Collaborative optimization analysis method for multi-station tower drum welding

Abstract

The invention relates to the technical field of process optimization, in particular to a collaborative optimization analysis method for multi-station tower drum welding. In order to solve the problem that cooperative control of tower drum multi-station welding is difficult, firstly, multi-mode sensing data of welding seam vision, arc sensing, thermal field distribution and welding voiceprint are loaded, a fusion evaluation model is constructed to obtain a welding seam deviation value and a sensor confidence score, and the health state of a sensor is judged according to a threshold value; and a corresponding fault-tolerant reconstruction control mode is triggered to obtain welding tracking control parameters, then a multi-station welding heat source model is loaded, thermal interference is decoupled in combination with the control parameters and thermal field data, multiple stations are cooperatively regulated and controlled, a cooperative control instruction is output and executed, and welding process optimization is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process optimization, in particular to a collaborative optimization analysis method for multi-station tower cylinder welding. BACKGROUND

[0002] In the multi-station welding process of large fan tower cylinders, due to the large size of the workpiece, the serious heat accumulation of multi-layer welding, and the mutual superposition of heat fields caused by simultaneous operation of multiple welding guns, welding deformation and stress concentration are easily induced. The existing technology mainly relies on single visual or arc sensor for trajectory tracking. For example, the patent document with publication number CN121361094A discloses a visual and arc fusion method based on confidence weighting, but it only corrects the trajectory of a single robot and does not consider the thermal coupling interference between multiple stations. The patent document with publication number CN103341685B proposes a fusion strategy that switches between magnetic control arc and vision according to the curvature of the weld, which improves the accuracy of complex trajectory tracking. However, the fusion decision relies on the geometric features of the workpiece, lacks fault tolerance mechanism for sensor failure, and does not involve multi-heat-source collaborative control.

[0003] Moreover, in the tower cylinder welding site, strong arc light and smoke often cause sensor signal loss. The existing methods cannot dynamically reconstruct the control mode, which may cause downtime or welding deviation. Therefore, there is an urgent need for a collaborative control method that can fuse multi-modal information, has sensor failure tolerance capability, and can decouple multi-station thermal interference. SUMMARY

[0004] The present application provides a collaborative optimization analysis method for multi-station tower cylinder welding to solve the problems of existing technology that cannot fuse multi-modal information, lack sensor failure tolerance capability, and cannot decouple multi-station thermal interference.

[0005] The technical solution of the present application to solve the above technical problems is as follows: In a first aspect, the present application provides a collaborative optimization analysis method for multi-station tower cylinder welding, comprising: S100: loading tower cylinder welding multi-modal perception data under a multi-station welding scenario, wherein the tower cylinder welding multi-modal perception data includes weld visual perception data, arc sensor perception data, thermal field distribution perception data, and welding voiceprint perception data; S200: constructing a multi-modal fusion evaluation model, inputting the tower cylinder welding multi-modal perception data into the multi-modal fusion evaluation model, obtaining a fused weld deviation and a plurality of sensor confidence scores, and the plurality of sensor confidence scores corresponding to the tower cylinder welding multi-modal perception data; S300: comparing the plurality of sensor confidence scores with a preset confidence warning threshold and a preset confidence failure threshold, respectively, dynamically determining the health status of each sensor, and obtaining a sensor health status set; S400: Trigger the corresponding fault-tolerant reconstruction control mode based on the sensor health state set and the fused weld deviation amount, and obtain the reconstructed weld tracking control parameters; S500: Load the multi-station welding heat source model, decouple and cooperatively control the thermal interference between the multi-stations based on the reconstructed weld tracking control parameters and the thermal field distribution perception data, obtain the multi-station cooperative control instructions, and execute the cooperative control of the tower cylinder welding process.

[0006] Optionally, load the tower cylinder welding multi-modal perception data under the multi-station welding scenario, including: Obtain the weld bevel edge contour data, bevel center point deviation data, and bevel width data collected by the laser vision sensor in the tower cylinder welding scenario, and add them to the weld visual perception data; Obtain the welding current instantaneous value, welding voltage instantaneous value, current standard deviation, voltage standard deviation, and short circuit frequency data collected by the arc sensor in the tower cylinder welding scenario, and add them to the arc sensor perception data; Obtain the molten pool center temperature data, molten pool width data, and heat affected zone temperature gradient data collected by the infrared thermal imager in the tower cylinder welding scenario, and add them to the thermal field distribution perception data; Obtain the welding process frequency band sound pressure level data collected by the microphone array in the tower cylinder welding scenario, and add it to the welding acoustic print perception data.

[0007] Optionally, construct a multi-modal fusion evaluation model, which includes: Collect a historical welding process data set of a preset tower cylinder model, wherein the historical welding process data set includes historical weld visual perception data, historical arc sensor perception data, historical thermal field distribution perception data, historical welding acoustic print perception data, and corresponding historical real weld deviation amount; Use the historical real weld deviation amount as supervision, use the historical weld visual perception data, historical arc sensor perception data, historical thermal field distribution perception data, and historical welding acoustic print perception data as input, use the deviation amount output node and the confidence of each sensor as output, train the initial deep confidence network model, and obtain the multi-modal fusion evaluation model with the output layer containing the output node.

[0008] Optionally, according to the confidence scores of several sensors, compare them with the preset confidence warning threshold and the preset confidence failure threshold respectively, dynamically determine the health state of each sensor, and obtain the sensor health state set, including: Compare the confidence score of any one sensor with the preset confidence warning threshold and the preset confidence failure threshold one by one; When the sensor confidence score is greater than or equal to the preset confidence warning threshold, the corresponding sensor is determined to be in a healthy state; When the sensor confidence score is less than the preset confidence warning threshold but greater than or equal to the preset confidence failure threshold, it is determined that the corresponding sensor is in an interference state; When the sensor confidence score is less than the preset confidence failure threshold, it is determined that the corresponding sensor is in a failure state; Merge the determination results of the health state, the interference state and the failure state of all sensors to form a sensor health state set.

[0009] Among them, based on the sensor health state set and the fusion weld seam deviation, the corresponding fault-tolerant reconstruction control mode is triggered to obtain the reconstructed welding tracking control parameters, including: When all sensors in the sensor health state set are determined to be in a healthy state, a full fusion tracking control mode is triggered, and the fusion weld seam deviation is directly converted into welding tracking control parameters; When the visual sensor in the sensor health state set is determined to be in an interference state, and the arc sensor, the thermal imager and the voiceprint sensor are not determined to be in a failure state, a degraded fusion tracking control mode is triggered, the weight of the weld visual perception data in the fusion calculation is reduced, the weight of the arc sensor perception data, the thermal field distribution perception data and the welding voiceprint perception data in the fusion calculation is increased, and the degraded fusion weld seam deviation is recalculated based on the adjusted weight to obtain welding tracking control parameters; When the visual sensor in the sensor health state set is determined to be in a failure state, a visual guidance-free control mode is triggered, historical weld path trajectory data is loaded, weld position calculation is performed in combination with the welding current instantaneous value, the welding voltage instantaneous value and the short-circuit frequency data in the arc sensor perception data, the calculated weld seam deviation is obtained, and the welding tracking control parameters are converted.

[0010] Optionally, a multi-station welding heat source model is loaded, the heat interference between the multi-stations is decoupled and cooperatively controlled based on the reconstructed welding tracking control parameters and the thermal field distribution perception data, and multi-station cooperative control instructions are obtained, including: The real-time welding power data and the real-time welding torch position data of each station are input into the multi-station welding heat source model to obtain tower overall transient thermal field distribution prediction data; When the molten pool temperature data of the first station exceeds the preset temperature abnormality threshold based on the thermal field distribution perception data, the molten pool temperature abnormality data of the first station is input into the multi-station welding heat source model for heat source backstepping to obtain a heat interference source identification result; When the heat interference source identification result determines that the molten pool temperature abnormality is caused by the thermal radiation interference of the second station, the heat shielding instruction including the welding torch angle adjustment parameter and the local cooling opening parameter is sent to the second station by the cloud cooperative controller; The cloud collaborative controller calibrates the arc starting point and the arc ending point of the multi-station girth weld according to the fusion weld deviation of each station at the current time, and obtains joint consistency calibration parameters; The heat shielding instruction and the joint consistency calibration parameters are combined to obtain a multi-station collaborative control instruction.

[0011] Optionally, the multi-station welding heat source model comprises the following steps: Collecting historical welding power data, historical welding torch position coordinates and corresponding historical tower hot imaging data of each station to construct a heat source model training set; Based on a double-ellipsoid heat source distribution function, an initial heat source model containing a heat source superposition factor is constructed, wherein the heat source superposition factor is used to represent the energy coupling effect of the heat sources of adjacent stations. The initial heat source model is parameterized by taking the historical tower hot imaging data as supervision and taking the historical welding power data and the historical welding torch position coordinates as input, and a multi-station welding heat source model is obtained.

[0012] Optionally, the method further comprises the following steps: When the confidence score of the plurality of sensors is less than the preset confidence failure threshold, it is determined that the sensor health state set is in a total failure state, an emergency shutdown protection mode is triggered, a shutdown alarm instruction is generated, and the welding process is terminated.

[0013] Optionally, the method further comprises the following steps: Obtaining the welding quality false negative rate caused by sensor failure and the welding quality false positive rate caused by environmental interference in the historical welding process of the tower welding scene; The ratio of the welding quality false negative rate to the welding quality false positive rate is calculated and set as a confidence threshold adjustment factor; According to the ratio of the preset reference adjustment factor to the confidence threshold adjustment factor, the preset confidence warning threshold and the preset confidence failure threshold are adjusted.

[0014] Optionally, the method further comprises the following steps: Obtaining real-time welding progress data and real-time weld reinforcement detection data of each station in the tower welding scene; When the real-time weld reinforcement detection data exceeds the preset reinforcement threshold, the corresponding position is identified as a repair welding area; According to the real-time welding progress data, the adjacent stations in an idle state or with leading welding progress are dispatched, a repair welding collaborative instruction is generated, and online repair welding work is performed on the repair welding area.

[0015] Through implementation of the present application, the tower drum welding multi-modal perception data in a multi-station welding scene can be realized, wherein the tower drum welding multi-modal perception data includes welding seam visual perception data, electric arc sensing perception data, heat field distribution perception data and welding acoustic print perception data; the information limitation of single type perception data is broken through, multi-modal data can reflect the real state of the welding process from different dimensions, and the analysis deviation caused by one-sidedness of single source data is avoided; meanwhile, the data is collected for the multi-station scene of tower drum welding, which matches the working condition demand of actual industrial production, and provides matched data source for subsequent multi-station collaborative regulation and control.

[0016] Through implementation of the present application, the tower drum welding multi-modal perception data can be input into the multi-modal fusion evaluation model to obtain fusion welding seam deviation and a plurality of sensor confidence scores corresponding to the tower drum welding multi-modal perception data; compared with deviation detection of a single sensor, the fusion welding seam deviation combines multi-dimensional data features, has higher detection accuracy, and can more truly reflect the actual welding seam deviation of the tower drum welding; meanwhile, quantitative determination of the working state of the sensor is realized, the short board that the prior art lacks sensor failure evaluation capability is broken through, and quantitative basis is provided for subsequent fault-tolerant control.

[0017] Through implementation of the present application, the health states of the sensors can be dynamically determined by comparing the plurality of sensor confidence scores with preset confidence warning thresholds and preset confidence failure thresholds respectively, and a sensor health state set can be obtained; fine and dynamic determination of the sensor state is realized instead of simple binary determination of validity / invalidity, and the intermediate state of the sensor disturbed by arc light, smoke and other environmental interference can be accurately identified; the formed health state set provides clear and explicit state basis for accurate triggering of a subsequent fault-tolerant reconstruction control mode, and avoids false triggering or missed triggering of fault-tolerant control.

[0018] Through implementation of the present application, the corresponding fault-tolerant reconstruction control mode can be triggered based on the sensor health state set and the fusion welding seam deviation, and the reconstructed welding tracking control parameters can be obtained; the fault-tolerant capability of sensor failure and disturbance is possessed, when the sensor is disturbed by the environment or fails, the machine does not need to be stopped, the data fusion weight is adjusted or the welding seam position is calculated through dynamic reconstruction of the control mode, and the continuity of the welding tracking control is ensured; the control parameters are customized for different sensor failure scenes, the welding trajectory tracking can still maintain high accuracy under non-ideal perception state, and the welding deviation problem caused by loss of sensor signal is avoided.

[0019] By implementing the present application, the multi-station welding heat source model can be loaded, the thermal interference between the multi-stations can be decoupled and cooperatively controlled based on the reconstructed welding tracking control parameters and the heat field distribution perception data, the multi-station cooperative control instructions can be obtained, and the cooperative control of the tower drum welding process can be performed. The present application breaks through the defects of the prior art that do not consider the multi-station thermal coupling interference, can accurately identify the thermal interference source and decouple the thermal interference through thermal shielding, parameter adjustment and other ways, and effectively avoids the welding deformation and stress concentration caused by the thermal field superposition of multi-welding gun operation. The multi-station arc striking points, arc collecting points and the like are uniformly calibrated, and the stations can be dispatched for online repair welding, the global cooperative control of multi-station welding is realized, and the consistency and quality stability of the overall welding of the tower drum are improved.

[0020] In summary, by implementing the present application, the global cooperative control of the multi-station welding process of the tower drum can be realized, the welding deformation and stress concentration problems caused by the thermal field superposition in the multi-station welding of the tower drum are effectively solved, the stop and welding deviation problems caused by the loss of sensor signals due to arc light and smoke are solved, the welding seam trajectory tracking accuracy and the quality consistency of multi-station welding are improved, the continuity, stability and efficiency of the tower drum welding process are ensured, and the welding quality and production efficiency of the large wind turbine tower drum are greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flowchart of a cooperative optimization analysis method for multi-station tower drum welding provided by the present application is shown. Figure 2 A flowchart of obtaining multi-station cooperative control instructions in a cooperative optimization analysis method for multi-station tower drum welding provided by the present application is shown. DETAILED DESCRIPTION

[0022] As shown in Figure 1 The embodiment of the present application provides a cooperative optimization analysis method for multi-station tower drum welding, which comprises: S100: loading tower drum welding multi-modal perception data in a multi-station welding scene, wherein the tower drum welding multi-modal perception data comprises welding seam visual perception data, arc sensing perception data, heat field distribution perception data and welding voiceprint perception data; S200: constructing a multi-modal fusion evaluation model, inputting the tower drum welding multi-modal perception data into the multi-modal fusion evaluation model, obtaining fusion welding seam deviation and a plurality of sensor confidence scores, and the plurality of sensor confidence scores corresponding to the tower drum welding multi-modal perception data; S300: comparing the plurality of sensor confidence scores with a preset confidence warning threshold and a preset confidence failure threshold, respectively, dynamically determining the health status of each sensor, and obtaining a sensor health status set; S400: Trigger the corresponding fault-tolerant reconstruction control mode based on the sensor health state set and the fused weld deviation amount, and obtain the reconstructed weld tracking control parameters; S500: Load the multi-station welding heat source model, decouple and cooperatively control the heat interference between the multi-stations based on the reconstructed weld tracking control parameters and the heat field distribution perception data, obtain the multi-station cooperative control instructions, and perform the cooperative control of the tower cylinder welding process.

[0023] In step S100 of the embodiment, the tower cylinder welding multi-modal perception data under the multi-station welding scene is loaded, including: Obtain the weld bevel edge profile data, bevel center point deviation data and bevel width data collected by the laser vision sensor under the tower cylinder welding scene, and add them into the weld visual perception data; Obtain the welding current instantaneous value, welding voltage instantaneous value, current standard deviation, voltage standard deviation and short circuit frequency data collected by the arc sensor under the tower cylinder welding scene, and add them into the arc sensing perception data; Obtain the molten pool center temperature data, molten pool width data and heat affected zone temperature gradient data collected by the infrared thermal imager under the tower cylinder welding scene, and add them into the heat field distribution perception data; Obtain the welding process frequency band sound pressure level data collected by the microphone array under the tower cylinder welding scene, and add them into the welding voiceprint perception data.

[0024] In step S100 of the embodiment, the purpose of the above steps is to provide comprehensive and multi-dimensional original input data for the multi-modal fusion evaluation model constructed subsequently, to realize multi-dimensional perception of the weld state, the welding process heat field and the sensor working state, to lay a data foundation for subsequent fusion weld deviation amount calculation, sensor health state determination, fault-tolerant reconstruction control mode triggering and multi-station heat interference decoupling and cooperative control, and to solve the problem that single sensing data in the tower cylinder welding scene is easily interfered by arc light and smoke, resulting in signal loss, and to ensure the accuracy and comprehensiveness of multi-station welding cooperative optimization analysis.

[0025] To achieve the above purpose, first, the weld bevel edge profile data, bevel center point deviation data and bevel width data collected by the laser vision sensor under the tower cylinder welding scene are obtained, and added into the weld visual perception data; The purpose of this step is to collect the geometric feature data related to the weld groove by the laser vision sensor, intuitively reflect the actual position of the weld and the shape of the groove, provide the core data of the visual dimension for the calculation of the weld deviation, realize the precise perception of the weld geometry, classify the collected data into weld visual perception data, complete the integration of this type of perception data, for example, the edge profile of the weld groove of a certain tower drum is collected as a smooth V-shaped profile, the deviation data of the groove center point relative to the standard position is 0.2mm, and the groove width data is 15mm. Add all these data to the weld visual perception data.

[0026] Then, the welding current instantaneous value, welding voltage instantaneous value, current standard deviation, voltage standard deviation and short circuit frequency data collected by the arc sensor in the scene of tower drum welding are obtained and added to the arc sensing perception data. The purpose of this step is to collect the electric parameter data related to the welding arc by the arc sensor, reflect the stability of the welding arc and the welding energy input state in the welding process, provide the key data of the arc dimension for subsequent fusion evaluation and weld position calculation, realize the real-time perception of the working state of the welding arc, classify the collected data into arc sensing perception data, complete the integration of this type of perception data, for example, the welding current instantaneous value is 320A, the welding voltage instantaneous value is 30V, the current standard deviation is 5A, the voltage standard deviation is 1V, and the short circuit frequency is 15Hz. Add all these data to the arc sensing perception data.

[0027] Then, the molten pool center temperature data, molten pool width data and heat affected zone temperature gradient data collected by the infrared thermal imager in the scene of tower drum welding are obtained and added to the thermal field distribution perception data. The purpose of this step is to collect the thermal parameter data of the welding molten pool and the heat affected zone by the infrared thermal imager, accurately reflect the thermal field distribution state in the welding process, provide the core data of the thermal field dimension for multi-station thermal interference decoupling and heat source model calculation, realize the real-time monitoring of the welding thermal field, classify the collected data into thermal field distribution perception data, complete the integration of this type of perception data, for example, the molten pool center temperature data of a certain station is 1500℃, the molten pool width data is 8mm, and the heat affected zone temperature gradient data is 200℃ / mm. Add all these data to the thermal field distribution perception data.

[0028] Further, the welding process frequency band sound pressure level data collected by the microphone array in the scene of tower drum welding is obtained and added to the welding soundprint perception data.

[0029] The purpose of this step is to collect acoustic fingerprint feature data during the welding process through the microphone array, to assist in reflecting the stability of the welding process from the acoustic fingerprint dimension, to supplement the dimension of multi-modal perception data, to improve the comprehensiveness and accuracy of subsequent multi-modal fusion evaluation, to classify the collected data into welding acoustic fingerprint perception data, to complete the integration of such perception data, for example, collecting 85dB of sound pressure level data in the 2kHz frequency band and 78dB of sound pressure level data in the 5kHz frequency band during the welding process, adding all these data to the welding acoustic fingerprint perception data.

[0030] In step S200 of the embodiment of the present application, a multi-modal fusion evaluation model is constructed, which includes the following steps: A historical welding process data set of a preset tower drum model is collected, wherein the historical welding process data set includes historical weld visual perception data, historical arc sensor perception data, historical thermal field distribution perception data, historical welding acoustic fingerprint perception data, and corresponding historical real weld deviation amount; The historical real weld deviation amount is used as supervision, the historical weld visual perception data, the historical arc sensor perception data, the historical thermal field distribution perception data, and the historical welding acoustic fingerprint perception data are used as input, the deviation amount output node and the sensor confidence are used as output, and an initial deep belief network model is trained to obtain a multi-modal fusion evaluation model with an output layer containing an output node.

[0031] In step S200 of the embodiment of the present application, the purpose of the above steps is to prepare sufficient and effective training data for training the multi-modal fusion evaluation model, and to train the initial deep belief network model in a supervised manner, so that the model has the ability to accurately calculate the fusion weld deviation amount from the multi-modal perception data and evaluate the sensor confidence score, providing reliable model support for subsequent sensor health state judgment and fault tolerance reconstruction control mode triggering, and ensuring the accuracy of data fusion evaluation in multi-station tower drum welding collaborative optimization analysis.

[0032] To achieve the above-mentioned purposes, first, a historical welding process data set of a preset tower drum model is collected, wherein the historical welding process data set includes historical weld visual perception data, historical arc sensor perception data, historical thermal field distribution perception data, historical welding acoustic fingerprint perception data, and corresponding historical real weld deviation amount; The purpose of this step is to provide real and comprehensive sample data for training the initial deep belief network model, ensuring the effectiveness and pertinence of model training, and achieving the implementation by collecting various types of historical perception data generated during the past welding process of the specified preset tower drum model and the corresponding historical real weld deviation, and the historical welding process data set needs to include historical weld visual perception data, historical arc sensor perception data, historical thermal field distribution perception data, historical welding voiceprint perception data, and corresponding historical real weld deviation, for example, collecting historical welding data of a certain 2.5MW wind turbine tower drum model, which includes historical weld bevel edge contour data, historical welding current instantaneous value data, historical molten pool center temperature data, and historical welding process frequency band sound pressure level data during the welding of the tower drum of this model, and the historical real weld deviation of the corresponding welding position is 0.3mm. These data are integrated into the historical welding process data set of the preset tower drum model.

[0033] Then, the initial deep belief network model is trained with the historical real weld deviation as the supervision, the historical weld visual perception data, the historical arc sensor perception data, the historical thermal field distribution perception data, and the historical welding voiceprint perception data as the input, and the deviation output node and the sensor confidence as the output, to obtain a multi-modal fusion evaluation model with the output layer containing the output node.

[0034] The purpose of this step is to let the initial deep belief network model learn the correlation between multi-modal perception data and fused weld deviation and sensor confidence through supervised deep learning training, so that the trained multi-modal fusion evaluation model can realize effective fusion and evaluation of multi-modal perception data. The implementation is to collect various types of historical perception data as input data of the input layer of the model, use the historical real weld deviation as the supervision for model training, set the deviation output node and the sensor confidence as the output content of the output layer of the model, and use the deep learning algorithm to iteratively train the initial deep belief network model until the output result of the initial deep belief network model meets the preset accuracy requirement, and finally obtain a multi-modal fusion evaluation model with the output layer containing the output node. For example, input the historical weld visual perception data, historical arc sensor perception data, historical thermal field distribution perception data, and historical welding voiceprint perception data of a certain preset tower drum model into the initial deep belief network model, use the corresponding historical real weld deviation of 0.2mm as the supervision, set the calculation result of the fused weld deviation and the confidence scores of the laser vision sensor, the arc sensor, the infrared thermal imager, and the microphone array as the model output, and after multiple rounds of iterative training, the error between the fused weld deviation output by the model and the historical real weld deviation is controlled within 0.05mm. At this time, the training is completed and the corresponding multi-modal fusion evaluation model is obtained.

[0035] For example, the multi-modal fusion evaluation model can be obtained by training an initial deep belief network model, using a historical welding process data set of a preset tower drum model as training samples, the samples containing historical welding seam vision, arc sensing, heat field distribution, welding acoustic fingerprint perception data and corresponding historical real welding seam deviation, the sample size being 5000 groups, and the training rounds being set to 300 rounds. The initial deep belief network model input is four types of historical perception data, the historical real welding seam deviation is supervised, and the output is the deviation output node and the confidence of each sensor. The hyperparameters are set as follows: learning rate 0.001, 3 hidden layers, neuron number 512, 256, 128, and ReLU activation function. The convergence criterion is that the average error between the fusion welding seam deviation output by the model and the historical real welding seam deviation is stably less than 0.05 mm, and there is no obvious decrease in the error for 20 consecutive iterations.

[0036] Further, the tower drum welding multi-modal perception data needs to be input into the multi-modal fusion evaluation model to obtain the fusion welding seam deviation and the confidence scores of several sensors, and the confidence scores of several sensors correspond to the tower drum welding multi-modal perception data; For example, the welding seam visual perception data collected has a V-shaped edge profile of the welding seam groove, a groove center point deviation of 0.3 mm, a groove width of 14 mm, the arc sensing perception data has a welding current instantaneous value of 310 A, a welding voltage instantaneous value of 29 V, a current standard deviation of 4 A, a voltage standard deviation of 0.8 V, and a short-circuit frequency of 14 Hz, the heat field distribution perception data has a molten pool center temperature of 1480℃, a molten pool width of 7.5 mm, and a heat affected zone temperature gradient of 190℃ / mm, and the welding acoustic fingerprint perception data has a sound pressure level of 84 dB in the 2 kHz frequency band and a sound pressure level of 77 dB in the 5 kHz frequency band. After inputting the multi-modal fusion evaluation model, the fusion welding seam deviation is obtained as 0.25 mm, and the confidence scores of several sensors are obtained as 92 points for the laser visual sensor, 95 points for the arc sensor, 93 points for the infrared thermal imager, and 90 points for the microphone array. Each score is matched with the corresponding tower drum welding multi-modal perception data.

[0037] In step S300 of the embodiment, the confidence scores of several sensors are compared with the preset confidence warning threshold and the preset confidence failure threshold, respectively, to dynamically determine the health status of each sensor and obtain a sensor health status set, including: The confidence scores of any one sensor are compared with the preset confidence warning threshold and the preset confidence failure threshold one by one; When the sensor confidence score is greater than or equal to the preset confidence warning threshold, the corresponding sensor is determined to be in a healthy state; When the sensor confidence score is less than the preset confidence warning threshold but greater than or equal to the preset confidence failure threshold, it is determined that the corresponding sensor is in an interference state; When the sensor confidence score is less than the preset confidence failure threshold, it is determined that the corresponding sensor is in a failure state. The determination results of the health state, the interference state and the failure state of all the sensors are combined to form a sensor health state set.

[0038] In step S300 of the embodiment, the purpose of the above steps is to accurately identify the real-time working state of each sensor in the tower drum welding process, to provide a direct basis for subsequent triggering of a corresponding fault-tolerant reconstruction control mode according to the sensor state, to solve the problem of signal abnormality or failure of the sensor in the welding field due to arc light and smoke interference, and to ensure the fault tolerance and stability of the welding tracking control.

[0039] To achieve the above purpose, firstly, the confidence score of any one sensor is compared with the preset confidence warning threshold and the preset confidence failure threshold one by one. The purpose of this step is to establish a quantitative comparison basis for the health state determination of each sensor, to preliminarily distinguish the state interval of the sensor through numerical comparison, and the implementation is to extract the confidence score of each sensor output by the multi-modal fusion evaluation model, and to perform a numerical size comparison operation with the preset confidence warning threshold and the confidence failure threshold in turn, for example, the preset confidence warning threshold is 80 points, the preset confidence failure threshold is 50 points, and the extracted laser vision sensor confidence score is 75 points. 75 points are compared with 80 points first, and then compared with 50 points, and the one-by-one comparison operation of the sensor is completed.

[0040] When the sensor confidence score is greater than or equal to the preset confidence warning threshold, it is determined that the corresponding sensor is in a healthy state. The purpose of this step is to determine the judgment standard of the sensor in the normal working state without interference and reliable data, and the implementation is that after the comparison of the sensor confidence score with the two thresholds is completed, if the score meets the condition of being greater than or equal to the confidence warning threshold, the working state of the sensor can be determined as a healthy state, for example, the preset confidence warning threshold is 80 points, the confidence failure threshold is 60 points, and the arc sensor confidence score is 95 points. 95 points are greater than the confidence warning threshold of 80 points, so the arc sensor is determined to be in a healthy state.

[0041] When the sensor confidence score is less than the preset confidence warning threshold but greater than or equal to the preset confidence failure threshold, it is determined that the corresponding sensor is in an interference state. The purpose of this step is to determine the state determination standard of the sensor disturbed by arc light, smoke dust and other environmental factors, and the reliability of the data decreases but is not completely invalid. The implementation is that after completing the threshold comparison, if the sensor confidence score is between the confidence warning threshold and the confidence failure threshold, the working state of the sensor can be determined as the interference state. For example, the confidence warning threshold is set to 80 points, and the confidence failure threshold is set to 60 points. If the infrared thermal imager confidence score is 70 points, which is less than the confidence warning threshold of 80 points and greater than the confidence failure threshold of 60 points, the infrared thermal imager is determined to be in an interference state.

[0042] When the sensor confidence score is less than the preset confidence failure threshold, the corresponding sensor is determined to be in a failure state. The purpose of this step is to determine the state determination standard of the sensor due to severe interference or equipment failure, resulting in signal loss and completely unreliable data. The implementation is that after completing the threshold comparison, if the sensor confidence score meets the condition of being less than the confidence failure threshold, the working state of the sensor can be determined as the failure state. For example, if the confidence warning threshold is set to 80 points and the confidence failure threshold is set to 60 points, the microphone array confidence score is 55 points, which is less than the confidence failure threshold of 60 points, so the microphone array is determined to be in a failure state.

[0043] Merge the determination results of the health state, interference state and failure state of all sensors to form a sensor health state set.

[0044] Integrate the independent state determination results of each sensor to form a unified and complete sensor state set, which provides comprehensive state reference for triggering the subsequent fault-tolerant reconstruction control mode. The implementation is to collect the state determination results of all sensors participating in perception such as laser vision sensor, arc sensor, infrared thermal imager and microphone array, integrate each sensor with the corresponding determination state, and form a structured sensor health state set.

[0045] In step S400 of the embodiment of the application, based on the sensor health state set and the fusion weld deviation, the corresponding fault-tolerant reconstruction control mode is triggered, and the reconstructed welding tracking control parameters are obtained, including: When all sensors in the sensor health state set are determined to be in a healthy state, the full fusion tracking control mode is triggered, and the fusion weld deviation is directly converted into welding tracking control parameters. When the sensor health state set determines that the visual sensor is in the interference state, and the arc sensor, the thermal imager and the acoustic sensor are all determined to be in the failure state, a degraded fusion tracking control mode is triggered, the weight of the weld visual perception data in the fusion calculation is reduced, the weight of the arc sensing perception data, the thermal field distribution perception data and the welding acoustic perception data in the fusion calculation is increased, and the degraded fusion weld deviation is recalculated based on the adjusted weight to obtain the welding tracking control parameter; When the sensor health state set determines that the visual sensor is in the failure state, a visual guidance control mode is triggered, historical weld path trajectory data is loaded, the weld position is calculated in combination with the welding current instantaneous value, the welding voltage instantaneous value and the short circuit frequency data in the arc sensing perception data, the calculated weld deviation is obtained, and the welding tracking control parameter is converted.

[0046] In step S400 of the embodiment, the above steps are used to dynamically match the welding tracking control strategy according to the actual working state of each sensor, avoid the welding trajectory deviation problem caused by sensor interference or failure through fault tolerance reconstruction, ensure that accurate welding tracking control parameters can be output under different sensor states, provide a reliable control basis for subsequent multi-station thermal interference decoupling and collaborative control, avoid tower cylinder welding downtime or welding deviation caused by sensor abnormalities, and ensure the continuity and welding precision of the multi-station tower cylinder welding process.

[0047] To achieve the above purpose, when the sensor health state set determines that all sensors are in the healthy state, a full fusion tracking control mode is triggered, and the fusion weld deviation is directly converted into the welding tracking control parameter; That is, after verifying the sensor health state set and confirming that the laser vision sensor, the arc sensor, the infrared thermal imager and the microphone array are in the healthy state, the full fusion tracking control mode is triggered, the fusion weld deviation output by the multi-modal fusion evaluation model is directly converted into the welding tracking control parameter such as the welding gun moving distance and the welding gun angle according to the preset parameter conversion rule. For example, when the sensor health state set determines that the laser vision sensor, the arc sensor, the infrared thermal imager and the microphone array are in the healthy state, the fusion weld deviation output by the multi-modal fusion evaluation model is 0.2 mm, and the 0.2 mm fusion weld deviation is directly converted into the welding tracking control parameter of compensating 0.2 mm along the center line of the groove and keeping the welding gun angle at 30° according to the preset rule.

[0048] The preset rule is a standardized mapping rule for converting the weld deviation amount to the welding tracking control parameter in the tower drum welding field, needs to be formulated in combination with the welding gun movement control logic, the groove size requirement and the welding process specification of the tower drum welding, and the core is to convert the value and direction of the fusion weld deviation amount into the control parameters such as the compensation distance of the welding gun in the horizontal / vertical direction, the angle fine adjustment amount of the welding gun, the travel speed correction value, and the like, while matching the preset welding precision standard of the station, to ensure that the deviation compensation is accurately adapted to the actual working condition of the tower drum welding. For example, the fusion weld deviation amount of 0.2 mm corresponds to the horizontal compensation of 1 mm of the welding gun, the deviation direction is left, the welding gun moves left, the deviation direction is right, the welding gun moves right, the angle and travel speed of the welding gun remain the default values.

[0049] Specifically, the formulation of the preset rule needs to be based on the actual working condition of the tower drum welding. First, the welding gun movement control hardware logic needs to be matched, and the parameter conversion basis is determined according to the displacement accuracy, angle adjustment range and speed adjustment step of the welding gun servo motor. Second, the tower drum weld groove size design requirement needs to be fitted, and the deviation compensation range is defined according to the groove angle, root gap and groove width. Third, the tower drum welding industry process specification and the corresponding tower drum model welding process file need to be followed, and the precision standard of different weld positions needs to be determined. The multi-station welding collaborative control requirement needs to be met to ensure that the conversion rules of all stations are unified and the joint consistency requirement of the girth weld is met.

[0050] The core mapping logic of the preset rule takes the value and direction of the fusion weld deviation amount as the core input, and establishes accurate mapping with the welding tracking control parameters. The deviation value is linearly converted according to the welding gun hardware displacement accuracy, directly corresponding to the compensation distance of the welding gun in the horizontal / vertical direction, and the larger the value, the more the compensation amount is adapted. The deviation direction determines the compensation direction of the welding gun, and the left / right deviation of the groove in the horizontal direction corresponds to the left / right movement of the welding gun, and the front / back deviation in the vertical direction corresponds to the front / back movement of the welding gun. And the adjustment type can be divided according to the deviation amount threshold, and only the displacement compensation of the welding gun is adjusted for small deviation, and the angle and travel speed of the welding gun are fine tuned for large deviation.

[0051] The preset rule also includes process constraints, which need to be strictly followed. The compensation distance of the welding gun should not exceed 1 / 2 of the single-sided gap of the groove to avoid welding deviation and incomplete fusion and other welding defects. The angle fine adjustment amount of the welding gun is controlled within ±5° to match the tower drum thick-wall weld pool forming requirement and ensure the penetration. The travel speed correction value of the welding gun fluctuates by no more than ±20% of the default value to maintain stable welding heat input and prevent heat accumulation or insufficient heat input. All converted control parameters need to meet the adjustment accuracy of the welding gun servo system, and there is no hardware incompatibility such as excessive range and non-integer step.

[0052] When the sensor health state set determines that the laser vision sensor is in an interference state, and the arc sensor, the thermal imager, and the acoustic sensor are all determined to be in a failure state, a degraded fusion tracking control mode is triggered, the weight of the weld visual perception data in the fusion calculation is reduced, the weights of the arc sensing perception data, the thermal field distribution perception data, and the welding acoustic perception data in the fusion calculation are increased, and a degraded fusion weld deviation is recalculated based on the adjusted weights to obtain a welding tracking control parameter. That is, first verify the sensor health state set, confirm that the laser vision sensor is in an interference state, and that the arc sensor, the infrared thermal imager, and the microphone array are all in a healthy or interference state and are not failed, then trigger the degraded fusion tracking control mode, reduce the fusion weight of the weld visual perception data according to the preset weight adjustment rule, increase the fusion weight of the arc sensing perception data, the thermal field distribution perception data, and the welding acoustic perception data, and then recalculate the fusion of the modal perception data based on the adjusted weights to obtain a degraded fusion weld deviation, and finally convert the deviation into a welding tracking control parameter according to the parameter conversion rule.

[0053] The preset weight adjustment rule takes the laser vision sensor interference and the remaining three types of sensors not being failed as the trigger premise, the quantitative basis is the health state of each sensor, the core rule is to reduce the weight of the weld visual perception data disturbed, and the weight of the arc sensing perception data, the thermal field distribution perception data, and the welding acoustic perception data is increased by the same amount, and the total weight of the four types of perception data is always 1.

[0054] For example, the sensor health state set determines that the laser vision sensor is in an interference state, the arc sensor, the infrared thermal imager, and the microphone array are all in a healthy state, the weld visual perception data weight in the original fusion calculation is 0.4, the arc sensing perception data, the thermal field distribution perception data, and the welding acoustic perception data weight are each 0.2, the weld visual perception data weight is reduced to 0.1 after adjustment, the arc sensing perception data, the thermal field distribution perception data, and the welding acoustic perception data weight are each increased to 0.3, and the degraded fusion weld deviation is recalculated to be 0.22 mm based on the new weight, and the deviation is converted into a welding tracking control parameter of 0.22 mm of the welding torch along the center line of the groove transverse compensation and a welding speed of 5 mm / s.

[0055] When the sensor health state set determines that the laser vision sensor is in a failure state, a vision-free guidance control mode is triggered, historical weld path trajectory data is loaded, the welding current instantaneous value, the welding voltage instantaneous value, and the short-circuit frequency data in the arc sensing perception data are combined to calculate the weld position, a calculated weld deviation is obtained, and a welding tracking control parameter is converted.

[0056] That is, first check the sensor health status set, confirm that the laser vision sensor is in a failure state, then trigger the visionless guidance control mode, load the historical weld path trajectory data of the same model tower drum at the same station from the database as the basis trajectory reference, extract the welding current instantaneous value, welding voltage instantaneous value and short circuit frequency data in the arc sensor sensing data, substitute into the preset weld position calculation algorithm, combine the historical trajectory data to calculate the actual weld position and the calculated weld deviation of the standard trajectory, and finally convert the deviation into welding tracking control parameters according to the parameter conversion rule.

[0057] The principle of the weld position calculation algorithm is based on the correlation between the arc sensor sensing data and the weld position, and the historical weld path trajectory data of the same model tower drum at the same station is used as the reference. By real-time changes of the welding current instantaneous value, the welding voltage instantaneous value and the short circuit frequency, the offset of the actual weld position relative to the standard trajectory is quantitatively determined, and the weld position calculation under the visionless guidance is realized.

[0058] The weld position calculation algorithm first calibrates the historical weld path trajectory data as the standard weld path, and then establishes a quantitative correspondence between the arc sensor sensing data and the weld position offset: when the arc sensor sensing data deviates from the threshold of the preset standard welding state, it is converted into the offset value of the weld transverse / longitudinal direction according to the arc sensor sensing data-offset conversion ratio, and if there is no threshold deviation, the historical weld path trajectory data is used as the reference to determine no offset.

[0059] The threshold of the preset standard welding state is obtained based on the arc sensor sensing data corresponding to the historical weld path trajectory data of the same model tower drum at the same station. First, the historical arc sensor sensing data of the same station is selected, which has qualified welding quality and meets the standard of weld forming, and the stable fluctuation interval of the welding current instantaneous value, the welding voltage instantaneous value and the short circuit frequency is extracted. The interval mean value is taken as the standard value, and the threshold is calibrated in the range of standard value ± 5%-10%. After verifying the threshold adaptability in the actual welding working condition, the threshold is determined as the final threshold of the preset standard welding state.

[0060] The arc sensor sensing data-offset conversion ratio is calibrated based on the historical weld path trajectory data and the corresponding historical arc sensor sensing data, for example, the current deviates from the standard value by 10A, which corresponds to the weld position offset of 0.05mm, the voltage deviates from 1V, which corresponds to the weld position offset of 0.05mm, and the short circuit frequency deviates from 3Hz, which corresponds to the weld position offset of 0.05mm.

[0061] For example, the sensor health status centralized laser vision sensor is in a failure state, the historical weld path trajectory data loaded to a certain ring seam station of a certain 2.5 MW fan tower drum is a standard trajectory, and the welding current instantaneous value, the welding voltage instantaneous value and the short circuit frequency in the arc sensor sensing data extracted are 320 A, 30 V and 15 Hz respectively. These electric parameters are substituted into the weld position calculation algorithm, and the arc sensor sensing data-offset conversion ratio obtained according to the historical weld path trajectory data is combined to calculate that the weld position deviation is 0.3 mm. The calculated weld deviation is converted into the welding tracking control parameters that the welding torch is longitudinally compensated along the center line of the groove by 0.3 mm and the welding wire feeding speed is adjusted to 8 m / min.

[0062] As shown in Figure 2 In step S500 of the embodiment of the present application, a multi-station welding heat source model is loaded, the thermal interference between the multi-stations is decoupled and cooperatively controlled based on the reconstructed welding tracking control parameters and the heat field distribution sensing data, and multi-station cooperative control instructions are obtained, including: The real-time welding power data and the real-time welding torch position data of each station are input into the multi-station welding heat source model to obtain the overall transient heat field distribution prediction data of the tower drum; When the molten pool temperature data of the first station exceeds the preset temperature abnormality threshold based on the heat field distribution sensing data, the molten pool temperature abnormality data of the first station is input into the multi-station welding heat source model for heat source backstepping to obtain a thermal interference source identification result; When the thermal interference source identification result determines that the molten pool temperature abnormality is caused by the thermal radiation interference of the second station, the cloud cooperative controller sends a heat shielding instruction to the second station, and the heat shielding instruction includes welding torch angle adjustment parameters and local cooling opening parameters; The cloud cooperative controller uniformly calibrates the arc striking point and the arc stopping point of the multi-station girth weld according to the fusion weld deviation of each station at the current time to obtain joint consistency calibration parameters; The heat shielding instruction and the joint consistency calibration parameters are merged to obtain the multi-station cooperative control instruction.

[0063] In step S500 of the embodiment of the present application, the purpose of the above steps is to realize the prediction of the overall heat field of the tower drum welding and the accurate identification of the thermal interference source through the multi-station welding heat source model, to take targeted heat shielding measures to eliminate the thermal coupling interference between the multi-stations, to uniformly calibrate the arc striking point and the arc stopping point of the multi-station girth weld to ensure the consistency of the weld joint, and finally to integrate the multi-station cooperative control instruction to realize the heat field and trajectory cooperative control of the multi-station tower drum welding, to solve the problems of heat field superposition, welding deformation and stress concentration caused by the simultaneous operation of multiple welding torches, and to ensure the overall quality and stability of the multi-station welding.

[0064] To achieve the above-mentioned purpose, first, the real-time welding power data and real-time welding torch position data of each station are input into the multi-station welding heat source model to obtain the overall transient thermal field distribution prediction data of the tower drum; That is, the real-time welding power data and real-time welding torch position data of each station in the multi-station welding scene are collected, and these data are synchronously input into the multi-station welding heat source model which has completed parameter identification. Through the operation of the double-ellipsoid heat source distribution function and the heat source superposition factor inside the multi-station welding heat source model, the overall transient thermal field distribution prediction data of the tower drum is output. For example, the real-time welding power data of three stations of the tower drum are collected as 30kW, 28kW and 32kW, and the real-time welding torch position data are collected as 0°, 120° and 240° at the tower drum girth joint. These data are input into the multi-station welding heat source model, and after operation, the overall transient thermal field distribution prediction data of the tower drum is obtained, in which the predicted temperature around the molten pool of the 0° station is 1550℃, the predicted temperature around the molten pool of the 120° station is 1520℃, and the predicted temperature around the molten pool of the 240° station is 1580℃.

[0065] When the molten pool temperature data of the first station is detected to exceed the preset temperature abnormality threshold based on the thermal field distribution perception data, the molten pool temperature abnormality data of the first station is input into the multi-station welding heat source model for heat source backstepping to obtain the thermal interference source identification result; That is, the thermal field distribution perception data collected by the infrared thermal imager is used to monitor the molten pool temperature data of each station in real time, and the molten pool temperature data is compared with the preset temperature abnormality threshold. When it is detected that the molten pool temperature data of the first station exceeds the threshold, the molten pool temperature abnormality data of the station is extracted and input into the multi-station welding heat source model. The model is used for heat source backstepping operation to analyze the cause of the temperature abnormality and identify the corresponding thermal interference source, and the thermal interference source identification result is output. For example, the preset molten pool temperature abnormality threshold is 1600℃, and the molten pool temperature data of the first station is detected to be 1650℃ based on the thermal field distribution perception data, which exceeds the temperature abnormality threshold. The molten pool temperature abnormality data of 1650℃ is input into the multi-station welding heat source model for heat source backstepping. After operation, the thermal interference source identification result is obtained, which determines that the temperature abnormality is caused by the thermal radiation interference of the second station.

[0066] The first station and the second station are not a one-to-one correspondence, but are two independent stations randomly selected in the multi-station tower drum welding scene. They are example station designations selected to illustrate the thermal interference decoupling and collaborative control logic. The first station is an abnormal station where the molten pool temperature data exceeds the preset threshold, and the second station is a thermal interference source station determined by the multi-station welding heat source model backstepping to cause the temperature abnormality. The correlation between the two is only determined by the actual occurrence of thermal radiation interference, and there is no fixed one-to-one matching relationship. It can be flexibly determined according to the actual occurrence of the thermal interference on site.

[0067] When the heat interference source identification result determines that the abnormality of the molten pool temperature is caused by the thermal radiation interference of the second station, the cloud collaborative controller sends a heat shielding instruction to the second station, and the heat shielding instruction includes a welding torch angle adjustment parameter and a local cooling opening parameter. That is, after the heat interference source identification result determines that the abnormality of the molten pool temperature is caused by the thermal radiation interference of the second station, the cloud collaborative controller generates a heat shielding instruction including a welding torch angle adjustment parameter and a local cooling opening parameter according to the degree and range of the thermal interference, and sends the instruction to the welding control system of the second station to guide the second station to perform a heat shielding operation. For example, when the heat interference source identification result determines that the abnormality of the molten pool temperature of the first station is caused by the thermal radiation interference of the second station, the cloud collaborative controller sends a heat shielding instruction to the second station, wherein the welding torch angle adjustment parameter is to adjust the welding torch by 15° away from the first station, and the local cooling opening parameter is to open the local cooling fan and adjust the cooling air volume to 5 cubic meters per minute.

[0068] Then, the cloud collaborative controller unifies the calibration of the start and end points of the multi-station girth weld according to the fusion weld deviation of each station at the current time to obtain joint consistency calibration parameters. That is, the cloud collaborative controller collects the fusion weld deviation of each station at the current time, and according to the welding process requirements of the tower ring weld, takes the standard start and end points as the reference, combines the deviation of each station to calculate the corresponding calibration adjustment value, and unifies the calibration of the start and end points of each station to finally generate the joint consistency calibration parameters. For example, the cloud collaborative controller collects the fusion weld deviation of the four stations of the tower at the current time as 0.2 mm, 0.15 mm, 0.22 mm, and 0.18 mm, takes the standard start points of the tower ring weld as 0°, 90°, 180°, and 270°, and the standard end points as 89°, 179°, 269°, and 359° as the reference, and combines the deviation of each station to unify the calibration to obtain the joint consistency calibration parameters. The first station start point is calibrated by +0.2 mm, and the end point is calibrated by +0.2 mm. The second station start point is calibrated by +0.15 mm, and the end point is calibrated by +0.15 mm. The third station start point is calibrated by +0.22 mm, and the end point is calibrated by +0.22 mm. The fourth station start point is calibrated by +0.18 mm, and the end point is calibrated by +0.18 mm.

[0069] The heat shielding instruction and the joint consistency calibration parameters are combined to obtain a multi-station collaborative control instruction.

[0070] All the heat shielding instructions issued by the cloud collaborative controller and the calculated joint consistency calibration parameters are summarized and integrated, classified and associated according to the station number, to form a multi-station collaborative control instruction containing the heat shielding operation parameters and joint calibration parameters of each station, and then sent to the welding control system of each station. For example, the heat shielding instruction of adjusting the welding torch angle by 15° and turning on the local cooling with a wind volume of 5m³ / min sent to the second station is combined with the joint consistency calibration parameters of the four stations to form a multi-station collaborative control instruction, which clearly specifies the calibration parameters of the starting and ending points of each station, and specifies the operation requirements of the second station to adjust the welding torch angle and turn on the local cooling.

[0071] In step S500 of the embodiment of the present application, loading the multi-station welding heat source model further includes: Collecting historical welding power data, historical welding torch position coordinates and corresponding historical tower hot imaging data of each station to construct a heat source model training set; Based on the double-ellipsoid heat source distribution function, an initial heat source model containing a heat source superposition factor is constructed, wherein the heat source superposition factor is used to represent the energy coupling effect of the heat sources of adjacent stations; Taking the historical tower hot imaging data as supervision and taking the historical welding power data and historical welding torch position coordinates as input, the initial heat source model is parameterized to obtain the multi-station welding heat source model.

[0072] The purpose of the above steps in step S500 of the embodiment of the present application is to build a precise and practical welding scene fitting heat source model for heat field prediction, heat interference source identification and heat collaborative control of multi-station tower welding. By collecting historical welding data to construct a training set, combining the double-ellipsoid heat source distribution function to build an initial model and completing parameter identification, the model can accurately represent the heat source energy coupling effect between multiple stations, accurately calculate the heat field distribution of tower welding, and provide reliable model support for subsequent multi-station heat interference decoupling and collaborative control.

[0073] To achieve the above purpose, first, the historical welding power data, historical welding torch position coordinates and corresponding historical tower hot imaging data of each station are collected to construct a heat source model training set; That is, collect the welding power data and welding torch position coordinate data of each station during the multi-station tower cylinder welding process, and collect the tower cylinder thermal imaging data corresponding to the two groups of data in time and station. The three types of data are arranged and integrated according to the preset format to form a structured heat source model training set. For example, the historical welding power data of three stations of the tower cylinder welding is 29kW, 31kW and 30kW, the historical welding torch position coordinates are 0°, 120° and 240° at the tower cylinder ring seam, and the corresponding historical tower cylinder thermal imaging data at the corresponding time and station is 1490℃, 1510℃ and 1500℃ at the center of the molten pool. After the data is associated and integrated according to the station, a heat source model training set is constructed.

[0074] Then, based on the double-ellipsoid heat source distribution function, an initial heat source model containing a heat source superposition factor is constructed, wherein the heat source superposition factor is used to represent the energy coupling effect of the heat sources of adjacent stations. That is, based on the double-ellipsoid heat source distribution function, a heat source superposition factor representing the energy coupling effect of the heat sources of adjacent stations is added to the function, the value range and operation logic of the factor are defined, and the spatial layout and heat source characteristics of multi-station welding are combined to build an initial multi-station welding heat source model. For example, based on the double-ellipsoid heat source distribution function, a heat source superposition factor with a value range of 0-1 is introduced, and the factor is set to 0.8 when the distance between the welding torches of adjacent stations is less than 500mm and 0.3 when the distance is greater than 500mm to represent the energy coupling effect of the heat sources of adjacent stations at different distances, and the construction of the initial heat source model is completed.

[0075] Then, the historical tower cylinder thermal imaging data is used as supervision, and the historical welding power data and historical welding torch position coordinates are used as input to perform parameter identification on the initial heat source model to obtain a multi-station welding heat source model.

[0076] That is, the historical welding power data and historical welding torch position coordinates in the heat source model training set are input into the initial heat source model, and the historical tower cylinder thermal imaging data is used as the supervision basis for model training. The double-ellipsoid heat source parameters, heat source superposition factors and other related parameters in the model are continuously iteratively adjusted through the parameter identification algorithm until the error between the heat field data output by the model and the historical tower cylinder thermal imaging data is controlled within the preset accuracy range, and finally a multi-station welding heat source model with optimized parameters is obtained. For example, the historical welding power data 29kW, 31kW and 30kW and the historical welding torch position coordinates 0°, 120° and 240° in the heat source model training set are input into the initial heat source model, and the corresponding historical tower cylinder thermal imaging data 1490℃, 1510℃ and 1500℃ at the center of the molten pool is used as supervision. The model parameters are iteratively adjusted through the parameter identification algorithm to make the error between the molten pool center temperature output by the model and the actual thermal imaging data not more than ±10℃. At this time, the parameter identification is completed and a multi-station welding heat source model is obtained.

[0077] In step S500 of the embodiment of the present application, the step further includes: When the confidence score of each sensor is less than the preset confidence failure threshold, it is determined that the sensor health state set is in a total failure state, an emergency shutdown protection mode is triggered, a shutdown alarm instruction is generated, and the welding process is terminated.

[0078] In step S500 of the embodiment of the present application, the purpose of the above step is to avoid quality problems or safety accidents caused by out-of-control welding in the extreme case that all sensor confidence scores are lower than the failure threshold and there is no valid sensing data. The implementation is to determine that the sensor health state set is in a total failure state, then trigger an emergency shutdown protection mode, generate a shutdown alarm instruction, and immediately terminate the welding process; for example, the confidence failure threshold is set to 60 points, if the confidence scores of the laser vision, electric arc, infrared thermal imager, and microphone array are all lower than 60 points, shutdown is triggered and the welding is terminated.

[0079] In step S500 of the embodiment of the present application, the step further includes: Obtain the welding quality false negative rate caused by sensor failure and the welding quality false positive rate caused by environmental interference in the historical welding process of the tower drum welding scene; Calculate the ratio of the welding quality false negative rate and the welding quality false positive rate, and set it as a confidence threshold adjustment factor; Adjust the preset confidence warning threshold and the preset confidence failure threshold according to the ratio of the preset reference adjustment factor and the confidence threshold adjustment factor.

[0080] In step S500 of the embodiment of the present application, the purpose of the above step is to dynamically adjust the preset confidence warning threshold and the confidence failure threshold based on the historical welding quality detection data of the tower drum welding scene, so that the threshold setting is more suitable for the actual welding environment characteristics and sensor working characteristics, avoiding sensor state misjudgment, welding quality false negative or false positive problems caused by fixed threshold, improving the accuracy of sensor health state judgment, and further ensuring the rationality of fault-tolerant reconstruction control mode triggering and the reliability of welding quality detection.

[0081] To achieve the above purpose, first, the welding quality false negative rate caused by sensor failure and the welding quality false positive rate caused by environmental interference in the historical welding process of the tower drum welding scene need to be obtained; That is, the quality detection records and sensor working state records of the past welding process of the tower drum welding scene are counted, the ratio of the number of cases of welding quality missing reports due to sensor failure to the total number of quality detection cases is screened, the welding quality missing report rate is obtained, and the ratio of the number of cases of welding quality false positives due to environmental interference to the total number of quality detection cases is screened, the welding quality false positive rate is obtained. For example, the nearly 1000 historical welding quality detection records of a certain tower drum welding scene are counted, there are 20 cases of welding quality missing reports due to sensor failure, the welding quality missing report rate is 2%, and there are 10 cases of welding quality false positives due to environmental interference, the welding quality false positive rate is 1%.

[0082] Then, the ratio of the welding quality missing report rate to the welding quality false positive rate is calculated, which is set as a confidence threshold adjustment factor. That is, the obtained welding quality missing report rate value is taken as the dividend, the welding quality false positive rate value is taken as the divisor, and division operation is performed, and the operation result is defined as the confidence threshold adjustment factor. For example, the welding quality missing report rate is 2% and the welding quality false positive rate is 1%, the ratio of 2% to 1% is calculated to obtain the result 2, and the result is set as the confidence threshold adjustment factor.

[0083] Then, the preset confidence warning threshold and the preset confidence failure threshold are adjusted according to the ratio of the preset reference adjustment factor to the confidence threshold adjustment factor.

[0084] That is, a fixed reference adjustment factor is preset, division operation is performed between the reference adjustment factor and the calculated confidence threshold adjustment factor to obtain a threshold adjustment coefficient, and then the coefficient is multiplied by the original preset confidence warning threshold and confidence failure threshold respectively to obtain the adjusted confidence warning threshold and confidence failure threshold. For example, the confidence reference adjustment factor is set to 1, the confidence threshold adjustment factor is calculated to be 2, the ratio of 1 to 2 is calculated to obtain the threshold adjustment coefficient 0.5, the original preset confidence warning threshold is 80 points, and the confidence failure threshold is 60 points. The adjustment coefficient 0.5 is multiplied by the two thresholds respectively to obtain the adjusted confidence warning threshold of 40 points and the confidence failure threshold of 30 points.

[0085] In step S500 of the embodiment of the application, the following steps are further included: Real-time welding progress data and real-time weld reinforcement detection data of each station in the tower drum welding scene are obtained. When the real-time weld reinforcement detection data exceeds the preset reinforcement threshold, the corresponding position is identified as a repair welding area. According to the real-time welding progress data, adjacent stations in an idle state or with leading welding progress are dispatched to generate repair welding coordination instructions and perform online repair welding operations on the repair welding area.

[0086] The purpose of the above steps in step S500 of the embodiment of the present application is to monitor and control the weld reinforcement in real time during the welding of the tower drum. By obtaining the welding progress and weld reinforcement data of each station, the reinforcement exceeding area is accurately identified, and then the online repair welding is carried out in combination with the welding progress of the adjacent station in idle or leading progress, so as to timely correct the welding quality defects and avoid subsequent rework, while fully utilizing the operation resources of multiple stations to ensure the overall quality and operation efficiency of the tower drum welding.

[0087] To achieve the above purpose, first, real-time welding progress data and real-time weld reinforcement detection data of each station under the scenario of tower drum welding are obtained; That is, the welding control system collects the welding completion length, welding ring seam completion angle and other data of each station to form real-time welding progress data, and at the same time, the weld reinforcement actual value of the welded area of each station is collected by the weld detection device to form real-time weld reinforcement detection data. For example, the system collects the real-time welding progress data of four stations of the tower drum welding as follows: the first station completes 60% of the ring seam, the second station completes 70% of the ring seam, the third station completes 90% of the ring seam, and the fourth station is in idle state. At the same time, the real-time weld reinforcement detection data of a certain section of the weld of the first station is 5mm, the weld reinforcement of the second station is 3mm, and the weld reinforcement of the third station is 2.8mm.

[0088] Then, when the real-time weld reinforcement detection data exceeds the preset reinforcement threshold, the corresponding position is identified as a repair welding area. That is, the real-time weld reinforcement detection data of each station collected is compared with the preset reinforcement threshold in value. If the reinforcement detection data of a certain position exceeds the reinforcement threshold, the position is marked and identified as a repair welding area that needs repair welding operation, for example, the preset weld reinforcement threshold of the tower drum welding is 4mm, the real-time weld reinforcement detection data of a certain section of the weld of the first station is 5mm, which exceeds the preset reinforcement threshold of 4mm. Therefore, the position of the section of the weld of the first station is identified as a repair welding area.

[0089] Then, according to the real-time welding progress data, the adjacent station in idle state or with leading welding progress is scheduled to generate a repair welding coordination instruction to perform online repair welding operation on the repair welding area.

[0090] That is, according to the real-time welding progress data of each station, the adjacent stations in the idle state or the welding progress leading position are screened out, the cloud collaborative controller combines the position, range and other information of the repair welding area to generate a repair welding collaborative instruction containing the repair welding position, repair welding process parameters and work station, and the instruction is sent to the scheduled station to perform online repair welding operation on the repair welding area, for example, the first station has a repair welding area, the adjacent fourth station is in the idle state, the third station is in the leading position and completes 90% of the welding progress, the idle fourth station is scheduled to perform repair welding operation, the cloud collaborative controller generates a repair welding collaborative instruction, the repair welding area is the weld segment at 60% of the first station ring seam, the repair welding current is 280A, and the repair welding voltage is 28V, and the fourth station performs online repair welding operation on the repair welding area according to the instruction.

Claims

1. A collaborative optimization analysis method for multi-station tower welding, characterized in that, include: S100: Load multi-modal perception data of tower welding in multi-station welding scenarios. The multi-modal perception data of tower welding includes weld visual perception data, arc sensing perception data, thermal field distribution perception data and welding acoustic perception data. S200: Construct a multimodal fusion evaluation model, input the multimodal sensing data of tower welding into the multimodal fusion evaluation model, obtain the fusion weld deviation and several sensor confidence scores, and the several sensor confidence scores correspond to the multimodal sensing data of tower welding. S300: Based on the confidence scores of several sensors, compare them with the preset confidence warning threshold and the preset confidence failure threshold respectively, dynamically determine the health status of each sensor, and obtain the sensor health status set; S400: Based on the sensor health status set and the fused weld deviation, trigger the corresponding fault-tolerant reconfiguration control mode to obtain the reconfigured welding tracking control parameters; S500: Loads a multi-station welding heat source model, and based on the reconstructed welding tracking control parameters and thermal field distribution sensing data, decouples and coordinates the thermal interference between multiple stations, obtains multi-station collaborative control commands, and executes collaborative control of the tower welding process.

2. The collaborative optimization analysis method for multi-station tower welding according to claim 1, characterized in that, Loading multimodal sensing data of tower welding in multi-station welding scenarios, including: In the scenario of tower welding, the weld groove edge contour data, groove center point deviation data and groove width data collected by the laser vision sensor are added to the weld visual perception data. In the scenario of tower welding, the instantaneous value of welding current, instantaneous value of welding voltage, standard deviation of current, standard deviation of voltage, and short-circuit frequency data collected by the arc sensor are added to the arc sensor sensing data. In the scenario of tower welding, the infrared thermal imager collects the center temperature data of the molten pool, the width data of the molten pool, and the temperature gradient data of the heat-affected zone, and adds the thermal field distribution sensing data. Acquire the frequency band sound pressure level data of the welding process collected by the microphone array in the tower welding scenario, and add it to the welding acoustic perception data.

3. The collaborative optimization analysis method for multi-station tower welding according to claim 1, characterized in that, The construction of a multimodal fusion evaluation model previously included: Collect historical welding process datasets for a preset tower model. The historical welding process datasets include historical weld visual perception data, historical arc sensing perception data, historical thermal field distribution perception data, historical welding acoustic perception data, and the corresponding historical actual weld deviation. Using historical real weld deviation as supervision, and taking historical weld visual perception data, historical arc sensing perception data, historical thermal field distribution perception data, and historical welding acoustic perception data as inputs, and using deviation output nodes and the confidence scores of each sensor as outputs, an initial deep confidence network model is trained to obtain a multimodal fusion evaluation model whose output layer includes output nodes.

4. The collaborative optimization analysis method for multi-station tower welding according to claim 1, characterized in that, Based on the confidence scores of several sensors, the health status of each sensor is dynamically determined by comparing them with preset confidence warning thresholds and preset confidence failure thresholds, thus obtaining a sensor health status set, including: The confidence score of any sensor is compared one by one with the preset confidence warning threshold and the preset confidence failure threshold. When the confidence score of a sensor is greater than or equal to the preset confidence warning threshold, the corresponding sensor is determined to be in a healthy state. When the sensor confidence score is less than the preset confidence warning threshold but greater than or equal to the preset confidence failure threshold, the corresponding sensor is determined to be in an interference state. When the confidence score of a sensor is less than the preset confidence failure threshold, the corresponding sensor is determined to be in a failure state. The results of the health status, interference status, and failure status of all sensors are combined to form a sensor health status set.

5. The collaborative optimization analysis method for multi-station tower welding according to claim 4, characterized in that, Based on the sensor health status set and the fused weld deviation, the corresponding fault-tolerant reconfiguration control mode is triggered to obtain the reconfigured welding tracking control parameters, including: When all sensors in the sensor health status set are determined to be in a healthy state, the full fusion tracking control mode is triggered, and the fusion weld deviation is directly converted into welding tracking control parameters. When the sensor health status is determined to be in an interference state by the visual sensor, and the arc sensor, thermal imager, and acoustic sensor are not determined to be in a failure state, the downgraded fusion tracking control mode is triggered. The weight of the weld visual perception data in the fusion calculation is reduced, while the weight of the arc sensor perception data, thermal field distribution perception data, and welding acoustic perception data in the fusion calculation is increased. Based on the adjusted weights, the downgraded fusion weld deviation is recalculated and converted into welding tracking control parameters. When the sensor health status is determined to be in a failed state by the visual sensor, the vision-free guidance control mode is triggered. Historical weld path trajectory data is loaded, and the weld position is estimated by combining the instantaneous values ​​of welding current, welding voltage, and short-circuit frequency data in the arc sensor data. The estimated weld deviation is then converted into welding tracking control parameters.

6. The collaborative optimization analysis method for multi-station tower welding according to claim 1, characterized in that, A multi-station welding heat source model is loaded. Based on the reconstructed welding tracking control parameters and thermal field distribution sensing data, thermal interference between multiple stations is decoupled and coordinated to obtain multi-station coordinated control commands, including: Input the real-time welding power data and real-time welding torch position data of each station into the multi-station welding heat source model to obtain the prediction data of the overall transient thermal field distribution of the tower. When the temperature data of the molten pool at the first station exceeds the preset temperature anomaly threshold, the abnormal temperature data of the molten pool at the first station is input into the multi-station welding heat source model to back-infer the heat source and obtain the heat interference source identification result. When the thermal interference source identification result determines that the abnormal temperature of the molten pool originates from the thermal radiation interference of the second station, the cloud-based collaborative controller sends a thermal shielding command to the second station. The thermal shielding command includes welding torch angle adjustment parameters and local cooling activation parameters. The cloud-based collaborative controller performs unified calibration of the arc start and end points of the circumferential welds at multiple workstations based on the fusion weld deviation at each workstation at the current moment, thereby obtaining joint consistency calibration parameters. By merging the thermal shielding command and the connector consistency calibration parameters, a multi-station collaborative control command is obtained.

7. The collaborative optimization analysis method for multi-station tower welding according to claim 1, characterized in that, Loading the multi-station welding heat source model also includes: Collect historical welding power data, historical welding torch position coordinates, and corresponding historical tower thermal imaging data for each workstation to construct a heat source model training set. Based on the double ellipsoidal heat source distribution function, an initial heat source model including a heat source superposition factor is constructed, where the heat source superposition factor is used to characterize the energy coupling effect of heat sources in adjacent workstations. Using historical tower thermal imaging data as supervision, and historical welding power data and historical welding torch position coordinates as input, the parameters of the initial heat source model are identified to obtain a multi-station welding heat source model.

8. The collaborative optimization analysis method for multi-station tower welding according to claim 1, characterized in that, Also includes: When the confidence scores of several sensors are all less than the preset confidence failure threshold, the sensor health status set is determined to be in a fully failed state, triggering the emergency shutdown protection mode, generating a shutdown alarm command and terminating the welding process.

9. The collaborative optimization analysis method for multi-station tower welding according to claim 1, characterized in that, Also includes: The historical welding process of tower welding was used to obtain the welding quality false alarm rate due to sensor failure and the welding quality false alarm rate due to environmental interference. Calculate the ratio of the welding quality false alarm rate to the welding quality false alarm rate, and set it as the confidence threshold adjustment factor; Based on the ratio of the preset baseline adjustment factor to the confidence threshold adjustment factor, the preset confidence warning threshold and the preset confidence failure threshold are adjusted.

10. The collaborative optimization analysis method for multi-station tower welding according to claim 1, characterized in that, Also includes: To obtain real-time welding progress data and real-time weld height detection data for each workstation in the tower welding scenario; When the real-time weld reinforcement detection data exceeds the preset reinforcement height threshold, the corresponding location is identified as a repair welding area; Based on real-time welding progress data, adjacent workstations that are idle or ahead in welding progress are scheduled to generate repair welding coordination instructions and perform online repair welding operations on the repair welding area.

Images