Mechanical arm laser processing process control method and system based on laser signal feedback

Abstract

The present application belongs to the field of laser precision machining and robot control technology, and relates to a mechanical arm laser machining process control method and system based on laser signal feedback, which synchronously collects molten pool coaxial photoelectric detection signal, joint encoding data and theoretical interpolation coordinates; through transforming the detection signal spectrum and filtering out the interference of plume radiation, the geometric characteristic parameters representing the shape evolution of the molten pool are obtained; the space vector difference between the physical coordinates of the machining head and the theoretical interpolation coordinates is calculated to determine the transient tracking error and map the power attenuation coefficient; the coefficient is used to reduce the laser transient output power in real time, so that the real-time laser energy density and the actual moving speed of the mechanical arm end are physically aligned, and the present application solves the time scale mismatching problem between the microsecond photoelectric response and the millisecond mechanical response by establishing the physical constraint of the energy output to the mechanical servo lag, effectively eliminates the local heat accumulation on the machining path and maintains the molten pool stability.

Description

Technical Field

[0001] This invention belongs to the field of laser precision machining and robot control technology, and particularly relates to a method and system for controlling the laser machining process of a robotic arm based on laser signal feedback. Background Technology

[0002] Currently, using industrial robots equipped with laser-guided ends for three-dimensional spatial contour processing has become a core path for manufacturing upgrading. However, existing technologies still have limitations in handling the dynamic coupling between spatial motion trajectory deviations and processing energy fields. 1. Pure kinematic trajectory compensation technology based on external precision measurement; for example, Chinese invention patent CN111546334A discloses an online pose error compensation method for industrial robots to reduce contour errors. It uses a laser tracker to measure the actual pose of the robotic arm end effector, combines the theoretical pose to calculate the position tracking error and convert it into contour error and joint angle error, which are then added to the drive node. The drawback is that this solution is essentially based on the spatial geometric feedback of offline calibration equipment. The decision logic lacks a rigid representation of the structural deformation induced by heat accumulation in laser processing operations. It does not map the dynamic physical deformation caused by thermal effects into the boundary constraints for subsequent trajectory correction, which may cause the system to output instructions that deviate from the actual processing trajectory during high-temperature continuous operation.

[0003] 2. Energy field closed-loop control technology based on single-point state parameters; for example, Chinese invention patent CN120560139A discloses a laser welding control system and control method, which monitors the surface temperature of the welded part in real time, adjusts the pulse width modulation duty cycle and working frequency, and dynamically controls the laser emission power and cooling equipment status. The drawback is that this technology focuses on the identification of the thermodynamic state of the weld pool and the open-loop prediction of energy, but fails to establish a closed-loop correlation between the identified thermodynamic characteristics and the multi-axis linkage spatial attitude. It cannot map the physical compression of the previously planned path by the sudden change in the molten pool morphology in real time, and it is difficult to cope with the risk of beam focus position drift under complex spatial curve processing conditions.

[0004] 3. Discrete automated processing architecture based on spatial physical isolation; for example, Chinese utility model patent CN203018918U discloses a laser processing equipment and laser processing system, which distributes laser processing platforms, feeding devices, and automated cantilever arms in the work area. The above-mentioned handling structure is used to exchange the processed materials between the material placement end and the processing end. The drawback is that such systems usually regard the material loading and unloading logistics and high-energy beam processing as discrete decision-making units, ignoring the vibration coupling effect of the robot body and the light-emitting components in long-term continuous operation; in long-cycle batch production, if there is no forced pruning and suppression mechanism for system micro-vibration, the tool center point is prone to cumulative drift, causing the final product to lose its value due to exceeding the tolerance limit.

[0005] Therefore, how to construct a processing control architecture with deep coupling perception capability of spatiotemporal physical characteristics, so that the error compensation logic converges to the real physical safety boundary throughout the entire processing cycle, is the technical problem to be solved by this invention. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art, solve the technical problem of energy input mismatch and actual displacement caused by mechanical servo lag in laser processing correction, and provide a robotic arm laser processing process control method and system based on laser signal feedback.

[0007] To achieve the above-mentioned objectives, the present invention provides a robotic arm laser processing control method based on laser signal feedback, comprising the following steps: Step 101, Synchronous acquisition of multi-dimensional machining data: The photoelectric sensing unit and the position detection module are used to synchronously acquire the coaxial photoelectric detection signal of the molten pool, the joint coding data output by the robotic arm joint drive device, and the theoretical interpolation coordinates sent in real time by the external CNC system. Step 102, reconstruction of molten pool geometric characterization parameters: frequency domain transformation is performed on the coaxial photoelectric detection signal of the molten pool and characteristic spectrum is extracted. The radiation noise interference generated by plasma plume is filtered out using a preset cutoff frequency threshold. The mid-to-low frequency characteristic components characterizing the thermal radiation intensity of liquid metal are extracted from the spectrum to reconstruct the molten pool geometric characterization parameters characterizing the evolution of the molten pool shape. Step 103, transient tracking error determination: The physical coordinates of the machining head are calculated based on the joint coding data, and the three-dimensional spatial vector deviation between the physical coordinates of the machining head and the theoretical interpolation coordinates is calculated to determine the transient tracking error that reflects the mass inertia and servo dynamic performance of the mechanical system. Step 104, Physical constraint mapping relationship establishment: Calculate the power attenuation coefficient based on the transient tracking error, and use the power attenuation coefficient to establish a physical constraint mapping relationship for dynamic tracking of the physical displacement response of the robotic arm on the correction path from the laser output power command. Step 105, Real-time modulation of laser transient power: The transient output power of the laser is reduced by using the power attenuation coefficient through the modulation interface of the laser controller. The real-time laser energy density is physically aligned with the actual moving speed of the robotic arm end effector by adjusting the laser pulse duty cycle, so as to eliminate local heat accumulation caused by motion lag.

[0008] In step 102 of this invention, the reconstruction of the geometric characterization parameters of the molten pool includes: determining the segmentation frequency of high-frequency components and mid-to-low-frequency components in the frequency domain space based on the difference characteristics of the radiation spectrum between the plasma plume and the liquid metal molten pool; and using a bandpass filtering algorithm to retain the mid-to-low-frequency components in the frequency band from 300Hz to 1500Hz.

[0009] In step 103 of this invention, determining the transient tracking error includes: extracting the dynamic current data and position loop deviation of the joint drive unit in real time; compensating for the elastic deformation caused by the mass inertia of the mechanical mechanism using the end effector dynamics model of the robotic arm; and calculating the three-dimensional spatial vector difference at the current sampling moment.

[0010] In step 104 of this invention, determining the power attenuation coefficient includes: defining a proportionality constant between the reference processing speed and the rated laser power; when the transient tracking error exceeds a preset geometric tolerance threshold, calculating the time change rate of the transient tracking error and mapping it to obtain the speed change on the correction path.

[0011] In step 105 of the present invention, reducing the transient output power of the laser includes: applying a power attenuation coefficient to the modulation port to reduce the peak power of the laser pulse, so that the laser power output curve is smoothly aligned with the deceleration curve of the actual moving speed of the mechanical end; and controlling the real-time laser energy density.

[0012] The method described in this invention further includes: extracting keyhole depth features and molten pool width features from the geometric characterization parameters of the molten pool; comparing the keyhole depth features with a preset stable molten depth threshold to determine whether there is a risk of collapse inside the molten pool; and if a risk of collapse is determined to exist, then superimposing a power compensation operator.

[0013] In step 101 of this invention, the acquisition frequency of joint coding data is not less than 2kHz. By using high-frequency sampling to match the dynamic response frequency of the laser continuous processing path during the three-dimensional space correction process, the calculation result of the power attenuation coefficient meets the real-time requirement of millisecond-level feedback.

[0014] The modulation accuracy of the transient output power of the laser described in this invention is from 1μs to 10μs. The laser waveform control algorithm with nanosecond-level response is used to track the motion trajectory deviation of the robotic arm in real time within the response range of 1ms to 10ms, thereby achieving time scale alignment between macroscopic motion and microscopic energy.

[0015] After step 105 of the present invention, the method further includes: inverting the weld morphology using the reconstructed molten pool geometric characterization parameters; and correcting the energy distribution weight in the subsequent correction path according to the inversion results to adapt to the thermally sensitive physical characteristics of aerospace thin-walled workpieces during thermal cycling.

[0016] The present invention also provides a control system for laser processing of a robotic arm based on laser signal feedback, comprising: The data synchronization acquisition module is used to synchronously acquire the coaxial photoelectric detection signal of the molten pool, joint coding data, and theoretical interpolation coordinates; The spectrum feature extraction module, connected to the data synchronous acquisition module, is used to perform frequency domain transformation on the coaxial photoelectric detection signal of the molten pool and filter out interference signals using a preset cutoff frequency threshold, so as to extract the mid-to-low frequency characteristic components that characterize the thermal radiation of liquid metal and reconstruct the geometric characterization parameters of the molten pool. The position deviation calculation module is used to calculate the physical coordinates of the machining head based on the joint coding data, and to calculate the three-dimensional spatial vector difference between the physical coordinates of the machining head and the theoretical interpolation coordinates in order to determine the transient tracking error. The energy constraint mapping module, connected to the position deviation calculation module, is used to determine the power attenuation coefficient based on the transient tracking error and establish a physical constraint mapping relationship between the laser output energy and the servo hysteresis of the mechanical mechanism. The power modulation control unit is used to reduce the transient output power of the laser through the laser modulation interface by utilizing the power attenuation coefficient, and to match the real-time laser energy density with the actual moving speed of the robotic arm end effector by adjusting the laser pulse duty cycle.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: Firstly, during the laser processing of the robotic arm, the method establishes a confidence assessment mechanism for interference in the processing environment by monitoring the fluctuation variance of the coaxial photoelectric feedback signal within a time sliding window. When the variance exceeds the preset confidence threshold, the system can identify non-characteristic noise caused by protective airflow disturbance or metal vapor flashing, and automatically freeze the update state of the laser waveform adjustment coefficient, maintaining only the motion compensation command. This logical judgment method avoids random interference signals misleading the closed-loop control loop, ensures the continuity and reliability of energy control logic during tracking and correction, and eliminates abnormal power jumps caused by false signal fluctuations.

[0018] Secondly, the method utilizes encoder feedback data from the servo motor of the robotic arm joint to calculate the actual physical coordinates of the machining head and compares them with the expected command coordinates to determine the transient position tracking error. This physical error is then converted into attenuation damping that characterizes the actual movement speed of the laser spot. This forces the laser output power to physically adapt to the servo hysteresis caused by the mass inertia of the mechanical system. When the robotic arm slows down due to drastic trajectory correction, the system synchronously reduces the transient laser power through a suppression function, ensuring that the linear energy density injected into the molten pool is always maintained within a safe threshold. This solves the time scale mismatch problem between microsecond-level photoelectric response and millisecond-level mechanical response in traditional control methods and avoids burn-through defects caused by excessive heat input on the correction path.

[0019] Third, the method performs spectral deconstruction on the coaxial photoelectric feedback signal in the frequency domain. By setting a cutoff threshold to filter out the high-frequency characteristic radiation generated by plasma plume, it extracts the mid-to-low frequency feature set characterizing the radiation of liquid metal. This feature decoupling logic based on the physical properties of material thermal radiation enables the system to penetrate the interference smoke generated by the high-energy beam and obtain the true geometric information of the molten pool contour and the thermophysical steady-state characteristics. This mechanism ensures that the generation of energy regulation commands is directly derived from the objective evolution of the material's melting state, rather than being controlled by unstable plume radiation, thus maintaining the consistency of the weld morphology during continuous processing. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the laser processing power and motion hysteresis compensation control process involved in the present invention; Figure 2 This is a schematic block diagram of the melt pool morphology feature identification based on spectral decoupling involved in the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings.

[0022] Example 1: This example discloses a method for controlling the laser processing of a robotic arm based on laser signal feedback, including the following steps: Step 101, Synchronous acquisition of multi-dimensional machining data: The photoelectric sensing unit and the position detection module are used to synchronously acquire the coaxial photoelectric detection signal of the molten pool, the joint coding data output by the robotic arm joint drive device, and the theoretical interpolation coordinates sent in real time by the external CNC system. Step 102, reconstruction of molten pool geometric characterization parameters: frequency domain transformation is performed on the coaxial photoelectric detection signal of the molten pool and characteristic spectrum is extracted. The radiation noise interference generated by plasma plume is filtered out using a preset cutoff frequency threshold. The mid-to-low frequency characteristic components characterizing the thermal radiation intensity of liquid metal are extracted from the spectrum to reconstruct the molten pool geometric characterization parameters characterizing the evolution of the molten pool shape. Step 103, transient tracking error determination: The physical coordinates of the machining head are calculated based on the joint coding data, and the three-dimensional spatial vector deviation between the physical coordinates of the machining head and the theoretical interpolation coordinates is calculated to determine the transient tracking error that reflects the mass inertia and servo dynamic performance of the mechanical system. Step 104, Physical constraint mapping relationship establishment: Calculate the power attenuation coefficient based on the transient tracking error, and use the power attenuation coefficient to establish a physical constraint mapping relationship for dynamic tracking of the physical displacement response of the robotic arm on the correction path from the laser output power command. Step 105, Real-time modulation of laser transient power: The transient output power of the laser is reduced by using the power attenuation coefficient through the modulation interface of the laser controller. The real-time laser energy density is physically aligned with the actual moving speed of the robotic arm end effector by adjusting the laser pulse duty cycle, so as to eliminate local heat accumulation caused by motion lag.

[0023] In step 102 of this embodiment, the reconstructed geometric characterization parameters of the molten pool include: determining the segmentation frequency of high-frequency components and mid-to-low-frequency components in the frequency domain space based on the difference characteristics of the radiation spectrum between the plasma plume and the liquid metal molten pool; and using a bandpass filtering algorithm to retain the mid-to-low-frequency components in the frequency band from 300Hz to 1500Hz.

[0024] In step 103 of this embodiment, determining the transient tracking error includes: extracting the dynamic current data and position loop deviation of the joint drive unit in real time; compensating for the elastic deformation caused by the mass inertia of the mechanical mechanism using the end-effector dynamics model of the robotic arm; and calculating the three-dimensional spatial vector difference at the current sampling moment.

[0025] In step 104 of this embodiment, determining the power attenuation coefficient includes: defining a proportional constant between the reference processing speed and the rated laser power; when the transient tracking error exceeds a preset geometric tolerance threshold, calculating the time change rate of the transient tracking error and mapping it to obtain the speed change on the correction path.

[0026] In step 105 of this embodiment, reducing the transient output power of the laser includes: applying a power attenuation coefficient to the modulation port to reduce the peak power of the laser pulse, so that the laser power output curve is smoothly aligned with the deceleration curve of the actual moving speed of the mechanical end; and controlling the real-time laser energy density.

[0027] The method described in this embodiment further includes: extracting keyhole depth features and molten pool width features from the geometric characterization parameters of the molten pool; comparing the keyhole depth features with a preset stable molten depth threshold to determine whether there is a risk of collapse inside the molten pool; if it is determined that there is a risk of collapse, then superimposing a power compensation operator.

[0028] In step 101 of this embodiment, the acquisition frequency of joint coding data is not less than 2kHz. By using high-frequency sampling to match the dynamic response frequency of the laser continuous processing path during the three-dimensional spatial correction process, the calculation result of the power attenuation coefficient meets the real-time requirement of millisecond-level feedback.

[0029] The modulation accuracy of the transient output power of the laser described in this embodiment is 1μs to 10μs. The laser waveform control algorithm with nanosecond-level response is used to track the motion trajectory deviation of the robotic arm in real time within the response range of 1ms to 10ms, thereby achieving time scale alignment between macroscopic motion and microscopic energy.

[0030] Following step 105 in this embodiment, the method further includes: inverting the weld morphology using the reconstructed molten pool geometric characterization parameters; and correcting the energy distribution weights in the subsequent correction path based on the inversion results to adapt to the thermally sensitive physical characteristics of aerospace thin-walled workpieces during thermal cycling.

[0031] This embodiment also provides a robotic arm laser processing control system based on laser signal feedback, including: The data synchronization acquisition module is used to synchronously acquire the coaxial photoelectric detection signal of the molten pool, joint coding data, and theoretical interpolation coordinates; The spectrum feature extraction module, connected to the data synchronous acquisition module, is used to perform frequency domain transformation on the coaxial photoelectric detection signal of the molten pool and filter out interference signals using a preset cutoff frequency threshold, so as to extract the mid-to-low frequency characteristic components that characterize the thermal radiation of liquid metal and reconstruct the geometric characterization parameters of the molten pool. The position deviation calculation module is used to calculate the physical coordinates of the machining head based on the joint coding data, and to calculate the three-dimensional spatial vector difference between the physical coordinates of the machining head and the theoretical interpolation coordinates in order to determine the transient tracking error. The energy constraint mapping module, connected to the position deviation calculation module, is used to determine the power attenuation coefficient based on the transient tracking error and establish a physical constraint mapping relationship between the laser output energy and the servo hysteresis of the mechanical mechanism. The power modulation control unit is used to reduce the transient output power of the laser through the laser modulation interface by utilizing the power attenuation coefficient, and to match the real-time laser energy density with the actual moving speed of the robotic arm end effector by adjusting the laser pulse duty cycle.

[0032] Example 2: In the application scenario of three-dimensional trajectory laser welding of thin-walled aerospace parts, when the system faces the working condition of dynamic thermal deformation of titanium alloy workpieces due to laser heat input and trajectory deviation caused by assembly tolerance, the method provided by this invention uses the end coaxial photoelectric sensing unit and position detection module to collect the coaxial photoelectric detection signal of the molten pool, joint coding data, and theoretical interpolation coordinates sent in real time by the external CNC system. The coaxial photoelectric detection signal of the molten pool is transformed to the frequency domain to extract the characteristic spectrum. At the same time, the radiation noise generated by plasma plume is filtered out with the cutoff frequency threshold. Thus, the mid-to-low frequency characteristic components characterizing the thermal radiation intensity of liquid metal are extracted, thereby reconstructing the geometric characterization parameters of the molten pool. These parameters are used to characterize the shape evolution of the molten pool.

[0033] The physical coordinates of the machining head are calculated using joint coding data, and the three-dimensional spatial vector deviation between the physical coordinates and the theoretical interpolation coordinates is calculated to determine the transient tracking error, which reflects the mass inertia and servo dynamic performance of the mechanical system. Based on the transient tracking error, the energy suppression damping factor is calculated. The formula for calculating the energy suppression damping factor is as follows: ,in, As an energy suppression damping factor, This represents the actual relative speed. To achieve the expected speed, the laser pulse duty cycle is adjusted through the modulation interface of the laser controller using an energy suppression damping factor, so that the real-time laser energy density is physically aligned with the actual moving speed of the robotic arm end effector, thereby suppressing local heat accumulation caused by motion lag.

[0034] The system monitors the fluctuation variance of the coaxial photoelectric feedback signal within a preset time window and compares the fluctuation variance with a preset confidence threshold. When the fluctuation variance is not less than the confidence threshold, the current state of the laser waveform adjustment coefficient is maintained and only the motion compensation command is maintained. This logic avoids random interference signals from misleading the energy control loop. Combined with the feedback constraint of the expected end acceleration, the spatial pose correction and energy distribution adjustment of the laser processing system are synchronized within the control cycle, maintaining the consistency of weld penetration depth on the correction path, and the system enters a stable processing state.

[0035] Example 3: This example verifies the synchronous coordination between energy distribution and mechanical pose correction during the welding of thin-walled titanium alloy parts. The test platform uses a six-axis industrial robotic arm with a continuous laser generator with a rated power of 2000W. The repeatability of the robotic arm is not less than 0.05mm and the maximum processing speed is 2.0m / s. The laser modulation frequency supports 100kHz. The test signal source is obtained through a coaxial photoelectric sensing unit with a sampling rate of 2000Hz. During the data acquisition process, Gaussian white noise with a signal-to-noise ratio of 20dB is superimposed on the original signal to simulate electromagnetic interference in the industrial environment.

[0036] Set sampling period A balance needs to be struck between the real-time performance of the control closed loop and the risk of sampling aliasing; its value is controlled by the natural vibration frequency of the mechanical system. With controller operation delay The interaction constraints, based on the Nyquist criterion, reserve a safety margin of 5 times, and the decision logic is defined as follows: To determine the sampling period based on the characteristic vibration frequency of 40Hz at the end of the robotic arm in this experiment. The value of 2ms meets the requirements for extracting dynamic pose deviation data. The experiment was divided into three groups. The control group used proportional-integral-derivative control logic to compensate only for the robot arm's pose. The experimental group of this invention used a control method with an energy suppression damping factor. The out-of-range control group set the energy suppression range outside the defined interval. Under the condition of 2.5mm trajectory deviation caused by the simulated assembly gap, due to the physical response delay of the servo system, the actual relative movement speed of the processing head on the correction path was... The laser output power decreased from the expected 1.20 m / s to 0.98 m / s. In the control group, the laser output power remained constant, resulting in a 22.4% increase in the linear energy density of the correction section. The measured molten pool radiation intensity signal fluctuated, and the fluctuation variance increased to 0.18.

[0037] The present invention uses physical coordinates derived from joint-coded data to calculate transient tracking error, and determines the energy suppression damping factor according to the following formula: ,in, As an energy suppression damping factor, This represents the actual relative speed. The sample solution obtained by this invention is the speed expected by the command. The value is 0.817. Based on this, the controller adjusts the laser pulse duty cycle, synchronously reducing the output power to 82.0% of the rated value. At this point, the variance of the radiant intensity fluctuation of the molten pool in the sample group of this invention remains between 0.04 and 0.06. The surface characteristics of the molten pool exhibit quasi-periodic evolution. The gradient response characteristics under different deviation intensities show correlated changes. When the trajectory tracking error is at the 0.5mm level, the system calculates... With a laser power of 0.96 and fine-tuning, the measured weld penetration deviation was 2.1%. When the tracking error increased to the 2.5mm level, the system limited the energy input, and the penetration deviation remained within 4.5%. In the out-of-range control group, when... When the overload of the adjustment weight causes the power voltage drop to exceed the compensation requirement of the actual moving speed, the low energy obtained by the molten pool will cause the keyhole to collapse. This set of data provides the verification basis for the limited parameter range to maintain the working window of welding steady state, establishes the physical constraint of energy output to mechanical servo lag, offsets the time scale mismatch caused by the mass inertia of the robotic arm, maintains the thermophysical balance of the fluid dynamics inside the molten pool under noisy conditions, and maintains the consistency of weld formation quality on the processing path.

[0038] Example 4: In this example, in a three-dimensional complex trajectory laser continuous welding scenario, maintaining physical steady state during the processing requires isolating the molten metal pool thermal radiation component from the coaxial photoelectric detection signal and establishing a deterministic mapping between mechanical pose deviation and laser energy. The controller receives the molten pool coaxial photoelectric detection signal and uses a fast Fourier transform algorithm and a Hanning sliding data window to transform the signal to the frequency domain. The length of the sliding sampling window is set to 512 sampling points. To determine the high-frequency cutoff threshold for filtering out plasma plume radiation noise, the controller calculates the power spectral density of the characteristic spectrum. The controller searches for the frequency domain inflection point where the first derivative of the power spectral density exceeds a preset attenuation rate threshold and establishes the corresponding frequency value as the high-frequency cutoff threshold. Based on the determined high-frequency cutoff threshold, the controller uses a Butterworth bandpass filter algorithm to retain the mid-to-low frequency characteristic components in the 300Hz to 1500Hz frequency band. By extracting the thermal radiation characteristic of the molten metal... The controller reconstructs the geometric characterization parameters of the molten pool by analyzing the low-to-mid-frequency characteristic components of the radiation intensity. The coaxial photoelectric sensing unit uses a four-quadrant photodetector with spatial resolution to collect the radiation signal from the molten pool. Since the frequency of liquid metal fluctuations in different regions of the molten pool is related to the surface tension and local mass distribution of its capillary waves, the controller establishes a mapping matrix between the frequency feature vector and the spatial coordinates. This mapping matrix is ​​pre-calibrated through offline experiments on the standard molten pool morphology, associating the amplitude intensity of a specific frequency band with different quadrant regions of the four-quadrant photosensitive surface. When the controller acquires the low-to-mid-frequency characteristic components after frequency domain transformation, it uses this mapping matrix to restore the energy weights at different characteristic frequencies to the corresponding spatial pixel coordinate system, thereby reconstructing the one-dimensional temporal fluctuation information into a radiation intensity matrix with spatial distribution characteristics. Through this transformation process, the system can identify the spatial contour information of the molten pool thermal radiation based solely on the spectral characteristics of the photoelectric detection signal, even without camera imaging.

[0039] During the motion compensation and energy modulation phase, the controller continuously monitors the calculated three-dimensional spatial vector deviation. When the transient tracking error exceeds the geometric tolerance benchmark of 0.1 mm, the controller calculates the time evolution rate of the transient tracking error to obtain the velocity deviation on the actual correction path. The controller updates the actual relative movement speed based on the velocity deviation and calculates the energy suppression damping factor. The specific calculation formula is as follows: ,in, As an energy suppression damping factor, This represents the actual relative speed. The expected speed of the instruction.

[0040] The controller directly outputs the energy suppression damping factor to the modulation port of the laser controller to proportionally reduce the peak power of the laser pulse. This signal extraction and parameter conversion mechanism eliminates the potential problems of parameter presets in the control loop, ensuring physical alignment between the real-time laser energy density and the actual moving speed of the robotic arm's end effector. This suppresses local heat accumulation caused by motion lag. In the physical quantization process of reconstructing the geometric characterization parameters of the molten pool, the controller retrieves the optical transformation matrix stored in the read-only memory and projects the photoelectric radiation intensity distribution after bandpass filtering onto the two-dimensional pixel coordinate system. A preset radiation intensity contour extraction algorithm is used to identify the edge contour of the liquid metal. The specific execution steps of this contour extraction algorithm are as follows: the radiation intensity distribution in the reconstructed two-dimensional pixel coordinate system is normalized to generate a grayscale matrix. Set initial threshold The algorithm uses the Sobel operator to perform gradient operations on the gray-level matrix to identify the set of feature points with the largest rate of change in radiation intensity. Then, it performs a radial scan from the center of the radiation energy outwards to find points where the intensity value equals... Furthermore, continuous pixels pointing outwards along the gradient direction are located sub-pixel using bilinear interpolation. An eight-neighborhood search algorithm connects discrete feature points into a closed curve. When the perimeter of this curve deviates from the preset empirical perimeter of the molten pool by less than 5%, the closed curve is identified as the contour line of the liquid metal edge. The maximum lateral span of the contour is calculated based on the imaging magnification of the coaxial optical path and the physical size of the photoelectric sensor pixels, serving as a geometric index characterizing the width of the molten pool. The multi-reflection enhancement effect within the keyhole is calculated by integrating the radiation energy of the high-brightness region at the center of the molten pool and combining it with the optical path attenuation coefficient. This allows for the extraction of the keyhole depth features. Based on the blackbody chamber radiation capture physical model, the thermal radiation inside the deep groove exhibits a nonlinear energy evolution mechanism due to multiple sidewall reflections. The controller calculates the multi-reflection enhancement effect and quantifies the physical quantity of the keyhole depth features based on this principle. The specific calculation formula is as follows: ,in, The physical quantity representing the keyhole depth calculated quantitatively has a constraint value of a real number greater than zero. The calibration attenuation constant represents the correlation between the reflectivity of the workpiece material surface and the detection wavelength. This represents the cumulative value of the real-time radiant energy integrated in the central high-brightness area acquired by the photoelectric sensing unit. The reference radiation energy constant of the smooth molten pool is pre-collected and solidified using a specific planar reference test plate. The conversion procedure directly maps the two-dimensional planar radiation flux value to the three-dimensional depth coordinates. This step-by-step transformation logic from spectral components to spatial geometric dimensions eliminates the influence of the nonlinear gain of the probe optical path on the determination of the physical state.

[0041] Example 5: In the pre-calibration scenario of a laser processing system, the controller acquires laser power response characteristics before performing complex trajectory welding to establish the correspondence between the modulation interface signal and the laser pulse duty cycle. Specific steps include driving the laser generator to output test pulses at a preset command power, simultaneously adjusting the modulation voltage in 0.5V steps within the hardware range, synchronously acquiring the reflected light intensity from the surface of a standard test board using a coaxial photoelectric sensor unit, and calculating the ratio of the measured radiation intensity to the command power at each voltage step point to determine the energy suppression damping factor. The corresponding hardware drive command values ​​at different correction stages are used to offset the control nonlinearity caused by the difference in the modulation characteristics of the laser source, so that the reduction in linear energy density is physically aligned with the attenuation of the actual moving speed of the robotic arm end.

[0042] When the system faces specific protective airflow environments or workpiece surface conditions, the controller uses a baseline noise assessment program to quantify and determine the confidence threshold for freezing the laser waveform adjustment. Specifically, with the laser generator off, the protective gas is activated and the robotic arm moves along a preset trajectory. A coaxial photoelectric sensor unit collects background noise signals, and the signal fluctuation variance is calculated within a sliding window containing 512 sampling points. The controller extracts the maximum value of the fluctuation variance within the assessment period and multiplies it by a coefficient of 3 to calculate the confidence threshold. Here, the coefficient 3 is selected based on a normal distribution statistical model. In principle, since background noise caused by random protective airflow disturbances and electromagnetic interference in the processing environment usually follows a Gaussian distribution, the fluctuation variance calculated within the silent calibration period represents the standard deviation level of the system noise. Selecting three times the standard deviation as the confidence boundary can cover all random fluctuations caused by non-processing features with a probability of 99.7%. When the fluctuation variance of the real-time signal exceeds this confidence threshold, the system has a very high statistical confidence to determine that the signal abrupt change is due to external abnormal interference rather than the physical evolution of the molten pool, thereby triggering the freezing logic of the laser adjustment coefficient to avoid oscillations in the closed-loop system. This quantization method is used to distinguish the signal characteristics generated by random airflow fluctuations and the physical evolution of the molten pool. By performing in-situ calibration of the background noise characteristics before the welding task starts, the system establishes the feedback boundary of the energy control loop in a noisy environment, thereby shielding interference signals exceeding this physical boundary in real time during the correction process, and maintaining the thermophysical balance inside the molten pool in a stable state.

[0043] Example 6: In this example, in the production line deployment scenario of thin-walled components for aerospace fuel tanks, the control system performs a calibration procedure for mechanical inertia and optical response timeliness before formal operation. It drives the robotic arm's end effector to execute a preset trajectory within a speed range of 0.2 m / s to 1.8 m / s and synchronously transmits joint encoding data. The controller compares the theoretical interpolated coordinates at the sampling time with the physical coordinates of the processing head to obtain the hysteresis displacement at different dynamic response points. The hysteresis displacement is used to linearly fit the velocity decay within the sampling period, thereby generating a velocity compensation matrix stored in memory. This matrix serves as the basis for calculating the energy suppression damping factor. The baseline data is used to provide a quantitative basis for motion hysteresis during the actual correction process, thereby offsetting the energy regulation delay caused by the physical inertia of the servo system.

[0044] To address the differences in spectral radiation across different metallic surfaces, the system executes a calibration procedure for the characteristic frequencies of the molten pool signal to determine the high-frequency cutoff frequency threshold. The procedure involves applying a power of 500W to 1500W to the surface of the target workpiece to generate a plasma plume. Simultaneously, a coaxial photoelectric sensor unit acquires the radiation spectrum signal and performs a fast Fourier transform. The controller identifies the frequency distribution in the characteristic spectrum that belongs to the plasma pulsation characteristics and determines the high-frequency cutoff frequency threshold by calculating the corner frequency from the falling edge of the power spectral density to the noise level. This value provides the physical segmentation point for the bandpass filtering algorithm, so that the reconstructed geometric characterization parameters of the molten pool exclude feather background interference and only contain the signal components reflecting the flow of liquid metal. The system adjusts the filter parameters based on this calibration result to maintain the consistency between the laser energy distribution and the physical evolution state of the molten pool on the correction path.

[0045] In the standardized procedure for initial geometric tolerance threshold calibration for different metal materials, the system bases the initial geometric tolerance threshold on the thermal diffusivity of the target material and the laser spot diameter. Determine the thermal equilibrium radius of the molten pool Utilizing the thermal equilibrium radius With the diameter of the light spot The product relationship is used to establish a pose deviation criterion for triggering energy suppression, and the geometric tolerance threshold is set to the laser spot diameter. The value is 20% to 35%, and specifically 0.1 mm in the welding application of thin-walled titanium alloy parts. This quantitative boundary establishes the physical trigger point for the conversion of the dynamic deviation of the mechanical system into the laser power adjustment loop, enabling the controller to pre-compensate the linear energy density by calculating the time evolution rate of the tracking error before the trajectory tracking error induces the instability of the molten pool.

Claims

1. A method for controlling the laser processing of a robotic arm based on laser signal feedback, characterized in that, Includes the following steps: Step 101: Use the photoelectric sensing unit and position detection module to synchronously collect the coaxial photoelectric detection signal of the molten pool, the joint coding data output by the robotic arm joint drive device, and the theoretical interpolation coordinates sent in real time by the external CNC system. Step 102: Perform frequency domain transformation on the coaxial photoelectric detection signal of the molten pool and extract the characteristic spectrum. Use a preset cutoff frequency threshold to filter out the radiation noise interference generated by the plasma plume. Extract the mid-to-low frequency characteristic components that characterize the thermal radiation intensity of the liquid metal to reconstruct the geometric characterization parameters of the molten pool that characterize the shape evolution of the molten pool. The reconstruction of the geometric characterization parameters of the molten pool includes: determining the segmentation frequency of the high-frequency component and the mid-to-low frequency component in the frequency domain space based on the difference characteristics of the radiation spectrum between the plasma plume and the liquid metal molten pool; and using a bandpass filtering algorithm to retain the mid-to-low frequency components in the frequency band from 300Hz to 1500Hz. Step 103: Calculate the physical coordinates of the machining head based on the joint coding data, and calculate the three-dimensional spatial vector deviation between the physical coordinates of the machining head and the theoretical interpolation coordinates, so as to determine the transient tracking error that reflects the mass inertia and servo dynamic performance of the mechanical system. Step 104: Calculate the power attenuation coefficient based on the transient tracking error, and use the power attenuation coefficient to establish a physical constraint mapping relationship for dynamic tracking of the physical displacement response of the robotic arm on the correction path from the laser output power command. Determining the power attenuation coefficient includes: defining the proportional constant between the reference processing speed and the rated laser power; when the transient tracking error exceeds the preset geometric tolerance threshold, calculate the time change rate of the transient tracking error and map it to obtain the speed change on the correction path. Step 105: The transient output power of the laser is reduced through the modulation interface of the laser controller using a power attenuation coefficient. The real-time laser energy density is physically aligned with the actual moving speed of the robotic arm end effector by adjusting the laser pulse duty cycle, thereby eliminating local heat accumulation caused by motion lag. The process of reducing the transient output power of the laser includes: applying the power attenuation coefficient to the modulation port to reduce the peak power of the laser pulse, so that the laser power output curve is smoothly aligned with the deceleration curve of the actual moving speed of the robotic arm end effector; controlling the real-time laser energy density; and the modulation accuracy of the transient output power of the laser is 1μs to 10μs. A laser waveform control algorithm with nanosecond-level response is used to track the motion trajectory deviation of the robotic arm in real time within the response range of 1ms to 10ms, thereby achieving time-scale alignment between macroscopic motion and microscopic energy.

2. The method for controlling the laser processing of a robotic arm based on laser signal feedback according to claim 1, characterized in that, In step 103, determining the transient tracking error includes: extracting the dynamic current data and position loop deviation of the joint drive unit in real time; compensating for the elastic deformation caused by the mass inertia of the mechanical mechanism using the end effector dynamics model of the robotic arm; and calculating the three-dimensional spatial vector difference at the current sampling moment.

3. The method for controlling the laser processing of a robotic arm based on laser signal feedback according to claim 1, characterized in that, The method also includes: extracting keyhole depth features and molten pool width features from the geometric characterization parameters of the molten pool; comparing the keyhole depth features with a preset stable molten depth threshold to determine whether there is a risk of collapse inside the molten pool; if a risk of collapse is determined, then superimposing a power compensation operator.

4. The method for controlling the laser processing of a robotic arm based on laser signal feedback according to claim 1, characterized in that, In step 101, the acquisition frequency of joint coding data is not less than 2kHz. By using high-frequency sampling to match the dynamic response frequency of the laser continuous processing path during the three-dimensional space correction process, the calculation result of the power attenuation coefficient meets the real-time requirement of millisecond-level feedback.

5. The method for controlling the laser processing of a robotic arm based on laser signal feedback according to claim 1, characterized in that, After step 105, the method further includes: inverting the weld morphology using the reconstructed molten pool geometric characterization parameters; and correcting the energy distribution weights in the subsequent correction path based on the inversion results to adapt to the thermally sensitive physical characteristics of aerospace thin-walled workpieces during thermal cycling.

6. A robotic arm laser processing process control system based on laser signal feedback, used to implement the robotic arm laser processing process control method based on laser signal feedback as described in claim 1, characterized in that, include: The data synchronization acquisition module is used to synchronously acquire the coaxial photoelectric detection signal of the molten pool, joint coding data, and theoretical interpolation coordinates; The spectrum feature extraction module, connected to the data synchronization acquisition module, is used to perform frequency domain transformation on the coaxial photoelectric detection signal of the molten pool and filter out interference signals using a preset cutoff frequency threshold to extract the mid-to-low frequency characteristic components characterizing the thermal radiation of liquid metal and reconstruct the geometric characterization parameters of the molten pool. The reconstruction of the molten pool geometric characterization parameters includes: determining the segmentation frequency between high-frequency and mid-to-low frequency components in the frequency domain based on the difference in radiation spectra between the plasma plume and the molten metal pool; and using a bandpass filtering algorithm to retain the mid-to-low frequency components within the 300Hz to 1500Hz frequency band. The position deviation calculation module is used to calculate the physical coordinates of the machining head based on the joint coding data, and to calculate the three-dimensional spatial vector difference between the physical coordinates of the machining head and the theoretical interpolation coordinates in order to determine the transient tracking error. The energy constraint mapping module, connected to the position deviation calculation module, is used to determine the power attenuation coefficient based on the transient tracking error and establish a physical constraint mapping relationship between the laser output energy and the servo hysteresis of the mechanical mechanism. The determination of the power attenuation coefficient includes: defining the proportionality constant between the reference processing speed and the rated laser power; when the transient tracking error exceeds the preset geometric tolerance threshold, calculating the time change rate of the transient tracking error and mapping it to obtain the speed change on the correction path. The power modulation and control unit is used to reduce the transient output power of the laser through the laser modulation interface using a power attenuation coefficient, and to match the real-time laser energy density with the actual moving speed of the robotic arm end effector by adjusting the laser pulse duty cycle. The voltage reduction of the transient output power includes: applying the power attenuation coefficient to the modulation port to reduce the peak power of the laser pulse, so that the laser power output curve is smoothly aligned with the deceleration curve of the actual moving speed of the robotic arm end effector; controlling the real-time laser energy density; and the modulation accuracy of the transient output power is from 1μs to 10μs. A laser waveform control algorithm with nanosecond-level response is used to track the trajectory deviation of the robotic arm in real time within the 1ms to 10ms response range, achieving time-scale alignment between macroscopic motion and microscopic energy.

Images