Welding robot welding path optimization method and system based on data driving

Abstract

The invention relates to the technical field of automatic operation, in particular to a welding robot welding path optimization method and system based on data driving. The method comprises the following steps: presetting a heat-machine coupling model and setting an initial sensitivity coefficient; welding current and arc voltage are collected in real time in the welding process, instantaneous thermal power is calculated, time integration is carried out, and an accumulated enthalpy value is obtained; inputting the accumulated enthalpy value into a heat-machine coupling model to obtain a theoretical deformation predicted value, and calculating an actually measured deformation estimated value based on a difference value between the arc voltage and the reference voltage; constructing a sliding time window, calculating a residual sequence, and integrating to obtain a drift characteristic value; and updating the sensitivity coefficient based on the drift characteristic value, and correcting the reference position coordinates of the welding robot by using the deformation compensation amount to generate a welding gun position compensation instruction. According to the method, submillimeter-level dynamic compensation of thermal deformation of the workpiece is achieved through a sensitivity self-adaptive correction mechanism of residual integration, and the tracking precision and consistency of a welding path are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of automated operation technology. More specifically, this application relates to a data-driven welding robot welding path optimization method and system. Background Technology

[0002] As the core pressure-bearing component of a water heater, the welding quality of the inner tank directly affects the product's safety and lifespan. Welding of the inner tank typically involves long circumferential seams at the end caps and longitudinal seams in the cylinder, using materials such as thin enamel steel or stainless steel sheets with wall thicknesses ranging from 1.0 mm to 3.0 mm. In actual production, welding robots generally perform welding operations according to a pre-taught fixed trajectory. However, due to the significant heat accumulation effect of thin-walled materials under continuous welding heat input, the workpiece undergoes nonlinear radial expansion and axial contraction, causing the actual weld trajectory to deviate in real time from the taught trajectory.

[0003] Existing welding path optimization technologies mainly fall into two categories. The first category is offline planning methods based on intelligent algorithms, such as using deep reinforcement learning algorithms to plan the path between weld endpoints and then optimizing the global point sequence using ant colony optimization. This type of method reduces the reliance on precise mathematical models, but its core is a deep learning black-box model, lacking interpretability in industrial settings and failing to dynamically and adaptively adjust for real-time path deviations caused by thermal deformation during welding. The second category is real-time tracking methods based on laser or vision sensors, which use laser sensors mounted at the welding torch tip to scan the weld morphology in real time and provide feedback on deviations for correction. However, this type of method is a hysteresis-based adjustment, heavily influenced by arc light and spatter interference, and cannot utilize historical welding data to achieve global predictive optimization of the path.

[0004] In summary, existing technologies have the following shortcomings: perception and optimization are disconnected; existing sensing devices can only passively correct current deviations and cannot actively predict path change trends; massive historical data such as current and voltage generated during welding are not converted into knowledge to improve path accuracy, causing the device to be unable to evolve with the increase of production batches; and nonlinear deformation caused by heat accumulation in the welding of thin-walled containers is difficult to compensate for at the sub-millimeter level using traditional geometric modeling methods. Summary of the Invention

[0005] The purpose of this application is to propose a data-driven welding robot welding path optimization method and system to solve the problem of insufficient welding path tracking accuracy in the prior art.

[0006] In the first aspect, the data-driven welding robot welding path optimization method provided in this application includes: pre-setting a thermo-mechanical coupling model, wherein the thermo-mechanical coupling model defines a nonlinear mapping relationship between unit cumulative heat input and workpiece surface deformation, and sets an initial sensitivity coefficient; during the welding process, welding current and arc voltage are collected in real time, and the product of the two is calculated to obtain instantaneous heat power, and the instantaneous heat power is integrated over time to obtain a cumulative enthalpy value; the cumulative enthalpy value is used as the input of the thermo-mechanical coupling model to obtain a theoretical deformation prediction value; simultaneously, the measured deformation estimation value is calculated based on the difference between the arc voltage and the reference voltage; a sliding time window is constructed, and the difference between the measured deformation estimation value and the corrected theoretical deformation prediction value is calculated within the sliding time window to obtain a residual sequence, and the residual sequence is integrated to obtain a drift characteristic value; the sensitivity coefficient is updated based on the drift characteristic value; the updated sensitivity coefficient is multiplied by the theoretical deformation prediction value to obtain a deformation compensation amount, and the deformation compensation amount is used to correct the reference position coordinates of the welding robot, generate a welding torch position compensation command, and drive the welding robot to perform path correction.

[0007] This application establishes a deformation prediction benchmark at the physical mechanism level through a pre-set thermo-mechanical coupling model, and simultaneously acquires theoretical deformation prediction values ​​based on energy integration and measured deformation estimation values ​​based on arc voltage during the welding process. The drift feature value is extracted by the sliding window integration of the residuals of the two to dynamically update the sensitivity coefficient, thereby combining the stability of the physical model with the adaptive capability of data-driven model, realizing sub-millimeter-level dynamic compensation for nonlinear thermal deformation during the welding process of thin-walled containers, and significantly improving the tracking accuracy and consistency of the welding path.

[0008] Optionally, the process of establishing the thermo-mechanical coupling model includes: obtaining the thermophysical and mechanical property parameters of the workpiece material; establishing an initial mapping function between heat input and deformation based on the heat conduction equation and thermoelastic constitutive relation; collecting measured deformation data under different heat input conditions through offline calibration experiments, fitting and correcting the parameters of the initial mapping function, and generating the thermo-mechanical coupling model.

[0009] This application obtains the thermophysical and mechanical properties of the workpiece material, establishes an initial mapping function based on the heat conduction equation and thermoelastic constitutive relation, and corrects the parameters through offline calibration experiments. This enables the thermo-mechanical coupling model to accurately reflect the nonlinear correspondence between heat input and deformation under specific material and tooling conditions, providing a reliable physical prediction benchmark for subsequent online compensation.

[0010] Optionally, the calculation process of the cumulative enthalpy value includes: reading the output values ​​of the welding current sensor and the arc voltage sensor at a preset sampling period; calculating the product of the welding current and the arc voltage to obtain the instantaneous thermal power; and performing local time integration on the instantaneous thermal power within a preset sliding time window to obtain the cumulative enthalpy value.

[0011] Optionally, the step of obtaining the theoretical deformation prediction value includes: inputting the cumulative enthalpy value into the input terminal of the thermo-mechanical coupling model; the thermo-mechanical coupling model outputs the theoretical deformation prediction value corresponding to the cumulative enthalpy value through interpolation or table lookup operations based on a pre-stored nonlinear mapping curve.

[0012] Optionally, the step of obtaining the measured deformation estimation value includes: obtaining the current arc voltage value and a preset reference voltage value; calculating the difference between the arc voltage value and the reference voltage value to obtain a voltage deviation; converting the voltage deviation into the arc length change from the welding torch tip to the workpiece surface based on the correspondence between arc length and voltage in the static characteristics of the arc; and using the absolute value of the arc length change as the measured deformation estimation value of the current workpiece surface.

[0013] This application converts voltage deviation into arc length change based on the correspondence between arc length and voltage in the static characteristics of the arc. It uses the welding arc itself as an indirect measurement method to obtain real-time deformation information of the workpiece surface, eliminating the need for additional contact displacement sensors. This reduces hardware costs while avoiding interference from arc light and spatter on optical sensors.

[0014] Optionally, the correction of the theoretical deformation prediction value includes: obtaining the sensitivity coefficient at the current moment; multiplying the theoretical deformation prediction value output by the thermo-mechanical coupling model with the sensitivity coefficient to obtain the corrected theoretical deformation prediction value.

[0015] Optionally, the sensitivity coefficient update process includes: multiplying the drift feature value by a preset adaptive learning rate to obtain the adjustment increment; adding the current sensitivity coefficient to the adjustment increment to obtain the updated sensitivity coefficient; the adaptive learning rate is dynamically adjusted according to the absolute value of the drift feature value, increasing the adaptive learning rate when the absolute value exceeds a preset threshold, and decreasing the adaptive learning rate when the absolute value is lower than a preset threshold.

[0016] This application achieves an automatic balance between compensation response speed and stability by dynamically adjusting the adaptive learning rate based on the absolute value of the drift characteristic value. When the deviation is large, the adjustment speed of the sensitivity coefficient is accelerated to quickly track changes in operating conditions, while the adjustment speed is slowed down when the deviation is small to avoid overcorrection and the introduction of oscillations.

[0017] Optionally, the generation of the welding torch position compensation command includes: reading the current reference position coordinates of the welding robot; multiplying the updated sensitivity coefficient by the theoretical deformation prediction value to obtain the deformation compensation amount; and adding the height direction component of the reference position coordinates to the deformation compensation amount to generate the target position coordinates in the height direction as the welding torch position compensation command.

[0018] Optionally, the optimization method further includes a safety protection step: setting an upper limit and a lower limit for the safety range of the sensitivity coefficient; when the updated sensitivity coefficient exceeds the safety range, it is determined to be an abnormal physical deformation state; under the abnormal physical deformation state, an abnormal alarm signal is output and the welding robot is switched to a degraded operation mode or an emergency shutdown is triggered.

[0019] This application, by setting a safe range for the sensitivity coefficient and triggering abnormal alarms and degraded operation when the limit is exceeded, can effectively identify abnormal working conditions such as sensor failure or workpiece perforation, prevent erroneous compensation commands from causing welding quality deterioration or equipment damage, and ensure the safety of industrial production.

[0020] In the second aspect, a data-driven welding robot welding path optimization system includes: processor; The memory stores computer instructions for a data-driven welding robot welding path optimization system, which, when executed by the processor, cause the system to perform the aforementioned data-driven welding robot welding path optimization method.

[0021] The beneficial effects of this application are as follows: This application establishes a deformation prediction benchmark at the physical mechanism level through a pre-set thermo-mechanical coupling model, and simultaneously obtains theoretical deformation prediction values ​​based on energy integration and measured deformation estimation values ​​based on arc voltage during the welding process. The drift feature value is extracted by the sliding window integral of the residual between the two to dynamically update the sensitivity coefficient, thereby combining the stability of the physical model with the adaptive capability of data-driven model, realizing sub-millimeter-level dynamic compensation for nonlinear thermal deformation during the welding process of thin-walled containers, and significantly improving the tracking accuracy and consistency of the welding path. Attached Figure Description

[0022] Figure 1 This is a flowchart of a data-driven welding robot welding path optimization method according to an embodiment of this application.

[0023] Figure 2 This is a key parameter monitoring diagram of a data-driven welding robot welding path optimization method according to an embodiment of this application.

[0024] Figure 3This is a deformation prediction and adaptive correction diagram of a data-driven welding robot welding path optimization method according to an embodiment of this application.

[0025] Figure 4 This is a comparison chart of the trajectory optimization effects of the data-driven welding robot welding path optimization method according to the embodiments of this application.

[0026] Figure 5 This is a structural block diagram of a data-driven welding robot welding path optimization system according to an embodiment of this application. Detailed Implementation

[0027] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Figure 1 The diagram shown is a flowchart of a data-driven welding robot welding path optimization method according to an embodiment of this application.

[0028] S1: Pre-set thermo-mechanical coupling model and collect welding parameters in real time to calculate cumulative enthalpy value.

[0029] First, a thermo-mechanical coupling model is pre-set, and welding parameters are collected in real time during the welding process to calculate the cumulative enthalpy. The thermo-mechanical coupling model defines a nonlinear mapping relationship between the unit cumulative heat input and the workpiece surface deformation, and is suitable for welding cylindrical thin-walled containers such as water heater inner tanks. In this embodiment, the workpiece material is a 1.2 mm thick enamel-lined steel plate, and the welding method is circumferential welding.

[0030] In establishing the thermo-mechanical coupling model, the thermophysical and mechanical properties of the workpiece material are first obtained. The thermophysical properties include the material's thermal conductivity, specific heat capacity, and density, while the mechanical properties include the elastic modulus and coefficient of thermal expansion. An initial mapping function between heat input and deformation is established based on the heat conduction equation and thermoelastic constitutive relation. This initial mapping function describes how, under given boundary conditions, the energy accumulated by the welding heat source forms a temperature field distribution within the workpiece through heat conduction, leading to thermal stress caused by the temperature gradient and ultimately resulting in measurable deformation on the workpiece surface. Measured deformation data under different heat input conditions are collected through offline calibration experiments. The parameters of the initial mapping function are fitted and corrected using the nonlinear least squares method to generate a calibrated thermo-mechanical coupling model. The initial sensitivity coefficient is set to 1.0, indicating that the proportional relationship between the model output and the actual deformation is one-to-one in the initial state.

[0031] After the welding process starts, the output values ​​of the welding current sensor and the arc voltage sensor are read at a preset sampling period. In this embodiment, the sampling period is set to 1 millisecond. The welding current sensor uses a Hall effect current transformer with a range of 0 to 500 amperes; the arc voltage sensor uses a differential voltage acquisition module with a range of 0 to 50 volts. The instantaneous thermal power is obtained by calculating the product of the welding current and the arc voltage.

[0032] The formula for calculating the cumulative enthalpy is as follows: ; in, This represents the cumulative enthalpy from the start of welding to the current time t, expressed in joules. This represents the welding current value collected at time τ, in amperes. The value of the arc voltage collected at time τ is expressed in volts; τ represents the integral variable, expressed in seconds; t represents the current time, expressed in seconds. This represents the preset sliding time window length, in seconds. In the digital implementation, the above continuous local integration is discretized using the trapezoidal numerical integration method. The preset sliding time window is maintained within each sampling period. The product of the current instantaneous thermal power and the sampling interval is added to the time window, and data outside the time window is discarded to achieve local time integration.

[0033] like Figure 2 The figure shows a key parameter monitoring diagram of the data-driven welding robot welding path optimization method according to an embodiment of this application. The high-frequency fluctuating curves in the figure represent the real-time acquired welding current and arc voltage, reflecting transient changes during the welding process. The figure also shows the result of local time integration of the instantaneous heat power obtained by multiplying the welding current and arc voltage within a preset sliding time window, i.e., the cumulative enthalpy value. This processing transforms the discrete electrical signal into a stable physical characteristic characterizing the heat input, providing crucial energy input data for the thermo-mechanical coupling model to output theoretical deformation prediction values.

[0034] S2: Theoretical deformation prediction and measured deformation estimation values ​​are obtained based on the thermo-mechanical coupling model and arc voltage.

[0035] After obtaining the cumulative enthalpy value, two parallel deformation estimation channels are executed simultaneously to obtain theoretical deformation prediction values ​​and measured deformation estimation values. These two channels independently estimate the workpiece deformation from the energy input dimension and the geometric observation dimension, respectively, providing orthogonal data sources for subsequent drift feature extraction.

[0036] The first channel is the theoretical deformation prediction channel. The current cumulative enthalpy value is input to the input of the thermo-mechanical coupling model. Based on a pre-stored nonlinear mapping curve, the thermo-mechanical coupling model outputs the theoretical deformation prediction value corresponding to the current cumulative enthalpy value through interpolation or table lookup operations. In this embodiment, the nonlinear mapping curve is stored in the controller's lookup table in the form of discrete data points, and intermediate values ​​are calculated using cubic spline interpolation. The theoretical deformation prediction value characterizes the deformation that the workpiece surface should produce according to the standard physical model under the current cumulative heat input conditions.

[0037] The second channel is for estimating the measured deformation. It acquires the current arc voltage value and a preset reference voltage value, and calculates the difference between the arc voltage value and the reference voltage value to obtain the voltage deviation. The reference voltage value is determined by averaging over the first few sampling periods after the welding arc stabilizes. In this embodiment, the arithmetic mean of the arc voltage within the first 500 milliseconds after welding arc ignition is used as the reference voltage. Based on the correspondence between arc length and voltage in the static characteristics of the arc, the voltage deviation is converted into the change in arc length from the welding torch tip to the workpiece surface. In gas metal arc welding (GMAW), the arc voltage and arc length are approximately linearly positively correlated; that is, when the workpiece surface bulges towards the welding torch due to thermal expansion, the arc length shortens, and the arc voltage decreases accordingly. The absolute value of this change in arc length is used as the estimated measured deformation value of the current workpiece surface.

[0038] The sensitivity coefficient at the current moment is obtained. The theoretical deformation prediction value output by the thermo-mechanical coupling model is multiplied by the sensitivity coefficient to obtain the corrected theoretical deformation prediction value. The introduction of the sensitivity coefficient allows the theoretical prediction value to be dynamically adjusted according to the actual working conditions, compensating for model deviations caused by factors such as batch differences in workpieces and wear of tooling fixtures.

[0039] S3: Construct a sliding time window to calculate the residual sequence and extract drift feature values.

[0040] After simultaneously acquiring theoretical deformation predictions and measured deformation estimates, a sliding time window is constructed to calculate the residual sequence and extract drift characteristics. The core purpose of this step is to extract the persistent model drift trend from the deviation between the measured deformation estimates and the corrected theoretical deformation predictions, and to filter out high-frequency noise caused by transient disturbances such as arc flash and splash impact.

[0041] A sliding time window with a preset time span is constructed. In this embodiment, the length of the sliding time window is set to 3 seconds, which includes data from 3000 sampling points. At each sampling moment, the difference between the current measured deformation estimate and the current corrected theoretical deformation prediction is calculated, and this difference is recorded as the residual value at the current moment. Within the sliding time window, the residual values ​​at all moments constitute a residual sequence.

[0042] The residual sequence within the sliding time window is integrated to extract drift feature values. The formula for calculating the drift feature values ​​is as follows: ; in, This represents the drift characteristic value at the current time t, in millimeters per second. This represents the estimated measured deformation at time τ, in millimeters. The sensitivity coefficient at time τ is dimensionless. This represents the theoretical deformation prediction value output by the thermo-mechanical coupling model at time τ, in millimeters. This represents the length of the sliding time window, in seconds. The significance of using integral calculations instead of instantaneous differences lies in the fact that integrals inherently possess low-pass filtering properties, capable of canceling out high-frequency residuals caused by instantaneous voltage jumps due to spatter impacting the arc during welding, retaining only the persistent deviation trend caused by changes in the overall physical properties of the workpiece. When the drift characteristic value continuously deviates from zero, it indicates that the sensitivity coefficient of the current thermo-mechanical coupling model can no longer accurately reflect the actual working conditions, requiring adaptive updates.

[0043] like Figure 3 The figure shows the deformation prediction and adaptive correction diagram of the data-driven welding robot welding path optimization method according to an embodiment of this application. This figure reveals the data-driven sensitivity coefficient update mechanism. The curves represent the theoretical deformation prediction value and the measured deformation estimation value, respectively. The dashed line illustrates the convergence process of the sensitivity coefficient over time. By constructing a sliding time window, the residual sequence of the two deformation values ​​is calculated, and the drift feature value is extracted, thereby dynamically updating the sensitivity coefficient. This mechanism ensures that the corrected theoretical deformation prediction value can accurately approximate the actual working condition, thus generating an accurate deformation compensation amount.

[0044] S4: Update the sensitivity coefficient based on the drift feature value and generate welding torch position compensation command to drive path correction.

[0045] After extracting the drift feature value, the sensitivity coefficient is updated based on the feature value and a welding gun position compensation command is generated to drive the welding robot to perform path correction.

[0046] The drift feature values ​​are input into the adaptive controller. The adaptive learning rate is dynamically adjusted based on the absolute value of the drift feature values: when the absolute value of the drift feature values ​​exceeds a preset threshold, it indicates a large model deviation and drastic changes in operating conditions; in this case, the adaptive learning rate is increased to accelerate the adjustment speed of the sensitivity coefficient. When the absolute value of the drift feature values ​​is below the preset threshold, it indicates a small model deviation and relatively stable operating conditions; in this case, the adaptive learning rate is decreased to avoid overcorrection and the introduction of unnecessary oscillations. In this embodiment, the preset threshold is set to 0.5 mm·s. The update formula for the sensitivity coefficient is as follows: ; in, This represents the updated sensitivity coefficient, which is dimensionless. This represents the sensitivity coefficient before the update, and is dimensionless. This represents the adaptive learning rate, measured in millimeters per second. This represents the drift characteristic value at the current moment, expressed in millimeters per second. The physical meaning of this update mechanism is that when the measured deformation is consistently greater than the corrected theoretical prediction value, the drift characteristic value is positive, the sensitivity coefficient increases, and the model becomes more sensitive to thermal input; conversely, the sensitivity coefficient decreases.

[0047] The current reference position coordinates of the welding robot are read. These reference position coordinates represent the theoretical spatial position where the welding torch tip should be when the welding robot is running according to the taught trajectory. The updated sensitivity coefficient is multiplied by the theoretical deformation prediction value to obtain the deformation compensation amount. The height component of the reference position coordinates is added to the deformation compensation amount to generate the target position coordinates in the height direction, which serves as the welding torch position compensation command. After receiving this compensation command, the motion controller of the welding robot adjusts the height position of the welding torch within the current interpolation cycle to maintain the optimal arc length distance between the welding torch tip and the workpiece surface.

[0048] After the sensitivity coefficient is updated, a safety protection step is executed. An upper and lower limit value for the safety range of the sensitivity coefficient is set; in this embodiment, the upper limit is set to 2.0 and the lower limit to 0.3. When the updated sensitivity coefficient exceeds the safety range, it is determined to be an abnormal physical deformation state. This state may be caused by abnormal conditions such as sensor malfunction, wire jamming, or workpiece perforation. In the abnormal physical deformation state, the controller outputs an abnormal alarm signal and switches the welding robot to a degraded operation mode, forcing the sensitivity coefficient back to the median value of the safety range and reducing the welding speed. If the abnormal state persists for multiple consecutive sampling cycles, an emergency shutdown is triggered to protect the equipment and workpiece safety.

[0049] Through the above steps, this application constructs a complete data-driven closed-loop compensation system. The thermo-mechanical coupling model at the physical level provides a stable large-directional prediction benchmark, while the residual integral and sensitivity adaptive mechanism at the data level continuously correct the deviation between the model and actual operating conditions. In the circumferential weld experiment of the water heater inner tank, compared with the traditional fixed-trajectory welding method, the optimized method of this application significantly improves the weld formation uniformity and welding quality consistency.

[0050] like Figure 4The figure shows a comparison of trajectory optimization effects of the data-driven welding robot welding path optimization method according to an embodiment of this application. This figure illustrates the path tracking effect after correcting the welding robot's reference position coordinates using deformation compensation. The welding torch position compensation command trajectory generated based on the thermomechanical coupling model and sensitivity coefficient closely follows the actual deformation of the workpiece surface, significantly reducing the path tracking error. The change in the root mean square error data proves that the method achieves dynamic compensation.

[0051] According to a second aspect of this application, a data-driven welding robot welding path optimization system is also provided. Figure 5 This is a structural block diagram of a data-driven welding robot welding path optimization system according to an embodiment of this application. Figure 5 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the data-driven welding robot welding path optimization method according to the first aspect of this application. The system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface. Their configurations and functions are known in the art and will not be described further here.

[0052] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in this application, based on the technical solution and inventive concept of this application, should be within the scope of protection of this application.

Claims

1. A data-driven welding robot welding path optimization method, characterized in that, The data-driven welding robot welding path optimization method includes: A pre-set thermo-mechanical coupling model is provided, which defines the nonlinear mapping relationship between the unit cumulative heat input and the workpiece surface deformation, and sets an initial sensitivity coefficient. During the welding process, the welding current and arc voltage are collected in real time, and the product of the two is calculated to obtain the instantaneous heat power. The instantaneous heat power is then integrated over time to obtain the cumulative enthalpy value. The cumulative enthalpy value is used as the input to the thermo-mechanical coupling model to obtain the theoretical deformation prediction value; at the same time, the measured deformation estimate value is calculated based on the difference between the arc voltage and the reference voltage. A sliding time window is constructed, and the difference between the measured deformation estimate and the corrected theoretical deformation prediction is calculated within the sliding time window to obtain a residual sequence. The residual sequence is then integrated to obtain the drift characteristic value. The sensitivity coefficient is updated based on the drift characteristic value; the updated sensitivity coefficient is multiplied by the theoretical deformation prediction value to obtain the deformation compensation amount; the deformation compensation amount is used to correct the reference position coordinates of the welding robot, generate the welding gun position compensation command, and drive the welding robot to perform path correction.

2. The data-driven welding robot welding path optimization method according to claim 1, characterized in that, The process of establishing the thermo-mechanical coupling model includes: Obtain the thermophysical and mechanical properties of the workpiece material; An initial mapping function between heat input and deformation is established based on the heat conduction equation and thermoelastic constitutive relation; Measured deformation data under different thermal input conditions were collected through offline calibration experiments. The parameters of the initial mapping function were fitted and corrected to generate the thermo-mechanical coupling model.

3. The data-driven welding robot welding path optimization method according to claim 1, characterized in that, The calculation process for the cumulative enthalpy value includes: The output values ​​of the welding current sensor and the arc voltage sensor are read at a preset sampling period; The instantaneous thermal power is obtained by calculating the product of the welding current and the arc voltage; The instantaneous thermal power is locally integrated within a preset sliding time window to obtain the cumulative enthalpy value.

4. The data-driven welding robot welding path optimization method according to claim 1, characterized in that, The steps for obtaining the theoretical deformation prediction value include: Input the cumulative enthalpy value into the input terminal of the thermo-mechanical coupling model; The thermo-mechanical coupling model is based on a pre-stored nonlinear mapping curve and outputs the theoretical deformation prediction value corresponding to the cumulative enthalpy value through interpolation or table lookup operations.

5. The data-driven welding robot welding path optimization method according to claim 1, characterized in that, The steps for obtaining the measured deformation estimate include: Obtain the current arc voltage value and the preset reference voltage value; The voltage deviation is obtained by calculating the difference between the arc voltage value and the reference voltage value; Based on the correspondence between arc length and voltage in the static characteristics of electric arc, the voltage deviation is converted into the change in arc length from the tip of the welding torch to the surface of the workpiece. The absolute value of the change in arc length is used as the estimated value of the measured deformation of the current workpiece surface.

6. The data-driven welding robot welding path optimization method according to claim 1, characterized in that, The correction to the theoretical deformation prediction value includes: Obtain the sensitivity coefficient at the current moment; The theoretical deformation prediction value output by the thermo-mechanical coupling model is multiplied by the sensitivity coefficient to obtain the corrected theoretical deformation prediction value.

7. The data-driven welding robot welding path optimization method according to claim 1, characterized in that, The update process for the sensitivity coefficient includes: The adjustment increment is obtained by multiplying the drift feature value by the preset adaptive learning rate; Add the current sensitivity coefficient to the adjustment increment to obtain the updated sensitivity coefficient; The adaptive learning rate is dynamically adjusted based on the absolute value of the drift feature value. When the absolute value exceeds a preset threshold, the adaptive learning rate is increased; when the absolute value is lower than the preset threshold, the adaptive learning rate is decreased.

8. The data-driven welding robot welding path optimization method according to claim 1, characterized in that, The generation of the welding torch position compensation command includes: Read the current reference position coordinates of the welding robot; The updated sensitivity coefficient is multiplied by the theoretical deformation prediction value to obtain the deformation compensation amount; The height component of the reference position coordinates is added to the deformation compensation amount to generate the target position coordinates in the height direction, which is used as the welding gun position compensation command.

9. The data-driven welding robot welding path optimization method according to claim 1, characterized in that, The optimization method also includes security protection steps: Set the upper and lower limits of the safe range for the sensitivity coefficient; When the updated sensitivity coefficient exceeds the safe range, it is determined to be an abnormal physical deformation state; In the abnormal physical deformation state, an abnormal alarm signal is output and the welding robot is switched to a degraded operation mode or an emergency shutdown is triggered.

10. A data-driven welding robot welding path optimization system, characterized in that, include: processor; A memory, wherein a computer program is stored; When the processor is configured to execute the computer program, it implements the data-driven welding robot welding path optimization method as described in any one of claims 1 to 9.

Images