Intelligent door and window surface treatment regulation and control system and method based on multi-sensor fusion

Abstract

The invention discloses a door and window surface treatment intelligent regulation and control system and method based on multi-sensor fusion. According to the method, through multi-dimensional cooperation of the self-learning optimization center, breakthrough improvement of the process quality and essential optimization of the production efficiency are achieved. According to the system, the high-precision simulation capability of the digital twins and a dynamic decision-making mechanism of reinforcement learning are deeply fused, so that fine changes of material characteristics, environmental conditions and equipment states can be sensed in real time in the door and window surface treatment process, and optimal process parameters are automatically generated. The traditional hysteresis quality which depends on manual experience for adjustment is thoroughly eliminated, and when the system treats complex working conditions such as special-shaped curved surfaces and composite base materials, not only can the uniformity of the nanoscale coating be maintained, but also the energy supply strategy can be dynamically optimized, and the energy consumption and the material loss can be obviously reduced. The cross-process global optimization capability ensures that the links of pretreatment, spraying, curing and the like form cooperative gain instead of isolated operation, and the debugging period of the complex process is greatly shortened through the full-process intelligence.

Description

Technical Field

[0001] The present invention belongs to the technical field of surface treatment of door and window production, and specifically relates to an intelligent control system and method for door and window surface treatment based on multi-sensor fusion. Background Art

[0002] The door and window surface treatment control system is an automated control system that integrates multiple advanced technologies to optimize the door and window surface treatment process. This system typically consists of a central controller, a sensor network, actuators, and a human-machine interface. As the core of the system, the central controller receives and processes data from various sensors and issues commands to the actuators based on preset programs and parameters, achieving precise control of the door and window surface treatment process. The sensor network monitors key parameters such as temperature, humidity, coating thickness, and gloss on the door and window surfaces in real time and feeds this data back to the central controller. Based on the controller's instructions, the actuators adjust operating parameters for spraying, baking, cooling, and other process steps to ensure the quality and efficiency of the door and window surface treatment. The human-machine interface provides an intuitive operating platform for operators to easily set parameters, monitor the process, and diagnose faults. The door and window surface treatment control system can effectively improve the quality and stability of door and window surface treatment, reduce production costs, and minimize environmental pollution. It is a key technical support for the transformation and upgrading of the door and window manufacturing industry towards intelligent and green manufacturing.

[0003] However, in existing technologies, traditional actuators rely on fixed programs to control mechanical motion trajectories, making it difficult to respond to complex surface morphology changes in real time. This results in coating uniformity being restricted by manual experience. At the same time, conventional decision-making systems lack autonomous optimization capabilities and require repeated trial and error when faced with new materials or sudden environmental interference. The debugging cycle is long and effective knowledge cannot be accumulated, which restricts process innovation and cross-scenario adaptability. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent control system and method for door and window surface treatment based on multi-sensor fusion in order to solve the above-mentioned problems.

[0005] The technical solution adopted by the present invention is as follows: an intelligent control system for door and window surface treatment based on multi-sensor fusion, the system comprising: a multimodal perception module, a heterogeneous data fusion module, a dynamic control execution module, a self-learning optimization hub module, a human-computer interaction module and a cloud management module;

[0006] The self-learning optimization central module is internally configured with: a digital twin submodule, a reinforcement learning decision submodule, a process knowledge base submodule, and an abnormality prediction submodule;

[0007] The sensor data output terminal of the multimodal perception module is directly connected to the original signal input terminal of the heterogeneous data fusion module, and the collected coating thickness and surface stress physical quantities are transmitted to the fusion module;

[0008] The decision output of the heterogeneous data fusion module is connected to the data input of the self-learning optimization hub module to provide an environmental state assessment result that has been time-space aligned and confidence-weighted;

[0009] The control command output of the self-learning optimization central module is connected to the control signal input of the dynamic control execution module to drive the robot arm trajectory and energy unit parameter adjustment. At the same time, the real-time status feedback end of the execution module is transmitted back to the closed-loop optimization interface of the central module to form a decision-making-execution data closed loop;

[0010] The data interaction end of the human-machine interaction module is bidirectionally connected to the state output end and the instruction input end of the self-learning center to realize process parameter visualization and manual intervention instruction injection;

[0011] The data aggregation end of the cloud management module receives the historical data stream of the multimodal perception module, the feature data set of the heterogeneous fusion module and the optimization model parameters of the self-learning center, and synchronizes the global process strategy and safety rules to each module through the configuration of the downstream end, forming a cross-level data collaboration network.

[0012] In a preferred embodiment, the multimodal sensing module works in tandem with a laser ranging sensor and a micro-piezoelectric film. The laser ranging unit uses the time-of-flight principle to scan the workpiece surface at a frequency of thousands of times per second to generate three-dimensional point cloud data. The micro-piezoelectric film is embedded in the end of the spraying equipment to sense the microscopic stress fluctuations during coating deposition in real time. After the two types of sensor data are aligned by timestamp, a complete physical model of the surface morphology is jointly constructed. When it is detected that the local thickness deviation exceeds the preset threshold, the compensation mechanism is immediately triggered to ensure spraying uniformity.

[0013] In a preferred embodiment, the heterogeneous data fusion module adopts a three-level processing architecture. The first layer performs adaptive noise reduction on the original signal to eliminate environmental electromagnetic interference and mechanical vibration noise; the middle layer extracts the spatiotemporal correlation characteristics of different sensors, performs time-domain convolution on the surface reflectivity curve of the optical sensor and the stress change spectrum of the mechanical sensor, and captures the coupling relationship between the two during the coating curing stage; the top layer establishes a decision model based on probabilistic reasoning, integrates the confidence weights of multi-source data, and when there is a contradiction between the temperature and humidity sensor and the infrared thermal imaging data, the system automatically calls the historical working condition library to resolve the conflict and generate a final credible environmental status assessment.

[0014] In a preferred embodiment, the dynamic control execution module is composed of four parts: a multi-axis linkage robotic arm system, an intelligent spraying device, an energy control unit and a motion control center. Each device works together to achieve precise surface treatment operations.

[0015] In a preferred embodiment, the digital twin submodule realizes deep interaction between the physical world and virtual space by constructing a simulation environment with virtual-real mapping; the system integrates multi-physics coupling models of fluid dynamics, thermodynamics, and mechanical stress fields, synchronizes sensor data on the production line in real time, and forms a high-precision dynamic simulation mirror;

[0016] The dynamic compensation formula is:

[0017]

[0018] in:

[0019] ΔP represents the dynamic compensation of model parameters, which is used to correct the deviation between the simulation model and actual production;

[0020] \η represents the adaptive learning rate, which is automatically adjusted according to the historical error fluctuation amplitude;

[0021] E sim / E real Represents the energy dissipation characteristic values of the simulation environment and the real production line;

[0022] x i represents the i-th type of parameter that affects the coating quality;

[0023] tSystem continuous operation time, used to enhance long-term stability;

[0024] τ represents the material characteristic time constant.

[0025] In a preferred embodiment, the reinforcement learning decision submodule is internally provided with:

[0026] The upper-level controller breaks down the complete surface treatment process into discrete stages: pretreatment, primer spraying, and curing. Each stage is equipped with an independent strategy network. The stage division is based on substrate status parameters and process progress fed back by real-time sensors. The upper layer uses a dynamic programming algorithm to determine the stage switching timing and global optimization goals.

[0027] The lower-level execution network is based on the proximal policy optimization framework. Each sub-strategy network contains a dual-channel value assessment module, which calculates the process quality benefit and energy consumption cost respectively. The strategy update adopts an offline-online hybrid training mechanism. The offline stage uses millions of virtual samples generated by digital twins for pre-training, and the online stage fine-tunes the network weights through incremental data from the actual production line. The safety protection mechanism is embedded in the entire decision-making process, including dual guarantees of hard constraints on parameter adjustment amplitude and soft constraints on real-time risk probability. The former limits the fluctuation range of key parameters such as spraying pressure and curing temperature, and the latter predicts the potential risks of the decision chain through Monte Carlo simulation. A shared experience pool is set up between each strategy network to realize parameter migration and knowledge reuse across process stages.

[0028] In a preferred embodiment, the bottom-layer material database of the process knowledge base submodule stores the molecular structure parameters and thermal expansion coefficients of aluminum alloys and glass substrates, the middle-layer process rule library encodes the coating viscosity-spraying distance conversion table and the temperature-curing rate correlation matrix physical relationship, and the top-layer case library includes typical workpiece processing records and abnormal handling plans; the knowledge graph uses an attribute graph model to realize data interconnection, the nodes represent materials, equipment, and environmental entities, and the edge relationships define the causal logic and constraints between process parameters.

[0029] In a preferred embodiment, the anomaly prediction submodule constructs a real-time risk prediction model based on deep learning by integrating multi-dimensional time series data from optical, mechanical, and environmental sensors. The system simultaneously analyzes micro-texture changes on the coating surface, substrate stress fluctuation curves, and environmental temperature and humidity evolution trends, and uses cross-modal feature extraction technology to identify anomaly patterns. For 12 common defects such as orange peel, pinholes, and sagging in coatings, the system has a built-in feature map library of tens of thousands of historical defect samples, and uses a time series convolutional network to capture the spatiotemporal correlation of early anomaly signals.

[0030] The multimodal risk index formula is:

[0031]

[0032] in:

[0033] R represents the real-time risk index, and a larger value indicates a higher probability of abnormality;

[0034] αk represents the dynamic weight coefficient (0-1) of the kth type of sensor (optical / mechanical / environmental), which is automatically adjusted according to the environmental stability;

[0035] F k (t) represents the normalized feature vector extracted by the k-th sensor at time t;

[0036] μk / δk represents the characteristic mean and standard deviation of the kth sensor under normal working conditions;

[0037] λ represents the time decay factor (default value is 0.05 / s), which strengthens the decision-making influence of recent data;

[0038] t0 represents the starting timestamp of the current processing stage.

[0039] In a preferred embodiment, the human-computer interaction module superimposes virtual process parameters and the real production line space through a head-mounted display device; the operator can call up the coating thickness heat map at any location by swiping with a gesture, and view the real-time stress distribution curve by clicking the device node with a fingertip;

[0040] The cloud management module cloud architecture adopts a hybrid mode of collaboration between edge computing nodes and central cloud. The edge nodes are deployed in the control cabinets of each production line, responsible for real-time processing of sensor raw data and executing low-latency control instructions; the central cloud aggregates global production information and stores ten years of process history records through a time series database.

[0041] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0042] 1. In the present invention, a breakthrough improvement in process quality and an essential optimization of production efficiency are achieved through the multi-dimensional collaboration of the self-learning optimization center. The system deeply integrates the high-precision simulation capabilities of digital twins with the dynamic decision-making mechanism of reinforcement learning, so that the surface treatment process of doors and windows can perceive subtle changes in material properties, environmental conditions and equipment status in real time, and automatically generate optimal process parameters. The lag of traditional reliance on manual experience adjustment is completely eliminated. When dealing with complex working conditions such as special-shaped curved surfaces and composite substrates, the system can not only maintain nano-level coating uniformity, but also dynamically optimize energy supply strategies, significantly reducing energy consumption and material loss. The global optimization capability across processes ensures that pretreatment, spraying, curing and other links form synergistic gains rather than operating in isolation. This full-process intelligence greatly compresses the debugging cycle of complex processes.

[0043] 2. In the present invention, the precision and adaptive optimization of the surface treatment process of doors and windows are achieved through the deep coordination of dynamic control execution modules and multi-sensor data. The high-precision robotic arm, combined with the real-time trajectory planning algorithm, can dynamically adjust the motion path according to the detection data of coating thickness and surface morphology, automatically increase the density of trajectory points in complex curved areas, and ensure the uniformity of spray coverage; the intelligent nozzle adjusts the atomized particle size and spray angle in milliseconds according to the material properties and changes in ambient temperature and humidity, effectively avoiding defects such as sagging and orange peel. The precise synchronization of the energy control unit and the mechanical movement enables the UV curing intensity and the infrared heating gradient to be intelligently matched with the spraying progress, which not only ensures the stability of the physical and chemical properties of the coating, but also greatly reduces ineffective energy consumption. This closed-loop linkage between the execution end, the perception end, and the decision-making end enables the system to always maintain process consistency when facing different substrates and changing working conditions, significantly improving the yield rate and equipment utilization rate, while reducing the frequency of manual intervention, and promoting the surface treatment process to autonomous optimization and upgrading of the entire process. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a block diagram of the overall system of the present invention;

[0045] Figure 2 This is a system block diagram of the self-learning optimization central module in the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0047] Reference Figure 1-2 ,

[0048] An intelligent control system for door and window surface treatment based on multi-sensor fusion, the system includes: a multimodal perception module, a heterogeneous data fusion module, a dynamic control execution module, a self-learning optimization center module, a human-computer interaction module and a cloud management module;

[0049] The internal configuration of the self-learning optimization hub module includes: digital twin submodule, reinforcement learning decision submodule, process knowledge base submodule and abnormality prediction submodule;

[0050] The sensor data output of the multimodal perception module is directly connected to the original signal input of the heterogeneous data fusion module, and the collected physical quantities such as coating thickness and surface stress are transmitted to the fusion module;

[0051] The decision output of the heterogeneous data fusion module is connected to the data input of the self-learning optimization hub module, providing an environmental status assessment result that has been time-space aligned and confidence-weighted.

[0052] The control command output of the self-learning optimization central module is connected to the control signal input of the dynamic control execution module to drive the robot arm trajectory and energy unit parameter adjustment. At the same time, the real-time status feedback of the execution module is transmitted back to the closed-loop optimization interface of the central module, forming a decision-making-execution data closed loop.

[0053] The data interaction end of the human-machine interaction module is bidirectionally connected to the status output end and command input end of the self-learning center to realize process parameter visualization and manual intervention command injection;

[0054] The data aggregation end of the cloud management module receives the historical data stream of the multimodal perception module, the feature data set of the heterogeneous fusion module and the optimization model parameters of the self-learning center, and synchronizes the global process strategy and safety rules to each module through the configuration of the downstream end, forming a cross-level data collaboration network.

[0055] The multimodal sensing module utilizes a laser ranging sensor and a micro-piezoelectric film to achieve nanoscale monitoring of coating thickness and dynamic capture of surface stress. The laser ranging unit uses the time-of-flight principle to scan the workpiece surface thousands of times per second, generating 3D point cloud data. The micro-piezoelectric film, embedded in the end of the spraying device, senses microscopic stress fluctuations during coating deposition in real time. The data from these two sensors is aligned using timestamps to construct a complete physical model of the surface morphology. When local thickness deviations exceeding a preset threshold are detected, a compensation mechanism is immediately triggered to ensure spray uniformity.

[0056] The heterogeneous data fusion module adopts a three-level processing architecture. The first layer performs adaptive noise reduction on the original signal to eliminate environmental electromagnetic interference and mechanical vibration noise; the middle layer extracts the spatiotemporal correlation characteristics of different sensors, such as performing time-domain convolution on the surface reflectivity curve of the optical sensor and the stress change spectrum of the mechanical sensor to capture the coupling relationship between the two during the coating curing stage; the top layer establishes a decision model based on probabilistic reasoning, integrating the confidence weights of multi-source data. When there is a contradiction between the temperature and humidity sensor and the infrared thermal imaging data, the system automatically calls the historical working condition library to resolve the conflict and generate a final credible environmental status assessment.

[0057] The dynamic control execution module converts coating thickness distribution data into spatial path instructions. The intelligent nozzle at the end of the robotic arm features a built-in micro-pressure regulating valve, dynamically adjusting the atomized particle size based on real-time surface roughness data. The spray path utilizes a fractal optimization strategy, automatically increasing trajectory point density in areas with sudden changes in curvature and extending the linear motion cycle in flat areas. Furthermore, the power output of the UV curing unit is matched in real time to the robotic arm's movement speed, ensuring consistent curing energy density at different locations.

[0058] It consists of four parts: a multi-axis linkage robotic arm system, an intelligent spraying device, an energy control unit and a motion control center. Each device works together to achieve precise surface treatment operations.

[0059] The six-degree-of-freedom robotic arm features a carbon fiber composite frame and six high-precision servo joints, each with a built-in torque sensor and harmonic reducer. The end-effector interface allows for quick replacement of process accessories such as spray heads and curing lamps. An inertial navigation module integrated into the arm base provides real-time compensation for position deviations caused by floor vibration. Each joint achieves a motion accuracy of ±0.02 mm, with a repeatability error of no more than ±5 microns.

[0060] The piezoelectric ceramic atomizing nozzle assembly consists of a ring-shaped piezoelectric ceramic oscillator and a titanium alloy nozzle. The oscillator generates a controllable atomized particle size of 5-50 microns at a high frequency of 20kHz. A spiral flow channel within the nozzle allows centrifugal force to achieve infinitely adjustable atomization angles from 30° to 90°. The accompanying two-component dynamic mixer utilizes a static mixing tube design, achieving uniform mixing of the base material and curing agent within 0.3 seconds.

[0061] The UV-IR hybrid energy unit consists of a matrix UV LED array and a carbon fiber infrared heating plate. The UV unit contains 256 independently controllable LED modules, each with millisecond-level power adjustment between 0-50W and wavelengths covering the 365nm-405nm range. The infrared heating plate has 96 built-in temperature control zones, utilizing an aluminum nitride ceramic substrate to achieve gradient temperature control up to 300°C, with a thermal response time of less than 1.5 seconds.

[0062] The motion control and feedback system is equipped with an industrial-grade motion control card and utilizes the EtherCAT bus to synchronously drive the motors in each joint of the robotic arm. A real-time trajectory planner updates motion commands 2000 times per second, dynamically correcting the theoretical trajectory based on actual end-point position data from a laser tracker. The safety protection system includes a triple safeguard mechanism: overload current monitoring, collision torque detection, and an emergency braking circuit. In the event of an abnormal condition, actuator power can be cut off within 50 milliseconds.

[0063] The digital twin submodule realizes the deep interaction between the physical world and the virtual space by constructing a simulation environment of virtual-reality mapping. The system integrates multi-physics field coupling models such as fluid dynamics, thermodynamics and mechanical stress fields, and synchronizes sensor data on the production line (including temperature, coating thickness, substrate stress, etc.) in real time to form a high-precision dynamic simulation mirror. Based on the physical and chemical parameters of hundreds of substrates and coatings stored in the material property library, the system can simulate microscopic processes such as coating flow and solidification deformation, and automatically adjust the simulation accuracy with the help of dynamic grid technology: enable micron-level grids to capture details at the spraying edge or curved surface area, and switch to millimeter-level grids in flat areas to improve computing efficiency. Through the parameter inversion algorithm, the system compares the real-time collected processing results with the simulation prediction values, automatically corrects the model deviation, and ensures that the error rate between the virtual environment and the real production line is less than 0.5%, significantly reducing the cost of trial and error.

[0064] The dynamic compensation formula is:

[0065]

[0066] in:

[0067] ΔP represents the dynamic compensation of model parameters, which is used to correct the deviation between the simulation model and actual production;

[0068] \η represents the adaptive learning rate, which is automatically adjusted according to the historical error fluctuation range (range 0.01~0.2); E sim / E real Represents the energy dissipation characteristic values of the simulation environment and the real production line;

[0069] x i Indicates the i-th type of parameters that affect the coating quality (such as spraying pressure, curing temperature, etc.);

[0070] tSystem continuous operation time, used to enhance long-term stability;

[0071] τ represents the material characteristic time constant.

[0072] The internal settings of the reinforcement learning decision submodule are:

[0073] The upper-level controller breaks down the complete surface treatment process into discrete stages, such as pretreatment, primer spraying, and curing. Each stage is equipped with an independent strategy network. Stage division is based on substrate status parameters and process progress fed back by real-time sensors. The upper layer uses a dynamic programming algorithm to determine stage switching timing and global optimization objectives.

[0074] The lower-level execution network is based on a proximal policy optimization framework. Each sub-policy network includes a dual-channel value assessment module, which calculates process quality benefits and energy costs, respectively. Policy updates utilize a hybrid offline-online training mechanism. The offline phase utilizes millions of virtual samples generated by the digital twin for pre-training, while the online phase fine-tunes network weights using incremental data from actual production lines. Safety protection mechanisms are embedded throughout the decision-making process, encompassing both hard constraints on parameter adjustment ranges and soft constraints on real-time risk probabilities. The former limits the fluctuation range of key parameters such as spray pressure and curing temperature, while the latter predicts potential risks in the decision chain through Monte Carlo simulation. A shared experience pool is established between each policy network to enable parameter migration and knowledge reuse across process stages.

[0075] The process knowledge base submodule's underlying material database stores molecular structure parameters and thermal expansion coefficients for substrates like aluminum alloy and glass. The middle-layer process rule library encodes physical relationships, such as coating viscosity-spraying distance conversion tables and temperature-curing rate correlation matrices. The top-level case library contains typical workpiece processing records and exception handling plans. The knowledge graph utilizes an attribute graph model to interconnect data. Nodes represent entities such as materials, equipment, and environments, while edges define the causal logic and constraints between process parameters. The rule generation engine incorporates two automated channels: induction and deduction. The induction channel extracts empirical formulas through statistical analysis of historical data, such as deriving a regulation rule requiring a 2% infrared power compensation for every 5% increase in humidity. The deduction channel translates the action sequences output by the reinforcement learning policy network into interpretable if-then control logic. The dynamic update mechanism leverages graph neural network technology to automatically identify implicit parameter associations in newly collected data. For example, discovering a nonlinear relationship between surface roughness and coating adhesion under specific lighting conditions automatically generates new process constraint entries. The knowledge distillation module regularly transforms the policy network's abstract decision patterns into standard process procedures, forming a collaborative human-machine decision-making loop.

[0076] The anomaly prediction submodule integrates multi-dimensional time-series data from optical, mechanical, and environmental sensors to construct a real-time risk prediction model based on deep learning. The system simultaneously analyzes microtexture changes on the coating surface, substrate stress fluctuations, and environmental temperature and humidity trends, identifying anomaly patterns using cross-modal feature extraction techniques. The system includes a built-in feature map library of tens of thousands of historical defect samples for 12 common defects, including orange peel, pinholes, and sagging. Using a time-series convolutional network, the system captures the spatiotemporal correlations of early anomaly signals. The system employs a progressive three-level response mechanism: dynamic compensation strategies are initiated when minor parameter deviations are detected; moderate risks trigger a process pause and self-check; and severe anomalies are predicted, with power cutoff and equipment lockout. To achieve millisecond-level response, the model is hardware-accelerated using edge computing units and, combined with an attention mechanism, dynamically assigns decision weights to different sensors, ensuring a latency of less than 200 milliseconds from data acquisition to control execution.

[0077] The multimodal risk index formula is:

[0078]

[0079] in:

[0080] R represents the real-time risk index, and a larger value indicates a higher probability of abnormality;

[0081] αk represents the dynamic weight coefficient (0-1) of the kth type of sensor (optical / mechanical / environmental), which is automatically adjusted according to the environmental stability;

[0082] F k (t) represents the normalized feature vector extracted by the k-th sensor at time t;

[0083] μk / δk represents the characteristic mean and standard deviation of the kth sensor under normal working conditions;

[0084] λ represents the time decay factor (default value is 0.05 / s), which strengthens the decision-making influence of recent data;

[0085] t0 represents the starting timestamp of the current processing stage.

[0086] The human-computer interaction module overlays virtual process parameters with the real production line space through a head-mounted display. Operators can use gestures to access coating thickness heat maps at any location and tap equipment nodes to view real-time stress distribution curves.

[0087] The cloud management module's cloud architecture adopts a hybrid mode of collaboration between edge computing nodes and central cloud. Edge nodes are deployed in the control cabinets of each production line, responsible for real-time processing of sensor raw data and executing low-latency control instructions; the central cloud aggregates global production information and stores ten years of process history records through a time series database.

[0088] A method for intelligent control of door and window surface treatment based on multi-sensor fusion, the method runs the intelligent control system for door and window surface treatment based on multi-sensor fusion of the above embodiment.

[0089] From the above we can know:

[0090] In the present invention, a breakthrough improvement in process quality and an essential optimization of production efficiency are achieved through the multi-dimensional collaboration of the self-learning optimization center. The system deeply integrates the high-precision simulation capabilities of digital twins with the dynamic decision-making mechanism of reinforcement learning, so that the surface treatment process of doors and windows can perceive subtle changes in material properties, environmental conditions and equipment status in real time, and automatically generate optimal process parameters. The lag of traditional reliance on manual experience adjustment is completely eliminated. When dealing with complex working conditions such as special-shaped curved surfaces and composite substrates, the system can not only maintain nano-level coating uniformity, but also dynamically optimize energy supply strategies, significantly reducing energy consumption and material loss. The global optimization capability across processes ensures that pretreatment, spraying, curing and other links form synergistic gains, rather than operating in isolation. This full-process intelligence greatly compresses the debugging cycle of complex processes.

[0091] In the present invention, the precision and adaptive optimization of the surface treatment process of doors and windows are achieved through the deep coordination of dynamic control execution modules and multi-sensor data. The high-precision robotic arm, combined with the real-time trajectory planning algorithm, can dynamically adjust the motion path according to the detection data of coating thickness and surface morphology, automatically increase the density of trajectory points in complex curved areas, and ensure the uniformity of spray coverage; the intelligent nozzle adjusts the atomized particle size and spray angle in milliseconds according to the material properties and environmental temperature and humidity changes, effectively avoiding defects such as sagging and orange peel. The precise synchronization of the energy control unit and the mechanical movement enables the UV curing intensity and the infrared heating gradient to be intelligently matched with the spraying progress, which not only ensures the stability of the physical and chemical properties of the coating, but also greatly reduces ineffective energy consumption. This closed-loop linkage between the execution end, the perception end, and the decision-making end enables the system to always maintain process consistency when facing different substrates and changing working conditions, significantly improving the yield rate and equipment utilization rate, while reducing the frequency of manual intervention, and promoting the surface treatment process to autonomous optimization and upgrading of the entire process.

[0092] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements that are inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0093] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An intelligent control system for door and window surface treatment based on multi-sensor fusion, characterized by: The system includes: a multimodal perception module, a heterogeneous data fusion module, a dynamic control execution module, a self-learning optimization center module, a human-computer interaction module and a cloud management module; The self-learning optimization central module is internally configured with: a digital twin submodule, a reinforcement learning decision submodule, a process knowledge base submodule, and an abnormality prediction submodule; The sensor data output terminal of the multimodal perception module is directly connected to the original signal input terminal of the heterogeneous data fusion module, and the collected coating thickness and surface stress physical quantities are transmitted to the fusion module; The decision output of the heterogeneous data fusion module is connected to the data input of the self-learning optimization hub module to provide an environmental state assessment result that has been time-space aligned and confidence-weighted; The control command output of the self-learning optimization central module is connected to the control signal input of the dynamic control execution module to drive the robot arm trajectory and energy unit parameter adjustment. At the same time, the real-time status feedback end of the execution module is transmitted back to the closed-loop optimization interface of the central module to form a decision-making-execution data closed loop; The data interaction end of the human-machine interaction module is bidirectionally connected to the state output end and the instruction input end of the self-learning center to realize process parameter visualization and manual intervention instruction injection; The data aggregation end of the cloud management module receives the historical data stream of the multimodal perception module, the feature data set of the heterogeneous fusion module and the optimization model parameters of the self-learning center, and synchronizes the global process strategy and safety rules to each module through the configuration of the downstream end, forming a cross-level data collaboration network.

2. The intelligent control system for door and window surface treatment based on multi-sensor fusion according to claim 1, characterized in that: The multimodal sensing module works in tandem with a laser ranging sensor and a micro-piezoelectric film. The laser ranging unit uses the time-of-flight principle to scan the workpiece surface thousands of times per second to generate three-dimensional point cloud data. A micro-piezoelectric film is embedded in the end of the spraying equipment to sense the micro-stress fluctuations during coating deposition in real time. After the two types of sensor data are aligned through timestamps, a complete physical model of the surface morphology is jointly constructed. When the local thickness deviation is detected to exceed the preset threshold, the compensation mechanism is immediately triggered to ensure spraying uniformity.

3. The intelligent control system for door and window surface treatment based on multi-sensor fusion according to claim 1, characterized in that: The heterogeneous data fusion module adopts a three-level processing architecture. The first layer performs adaptive noise reduction on the original signal to eliminate environmental electromagnetic interference and mechanical vibration noise. The middle layer extracts the spatiotemporal correlation characteristics of different sensors and performs time-domain convolution on the surface reflectivity curve of the optical sensor and the stress variation spectrum of the mechanical sensor to capture the coupling relationship between the two during the coating curing stage. The top layer establishes a decision-making model based on probabilistic reasoning and integrates the confidence weights of multi-source data. When there is a contradiction between the temperature and humidity sensor and the infrared thermal imaging data, the system automatically calls the historical operating condition library to resolve the conflict and generate a final credible environmental status assessment.

4. The intelligent control system for door and window surface treatment based on multi-sensor fusion according to claim 1, characterized in that: The dynamic control execution module consists of four parts: a multi-axis linkage robotic arm system, an intelligent spraying device, an energy control unit and a motion control center. Each device works together to achieve precise surface treatment operations.

5. The intelligent control system for door and window surface treatment based on multi-sensor fusion according to claim 1, characterized in that: The digital twin submodule realizes deep interaction between the physical world and the virtual space by constructing a simulation environment with virtual-reality mapping; The system integrates multi-physics coupling models of fluid dynamics, thermodynamics and mechanical stress fields, synchronizes sensor data on the production line in real time, and forms a high-precision dynamic simulation mirror; The dynamic compensation formula is: in: ΔP represents the dynamic compensation of model parameters, which is used to correct the deviation between the simulation model and actual production; \η represents the adaptive learning rate, which is automatically adjusted according to the historical error fluctuation range; E sim / E real Represents the energy dissipation characteristic values of the simulation environment and the real production line; x i represents the i-th type of parameters that affect the coating quality; tSystem continuous operation time, used to enhance long-term stability; τ represents the material characteristic time constant.

6. The intelligent control system for door and window surface treatment based on multi-sensor fusion according to claim 1, characterized in that: The internal configuration of the reinforcement learning decision submodule includes: The upper-level controller breaks down the complete surface treatment process into discrete stages: pretreatment, primer spraying, and curing. Each stage is equipped with an independent strategy network. The stage division is based on substrate status parameters and process progress fed back by real-time sensors. The upper layer uses a dynamic programming algorithm to determine the stage switching timing and global optimization goals. The lower-level execution network is based on the proximal strategy optimization framework. Each sub-strategy network contains a dual-channel value evaluation module to calculate process quality benefits and energy consumption costs respectively. Strategy updates utilize an offline-online hybrid training mechanism. In the offline phase, pre-training is performed using millions of virtual samples generated by digital twins. In the online phase, network weights are fine-tuned using incremental data from actual production lines. The safety protection mechanism is embedded in the entire decision-making process, including dual guarantees of hard constraints on parameter adjustment range and soft constraints on real-time risk probability. The former limits the fluctuation range of key parameters such as spraying pressure and curing temperature, while the latter predicts the potential risks of the decision-making chain through Monte Carlo simulation; a shared experience pool is set up between each strategy network to realize parameter migration and knowledge reuse across process stages.

7. The intelligent control system for door and window surface treatment based on multi-sensor fusion according to claim 1, characterized in that: The bottom-layer material database of the process knowledge base submodule stores the molecular structure parameters and thermal expansion coefficients of aluminum alloys and glass substrates; the middle-layer process rule library encodes the coating viscosity-spraying distance conversion table and the physical relationship of the temperature-curing rate association matrix; the top-layer case library includes typical workpiece processing records and abnormal handling plans; the knowledge graph uses an attribute graph model to realize data interconnection, the nodes represent materials, equipment, and environmental entities, and the edge relationships define the causal logic and constraints between process parameters.

8. The intelligent control system for door and window surface treatment based on multi-sensor fusion according to claim 1, characterized in that: The anomaly prediction submodule builds a real-time risk prediction model based on deep learning by integrating multi-dimensional time series data from optical, mechanical, and environmental sensors. The system simultaneously analyzes micro-texture changes on the coating surface, substrate stress fluctuation curves, and environmental temperature and humidity evolution trends, using cross-modal feature extraction technology to identify anomaly patterns. For 12 common defects such as orange peel, pinholes, and sagging in coatings, the system has a built-in feature map library of tens of thousands of historical defect samples, capturing the spatiotemporal correlation of early anomaly signals through a time series convolutional network. The multimodal risk index formula is: in: R represents the real-time risk index, and a larger value indicates a higher probability of abnormality; αk represents the dynamic weight coefficient (0-1) of the kth sensor (optical / mechanical / environmental), which is automatically adjusted according to the environmental stability; F k (t) represents the normalized feature vector extracted by the k-th sensor at time t; μk / δk represents the characteristic mean and standard deviation of the kth sensor under normal working conditions; λ represents the time decay factor (default value is 0.05 / s), which strengthens the decision-making influence of recent data; t0 represents the starting timestamp of the current processing stage.

9. The intelligent control system for door and window surface treatment based on multi-sensor fusion according to claim 1, characterized in that: The human-computer interaction module superimposes virtual process parameters and the real production line space through a head-mounted display device; the operator can use gestures to call up the coating thickness heat map at any location, and click on the equipment node with a fingertip to view the real-time stress distribution curve; The cloud management module's cloud architecture adopts a hybrid mode of collaboration between edge computing nodes and central cloud. The edge nodes are deployed in the control cabinets of each production line, responsible for real-time processing of sensor raw data and executing low-latency control instructions; the central cloud aggregates global production information and stores ten years of process history records through a time series database.

10. An intelligent control method for door and window surface treatment based on multi-sensor fusion, characterized by: The method operates the intelligent control system for door and window surface treatment based on multi-sensor fusion as described in any one of claims 1 to 9.

Images