Online visual monitoring and compensating system for welding track of robot

Abstract

The invention relates to the technical field of robot welding control, and discloses an online visual monitoring and compensation system for a robot welding track, which comprises an excitation module, a sensing dimension reduction module, a dual-channel phase-locked demodulation module and a normalization compensation control module, the excitation module drives the galvanometer to apply periodic swing, the sensing dimension reduction module extracts a one-dimensional time-varying intensity signal, the dual-channel phase-locked demodulation module obtains a fundamental component and a second harmonic component of the signal in parallel, and the normalization compensation control module generates a compensation signal based on the ratio of the fundamental component to the second harmonic component. The second harmonic component is used for representing the optical response gain in the welding process, real-time normalization processing is carried out on the geometric deviation represented by the fundamental frequency component, the mechanism enables the system to avoid gain drift caused by workpiece materials and laser power changes, and self-adaption and stable tracking under variable working conditions are achieved.

Description

Technical Field

[0001] This invention relates to an online visual monitoring and compensation system for robot welding trajectory, belonging to the field of robot welding control technology. Background Technology

[0002] Currently, robotic laser welding, with its concentrated energy and tiny spot size, is widely used in automated production to achieve high-efficiency and high-quality connections. Especially in industries with stringent welding quality requirements, its application has become a mainstream technology. However, the physical characteristics of this process also make it extremely sensitive to the relative position of the laser spot and the weld. Any tiny trajectory deviation, caused by workpiece tolerances, assembly gaps, or tooling positioning errors, can directly lead to quality defects such as incomplete penetration or weak welds. To ensure welding quality, existing technologies generally rely on high-cost preconditions, namely, passively ensuring positioning accuracy through high-precision incoming parts and high-rigidity, high-cost precision tooling fixtures. This rigid binding of welding quality to the high cost of pre-processing limits the wider application of this technology.

[0003] The industry has attempted to introduce pre-welding or in-welding visual sensing systems to actively locate weld seam trajectories. However, this approach increases system complexity and cost. Furthermore, it largely relies on image recognition of cold-state geometric features of the workpiece, making it difficult to effectively address the dynamic thermal deformation inevitably caused by heat input during welding. Additionally, the intense plasma light, metal spatter, and fumes generated during welding also interfere with the imaging quality of visual sensors and the reliability of algorithms. Even some solutions attempting real-time online monitoring have not escaped these limitations, still relying on external visual recognition of geometric features. For example, [the following is an example of a solution with authorization announcement number CN]. Chinese invention patent 105345264B discloses a real-time online monitoring system for laser welding of complex curved surface components. The system attempts to capture the relative shape and position of the welding head and the workpiece in real time using an external CCD camera, and extract the geometric features of the light spot and the welding position through image processing, and then calculate the compensation vector. The fundamental flaw of this approach is that its sensing logic is based on the ideal premise of being able to clearly image and accurately identify geometric features. However, under real welding conditions, the feature information used for positioning is easily submerged or interfered with by strong plasma light, metal spatter and fumes, making it difficult for the system to operate stably and reliably under strong interference conditions.

[0004] Therefore, the technical problem to be solved by this invention is to provide a system and method that utilizes information from the welding process itself to perform real-time online monitoring and compensation for trajectory deviations caused by workpiece tolerances, assembly errors, and dynamic thermal deformation in a highly anti-interference manner, thereby reducing dependence on high-cost pre-tooling and high-precision incoming materials. Summary of the Invention

[0005] This invention provides an online visual monitoring and compensation system for robot welding trajectories. Its main purpose is to solve the problem that existing technologies rely on high-cost pre-precision calibration and are difficult to perform real-time and reliable compensation under strong interference conditions when dealing with trajectory deviations caused by workpiece tolerances, assembly errors, and dynamic thermal deformation.

[0006] To achieve the above objectives, this invention provides an online visual monitoring and compensation system for robot welding trajectories, applicable to robots containing laser welding heads. The laser welding head includes a scanning galvanometer and a coaxial vision sensor. The system includes: The excitation module is used to drive the scanning galvanometer to apply periodic oscillation modulation to the laser spot at a preset oscillation frequency. The perception dimension reduction module is used to acquire visual flow data of the interaction area between the laser spot and the workpiece through a coaxial vision sensor, and extract a one-dimensional time-varying intensity signal from the visual flow data. The dual-channel phase-locked demodulation module is used to perform two-way demodulation operations on a one-dimensional time-varying intensity signal in parallel. The first demodulation channel demodulates the one-dimensional time-varying intensity signal with a preset swing frequency as the first reference reference to obtain the fundamental frequency component characterizing the geometric deviation between the swing modulation center point and the actual weld. The second demodulation channel demodulates the one-dimensional time-varying intensity signal with twice the preset swing frequency as the second reference reference to obtain the second harmonic component characterizing the optical response gain of the interaction region. The normalization compensation control module is used to calculate a normalized trajectory deviation based on the ratio of the fundamental frequency component to the second harmonic component, and generate a compensation signal based on the normalized trajectory deviation to drive the scanning galvanometer to adjust the center point of the periodic oscillation modulation.

[0007] Preferably, the normalization compensation control module is used to calculate the normalized trajectory deviation. Normalized trajectory deviation The calculation rules follow: ;in The amplitude of the fundamental frequency component. This represents the amplitude of the second harmonic component.

[0008] Preferably, the perception dimensionality reduction module is used to perform averaging processing on the preset region of interest in the visual stream data to extract a one-dimensional time-varying intensity signal, and the dual-channel phase-locked demodulation module is used to execute the digital lock-in amplifier algorithm and periodically oscillate to modulate a linear oscillation perpendicular to the actual weld direction.

[0009] Preferably, the excitation module is used to apply periodic swing modulation with a dynamically changing swing frequency, and the dual-channel phase-locked demodulation module is used to: synchronously acquire the dynamically changing swing frequency as a first reference reference, and synchronously acquire twice the frequency of the dynamically changing swing frequency as a second reference reference, and demodulate the fundamental frequency component and the second harmonic component based on the first reference reference and the second reference reference, respectively.

[0010] Preferably, the dynamically changing oscillation frequency is a frequency chirp signal that performs periodic linear scanning within a preset frequency range.

[0011] Preferably, the system further includes a process diagnosis module, which is used to receive one-dimensional time-varying intensity signals in parallel and perform statistical analysis on the one-dimensional time-varying intensity signals to obtain statistical characteristic parameters, including the mean and variance of the one-dimensional time-varying intensity signals; the process diagnosis module is used to compare the statistical characteristic parameters with preset process thresholds to determine whether there is a welding process failure caused by excessive workpiece gap.

[0012] Preferably, the process diagnostic module is used to filter out the preset oscillation frequency and its harmonic components by using a band-stop filter before performing statistical analysis on the one-dimensional time-varying intensity signal. The statistical characteristic parameter is the signal variance of the one-dimensional time-varying intensity signal within a preset time window.

[0013] Preferably, the process diagnostic module is used to generate an alarm signal when a welding process failure is determined to exist. The alarm signal is used to stop the operation of the normalization compensation control module.

[0014] Preferably, the system is used to pre-execute the excitation module, the perception dimensionality reduction module, and the dual-channel phase-locked demodulation module under cold conditions before the welding laser is turned on, in order to obtain a fundamental frequency artifact component. The fundamental frequency artifact component is the baseline deviation of the system optical artifact caused by the periodic oscillation modulation itself. The normalization compensation control module is used to: after the welding laser is turned on, subtract the fundamental frequency artifact component from the fundamental frequency component demodulated by the dual-channel phase-locked demodulation module to obtain the true fundamental frequency component, and calculate the normalized trajectory deviation based on the ratio of the true fundamental frequency component to the second harmonic component.

[0015] Preferably, the coaxial vision sensor is a standard coaxial camera, and the system uses the phase-locked logic of the dual-channel phase-locked demodulation module to suppress plasma light and splash noise.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This scheme establishes a process modulation-based sensing method, which utilizes the periodic oscillation of the existing scanning galvanometer in the laser welding system as an active excitation, and reduces the multidimensional image information acquired by the coaxial vision sensor to a one-dimensional time-varying intensity signal for processing. The system obtains the trajectory deviation by phase-locked demodulation of a specific frequency component in the one-dimensional signal, avoiding the complex two-dimensional image processing and feature recognition links in traditional vision schemes. The operation of its positioning mechanism does not depend on a clear image of the weld geometry.

[0017] 2. The sensing and demodulation mechanism adopted in this solution makes it resistant to optical interference during the welding process. The system locks onto the deviation signal that is strictly synchronized with the active excitation frequency. The optical noise generated by the high-intensity plasma or random spatter during welding usually has no correlation with the specific excitation frequency. Therefore, it is regarded as an uncorrelated signal and suppressed during the phase-locked demodulation process, thus ensuring the stability of trajectory compensation under strong interference conditions.

[0018] 3. The real-time compensation mechanism of this solution enables it to correct dynamic trajectory deviations caused by workpiece tolerances, assembly gaps, or welding thermal deformation online. This capability reduces the dependence of welding quality on the absolute precision provided by preceding processes such as component processing and tooling clamping, and provides a feasible way to solve the technical and economic contradiction between high-quality output and high-precision preceding costs that is common in automated welding. Attached Figure Description

[0019] Figure 1 This is a functional block diagram of the robot welding trajectory normalization compensation system of the present invention; Figure 2 This is a failure data graph of the control system of the present invention without normalization processing; Figure 3 This diagram illustrates the technical challenges faced by the robot welding trajectory compensation of this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are only some embodiments of this invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] The online visual monitoring and compensation system for robot welding trajectory disclosed in this invention, as an online control system applied to robot laser welding heads, comprises four core modules working in tandem: an excitation module, a perception dimensionality reduction module, a dual-channel phase-locked demodulation module, and a normalization compensation control module. The system's operating logic involves the excitation module actively applying a known periodic oscillation modulation to the laser spot; the perception dimensionality reduction module using a coaxial vision sensor to capture the optical echo when this modulation interacts with the workpiece, reducing it from high-dimensional visual flow data to a one-dimensional time-varying intensity signal; the dual-channel phase-locked demodulation module using the phase-locked amplification principle to demodulate in parallel from this one-dimensional signal the fundamental frequency component representing geometric deviation and the second harmonic component representing optical response gain; finally, the normalization compensation control module calculates the ratio of these two components to obtain a normalized trajectory deviation unaffected by process gain drift, and generates a compensation signal based on this, feeding it back to the scanning galvanometer to adjust the oscillation center point. Real-time closed-loop tracking of the weld is achieved. In a specific application scenario, taking laser welding of power battery shells or automotive body-in-white as an example, the system faces challenges not only in the initial trajectory deviation caused by workpiece material tolerances or poor tooling fixture positioning, but more importantly, in the welding process, high energy input inevitably leads to dynamic thermal deformation of the workpiece. Simultaneously, welding different batches of materials or adjusting the laser power can cause drastic changes in the optical properties of the welding area, i.e., optical response gain drift. This system is designed to address these dynamic and variable welding conditions. The excitation module, typically integrated into the digital controller of the laser welding head in engineering implementation, sends control commands to the scanning galvanometer of the welding head. These commands force the scanning galvanometer to superimpose a high-frequency, small-amplitude periodic oscillation modulation on the laser focus while the robot body executes the main welding trajectory. In a preferred embodiment, this oscillation modulation is set to a linear oscillation perpendicular to the actual weld direction, with a preset oscillation frequency... , is a known value of this system, the frequency can be set to 800Hz, and its swing amplitude is usually set in the range of 0.1mm to 0.4mm.

[0022] The perception dimensionality reduction module connects to a coaxial vision sensor at its input. In a preferred embodiment, this sensor is a standard coaxial camera used to acquire real-time 2D visual flow data of the area where the laser spot interacts with the workpiece. During the welding process, this visual flow data is filled with strong noise generated by plasma light, metal spatter, etc. To avoid complex and easily interfered 2D image recognition, the core task of this module is to reduce the dimensionality of the information. The specific implementation path is as follows: in each frame of the acquired 2D image, a preset region of interest (ROI) covering the laser point of action and the molten pool is locked, and all pixel values ​​within the ROI are averaged, i.e., the average brightness is calculated, thereby compressing the image information of that frame into a scalar value. With continuous acquisition of the image stream, the module finally outputs a one-dimensional time-varying intensity signal. The sampling rate of this signal is consistent with the frame rate of the coaxial camera; the dual-channel phase-locked demodulation module is the core of the system's signal processing, and its function is to demodulate signals from noise-inundated areas. In a preferred embodiment, the precise extraction of specific frequency components from the signal is achieved by executing a digital lock-in amplifier algorithm. This algorithm utilizes the correlation between the signal and a known reference frequency to suppress all unrelated noise. The module performs two demodulation operations in parallel: the first demodulation channel receives... The signal is given at a preset oscillation frequency set by the excitation module. This frequency can be set to 800Hz as its primary reference; this channel is used for demodulation. The frequency in the signal is strictly equal to The physical meaning of the fundamental frequency component is that... amplitude The optical response is proportional to the magnitude of the geometric deviation of the oscillation center point from the actual weld center, while its phase indicates the direction of the deviation; when the reference is centered, the optical response is symmetrical. Component amplitude Approaching zero, the second demodulation channel also receives... The signal, but at twice the preset oscillation frequency. As its second reference benchmark, When the frequency is 800Hz, this double harmonic is 1600Hz; this channel is used for demodulation. The frequency in the signal is strictly equal to The component; its physical meaning is that when the swing is symmetrically centered, the optical response reaches its peak on both sides of the swing, making the main frequency of the signal appear as Therefore, this second harmonic component The amplitude represents the overall optical response intensity of the welding process, i.e., the optical response gain.

[0023] The normalization compensation control module is used to solve problems that depend solely on the amplitude of the fundamental frequency component. When performing control, it is subject to optical response gain, i.e., the amplitude of the second harmonic component. This module receives the fundamental frequency component amplitude from the first demodulation channel output of the dual-channel phase-locked demodulation module. Second harmonic component amplitude of the second demodulation channel output Its core procedure is to perform a normalization calculation, that is, to calculate a normalized trajectory deviation based on the ratio of the fundamental frequency component to the second harmonic component. In a preferred embodiment, the calculation rule follows: ,because and Both originate from the same optical process. When changes in operating conditions (such as increased material reflectivity) lead to an increase in overall gain, both will increase proportionally, and their ratio will be... Then it remains stable, thus This module becomes a robust control variable that is only related to geometric deviation, based on... The amplitude and phase direction of the fundamental frequency component are used to generate a compensation signal, which is a quasi-DC or low-frequency correction signal. This signal is fed back to drive the scanning galvanometer to adjust the center point of its periodic oscillation modulation in real time. The control objective is to always keep the amplitude and phase direction of the fundamental frequency component in the correct position. Drive it to near zero, that is, reduce the amplitude of the fundamental frequency component. The system drives the laser beam to near zero, thus achieving adaptive and stable tracking of the weld seam. To eliminate systematic optical artifact interference unrelated to the welding process that may be caused by the movement of the scanning galvanometer itself on the coaxial optical path, this system also performs a pre-weld calibration. This procedure is performed under cold conditions before the welding laser is turned on, when the laser spot has not physically interacted with the workpiece. The system pre-executes the first demodulation channel of the excitation module, the perception dimensionality reduction module, and the dual-channel phase-locked demodulation module. Any non-zero signal demodulated at this time is regarded as the fundamental frequency artifact component caused by the oscillation modulation itself, i.e., the system optical artifact baseline deviation. This fundamental frequency artifact component is stored. After the welding laser is turned on, the normalization compensation control module is used to extract the fundamental frequency component demodulated in real time. In the process, the stored fundamental frequency artifact component is subtracted to obtain the true fundamental frequency component; finally, the normalized trajectory deviation is calculated based on the ratio of the true fundamental frequency component to the second harmonic component.

[0024] Considering certain welding processes, taking the welding of galvanized sheets as an example, it is possible to generate welding processes with specific frequencies. Strong periodic process noise, if and If they get too close, they will interfere with phase-locked demodulation. Therefore, the excitation module is used to operate at a dynamically changing oscillation frequency. A periodic oscillation modulation is applied; in a preferred embodiment, the dynamically changing oscillation frequency is a frequency chirp signal that performs a periodic linear scan within a preset frequency range, which can be set to 1000Hz to 1100Hz; correspondingly, a dual-channel phase-locked demodulation module is used to synchronously acquire the dynamically changing frequency. As the first reference benchmark, its double frequency is acquired simultaneously. As a second reference standard, and based on these two real-time reference standards, the fundamental frequency component and the second harmonic component are demodulated respectively; due to process noise The frequency is fixed, while the demodulation reference of this system is moving, and this noise... Unable to be synchronously locked, thus effectively suppressed, ensuring the system's robustness under strong periodic process noise interference; furthermore, to address the issue that the main scheme can only track the geometric center but cannot determine the process state or whether the gap is too large, this system, in a preferred embodiment, also includes a process diagnosis module. This process diagnosis module is used to receive one-dimensional time-varying intensity signals from the sensing and dimensionality reduction module in parallel. The analysis logic of this module complements that of the main scheme: it filters out the preset oscillation frequency using a band-stop filter. and its second harmonic components This is because these are tracking signals, not process status signals; this module filters out the excitation signals... The signal undergoes statistical analysis, and it mainly contains residual DC components and broadband noise components; the statistical characteristic parameters preferably include... The signal mean, which characterizes the average brightness of the molten pool, is compared with... The signal variance within a preset time window characterizes the intensity of molten pool fluctuations, i.e., broadband noise. The physical principle is that a healthy molten pool has a high mean and high variance, while when the workpiece gap is too large, causing a weld void, the molten pool collapses, the keyhole disappears, and both the signal mean and variance decrease. Therefore, this process diagnostic module compares the acquired statistical characteristic parameters, taking variance as an example, with a preset process threshold to determine if there is a welding process failure due to excessive workpiece gap. In a preferred embodiment, when a welding process failure is determined to exist, this module generates an alarm signal. This alarm signal is used to stop the operation of the normalization compensation control module or stop laser emission and robot movement, thereby avoiding continued invalid tracking on unweldable gaps.

[0025] Example 1: This example illustrates the disclosed technical solution in a specific operational scenario of an intelligent welding system with multivariate interference. In a typical application within a field, such as the automated welding of a complex automotive structural component assembled from high-reflectivity aluminum alloy and low-reflectivity coated steel plates, the system faces a complex set of technical challenges: the workpiece has initial geometric tolerances during assembly, and during welding, high energy input causes dynamic thermal deformation of the entire structural component, requiring the system to possess real-time trajectory tracking capabilities; when the welding trajectory transitions from the aluminum alloy region to the coated steel plate region, the significant differences in the reflectivity, absorptivity, and plasma morphology of the two materials cause a dramatic jump in the overall signal intensity acquired by the coaxial vision sensor—i.e., the optical response gain—several times. Under these conditions, a signal relying solely on the fundamental frequency component... Using amplitude to determine deviation in control methods presents challenges. For example, with a constant geometric deviation of 0.2mm, in the low-gain aluminum alloy region, the system might demodulate an amplitude of... of The signal, but when entering the high-gain coated steel plate area, the same 0.2mm deviation will produce an amplitude of of The signal; this gain drift leads to When a stable quantization relationship is lost between the signal and the actual geometric deviation, the controller may experience severe overshoot or misjudgment, leading to unstable tracking control. In this scenario, the system of this invention demonstrates a cooperative suppression capability against multivariate disturbances through its internal mechanism. After system startup, the excitation module swings at a preset frequency... The scanning galvanometer is driven, and the sensing and dimensionality reduction module acquires a one-dimensional time-varying intensity signal. When the weld joint is located in the aluminum alloy region (low gain) and there is a deviation of 0.2mm, the dual-channel phase-locked demodulation module works in parallel: the first demodulation channel demodulates the fundamental frequency component with a smaller amplitude. Meanwhile, the second demodulation channel demodulates a second harmonic component with a similarly small amplitude. ,Should The magnitude objectively characterizes the low optical response gain under the current operating condition; the normalized compensation control module then calculates the normalized trajectory deviation. .

[0026] When the welding trajectory moves to the high-gain area of ​​the coated steel plate, although the geometric deviation remains at 0.2mm, the drastic change in gain causes the outputs of both channels to increase proportionally and significantly, becoming... and At this point, the trajectory deviation calculated by the normalized compensation control module is: ;because and Both originate from the same optical process, and their changes are proportional, resulting in The calculation results and The calculation results are highly consistent in numerical value. The system utilizes the second harmonic component in this way. As a calibration scale for real-time characterizing process gain, it is used for the fundamental frequency component characterizing geometric deviation. Performing periodic dynamic normalization processing, this and The coordinated operation and normalization logic between them resolve the locking contradiction between the geometric deviation signal and the optical gain drift, resulting in a normalized trajectory deviation output by the system. It is only related to the actual geometric deviation, and is not related to the drastic changes in gain caused by workpiece material, laser power, surface condition, etc.

[0027] Example 2: This example provides experimental data to verify the online visual monitoring and compensation system for the robot welding trajectory of the present invention. When the welding conditions change drastically, it utilizes a dual-channel phase-locked demodulation module and a normalized compensation control module to work together to achieve the effectiveness of adaptive and stable tracking of the working conditions. The experimental platform is built on a robotic laser welding workstation, which includes a six-axis robot, a laser welding head containing a scanning galvanometer, and a standard coaxial vision sensor integrated into the optical path of the welding head. The image acquisition frame rate of the coaxial vision sensor is set to 4000Hz. The control system used in the experiment is the system of the present invention, and its excitation module is set to a preset oscillation frequency. The frequency was 800Hz, and the swing amplitude was 0.3mm. The test object was a specially made flat workpiece spliced ​​from two materials: a high-reflectivity 6061 aluminum alloy on the left and a low-reflectivity SPCC steel plate with a black coating on the right. The two were welded flat to form a clear geometric boundary, which was the actual weld seam in the test. The purpose of the test was to verify the stability of the normalization mechanism of the system of the present invention compared with the traditional single-channel mechanism when the optical response gain undergoes a step change. For this purpose, two sets of tests were set up: one was a control group, whose control system only used the fundamental frequency component obtained by the first demodulation channel. The amplitude is used as a representation of the trajectory deviation, that is... The other group is the experimental group of this invention, whose control system adopts a normalized compensation control module based on the fundamental frequency component. With second harmonic components The ratio is used to calculate the normalized trajectory deviation, i.e. During the experiment, a constant physical deviation of 0.2 mm was artificially introduced, meaning that the trajectory followed by the robot carrying the welding head was always parallel to the actual weld boundary of the workpiece, with an offset of 0.2 mm. Welding was started, and the welding head was moved at a speed of 5 m / min from the low-gain region of the high-reflectivity aluminum alloy area to the high-gain region of the low-reflectivity coated steel plate area. The demodulated component signals of the two systems and the final calculated trajectory deviation were continuously recorded. Before and after the welding head crossed the material boundary, the system collected and recorded the instantaneous amplitude of the key signals, as shown in Table 1.

[0028] Table 1: Comparative Experimental Data of Signal Response Across Material Boundaries Experimental data show that in the low-gain region of the aluminum alloy, due to a physical deviation of 0.2mm, both groups demodulated a fundamental frequency component of 1.12. and the second harmonic component of 5.45. At this time, the output of the control group was 1.12, and the output of the experimental group of this invention was... for When the welding head enters the high-gain region of the coated steel plate, although the physical deviation remains at 0.2mm, the optical response gain increases dramatically, leading to... Increased to 4.38, The output of the control group increased accordingly to 21.30; at this point, the output of the control group... (Right now As the gain increases to 4.38, this value amplifies the physical deviation. Compared to 1.12 in the low-gain region, the deviation reading changes by approximately 291%, which will lead to overshoot in closed-loop control. In contrast, the output of the experimental group of this invention... for The calculation result is basically consistent with 0.206 in the low-gain region; experimental data show that the dual-channel phase-locked demodulation and normalization compensation control mechanism adopted in this invention utilizes the second harmonic component. As a real-time characterization of the process optical response gain, and for the fundamental frequency component Normalization was performed, and the resulting normalized trajectory deviation was obtained. This avoids gain drift interference caused by changes in the reflectivity of the workpiece material.

[0029] Example 3: To further verify the necessity of the normalization mechanism of the present invention from a reverse perspective, this example is set as a comparative example. The test platform, workpiece material, and welding process parameters used in this example include a robotic welding workstation, an aluminum alloy-steel plate spliced ​​workpiece, and a preset oscillation frequency of 800Hz. The welding speed of 5 m / min is consistent with that in Example 2. This example uses a control system that does not include the second demodulation channel in the dual-channel phase-locked demodulation module of this invention, nor does it include the normalization compensation control module. This control system represents a conventional single-channel tracking method, that is, using only the fundamental frequency component demodulated by the first demodulation channel. As the only trajectory deviation and the Signal( A standard proportional-integral (PI) closed-loop controller was directly input to drive the scanning galvanometer for trajectory compensation. The test process was used to simulate the actual response of the control system when it encountered the step condition of Example 2 under closed-loop control. At the beginning of the test, the welding head was aligned with the actual weld with a physical deviation of ~0mm in the low-gain region of the aluminum alloy area and tracked it stably. The welding head was made to cross the boundary at a uniform speed of 5m / min and enter the high-gain region of the coated steel plate area. The key state parameters of its closed-loop control process were continuously recorded, as shown in Table 2.

[0030] Table 2: Closed-loop response data of the comparison system at gain step. Experimental data shows that at time T0, the control system can stably track the actual weld in the aluminum alloy region (low gain region), where a tiny physical deviation of 0.01 mm will cause... The signal (0.05) is close to zero; at time T1, the weld joint crosses the material boundary into the high-gain region, although the physical deviation remains at 0.01 mm. The signal jumps instantaneously to 0.20 due to the step gain of the optical response. The PI controller of the control system receives this spurious deviation signal, amplified approximately fourfold, at time T2 and misinterprets it as a large geometric deviation. Consequently, at time T3, it outputs a strong reverse compensation action, causing the scanning galvanometer to violently push the laser spot to the other side of the weld, resulting in a physical overcorrection of -0.15 mm. This new, reverse physical deviation is then amplified by the high gain, causing the system to fall into continuous uncontrolled oscillations at times T4 and T5, unable to establish a stable molten pool at the weld center. The experimental results of this embodiment indicate that the lack of a second harmonic component... A single-channel control system that undergoes normalization cannot simultaneously adapt its control loop parameters to two gain conditions. The system is designed to be unable to distinguish between the actual geometric deviation and the optical response gain drift of the process. When faced with a step change in the welding condition, its closed-loop control will fail due to erroneous signal amplification.

[0031] Example 4: This example combines Figures 1 to 3 Description of the online visual monitoring and compensation system for robot welding trajectory, such as... Figure 1 As shown, the system uses an excitation module to send control commands to a scanning galvanometer in the external hardware of the laser welding head to apply periodic oscillations. A coaxial vision sensor acquires visual flow data from the laser-affected area of ​​the workpiece and sends it to a perception dimensionality reduction module. This module extracts a one-dimensional time-varying intensity signal and sends it in parallel to a dual-channel phase-locked demodulation module and a process diagnostic module. The dual-channel phase-locked demodulation module demodulates the fundamental frequency component, representing geometric deviation, and the second harmonic component, representing optical gain, and sends both components to a normalization compensation control module. The normalization compensation control module calculates the normalized deviation based on the ratio of the two components and generates a compensation signal to feed back to the scanning galvanometer to adjust the oscillation center point. At the same time, the process diagnostic module statistically analyzes the mean and variance of the one-dimensional time-varying intensity signal and generates an alarm signal when a process failure is detected, thereby stopping the operation of the normalization compensation control module.

[0032] like Figure 2 As shown in the figure, the horizontal axis represents time points, from T0 to T5, and the vertical axis represents numerical values. The solid curve represents the physical deviation in mm, and the dashed curve represents the measured values ​​by the system. Data shows that, between times T1 and T2, although the physical deviation is close to zero, the measured... The spurious peaks generated by gain changes cause a severe reverse physical overshoot in the system at time T3, followed by sustained oscillations at times T4 and T5; for example... Figure 3 As shown, in robotic laser welding, the core dilemma is that trajectory deviation is difficult to compensate in real time and reliably due to the combined effects of multiple factors, including high cost upfront reliance on high-precision incoming parts, high-rigidity precision tooling and workpiece tolerances, dynamic thermal deformation of the workpiece caused by the heat input of the welding process, process noise interference such as strong plasma light, metal spatter and dust, as well as limitations of traditional vision such as reliance on cold-state geometric features, low algorithm reliability and high system complexity.

[0033] Example 5: This example discloses the key algorithm implementation path and parameter calibration procedure used in the engineering deployment of the system of the present invention to ensure the reproducibility and robustness of the system. The core function of the system of the present invention relies on the precise coordination of the perception dimensionality reduction module and the dual-channel phase-locked demodulation module, as well as the reasonable setting of two key parameters: preset oscillation frequency and process threshold. The perception dimensionality reduction module reduces 2D visual flow data to one-dimensional time-varying intensity signals. The specific operating procedure is as follows: at time Acquire a frame captured by the coaxial vision sensor Pixel image frame It locks a preset region of interest covering the laser's point of effect; the module then traverses all nodes within that region of interest. 1 pixel The total brightness is obtained by summing up the brightness values. Finally, by calculating the average value ,get One-dimensional time-varying intensity signal at time 10:00 The sampling point value; the dual-channel phase-locked demodulation module, taking its first demodulation channel as an example, is used to demodulate the fundamental frequency component. The core operation flow of the digital lock-in amplifier algorithm is as follows: the input is... The continuous sampling sequence and the predetermined preset swing frequency and sampling period During processing, the module generates two digital quadrature reference signals that are synchronized with the excitation module. and ; input signal Multiplying each signal by the two reference signals yields two mixed signals. and The two mixed signals are each passed through a digital low-pass filter to remove high-frequency components, resulting in two quasi-DC components. and Finally, through calculation The amplitude of the fundamental frequency component is output, which represents the magnitude of the deviation; the implementation path of the second demodulation channel is the same, except that its reference reference is replaced with... That is, to determine the preset oscillation frequency In a preferred embodiment, this system employs a systematic operating condition noise spectrum analysis procedure to determine... The value; this procedure is performed before actual welding, without applying oscillation modulation ( Under these conditions, the laser is activated and a stable molten pool is established on the workpiece to simulate real working conditions; at this time, the perception and dimension reduction module collects a background noise signal containing only interference such as plasma fluctuations and splashes. The system then... Perform a Fast Fourier Transform (FFT) to obtain the background noise spectrum under this operating condition. By analyzing this The system automatically identifies the main frequency bands where noise energy is concentrated, such as plasma characteristic frequencies. And within a usable engineering frequency range, such as 500Hz to 1500Hz, search for a frequency located at... The frequency of the spectral trough, that is, the frequency far from all major noise peaks, is defined as... This procedure ensures the excitation signal Works in a quiet window with minimal background noise.

[0034] To determine the process thresholds in the process diagnostic module, this system employs a statistical calibration procedure based on operating condition comparison. Specifically, in this implementation, a digital band-stop filter is configured, and its center frequency is automatically set to the value determined by the aforementioned procedure. This is used to filter out excitation signals; operators or automated program control systems perform calibration welding under two different working conditions: Condition 1, healthy weld pool, involves welding on standard butt workpieces with a gap of less than 0.1 mm, collecting the signal variance readings of 1000 filtered signals to form a healthy dataset. Condition 2: Melt pool collapse. On a workpiece with an artificially set excessively large gap (greater than 0.8 mm), no-welding is performed. Similarly, 1000 signal variance readings are collected to form a collapse dataset. The system then performs statistical analysis on these two sets of data, and based on a preset confidence level, such as ensuring that 99.9% of crash states can be identified, it analyzes... and Data distribution, such as calculating the mean. and standard deviation To determine a process threshold for distinguishing between the two process states. ,Should Can be set to This procedure provides a statistically based and reproducible judgment criterion for the process diagnostic module.

[0035] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0036] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An online visual monitoring and compensation system for robot welding trajectory, applied to a robot comprising a laser welding head, the laser welding head comprising a scanning galvanometer and a coaxial vision sensor, characterized in that, The system comprises: an excitation module configured to drive the scanning galvanometer to apply a periodic wobble modulation to the laser spot at a preset wobble frequency; a perception dimension reduction module configured to acquire, by the coaxial vision sensor, visual flow data of an interaction region between the laser spot and the workpiece, and extract a one-dimensional time-varying intensity signal from the visual flow data; a dual-channel phase-locked demodulation module configured to perform two-way demodulation operations on the one-dimensional time-varying intensity signal in parallel, wherein a first demodulation channel takes a preset wobble frequency as a first reference, demodulates the one-dimensional time-varying intensity signal to obtain a fundamental component representing a geometric deviation between a center point of the periodic wobble modulation and an actual weld seam; and a second demodulation channel takes a double frequency of the preset wobble frequency as a second reference, demodulates the one-dimensional time-varying intensity signal to obtain a second harmonic component representing an optical response gain of the interaction region; a normalization compensation control module configured to calculate a normalized trajectory deviation based on a ratio of the fundamental component to the second harmonic component, and generate a compensation signal based on the normalized trajectory deviation to drive the scanning galvanometer to adjust the center point of the periodic wobble modulation.

2. The system for online visual monitoring and compensation of robotic welding trajectories according to claim 1, characterized in that, The normalization compensation control module is configured to calculate a normalized trajectory deviation amount The calculation rule of the normalized trajectory deviation amount follows: ; wherein, is the amplitude of the fundamental component, is the amplitude of the second harmonic component.

3. The system for online visual monitoring and compensation of robotic welding trajectories of claim 1, wherein, The perception dimension reduction module is configured to perform averaging processing on a preset region of interest in the visual flow data to extract the one-dimensional time-varying intensity signal, the dual-channel phase-locked demodulation module is configured to perform a digital phase-locked amplifier algorithm, and the periodic wobble modulation is a linear wobble perpendicular to the actual weld seam.

4. The system for online visual monitoring and compensation of robotic welding trajectories of claim 1, wherein, The excitation module is configured to apply the periodic wobble modulation at a dynamically changing wobble frequency, and the dual-channel phase-locked demodulation module is configured to: synchronously acquire the dynamically changing wobble frequency as the first reference, synchronously acquire a double frequency of the dynamically changing wobble frequency as the second reference, and demodulate the fundamental component and the second harmonic component based on the first reference and the second reference, respectively.

5. The system for online visual monitoring and compensation of a robot welding trajectory according to claim 4, characterized in that, The dynamically changing wobble frequency is a frequency chirp signal performing a periodic linear scan in a preset frequency range.

6. The system for online visual monitoring and compensation of robotic welding trajectories of claim 1, wherein, The system further comprises a process diagnosis module configured to receive the one-dimensional time-varying intensity signal in parallel, and perform statistical analysis on the one-dimensional time-varying intensity signal to obtain statistical characteristic parameters, including a mean value of the one-dimensional time-varying intensity signal and a variance of the one-dimensional time-varying intensity signal. The process diagnosis module is configured to compare the statistical characteristic parameters with preset process thresholds to determine whether a welding process failure caused by an excessively large workpiece gap exists. The process diagnosis module is configured to filter out the preset wobble frequency and harmonic components thereof by a band-stop filter before performing the statistical analysis on the one-dimensional time-varying intensity signal, and the statistical characteristic parameters are signal variances of the one-dimensional time-varying intensity signal in a preset time window.

7. The system for online visual monitoring and compensation of a robot welding trajectory according to claim 6, characterized in that, The process diagnosis module is configured to generate an alarm signal when the welding process failure is determined to exist, and the alarm signal is configured to suspend the operation of the normalization compensation control module.

8. The system for online visual monitoring and compensation of robotic welding trajectories according to claim 6, wherein, ​ 9. The system for online visual monitoring and compensation of robotic welding trajectories of claim 1, wherein, The system is used to execute the excitation module, the perception dimension reduction module and the double-channel phase-locked demodulation module in advance under the cold state before the welding laser is turned on, so as to obtain a fundamental frequency artifact component, the fundamental frequency artifact component is a system optical artifact baseline deviation amount caused by the periodic swing modulation itself, and the normalization compensation control module is used to: after the welding laser is turned on, the fundamental frequency artifact component is subtracted from the fundamental frequency component demodulated from the double-channel phase-locked demodulation module, so as to obtain a true fundamental frequency component, and based on the ratio of the true fundamental frequency component to the second harmonic component, the normalized trajectory deviation amount is calculated.

10. The system for online visual monitoring and compensation of robotic welding trajectories of claim 1, wherein, The coaxial vision sensor is a standard coaxial camera.

Images