A flexible fabric multi-layer spreading positioning method based on neural computing

Abstract

The present application relates to the technical field of flexible fabric automation processing, and particularly relates to a flexible fabric multi-layer spreading positioning method based on neural computing. In the present application, the physical characteristic parameters of the fabric are obtained and the initial parameters of the model are configured in the pre-preparation stage, the multi-dimensional perception module is used to collect data such as contact pressure and texture deformation in the layered spreading and multi-layer stacking stage, the micro-dislocation and positioning deviation are predicted in combination with the layered coupling neural computing model, the adaptive execution adjustment module is used for layered independent compensation and tension adjustment, and finally the model parameters are updated through iterative optimization to form a closed-loop control. The present application effectively solves the problem that in the prior art, the micro-dislocation is accumulated and the overall positioning accuracy is low due to interlayer friction and nonlinear slip in multi-layer fabric spreading, can dynamically monitor and accurately control the interlayer action and positioning deviation, realizes layered accurate control of each fabric layer, improves the multi-layer flexible fabric spreading positioning accuracy, and meets the high-precision assembly or composite processing requirements.

Description

Technical Field

[0001] This invention relates to the field of automated processing technology for flexible fabrics, specifically a method for multi-layer spreading and positioning of flexible fabrics based on neural computing. Background Technology

[0002] Automated deployment and positioning of flexible fabrics is a crucial step in textile, garment manufacturing, and soft robotics. Traditional methods typically rely on mechanical grippers or vacuum suction devices to fix and flatten single or multiple layers of fabric. These technologies are based on pre-set paths and force control, utilizing sensors to detect fabric tension and positional deviations. With the development of neural networks and computational models, some systems have begun to incorporate visual feedback and lightweight control algorithms to improve their adaptability to fabric deformation. However, due to the anisotropic, low-stiffness, and easily deformable physical properties of fabrics, their dynamic response during multi-layer stacking remains difficult to predict and control accurately.

[0003] In existing technologies, while neural computation-based control methods can achieve a rough estimate of local deformation in single-layer fabrics, when dealing with multi-layer fabrics, the coupling of interlayer contact friction and nonlinear slip effects leads to unpredictable microscopic misalignments between fabric layers. These misalignments gradually accumulate during the fabric spreading process, eventually causing a decrease in overall positioning accuracy. Especially when the number of fabric layers increases or the material differences are significant, existing models cannot accurately perceive and compensate for the implicit displacements caused by interlayer interactions, making it difficult for multi-layer fabrics to meet the requirements of high-precision assembly or composite processing. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a flexible fabric multilayer laying and positioning method based on neural computing. This method solves the problem that, compared with existing technologies, multilayer fabric laying results in the accumulation of microscopic misalignments and low overall positioning accuracy due to interlayer friction and nonlinear slippage, which cannot meet the requirements of high-precision assembly or composite processing.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for multi-layer laying and positioning of flexible fabrics based on neural computing, comprising the following steps:

[0006] S1. Pre-preparation stage: The elastic modulus, friction coefficient and anisotropy coefficient of the flexible fabric to be laid are obtained through the fabric physical property pre-sensing unit; and the equipment calibration of the multi-dimensional sensing module and the initial parameter configuration of the hierarchical coupled neural computing model are completed.

[0007] S2, Layered Spreading Stage: The spreading actuator is driven to spread the first layer of flexible fabric along a preset path; the multi-dimensional sensing module collects in real time the contact pressure distribution data, fabric surface texture deformation and slip trajectory, and infrared thermal imaging temperature distribution data of the first layer of flexible fabric; the layered coupled neural computing model outputs the positioning deviation compensation amount of the first layer of flexible fabric; the adaptive execution adjustment module adjusts the spreading actuator and the multi-node collaborative tension adjustment system according to the compensation amount to ensure the positioning accuracy of the first layer of flexible fabric;

[0008] S3. In the multi-layer stacking stage, subsequent flexible fabric layers are stacked sequentially. For each layer of fabric stacked, the multi-dimensional sensing module focuses on collecting contact interaction data between the newly added layer and the lower layer. The layered coupled neural computing model predicts the micro-dislocation amount and overall positioning deviation between the newly added layer and the already laid layer. The adaptive execution adjustment module performs independent micro-displacement compensation and tension adjustment for the newly added layer and the already laid layer, respectively, to achieve a real-time closed loop of laying, detection and compensation.

[0009] S4. In the iterative optimization stage, after all layers of flexible fabric are laid out, the overall positioning accuracy is detected, and the positioning error data is fed back to the layered coupled neural computing model and the model parameters are updated.

[0010] Furthermore, the multi-dimensional perception module includes a micro-force-vision fusion perception unit, which comprises a distributed micro-force sensor array, a high-frame-rate polarized light vision camera, and an infrared thermal imaging sensor. The distributed micro-force sensor array is deployed on the contact surface of the spreading platform and the surface of the spreading actuator to collect real-time contact pressure distribution data between the flexible fabric layers. The high-frame-rate polarized light vision camera captures the texture deformation and interlayer slippage trajectory of the flexible fabric surface and uses the characteristics of polarized light to distinguish the boundary contours of different flexible fabric layers. The infrared thermal imaging sensor is positioned at key nodes in the spreading path to assist in determining the dynamic trend of the interlayer slippage.

[0011] Furthermore, the multi-dimensional perception module further includes a fabric physical property pre-sensing and modeling unit. The fabric physical property pre-sensing and modeling unit obtains key physical parameters of the elastic modulus, friction coefficient, and anisotropy coefficient of the flexible fabric to be laid out through a non-contact sound wave propagation speed detection method and a surface roughness scanning method. The key physical parameters are used as initial conditions input into the layered coupled neural computing model.

[0012] Furthermore, the hierarchical coupled neural computation model includes a two-branch convolutional attention neural prediction model, which includes an interlayer interaction prediction branch. The interlayer interaction prediction branch takes the contact pressure distribution data collected by the distributed micro-force sensing array and the interlayer slip trajectory captured by polarized light vision as input, and focuses on the interlayer pressure change and slip significant region through the convolutional attention mechanism to establish an interlayer friction-slip coupled dynamic model, and predicts the micro-dislocation amount and accumulation trend of each flexible fabric layer in different laying stages in real time.

[0013] Furthermore, the dual-branch convolutional attention neural prediction model further includes an overall positioning compensation branch; the overall positioning compensation branch uses the temperature distribution data of the infrared thermal imaging sensor and the physical property parameters of the fabric as auxiliary inputs, combined with the preset trajectory of the laying path, and outputs the real-time compensation amount of each flexible fabric layer through the attention mechanism to associate the mapping relationship between interlayer micro-misalignment and overall positioning deviation, thereby realizing the linkage calculation of micro-misalignment and macro-positioning.

[0014] Furthermore, the hierarchical coupled neural computing model further includes an online adaptive model iterative optimization mechanism; the online adaptive model iterative optimization mechanism introduces a reinforcement learning iterative strategy, and during the multi-layer unfolding process, the actual localization error after each unfolding is used as a feedback signal to dynamically adjust the weight parameters of the dual-branch convolutional attention neural prediction model; the online adaptive model iterative optimization mechanism establishes a real-time backup and backtracking mechanism for model parameters, and when the localization error is detected to exceed the threshold, the historical optimal parameters are automatically called for recalculation.

[0015] Furthermore, the adaptive execution adjustment module includes a distributed magnetorheological flexible actuator. The surface of the actuator is covered with a magnetorheological flexible material. The stiffness and surface friction coefficient of the flexible material are dynamically changed by adjusting the magnetic field strength. The actuator, combined with the compensation amount output by the hierarchical coupled neural computing model, independently controls the laying actuators of different flexible fabric layers. By enhancing the local magnetic field strength, the friction between the actuator and the flexible fabric is increased, and slippage is suppressed. By adjusting the movement speed and force of the actuator in the corresponding area, layered independent micro-displacement compensation is achieved.

[0016] Furthermore, the adaptive execution adjustment module further includes a multi-node collaborative tension adjustment system; the multi-node collaborative tension adjustment system sets up multiple sets of independently adjustable tension control nodes around the spreading platform, each node drives a flexible tension roller through a servo motor, and adjusts the tension of the corresponding node in real time according to the tension requirements of each flexible fabric layer output by the layered coupled neural computing model, forming a collaborative control of local tension fine-tuning and overall spreading flatness.

[0017] Furthermore, the multi-dimensional sensing module and the hierarchical coupled neural computing module form a closed-loop control through real-time data interaction; the physical environment data and fabric state data captured by the multi-dimensional sensing module serve as the input to the hierarchical coupled neural computing module, and the compensation amount output by the hierarchical coupled neural computing module serves as the adjustment instruction of the adaptive execution adjustment module.

[0018] Furthermore, the multi-dimensional perception module, the hierarchical coupled neural computing module, and the adaptive execution adjustment module form a closed-loop control system through real-time data interaction; the closed-loop control system dynamically monitors, predicts, and adjusts the interlayer friction slippage, fabric physical properties, and overall positioning deviation during the multi-layer laying of the flexible fabric, thereby achieving precise hierarchical control of each fabric layer and maintaining synchronous data updates.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] This invention acquires the physical property parameters of the fabric and configures the initial parameters of the model during the pre-preparation stage. During the layered laying and multi-layer stacking stages, a multi-dimensional sensing module collects data such as contact pressure and texture deformation in real time. Combined with a layered coupled neural computing model, it predicts microscopic misalignment and positioning deviation. An adaptive execution adjustment module then performs independent compensation and tension adjustment for each layer. Finally, the model parameters are updated through iterative optimization to form a closed-loop control. This effectively solves the problem of low overall positioning accuracy in multi-layer fabric laying due to the accumulation of microscopic misalignment caused by interlayer friction and nonlinear slippage. It can dynamically monitor and accurately control interlayer interactions and positioning deviations, achieving precise layered control of each fabric layer, improving the laying and positioning accuracy of multi-layer flexible fabrics, and meeting the needs of high-precision assembly or composite processing. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 This invention provides a method for positioning and spreading multiple layers of flexible fabric based on neural computing, comprising the following steps:

[0024] S1. Pre-preparation stage: The elastic modulus, friction coefficient and anisotropy coefficient of the flexible fabric to be laid are obtained through the fabric physical property pre-sensing unit; and the equipment calibration of the multi-dimensional sensing module and the initial parameter configuration of the hierarchical coupled neural computing model are completed.

[0025] S2, Layered Spreading Stage: The spreading actuator is driven to spread the first layer of flexible fabric according to a preset path; the multi-dimensional sensing module collects in real time the contact pressure distribution data, fabric surface texture deformation and slip trajectory, and infrared thermal imaging temperature distribution data of the first layer of flexible fabric; the layered coupled neural computing model outputs the positioning deviation compensation amount of the first layer of flexible fabric; the adaptive execution adjustment module adjusts the spreading actuator and the multi-node collaborative tension adjustment system according to the compensation amount to ensure the positioning accuracy of the first layer of flexible fabric.

[0026] S3, Multi-layer stacking stage: Subsequent flexible fabric layers are stacked sequentially; for each layer of fabric stacked, the multi-dimensional sensing module focuses on collecting contact interaction data between the new layer and the lower layer; the layered coupled neural computing model predicts the micro-dislocation amount and overall positioning deviation between the new layer and the already laid layer; the adaptive execution adjustment module performs independent micro-displacement compensation and tension adjustment for the new layer and the already laid layer respectively, realizing a real-time closed loop of laying, detection and compensation;

[0027] S4. In the iterative optimization stage, after all layers of flexible fabric are laid out, the overall positioning accuracy is detected, and the positioning error data is fed back to the layered coupled neural computing model and the model parameters are updated.

[0028] Specifically, this method is used in automated textile and apparel production lines to handle the laying and positioning of multi-layered cotton-linen blended flexible fabrics. First, in the pre-preparation stage, the fabric physical property pre-sensing unit uses a non-contact acoustic wave detector to detect the cotton-linen blended fabric, obtaining its elastic modulus, coefficient of friction, and anisotropy coefficient. Simultaneously, the distributed micro-force sensor array and high-frame-rate polarized light vision camera in the multi-dimensional sensing module are calibrated to ensure data acquisition accuracy. The acquired fabric physical property parameters are then input into a hierarchically coupled neural computation model to complete the initial parameter configuration.

[0029] Next, the layering and spreading stage begins, driving the spreading actuator to spread the first layer of cotton-linen blended fabric along a preset linear spreading path. In the multi-dimensional sensing module, a distributed micro-force sensor array collects real-time data on the contact pressure distribution between the first layer of fabric and the spreading platform. A high-frame-rate polarized light vision camera captures the fabric surface texture deformation and slippage trajectory at 200 frames per second. An infrared thermal imaging sensor collects temperature distribution data at three key nodes: the start, midpoint, and end point of the spreading path. After receiving this data, the layered coupled neural computing model calculates and outputs the positioning deviation compensation amount for the first layer of fabric through internal algorithms. The adaptive execution adjustment module adjusts the movement speed of the spreading actuator based on this compensation amount, for example, adjusting the original movement speed from 5 cm / s to 4.8 cm / s, while simultaneously adjusting the tension of each node in the multi-node collaborative tension adjustment system to ensure the positioning accuracy of the first layer of fabric.

[0030] Then, the multi-layer stacking stage begins, with subsequent cotton-linen blended fabric layers being added sequentially. With each layer, the multi-dimensional sensing module focuses on collecting contact interaction data such as the contact pressure distribution and inter-layer slippage trajectory between the new layer and the layer below. Based on this data, the layered coupled neural computing model predicts the microscopic misalignment and overall positioning deviation between the new layer and the already laid layers. The adaptive execution adjustment module compensates for the micro-displacement of the new layer by adjusting the force of the laying actuator, for example, reducing the actuator force from 5 Newtons to 4.5 Newtons. Simultaneously, it fine-tunes the tension of the already laid layers, achieving a real-time closed loop for laying, detection, and compensation.

[0031] Finally, in the iterative optimization phase, after all layers of fabric have been laid out, an accuracy testing device is used to check the overall positioning accuracy. If a positioning error is found, the error data is fed back to the layered coupled neural computing model. The model updates its internal parameters based on the error data to improve positioning accuracy in subsequent fabric laying processes. This embodiment effectively solves the problem of low positioning accuracy caused by interlayer friction and nonlinear slippage in traditional methods for laying out multi-layer fabrics.

[0032] In this embodiment, the multi-dimensional perception module includes a micro-force-vision fusion perception unit, which comprises a distributed micro-force sensor array, a high-frame-rate polarized light vision camera, and an infrared thermal imaging sensor. The distributed micro-force sensor array is deployed on the contact surface of the spreading platform and the surface of the spreading actuator to collect real-time contact pressure distribution data between the flexible fabric layers. The high-frame-rate polarized light vision camera captures the texture deformation and interlayer slippage trajectory of the flexible fabric surface and uses the characteristics of polarized light to distinguish the boundary contours of different flexible fabric layers. The infrared thermal imaging sensor is positioned at key nodes in the spreading path to assist in determining the dynamic trend of interlayer slippage.

[0033] Specifically, the micro-force-vision fusion sensing unit in the multi-dimensional perception module uses piezoelectric sensing elements in its distributed micro-force sensing array. These elements are deployed across the entire contact surface of the spreading platform and the clamping surface of the spreading actuator. The spacing between each sensing unit in the array is set to 2 mm, enabling real-time acquisition of contact pressure distribution data at different locations between the flexible fabric layers. The high-frame-rate polarized light vision camera has a frame rate of 300 frames per second, and its lens focal length is adjusted to a value suitable for capturing fabric surface details, such as a 50 mm focal length. During spreading, the camera can clearly capture the texture deformation of the flexible fabric surface. Simultaneously, utilizing the characteristics of polarized light, it distinguishes the boundary contours of different flexible fabric layers, avoiding detection errors caused by blurred interlayer boundaries. The infrared thermal imaging sensor has a resolution of 640×512 pixels and is fixedly installed at key nodes in the spreading path, such as the four corners of the spreading platform. By detecting temperature changes between fabric layers, it helps determine the dynamic trend of interlayer slippage. When interlayer slippage occurs, local friction causes a temperature increase, which the sensor can promptly capture and transmit data.

[0034] In this embodiment, the multi-dimensional perception module further includes a fabric physical property pre-sensing and modeling unit. The fabric physical property pre-sensing and modeling unit obtains key physical parameters such as the elastic modulus, friction coefficient, and anisotropy coefficient of the flexible fabric to be laid out by using a non-contact sound wave propagation speed detection method and a surface roughness scanning method. The key physical parameters are used as initial conditions to input the hierarchical coupled neural computing model.

[0035] Specifically, the fabric physical property pre-sensing and modeling unit is equipped with a non-contact acoustic wave propagation velocity detector and a surface roughness scanner. When inspecting the flexible fabric to be laid, the non-contact acoustic wave propagation velocity detector emits sound waves of a specific frequency, such as 20 kHz, onto the fabric surface. By detecting the propagation speed of the sound waves within the fabric, and combining this with a pre-set algorithm, the elastic modulus of the fabric is calculated. The surface roughness scanner uses laser scanning to scan the fabric surface with a scanning accuracy set to 0.1 micrometers. Based on the fabric surface contour data obtained from the scan, the coefficient of friction and anisotropy coefficient of the fabric are calculated. These key physical parameters—elastic modulus, coefficient of friction, and anisotropy coefficient—are used as initial conditions input into the hierarchical coupled neural computation model, providing basic data support for subsequent model calculations and ensuring that the model can more accurately predict and calculate the fabric laying process.

[0036] In this embodiment, the layered coupled neural computation model includes a two-branch convolutional attention neural prediction model, which includes an interlayer interaction prediction branch. The interlayer interaction prediction branch takes the contact pressure distribution data collected by the distributed micro-force sensing array and the interlayer slip trajectory captured by polarized light vision as input. It focuses on the interlayer pressure change and significant slip regions through the convolutional attention mechanism, establishes an interlayer friction-slip coupled dynamic model, and predicts the microscopic misalignment and accumulation trend of each flexible fabric layer at different spreading stages in real time.

[0037] Specifically, in the hierarchical coupled neural computation model, the dual-branch convolutional attention neural prediction model uses contact pressure distribution data collected by a distributed micro-force sensing array and interlayer slip trajectories captured by polarized light vision as input data for its interlayer interaction prediction branch. First, the input contact pressure distribution data is normalized. Let the original value of the contact pressure distribution data be... The maximum value is The minimum value is Normalized contact pressure data The calculation formula is: ;

[0038] in, Raw contact pressure data collected by a distributed micro-force sensor array, in Pascals; This is the maximum value in the original contact pressure data, in Pascals. This is the minimum value in the original contact pressure data, in Pascals; These are normalized contact pressure data, dimensionless.

[0039] To convert normalized data into physically meaningful microscopic displacements, a preset conversion coefficient needs to be introduced: for contact pressure data, a pressure-displacement conversion coefficient is set. The unit is millimeters per pascal. This coefficient is obtained through pressure-deformation experiments on flexible fabrics of the same material and is used to convert the pressure effect corresponding to the normalized contact pressure into a microscopic displacement contribution. For the slip trajectory data, set the slip-displacement conversion coefficient. The unit is millimeters per pixel. This coefficient is obtained based on the pixel resolution calibration of a high frame rate polarized light vision camera and is used to convert the pixel data of the slip trajectory into a microscopic displacement contribution. ,in These are the normalized pixel values ​​of the sliding trajectory.

[0040] Next, through a convolutional attention mechanism, the model automatically focuses on regions of abrupt changes in interlayer pressure and significant slip, strengthening weight allocation in these regions. Then, an interlayer friction-slip coupled dynamic model is established, with the following formula for predicting micro-dislocation: ;

[0041] in, The microscopic misalignment of each flexible fabric layer is expressed in millimeters. and These are the weighting coefficients for the contact pressure distribution data and the slip trajectory data, respectively. They are dimensionless and obtained through model training, for example... , ; This represents the pressure-induced microscopic displacement, expressed in millimeters. The value represents the microscopic displacement caused by slip, expressed in millimeters. This is the model error term, measured in millimeters. Its value is controlled within a small range through model training and optimization, typically around [value missing]. millimeters;

[0042] To verify the physical rationality of the formula, specific parameter examples can be used for illustration: If Pascal, mm / Pascal ,but millimeters; if Pixels millimeters per pixel millimeters; combination , , millimeters, can be obtained The dimensions are all in millimeters, conforming to the logical requirement of adding physical quantities. This model can predict the microscopic misalignment and cumulative trend of each flexible fabric layer at different laying stages in real time, providing data for subsequent compensation and adjustment.

[0043] In this embodiment, the dual-branch convolutional attention neural prediction model further includes an overall positioning compensation branch. The overall positioning compensation branch uses temperature distribution data from an infrared thermal imaging sensor and fabric physical property parameters as auxiliary inputs. Combined with the preset trajectory of the spreading path, it uses an attention mechanism to associate the mapping relationship between interlayer micro-misalignment and overall positioning deviation, and outputs the real-time compensation amount of each flexible fabric layer, thereby realizing the linkage calculation of micro-misalignment and macro-positioning.

[0044] Specifically, the global localization compensation branch of the dual-branch convolutional attention neural prediction model uses temperature distribution data and fabric physical property parameters collected by an infrared thermal imaging sensor as auxiliary input data. First, the temperature distribution data is processed, assuming the original temperature distribution data value is... The maximum value is The minimum value is Normalized temperature data The calculation formula is: ;

[0045] in, Raw temperature data collected by an infrared thermal imaging sensor, in degrees Celsius; This represents the maximum value in the original temperature data, in degrees Celsius. This is the minimum value in the original temperature data, in degrees Celsius. These are normalized temperature data, dimensionless.

[0046] Elastic modulus in the physical properties of fabric coefficient of friction and anisotropy coefficient Normalization is also performed, similar to the processing method used for the temperature data mentioned above. Then, it is combined with the preset trajectory data of the spreading path. By linking the mapping relationship between interlayer micro-misalignment and overall positioning deviation through the attention mechanism, a calculation model for the overall positioning compensation amount is established, and the formula is: ;

[0047] in, The real-time compensation amount for each flexible fabric layer is expressed in millimeters. , and These are the weighting coefficients for the microscopic misalignment, auxiliary input data, and preset trajectory data, respectively. They are dimensionless and determined through model training. For example... , , ; The function is based on the microscopic misalignment quantity, describing the impact of microscopic misalignment on the overall positioning compensation, with the unit being millimeters. A function based on normalized auxiliary input data describes the impact of this data on overall positioning compensation, in millimeters; This provides the preset trajectory data for the unfolding path, in millimeters. The formula outputs the real-time compensation amount for each flexible fabric layer, enabling coordinated calculation of microscopic misalignment and macroscopic positioning, thus improving overall positioning accuracy.

[0048] In this embodiment, the hierarchical coupled neural computing model further includes an online adaptive model iterative optimization mechanism. The online adaptive model iterative optimization mechanism introduces a reinforcement learning iterative strategy. During the multi-layer unfolding process, the actual localization error after each unfolding is used as a feedback signal to dynamically adjust the weight parameters of the dual-branch convolutional attention neural prediction model. The online adaptive model iterative optimization mechanism establishes a real-time backup and backtracking mechanism for model parameters. When the localization error is detected to exceed the threshold, the historical optimal parameters are automatically called and recalculated.

[0049] Specifically, an online adaptive iterative optimization mechanism for the hierarchically coupled neural computation model is introduced, employing a reinforcement learning iterative strategy. During the multi-layer unfolding process, the actual localization error is detected after each unfolding step. This is used as a feedback signal input into the optimization mechanism. Let the current weight parameters of the model be... The weight parameters are adjusted based on the feedback signal. The calculation formula is:

[0050] in, These are the weight parameters of the model before iteration, and are dimensionless. These are the dimensionless weight parameters of the model after iteration. The learning rate is dimensionless and ranges from 0.001 to 0.01, and can be adjusted according to the actual situation. For weight parameters and positioning error loss function for variables The gradient is dimensionless.

[0051] Simultaneously, this optimization mechanism establishes a real-time backup and backtracking mechanism for model parameters. At fixed time intervals, such as 10 seconds, the current model parameters are backed up and stored. When a positioning error exceeding a preset threshold is detected, for example, an error exceeding 0.5 millimeters, the optimal parameters from the historical backup are automatically retrieved for recalculation. This ensures the model maintains consistently good prediction and computational performance, preventing a decrease in positioning accuracy due to parameter anomalies.

[0052] In this embodiment, the adaptive execution adjustment module includes a distributed magnetorheological flexible actuator. The surface of the actuator is covered with a magnetorheological flexible material. The stiffness and surface friction coefficient of the flexible material are dynamically changed by adjusting the magnetic field strength. The actuator, combined with the compensation amount output by the hierarchical coupled neural computing model, independently controls the laying actuator of different flexible fabric layers. By increasing the local magnetic field strength, the friction between the actuator and the flexible fabric is increased, and slippage is suppressed. By adjusting the movement speed and force of the actuator in the corresponding area, layered independent micro-displacement compensation is achieved.

[0053] Specifically, the distributed magnetorheological flexible actuator in the adaptive execution adjustment module has a surface-coated magnetorheological flexible material made of a mixture of granular ferromagnetic material and polymer. During the deployment process, the magnetic field strength applied to the actuator is adjusted based on the compensation amount output by the hierarchical coupled neural computation model. Let the initial magnetic field strength be... Adjusted magnetic field strength The calculation formula is: ;

[0054] in, The magnetic field strength before adjustment, expressed in Tesla; The adjusted magnetic field strength is expressed in Tesla. This is the magnetic field strength adjustment coefficient, measured in Tesla per millimeter, determined experimentally, for example... Tesla / millimeter; This represents the compensation amount output by the hierarchically coupled neural computation model, expressed in millimeters.

[0055] By adjusting the magnetic field strength, the stiffness and surface friction coefficient of the flexible material are dynamically changed. When it is necessary to increase the friction between the actuator and the flexible fabric to suppress slippage, the magnetic field strength is increased, for example, from 0.3 Tesla to 0.5 Tesla, which increases the stiffness of the flexible material and improves the surface friction coefficient. When micro-displacement compensation is required, the movement speed and force of the actuator in the corresponding area are adjusted according to the compensation amount, for example, the movement speed is adjusted from 6 cm / s to 5.5 cm / s, and the force is adjusted from 6 N to 5.8 N. The laying actuators of different flexible fabric layers are independently controlled to ensure the positioning accuracy of each fabric layer.

[0056] In this embodiment, the adaptive execution adjustment module further includes a multi-node collaborative tension adjustment system. The multi-node collaborative tension adjustment system sets up multiple sets of independently adjustable tension control nodes around the spreading platform. Each node drives a flexible tension roller through a servo motor. According to the tension requirements of each flexible fabric layer output by the hierarchical coupled neural computing model, the tension of the corresponding node is adjusted in real time to form a collaborative control of local tension fine-tuning and overall spreading flatness.

[0057] Specifically, the multi-node collaborative tension adjustment system of the adaptive execution adjustment module has eight independently adjustable tension control nodes evenly arranged around the spreading platform. Each node is equipped with a servo motor and a flexible tension roller. The servo motor has a power of, for example, 500 watts, and the flexible tension roller has a diameter of, for example, 50 millimeters. During the spreading process, the layered coupled neural computing model outputs the corresponding tension requirements according to the characteristics of each flexible fabric layer. For example, the tension requirement is set to 2 Newtons for the first thin fabric layer and 3 Newtons for the second thicker fabric layer.

[0058] A servo motor drives a flexible tension roller to rotate according to tension requirements, adjusting the tension level. Let the current tension be... The target tension is The tension adjustment formula is: ;

[0059] in, The tension before adjustment, measured in Newtons (N). The adjusted tension is expressed in Newtons (N). This is the tension adjustment coefficient, dimensionless, with a value of, for example, 0.3. This coefficient controls the rate of tension adjustment, preventing sudden tension changes from damaging the fabric. Through independent adjustment of each node, a synergistic control is achieved between local tension fine-tuning and overall spreading and smoothing, ensuring that the multi-layered fabric is spread out smoothly without wrinkles or misalignments.

[0060] In this embodiment, the multi-dimensional perception module and the hierarchical coupled neural computing module form a closed-loop control through real-time data interaction; the physical environment data and fabric state data captured by the multi-dimensional perception module serve as the input to the hierarchical coupled neural computing module, and the compensation amount output by the hierarchical coupled neural computing module serves as the adjustment instruction of the adaptive execution adjustment module.

[0061] Specifically, the multi-dimensional sensing module and the hierarchically coupled neural computing module interact in real time via industrial Ethernet, with a data transmission rate set to 1000Mbps to ensure fast and accurate data transmission, forming a closed-loop control. During the spreading process, the distributed micro-force sensor array, high-frame-rate polarized light vision camera, and infrared thermal imaging sensor in the multi-dimensional sensing module continuously capture physical environment data, such as the temperature and humidity data of the spreading platform, as well as fabric state data, such as the fabric's contact pressure, texture deformation, and slip trajectory, and transmit this data to the hierarchically coupled neural computing module in real time.

[0062] After receiving data, the layered coupled neural computing module processes the data using its internal algorithm model, outputting compensation amounts for each flexible fabric layer. These compensation amounts serve as adjustment commands for the adaptive execution adjustment module, which is transmitted in real-time to guide adjustments to the spreading actuator and the multi-node collaborative tension adjustment system. This closed-loop control with real-time data interaction enables timely responses to changes during fabric spreading, ensuring positioning accuracy.

[0063] In this embodiment, the multi-dimensional perception module, the hierarchical coupled neural computing module, and the adaptive execution adjustment module form a closed-loop control system through real-time data interaction. The closed-loop control system dynamically monitors, predicts, and adjusts the interlayer friction slippage, fabric physical properties, and overall positioning deviation during the multi-layer laying of the flexible fabric, thereby achieving precise hierarchical control of each fabric layer and keeping the data updated synchronously.

[0064] Specifically, the multi-dimensional sensing module, the hierarchical coupled neural computing module, and the adaptive execution adjustment module are connected through the industrial control system to form a complete closed-loop control system. During the multi-layer deployment of the flexible fabric, the multi-dimensional sensing module dynamically monitors the interlayer friction and slippage, changes in the physical properties of the fabric, and overall positioning deviation in real time, and transmits the monitored data to the hierarchical coupled neural computing module.

[0065] The layered coupled neural computing module analyzes and predicts data, calculates corresponding adjustment parameters, and transmits these parameters to the adaptive execution adjustment module. The adaptive execution adjustment module adjusts the spreading actuator and the multi-node collaborative tension adjustment system according to the adjustment parameters, achieving dynamic adjustment of interlayer friction slippage, fabric physical properties, and overall positioning deviation. Simultaneously, all modules maintain synchronized data updates. For example, the multi-dimensional sensing module transmits the latest monitoring data to the computing module every 0.1 seconds, and the computing module transmits the latest adjustment parameters to the execution module every 0.1 seconds, ensuring that the entire system can control the fabric spreading process in real time and accurately, achieving precise layered control of each fabric layer.

[0066] In summary, this invention acquires the physical property parameters of the fabric and configures the initial parameters of the model during the pre-preparation stage. During the layered laying and multi-layer stacking stages, a multi-dimensional sensing module collects data such as contact pressure and texture deformation in real time. Combined with a layered coupled neural computing model, it predicts microscopic misalignment and positioning deviation. An adaptive execution adjustment module then performs independent compensation and tension adjustment for each layer. Finally, iterative optimization updates the model parameters to form a closed-loop control. This effectively solves the problem of low overall positioning accuracy in multi-layer fabric laying due to the accumulation of microscopic misalignment caused by interlayer friction and nonlinear slippage. It can dynamically monitor and precisely control interlayer interactions and positioning deviations, achieving precise layered control of each fabric layer, improving the positioning accuracy of multi-layer flexible fabric laying, and meeting the needs of high-precision assembly or composite processing.

[0067] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0068] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for multi-layer laying and positioning of flexible fabrics based on neural computing, characterized in that: Includes the following steps: S1. Pre-preparation stage: The elastic modulus, friction coefficient and anisotropy coefficient of the flexible fabric to be laid are obtained through the fabric physical property pre-sensing unit; and the equipment calibration of the multi-dimensional sensing module and the initial parameter configuration of the hierarchical coupled neural computing model are completed. S2, Layered spreading stage: The spreading actuator is driven to spread the first layer of flexible fabric according to a preset path; The multi-dimensional sensing module collects the contact pressure distribution data, fabric surface texture deformation and slip trajectory, and infrared thermal imaging temperature distribution data of the first layer of flexible fabric in real time. The hierarchical coupled neural computing model outputs the positioning deviation compensation amount of the first layer of flexible fabric; the adaptive execution adjustment module adjusts the spreading execution mechanism and the multi-node collaborative tension adjustment system according to the compensation amount to ensure the positioning accuracy of the first layer of flexible fabric. S3. In the multi-layer stacking stage, subsequent flexible fabric layers are stacked sequentially. For each layer of fabric stacked, the multi-dimensional sensing module focuses on collecting contact interaction data between the newly added layer and the lower layer. The layered coupled neural computing model predicts the micro-dislocation amount and overall positioning deviation between the newly added layer and the already laid layer. The adaptive execution adjustment module performs independent micro-displacement compensation and tension adjustment for the newly added layer and the already laid layer, respectively, to achieve a real-time closed loop of laying, detection and compensation. S4. In the iterative optimization stage, after all layers of flexible fabric are laid out, the overall positioning accuracy is detected, and the positioning error data is fed back to the layered coupled neural computing model and the model parameters are updated.

2. The method for multi-layer laying and positioning of flexible fabrics based on neural computing according to claim 1, characterized in that, The multi-dimensional perception module includes a micro-force-vision fusion perception unit, which comprises a distributed micro-force sensor array, a high-frame-rate polarized light vision camera, and an infrared thermal imaging sensor. The distributed micro-force sensor array is deployed on the contact surface of the spreading platform and the surface of the spreading actuator to collect real-time contact pressure distribution data between the flexible fabric layers. The high-frame-rate polarized light vision camera captures the texture deformation and interlayer slippage trajectory of the flexible fabric surface and uses the characteristics of polarized light to distinguish the boundary contours of different flexible fabric layers. The infrared thermal imaging sensor is positioned at key nodes in the spreading path to assist in determining the dynamic trend of the interlayer slippage.

3. The method for multi-layer laying and positioning of flexible fabrics based on neural computing according to claim 1, characterized in that, The multi-dimensional perception module further includes a fabric physical property pre-sensing and modeling unit. The fabric physical property pre-sensing and modeling unit obtains key physical parameters such as the elastic modulus, friction coefficient, and anisotropy coefficient of the flexible fabric to be laid out by using a non-contact sound wave propagation speed detection method and a surface roughness scanning method. The key physical parameters are used as initial conditions input into the layered coupled neural computing model.

4. The method for multi-layer laying and positioning of flexible fabrics based on neural computing according to claim 2, characterized in that, The hierarchical coupled neural computation model includes a dual-branch convolutional attention neural prediction model, which includes an interlayer interaction prediction branch. The interlayer interaction prediction branch takes the contact pressure distribution data collected by the distributed micro-force sensing array and the interlayer slip trajectory captured by the high frame rate polarized light vision camera as input. It focuses on the interlayer pressure abrupt change and significant slip regions through the convolutional attention mechanism, establishes an interlayer friction-slip coupled dynamic model, and predicts the microscopic misalignment and accumulation trend of each flexible fabric layer in different laying stages in real time.

5. The method for multi-layer laying and positioning of flexible fabric based on neural computing according to claim 4, characterized in that, The dual-branch convolutional attention neural prediction model further includes an overall positioning compensation branch; the overall positioning compensation branch uses the temperature distribution data of the infrared thermal imaging sensor and the physical property parameters of the fabric as auxiliary inputs, combined with the preset trajectory of the laying path, and outputs the real-time compensation amount of each flexible fabric layer through the attention mechanism to associate the mapping relationship between interlayer micro-misalignment and overall positioning deviation, thereby realizing the linkage calculation of micro-misalignment and macro-positioning.

6. The method for multi-layer laying and positioning of flexible fabric based on neural computing according to claim 4, characterized in that, The hierarchical coupled neural computation model further includes an online adaptive model iterative optimization mechanism; the online adaptive model iterative optimization mechanism introduces a reinforcement learning iterative strategy, and during the multi-layer unfolding process, the actual localization error after each unfolding is used as a feedback signal to dynamically adjust the weight parameters of the dual-branch convolutional attention neural prediction model; the online adaptive model iterative optimization mechanism establishes a real-time backup and backtracking mechanism for model parameters, and when the localization error is detected to exceed the threshold, the historical optimal parameters are automatically called and recalculated.

7. The method for multi-layer laying and positioning of flexible fabrics based on neural computing according to claim 1, characterized in that, The adaptive execution adjustment module includes a distributed magnetorheological flexible actuator. The surface of the actuator is covered with a magnetorheological flexible material. The stiffness and surface friction coefficient of the flexible material are dynamically changed by adjusting the magnetic field strength. The actuator, combined with the compensation amount output by the hierarchical coupled neural computing model, independently controls the laying actuators of different flexible fabric layers. By increasing the local magnetic field strength, the friction between the actuator and the flexible fabric is increased, and slippage is suppressed. By adjusting the movement speed and force of the actuator in the corresponding area, layered independent micro-displacement compensation is achieved.

8. The method for multi-layer laying and positioning of flexible fabrics based on neural computing according to claim 1, characterized in that, The adaptive execution adjustment module further includes a multi-node collaborative tension adjustment system; the multi-node collaborative tension adjustment system sets up multiple sets of independently adjustable tension control nodes around the spreading platform, each node drives a flexible tension roller through a servo motor, and adjusts the tension of the corresponding node in real time according to the tension requirements of each flexible fabric layer output by the layered coupled neural computing model, forming a collaborative control of local tension fine-tuning and overall spreading flatness.

9. A method for multi-layer laying and positioning of flexible fabrics based on neural computing according to claim 1, characterized in that, The multi-dimensional sensing module and the hierarchical coupled neural computing module form a closed-loop control through real-time data interaction; the physical environment data and fabric state data captured by the multi-dimensional sensing module are used as inputs to the hierarchical coupled neural computing module, and the compensation amount output by the hierarchical coupled neural computing module is used as the adjustment command of the adaptive execution adjustment module.

10. The method for multi-layer laying and positioning of flexible fabric based on neural computing according to claim 1, characterized in that, The multi-dimensional perception module, the hierarchical coupled neural computing module, and the adaptive execution adjustment module form a closed-loop control system through real-time data interaction. The closed-loop control system dynamically monitors, predicts, and adjusts the interlayer friction slippage, fabric physical properties, and overall positioning deviation during the multi-layer laying of the flexible fabric, thereby achieving precise hierarchical control of each fabric layer and keeping the data updated synchronously.

Images