An arc additive manufacturing online monitoring and real-time control method and system

Abstract

The application provides an electric arc additive manufacturing online monitoring and real-time control method and system, and relates to the technical field of metal additive manufacturing process monitoring and control. In the manufacturing process, a multi-modal sensor is deployed, electric arc voltage, current, molten pool morphology, temperature field and forming layer geometric feature data are synchronously collected, and original multi-modal data sets are formed under a unified time reference. Through signal processing and image preprocessing, energy input, molten pool morphology, temperature distribution and geometric deviation and other features are extracted, forming state vectors are fused, and deviation amounts are obtained by matching with preset process models. Based on the deviation amounts, an adaptive control algorithm is used to calculate online adjustment amounts of welding current, wire feeding speed, moving trajectory and layer thickness, and the adjustment amounts are converted into control instructions to act on actuators, so that online closed-loop control of the forming process is realized, and forming stability, consistency and geometric precision are improved.

Description

Technical Field

[0001] This invention relates to the field of monitoring and control technology for metal additive manufacturing processes, and in particular to a method and system for online monitoring and real-time control of arc additive manufacturing. Background Technology

[0002] Existing arc additive manufacturing typically uses preset process parameters for forming control, and process monitoring is mostly limited to single signals or manual observation, such as monitoring only the arc voltage and current or observing only the molten pool image. It is difficult to simultaneously characterize the arc energy input, molten pool morphology evolution, temperature field distribution, and geometric deviation of the forming layer on the same time reference, thus making it difficult to identify fluctuations in the forming state and early signs of defects in a timely manner.

[0003] As the requirements for geometric accuracy and consistency of complex metal components increase, the development trend of arc additive manufacturing is to introduce multi-source sensing such as vision, thermal imaging and electrical signals, combine signal processing and image recognition to achieve online quality assessment, and further use the monitoring results for real-time closed-loop control to dynamically adjust welding current, wire feed speed and motion path during the forming process to achieve stable deposition and controllable dimensions.

[0004] However, existing solutions generally suffer from problems such as asynchronous multi-source data, lack of unified state representation in feature extraction and fusion, and disconnect between monitoring and control. This makes it difficult to generate control quantities in real time based on the forming state, and parameter adjustments are delayed or excessive, which can easily lead to cumulative layer height errors, unstable molten pools, forming defects and reduced consistency, thus restricting the stable forming and efficiency improvement of complex components. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the purpose of this invention is to provide an online monitoring and real-time control method and system for arc additive manufacturing, which realizes quantitative characterization and online closed-loop control of the forming state of arc additive manufacturing based on multimodal data fusion, thereby improving the stability, consistency and geometric accuracy of the forming process.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for online monitoring and real-time control in electric arc additive manufacturing includes: In the process of arc additive manufacturing, multimodal sensors are deployed to simultaneously collect arc voltage data, arc current data, molten pool morphology data, temperature field data, and geometric feature data of the forming layer. The data collected by the multimodal sensors are written with a unified time reference timestamp to form the original multimodal dataset. Signal processing and image preprocessing are performed on the original multimodal dataset to obtain a preprocessed dataset for feature calculation; Based on the preprocessed dataset, a feature set characterizing the process parameters and forming state is extracted; the feature set includes at least the energy input features obtained from the arc voltage data and the arc current data, the molten pool morphology features obtained from the molten pool morphology data, the temperature distribution features obtained from the temperature field data, and the geometric deviation features obtained from the forming layer geometric feature data; Feature alignment and fusion are performed on the feature set to generate a forming state vector, and the forming state vector is matched with the target state corresponding to the preset process model to obtain the deviation amount used for control. Based on the aforementioned deviation, an adaptive control algorithm is used to calculate the online adjustment amount; the online adjustment amount includes welding current adjustment, wire feed speed adjustment, movement trajectory adjustment, and layer thickness adjustment; The online adjustment amount is converted into a control command and sent to the actuator, so that the actuator adjusts the welding current, wire feeding speed, movement trajectory and layer thickness online accordingly.

[0007] Preferably, the multimodal sensor includes an arc voltage sensor, an arc current sensor, a molten pool morphology sensor, a temperature field sensor, and a forming layer geometric feature sensor; the arc voltage sensor is used to collect the arc voltage data, the arc current sensor is used to collect the arc current data; the molten pool morphology sensor is used to collect the molten pool morphology data, the temperature field sensor is used to collect the temperature field data; and the forming layer geometric feature sensor is used to collect the forming layer geometric feature data.

[0008] Preferably, the data collected by the multimodal sensor is written with a timestamp based on a unified time base to form an original multimodal dataset, including: Acquire a clock signal and determine the clock signal as the unified time reference; According to the preset synchronous triggering mechanism, the multimodal sensor is triggered to complete a synchronous acquisition at the same acquisition time, and the arc voltage data, arc current data, molten pool morphology data, temperature field data and forming layer geometric feature data corresponding to the same acquisition time are obtained; The timestamps corresponding to the acquisition time under the unified time reference are respectively written into the arc voltage data, the arc current data, the molten pool morphology data, the temperature field data, and the geometric feature data of the forming layer, and the data after writing the timestamps are aggregated to form the original multimodal dataset.

[0009] Preferably, signal processing and image preprocessing are performed on the original multimodal dataset to obtain a preprocessed dataset for feature calculation, including: Anomaly removal and filtering processes are performed on the arc voltage data and the arc current data to obtain preprocessed arc voltage data and preprocessed arc current data. The molten pool morphology data is subjected to denoising and molten pool region extraction to obtain molten pool morphology preprocessed data; The temperature field data is subjected to denoising and temperature calibration to obtain preprocessed temperature field data. The geometric feature data of the forming layer are subjected to denoising and geometric calibration to obtain geometric feature preprocessing data; The preprocessed data of arc voltage, arc current, molten pool morphology, temperature field, and geometric features are combined to obtain the preprocessed dataset.

[0010] Preferably, the method for determining the energy input characteristics includes: The instantaneous power sequence is calculated based on the arc voltage data and arc current data corresponding to the same timestamp to obtain the instantaneous power characteristics; The instantaneous power sequence is integrated within a preset time window corresponding to the timestamp to obtain the energy input characteristics.

[0011] Preferably, the feature set characterizing process parameters and forming state is extracted based on the preprocessed dataset, including: Based on the preprocessed data of the molten pool morphology, the molten pool width feature, molten pool length feature, and molten pool area feature are extracted as the molten pool morphology features; Based on the preprocessed temperature field data, the highest temperature feature and temperature gradient feature are extracted as the temperature distribution feature; Based on the geometric feature preprocessing data, the measured layer thickness and the measured forming width are extracted, and the target layer thickness and the target forming width are determined based on the target state corresponding to the preset process model. The layer thickness deviation feature is obtained based on the difference between the measured layer thickness and the target layer thickness, and the forming width deviation feature is obtained based on the difference between the measured forming width and the target forming width. The layer thickness deviation feature and the forming width deviation feature are used as the geometric deviation feature.

[0012] Preferably, feature alignment and fusion are performed on the feature set to generate a shaped state vector, including: Based on the timestamp, each feature in the feature set is time-aligned to obtain a synchronous feature vector corresponding to the same acquisition time. The synchronization feature vector is normalized to obtain a normalized synchronization feature vector; The normalized synchronous feature vector is weighted and fused based on a preset fusion weight to output the formed state vector; wherein, the preset fusion weight is the weight coefficient corresponding to each feature in the weighted fusion.

[0013] Preferably, the forming state vector is matched with the target state corresponding to the preset process model to obtain the deviation amount used for control, including: Based on the preset process model, a target state vector corresponding to the current timestamp is determined; the target state vector is a vectorized representation of the target state. The distance between the formed state vector and the target state vector is calculated to obtain the deviation vector; The deviation amount is determined based on the deviation vector; the deviation amount includes energy input deviation, molten pool morphology deviation, temperature distribution deviation, and geometric deviation.

[0014] Preferably, based on the deviation, an adaptive control algorithm is used to calculate the online adjustment amount, including: Based on the deviation, a control mapping relationship is established between the deviation and the welding current adjustment, the wire feed speed adjustment, the movement trajectory adjustment, and the layer thickness adjustment; The control parameters in the control mapping relationship are adaptively updated based on the deviation, and the online adjustment amount is output. A preset amplitude constraint and a preset rate of change constraint are applied to the online adjustment amount to obtain the online adjustment amount that satisfies the constraints.

[0015] Preferably, the online adjustment quantity is converted into a control command and sent to the actuator, causing the actuator to adjust the welding current, wire feed speed, movement trajectory, and layer thickness online accordingly, including: The welding current adjustment amount is converted into a welding current control command, the wire feed speed adjustment amount is converted into a wire feed speed control command, the movement trajectory adjustment amount is converted into a movement trajectory control command, and the layer thickness adjustment amount is converted into a layer thickness control command. Each of the control commands is then sent to the actuator to implement the corresponding online adjustment.

[0016] An online monitoring and real-time control system for electric arc additive manufacturing includes: The multimodal synchronous acquisition and timestamp calibration unit is used to deploy multimodal sensors during the arc additive manufacturing process, synchronously acquire arc voltage data, arc current data, molten pool morphology data, temperature field data and forming layer geometric feature data, and write timestamps with a unified time reference to the data acquired by the multimodal sensors to form the original multimodal dataset; The signal processing and image preprocessing unit is used to perform signal processing and image preprocessing on the original multimodal dataset to obtain a preprocessed dataset for feature calculation. The process-state feature extraction unit is used to extract a set of features characterizing process parameters and forming state based on the preprocessed dataset; the feature set includes at least the energy input features obtained from the arc voltage data and the arc current data, the molten pool morphology features obtained from the molten pool morphology data, the temperature distribution features obtained from the temperature field data, and the geometric deviation features obtained from the forming layer geometric feature data. The feature alignment and fusion and state deviation calculation unit is used to perform feature alignment and fusion on the feature set, generate a forming state vector, and match the forming state vector with the target state corresponding to the preset process model to obtain the deviation amount used for control. An adaptive control online adjustment calculation unit is used to calculate the online adjustment amount based on the deviation amount using an adaptive control algorithm; the online adjustment amount includes welding current adjustment amount, wire feed speed adjustment amount, movement trajectory adjustment amount, and layer thickness adjustment amount; The control command generation and execution mechanism drive unit is used to convert the online adjustment amount into control commands and send them to the execution mechanism, so that the execution mechanism can adjust the welding current, wire feeding speed, movement trajectory and layer thickness online accordingly.

[0017] The present invention discloses the following technical effects: This invention simultaneously collects arc voltage, arc current, molten pool morphology, temperature field, and geometric features of the forming layer during the arc additive manufacturing process, and forms an original multimodal dataset under a unified time reference. This enables the energy input state, molten pool evolution state, and geometric forming state to be consistently characterized within the same temporal framework, avoiding the state judgment deviation caused by the asynchronous time of multi-source monitoring information in the prior art, thereby improving the accuracy and usability of process monitoring results.

[0018] This invention transforms complex multi-source raw data into a set of features that can be used for control by performing signal processing and image preprocessing on the original multimodal dataset, and further extracting energy input features, melt pool morphology features, temperature distribution features and geometric deviation features. This enables the changes in process parameters and the fluctuations in forming state to be characterized in the form of quantitative features, overcoming the problem in the prior art that relies on only a single signal or experience judgment and is difficult to fully reflect the forming state.

[0019] This invention generates a unified forming state vector by performing feature alignment and fusion on the feature set, and matches the forming state vector with the target state corresponding to the preset process model to obtain the deviation amount used for control. This allows the degree to which the forming state deviates from the target state to be directly quantified, providing a clear basis for subsequent control and avoiding the defects of lack of direct correlation between monitoring and control and unclear control basis in the prior art.

[0020] Based on the aforementioned deviation, this invention employs an adaptive control algorithm to calculate the online adjustment of welding current, wire feed speed, movement trajectory, and layer thickness. This allows the control parameters to be dynamically updated according to the real-time forming state, thereby reducing problems such as molten pool instability, cumulative layer height error, and decreased geometric accuracy caused by parameter adjustment lag or over-adjustment, and improving the stability and consistency of the arc additive manufacturing process.

[0021] This invention converts the online adjustment amount into control commands and sends them to the actuator, forming a closed-loop adjustment mechanism from data acquisition and state characterization to control feedback. This enables the forming process to correct deviations in real time during continuous manufacturing, effectively reducing the defect rate, improving the forming accuracy and production efficiency of complex metal components, and enhancing the engineering applicability of the electric arc additive manufacturing process. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart of the method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system structure provided in an embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] The purpose of this invention is to provide an online monitoring and real-time control method and system for electric arc additive manufacturing. By combining real-time monitoring, state matching and adaptive adjustment, the probability of forming defects is reduced and the manufacturing quality and production efficiency of complex metal components are improved.

[0026] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Figure 1 The method flowchart provided in the embodiments of the present invention is as follows: Figure 1 As shown, the present invention provides an online monitoring and real-time control method for arc additive manufacturing, comprising: Step 100: Deploy multimodal sensors during the arc additive manufacturing process to simultaneously collect arc voltage data, arc current data, molten pool morphology data, temperature field data, and geometric feature data of the forming layer, and write timestamps with a unified time reference to the data collected by the multimodal sensors to form the original multimodal dataset; Step 200: Perform signal processing and image preprocessing on the original multimodal dataset to obtain a preprocessed dataset for feature calculation; Step 300: Extract a feature set characterizing process parameters and forming state based on the preprocessed dataset; the feature set includes at least the energy input features obtained from arc voltage data and arc current data, the molten pool morphology features obtained from molten pool morphology data, the temperature distribution features obtained from temperature field data, and the geometric deviation features obtained from forming layer geometric feature data; Step 400: Perform feature alignment and fusion on the feature set to generate a forming state vector, and match the forming state vector with the target state corresponding to the preset process model to obtain the deviation amount used for control; Step 500: Based on the deviation, an adaptive control algorithm is used to calculate the online adjustment amount; the online adjustment amount includes the welding current adjustment amount, wire feed speed adjustment amount, movement trajectory adjustment amount, and layer thickness adjustment amount; Step 600: Convert the online adjustment quantity into a control command and send it to the actuator, so that the actuator can adjust the welding current, wire feed speed, movement trajectory and layer thickness online accordingly.

[0028] Specifically, in step 100 of this embodiment, multimodal sensors are deployed during the arc additive manufacturing process, and synchronous acquisition and timestamp writing of multi-source data are completed under a unified time reference. The multimodal sensors include at least an arc voltage sensor for acquiring arc voltage data, an arc current sensor for acquiring arc current data, a molten pool morphology sensor for acquiring molten pool morphology data, a temperature field sensor for acquiring temperature field data, and a forming layer geometric feature sensor for acquiring forming layer geometric feature data. In this embodiment, the time stamp used to identify the data acquisition time is defined as a timestamp, which is used for time alignment and corresponding association of different types of data under the same time reference. Preferably, this embodiment selects a continuous and stable clock signal as the unified time reference and performs timing based on the unified time reference to ensure that different types of data have a consistent time reference relationship; more preferably, the timing resolution of the unified time reference is set to not less than 1 microsecond to meet the time accuracy requirements of arc energy fluctuations and molten pool state changes.

[0029] After establishing a unified time reference, this embodiment triggers the multimodal sensors to complete a synchronous acquisition at the same acquisition time based on a preset synchronous triggering mechanism. The synchronous triggering mechanism refers to this embodiment determining the acquisition trigger time under a unified time reference, and simultaneously sending acquisition trigger signals to the arc voltage sensor, arc current sensor, molten pool morphology sensor, temperature field sensor, and forming layer geometric feature sensor at that acquisition trigger time. This causes each sensor to output corresponding arc voltage data, arc current data, molten pool morphology data, temperature field data, and forming layer geometric feature data at the same acquisition time. To ensure the clarity of the technical meaning of "same acquisition time," this embodiment limits the acquisition time deviation of various types of data to no more than 100 microseconds, and defines various types of data acquired at the same acquisition trigger time as a group of synchronously acquired data. Preferably, this embodiment assigns a unique data group number to each group of synchronously acquired data to characterize the sequential relationship between different acquisition times.

[0030] After obtaining various types of data corresponding to the same acquisition time, this embodiment writes the timestamps corresponding to the acquisition time under a unified time reference into the arc voltage data, arc current data, molten pool morphology data, temperature field data, and forming layer geometric feature data, respectively. The data after writing the timestamps are then aggregated to form the original multimodal dataset. Writing the timestamps refers to adding a corresponding timestamp field to each piece of arc voltage data, arc current data, and each frame of molten pool morphology data, temperature field data, and forming layer geometric feature data, enabling different types of data to be time-aligned based on the timestamps. The original multimodal dataset refers to a data set organized by multiple types of data according to timestamps and data group numbers, including at least arc voltage data sequences, arc current data sequences, molten pool morphology data sequences, temperature field data sequences, and forming layer geometric feature data sequences. Preferably, this embodiment sets the continuous recording duration of one original multimodal dataset to 10 to 300 seconds, and performs an integrity check on the original multimodal dataset after recording to confirm that each data group number corresponds to complete multiple types of data, thereby providing a reliable data foundation for subsequent feature calculations.

[0031] Optionally, in this embodiment, during step 200, signal processing and image preprocessing are performed on the original multimodal dataset formed in step 100 to obtain a preprocessed dataset that can be used for subsequent feature calculation. The original multimodal dataset includes at least arc voltage data, arc current data, molten pool morphology data, temperature field data, and forming layer geometric feature data, all of which carry timestamps under a unified time reference. This embodiment sets corresponding preprocessing procedures for different types of data to eliminate the impact of acquisition noise, abnormal interference, and scale bias on subsequent feature extraction, thereby ensuring that all types of data are comparable and consistent under the same time reference. Preferably, in this embodiment, the processing of the original multimodal dataset is performed sequentially using timestamps as the index, ensuring that the preprocessing results of each type of data maintain a one-to-one correspondence with the original acquisition time.

[0032] For arc voltage and arc current data, this embodiment performs outlier removal and filtering to obtain preprocessed arc voltage and arc current data. Outlier removal involves identifying and removing data points that significantly deviate from the normal range of variation to avoid the impact of transient interference or acquisition errors on subsequent analysis. Preferably, this embodiment identifies and removes data points in a continuous sampling sequence that exceed three times the variation range of adjacent data points. The filtering process suppresses high-frequency noise while retaining effective components reflecting the trend of arc energy changes. Preferably, this embodiment uses a sliding time window to smooth the arc voltage and arc current data. The width of the sliding time window is set to 5 to 20 consecutive sampling points, thereby reducing the impact of noise without weakening the main variation characteristics.

[0033] This embodiment performs corresponding image preprocessing and calibration processing on the molten pool morphology data, temperature field data, and forming layer geometric feature data, respectively. Specifically, denoising and molten pool region extraction processing are performed on the molten pool morphology data to obtain molten pool morphology preprocessing data that can clearly characterize the molten pool outline; preferably, the molten pool region extraction processing includes determining the molten pool region boundary based on grayscale or brightness differences and distinguishing the molten pool region from the background region. Denoising and temperature calibration processing are performed on the temperature field data to eliminate environmental interference and convert the temperature field data into temperature field preprocessing data with a uniform temperature scale; preferably, the temperature calibration processing includes correcting the temperature field data according to a pre-set reference temperature. Denoising and geometric calibration processing are performed on the forming layer geometric feature data to obtain geometric feature preprocessing data that can accurately reflect the actual geometric dimensions of the forming layer; wherein, the geometric calibration processing is used to convert the collected geometric feature data into actual size dimensions. Finally, in this embodiment, the preprocessed data of arc voltage, arc current, molten pool morphology, temperature field, and geometric features are collected to form the preprocessed dataset, providing a unified data foundation for feature extraction in step 300.

[0034] Furthermore, in this embodiment, during step 300, a feature set characterizing the process parameters and forming state is extracted based on the preprocessed dataset obtained in step 200, and the feature set is made to correspond one-to-one with the timestamp under a unified time reference. The feature set includes at least energy input features, molten pool morphology features, temperature distribution features, and geometric deviation features. Specifically, the energy input features characterize the arc energy input level and its change over time; the molten pool morphology features characterize the shape and coverage of the molten pool; the temperature distribution features characterize the thermal state distribution of the molten pool and its neighborhood; and the geometric deviation features characterize the degree of deviation of the geometric dimensions of the forming layer from the target state. This embodiment processes the preprocessed dataset time-by-time using timestamps as indexes, ensuring that each acquisition time outputs the feature set corresponding to that acquisition time, thereby providing a calculable data basis for the alignment, fusion, and control of the forming state in subsequent steps. Preferably, this embodiment uses 1 millisecond to 20 milliseconds as the interval range between data corresponding to two adjacent timestamps to ensure that the features can reflect the rapidly changing dynamic process during arc additive manufacturing.

[0035] Regarding the determination of energy input characteristics, this embodiment constructs an instantaneous power sequence based on the preprocessed arc voltage data and preprocessed arc current data corresponding to the same timestamp, and obtains instantaneous power characteristics from the instantaneous power sequence. The "instantaneous power sequence" refers to a sequence of power values ​​obtained by multiplying the preprocessed arc voltage data and preprocessed arc current data point-by-point within a continuous timestamp range. Point-by-point multiplication means establishing a one-to-one correspondence between the arc voltage and arc current values ​​at the same timestamp, and using their product as the instantaneous power value at that timestamp. To avoid instability in energy assessment due to instantaneous fluctuations, this embodiment performs continuous calculations on the instantaneous power sequence within a short time range. Preferably, the short time range covers at least 10 consecutive timestamps of data points to form instantaneous power characteristics that can characterize the energy change trend. More preferably, this embodiment interpolates missing points using instantaneous power values ​​corresponding to adjacent timestamps to ensure the continuity of the instantaneous power sequence, thereby ensuring the repeatability of the subsequent integration process.

[0036] In further determining the energy input characteristics, this embodiment integrates the instantaneous power sequence within a preset time window corresponding to a timestamp to obtain the energy input characteristics. The "preset time window" refers to a continuous time period centered on or starting from a certain timestamp, used to convert the instantaneous power sequence within that time period into an energy accumulation. Preferably, the length of the preset time window is set to 20 milliseconds to 200 milliseconds to balance the ability to respond to rapid fluctuations and the ability to suppress noise. The "integration" in this embodiment is achieved through discrete accumulation, that is, the instantaneous power values ​​corresponding to each timestamp within the preset time window are accumulated in chronological order and combined with the time interval between adjacent timestamps to convert into an energy accumulation. The time interval is determined by the difference between adjacent timestamps under a unified time reference. Preferably, when the interval between adjacent timestamps remains constant, this embodiment uses a fixed time interval for the conversion; when there are slight changes in the interval between adjacent timestamps, this embodiment uses the actual interval between each adjacent timestamp for the conversion. Through the above processing, this embodiment can output the energy input feature corresponding to the time window at the end of each preset time window, and associate and store the energy input feature with the end timestamp of the preset time window so that it can be aligned with the time reference of the molten pool morphology feature, temperature distribution feature and geometric deviation feature in the future.

[0037] Regarding the determination of molten pool morphology features, temperature distribution features, and geometric deviation features, this embodiment extracts features based on molten pool morphology preprocessing data, temperature field preprocessing data, and geometric feature preprocessing data, respectively, and forms the feature set. Specifically, this embodiment determines the molten pool region boundary for the molten pool morphology preprocessing data corresponding to each timestamp, and calculates the molten pool width feature, molten pool length feature, and molten pool area feature at the same scale. The molten pool width feature is the maximum span of the molten pool region in a preset horizontal direction, the molten pool length feature is the maximum span of the molten pool region in a preset vertical direction, and the molten pool area feature is the actual area corresponding to the number of pixels within the molten pool region after geometric calibration. Preferably, this embodiment sets the conversion ratio between pixels and actual length to 0.02 mm to 0.20 mm per pixel during geometric calibration to ensure that the dimensional dimensions of the molten pool width feature and the molten pool length feature are clearly defined. For the temperature field preprocessing data corresponding to each time stamp, this embodiment extracts the highest temperature feature and calculates the temperature gradient feature. The highest temperature feature is the maximum temperature value in the temperature field preprocessing data, and the temperature gradient feature is the rate of temperature change along a preset direction. Preferably, the preset direction includes the direction along the centerline of the molten pool or a direction perpendicular to the centerline of the molten pool. For the geometric feature preprocessing data corresponding to each time stamp, this embodiment extracts the measured layer thickness and the measured forming width, and determines the target layer thickness and the target forming width based on the target state corresponding to the preset process model. The target layer thickness and the target forming width are given by the preset process model under the current layer number and the current path segment. Subsequently, this embodiment uses the difference between the measured layer thickness and the target layer thickness as the layer thickness deviation feature, and the difference between the measured forming width and the target forming width as the forming width deviation feature. The layer thickness deviation feature and the forming width deviation feature are then written into the feature set as geometric deviation features. Through the above process, this embodiment forms a feature set at each time stamp, including energy input characteristics, melt pool morphology characteristics, temperature distribution characteristics, and geometric deviation characteristics, thereby achieving quantitative characterization of process parameters and forming state.

[0038] As an optional implementation, when constructing an instantaneous power sequence based on preprocessed arc voltage and arc current data corresponding to the same timestamp, this embodiment calculates the corresponding arc voltage and arc current values ​​point-by-point at each timestamp to obtain instantaneous power characteristics. Based on this, the instantaneous power sequence is integrated within a preset time window corresponding to the timestamp to obtain energy input characteristics. Specifically, the instantaneous power characteristics can be determined as follows: When constructing an instantaneous power sequence based on preprocessed arc voltage and arc current data corresponding to the same time segment, this embodiment calculates the corresponding arc voltage and arc current values ​​point-by-point at each time segment to obtain instantaneous power characteristics. Based on this, the instantaneous power sequence is integrated within a preset time window corresponding to the time segment to obtain energy input characteristics. Specifically, the instantaneous power characteristics can be determined as follows: in, For time cut Corresponding instantaneous power characteristics; For the time segment Corresponding arc voltage preprocessing data; For the time segment The corresponding arc current preprocessing data; the time segment Determined by a unified time reference, it is used to identify the time when the corresponding data was collected.

[0039] After obtaining the instantaneous power sequence, this embodiment integrates the instantaneous power sequence within a preset time window corresponding to the time truncation to obtain energy input characteristics. The integration is achieved through discrete accumulation, specifically as follows: in, To cut off by time The energy input characteristics at the termination time; For time cut Corresponding instantaneous power characteristics; The time interval between adjacent time segments is determined by the difference between adjacent time segments under a unified time base. The number of time segments included within the preset time window is used to limit the range of instantaneous power data participating in the integration calculation.

[0040] When extracting molten pool morphology features based on molten pool morphology preprocessing data, this embodiment first determines the molten pool region boundary and maps the pixel scale to the actual size scale under a unified geometric calibration scale, thereby obtaining molten pool width features, molten pool length features, and molten pool area features. Specifically, the molten pool width feature and molten pool length feature can be expressed as follows: in, This refers to the width of the molten pool. Characterized by the length of the molten pool; This represents the maximum pixel span of the molten pool region in the preset horizontal direction. This represents the maximum pixel span of the molten pool region in the preset vertical direction. This is a geometric calibration coefficient between pixels and actual length, used to convert pixel scale to actual size scale.

[0041] Based on this, this embodiment maps the number of pixels within the molten pool region to the actual area to obtain the molten pool area feature. The calculation method is as follows: in, Characteristics of the molten pool area; This represents the number of pixels within the molten pool area. The geometric calibration coefficient is used to complete the mapping from pixel area to actual area.

[0042] When extracting temperature distribution features based on temperature field preprocessing data, this embodiment takes the maximum value of all effective temperature points in the temperature field as the highest temperature feature, and calculates the rate of change between adjacent temperature points in a preset direction to characterize the temperature gradient feature. Its representation is as follows: in, It is a temperature gradient feature; This represents the highest temperature value in the temperature field preprocessing data. The lowest temperature value in the direction corresponding to the highest temperature value; The spatial distance between the highest temperature value and the lowest temperature value is determined by the geometrically calibrated spatial scale. In determining geometric deviation features based on geometric feature preprocessing data, this embodiment uses the difference between the measured layer thickness and the target layer thickness as the layer thickness deviation feature, and the difference between the measured forming width and the target forming width as the forming width deviation feature. Their calculation methods are as follows: in, This is a characteristic of layer thickness deviation; The measured layer thickness is obtained from geometric feature preprocessing data; The target layer thickness is determined by the target state corresponding to the preset process model; This refers to the characteristic of forming width deviation; To measure the forming width; The target forming width is determined by the preset process model.

[0043] Furthermore, in step 400 of this embodiment, when performing "time alignment of each feature in the feature set based on the timestamp to obtain the synchronous feature vector corresponding to the same acquisition time", the energy input feature, molten pool morphology feature, temperature distribution feature and geometric deviation feature extracted under the same timestamp are concatenated in a fixed order to form a synchronous feature vector, represented as: in, For timestamps The corresponding synchronization feature vector; A timestamp under a unified time base; The energy input feature corresponding to the timestamp; , , These are the melt pool width feature, melt pool length feature, and melt pool area feature corresponding to the timestamp, respectively. The highest temperature feature corresponding to the timestamp; The temperature gradient feature corresponding to the timestamp; The layer thickness deviation feature corresponding to the timestamp; This refers to the forming width deviation feature corresponding to the timestamp. In this embodiment, when performing "normalization processing on the synchronization feature vector to obtain a normalized synchronization feature vector", to avoid the influence of differences in the dimensions of different features on the fusion result, the components of the synchronization feature vector are subjected to minimum-maximum normalization to obtain a normalized synchronization feature vector: in, For timestamps Next Normalization results of each feature component; Synchronization feature vector The One component; and The first The minimum and maximum values ​​of each feature component within a preset statistical interval, wherein the preset statistical interval is preferably the data corresponding to the most recent 200 consecutive timestamps; This is the normalized synchronization feature vector. In this embodiment, when performing "weighted fusion of the normalized synchronization feature vector based on preset fusion weights to output the formed state vector", the normalized synchronization feature vector is linearly weighted according to the preset fusion weights to obtain the formed state vector, which is expressed as: in, For timestamps The corresponding forming state vector; For timestamps The corresponding normalized synchronization feature vector; To preset the fusion weight matrix, It consists of the weight coefficients corresponding to each feature in the weighted fusion, and In this embodiment, a constant matrix that is predetermined and maintained and does not change with the timestamp is used to map the normalized synchronization feature vector to the shaped state vector space.

[0044] In this embodiment, when performing the step of "matching the forming state vector with the target state corresponding to the preset process model to obtain the deviation for control," the target state vector corresponding to the current time segment is first determined based on the preset process model. Then, a distance metric is performed to obtain the deviation vector, and the deviation amount is given accordingly. The target state vector and the deviation vector are respectively represented as follows: in, For the preset process model at timestamp The corresponding target state vector; For timestamps The corresponding forming state vector; For timestamps The corresponding deviation vector; The distance metric result is used to characterize the overall deviation between the formed state vector and the target state vector; Let be the dimension of the target state vector, and be the same as . The dimensions are consistent. When determining the deviation based on the deviation vector, this embodiment decomposes the deviation vector into energy input deviation, molten pool morphology deviation, temperature distribution deviation, and geometric deviation according to the feature source, and uses the absolute value or norm of the corresponding component as the deviation, expressed as: in, This refers to the energy input deviation. This refers to the deviation in molten pool morphology. This refers to the temperature distribution deviation. This refers to the geometric deviation. Deviation vector The Each component, and The deviation component corresponding to the energy input characteristics, to The deviation components corresponding to the molten pool width characteristics, molten pool length characteristics, and molten pool area characteristics. and The deviation component corresponding to the highest temperature characteristic and the temperature gradient characteristic. and The deviation components of the corresponding layer thickness deviation characteristics and forming width deviation characteristics.

[0045] In this embodiment, during step 500, an adaptive control algorithm is used to calculate online adjustment quantities based on the deviation obtained in step 400. These online adjustment quantities include welding current adjustment, wire feed speed adjustment, movement trajectory adjustment, and layer thickness adjustment. This embodiment first establishes a control mapping relationship between the deviation and each online adjustment quantity. The energy input deviation, molten pool morphology deviation, temperature distribution deviation, and geometric deviation are used as control inputs, while the welding current adjustment, wire feed speed adjustment, movement trajectory adjustment, and layer thickness adjustment are used as control outputs. This allows the deviation to be converted into correction quantities for process parameters. In this embodiment, the control mapping relationship is defined as a control parameterization relationship. This relationship characterizes the response of deviation changes to each online adjustment quantity and provides the updated objects for subsequent adaptive updates. To ensure the control mapping relationship is executable from the initial stage, preferably, this embodiment sets the initial control parameters of the control mapping relationship to such that each online adjustment quantity outputs zero when the deviation is zero and exhibits a monotonically increasing adjustment trend as the deviation increases.

[0046] After obtaining the control mapping relationship, this embodiment adaptively updates the control parameters in the control mapping relationship based on the deviation amount to output an online adjustment amount. A preset amplitude constraint and a preset rate of change constraint are applied to the output online adjustment amount to obtain an online adjustment amount that satisfies the constraints. The adaptive update refers to updating the control parameters of the control parameterization relationship in each control cycle based on the change in deviation amount between the current control cycle and the previous control cycle. This allows the control mapping relationship to dynamically adjust as the forming state changes, thereby reducing the risk of parameter adjustment lag or over-adjustment. The preset amplitude constraint is used to limit the maximum allowable adjustment range of the welding current adjustment, wire feed speed adjustment, movement trajectory adjustment, and layer thickness adjustment. Preferably, the upper limit of the amplitude of the welding current adjustment is set to no more than 50 amperes, the upper limit of the amplitude of the wire feed speed adjustment is set to no more than 2 meters per minute, the upper limit of the amplitude of the layer thickness adjustment is set to no more than 0.5 millimeters, and the upper limit of the amplitude of the movement trajectory adjustment is set to no more than 1 millimeter. The preset rate of change constraint is used to limit the rate of change of the online adjustment within adjacent control cycles. Preferably, the upper limit of the rate of change of the welding current adjustment is set to no more than 10 amperes per second, the upper limit of the rate of change of the wire feed speed adjustment is set to no more than 0.5 meters per minute per second, the upper limit of the rate of change of the layer thickness adjustment is set to no more than 0.1 millimeters per second, and the upper limit of the rate of change of the movement trajectory adjustment is set to no more than 0.2 millimeters per second, so that the obtained online adjustment satisfies responsiveness while avoiding introducing abrupt disturbances to the forming process.

[0047] As an example, in this embodiment, when performing the "establishment of a control mapping relationship between the deviation amount and the welding current adjustment amount, wire feed speed adjustment amount, movement trajectory adjustment amount, and layer thickness adjustment amount based on the deviation amount", the deviation amounts obtained in step 400 are used as control inputs, and an online adjustment amount vector is generated in a linear mapping manner, which is represented as follows: in, For timestamps The corresponding online adjustment vector consists of the following components, in order: welding current adjustment, wire feed speed adjustment, movement trajectory adjustment, and layer thickness adjustment. For timestamps The corresponding deviation vector has components that are, in order, energy input deviation, molten pool morphology deviation, temperature distribution deviation, and geometric deviation. The control mapping matrix is ​​used to characterize the mapping relationship between each deviation and each online adjustment. In this embodiment, the control mapping matrix is ​​initially set to a diagonal or quasi-diagonal form to ensure that different types of deviations preferentially act on the corresponding process parameter adjustment.

[0048] When performing the action of "adaptively updating the control parameters in the control mapping relationship based on the deviation and imposing constraints on the online adjustment," this embodiment updates the control mapping matrix cycle by cycle and imposes amplitude and rate-of-change restrictions on the online adjustment after the update, as shown below: in, This is the control mapping matrix corresponding to the previous timestamp; The adaptive update step size is used to limit the update magnitude of the control mapping matrix; This is the online adjustment vector after applying amplitude constraints; This is the maximum allowable amplitude vector for the online adjustment quantity; This is the final online adjustment vector after applying the rate of change constraint; This is the vector of the maximum permissible rate of change of the online adjustment quantity within adjacent control cycles; This indicates amplitude saturation calculation performed on a component basis; This indicates a rate-of-change constraint calculation performed on a component basis. Through the above processing, this embodiment obtains an online adjustment amount that satisfies stability constraints while responding to changes in the deviation amount, which is used for real-time adjustment of welding current, wire feed speed, movement trajectory and layer thickness in subsequent steps.

[0049] In this embodiment, during step 600, the online adjustment quantity obtained in step 500 is converted into a control command that can be recognized and executed by the actuator, and the control command is sent to the actuator to achieve online adjustment of welding current, wire feed speed, movement trajectory, and layer thickness. The online adjustment quantity includes welding current adjustment, wire feed speed adjustment, movement trajectory adjustment, and layer thickness adjustment. Each online adjustment quantity is calculated by an adaptive control algorithm in the previous control cycle and satisfies amplitude and rate of change constraints. In this embodiment, a "control command" is defined as parameterized instruction information used to characterize the adjustment result of the target process parameters. The control command is used to instruct the actuator to achieve the target adjustment state in the current control cycle, thereby ensuring a one-to-one correspondence between the actuator's actions and the online adjustment quantities.

[0050] Specifically, this embodiment converts welding current adjustment amounts into welding current control commands, which instruct the actuator to adjust the welding current relative to the current welding current setpoint within the current control cycle; it also converts wire feed speed adjustment amounts into wire feed speed control commands, which instruct the actuator to adjust the wire feed speed relative to the current wire feed speed setpoint within the current control cycle; it converts movement trajectory adjustment amounts into movement trajectory control commands, which instruct the actuator to perform position offset or path correction based on the current movement path; and it converts layer thickness adjustment amounts into layer thickness control commands, which instruct the actuator to adjust deposition conditions to converge the layer thickness of the subsequent forming layer towards the target state. Through these methods, this embodiment ensures that each online adjustment amount is converted into a semantically clear and uniquely targeted control command, avoiding ambiguity in the control commands during execution.

[0051] After generating the control commands, this embodiment sends each control command to the actuator to implement online adjustments accordingly. The actuator includes an arc energy adjustment component that responds to welding current control commands, a wire feeding adjustment component that responds to wire feed speed control commands, a motion execution component that responds to movement trajectory control commands, and a forming adjustment component that responds to layer thickness control commands. Each actuator completes the corresponding parameter adjustment within the current control cycle based on the received control commands. This embodiment continuously converts the online adjustment quantity into control commands and applies them to the actuators by cyclically executing steps 100 to 600 during the continuous forming process, thereby forming a closed-loop adjustment process based on the forming state as feedback, achieving real-time coordinated control of welding current, wire feed speed, movement trajectory, and layer thickness during arc additive manufacturing.

[0052] Corresponding to the above methods, such as Figure 2 As shown, this embodiment also provides an online monitoring and real-time control system for electric arc additive manufacturing, including: The multimodal synchronous acquisition and timestamp calibration unit is used to deploy multimodal sensors in the arc additive manufacturing process to synchronously acquire arc voltage data, arc current data, molten pool morphology data, temperature field data and forming layer geometric feature data, and write timestamps with a unified time reference to the data acquired by the multimodal sensors to form the original multimodal dataset; The signal processing and image preprocessing unit is used to perform signal processing and image preprocessing on the original multimodal dataset to obtain a preprocessed dataset for feature calculation. The process-state feature extraction unit is used to extract a set of features characterizing process parameters and forming state based on the preprocessed dataset. The feature set includes at least the energy input features obtained from arc voltage data and arc current data, the molten pool morphology features obtained from molten pool morphology data, the temperature distribution features obtained from temperature field data, and the geometric deviation features obtained from forming layer geometric feature data. The feature alignment and fusion and state deviation calculation unit is used to perform feature alignment and fusion on the feature set, generate a forming state vector, and match the forming state vector with the target state corresponding to the preset process model to obtain the deviation amount used for control. The adaptive control online adjustment calculation unit is used to calculate the online adjustment based on the deviation using an adaptive control algorithm. The online adjustment includes welding current adjustment, wire feed speed adjustment, movement trajectory adjustment, and layer thickness adjustment. The control command generation and execution mechanism drive unit is used to convert online adjustment quantities into control commands and send them to the execution mechanism, so that the execution mechanism can adjust the welding current, wire feeding speed, movement trajectory and layer thickness online accordingly.

[0053] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0054] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for online monitoring and real-time control in electric arc additive manufacturing, characterized in that, include: In the process of arc additive manufacturing, multimodal sensors are deployed to simultaneously collect arc voltage data, arc current data, molten pool morphology data, temperature field data, and geometric feature data of the forming layer. The data collected by the multimodal sensors are written with a unified time reference timestamp to form the original multimodal dataset. Signal processing and image preprocessing are performed on the original multimodal dataset to obtain a preprocessed dataset for feature calculation; Based on the preprocessed dataset, a feature set characterizing process parameters and forming state is extracted; The feature set includes at least the energy input feature obtained from the arc voltage data and the arc current data, the molten pool morphology feature obtained from the molten pool morphology data, the temperature distribution feature obtained from the temperature field data, and the geometric deviation feature obtained from the forming layer geometric feature data; Feature alignment and fusion are performed on the feature set to generate a forming state vector, and the forming state vector is matched with the target state corresponding to the preset process model to obtain the deviation amount used for control. Based on the aforementioned deviation, an adaptive control algorithm is used to calculate the online adjustment amount; the online adjustment amount includes welding current adjustment, wire feed speed adjustment, movement trajectory adjustment, and layer thickness adjustment; The online adjustment amount is converted into a control command and sent to the actuator, so that the actuator adjusts the welding current, wire feeding speed, moving trajectory and layer thickness online accordingly. The forming state vector is matched with the target state corresponding to the preset process model to obtain the deviation amount used for control, including: Based on the preset process model, a target state vector corresponding to the current timestamp is determined; the target state vector is a vectorized representation of the target state. The distance between the formed state vector and the target state vector is calculated to obtain the deviation vector; The deviation amount is determined based on the deviation vector; the deviation amount includes energy input deviation, molten pool morphology deviation, temperature distribution deviation, and geometric deviation. Based on the aforementioned deviation, an adaptive control algorithm is used to calculate the online adjustment amount, including: Based on the deviation, a control mapping relationship is established between the deviation and the welding current adjustment, the wire feed speed adjustment, the movement trajectory adjustment, and the layer thickness adjustment; The control parameters in the control mapping relationship are adaptively updated based on the deviation, and the online adjustment amount is output. A preset amplitude constraint and a preset rate of change constraint are applied to the online adjustment amount to obtain the online adjustment amount that satisfies the constraints; Each deviation is used as a control input, and an online adjustment vector is generated using a linear mapping method. The control mapping matrix is ​​used to characterize the mapping relationship between each deviation and each online adjustment. The control mapping matrix is ​​initially set to a diagonal or quasi-diagonal form to ensure that different types of deviations preferentially act on the corresponding process parameter adjustments. In each control cycle, the control parameters of the control parameterization relationship are updated according to the change of the deviation between the current deviation and the deviation in the previous control cycle, and the control mapping matrix is ​​updated cycle by cycle.

2. The online monitoring and real-time control method for arc additive manufacturing according to claim 1, characterized in that, The multimodal sensor includes an arc voltage sensor, an arc current sensor, a molten pool morphology sensor, a temperature field sensor, and a forming layer geometric feature sensor; the arc voltage sensor is used to collect the arc voltage data, the arc current sensor is used to collect the arc current data; the molten pool morphology sensor is used to collect the molten pool morphology data, the temperature field sensor is used to collect the temperature field data; and the forming layer geometric feature sensor is used to collect the forming layer geometric feature data.

3. The online monitoring and real-time control method for arc additive manufacturing according to claim 1, characterized in that, The data collected by the multimodal sensor is written with a timestamp based on a unified time base to form the original multimodal dataset, including: Acquire a clock signal and determine the clock signal as the unified time reference; According to the preset synchronous triggering mechanism, the multimodal sensor is triggered to complete a synchronous acquisition at the same acquisition time, and the arc voltage data, arc current data, molten pool morphology data, temperature field data and forming layer geometric feature data corresponding to the same acquisition time are obtained; The timestamps corresponding to the acquisition time under the unified time reference are respectively written into the arc voltage data, the arc current data, the molten pool morphology data, the temperature field data, and the geometric feature data of the forming layer, and the data after writing the timestamps are aggregated to form the original multimodal dataset.

4. The online monitoring and real-time control method for arc additive manufacturing according to claim 1, characterized in that, Signal processing and image preprocessing are performed on the original multimodal dataset to obtain a preprocessed dataset for feature calculation, including: Anomaly removal and filtering processes are performed on the arc voltage data and the arc current data to obtain preprocessed arc voltage data and preprocessed arc current data. The molten pool morphology data is subjected to denoising and molten pool region extraction to obtain molten pool morphology preprocessed data; The temperature field data is subjected to denoising and temperature calibration to obtain preprocessed temperature field data. The geometric feature data of the forming layer are subjected to denoising and geometric calibration to obtain geometric feature preprocessing data; The preprocessed data of arc voltage, arc current, molten pool morphology, temperature field, and geometric features are combined to obtain the preprocessed dataset.

5. The online monitoring and real-time control method for arc additive manufacturing according to claim 1, characterized in that, The methods for determining the energy input characteristics include: The instantaneous power sequence is calculated based on the arc voltage data and arc current data corresponding to the same timestamp to obtain the instantaneous power characteristics; The instantaneous power sequence is integrated within a preset time window corresponding to the timestamp to obtain the energy input characteristics.

6. The online monitoring and real-time control method for arc additive manufacturing according to claim 4, characterized in that, Based on the preprocessed dataset, a feature set characterizing process parameters and forming state is extracted, including: Based on the preprocessed data of the molten pool morphology, the molten pool width feature, molten pool length feature, and molten pool area feature are extracted as the molten pool morphology features; Based on the preprocessed temperature field data, the highest temperature feature and temperature gradient feature are extracted as the temperature distribution feature; Based on the geometric feature preprocessing data, the measured layer thickness and the measured forming width are extracted, and the target layer thickness and the target forming width are determined based on the target state corresponding to the preset process model. The layer thickness deviation feature is obtained based on the difference between the measured layer thickness and the target layer thickness, and the forming width deviation feature is obtained based on the difference between the measured forming width and the target forming width. The layer thickness deviation feature and the forming width deviation feature are used as the geometric deviation feature.

7. The online monitoring and real-time control method for arc additive manufacturing according to claim 1, characterized in that, Perform feature alignment and fusion on the feature set to generate a shaped state vector, including: Based on the timestamp, each feature in the feature set is time-aligned to obtain a synchronous feature vector corresponding to the same acquisition time. The synchronization feature vector is normalized to obtain a normalized synchronization feature vector; The normalized synchronous feature vector is weighted and fused based on a preset fusion weight to output the formed state vector; wherein, the preset fusion weight is the weight coefficient corresponding to each feature in the weighted fusion.

8. An online monitoring and real-time control system for electric arc additive manufacturing, characterized in that, The system for implementing the method as described in any one of claims 1 to 7 comprises: The multimodal synchronous acquisition and timestamp calibration unit is used to deploy multimodal sensors during the arc additive manufacturing process, synchronously acquire arc voltage data, arc current data, molten pool morphology data, temperature field data and forming layer geometric feature data, and write timestamps with a unified time reference to the data acquired by the multimodal sensors to form the original multimodal dataset; The signal processing and image preprocessing unit is used to perform signal processing and image preprocessing on the original multimodal dataset to obtain a preprocessed dataset for feature calculation. The process-state feature extraction unit is used to extract a set of features characterizing process parameters and forming state based on the preprocessed dataset; the feature set includes at least the energy input features obtained from the arc voltage data and the arc current data, the molten pool morphology features obtained from the molten pool morphology data, the temperature distribution features obtained from the temperature field data, and the geometric deviation features obtained from the forming layer geometric feature data. The feature alignment and fusion and state deviation calculation unit is used to perform feature alignment and fusion on the feature set, generate a forming state vector, and match the forming state vector with the target state corresponding to the preset process model to obtain the deviation amount used for control. An adaptive control online adjustment calculation unit is used to calculate the online adjustment amount based on the deviation amount using an adaptive control algorithm; the online adjustment amount includes welding current adjustment amount, wire feed speed adjustment amount, movement trajectory adjustment amount, and layer thickness adjustment amount; The control command generation and execution mechanism drive unit is used to convert the online adjustment amount into control commands and send them to the execution mechanism, so that the execution mechanism can adjust the welding current, wire feeding speed, movement trajectory and layer thickness online accordingly.

Images