An intelligent dynamic chip capturing system based on multi-modal perception and digital twinning and a control method thereof

Abstract

This invention discloses an intelligent dynamic chip-collecting system and its control method based on multimodal perception and digital twin, belonging to the field of automation control technology. The system includes a multimodal perception module, a dynamic follow-up suction execution module, a multi-stage separation and recovery module, and a digital twin and intelligent control module. The multimodal perception module is used to acquire workpiece point cloud data, grinding contact force, and dust concentration in the working area. The dynamic follow-up suction execution module is used to dynamically adjust the direction and opening of the suction port to follow the direction of chip splashing. Through multimodal sensor fusion and digital twin closed-loop decision control, the adjustable-direction suction unit at the end dynamically matches the chip splashing path. Combined with multi-condition adaptive suction adjustment, self-learning optimization mechanism, and three-stage separation and recovery and predictive filter maintenance technology, the system solves the core contradiction of the mismatch between the static characteristics of traditional chip-collecting devices and the fully dynamic characteristics of robotic polishing, significantly improving chip capture efficiency.

Description

Technical Field

[0001] This invention relates to the field of automation control technology, and in particular to an intelligent dynamic chip removal system and its control method based on multimodal sensing and digital twins. Background Technology

[0002] In high-end manufacturing sectors such as rail transit equipment, precision hardware processing, and automotive and aerospace manufacturing, robotic automated polishing and grinding of complex curved surface parts has become the mainstream processing method. However, the grinding process generates a large amount of metal shavings with a wide particle size range, high-temperature oxidized dust, and submicron-sized inhalable particles. These pollutants not only deteriorate the workshop working environment and cause occupational health risks such as pneumoconiosis for operators, but also adhere to the surfaces of precision testing instruments and robot joints, causing equipment accuracy drift. They can even enter the equipment through moving parts such as transmission chains and guide rails, exacerbating the wear and fatigue failure of key components and seriously threatening the long-term stable operation and production safety of the production line.

[0003] Currently, the existing technical solutions and their limitations for handling debris during the polishing and grinding process are mainly reflected in the following aspects: 1. Manual Offline Cleaning Method: This method relies on operators using brooms, compressed air, and other tools to manually clean the workpiece after each batch of grinding operations. This method is a non-continuous operation with low cleaning efficiency, and it cannot match the continuous production rhythm of automated production lines. When using compressed air to blow away dust, submicron-sized dust will be diffused throughout the workshop space, causing serious secondary dust pollution. At the same time, manual cleaning has inherent blind spots, making it difficult to thoroughly remove residual debris from the back of the workpiece, inner cavities, and corners of complex structures. These residues will flow with the workpiece to subsequent precision machining or assembly processes, resulting in secondary scratches on the workpiece surface, assembly gap deviations, and other quality defects.

[0004] 2. Fixed-point dust collection solution: Dust collection pipes and dust collection hoods are fixedly installed around the grinding station. Because the working point of the robot grinding operation dynamically changes with the complex curved surface machining trajectory, the fixed dust collection port cannot follow the grinding heat source and the center of debris splash in real time. It can only be effective when the grinding point is within the effective coverage area of ​​the dust collection port. For the far-end machining area of ​​large workpieces and the back-splash area of ​​complex curved surfaces, the dust collection efficiency is usually less than 30%, basically unable to achieve effective collection.

[0005] 3. Rigid Follow-up Dust Collection Solution: This solution integrates a fixed-direction dust collection port onto the robot's end effector, achieving rigid synchronous follow-up with the grinding tool. However, the fixed orientation and relative position of the dust collection port prevent dynamic adjustment of the suction angle based on the direction of debris splashing under different processing postures. This results in a large amount of debris splashed laterally and backward being unable to be effectively captured. Furthermore, the system operates with a constant airflow, failing to adaptively adjust negative pressure and airflow according to the grinding volume and dust generation intensity of different processes such as rough grinding, semi-fine grinding, and fine grinding. This leads to insufficient suction during rough grinding and redundant energy consumption during fine grinding, resulting in a low system energy efficiency ratio.

[0006] 4. Wet Spraying Treatment Solution: This method collects grinding debris by spraying water mist to suppress dust or rinsing with coolant. However, it generates a large amount of mixed slurry wastewater containing abrasive, metal powder, and cutting fluid. This requires the construction of complex solid-liquid separation, wastewater treatment, and abrasive recovery systems, resulting in high initial investment and daily operation and maintenance costs. Furthermore, residual moisture can cause corrosion of steel workpieces, necessitating additional drying and rust prevention processes. This method is completely unsuitable for dry, oil-free polishing processes required in aerospace, precision mold making, and other fields.

[0007] In summary, existing debris handling technologies generally suffer from a core contradiction: the static / semi-static characteristics of dust collection devices are mismatched with the fully dynamic characteristics of robotic polishing operations. Furthermore, none of them have established a real-time perception and closed-loop control mechanism for the intensity of dust generation during polishing, the trajectory of debris splashes, and the state of residue accumulation at the workstation. As a result, the residue handling process has become a key bottleneck restricting the continuous operation of automated production lines, the consistency of product quality, and the safety of the working environment. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides an intelligent dynamic chip collection system and its control method based on multimodal sensing and digital twins. By using multimodal sensor fusion and digital twin closed-loop decision control to drive the adjustable end-effector suction unit to dynamically match the chip splash path, and combining multi-condition adaptive suction adjustment, self-learning optimization mechanism, and three-stage separation and recycling and predictive maintenance technology for filter elements, this invention solves the core contradiction of the mismatch between the static characteristics of traditional chip collection devices and the fully dynamic characteristics of robotic polishing. This significantly improves chip capture efficiency, effectively improves the working environment, ensures product processing quality, and guarantees the long-term stable operation of automated production lines.

[0009] In a first aspect, the present invention provides an intelligent dynamic dust collection system based on multimodal perception and digital twin, the system comprising a multimodal perception module, a dynamic follow-up dust collection execution module, a multi-level separation and recycling module, and a digital twin and intelligent control module; The multimodal sensing module integrates a 3D vision sensor, a force sensor, a dust concentration sensor, and a differential pressure sensor to acquire point cloud data of the workpiece surface, grinding contact force signals, dust concentration in the working area, and resistance status of the separation module. The dynamic follow-up vacuuming execution module includes an adjustable end vacuuming unit, a fixed vacuuming port array, and a variable frequency fan. It is used to respond to the instructions of the digital twin and intelligent control module and dynamically adjust the orientation and opening of the vacuuming port to follow the direction of debris splashing. The multi-stage separation and recovery module includes an electromagnetic separation unit, a cyclone separation unit, and a HEPA high-efficiency filter unit, which are used to perform graded separation treatment on the debris-containing airflow. The digital twin and intelligent control module estimates the debris splash direction in real time based on the splash direction fusion algorithm and Kalman filter, and controls the adjustable suction port angle of the end-effector to dynamically track the debris splash direction.

[0010] Preferably, the multimodal sensing module includes: The 3D vision sensor is deployed above the workstation or at the end of the robotic arm to pre-scan and acquire point cloud data of the workpiece surface, and to detect the spatial distribution of the debris accumulation area in real time, thereby constructing a three-dimensional debris distribution model. The force sensor is installed between the end of the robotic arm and the grinding tool to collect the normal and tangential components of the grinding contact force in real time. The dust concentration sensor is deployed around the work area to monitor the concentration of suspended particulate matter in the air. The differential pressure sensor is installed on both the front and rear sides of the filter unit to monitor the degree of filter clogging in real time.

[0011] Preferably, the dynamic follow-up vacuuming execution module includes an adjustable end vacuuming unit, a fixed vacuuming port array, and a variable frequency fan; The adjustable suction unit includes a rotary drive mechanism, an angle encoder, and an adjustable suction port. The rotary drive mechanism is used to drive the adjustable suction port to rotate around the sixth axis of the robot to a target angle. The angle encoder is used to provide real-time feedback on the actual angle of the adjustable suction port. The adjustable suction port is used to dynamically adjust the suction direction according to the angle of debris splash under different processing postures. The fixed suction port array is distributed among several electric valve suction ports on the workstation, and is used to control the opening and closing of the corresponding valves by looking up a table based on the TCP coordinates of the robot end effector. The variable frequency fan adopts variable frequency speed control technology, which adjusts the operating speed in real time according to the grinding process stage, dust concentration in the working area and filter element pressure difference. It also supports pulse backflushing mode to achieve automatic cleaning of the filter element.

[0012] Preferably, the decision-making logic operating within the digital twin and intelligent control module is based on the following splash direction estimation model, the relevant calculation formula of which is as follows: ; In the formula, The main direction vector of debris splash; The grinding tangent direction vector is obtained by projecting the robot's TCP velocity vector onto the tangent plane at the contact point. This is the normalized gravity vector, pointing towards the ground. and These are empirical weighting coefficients, representing the weights of the influence of grinding motion dominance and gravity assistance on the spatter direction, respectively. The grinding tangent direction vector is calculated using a projection algorithm based on point cloud normal vectors. The calculation formula is as follows: ; In the formula, The robot's TCP linear velocity vector is acquired in real time by the robot controller. The contact point normal vector is extracted from the point cloud model of the workpiece surface obtained by the 3D vision sensor. This is the grinding tangent direction vector.

[0013] Preferably, the digital twin and intelligent control module includes: The splash direction fusion algorithm is used to perform vector synthesis of the grinding motion direction and the gravity direction. The calculation formula is as follows: ; In the formula, This is the main direction vector of the synthesized splash. These are the weighting coefficients; It is a unit vector in the direction of velocity, i.e. ; It is the unit vector in the direction of gravity, i.e. ; The Kalman filter is used to smooth the input force signal, and its state update equation is: ; In the formula, This is the optimal contact force state estimate after filtering at the current moment; This is the prior prediction of the current time based on the state at the previous time step; This is the Kalman gain matrix, used to assign weights between the prediction model and the observed data; The original observation vectors are provided by the six-dimensional force sensor; This is the state transition matrix, used to describe the relationship between the system state and time.

[0014] Preferably, the digital twin and intelligent control module is configured with a self-learning optimization module. This module uses historical operational data to iteratively correct the threshold parameters in the control decision. The corrected adaptive threshold follows the following logic: ; In the formula, The control threshold for final execution; The preset threshold value at the factory; This is an influencing factor on the properties of grinding materials, used to compensate for the grinding differences between different materials, especially when machining materials with high hardness. The factor should be appropriately increased to enhance the clipping intensity; This is a workpiece size correction factor used to adjust chip-catching strategies to accommodate workpieces of different specifications.

[0015] Preferably, the multi-stage separation and recycling module includes: The electromagnetic separation unit uses a permanent magnet or electromagnet to generate a strong magnetic field to adsorb ferromagnetic debris in the debris-containing airflow, and achieves automatic unloading by timed power-off. The cyclone separation unit uses centrifugal force to inertially settle particles larger than a preset threshold. The HEPA high-efficiency filtration unit uses H13 grade high-efficiency filter paper for filtration.

[0016] Preferably, the system also has the following auxiliary control functions: Adaptive airflow logic: The digital twin and intelligent control module adjusts the variable frequency fan speed in real time based on the material removal rate estimation results to achieve on-demand air supply, increasing airflow in the coarse grinding stage and reducing energy consumption in the fine grinding stage; Stacking detection and automatic cleaning logic: The 3D vision sensor scans the workbench with point cloud, generates a height deviation map by registering it with a standard point cloud model, identifies the stacking area, and drives the robot to perform fixed-point cleaning. Predictive maintenance logic: Based on the linear relationship between the cumulative material removal amount and the rate of increase in filter element pressure difference, the remaining service life of the HEPA filter element is predicted, and a replacement warning is sent in advance.

[0017] Secondly, the present invention also provides a control method for an intelligent dynamic chip-catching system based on multimodal sensing and digital twins, applied to the intelligent dynamic chip-catching system based on multimodal sensing and digital twins as described above, the method comprising: S100 continuously acquires point cloud data of the workpiece surface, grinding contact force signal, dust concentration and pressure difference data of the working area, and uses Kalman filter for noise reduction. S200: Based on the current robot motion state and contact force vector, call the splash direction fusion algorithm to calculate the main direction of debris splash; S300: Determine the deviation between the main direction of debris splash and the current angle of the end suction port. If the deviation exceeds the preset deviation threshold, drive the adjustable suction port to rotate to the target angle to achieve dynamic tracking. S400: Adjust the variable frequency fan speed based on the estimated material removal rate, while monitoring the filter element pressure difference and dust collection box material level. S500 performs graded separation treatment on the airflow containing debris and records the cleaning log data.

[0018] Compared with the prior art, the present invention has the following advantages and beneficial effects: 1. This invention comprehensively collects workpiece surface morphology, grinding contact state, and working environment data through multimodal sensing fusion technology. Combined with the chip splash direction estimation and fusion algorithm built by the digital twin platform, it drives the adjustable end-effector dust collection unit to track the chip splash trajectory in real time. This solves the core contradiction of the mismatch between the static characteristics of traditional chip collection devices and the full dynamic characteristics of robotic polishing operations, significantly improves the total chip capture rate, significantly reduces the concentration of suspended dust in the working area, curbs occupational health risks caused by dust diffusion from the source, reduces secondary scratches and assembly gap deviations caused by residual chips on the workpiece surface, and avoids precision drift and wear of key components caused by chips intruding into the moving parts of precision equipment.

[0019] 2. By pre-setting multi-condition adaptive working modes, combined with variable frequency fan speed adjustment, dynamic control of dust inlet opening, and on-demand opening and closing technology of fixed dust inlet array, the suction power and dust collection range are dynamically matched according to the dust generation intensity of different process stages such as rough grinding and fine grinding, solving the problems of insufficient suction power in the rough grinding stage and redundant energy consumption in the fine grinding stage of traditional systems; at the same time, a self-learning optimization module based on weighted least squares method is integrated, which automatically optimizes the control strategy and decision threshold by continuously accumulating operation data of workpieces of different materials and specifications, realizing adaptive adaptation to various complex processing scenarios, and comprehensively improving the overall energy efficiency ratio and versatility of the system.

[0020] 3. Through electromagnetic-cyclone-high-efficiency filtration three-stage separation technology, efficient classification, separation and recycling of debris and dust of different particle sizes are achieved, ensuring that the treated air meets emission standards; at the same time, with real-time differential pressure monitoring and pulse backflushing automatic cleaning mechanism, predictive maintenance of filter elements is achieved, timely restoration of system filtration efficiency is achieved, performance degradation during long-term operation is avoided, the workload of manual maintenance is greatly reduced, the continuous operation cycle of the system is extended, and a reliable guarantee is provided for the long-term stable operation of automated production lines. Attached Figure Description

[0021] Figure 1 This is a structural block diagram of an intelligent dynamic debris-collecting system based on multimodal perception and digital twins.

[0022] Figure 2 This is a flowchart illustrating a control method for an intelligent dynamic debris-collecting system based on multimodal sensing and digital twins. Detailed Implementation

[0023] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention. It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0024] Please see Figures 1-2 This embodiment proposes an intelligent dynamic chip collection system and its control method based on multimodal perception and digital twins, applied to robotic polishing and grinding stations in fields such as rail transit equipment, hardware processing, and automobile manufacturing. The aim is to resolve the contradiction between the static characteristics of the chip collection device and the dynamic characteristics of chip splashing during polishing operations through multimodal sensor fusion, grinding mechanism modeling, and digital twin closed-loop decision control. The system includes a multimodal perception module, a dynamic follow-up dust collection execution module, a multi-level separation and recovery module, and a digital twin and intelligent control module. Each module achieves high-bandwidth, low-latency data interaction via an industrial Ethernet bus and fieldbus, and functional safety is ensured by a dual-redundant power supply architecture.

[0025] At the mechanical structure level, the dynamic follow-up suction module is fixedly connected to the robot workstation base via a rigid bracket. Internally, it houses an adjustable end-effector suction unit, a fixed suction port array, and a variable frequency fan. The adjustable end-effector suction unit includes a rotary drive mechanism, an angle encoder, and an adjustable suction port. The rotary drive mechanism, controlled by a digital twin and intelligent control module, drives the adjustable suction port to rotate around the robot's sixth axis to a target angle. The angle encoder provides real-time feedback on the actual angle of the adjustable suction port, forming a closed-loop position control to ensure the adjustable suction port accurately tracks the direction of debris splash. The adjustable suction port, mounted at the end of the dynamic follow-up suction module, utilizes a streamlined venturi tube. The structure, with its geometry optimized for specific fluid dynamics, maintains stable suction efficiency across a wide flow rate range, effectively reducing airflow resistance and noise. Each adjustable suction port is supported on the output shaft of the rotary drive mechanism via a precision bearing assembly, and the orientation of each port is controlled by an independent angle servo drive unit. This allows for dynamic adjustment of the suction direction based on the debris splash angle under different processing postures, maximizing the capture of lateral, backward, and oblique debris splashes. When the system detects a significant change in the debris splash direction, the angle servo drive unit quickly adjusts the adjustable suction port to the target angle for real-time tracking. In the finishing grinding stage, where grinding intensity is lower, the suction port opening is appropriately reduced to lower energy consumption.

[0026] The multimodal sensing module, serving as the system's data source, integrates a 3D vision sensor, a force sensor, a dust concentration sensor, and a differential pressure sensor. The 3D vision sensor, deployed above the workstation or at the end of the robotic arm, pre-scans and acquires point cloud data of the workpiece surface, and detects the spatial distribution of debris accumulation areas in real time, constructing a dynamic three-dimensional debris distribution model. The force sensor, a six-dimensional force sensor, is installed between the end of the robotic arm and the grinding tool, and collects the normal and tangential components of the grinding contact force in real time, providing data support for material removal rate estimation. The dust concentration sensor is deployed around the work area to monitor the concentration of suspended particulate matter in the air, providing a basis for airflow adjustment. The differential pressure sensor is installed on both sides of the filter unit to monitor the filter element clogging level in real time and evaluate filtration efficiency. All sensing data is transmitted to the digital twin and intelligent control module via shielded twisted-pair cables or a dedicated industrial bus.

[0027] The digital twin and intelligent control module adopts a dual-core heterogeneous processor architecture. The first core is a real-time control core, which is responsible for running high-frequency sampling, Kalman filtering algorithm and PID control of servo loop; the second core is a strategy operation core, which is responsible for running decision logic based on splash direction estimation model, self-learning optimization algorithm and field communication protocol stack. This hardware design ensures that the system has extremely high real-time response speed when processing complex multimodal data.

[0028] The core decision-making logic of the digital twin and intelligent control module is based on a splash direction estimation model. The system must ensure that the adjustable suction port at the end can track the splash direction of debris in real time, achieving real-time matching between the debris collection position and the splash path. The splash direction estimation model follows the following mathematical expression: ; In the formula, The main direction vector of debris splash serves as the target direction for controlling the angle of the end suction port. The grinding tangent direction vector is obtained by projecting the robot's TCP velocity vector onto the tangent plane at the contact point. This is the normalized gravity vector, pointing towards the ground. and These are empirical weighting coefficients, representing the influence of grinding motion dominance and gravity assistance on the spatter direction, respectively. Their value range is typically set to... , The specific values ​​are determined based on the actual grinding process parameters.

[0029] To accurately calculate the grinding tangent direction vector mentioned above, this invention employs a projection algorithm based on point cloud normal vectors, and the calculation formula is as follows: ; In the formula, The robot's TCP linear velocity vector is acquired in real time by the robot controller. The contact point normal vector is extracted from the point cloud model of the workpiece surface obtained by 3D vision pre-scanning; This is the calculated grinding tangent direction vector. Through this projection algorithm, the system can accurately determine the dominant influence of grinding motion on the direction of chip ejection.

[0030] Furthermore, to obtain accurate synthetic spatter direction during dynamic grinding, the digital twin and intelligent control module executes a spatter direction fusion algorithm based on weighted synthesis. This algorithm can perform vector synthesis of the grinding motion direction and the gravity direction; the expression for the main synthetic spatter direction is: ; In the formula, This is the main direction vector of the synthesized splash. and These are the weighting coefficients; It is a unit vector in the direction of velocity, i.e. ; It is the unit vector in the direction of gravity, i.e. Through this fusion algorithm, the system can achieve a balance between grinding motion dominance and gravity assistance, obtaining a more realistic estimate of the spatter direction.

[0031] Considering the presence of random high-frequency noise in the contact force signal during grinding, the system uses a Kalman filter to smooth the input force signal. The state update equation of the Kalman filter is defined as follows: ; In the formula, This is the optimal contact force state estimate after filtering at the current moment; This is the prior prediction value made for the current time based on the state at the previous time step; This is the Kalman gain matrix, used to assign weights between the prediction model and the observed data; The original observation vectors are provided by the six-dimensional force sensor; This is the state transition matrix, describing the relationship between the system state and time. Through this filtering mechanism, the angle servo drive unit can obtain stable control input, effectively avoiding frequent adjustment operations of the suction port under high-frequency disturbances and reducing fatigue load on the mechanical structure.

[0032] The multi-stage separation and recovery module serves as the hub for dust-laden airflow treatment, integrating three-stage separation units. The first stage is an electromagnetic separation unit, which uses permanent magnets or electromagnets to generate a strong magnetic field to adsorb ferromagnetic debris in the dust-laden airflow and automatically unloads the material by timed power-off. The second stage is a cyclone separation unit, which uses centrifugal force to achieve inertial sedimentation of large particles with a diameter greater than 10μm. The third stage is a HEPA high-efficiency filtration unit, which uses H13 grade high-efficiency filter paper with a filtration efficiency of over 99.95%, ensuring that the treated clean air meets emission standards.

[0033] In practical implementation, this invention pre-sets multiple adaptive working modes to cope with complex grinding environments. In coarse grinding mode, the system prioritizes chip collection efficiency. At this time, the variable frequency fan operates at high speed, the end dust suction port is at its maximum opening, and the entire fixed dust suction port array is open to maximize dust collection capacity. In fine grinding mode, the system enters energy-saving mode, the variable frequency fan speed is reduced, the end dust suction port is precisely oriented according to the splash direction, and the fixed dust suction port array is opened as needed, reducing energy consumption while ensuring the cleanliness of the machined surface.

[0034] As a key engineering optimization of this invention, the digital twin and intelligent control module integrates a self-learning optimization module based on the weighted least squares method. This module can dynamically correct the decision threshold in the control strategy based on the operation data of workpieces with different materials and specifications; the corrected adaptive threshold follows the calculation formula: ; In the formula, This is the control threshold that the system will ultimately execute. The preset threshold value at the factory; This is an influencing factor on the properties of grinding materials, used to compensate for the grinding differences of different materials (such as 45# steel, aluminum alloy, stainless steel, etc.). When machining materials with high hardness, this factor is appropriately increased to enhance the chip-catching strength. This is a workpiece size correction factor used to adjust chip-catching strategies to accommodate workpieces of different specifications.

[0035] The technical effects of the present invention will be quantitatively demonstrated through specific embodiments below.

[0036] In this embodiment, the experimental object is a standard medium-sized automobile engine cylinder block forging, made of 45# steel, with a surface roughness requirement of Ra1.6μm. The intelligent dynamic chip collection system is deployed at the robotic polishing station, with a pneumatic polishing tool installed at the end of the robotic arm, and an adjustable dust collection unit fixed above the polishing tool.

[0037] The robotic arm is set to perform rough grinding on the workpiece surface at a feed rate of 200 mm / s, with a grinding depth of 0.3 mm. A 3D vision sensor acquires a point cloud model of the workpiece surface in real time, and a six-dimensional force sensor measures a normal grinding force of approximately 15 N. The digital twin and intelligent control module uses the robot's TCP velocity vector... m / s and contact point normal vector The grinding tangent direction vector is calculated. Combined with weighting coefficients , The system calculates the main splash direction vector. The end suction port is adjusted to the target angle according to this direction vector, at which point the debris capture rate is measured to be 92%.

[0038] Comparative Case 1: Under the same grinding parameters, a fixed dust collection device was used. Because the angle of the dust collection port could not be dynamically adjusted according to the direction of the splashes, the fixed device could only capture debris splashed in a single direction during the grinding process, with a debris capture rate of only 58%, and a large amount of debris splashed to the area around the worktable.

[0039] Comparative Case 2: The robotic arm operates under the same conditions, but without activating any chip collection devices.

[0040] As can be seen from the experimental data in the table, the intelligent dynamic chip collection system provided by this invention achieved a chip capture rate of 92% under the operating conditions of this embodiment. This indicates that while the system achieves accurate estimation of the splash direction through multimodal perception fusion, it can also achieve real-time matching between the chip collection position and the splash path through the dynamic angle adjustment mechanism of the end suction port, significantly improving the reliability of the residue treatment process. In contrast, the fixed device in Comparative Case 1, due to its inability to adaptively adjust the angle, had a capture rate of only 58%, resulting in poor chip collection and secondary pollution of the working environment.

[0041] Furthermore, in order to solve the problem of filter clogging after long-term operation, the multi-stage separation and recovery module of the present invention has a predictive maintenance function; when the differential pressure sensor detects that the pressure difference before and after the filter unit exceeds a predetermined threshold and the dust concentration remains high, the digital twin and intelligent control module triggers an automatic cleaning program, controls the electromagnetic separation unit to disconnect power and unload material at regular intervals, and drives the variable frequency fan to enter the pulse backflushing mode, using high-pressure airflow to backflush the HEPA filter element and restore filtration efficiency.

[0042] In multi-robot collaborative control applications, this system can be configured with multiple distributed, adjustable-direction end-effector dust collection units for large, complex curved workpieces. The digital twin and intelligent control module employs a spatial grid-based collaborative algorithm to differentiate the dust collection direction of each unit. The dust collection unit located above the workpiece primarily captures upward-splashing high-temperature debris, while units located on the sides focus on capturing particles splashing tangentially, thus achieving omnidirectional debris capture at the workpiece level. This collaborative logic follows the following control equations: ; In the formula, It is a comprehensive chip collection efficiency evaluation index for multiple robots, used to measure the quantitative value of the collaborative effect of multiple dust collection units; This represents the total number of distributed vacuum units; For the first The weighting coefficient of each dust collection unit under the current working condition is dynamically allocated based on the dust collection unit's ability to cover different areas of the workpiece surface. To correspond to the current splash direction of the suction unit and the angle of the suction port The chip removal efficiency function is given below.

[0043] In terms of system safety, this invention uses an industrial bus to communicate with the robot controller in real time. When an emergency stop signal or robot collision warning is detected, the digital twin and intelligent control module will trigger the highest priority emergency command. Through the internally preset reset spring or high-speed drive circuit, all end suction ports will be retracted to a safe position in a very short time (usually less than 200ms) to ensure that the robot's safe operation characteristics are not affected.

[0044] In summary, this invention, through precise mechanical structure design, multi-dimensional environmental perception, and decision-making algorithms based on rigorous mathematical models, successfully resolves the contradiction between the static characteristics of the chip collection device and the dynamic characteristics of chip splashing in polishing operations. It not only achieves efficient chip capture and graded processing at the physical level, but also achieves deep integration of sensing data and actuators at the control level, providing a complete technical approach for automated polishing and cleaning in fields such as rail transit equipment, hardware processing, and automobile manufacturing.

[0045] In another extended embodiment of the invention, the end-effector adjustable suction unit is further configured as an intelligent structure with adaptive adjustment function. When the intensity of debris splashing is detected to exceed a preset threshold, the suction port automatically increases its opening to increase the suction flow rate; conversely, when the splashing intensity decreases, the suction port automatically decreases its opening to achieve on-demand power supply. This combination of flow rate-level adaptation and angle-level dynamic tracking constitutes a dual safety defense for the system.

[0046] The digital twin and intelligent control module also possesses self-learning capabilities based on historical data. By continuously recording sensing data and execution feedback during the operation, the system can automatically identify the influence of different workpiece materials and surface shapes on the spatter direction, and optimize the settings of weighting coefficients and threshold parameters. Using this prior information, the system can pre-set the chip-catching strategy before processing a new batch of workpieces, further improving the global optimization of chip-catching efficiency.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent dynamic chip removal system based on multimodal sensing and digital twin, characterized in that, It includes a multimodal sensing module, a dynamic follow-up vacuuming execution module, a multi-level separation and recycling module, and a digital twin and intelligent control module; The multimodal sensing module integrates a 3D vision sensor, a force sensor, a dust concentration sensor, and a differential pressure sensor to acquire point cloud data of the workpiece surface, grinding contact force signals, dust concentration in the working area, and resistance status of the separation module. The dynamic follow-up vacuuming execution module includes an adjustable end vacuuming unit, a fixed vacuuming port array, and a variable frequency fan. It is used to respond to the instructions of the digital twin and intelligent control module and dynamically adjust the orientation and opening of the vacuuming port to follow the direction of debris splashing. The multi-stage separation and recovery module includes an electromagnetic separation unit, a cyclone separation unit, and a HEPA high-efficiency filter unit, which are used to perform graded separation treatment on the debris-containing airflow. The digital twin and intelligent control module estimates the debris splash direction in real time based on the splash direction fusion algorithm and Kalman filter, and controls the adjustable suction port angle of the end-effector to dynamically track the debris splash direction.

2. The intelligent dynamic debris collection system based on multimodal perception and digital twin as described in claim 1, characterized in that, The multimodal sensing module includes: The 3D vision sensor is deployed above the workstation or at the end of the robotic arm to pre-scan and acquire point cloud data of the workpiece surface, and to detect the spatial distribution of the debris accumulation area in real time, thereby constructing a three-dimensional debris distribution model. The force sensor is installed between the end of the robotic arm and the grinding tool to collect the normal and tangential components of the grinding contact force in real time. The dust concentration sensor is deployed around the work area to monitor the concentration of suspended particulate matter in the air. The differential pressure sensor is installed on both the front and rear sides of the filter unit to monitor the degree of filter clogging in real time.

3. The intelligent dynamic chip removal system based on multimodal perception and digital twin as described in claim 2, characterized in that, The dynamic follow-up vacuuming execution module includes an adjustable end vacuuming unit, a fixed vacuuming port array, and a variable frequency fan; The adjustable suction unit includes a rotary drive mechanism, an angle encoder, and an adjustable suction port. The rotary drive mechanism is used to drive the adjustable suction port to rotate around the sixth axis of the robot to a target angle. The angle encoder is used to provide real-time feedback on the actual angle of the adjustable suction port. The adjustable suction port is used to dynamically adjust the suction direction according to the angle of debris splash under different processing postures. The fixed suction port array is distributed among several electric valve suction ports on the workstation, and is used to control the opening and closing of the corresponding valves by looking up a table based on the TCP coordinates of the robot end effector. The variable frequency fan adopts variable frequency speed control technology, which adjusts the operating speed in real time according to the grinding process stage, dust concentration in the working area and filter element pressure difference. It also supports pulse backflushing mode to achieve automatic cleaning of the filter element.

4. The intelligent dynamic chip removal system based on multimodal perception and digital twin as described in claim 3, characterized in that, The decision-making logic running inside the digital twin and intelligent control module is based on the following splash direction estimation model, and the relevant calculation formulas of the model are as follows: ； In the formula, The main direction vector of debris splash; The grinding tangent direction vector is obtained by projecting the robot's TCP velocity vector onto the tangent plane at the contact point. This is the normalized gravity vector, pointing towards the ground. and These are empirical weighting coefficients, representing the weights of the influence of grinding motion dominance and gravity assistance on the spatter direction, respectively. The grinding tangent direction vector is calculated using a projection algorithm based on point cloud normal vectors. The calculation formula is as follows: ； In the formula, The robot's TCP linear velocity vector is acquired in real time by the robot controller. The contact point normal vector is extracted from the point cloud model of the workpiece surface obtained by the 3D vision sensor. This is the grinding tangent direction vector.

5. The intelligent dynamic chip removal system based on multimodal perception and digital twin as described in claim 4, characterized in that, The digital twin and intelligent control module includes: The splash direction fusion algorithm is used to perform vector synthesis of the grinding motion direction and the gravity direction. The calculation formula is as follows: ； In the formula, This is the main direction vector of the synthesized splash. These are the weighting coefficients; It is a unit vector in the direction of velocity, i.e. ; It is the unit vector in the direction of gravity, i.e. ; The Kalman filter is used to smooth the input force signal, and its state update equation is: ； In the formula, This is the optimal contact force state estimate after filtering at the current moment; This is the prior prediction of the current time based on the state at the previous time step; This is the Kalman gain matrix, used to assign weights between the prediction model and the observed data; The original observation vectors are provided by the six-dimensional force sensor; This is the state transition matrix, used to describe the relationship between the system state and time.

6. The intelligent dynamic chip removal system based on multimodal perception and digital twin as described in claim 5, characterized in that, The digital twin and intelligent control module is equipped with a self-learning optimization module. This module uses historical operational data to iteratively correct the threshold parameters in the control decision. The corrected adaptive threshold follows the following logic: ； In the formula, The control threshold for final execution; The preset threshold value at the factory; This is an influencing factor on the properties of grinding materials, used to compensate for the grinding differences between different materials, especially when machining materials with high hardness. The factor should be appropriately increased to enhance the clipping intensity; This is a workpiece size correction factor used to adjust chip-catching strategies to accommodate workpieces of different specifications.

7. The intelligent dynamic chip removal system based on multimodal perception and digital twin as described in claim 1, characterized in that, The multi-stage separation and recycling module includes: The electromagnetic separation unit uses a permanent magnet or electromagnet to generate a strong magnetic field to adsorb ferromagnetic debris in the debris-containing airflow, and achieves automatic unloading by timed power-off. The cyclone separation unit uses centrifugal force to inertially settle particles larger than a preset threshold. The HEPA high-efficiency filtration unit uses H13 grade high-efficiency filter paper for filtration.

8. The intelligent dynamic chip removal system based on multimodal perception and digital twin as described in claim 1, characterized in that, The system also has the following auxiliary control functions: Adaptive airflow logic: The digital twin and intelligent control module adjusts the variable frequency fan speed in real time based on the material removal rate estimation results to achieve on-demand air supply, increasing airflow in the coarse grinding stage and reducing energy consumption in the fine grinding stage; Stacking detection and automatic cleaning logic: The 3D vision sensor scans the workbench with point cloud, generates a height deviation map by registering it with a standard point cloud model, identifies the stacking area, and drives the robot to perform fixed-point cleaning. Predictive maintenance logic: Based on the linear relationship between the cumulative material removal amount and the rate of increase in filter element pressure difference, the remaining service life of the HEPA filter element is predicted, and a replacement warning is sent in advance.

9. A control method for an intelligent dynamic chip-collecting system based on multimodal sensing and digital twin, characterized in that: The method, applied to the intelligent dynamic chip-catching system based on multimodal sensing and digital twins as described in any one of claims 1-8, comprises: S100 continuously acquires point cloud data of the workpiece surface, grinding contact force signal, dust concentration and pressure difference data of the working area, and uses Kalman filter for noise reduction. S200: Based on the current robot motion state and contact force vector, call the splash direction fusion algorithm to calculate the main direction of debris splash; S300: Determine the deviation between the main direction of debris splash and the current angle of the end suction port. If the deviation exceeds the preset deviation threshold, drive the adjustable suction port to rotate to the target angle to achieve dynamic tracking. S400: Adjust the variable frequency fan speed based on the estimated material removal rate, while monitoring the filter element pressure difference and dust collection box material level. S500 performs graded separation treatment on the airflow containing debris and records the cleaning log data.

Images