Furniture processing control system based on artificial intelligence

Abstract

The invention discloses a furniture processing control system based on artificial intelligence, and relates to the technical field of intelligent manufacturing and industrial automation. According to the system, cutter vibration frequency spectrum, plate texture features and environment temperature and humidity data are collected in real time through a distributed sensor array, multi-source data are fused through a space-time attention mechanism, and a joint feature matrix is generated. And based on the joint feature matrix, utilizing a depth map neural network to deconstruct a topology constraint relation of non-standard customization requirements, and generating an initial processing parameter set. And driving the digital twin model to perform virtual processing according to the initial processing parameter set, and predicting a processing node deformation error and generating a compensation vector in combination with the three-dimensional laser point cloud and infrared thermal imaging data. The compensation vector and real-time working condition data are received, a path cost function is evaluated through Monte Carlo tree search and a time sequence convolutional network, the cutter feeding speed and the cutting depth are corrected, and dynamic regulation and control of cutter path parameters are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing and industrial automation, and particularly to a furniture processing control system based on artificial intelligence. Background Technique

[0002] With the acceleration of the individualization of consumer demand and the intelligent transformation of the manufacturing industry, the furniture industry is facing a profound transformation from large-scale production to customized services. Consumers' demand for non-standard furniture continues to grow, requiring enterprises to quickly respond to diverse design requirements while taking into account delivery speed and product quality. Under the requirements of environmental protection policies and sustainable development, the industry urgently needs to solve the problems of material waste and high energy consumption. The traditional furniture manufacturing mode relies on manual experience and fixed process flows, making it difficult to adapt to dynamic order changes, resulting in low production efficiency and insufficient resource utilization, which has become the key bottleneck restricting the industry's upgrade.

[0003] However, the existing traditional furniture processing control system relies on fixed process parameters and manual experience, with insufficient flexibility leading to low efficiency in changeover adjustment and difficulty in adapting to the processing of non-standard parts; precision control overly relies on manual sampling inspection, resulting in unstable processing quality of complex components; the equipment energy consumption management is extensive, with serious losses during no-load operation; the intelligent level is insufficient, and the process planning relies on a preset rule engine, making it difficult to dynamically respond to real-time working conditions, resulting in material waste and limited production efficiency. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a furniture processing control system based on artificial intelligence, which solves the problems in the above background technique.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A furniture processing control system based on artificial intelligence includes the following modules: a multi-modal perception module, a process parameter generation module, a dynamic error compensation module, and a collaborative control module; the multi-modal perception module is used to collect multi-source heterogeneous data in real time through a distributed sensor array, including tool vibration spectrum, board texture features, and environmental temperature and humidity data, and fuse and encode the multi-source heterogeneous data through a spatio-temporal attention mechanism to generate a joint feature matrix; the process parameter generation module is used to deconstruct the topological constraint relationship of non-standard customization requirements based on the joint feature matrix, and generate an initial processing parameter set of cutting force threshold and path smoothness in combination with a parameter decoupling algorithm; the dynamic error compensation module is used to drive a digital twin model for virtual processing according to the initial processing parameter set, synchronously fuse three-dimensional laser point cloud and infrared thermal imaging data, predict the deformation error of future processing nodes through the gated recurrent unit of a recurrent neural network, and generate an error compensation vector; the collaborative control module is used to receive the error compensation vector and real-time working condition data, generate candidate paths through Monte Carlo tree search, evaluate the comprehensive cost function of the paths, extract vibration spectrum features through a temporal convolutional network, generate correction amounts of tool feed speed and cutting depth, and dynamically adjust the tool path parameters.

[0006] Furthermore, the specific process of fusing and encoding multi-source heterogeneous data through a spatio-temporal attention mechanism to generate a joint feature matrix is as follows: Collect the three-dimensional vibration signals of the tool through sensors deployed in a distributed manner, extract the time-frequency features of the tool vibration spectrum through short-time Fourier transform, synchronously capture the surface images of the board through an industrial camera array, extract the board texture features through a multi-scale convolutional neural network, and fuse the environmental temperature and humidity data collected by the temperature and humidity sensors; After spatio-temporal alignment of the multi-source heterogeneous data, dynamically model the correlation between vibration time series features and texture spatial features through a multi-head attention mechanism, calculate the cross-modal similarity through dot product and generate a weight matrix, and adjust the feature fusion weight in combination with environmental parameters; Reduce the dimension and compress the fused features through an autoencoder, and then extract local spatio-temporal patterns through a spatio-temporal convolutional network to generate a joint feature matrix.

[0007] Furthermore, the specific process of deconstructing the topological constraint relationship of non-standard customization requirements through a deep graph neural network is as follows: Convert the CAD drawing of the furniture into graph structure data, where the nodes represent furniture components, including coordinates, curvatures, and connection types, and the edges represent the connection relationships between components; Iteratively aggregate the spatial features of adjacent nodes through a multi-layer graph convolutional network, enhance the non-linear expression ability through an activation function, and output a topological constraint matrix to clarify the processing sequence and spatial dependence relationship of the components.

[0008] Furthermore, the specific process of generating the initial machining parameter set of cutting force threshold and path smoothness by combining the parameter decoupling algorithm is as follows: Taking the topological constraint matrix and material hardness parameters as inputs, a mixed-integer programming model is constructed with the objective function of minimizing the cutting force deviation and the path smoothness is constrained to meet the machining requirements. The objective function is solved by the branch and bound algorithm to generate the initial parameter set including the cutting force threshold, path curvature radius and feed rate.

[0009] Furthermore, the specific process of driving the digital twin model for virtual machining according to the initial machining parameter set and synchronously fusing the three-dimensional laser point cloud and infrared thermal imaging data is as follows: Load the initial machining parameters in the digital twin environment to drive the virtual tool to execute the planned path. Synchronously collect the high-frame-rate three-dimensional point cloud data through the lidar and perform spatio-temporal alignment with the tool pose. At the same time, obtain the temperature field distribution map of the infrared thermal imager and extract the thermal gradient characteristics of the tool-workpiece contact area to provide multi-modal input for deformation prediction.

[0010] Furthermore, the specific process of predicting the deformation error of future machining nodes by the gated recurrent unit of the recurrent neural network and generating the error compensation vector is as follows: Construct the three-dimensional point cloud data, thermal gradient characteristics and vibration signals into a spatio-temporal sequence and input it into the bidirectional gated recurrent unit network. The forward gated recurrent unit network captures the historical error propagation law, and the backward gated recurrent unit network predicts the future trend. After fusing the bidirectional hidden states, the deformation prediction value of future machining nodes is output to generate the error compensation vector including the three-dimensional displacement compensation amount and cutting force adjustment.

[0011] Furthermore, the specific process of generating candidate paths by Monte Carlo tree search and evaluating the comprehensive cost function of the paths is as follows: Expand the candidate path set with the current tool position as the root node, define the comprehensive cost function, including the predicted deformation error, material utilization rate and tool wear rate, and dynamically adjust the weights of each sub-item according to the sheet hardness. Screen the Pareto optimal path through the simulated annealing algorithm to balance the contradictory relationship between machining accuracy and resource consumption.

[0012] Furthermore, the specific process of extracting the vibration spectrum characteristics by the temporal convolutional network and generating the correction amounts of the tool feed rate and cutting depth is as follows: Perform wavelet packet decomposition on the vibration signal of the selected path and extract the multi-band energy characteristics as the temporal input. According to the temporal convolutional network to model the temporal correlation between the vibration characteristics and machining parameters, predict the real-time correction amounts of the tool feed rate and cutting depth, and generate the dynamic adjustment instruction.

[0013] Furthermore, the specific process of dynamically regulating the tool path parameters is as follows: Verify the physical feasibility of the control instructions in the digital twin system, including tool stiffness simulation, material fracture toughness detection, and motion interference analysis; After verification, send the instructions to the numerical control system for execution through the industrial communication protocol, and monitor the actual machining error in real time. If it exceeds the set threshold, trigger the recalculation of the compensation vector and update the process parameter set to form a closed-loop control.

[0014] The present invention has the following beneficial effects:

[0015] (1) An artificial intelligence-based furniture processing control system realizes the deep coupling of tool vibration, board texture, and environmental parameters through the spatio-temporal attention mechanism of the multi-modal perception module, significantly improving the information density and robustness of data representation. Based on the topological deconstruction ability of the deep graph neural network and the parameter decoupling algorithm, the process parameter generation module breaks through the bottleneck of converting non-standard customization requirements into an executable process chain, shortens the process adaptation time for non-standard parts, and at the same time ensures the precise matching of the cutting force threshold and path smoothness, reducing the risk of machining errors from the source.

[0016] (2) An artificial intelligence-based furniture processing control system. The dynamic error compensation module realizes the early perception and active compensation of machining deformation errors through digital twin-driven multi-modal prediction technology, effectively reducing material waste and rework rate; The collaborative control module adopts a multi-algorithm collaborative optimization strategy to synchronously balance multiple objectives of machining accuracy, resource utilization, and equipment loss, significantly improving the machining quality and material utilization rate of complex components, while reducing equipment energy consumption, forming an intelligent machining closed-loop control ability with both high precision and sustainability.

[0017] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of an artificial intelligence-based furniture processing control system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The embodiment of the present application solves the problems of difficult adaptation of flexible production, insufficient machining accuracy, extensive energy consumption management, and lack of real-time dynamic regulation ability in traditional furniture manufacturing through an artificial intelligence-based furniture processing control system. By integrating tool, board, and environmental data through the multi-modal perception module, the process parameter generation module realizes the intelligent conversion of non-standard requirements, the dynamic error compensation module predicts deformation errors in advance, and the collaborative control module optimizes the multi-objective machining path, forming a full-process closed-loop from data perception to dynamic regulation, significantly improving the quality stability, resource utilization rate, and response efficiency of customized furniture production.

[0020] The overall idea of the solution in the embodiments of this application is as follows:

[0021] Collect multi-source heterogeneous data in real time through a distributed sensor array, including tool vibration spectra, sheet texture features, and environmental temperature and humidity data, and fuse and encode the multi-source heterogeneous data through a spatio-temporal attention mechanism to generate a joint feature matrix.

[0022] Based on the joint feature matrix, deconstruct the topological constraint relationship of non-standard customization requirements through a deep graph neural network, and combine a parameter decoupling algorithm to generate an initial set of machining parameters for cutting force thresholds and path smoothness.

[0023] Drive a digital twin model for virtual machining according to the initial set of machining parameters, synchronously fuse three-dimensional laser point cloud and infrared thermal imaging data, and predict the deformation error of future machining nodes through the gated recurrent unit of a recurrent neural network to generate an error compensation vector.

[0024] Receive the error compensation vector and real-time working condition data, generate candidate paths through Monte Carlo tree search, evaluate the comprehensive cost function of the paths, extract vibration spectrum features through a temporal convolutional network, generate correction amounts for tool feed speed and cutting depth, and dynamically adjust the tool path parameters.

[0025] Please refer to Figure 1 , an embodiment of the present invention provides a technical solution: an artificial intelligence-based furniture processing control system, including the following modules: a multi-modal perception module, a process parameter generation module, a dynamic error compensation module, and a collaborative control module; the multi-modal perception module is used to collect multi-source heterogeneous data in real time through a distributed sensor array, including tool vibration spectra, sheet texture features, and environmental temperature and humidity data, and fuse and encode the multi-source heterogeneous data through a spatio-temporal attention mechanism to generate a joint feature matrix; the process parameter generation module is used to deconstruct the topological constraint relationship of non-standard customization requirements based on the joint feature matrix through a deep graph neural network, and combine a parameter decoupling algorithm to generate an initial set of machining parameters for cutting force thresholds and path smoothness; the dynamic error compensation module is used to drive a digital twin model for virtual machining according to the initial set of machining parameters, synchronously fuse three-dimensional laser point cloud and infrared thermal imaging data, and predict the deformation error of future machining nodes through the gated recurrent unit of a recurrent neural network to generate an error compensation vector; the collaborative control module is used to receive the error compensation vector and real-time working condition data, generate candidate paths through Monte Carlo tree search, evaluate the comprehensive cost function of the paths, extract vibration spectrum features through a temporal convolutional network, generate correction amounts for tool feed speed and cutting depth, and dynamically adjust the tool path parameters.

[0026] In this implementation plan, the multi-modal perception module: collects multi-source heterogeneous data of the processing site in real time through a distributed sensor array (such as high-frequency acceleration sensors, industrial cameras, temperature and humidity sensors), including: tool vibration spectrum: the vibration frequency and amplitude characteristics of the tool during the processing, reflecting the tool state and processing stability; board texture features: visual features such as the texture direction and knot distribution on the wood surface, affecting the cutting path planning; environmental temperature and humidity data: the temperature and humidity parameters of the processing area, affecting material deformation and tool performance. The spatio-temporal attention mechanism: through dynamic weight allocation, fuses vibration (time-series data), texture (spatial data) and environmental parameters (auxiliary variables), solves the problem of information fragmentation in traditional single-modal perception, and generates a joint feature matrix representing the overall processing state. The process parameter generation module: converts non-standard customized requirements (such as special-shaped furniture design) into executable processing parameters. The deep graph neural network (GNN): analyzes the topological constraint relationships of furniture CAD drawings (such as the component connection order and spatial dependency relationships), converts the design drawings into graph-structured data, where nodes represent component attributes and edges represent connection relationships; the parameter decoupling algorithm: disassembles complex non-standard requirements into standardized parameters that can be independently controlled (such as cutting force threshold, path smoothness), and generates an initial set of processing parameters that meet safety and quality requirements. The dynamic error compensation module: predicts the errors that may occur in actual processing in a virtual environment and generates compensation instructions in advance. The digital twin model: a virtual mirror of the physical production line, simulates the processing process after loading the initial parameters, and exposes potential problems in advance; the gated recurrent unit (GRU): a type of recurrent neural network that predicts the deformation errors of future processing nodes (such as size deviations caused by the thermal expansion and contraction of wood) through historical data, and generates an error compensation vector (three-dimensional displacement correction amount and cutting force adjustment value). The collaborative control module: dynamically adjusts the tool path parameters to balance multiple objectives such as processing accuracy, material utilization rate, and equipment loss. Monte Carlo tree search (MCTS): a heuristic search algorithm that generates candidate processing paths and evaluates their comprehensive costs (such as accuracy errors, material waste, tool wear); the temporal convolutional network (TCN): analyzes the temporal characteristics of vibration signals, generates real-time correction amounts for the tool feed speed and cutting depth, suppresses vibration interference, and improves processing stability. The distributed sensor array: multiple sensors are deployed according to spatial distribution, and cooperate to collect physical signals in different dimensions (such as vibration, images, environmental parameters), forming a comprehensive perception of the processing environment. Non-standard customized requirements: customer personalized design requirements (such as special-shaped curved surfaces, special connection structures), which cannot directly use standardized process parameters and need to be dynamically adapted. Topological constraint relationships: the spatial connection rules of furniture components (such as the order of mortise and tenon structures), which determine the dependence and priority of processing steps. The digital twin model: a virtual copy of the physical processing system, which realizes the visual prediction and optimization of the processing process through real-time data mapping and simulation.Pareto optimal path: A path in multi-objective optimization where one objective cannot be further optimized without sacrificing other objectives, representing the best balance point among accuracy, efficiency, and resource utilization.

[0027] Specifically, the specific process of fusing and encoding multi-source heterogeneous data through a spatio-temporal attention mechanism to generate a joint feature matrix is as follows: The three-dimensional vibration signal of the tool is collected by distributed sensors, and the time-frequency features of the tool vibration spectrum are extracted through short-time Fourier transform. Meanwhile, the surface image of the sheet is captured by an industrial camera array, and the texture features of the sheet are extracted through a multi-scale convolutional neural network. The environmental temperature and humidity data collected by the temperature and humidity sensors are also fused. After spatio-temporal alignment of the multi-source heterogeneous data, the correlation between the vibration time-series features and the texture spatial features is dynamically modeled through a multi-head attention mechanism. The cross-modal similarity is calculated through dot product to generate a weight matrix, and the feature fusion weight is adjusted in combination with environmental parameters. The fused features are compressed by an autoencoder, and then the local spatio-temporal patterns are extracted through a spatio-temporal convolutional network to generate a joint feature matrix.

[0028] In this implementation scheme, the multi-modal data fusion and encoding objective is to fuse the tool vibration, sheet texture, and environmental data to generate a joint feature matrix with high information density. Parameter explanation: x(τ): The three-dimensional vibration signal of the tool (unit: m / s 2 ), representing the dynamic behavior of the tool. w(τ): The Hanning window function (window length 10 ms), used to intercept local segments of the vibration signal. t: Time point (unit: second), indicating the current moment of signal analysis. f: Frequency (unit: Hz), reflecting the spectral components of the vibration signal. F v (t, f): The time-frequency energy matrix, describing the distribution of vibration energy over time and frequency. Parameter explanation: The query vector of the vibration time-series feature, mapped to a vector of dimension d. The key vector of the texture spatial feature, mapped to a vector of dimension d. d: Feature dimension (default 64), controlling the complexity of attention calculation. A a,j : The modal correlation weight, reflecting the correlation strength between the vibration feature a and the texture feature j. M = LayerNorm(F v + A·F t + E): The environmental parameter encoding vector (temperature and humidity data), with the same dimension as F v aligned. LayerNorm: Layer normalization operation, eliminating the dimensional differences of different sensors. M ∈ R T×d : The joint feature matrix, providing input for downstream modules.

[0029] Specifically, the specific process of deconstructing the topological constraint relationship of non-standard customization requirements through a deep graph neural network is as follows: Convert the CAD drawings of furniture into graph structure data, where nodes represent furniture components, including coordinates, curvature, and connection types, and edges represent the connection relationships between components; Iteratively aggregate the spatial features of adjacent nodes through a multi-layer graph convolutional network, enhance the non-linear expression ability through activation functions, and output a topological constraint matrix to clarify the processing order and spatial dependence relationship of components.

[0030] In this implementation plan, node definition: Each furniture component (such as a table leg, a panel) is defined as a node in the graph. Node features: Include geometric attributes (coordinates (x, y, z), curvature κ) and connection types (such as mortise and tenon, screws). Edge definition: If there is a physical connection between two components, an edge is added. Edge features: Can include information such as connection strength, direction (if complex modeling is required). Feature aggregation formula of the multi-layer graph convolutional network (GCN): Parameter explanation: The hidden feature vector of node i in the l-th layer (the initial layer is the original node feature). Learnable weight matrix for feature transformation. N(i): The set of adjacent nodes of node i (i.e., directly connected components). σ: GELU activation function, the formula is σ(x) = x × Φ(x), where Φ(x) is the standard Gaussian cumulative distribution function, enhancing the non-linear expression ability. Normalization term: Alleviate the influence of node degree differences on feature aggregation. Physical meaning: Feature aggregation: Each node learns the spatial dependence relationship between components by aggregating the features of adjacent nodes (for example: the chair legs need to be processed first, and the seat surface needs to be processed after the chair legs). Multi-layer stacking: Through 3 layers of graph convolution (l = 0, 1, 2), gradually expand the receptive field to capture the global topological constraints. Topological constraint matrix generation: Represents the feature vector of node i in the final layer L. The sigmoid function maps the node feature similarity to a probability value (range 0 - 1). The threshold 0.5 is used to determine whether component j needs to be processed before i (C ij = 1 represents a dependency relationship). If the chair leg node i and the seat surface node j satisfy C ij = 1, then the processing order is: chair legs → seat surface.

[0031] Specifically, taking the topological constraint matrix and material hardness parameters as inputs, construct a mixed integer programming model with minimizing the cutting force deviation as the objective function, and constrain the path smoothness to meet the processing requirements; Solve the objective function through the branch and bound algorithm to generate an initial parameter set including cutting force thresholds, path curvature radii, and feed speeds.

[0032] In this implementation plan, the construction and solution objective of the mixed-integer programming model is to generate safe and efficient initial machining parameters (cutting force, path curvature, feed rate). Formulas and parameters: Parameter explanation: min P Indicates that the objective is to minimize the cutting force parameter, P force : Cutting force parameter, which needs to approximate the maximum cutting force allowed by the material. F max (H m ): Material hardness H m The corresponding maximum allowable cutting force. P path : Path smoothness parameter to avoid tool breakage caused by sharp turns. τ smooth : Path smoothness threshold, determined by the material toughness. F safe (H m ): Safe cutting force threshold. Solving method: Branch and bound algorithm: Traverse the feasible solution space through tree-like search, prune invalid branches, and generate an initial parameter set that meets the constraints. s.t. means "subject to the following constraints".

[0033] Specifically, driving the digital twin model for virtual machining according to the initial machining parameter set, the specific process of synchronously fusing three-dimensional laser point cloud and infrared thermal imaging data is as follows: Load the initial machining parameters in the digital twin environment and drive the virtual tool to execute the planned path; Synchronously collect high-frame-rate three-dimensional point cloud data through lidar and perform spatio-temporal alignment with the tool pose. At the same time, obtain the temperature field distribution map of the infrared thermal imager, extract the thermal gradient characteristics of the tool-workpiece contact area, and provide multi-modal input for deformation prediction.

[0034] In this implementation plan, load the initial machining parameters (cutting force, path, speed) in the digital twin environment, drive the virtual tool to move along the planned path, and real-time simulate the physical machining process; Synchronously scan the workpiece surface through lidar at a high frame rate (≥200Hz) to generate three-dimensional point cloud data with millimeter-level accuracy, and strictly align it with the virtual tool pose through the coordinate transformation matrix to ensure geometric space consistency; At the same time, the infrared thermal imager captures the temperature field distribution of the tool-workpiece contact area at a rate of 30 frames per second, extracts the thermal gradient characteristics (temperature change rate), and characterizes the thermal stress distribution during the machining process. The point cloud data provides geometric deformation information, and the thermal gradient data reflects the thermodynamic effect. The two are fused to form multi-modal input, providing real-time and high-precision working condition data for the subsequent deformation prediction model, and realizing the deep coupling of the physical world and the virtual mirror.

[0035] Specifically, the specific process of predicting the deformation error of future machining nodes through the gated recurrent unit of the recurrent neural network and generating the error compensation vector is as follows: Construct the three-dimensional point cloud data, thermal gradient features, and vibration signals into a spatio-temporal sequence and input it into the bidirectional gated recurrent unit network; capture the historical error propagation law through the forward gated recurrent unit network, predict the future trend through the reverse gated recurrent unit network, and output the deformation prediction value of the future machining node after fusing the bidirectional hidden states, generating an error compensation vector including three-dimensional displacement compensation and cutting force adjustment.

[0036] In this implementation, the multi-modal sensor data is integrated into a temporal input to capture the spatio-temporal correlation in the machining process. Input data construction (spatio-temporal sequence generation) Objective: Integrate the multi-modal sensor data into a temporal input to capture the spatio-temporal correlation in the machining process. Formulas and parameters: Parameter explanation: The three-dimensional point cloud coordinates at time step t, representing the geometric deformation of the workpiece surface. The thermal gradient vector at time step t, reflecting the temperature change rate in the tool-workpiece contact area. The vibration signal frequency feature at time step t, representing the dynamic behavior of the tool. d input : Total dimension of input features. Bidirectional gated recurrent unit (BiGRU) network structure Objective: Model the historical law and future trend through the forward and reverse GRUs respectively, and fuse them to predict the deformation error. Formulas and parameters: Forward GRU (modeling historical law) Parameter explanation: The hidden state of the forward GRU at time step t, encoding the historical temporal information. X t : Current input feature vector (dimension 8). Reverse GRU (predicting future trend) Parameter explanation: The hidden state of the reverse GRU at time step t, encoding the future temporal information. Hidden state fusion and prediction output Parameter explanation: W0 ∈ R 4×256 : Output layer weight matrix, mapping the 256-dimensional concatenated features to a 4-dimensional compensation vector. b0 ∈ R 4 : Output layer bias term. The predicted compensation vector, where: δx, δy, δz: Three-dimensional displacement compensation; δforce: Cutting force adjustment. k = 30: Prediction step. 3. Internal calculation of GRU unit (taking a single time step as an example) Formula: z t = σ(W z ·[h t-1 , X t +b z )(Update gate) r t = σ(Wr · [h t-1 , X t + b r )(Reset gate) (Hidden state update) Parameter explanation: W z , W r , W h ∈R 128×(128+8) : Weight matrix (input dimension 8 + hidden state 128). b z , b r , b h ∈R 128 : Bias term. σ: Sigmoid function (output 0 to 1, controlling the information flow ratio). ⊙: Element-wise multiplication (Hadamard product).

[0037] Specifically, candidate paths are generated through Monte Carlo tree search, and the specific process of evaluating the comprehensive cost function of the paths is as follows: Taking the current tool position as the root node, expand the candidate path set, define the comprehensive cost function, including predicted deformation error, material utilization rate, and tool wear rate, and dynamically adjust the weights of each sub-item according to the hardness of the sheet; Screen the Pareto-optimal paths through the simulated annealing algorithm to balance the contradictory relationship between machining accuracy and resource consumption.

[0038] In this implementation scheme, the objectives of Monte Carlo tree search for generating candidate paths and comprehensive cost evaluation are: Generate and evaluate candidate machining paths, and balance machining accuracy, material utilization rate, and tool life. Steps and formulas: Candidate path generation: Root node: Current tool position p current =(x0, y0, z0). Path expansion: Generate candidate path sets R1, R2,..., R N through neighborhood discretization (grid resolution δ = 1mm). Comprehensive cost function: Parameter explanation: Δ pred : Predicted deformation error. η material ∈[0, 1]: Material utilization rate (effective cutting area / total area of the sheet). Tool wear rate (E is the tool elastic modulus, A = πr 2 ). Dynamic weight (based on the hardness H of the sheet m ): Simulated annealing screening: Energy function: E(R i ) = C(R i ) + T·∑ j λ j · max(0, g j (R i )) T: Temperature parameter. g j (R i ): Path constraint (such as curvature radius ≥ 10mm).

[0039] Specifically, the specific process of extracting vibration spectrum features through a temporal convolutional network and generating correction amounts for the tool feed speed and cutting depth is as follows: Perform wavelet packet decomposition on the vibration signals of the selected path, and extract multi-band energy features as the temporal input; According to the temporal convolutional network to model the temporal correlation between vibration features and machining parameters, predict the real-time correction amounts of the tool feed speed and cutting depth, and generate dynamic adjustment instructions.

[0040] In this implementation plan, the generation of vibration correction amounts by the temporal convolutional network (TCN): Suppress vibration interference and dynamically adjust the feed speed and cutting depth. Steps and formulas: Wavelet packet decomposition: E k = ∑ n |W k [n]| 2 ; Parameter explanation: W k [n]: Wavelet coefficients of the k-th sub-band (decomposition level = 4). E k : Sub-band energy (characterizing the vibration spectrum distribution). TCN correction amount prediction: Parameter explanation: τ = 10: Temporal window length. Δv feed : Feed speed correction amount. Δd cut : Cutting depth correction amount.

[0041] Specifically, the specific process of dynamically regulating the tool path parameters is as follows: Verify the physical feasibility of the control instructions in the digital twin system, including tool stiffness simulation, material fracture toughness detection, and motion interference analysis; After verification, send the instructions to the numerical control system for execution through the industrial communication protocol, and monitor the actual machining error in real time. If it exceeds the set threshold, trigger the recalculation of the compensation vector and update the process parameter set to form a closed-loop control.

[0042] In this implementation plan, the goal of dynamic regulation and closed-loop verification of the tool path: Ensure the physical feasibility and safety of the control instructions. Steps and formulas: Tool stiffness verification: Parameter explanation: F cut : Actual cutting force. L: Tool overhang length. (d is the tool diameter). σ yield : Yield strength of the tool material. Motion interference detection: min(||p tool -p obstacle ||≥5mm); Parameter explanation: p tool =(x t ,y t ,z t ): Coordinates of the tool tip. p obstacle: Coordinates of obstacles (fixtures, unprocessed areas). Closed-loop control logic: Instruction issuance: Transmit instructions to the numerical control system through the OPC UA protocol. Error monitoring: Trigger recalculation of the compensation vector if the set threshold is exceeded.

[0043] In summary, the present application has at least the following effects:

[0044] This system realizes the deep fusion of tool vibration, plate texture and environmental data through the spatio-temporal attention mechanism of the multi-modal perception module, generates a joint feature matrix with high information density, and significantly improves the machining state characterization ability; the process parameter generation module analyzes the topological constraint relationship of non-standard customization requirements based on the depth graph neural network, and generates safe and efficient initial parameters in combination with the mixed integer programming algorithm to adapt to the efficiency improvement; the dynamic error compensation module predicts the deformation error through the digital twin-driven bidirectional GRU network and generates a compensation vector in advance to improve the machining accuracy; the cooperative control module uses Monte Carlo tree search and temporal convolutional network to jointly optimize the path parameters, realizes the multi-objective balance of improving the plate utilization rate, prolonging the tool life and reducing the energy consumption, and through the digital twin verification and closed-loop regulation mechanism, ensures the machining safety and real-time performance, forms a complete intelligent control closed-loop from perception, decision-making to execution, and promotes the transformation of the furniture manufacturing industry to flexibility, high precision and sustainable development.

[0045] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0046] The present invention is described with reference to the flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0047] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means embodying the functionality specified in the flowchart Figure 1 a flowchart or multiple flowcharts and / or block Figure 1 a block or multiple blocks.

[0048] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functionality specified in the flowchart Figure 1 a flowchart or multiple flowcharts and / or block Figure 1 a block or multiple blocks.

[0049] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made by those skilled in the art once they learn the basic inventive concept. Therefore, the appended claims are intended to be construed to cover the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0050] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A furniture processing control system based on artificial intelligence, characterized in that: It includes the following modules: multimodal perception module, process parameter generation module, dynamic error compensation module, and collaborative control module; The multimodal perception module is used to collect multi-source heterogeneous data in real time through a distributed sensor array, including tool vibration spectrum, plate texture characteristics and environmental temperature and humidity data, and fuse and encode the multi-source heterogeneous data through a spatiotemporal attention mechanism to generate a joint feature matrix; The process parameter generation module is used to deconstruct the topological constraint relationship of non-standard customization requirements through a deep graph neural network based on a joint feature matrix, and generate an initial processing parameter set of cutting force threshold and path smoothness in combination with a parameter decoupling algorithm; The dynamic error compensation module is used to drive the digital twin model to perform virtual processing according to the initial processing parameter set, synchronously fuse the three-dimensional laser point cloud and infrared thermal imaging data, predict the deformation error of the future processing node through the gated recurrent unit of the recurrent neural network, and generate an error compensation vector; The collaborative control module is used to receive error compensation vectors and real-time working condition data, generate candidate paths through Monte Carlo tree search, evaluate the comprehensive cost function of the path, extract vibration spectrum features through a temporal convolutional network, generate corrections for tool feed speed and cutting depth, and dynamically control tool path parameters.

2. The artificial intelligence-based furniture processing control system according to claim 1, characterized in that: The specific process of fusing and encoding multi-source heterogeneous data through the spatiotemporal attention mechanism and generating a joint feature matrix is ​​as follows: The three-dimensional vibration signal of the tool is collected through distributed sensors, and the time-frequency characteristics of the tool vibration spectrum are extracted through short-time Fourier transform. The surface image of the plate is captured synchronously through an industrial camera array, and the texture features of the plate are extracted through a multi-scale convolutional neural network. The ambient temperature and humidity data collected by the temperature and humidity sensor are integrated; After the multi-source heterogeneous data are aligned in time and space, the correlation between the vibration time series features and the texture space features is dynamically modeled through the multi-head attention mechanism, the cross-modal similarity is calculated through the dot product and the weight matrix is ​​generated, and the feature fusion weight is adjusted in combination with the environmental parameters; The fused features are compressed and reduced in dimension through the autoencoder, and then the local spatiotemporal patterns are extracted through the spatiotemporal convolutional network to generate a joint feature matrix.

3. The artificial intelligence-based furniture processing control system according to claim 2, characterized in that: The specific process of deconstructing the topological constraint relationship of non-standard customization requirements through deep graph neural network is as follows: Convert the furniture CAD drawings into graph structure data, where nodes represent furniture components, including coordinates, curvature and connection types, and edges represent the connection relationship between components; The spatial features of adjacent nodes are iteratively aggregated through a multi-layer graph convolutional network, the nonlinear expression ability is enhanced through the activation function, and the topological constraint matrix is ​​output to clarify the component processing sequence and spatial dependency.

4. The artificial intelligence-based furniture processing control system according to claim 3 is characterized in that: The specific process of generating the initial machining parameter set of cutting force threshold and path smoothness by combining the parameter decoupling algorithm is as follows: The topological constraint matrix and material hardness parameters are used as input to construct a mixed integer programming model, with minimizing the cutting force deviation as the objective function, and constraining the path smoothness to meet the machining requirements; The objective function is solved by the branch and bound algorithm to generate an initial parameter set including the cutting force threshold, path curvature radius and feed speed.

5. The artificial intelligence-based furniture processing control system according to claim 4, characterized in that: The specific process of driving the digital twin model for virtual processing based on the initial processing parameter set and synchronously fusing the 3D laser point cloud and infrared thermal imaging data is as follows: Load the initial machining parameters in the digital twin environment and drive the virtual tool to execute the planned path; High-frame-rate three-dimensional point cloud data is collected synchronously through the lidar and aligned in time and space with the tool posture. At the same time, the temperature field distribution map of the infrared thermal imager is obtained to extract the thermal gradient characteristics of the tool-workpiece contact area, providing multi-modal input for deformation prediction.

6. The artificial intelligence-based furniture processing control system according to claim 5, characterized in that: The specific process of predicting the deformation error of future processing nodes through the gated recurrent unit of the recurrent neural network and generating the error compensation vector is as follows: The three-dimensional point cloud data, thermal gradient features and vibration signals are constructed into a spatiotemporal sequence and input into a bidirectional gated recurrent unit network; The historical error propagation law is captured through the forward gated recurrent unit network, and the future trend is predicted through the reverse gated recurrent unit network. After fusing the bidirectional hidden states, the deformation prediction value of the future processing node is output to generate an error compensation vector including three-dimensional displacement compensation and cutting force adjustment.

7. The artificial intelligence-based furniture processing control system according to claim 6, characterized in that: The specific process of generating candidate paths through Monte Carlo tree search and evaluating the comprehensive cost function of the path is as follows: The candidate path set is expanded with the current tool position as the root node, and a comprehensive cost function is defined, including the predicted deformation error, material utilization rate and tool wear rate, and the weight of each sub-item is dynamically adjusted according to the hardness of the plate; The Pareto optimal path is screened through the simulated annealing algorithm to balance the contradictory relationship between processing accuracy and resource loss.

8. The artificial intelligence-based furniture processing control system according to claim 7, characterized in that: The specific process of extracting vibration spectrum features through a time series convolutional network and generating correction values ​​for tool feed speed and cutting depth is as follows: The vibration signal of the selected path is decomposed by wavelet packet, and the multi-band energy features are extracted as time series input; The temporal correlation between vibration characteristics and machining parameters is modeled by temporal convolutional network, the real-time correction amount of tool feed speed and cutting depth is predicted, and dynamic adjustment instructions are generated.

9. The artificial intelligence-based furniture processing control system according to claim 8, characterized in that: The specific process of dynamically adjusting tool path parameters is as follows: Verify the physical feasibility of control instructions in the digital twin system, including tool stiffness simulation, material fracture toughness detection and motion interference analysis; After verification, the instructions are sent to the CNC system for execution through the industrial communication protocol, and the actual processing error is monitored in real time. If it exceeds the set threshold, the compensation vector recalculation is triggered, and the process parameter set is updated to form a closed-loop control.

Images