A three-dimensional stone sculpture robot intelligent processing system and method

Abstract

The present application relates to the technical fields of intelligent manufacturing and stone processing, in particular to a three-dimensional stone carving robot intelligent processing system and method, the system includes an industrial robot, through a six-dimensional force sensor arranged between the end flange and the cutter chuck, real-time acquisition of three-direction cutting force and torque data during the carving process; an acceleration sensor array arranged on the end effector or the spindle box is used to obtain three-axis vibration signals, and the device temperature is monitored by relying on the attached temperature sensor to complete the sensor temperature drift error compensation. The intelligent information management platform includes three subsystems of signal processing, control decision and execution, and completes the filtering, amplification, analog-digital conversion and feature fusion processing of multi-source sensing signals. The platform constructs a force-position mapping model based on a neural network, combines the model reference adaptive control principle, and generates position compensation and control compensation in real time according to the machining trajectory and cutting force deviation. Finally, the six-axis linkage precision machining is realized by the robot controller driving the device.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing and stone processing technology, specifically to an intelligent processing system and method for three-dimensional stone carving robots. Background Technology

[0002] Stone carving art has a long history. Traditional processing methods rely on manual carving by craftsmen, which has problems such as high labor intensity, low production efficiency, and poor product quality consistency. With the maturity of industrial robot technology, its application to stone carving has become an industry trend. However, there are the following technical challenges in the stone carving process: (1) Large fluctuations in cutting force: Stone (especially granite and marble) is a hard and brittle material. During the carving process, the cutting force exhibits significant nonlinear characteristics. The cutting force can reach 500N-2000N in the roughing stage, and is accompanied by drastic changes, which can easily induce elastic deformation and chatter at the end of the robot. (2) Limited accuracy of traditional position control: In the existing technology, the robot mainly relies on offline programming position control. The controller outputs a fixed sequence of joint angles, which cannot be adjusted in real time according to the actual cutting state. Under the disturbance of cutting force, the position deviation of the end can reach 0.5mm-2mm, which is difficult to meet the strict requirements of the finishing process for surface quality. (3) Separation of vibration monitoring and force control: Although some existing systems are equipped with vibration sensors or force sensors, each sensor works independently and lacks an effective information fusion mechanism. It is impossible to establish an intrinsic correlation model of "force-vibration-position", resulting in a crude compensation strategy and a delayed response.

[0003] Therefore, there is an urgent need for an intelligent processing system for three-dimensional stone carving robots that can sense cutting force and vibration in real time, dynamically establish a force-position mapping model, and correct the robot trajectory online, so as to improve the processing accuracy and surface quality of stone carving. Summary of the Invention

[0004] To address the aforementioned technical problems, the present invention aims to overcome the shortcomings of existing technologies and provide an intelligent processing system and method for three-dimensional stone carving robots, enabling real-time compensation for dynamic errors during the stone carving process and improving processing accuracy and surface quality.

[0005] This invention provides an intelligent processing system and method for three-dimensional stone carving robots, including multiple industrial robots and an intelligent information management platform. Each industrial robot is equipped with a sensing subsystem, which includes a six-dimensional force sensor, an acceleration sensor array, and a temperature sensor. The six-dimensional force sensor is positioned between the robot's end effector flange and the tool holder to collect real-time data on the three-directional cutting forces Fx, Fy, and Fz, and the three-directional torques Mx, My, and Mz during the stone carving process. The acceleration sensor array is positioned on the side wall of the robot's end effector or spindle housing to collect three-axis vibration signals Ax, Ay, and Az. The temperature sensor is attached to the housing of the six-dimensional force sensor to monitor the sensor's operating temperature for temperature drift compensation. The intelligent information management platform includes a signal processing subsystem, a control decision subsystem, and an execution subsystem. The signal processing subsystem includes a signal conditioning module and an edge computing unit. The signal conditioning module is connected to the sensing subsystem to perform signal filtering, amplification, and analog-to-digital conversion. The edge computing unit is used for feature extraction and feature fusion of multi-source signals. The control decision subsystem includes a force-position mapping model and an adaptive compensation controller. The force-position mapping model is based on a neural network algorithm and uses multi-dimensional feature vectors and the current end position as input and output position compensation quantities. The adaptive compensation controller is based on the model reference adaptive control principle and uses the deviation between the desired trajectory and the actual cutting force as input and output real-time compensation control quantities. The execution subsystem includes a robot controller. The robot controller communicates with the control decision subsystem, receives the corrected end position command, and drives the industrial robot to perform six-axis linkage motion.

[0006] Preferably, the intelligent information management platform uses digital twin technology to achieve real-time simulation and monitoring of the processing process, and supports virtual debugging and process optimization.

[0007] Preferably, the six-dimensional force sensor uses a resistance strain gauge or piezoelectric sensing element, with a sampling frequency of not less than 5kHz, a static accuracy of not more than 0.5%FS, and an inter-channel crosstalk of not more than 2%.

[0008] Preferably, the accelerometer is a MEMS triaxial accelerometer with a range of ±50g and a frequency response range of 0.5Hz to 5kHz.

[0009] Preferably, the feature extraction includes time-domain analysis and frequency-domain analysis: time-domain analysis calculates the peak factor, root mean square value, and kurtosis coefficient of each channel signal; frequency-domain analysis uses fast Fourier transform to extract the main frequency component and the energy proportion of a specified frequency band.

[0010] Preferably, the force-position mapping model adopts a BP neural network or a one-dimensional convolutional neural network architecture, including an input layer, several hidden layers and an output layer. The number of neurons in the input layer is 11, and the number of neurons in the output layer is 6, corresponding to the three-directional position compensation quantities Δx, Δy, Δz and the three-directional attitude compensation quantities ΔRx, ΔRy, ΔRz.

[0011] Preferably, it also includes a machining status recognition module, which is used to automatically identify the current machining stage based on the real-time cutting force amplitude and vibration energy, and switch the corresponding compensation strategy.

[0012] Preferably, the processing stage includes a roughing stage, a semi-finishing stage, and a finishing stage; the roughing stage adopts a lookup table method for rapid compensation; the semi-finishing stage adopts a neural network inference compensation strategy; and the finishing stage adopts a fine compensation strategy that combines neural network inference and adaptive control.

[0013] A processing method based on the above-mentioned intelligent processing workshop system for three-dimensional stone carving robots is characterized by the following steps:

[0014] Step S1: System initialization, loading calibration parameters and pre-trained force-position mapping model weights, setting force and vibration thresholds for each processing stage; Step S2: The sensing subsystem acquires six-dimensional force signals, three-axis vibration signals, end-effector position signals, and temperature signals in real time; Step S3: The signal processing subsystem performs low-pass filtering and power frequency notch filtering on the original signal to eliminate noise interference, extract time-domain and frequency-domain features, and perform feature fusion. Step S4: The control decision subsystem obtains the initial compensation amount by reasoning through the force-position mapping model based on the fused feature vector and the current end position, and then performs real-time correction through the adaptive compensation controller to obtain the final compensation amount; Step S5: The execution subsystem adds the compensation amount to the original trajectory command, generates the corrected end position command, and sends it to the robot controller; Step S6: Repeat steps S2 to S5 until the processing is complete.

[0015] The present invention has the following beneficial effects: (1) Significantly improved machining accuracy: Through end force-vibration fusion perception and force-position mapping model, end chatter error caused by cutting force disturbance can be eliminated in real time, and the machining accuracy is improved by more than 50% compared with traditional position control. The surface roughness Ra of the finished surface can reach within 0.8μm; (2) Extended tool life: Dynamic compensation avoids the chipping phenomenon caused by overcutting, and the average tool life is extended by more than 30%; (3) Strong adaptability: The system can automatically switch the compensation strategy according to the machining stage (roughing / semi-finishing / finishing) without manual intervention and has good working condition adaptability; (4) Fast abnormal response speed: Through the localized inference of the edge computing unit, the control cycle can be controlled within 10ms, which meets the real-time requirements. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a structural diagram of the intelligent processing system for three-dimensional stone carving robots of the present invention; Figure 2 This is a flowchart illustrating the intelligent processing method of the three-dimensional stone carving robot of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 a part of the embodiments of the present invention, not all of them. 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. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. 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.

[0019] Example The following are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the following embodiments. All technical solutions that fall within the scope of the present invention are within the scope of protection of the present invention.

[0020] Reference manual attached Figure 1 This invention provides a three-dimensional stone carving robot intelligent processing system, including multiple industrial robots and an intelligent information management platform, specifically including a perception subsystem, a signal processing subsystem, a control decision subsystem and an execution subsystem.

[0021] The six-dimensional force sensor of the sensing subsystem uses a resistance strain gauge sensing element with measurement ranges set as Fx / Fy: 0~800N, Fz: 0~1500N, Mx / My: 0~10Nm, Mz: 0~15Nm, sampling frequency set to 10kHz, and static accuracy of 0.3%FS. The six-dimensional force sensor is installed between the robot end effector and the tool holder via a standard robot end flange interface. Specifically, the upper end face of the six-dimensional force sensor is bolted to the robot's sixth-axis end flange, and the lower end face is connected to the engraving tool via a BT50 tool holder interface.

[0022] The accelerometer array uses a triaxial MEMS accelerometer with a range of ±50g and a frequency response of 0.5Hz to 5kHz. It is mounted on the side wall of the spindle housing and connected to the signal conditioning module via a signal cable. The sensitive axes of the triaxial accelerometer point in the X, Y, and Z orthogonal directions to obtain complete spatial vibration information.

[0023] The temperature sensor uses a PT100 platinum resistance thermometer with an accuracy class of A. It is attached to the surface of the housing of the six-dimensional force sensor to monitor the sensor's operating temperature in real time and compensate for measurement deviations caused by temperature drift.

[0024] The joint angle encoder is integrated into each joint axis of the robot. It adopts a 23-bit absolute encoder with a resolution of 0.001°. It reads the current joint angle values ​​through the built-in interface of the robot controller and calculates the position and orientation of the robot end effector in the Cartesian coordinate system.

[0025] The signal processing subsystem includes a signal conditioning module and an edge computing unit. The signal conditioning module integrates a low-pass filter (500Hz cutoff frequency), a 50Hz / 60Hz power frequency notch filter, and a 16-bit analog-to-digital converter (ADC) to preprocess the six-dimensional force signal and the three-axis vibration signal. The edge computing unit 6 uses an Intel Core i7 processor and runs the Linux RT real-time operating system, and is responsible for feature extraction and model inference.

[0026] The control decision-making subsystem includes a force-position mapping model, an adaptive compensation controller, and a machining state recognition module. The force-position mapping model adopts a BP neural network architecture, including 1 input layer (11 neurons), 2 hidden layers (each layer has 32 neurons, using the ReLU activation function), and 1 output layer (6 neurons, corresponding to the compensation amount of 6 degrees of freedom). The adaptive compensation controller 8 adopts the MRAC structure, and the reference model is a second-order system with a natural frequency ωn = 20 rad / s and a damping ratio ζ = 0.7. The machining state recognition module judges the machining stage according to the root mean square value of the real-time cutting force: when RMS > 15 N, it is judged as the rough machining stage; when 5 N < RMS ≤ 15 N, it is judged as the semi-finishing stage; when RMS ≤ 5 N, it is judged as the finishing stage.

[0027] The execution subsystem includes a robot controller and a spindle servo drive. The robot controller communicates with the edge computing unit via the EtherCAT bus, and the communication cycle is set to 0.5 ms to ensure real-time requirements. The six-axis industrial robot 11 selects the ABB IRB 6700 series, with a load of 300 kg, an arm span of 3.2 m, and a repeat positioning accuracy of ±0.05 mm. The spindle servo drive 12 controls the spindle speed range of 12000 - 24000 rpm.

[0028] As shown in the attached Figure 2 to the specification, this embodiment provides an intelligent machining method for a three-dimensional stone carving robot, including the following steps: S1. System initialization: After power-on, the edge computing unit loads the pre-trained weight file of the force-position mapping model (in.pb format), and performs zero calibration and temperature drift compensation parameter setting for the six-axis force sensor. At the same time, the robot controller completes the homing operation of each joint to confirm the ready state of the system.

[0029] S2. Real-time perception: The six-axis force sensor collects six-axis force signals at a sampling rate of 10 kHz, the triaxial accelerometers 2a - 2c collect vibration signals at a sampling rate of 5 kHz, and the joint angle encoders update the end position data in real time. The above data is transmitted to the signal conditioning module through the hard-wired analog interface, and after analog-to-digital conversion, it is sent to the edge computing unit.

[0030] S3. Signal Processing: The edge computing unit preprocesses the original signal. First, high-frequency noise is removed by a Butterworth low-pass filter with a cutoff frequency of 200Hz. Then, power frequency interference is eliminated by a 50Hz notch filter. Next, feature extraction is performed on the filtered signal, including: time-domain features—root mean square (RMS), peak value, and kurtosis of the six-dimensional force; frequency-domain features—energy proportion at the spindle rotation frequency and its harmonics (obtained through FFT analysis). Finally, the 10-dimensional fused features (6-dimensional force features + 3-dimensional vibration features + 1-dimensional temperature features) are combined with the current end position to form an 11-dimensional input vector.

[0031] S4. Model Inference: The 11-dimensional input vector is fed into the BP neural network for inference. During the forward propagation, the model undergoes a nonlinear transformation through two hidden layers, outputting 6-DOF position compensation quantities Δx, Δy, Δz, ΔRx, ΔRy, and ΔRz. The inference process is accelerated using INT8 quantization, with a single inference cycle taking approximately 3ms.

[0032] S5. Compensation Control: The adaptive compensation controller 8 receives the initial compensation amount output from the force-position mapping model 7 and performs adaptive correction based on the current cutting force deviation. The controller uses incremental output, superimposing the corrected compensation amount onto the original trajectory command. Let the original trajectory command be P_ref and the corrected command be P_cmd, then: P_cmd = P_ref + K_adapt × ΔP_model + ΔP_controller Where K_adapt is the adaptive gain coefficient, ΔP_model is the compensation amount of the neural network output, and ΔP_controller is the output of the adaptive controller.

[0033] S6. Command Issuance: The corrected end-effector position command is sent to the robot controller via the EtherCAT bus. After the controller performs inverse kinematics calculation, it outputs the joint angle commands to drive the six-axis industrial robot to execute the corrected trajectory.

[0034] S7. Iterative Loop: Repeat steps S2 to S6 until the processing path is completed. Within each control cycle (approximately 5ms), the system continuously performs closed-loop control involving perception, reasoning, and compensation.

[0035] Differentiated compensation strategies are adopted for different stages of stone carving: (1) Roughing stage: The material removal is large in this stage, the cutting force is as high as 500N~1500N, and the vibration is severe. The system adopts a fast compensation strategy based on the lookup table method. According to the preset cutting parameters (feed speed, cutting depth, spindle speed) and the historical data table of force-position deviation, the corresponding compensation amount is directly indexed, and the response speed is fast (about 2ms).

[0036] (2) Semi-finishing stage: The amount of material removed in this stage is moderate, and the cutting force is 100N~500N. The system switches to a neural network inference compensation strategy, uses an offline-trained BP model for fine compensation, and combines it with an adaptive controller for online fine-tuning.

[0037] (3) Finishing stage: This stage aims to improve surface quality, with cutting force less than 100N. The system activates high-precision compensation mode, reduces the output gain of the neural network model, and increases the integral term coefficient of the adaptive controller to eliminate steady-state error. The end position deviation after compensation is controlled within ±0.05mm, and the surface roughness Ra≤0.8μm.

[0038] (4) Abnormal protection: The system monitors the root mean square value of the vibration signal and the peak value of the six-dimensional force in real time. If the vibration RMS exceeds the threshold (roughing stage: 5g; finishing stage: 0.5g) or the cutting force exceeds 90% of the range, an emergency stop signal is immediately triggered, the robot controller 10 controls the robot to enter a safe stop state, and at the same time alarms the operator to check.

[0039] The above description is merely a preferred embodiment of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention by those skilled in the art within the scope of the technology disclosed in the present invention using this concept shall be deemed as an infringement of the protection scope of the present invention.

Claims

1. A stereoscopic stone sculpture robot intelligent processing system, characterized in that, The system includes multiple industrial robots and an intelligent information management platform. Each industrial robot is equipped with a sensing subsystem, which includes a six-dimensional force sensor, an acceleration sensor array, and a temperature sensor. The six-dimensional force sensor is positioned between the robot's end effector flange and the tool holder to collect real-time data on the three-axis cutting forces Fx, Fy, and Fz, and the three-axis torques Mx, My, and Mz during the stone carving process. The acceleration sensor array is located on the side wall of the robot's end effector or spindle housing to collect three-axis vibration signals Ax, Ay, and Az. The temperature sensor is attached to the housing of the six-dimensional force sensor to monitor the sensor's operating temperature for temperature drift compensation. The intelligent information management platform includes a signal processing subsystem, a control decision subsystem, and an execution subsystem. The signal processing subsystem includes... The signal conditioning module is connected to the sensing subsystem to perform signal filtering, amplification, and analog-to-digital conversion. The edge computing unit is used for feature extraction and feature fusion of multi-source signals. The control decision subsystem includes a force-position mapping model and an adaptive compensation controller. The force-position mapping model is based on a neural network algorithm and uses multi-dimensional feature vectors and the current end position as input and output position compensation quantities. The adaptive compensation controller is based on the model reference adaptive control principle and uses the deviation between the desired trajectory and the actual cutting force as input and output real-time compensation control quantities. The execution subsystem includes a robot controller that communicates with the control decision subsystem, receives the corrected end position command, and drives the industrial robot to perform six-axis linkage motion.

2. The stereolithography robotic intelligent processing workshop system according to claim 1, wherein, The intelligent information management platform uses digital twin technology to achieve real-time simulation and monitoring of the processing, and supports virtual debugging and process optimization.

3. The intelligent processing workshop system for three-dimensional stone carving robots according to claim 1, characterized in that, The six-dimensional force sensor uses a resistance strain gauge or piezoelectric sensing element, with a sampling frequency of not less than 5kHz, a static accuracy of not more than 0.5%FS, and an inter-channel crosstalk of not more than 2%.

4. The intelligent processing workshop system for three-dimensional stone carving robots according to claim 1, characterized in that, The accelerometer is a MEMS triaxial accelerometer with a range of ±50g and a frequency response range of 0.5Hz to 5kHz.

5. The intelligent processing workshop system for three-dimensional stone carving robots according to claim 1, characterized in that, The feature extraction includes time-domain analysis and frequency-domain analysis: time-domain analysis calculates the peak factor, root mean square value, and kurtosis coefficient of each channel signal; frequency-domain analysis uses fast Fourier transform to extract the main frequency component and the energy proportion of a specified frequency band.

6. The intelligent processing workshop system for three-dimensional stone carving robots according to claim 1, characterized in that, The force-position mapping model adopts a BP neural network or a one-dimensional convolutional neural network architecture, which includes an input layer, several hidden layers and an output layer. The number of neurons in the input layer is 11 and the number of neurons in the output layer is 6, corresponding to the three-directional position compensation quantities Δx, Δy, Δz and the three-directional attitude compensation quantities ΔRx, ΔRy, ΔRz.

7. The intelligent processing workshop system for three-dimensional stone carving robots according to claim 1, characterized in that, It also includes a machining status recognition module, which is used to automatically identify the current machining stage based on the real-time cutting force amplitude and vibration energy, and switch the corresponding compensation strategy.

8. The intelligent processing workshop system for three-dimensional stone carving robots according to claim 7, characterized in that, The processing stages include roughing, semi-finishing, and finishing. The roughing stage employs a lookup table method for rapid compensation; the semi-finishing stage employs a neural network inference compensation strategy; and the finishing stage employs a fine compensation strategy that combines neural network inference with adaptive control.

9. A processing method based on the intelligent processing workshop system for three-dimensional stone carving robots according to any one of claims 1 to 8, characterized in that, Includes the following steps: Step S1: System initialization, loading calibration parameters and pre-trained force-position mapping model weights, setting force and vibration thresholds for each processing stage; Step S2: The sensing subsystem acquires six-dimensional force signals, three-axis vibration signals, end-effector position signals, and temperature signals in real time; Step S3: The signal processing subsystem performs low-pass filtering and power frequency notch filtering on the original signal to eliminate noise interference, extract time-domain and frequency-domain features, and perform feature fusion. Step S4: The control decision subsystem obtains the initial compensation amount by reasoning through the force-position mapping model based on the fused feature vector and the current end position, and then performs real-time correction through the adaptive compensation controller to obtain the final compensation amount; Step S5: The execution subsystem adds the compensation amount to the original trajectory command, generates the corrected end position command, and sends it to the robot controller; Step S6: Repeat steps S2 to S5 until the processing is complete.

Images