏ and power density P / V; If ▏▽T▏>Threshold T or P / V>Threshold P , it is divided into the critical area; otherwise it is divided into the non-critical area. T is the temperature gradient threshold (5℃ / cm), Threshold P is the power density threshold (0.1W / cm 3 ).

[0061] 2. Data-Physics Hybrid Training Module: A fusion of data-driven and physics-driven methods a) Physically enhanced neural network: The FSI simulation results are used as neural network training data, and physical conservation laws are added as constraints to the loss function: , where λ 1 = 0.1, λ 2 =0.05 is the weight coefficient.

[0062] b) Data-assisted adaptive standard: using neural network prediction results to assist the adaptive grid adjustment of the FSI algorithm: , where η FEM is the traditional error estimate, |T FEM -T NN | is the difference between the finite element solution and the neural network prediction, and β = 0.3 is the mixing coefficient.

[0063] 3. Adaptive computing strategy module: such as Fig.11 As shown, a three-level computing architecture is implemented to dynamically adjust the computing strategy according to the working conditions. a) Real-time monitoring layer (high frequency, low precision): direct prediction using pre-trained neural networks; computational complexity O(m), response time <10ms; suitable for normal working condition monitoring.

[0064] b) State assessment layer (medium frequency, medium precision): Use the PSO-NN algorithm for fast optimization prediction; computational complexity O(n·m), response time <1s; suitable for load changes or temperature fluctuations.

[0065] c) Deep analysis layer (low frequency, high precision): full simulation using FSI; computational complexity O(n 3 ), response time 10-100s; suitable for abnormal conditions or regular system evaluation.

[0066] d) Rules for automatic switching of calculation levels: .

[0067] 4. Complementary verification mechanism module: verify the consistency of different algorithm results a) Anomaly detection: When the difference between the prediction results of two algorithms exceeds a threshold, an alarm is triggered: ; .

[0068] b) Credibility assessment: Dynamically evaluate the prediction credibility based on the consistency of the two algorithms ; c) Self-calibration system: continuously corrects model parameters through actual temperature monitoring data Physical parameter identification: ; Neural network adaptation: .

[0069] Embodiment 5: Control execution unit The control execution unit realizes intelligent control of the control cabinet cooling system according to the results of the collaborative optimization unit, including the following modules: 1. Hotspot warning module: predicts the formation of hotspots before the actual temperature rises.

[0070] a) Hotspot risk assessment: , where T critical is the critical temperature threshold (set to 70°C).

[0071] b) Risk probability calculation: Based on historical data and current status assessment: ; The sigmoid activation function is ; where w 1 ,w 2 ,w 3is the weight parameter, b is the bias parameter, which is obtained through historical data learning; : The risk probability of the temperature exceeding the critical value; : current temperature; : Temperature change rate; : Current load power.

[0072] 2. Predictive cooling control module: Activate directional cooling in advance based on hotspot risks.

[0073] a) Cooling power calculation: ; where T safe is the safety temperature threshold (set to 50°C), K is the proportionality coefficient, and α is the nonlinear exponent (set to 1.5).

[0074] b) Cooling strategy: If HotspotRisk<0.3, maintain basic cooling; if 0.3≤HotspotRisk<0.7, enhance cooling of target area; if HotspotRisk≥0.7, activate maximum cooling capacity and issue an early warning.

[0075] 3. Cooling resource allocation module: optimizes cooling system energy usage.

[0076] a) Cooling demand calculation: ; where ρ is the density, c p is the specific heat capacity, T Target is the target temperature.

[0077] b) Optimal allocation of cooling power: ; where P i is the power of the i-th cooling device, Q i is the corresponding cooling capacity.

[0078] c) Solution method: Use Lagrange multiplier method and KKT conditions to solve the optimization problem.

[0079] 4. Maintenance decision support module: predict component life and provide maintenance recommendations.

[0080] a) Component life prediction: based on Arrhenius equation ; where L is the life expectancy, L 0 is the base lifetime, Ea is the activation energy, which is 0.7-1.1 eV, k is the Boltzmann constant, T 0 is the reference temperature (usually 25°C), and T is the predicted operating temperature.

[0081] b) Optimal maintenance time recommendations: ; where C maintenance is the maintenance cost, C failure is the failure cost, P failureis the failure probability, To find the t that minimizes the expression in the brackets.

[0082] Example 6: Algorithm Collaborative Working Mechanism like Fig.12 As shown, the core innovation of the present invention lies in the collaborative working mechanism of the two algorithms, which mainly includes the following aspects: 1. Bidirectional data interaction: bidirectional information flow between the two algorithms a) Data flow from FSI to NN: FSI simulation results serve as training data for NN; FSI physical constraints are embedded in the loss function of NN; the temperature field calculated by FSI is used to verify NN predictions.

[0083] b) Data flow from NN to FSI: NN prediction results provide initial conditions for FSI; NN participates in the adaptive meshing strategy of FSI; NN assists in determining when FSI needs to be recalculated.

[0084] 2. Incremental update mechanism: The two algorithms promote each other and continuously optimize.

[0085] a) FSI incremental update neural network: Where γ is the learning rate (0.01-0.05), D FSI New data generated for FSI.

[0086] b) Neural network assisted FSI initial value setting: ; This method can significantly accelerate the convergence process of FSI.

[0087] 3. Time series fusion mechanism: consider the impact of historical data on current evaluation.

[0088] a) Timing fusion formula: ; For this application: ; where Z T is the fusion result at the current moment, Z t-i is the fusion result before i time units.

[0089] b) Exponential smoothing forecast: used for short-term trend forecast: ; where α is the smoothing factor (0.3-0.4).

[0090] The following example calculation process is used to verify the thermal management system of the intelligent control system of the laser cutting machine control cabinet described in the present invention.

[0091] 1. Calculation parameter setting The laser cutting machine control cabinet is used as the main body, and the size of the control cabinet is 800mm×600mm×1800mm (width×depth×height), and components such as a main control board, a driver, a power supply and a heat dissipation device are installed inside. The control cabinet adopts the structure described in the present invention, including a cabinet body (1), a cabinet door (2) and a ventilation slot (3). There are 5 assembly racks (5) inside the control cabinet, and different functional components are installed in each assembly rack, wherein the driver with larger power is located on the second and third assembly racks.

[0092] In order to verify the effectiveness of the proposed method, the following detailed calculation process is carried out, which is based on the particle swarm-neural network hybrid algorithm for heat pipe network flow prediction and the adaptive mesh finite element algorithm based on fluid-structure interaction. Through the synergy of the two algorithms, accurate prediction and control of the temperature field are achieved.

[0093] 2. Initial parameter setting 2.1 Environmental and working conditions parameters: Ambient temperature: T ambient =25℃; relative humidity: RH=45%; atmospheric pressure: P atm =101.325kPa; Total power of electronic components in the control cabinet: P Total =4500W; Single component power distribution: Main control board: P 1 =150W; Driver 1: P 2 =1200W; Driver 2: P 3 =1100W; Power module: P 4 =950W; other components: P 5 =1100W.

[0094] 2.2 Material thermophysical parameters: Air thermal properties (25°C): Density: ρ air =1.184kg / m 3 Specific heat capacity: c Pair =1005J / (kg·K); thermal conductivity: k air =0.0261W / (m·K); dynamic viscosity: μ air =1.85×10^-5Pa·s.

[0095] Thermal properties of cabinet material (steel plate): Density: ρ steel =7850kg / m 3 Specific heat capacity: c Psteel =475J / (kg·K); thermal conductivity: k steel =45W / (m·K) Thermophysical parameters of main materials (aluminum alloy) of electronic components: Density: ρ al =2700kg / m3 Specific heat capacity: c Pal =897J / (kg·K); thermal conductivity: k al =237W / (m·K).

[0096] PCB board thermal physical parameters: density: ρ Pcb =1850kg / m 3 Specific heat capacity: c P-Pcb =950J / (kg·K); thermal conductivity: k Pcb =0.3W / (m·K).

[0097] 2.3 Geometric parameters of heat dissipation structure Ventilation slot size: 20mm×100mm×10mm (width×length×depth), 8 slots are set on the bottom side of each cabinet door; The size of the long strip heat dissipation holes is 10mm×150mm (width×length), with 12 holes on each side of the side panel and 8 holes on each side of the cover plate; Ventilation area: A vent =0.016m 2 (Ventilation slot)+0.048m 2 (Heat dissipation hole) = 0.064m 2 .

[0098] 3. Calculation process of particle swarm-neural network hybrid algorithm for heat pipe network flow prediction 3.1 Neural Network Structure Design Construct a neural network for predicting temperature field with the following structure: Input layer: 14 neurons (including geometric position coordinates x, y, z, assembly rack number, component power, ambient temperature, etc.); Hidden layer 1: 32 neurons, using ReLU activation function; Hidden layer 2: 16 neurons, using ReLU activation function; Output layer: 1 neuron (predicted temperature T); Total number of neural network parameters: N Params =(14×32+32)+(32×16+16)+(16×1+1)=448+528+17=993 parameters.

[0099] 3.2 Training Data Preparation To train the neural network, the following data is prepared: Historical operation data: 200 groups (temperature distribution collected under different power and environmental conditions); Simplified model simulation data: 300 groups (generated according to the basic thermodynamic equations) totaling 500 groups of training data, each group contains temperature monitoring point data at 30 different locations in the control cabinet.

[0100] 3.3 Particle swarm optimization algorithm parameter setting: Number of particles: n Particles =40; maximum number of iterations: maxiter=150; inertia weight: w=0.8; cognitive parameter: c 1 =2.0; Social parameter: c 2 =2.0; Particle dimension: dim=993 (equal to the total number of neural network parameters).

[0101] 3.4 PSO-NN algorithm calculation process Step 1: Initialize the particle swarm 40 particles are randomly generated, each particle represents a set of neural network parameters. Taking the first particle as an example, the initial position part value is: 1 (0)=[0.12,-0.25,0.31,...,0.08] Step 2: Evaluate initial particle fitness For each particle representing the neural network, the mean square error (MSE) is calculated using the training data: The first particle (initial): MSE 1 =53.2℃ 2 Best initial particle (23rd): MSE 23 =41.7℃, the global optimal position is initialized to the position of the 23rd particle: g best =x 23 (0) Step 3: Iterative Optimization Taking the 50th iteration as an example, calculate the speed and position update of the first particle: Calculation speed: v 1 (50)=0.8×v 1 (49)+2.0×0.63×[p best,1 -x 1 (49)]+2.0×0.41×[g best -x 1 (49)]; The random number r 1 =0.63,r 2 =0.41, p best,1 is the best historical position of the first particle.

[0102] Speed ​​calculation result (partial value):v 1 (50)=[0.05,-0.12,0.09,...,0.03] Update position:x 1 (50)=x 1 (49)+v1 (50)=[0.28,-0.31,0.42,...,0.15] Step 4: Evaluate the updated fitness Mean square error of the first particle after update: MSE 1 =18.4℃ 2 The global optimal particle after the 50th iteration (the 17th): MSE 17 =7.2℃ 2 .

[0103] Step 5: Algorithm convergence After 150 iterations, the algorithm converges to the global optimal solution: Final mean square error: MSE final =2.1℃ 2 ; Mean absolute error: MAE=1.1℃.

[0104] 3.5 Constraints on the Heat Diffusion Equation To ensure physical rationality, the heat diffusion equation is used as a constraint of the neural network: ; Discretization using finite difference method: ; in: T n i,j,k Represents the temperature at position (i,j,k) at time step n Δx, Δy, Δz are the spatial step lengths, all of which are 0.05m Δt is the time step, which is 0.5s α is the thermal diffusion coefficient, which is taken as air =k air / (ρ air ·c Pair )=2.19×10 -5 m 2 / s 3.6 PSO-NN calculation results After the PSO-NN algorithm optimization, the temperature distribution prediction model in the control cabinet was obtained. The third assembly rack (where the driver 2 is located): ambient temperature 25℃, under full load conditions: predicted maximum temperature: T maxPred =72.4℃; Hot spot location: center of heat sink of driver 2.

[0105] Temperature field distribution: temperature of the heat sink of driver 2: 72.4°C; temperature 10 cm around driver 2: 65.2°C; average temperature inside the assembly rack: 58.6°C; temperature at the edge of the assembly rack: 48.3°C.

[0106] 4. Calculation process of adaptive mesh finite element algorithm based on fluid-structure interaction 4.1 Geometry model and initial mesh A simplified 3D model is established based on the actual structure of the control cabinet. The initial mesh parameters are: Initial grid size: 10 mm; Minimum grid size: 0.5mm; Initial number of grid cells: 124,350; Initial number of grid nodes: 156,428.

[0107] 4.2 Boundary condition setting Inlet boundary (ventilation slot): natural convection boundary condition vin = 0.15m / s (initial velocity estimate) T in =25℃(ambient temperature) Outlet boundary (heat dissipation hole): pressure outlet p out =0Pa(relative pressure) Component surface: fixed heat flow boundary q 1 =150W; A 1 =4166.7W / m 2 (Main control panel) 2 =1200W; A 2 =12000W / m 2 (Driver 1);q 3 =1100W; A 3 =11000W / m 2 (Driver 2);q 4 =950W; A 4 =9500W / m 2 (power module); q 5 =1100W; A 5 =5500W / m 2 (other components). i Represents the surface area of ​​each component.

[0108] Cabinet outer wall: convection boundary condition h out =5W / (m 2 ·K) (natural convection heat transfer coefficient); T amb =25℃(ambient temperature).

[0109] 4.3 FSI solution process Step 1: Fluid flow calculation Solve the Navier-Stokes equations: and ; Use SIMPLE algorithm to solve: guess the initial flow field u and p; solve the momentum equation to get the predicted velocity u** ; Calculate the pressure correction equation; Update the pressure field p new =p * +p'; Update velocity field u new =u ** +u'; Check convergence and return if it does not converge.

[0110] After 200 iterations, the steady-state flow field is obtained: the average flow velocity in the control cabinet: v avg =0.21m / s; Flow velocity in hot spot area (driver 2): v hotspot =0.32m / s; flow rate: Q=0.0134m 3 / s.

[0111] Step 2: Heat conduction calculation Solve the energy equation: ; Use implicit finite volume method to solve: establish discrete equations of control volume; construct coefficient matrix; solve algebraic equations; check convergence.

[0112] After 150 iterations, the steady-state temperature field is obtained: Maximum temperature: T maxFSI =75.8°C (at driver 2); Average temperature in control cabinet: T avgFSI =47.2℃.

[0113] Step 3: Structural thermal stress calculation Solve the linear elastic equations: ; The thermal stress , E is the elastic modulus, and α is the thermal expansion coefficient.

[0114] Calculation results: Maximum thermal stress: σ max =24.2MPa (located at the driver 2 mounting bracket); Maximum displacement: d max =0.32mm Step 4: Adaptive Mesh Refinement First Mesh Refinement: Computing the Error Estimator , where e is the temperature field gradient error. The marking error is greater than the threshold η Threshold = 0.01 area for refinement.

[0115] Mesh parameters after refinement: Number of mesh units: 186425; Number of mesh nodes: 229687; Mesh size near driver 2: 2.5 mm Resolve the flow field and temperature field: Maximum temperature: T maxFSI1 =78.2℃ (located at driver 2); average temperature in the control cabinet: T avgFSI1=48.4℃.

[0116] Second mesh refinement: According to the formula Calculate the new grid size, where η target =0.003, p=2. Parameters of refined mesh: Number of mesh elements: 267842; Number of mesh nodes: 342156; Mesh size near driver 2: 0.8 mm Resolve the flow field and temperature field: Maximum temperature: T maxFSI2 =79.3℃ (located at driver 2); average temperature in the control cabinet: T avgFSI2 =48.7℃ The third mesh refinement (final result): number of mesh elements: 325674; number of mesh nodes: 412389; mesh size near driver 2: 0.5 mm.

[0117] Final calculation result: Maximum temperature: T maxFSIfinal =79.5℃ (located at driver 2); average temperature in the control cabinet: T avgFSIfinal =48.8℃ 4.4 FSI calculation results Detailed temperature distribution based on FSI algorithm: The maximum temperature of the driver 2 surface: 79.5℃; Driver 2 heat sink average temperature: 76.2°C; Air temperature 10cm away from driver 2: 62.6℃; Average temperature of the 3rd assembly rack: 59.4℃; The inner surface temperature of the cabinet door: 39.2℃; Cabinet outer surface temperature: 32.6℃.

[0118] 5. Collaborative Calculation Process of Two Algorithms 5.1 Multi-scale Fusion Computation Divide the computational domain into critical areas and non-critical areas: Key area: near driver 1 and driver 2 (temperature gradient greater than 5°C / cm), calculated using the FSI algorithm; Non-critical area: The remaining area is calculated using the PSO-NN algorithm.

[0119] Step 1: Initial temperature field estimation The PSO-NN algorithm is used to quickly generate the initial temperature field of the entire control cabinet, with a calculation time of 0.12 seconds. Result: Estimated maximum temperature: T maxinit =72.4℃; Average temperature in the control cabinet: T avginit =46.5℃.

[0120] Step 2: FSI refinement calculation of key areas Using the PSO-NN result as the initial value, the FSI refinement calculation is performed on the driver 1 and driver 2 areas, and the calculation time is 35 seconds. Result: The maximum temperature of driver 2: T maxFSIlocal =79.5℃; Driver 1 maximum temperature: T maxFSIlocal2 =77.2℃.

[0121] Step 3: Regional result fusion The results of the key and non-key areas are fused using the weighted average method: ; The weight function w(x, y, z) is 1 in the center of the key area and gradually decreases to 0 outward, using a smooth transition function: d(x,y,z) is the distance from point (x,y,z) to the center of the key area, and L=0.15m is the characteristic length.

[0122] Fusion result: Final maximum temperature: T maxfinal =79.5℃; Average temperature in the control cabinet: T avgfinal =49.2℃.

[0123] 5.2 Physically enhanced neural network training Use FSI calculation results to enhance neural network training and design physical enhancement loss function: ;in: MSE(T Pred ,T FSI ) is the mean square error between the predicted temperature and the FSI result Constrained residual for the heat diffusion equation |▽·u| is the residual of the fluid incompressibility constraint λ 1 = 0.1, λ 2 =0.05 is the weight coefficient Training results: original neural network prediction mean square error: MSE original =48.3℃ 2 ; Prediction mean square error after physical enhancement: MSE enhanced =12.7℃ 2 ; Prediction accuracy improved: 73.7%.

[0124] 5.3 Adaptive Computing Strategy A three-level computing architecture is used to address different temperature gradients: Case 1: Stable operating condition (maximum temperature gradient < 3℃ / cm): Direct prediction using pre-trained neural network; Calculation time: 0.008 s; Temperature prediction error: ±1.5℃ Case 2: Load change operating condition (maximum temperature gradient is 6℃ / cm): Optimized prediction using PSO-NN algorithm; Calculation time: 0.6 s; Temperature prediction error: ±0.8℃ Case 3: Extreme operating condition (maximum temperature gradient is 10℃ / cm): Collaborative calculation using FSI full simulation and PSO-NN; Calculation time: 28 s; Temperature prediction error: ±0.3℃.

[0125] 5.4 Complementary verification mechanism calculation Taking the driver 2 area as an example, compare the prediction results of PSO-NN and FSI algorithms: PSO-NN predicted temperature: T NN = 72.4℃; FSI calculated temperature: T FSI = 79.5℃; Temperature difference: ΔT = 7.1℃; Relative difference: ΔT / T FSI = 8.9%; Anomaly detection calculation: |T NN - T FSI | = 7.1℃; 0.15·max(T FSI ) = 0.15×79.5℃ = 11.9℃.

[0126] Since 7.1℃ < 11.9℃, the anomaly alarm is not triggered.

[0127] Confidence evaluation: = exp(-0.5·|T NN - T FSI | / T FSI ) = exp(-0.5×0.089) = exp(-0.0445) = 0.957 The confidence of the algorithm prediction is 95.7%.

[0128] 5.5 Collaborative calculation results Temperature field prediction results: Highest temperature: T max = 79.5℃ (driver 2); Average temperature inside the control cabinet: T avg = 49.2℃; Temperature distribution error (compared with actual measurement): Average error ±0.8℃, Maximum error ±1.6℃.

[0129] Hotspot warning results: Hotspot risk assessment: HotspotRisk=0.68; Early warning time: 37 seconds.

[0130] Cooling control optimization: Optimal cooling power: P cooling =820W.

[0131] Energy saving effect: Save 23.5% energy compared with traditional constant temperature control.

[0132] Equipment life impact assessment: Component life prediction: =40000 exp(-0.8 / 8.617×10^-5·(1 / 298.15-1 / 352.65))=40000·exp(-0.8 / 8.617×10^-5·(-0.000516))=40000·exp(47.93)=40000·0.618=24720 hours Life prediction after using collaborative algorithm control: When the maximum temperature drops to 70℃ (343.15K): L=40000 exp(-0.8 / 8.617×10^-5·(1 / 298.15-1 / 343.15))=40000·exp(-0.8 / 8.617×10^-5·(-0.000438))=40000·exp(40.67)=40000·0.666=26640 hours Therefore, the lifespan improvement rate is 7.8%.

[0133] 6. Verification and comparison of calculation results To verify the effectiveness of the algorithm, the calculated results are compared with the actual measured data: 6.1 Comparison of temperature field prediction accuracy

[0134] Calculation average error: PSO-NN algorithm: 3.72℃; FSI algorithm: 0.88℃; collaborative algorithm: 0.66℃ 6.2 Computational Efficiency Comparison

[0135] 6.3 Control effect comparison

[0136] 7. Calculation conclusion Through the above calculation process, the following conclusions can be drawn: Temperature prediction accuracy: The average prediction error of the collaborative algorithm is 0.66°C, which is significantly improved compared with the PSO-NN algorithm (3.72°C) and FSI algorithm (0.88°C) used alone.

[0137] Computational efficiency: The collaborative algorithm has an average computational time of 28 seconds, which is 6.4 times faster than the FSI algorithm (180 seconds). Although it is slower than the PSO-NN algorithm (0.12 seconds), its accuracy is greatly improved.

[0138] Control effect: The temperature fluctuation range under collaborative algorithm control is ±2.8℃, which is significantly lower than traditional control (±8.5℃) and PSO-NN control (±4.2℃).

[0139] Energy efficiency: The collaborative algorithm consumes 9.5 kWh of energy per day, which is 23.4% less than traditional control and 12.0% less than PSO-NN control.

[0140] Early warning capability: The collaborative algorithm can predict the formation of hot spots 37 seconds in advance, providing sufficient response time for the control system.

[0141] Equipment life: The expected life of equipment under collaborative algorithm control is increased by 7.8%, which has good economic value.

[0142] In summary, the present invention can effectively improve the temperature prediction accuracy, control effect and energy efficiency of the intelligent control system of the laser cutting machine control cabinet, while reducing the equipment failure rate and extending the service life of the equipment, which is of great significance.

Claims

1. An intelligent control system for a laser cutting machine control cabinet, characterized in that: include: A control cabinet body, the control cabinet body comprising a cabinet body (1), cabinet doors (2) being hingedly connected to both sides of the front of the cabinet body (1), a plurality of ventilation slots (3) being provided on the lower sides of the cabinet doors (2), a plurality of assembly racks (5) being installed in the cabinet body (1), the assembly racks (5) comprising side panels (7), the side panels (7) and the cover plate (15) being provided with a plurality of long strip-shaped heat dissipation holes (16); A temperature monitoring unit, used to collect temperature data inside the control cabinet body; A temperature prediction processing unit, including a heat pipe network flow prediction module, which predicts the heat flow distribution inside the control cabinet based on a particle swarm-neural network hybrid algorithm; An airflow simulation unit, including a fluid-structure interaction simulation module, wherein the fluid-structure interaction simulation module simulates the interaction between the airflow and the structure inside the control cabinet based on an adaptive mesh finite element algorithm; A collaborative optimization unit, used to combine the output results of the temperature prediction processing unit and the airflow simulation unit to achieve intelligent management of the temperature and airflow inside the control cabinet; The control execution unit is used to adjust the cooling system of the control cabinet according to the optimization result of the collaborative optimization unit to maintain the temperature inside the control cabinet within a preset range.

2. The laser cutting machine control cabinet intelligent control system according to claim 1 is characterized in that: The heat pipe network flow prediction module uses the following steps to predict the temperature field: a) Establish the heat diffusion equation model: , where T is the temperature field, α is the thermal conductivity, and q is the heat source term; b) constructing a neural network structure for fitting the temperature field distribution; c) Optimize neural network weights using particle swarm optimization algorithm: , where x i represents the neural network weight, v i represents the particle speed, w represents the inertia weight, c1 and c2 represent acceleration constants, r1 and r2 represent random numbers, p best,i represents the optimal position of an individual, g best represents the global optimal position.

3. The intelligent control system for a laser cutting machine control cabinet according to claim 1 is characterized in that: The fluid-structure interaction simulation module uses the following equations for simulation: a) Fluid part calculation: and , where u is the flow velocity, p is the pressure, ρ is the fluid density, μ is the dynamic viscosity, and f is the volume force; b) Structural calculation: , where d is the structural displacement, ρ s is the structural density, σ is the stress tensor, f s is the external force; c) Adaptive grid adjustment: , where h is the grid size, η is the error estimator, and p is the polynomial order in the finite element space.

4. The laser cutting machine control cabinet intelligent control system according to claim 1 is characterized in that: The collaborative optimization unit comprises: A multi-scale fusion module, used to realize the fusion of micro-scale and macro-scale thermal flow models, wherein the micro-scale model is realized by the fluid-structure interaction simulation module, and the macro-scale model is realized by the heat pipe network flow prediction module; Data-physics hybrid training module for building physically enhanced neural networks and data-assisted adaptive finite element models, including: Physically enhanced neural network loss function: , where MSE represents mean square error, λ1 and λ2 are weight coefficients; Data-Assisted Adaptive Standards: , where η FEM is the traditional error estimate, β is the mixing coefficient, T FEM is the finite element solution, T NN Predict results for the neural network; Adaptive computing strategy module, implementing a three-level computing architecture, including: Real-time monitoring layer: Direct prediction using pre-trained neural networks, with a response time of less than 10 milliseconds; State assessment layer: Use particle swarm-neural network algorithm to optimize prediction, with a response time of less than 1 second; Deep analysis layer: Uses full fluid-structure interaction simulation with a response time of 10-100 seconds; automatically switches calculation levels based on temperature gradient indicators.

5. The intelligent control system for a laser cutting machine control cabinet according to claim 1 is characterized in that: A complementary verification mechanism module is provided between the temperature prediction processing unit and the airflow simulation unit, for verifying the consistency of the prediction results of the two units, and the complementary verification mechanism module includes: Anomaly detection logic, through the formula Determine whether to trigger an abnormal alarm, where T NN is the predicted temperature of the heat pipe network flow prediction module, T FSI is the predicted temperature of the fluid-structure interaction simulation module, and δ is the threshold coefficient; Credibility evaluation logic, through the formula Calculate the credibility of the prediction results, where κ is the scaling factor; Self-calibration mechanism, which continuously corrects model parameters through actual temperature monitoring data, including: Physical parameter identification: , where α is the physical parameter and μ is the update step size; Neural network adaptation: , where ω1 and ω2 are weight coefficients, T measured To actually measure the temperature.

6. The laser cutting machine control cabinet intelligent control system according to claim 1 is characterized in that: The control execution unit comprises: Hotspot early warning module predicts the risk of hotspot formation through collaborative algorithms: , where T critical is the critical temperature threshold; Predictive cooling control module, pre-activating directional cooling based on hotspot risk: , where T safe is the safety temperature threshold; Cooling resource allocation module, through optimization problem To achieve the optimal distribution of cooling power, where P i is the power of the i-th cooling device, Q i is the corresponding cooling capacity, Q cooling For cooling needs; Maintenance decision support module, through component life prediction equation Predict the life of electronic components, where L is the expected life, L0 is the reference life, Ea is the activation energy, k is the Boltzmann constant, T0 is the reference temperature, and T is the predicted operating temperature.

7. The intelligent control system for a laser cutting machine control cabinet according to claim 4 is characterized in that: The adaptive computing strategy module automatically selects the computing level according to the following rules: ,in Indicates the maximum value of temperature gradient, Threshold1 and Threshold2 are preset thresholds, and Level is the calculation level.

8. The intelligent control system for a laser cutting machine control cabinet according to claim 5 is characterized in that: The temperature prediction processing unit also includes an incremental update mechanism to achieve self-optimization of the algorithm in the following ways: FSI incrementally updates the neural network: , where W represents the neural network weight, γ is the learning rate, and D FSI Data generated for fluid-structure interaction simulations; Neural network assisted FSI initial value setting: , where T0, u0, and p0 are the initial conditions of temperature, velocity, and pressure, respectively.

9. The laser cutting machine control cabinet intelligent control system according to claim 6, characterized in that: The control execution unit also includes a timing optimization module, which considers historical temperature data through a timing fusion mechanism: , where Z T is the fusion result at the current moment, Z (t-i) is the fusion result before i time units, α i is the time attenuation coefficient, satisfying .

Images