Dissimilar material welding interface control method and system based on ultrasonic and negative pressure cooperative regulation and control

Abstract

The invention relates to the technical field of dissimilar material welding interface control, and particularly discloses a dissimilar material welding interface control method and system based on ultrasonic and negative pressure coordinated regulation, and the method comprises the following steps: S1, aligning and assembling the to-be-welded surfaces of a first dissimilar material workpiece and a second dissimilar material workpiece, the welding rod is arranged in a closed welding chamber; s2, the welding cavity is vacuumized, so that a preset negative pressure environment is formed and maintained in the welding cavity; s3, in the negative pressure environment, preset static pressure is applied to a workpiece welding area; and S4, applying the static pressure at the same time. According to the dissimilar material welding interface control method and system based on ultrasonic and negative pressure cooperative regulation and control, the production cost is directly reduced through nearly zero-defect manufacturing, the service life of a product is prolonged due to the fact that the quality of a connector is greatly improved, and a key technical support is provided for reliable connection of dissimilar materials in the fields of aerospace, new energy automobiles and the like.

Description

Technical Field

[0001] This invention relates to the field of dissimilar material welding interface control technology, specifically a dissimilar material welding interface control method and system based on the coordinated regulation of ultrasound and negative pressure. Background Technology

[0002] Welding of dissimilar materials is a core technology in advanced manufacturing fields such as aerospace and new energy vehicles. However, the inherent differences in their physical and chemical properties make the weld interface prone to defects such as brittle compounds, porosity, and microcracks, which seriously restricts the quality of the joint. The synergistic application of ultrasonic-assisted welding and negative pressure environment, which breaks the oxide film through high-frequency vibration and combines it with vacuum to suppress high-temperature oxidation, provides an effective path to improve interface quality. However, existing technologies still face serious challenges: First, process control heavily relies on preset parameters and lacks real-time perception and response capabilities to dynamic changes during welding, resulting in control lag; second, quality assessment relies on post-weld non-destructive testing, which can only achieve "post-screening" rather than "in-process intervention," resulting in resource waste; more importantly, the welding process involves complex coupling of multiple physical fields, which traditional models cannot accurately describe and predict. Although some studies have attempted to introduce single sensors such as infrared thermometry or purely data-driven machine learning models for monitoring and early warning, these methods either lack generalization ability due to the lack of physical mechanism guidance or fail to capture weak precursors of defect initiation due to shallow feature representation, resulting in limited predictive ability and an inability to provide an effective early warning window of ≥0.5 seconds to support active control. Therefore, developing an intelligent control method that can deeply integrate physical mechanisms and multimodal sensing data to achieve accurate and advanced defect prediction has become an urgent need to break through existing technological bottlenecks and promote the welding of dissimilar materials from "passive control" to "active intelligent regulation". Summary of the Invention

[0003] To solve the above technical problems, the present invention is implemented through the following technical solution: a method and system for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure, comprising the following steps: Step S1: Align and assemble the surfaces of the first dissimilar material workpiece and the second dissimilar material workpiece to be welded, and place them in a sealed welding chamber. Step S2: Evacuate the welding chamber to create and maintain a preset negative pressure environment inside; Step S3: Under the negative pressure environment, apply a preset static pressure to the welding area of ​​the workpiece; Step S4: While applying the static pressure, apply high-frequency ultrasonic vibration energy to the welding area of ​​the workpiece; Step S5: During the welding process, dynamic information of the welding process is collected in real time through a multi-sensor fusion platform; Step S6: Input the dynamic information into the pre-trained deep learning model to predict in real time the probability of defects occurring at the welding interface within a preset time window in the future. Step S7: When the predicted defect probability exceeds a preset threshold, generate an early warning signal and / or adjust at least one weld in real time.

[0004] Preferably, in step S5, the multi-sensor fusion platform includes: A high-speed vision sensor with a frame rate of no less than 1000fps is used to capture the dynamics of the molten pool, plastic flow, and spatter behavior in the welding area. Infrared temperature sensor, with a temperature measurement range covering 300℃ to 1800℃, is used to monitor the temperature field distribution at the welding interface in real time. The spectral analysis sensor, with an analysis band covering 200nm to 900nm, is used to detect changes in plasma electron temperature, density, and spectral intensity of characteristic elements during the welding process.

[0005] Preferably, in step S6, the pre-trained deep learning model is a neural network model based on the LSTM-CNN dual attention mechanism; The CNN module in the neural network model based on the LSTM-CNN dual attention mechanism is used to extract spatial features from the high-speed visual image and the infrared temperature field image; The LSTM module in the neural network model based on the LSTM-CNN dual attention mechanism is used to learn the dynamic evolution of the multi-channel sensing data over time. The dual attention mechanism in the neural network model based on LSTM-CNN includes a spatial attention module and a temporal attention module, which are used to adaptively weight the key spatial regions and key temporal nodes that contribute the most to defect prediction.

[0006] Preferably, in step S6, the model can predict micron-level defects such as pores and cracks, and the prediction time is not less than 0.5 seconds earlier than the actual occurrence time of the defect.

[0007] Preferably, the deep learning model described in step S6 can also be replaced by a spatiotemporal physical information fusion Transformer prediction model, and the specific working process is as follows: Multimodal physical feature embedding includes physical field reconstruction, physical constraint encoding, heterogeneous graph construction, and graph attention fusion; Causal convolution and look-ahead prediction, including time series prediction; Uncertainty quantification and decision output, including probability and confidence level output.

[0008] Preferably, in step S7, the welding process parameters adjusted in real time include one or more of the following: the magnitude of the static pressure, the amplitude, power, or duration of the high-frequency ultrasonic vibration energy.

[0009] A control system for dissimilar material welding interfaces based on the coordinated regulation of ultrasound and negative pressure, comprising: A welding chamber, the interior of which is used to accommodate and fix the first dissimilar material workpiece and the second dissimilar material workpiece; A vacuum system, connected to the welding chamber, is used to evacuate the welding chamber to create and maintain the negative pressure environment. A pressure application device, the pressure head of which extends into the welding chamber, is used to apply the static pressure to the welding area of ​​the workpiece; An ultrasonic energy device, wherein the ultrasonic transducer is coupled to the pressure head, is used to generate and transmit the high-frequency ultrasonic vibration energy to the workpiece welding area. A multi-sensor fusion platform, integrated into the welding chamber, is used to collect dynamic information of the welding process in real time. It includes at least a high-speed vision sensor, an infrared temperature sensor, and a spectral analysis sensor. AI prediction and control unit, including: The data processing module is used to receive and synchronously fuse multimodal data from the multi-sensor fusion platform; The deep learning model module loads a pre-trained neural network model based on the LSTM-CNN dual attention mechanism, which is used to perform real-time prediction of defect probability based on the fused data. The decision control module is used to generate early warning signals based on the prediction results and / or send parameter adjustment instructions to the pressure application device and the ultrasonic energy device; The central controller is electrically connected to the vacuum system, pressure application device, ultrasonic energy device, and AI prediction and control unit, and is used to coordinate and control the operation of the entire system.

[0010] Preferably, the high-speed vision sensor, infrared temperature sensor, and spectral analysis sensor are aligned with the workpiece welding area through an observation window or a specially designed sealed interface on the welding chamber.

[0011] Preferably, the AI ​​prediction and control unit exchanges data in real time with the central controller, pressure application device and ultrasonic energy device via fieldbus or industrial Ethernet to ensure that the response time of the control loop is less than 50 milliseconds.

[0012] It has the following beneficial effects: By combining a multi-sensor fusion platform with a deep learning model, this system achieves, for the first time, accurate pre-production prediction (≥0.5 seconds in advance) of micron-level defects at welding interfaces. This breaks through the limitations of traditional post-production detection, enabling proactive intervention and significantly improving product quality and consistency. The spatiotemporal physical information fusion Transformer model, by embedding physical laws such as mass conservation and energy conservation into the learning process, overcomes the shortcomings of weak generalization ability and poor interpretability of purely data-driven models, achieving high-precision and reliable prediction. This provides profound insights for process optimization. The system constructs a complete intelligent closed loop of "perception-decision-execution" by quantifying prediction uncertainties and making hierarchical decisions. It responds promptly and controls precisely, greatly improving the level and efficiency of production automation. This solution not only directly reduces production costs through near-zero-defect manufacturing but also extends product life due to the extreme improvement in joint quality, providing key technical support for reliable connection of dissimilar materials in aerospace, new energy vehicles, and other fields. Detailed Implementation

[0013] The technical solutions in the embodiments of the present invention will be clearly and completely described below. 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.

[0014] In a first embodiment, the present invention provides a technical solution: a control system for dissimilar material welding interfaces based on the coordinated regulation of ultrasound and negative pressure, comprising: A welding chamber, the interior of which is used to accommodate and fix the first dissimilar material workpiece and the second dissimilar material workpiece; A vacuum system, connected to the welding chamber, is used to evacuate the welding chamber to create and maintain the negative pressure environment. A pressure application device, the pressure head of which extends into the welding chamber, is used to apply the static pressure to the welding area of ​​the workpiece; An ultrasonic energy device, wherein the ultrasonic transducer is coupled to the pressure head, is used to generate and transmit the high-frequency ultrasonic vibration energy to the workpiece welding area. A multi-sensor fusion platform, integrated on the welding chamber, is used to collect dynamic information of the welding process in real time. It includes at least a high-speed vision sensor, an infrared temperature sensor, and a spectral analysis sensor. The high-speed vision sensor, infrared temperature sensor, and spectral analysis sensor are aligned with the workpiece welding area through an observation window or a specially designed sealed interface on the welding chamber. AI prediction and control unit, including: The data processing module is used to receive and synchronously fuse multimodal data from the multi-sensor fusion platform; The deep learning model module loads a pre-trained neural network model based on the LSTM-CNN dual attention mechanism, which is used to perform real-time prediction of defect probability based on the fused data. The decision control module is used to generate early warning signals based on the prediction results and / or send parameter adjustment instructions to the pressure application device and the ultrasonic energy device; The central controller is electrically connected to the vacuum system, pressure application device, ultrasonic energy device, and AI prediction and control unit, and is used to coordinate and control the operation of the entire system. The AI ​​prediction and control unit exchanges data in real time with the central controller, pressure application device and ultrasonic energy device via fieldbus or industrial Ethernet to ensure that the response time of the control loop is less than 50 milliseconds.

[0015] A method for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure includes the following steps: Step S1: Align and assemble the surfaces of the first dissimilar material workpiece and the second dissimilar material workpiece to be welded, and place them in a sealed welding chamber. Step S2: Evacuate the welding chamber to create and maintain a preset negative pressure environment inside; Step S3: Under the negative pressure environment, apply a preset static pressure to the welding area of ​​the workpiece; Step S4: While applying the static pressure, apply high-frequency ultrasonic vibration energy to the welding area of ​​the workpiece; Step S5: During the welding process, dynamic information of the welding process is collected in real time through a multi-sensor fusion platform; Step S6: Input the dynamic information into the pre-trained deep learning model to predict in real time the probability of defects occurring at the welding interface within a preset time window in the future. The pre-trained deep learning model is a neural network model based on the LSTM-CNN dual attention mechanism. The CNN module in the neural network model based on the LSTM-CNN dual attention mechanism is used to extract spatial features from the high-speed visual image and the infrared temperature field image; The LSTM module in the neural network model based on the LSTM-CNN dual attention mechanism is used to learn the dynamic evolution of the multi-channel sensing data over time. The dual attention mechanism in the neural network model based on LSTM-CNN includes a spatial attention module and a temporal attention module, which are used to adaptively weight the key spatial regions and key temporal nodes that contribute the most to defect prediction. The specific work process is as follows: Multimodal data synchronization and preprocessing: Data reception and alignment: The system simultaneously receives three raw data streams from a high-speed camera, an infrared thermometer, and a spectrometer; Since different sensors have different acquisition frequencies and delays, a time synchronization module is first needed to stamp each frame of data with a unified high-precision timestamp to ensure that the visual, temperature and spectral information are strictly corresponding when analyzing the welding status at the "same moment". Data preprocessing: High-speed visual imaging: Perform operations such as image denoising, contrast enhancement, and region of interest cropping to highlight the molten pool outline, plastic flow area, and possible spatter; Infrared temperature field data: Converting the raw thermal radiation signal into accurate temperature values ​​and generating a temperature distribution matrix may require emissivity calibration to eliminate the influence of material surface conditions; Spectral data: The acquired raw spectra are denoised, baseline corrected and normalized to extract the intensity or ratio of specific characteristic spectral lines; High-dimensional feature extraction and fusion: Spatial feature extraction (handled by the CNN module): The preprocessed high-speed image and infrared thermal image are fed into a convolutional neural network; CNNs automatically and hierarchically learn spatial features in images through multiple convolutional and pooling layers: Shallow features: such as edges, textures, and simple shapes; Deeper characteristics: such as complex patterns and organizational structures; Time series feature extraction (handled by the LSTM module): Sensor data from multiple consecutive time steps (including high-level spatial feature vectors and spectral feature vectors extracted by CNN, as well as raw temperature statistics such as maximum temperature and temperature difference) are arranged in chronological order to form a multivariate time series. This sequence is fed into a Long Short-Term Memory (LSTM) network. LSTM, with its unique gating mechanism, is able to learn and remember dynamic evolution patterns over extended periods during the welding process, such as: Abnormal fluctuation trends in the temperature field; The continuous increase or decrease of a specific spectral intensity over the past 0.5 seconds; The gradual change in the molten pool morphology from stable to violently oscillating; Dual attention mechanism weighting: Spatial Attention Module: This module operates on the feature map extracted by the CNN. It automatically learns and generates an "attention weight map". Regions with high weights mean that the model believes that the spatial information at these locations is most critical for the defect prediction at the current moment. Temporal Attention Module: This module operates on the hidden state sequence of the LSTM. It automatically evaluates which state(s) in the past period are most important for predicting future defects. For example, it may find that a temperature drop 0.3 seconds ago is the strongest precursor to crack formation, and thus give higher weight to the features at that time point. After attention weighting, the model no longer treats all spatial and temporal information equally, but focuses on the "key spatiotemporal" information most relevant to the defect, which greatly improves the accuracy and robustness of the prediction. Defect probability prediction and output: Feature fusion and decision-making: The high-dimensional features, after being weighted by CNN (spatial), LSTM (temporal) and attention mechanisms, are concatenated into a comprehensive feature vector that integrates spatiotemporal multimodal information; The combined feature vector is fed into one or more fully connected layers for final classification or regression calculation; Probability output: The final output layer of the model typically uses the Sigmoid or Softmax activation function, outputting one or more values ​​between 0 and 1, for example: [Porosity probability: 0.92, Crack probability: 0.15, No defect detected probability: 0.08] This output clearly indicates that, based on all sensor information from the current and past periods, the model predicts that within a future preset time window, the probability of porosity at the weld interface is 92%, and the probability of crack formation is 15%. Step S7: When the predicted defect probability exceeds a preset threshold, generate an early warning signal and / or adjust at least one weld in real time.

[0016] In a second embodiment, the present invention provides a technical solution: a method for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure, comprising the following steps: Step S1: Align and assemble the surfaces of the first dissimilar material workpiece and the second dissimilar material workpiece to be welded, and place them in a sealed welding chamber. Step S2: Evacuate the welding chamber to create and maintain a preset negative pressure environment inside; Step S3: Under the negative pressure environment, apply a preset static pressure to the welding area of ​​the workpiece; Step S4: While applying the static pressure, apply high-frequency ultrasonic vibration energy to the welding area of ​​the workpiece; Step S5: During the welding process, dynamic information of the welding process is collected in real time through a multi-sensor fusion platform; Step S6: Input the dynamic information into the pre-trained deep learning model to predict in real time the probability of defects occurring at the welding interface within a preset time window. The pre-trained deep learning model is a spatiotemporal physical information fusion Transformer prediction model, and the specific working process is as follows: Multimodal physical feature embedding: Physical field reconstruction: Input: Raw multi-sensor data stream; process: The molten pool velocity field U(x,y,t) is calculated from high-speed visual images using an optical flow algorithm. The infrared thermal image is converted into a temperature field T(x,y,t), and its gradient is calculated to obtain the heat flux density field ▽T; Extract the temporal signal I_λ(t) of the spectral intensity of a specific element from the spectral data and map it as a proxy feature of the energy density distribution E(x,y,t) of the welding area; Output: A series of field data with clear physical meaning, such as [flow velocity field, temperature field, energy density field]. This step elevates the original pixel data into physical quantities, laying the foundation for subsequent fusion of physical knowledge. Spatiotemporal coding guided by physical information: Physical constraint coding: Input: The physical field data generated in step S6.1; process: Using multiple small physical information neural networks, each PINN learns a simplified physical conservation law, for example: Mass-Conserving PINN: The input is a flow velocity field U, and its loss function encourages the model to learn feature representations where ▽·U≈0; Energy Conservation PINN: The inputs are a temperature field T and an energy field E, and its loss function encourages model learning. Feature representation; These PINNs are not used directly for prediction, but rather as feature extractors. The hidden states they output are "features" that have been "corrected" by the laws of physics, which we call F_physics. Transformer-based multi-source heterogeneous graph attention fusion: Heterogeneous graph construction: Input: Raw data features F_raw and physics-guided features F_physics; process: The welding area is discretized in space as a graph structure G=(V,E). Each node v_i represents a spatial location, and the node features include F_raw_i and F_physics_i for that location; Edge E represents the connection between positions; This is a heterogeneous graph because the nodes contain both data-driven and physical-driven features. Graph attention fusion: Use the Heterogeneous Graph Transformer to process this graph; The model calculates an attention score between any two nodes in the graph. This score depends not only on the similarity of their data features but also on the compatibility of their physical features. For example, if one node experiences a rapid temperature increase (F_physics) and another node exhibits abnormal eddies (F_raw), even if they are not spatially adjacent, the attention mechanism may assign them high connection weights because this could indicate the initiation of thermal cracks. Through multiple graph attention layers, the model achieves deep fusion of data features and physical features in the spatiotemporal graph structure, generating fused features F_fused; Causal convolution and forward prediction: Time series prediction: Input: The fused features F_fused(t) output from step S6.3, arranged in time series; process: This time series is processed using a causal convolutional network. Causal convolution ensures that the output at prediction time t depends only on information up to and before t, meeting the requirements of real-time prediction; The network takes the current time t_0 as the endpoint and reviews the fusion features within a time window [t_0-τ,t_0]. The network's ultimate task is not classification, but multi-step regression prediction. It directly outputs the defect-forming dynamic parameters at each time point within a future time window of ∆t ≥ 0.5s, for example: P_porosity(t_0+0.5): The probability of stomatal formation at time t_0+0.5s; C_crack_length(t_0+0.7): The predicted crack length at time t_0+0.7s; T_risk(t_0+1.0): The overall risk index at time t_0+1.0s; Uncertainty Quantification and Decision Output: Probability and confidence output: The model uses probabilistic prediction outputs. For each predicted value, it outputs not only a mean but also an uncertainty interval. Final output: A structured forecast report; Step S7: When the predicted defect probability exceeds a preset threshold, generate an early warning signal and / or adjust at least one weld in real time.

[0017] Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art and related fields based on the embodiments of the present invention without inventive effort should fall within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described and explained in the present invention, unless otherwise specified or limited, shall be implemented according to conventional means in the art.

Claims

1. A method for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure, characterized in that, Includes the following steps: Step S1: Align and assemble the surfaces of the first dissimilar material workpiece and the second dissimilar material workpiece to be welded, and place them in a sealed welding chamber. Step S2: Evacuate the welding chamber to create and maintain a preset negative pressure environment inside; Step S3: Under the negative pressure environment, apply a preset static pressure to the welding area of ​​the workpiece; Step S4: While applying the static pressure, apply high-frequency ultrasonic vibration energy to the welding area of ​​the workpiece; Step S5: During the welding process, dynamic information of the welding process is collected in real time through a multi-sensor fusion platform; Step S6: Input the dynamic information into the pre-trained deep learning model to predict in real time the probability of defects occurring at the welding interface within a preset time window in the future. Step S7: When the predicted defect probability exceeds a preset threshold, generate an early warning signal and / or adjust at least one weld in real time.

2. The method for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure according to claim 1, characterized in that, In step S5, the multi-sensor fusion platform includes: A high-speed vision sensor with a frame rate of no less than 1000fps is used to capture the dynamics of the molten pool, plastic flow, and spatter behavior in the welding area. Infrared temperature sensor, with a temperature measurement range covering 300℃ to 1800℃, is used to monitor the temperature field distribution at the welding interface in real time. The spectral analysis sensor, with an analysis band covering 200nm to 900nm, is used to detect changes in plasma electron temperature, density, and spectral intensity of characteristic elements during the welding process.

3. The method for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure according to claim 1, characterized in that, In step S6, the pre-trained deep learning model is a neural network model based on the LSTM-CNN dual attention mechanism; The CNN module in the neural network model based on the LSTM-CNN dual attention mechanism is used to extract spatial features from the high-speed visual image and the infrared temperature field image; The LSTM module in the neural network model based on the LSTM-CNN dual attention mechanism is used to learn the dynamic evolution of the multi-channel sensing data over time. The dual attention mechanism in the neural network model based on LSTM-CNN includes a spatial attention module and a temporal attention module, which are used to adaptively weight the key spatial regions and key temporal nodes that contribute the most to defect prediction.

4. The method for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure according to claim 3, characterized in that, In step S6, the model can predict micron-level defects such as pores and cracks, and the prediction time is no less than 0.5 seconds earlier than the actual occurrence time of the defect.

5. The method for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure according to claim 1, characterized in that, The deep learning model mentioned in step S6 can also be replaced by a spatiotemporal physical information fusion Transformer prediction model. The specific working process is as follows: Multimodal physical feature embedding includes physical field reconstruction, physical constraint encoding, heterogeneous graph construction, and graph attention fusion; Causal convolution and look-ahead prediction, including time series prediction; Uncertainty quantification and decision output, including probability and confidence level output.

6. The method for controlling the interface of dissimilar materials welding based on the synergistic regulation of ultrasound and negative pressure according to claim 1, characterized in that, In step S7, the welding process parameters that are adjusted in real time include one or more of the following: the magnitude of the static pressure, the amplitude, power, or duration of the high-frequency ultrasonic vibration energy.

7. A control system for dissimilar material welding interfaces based on the coordinated regulation of ultrasound and negative pressure, characterized in that, include: A welding chamber, the interior of which is used to accommodate and fix the first dissimilar material workpiece and the second dissimilar material workpiece; A vacuum system, connected to the welding chamber, is used to evacuate the welding chamber to create and maintain the negative pressure environment. A pressure application device, the pressure head of which extends into the welding chamber, is used to apply the static pressure to the welding area of ​​the workpiece; An ultrasonic energy device, wherein the ultrasonic transducer is coupled to the pressure head, is used to generate and transmit the high-frequency ultrasonic vibration energy to the workpiece welding area. A multi-sensor fusion platform, integrated into the welding chamber, is used to collect dynamic information of the welding process in real time. It includes at least a high-speed vision sensor, an infrared temperature sensor, and a spectral analysis sensor. AI prediction and control unit, including: The data processing module is used to receive and synchronously fuse multimodal data from the multi-sensor fusion platform; The deep learning model module loads a pre-trained neural network model based on the LSTM-CNN dual attention mechanism, which is used to perform real-time prediction of defect probability based on the fused data. The decision control module is used to generate early warning signals based on the prediction results and / or send parameter adjustment instructions to the pressure application device and the ultrasonic energy device; The central controller is electrically connected to the vacuum system, pressure application device, ultrasonic energy device, and AI prediction and control unit, and is used to coordinate and control the operation of the entire system.

8. A control system for dissimilar material welding interfaces based on the coordinated regulation of ultrasound and negative pressure according to claim 7, characterized in that, The high-speed vision sensor, infrared temperature sensor, and spectral analysis sensor are aligned with the workpiece welding area through an observation window or a specially designed sealed interface on the welding chamber.

9. A control system for dissimilar material welding interfaces based on the coordinated regulation of ultrasound and negative pressure according to claim 8, characterized in that, The AI ​​prediction and control unit exchanges data in real time with the central controller, pressure application device and ultrasonic energy device via fieldbus or industrial Ethernet to ensure that the response time of the control loop is less than 50 milliseconds.