Lithium battery ultrasonic welding experimental machine and welding parameter automatic optimization method

Abstract

The application discloses a lithium battery ultrasonic welding experiment machine and a welding parameter automatic optimization method, and belongs to the technical field of lithium battery manufacturing equipment. The AI background processing module comprises a data storage unit, a data cleaning and screening unit, a big data analysis unit and an AI optimization algorithm unit. The AI optimization algorithm unit adopts a BP neural network and a multi-objective particle swarm optimization combined algorithm. Based on the welding parameter and welding quality data collected in real time by multiple stations, the parameter-quality correlation model is automatically cleaned, screened, established and iteratively optimized, and the optimal welding parameter combination is output. The parameter output execution module sends the optimized parameters to each station for execution and feeds back the actual parameters and quality data. The application realizes full-automatic and accurate recording of multi-station welding parameters of a large turntable and AI-driven optimization. The parameter optimization cycle is greatly shortened, the welding qualification rate is significantly improved, and the experiment efficiency and welding quality consistency are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of lithium battery manufacturing equipment technology, and in particular to a lithium battery ultrasonic welding test machine and an automatic optimization method for welding parameters. Background Technology

[0002] In lithium battery manufacturing, ultrasonic welding is a core process for connecting tabs, electrodes, and current collectors. The welding quality directly determines the safety, cycle life, and energy density of the lithium battery. Precise control of welding parameters, such as welding pressure, energy, amplitude, and welding time, is crucial. Currently, most ultrasonic welding testing machines for lithium batteries adopt a single-station or simple multi-station design, relying mainly on manual setting of welding parameters and manual recording of experimental data, which presents the following drawbacks: First, the recording of parameters at multiple workstations is incomplete and inefficient. Existing large rotary multi-workstation experimental machines cannot achieve real-time synchronous recording of welding parameters at all workstations. They can only manually record some parameters of key workstations, such as welding time and amplitude, omitting key response parameters such as welding pressure fluctuation, maximum power, and welding depth. Moreover, manual recording is prone to errors and omissions, and cannot form a complete experimental data chain, making it difficult to support parameter optimization analysis. Second, parameter optimization relies on human experience, resulting in low accuracy and poor repeatability. In existing technologies, the optimization of welding parameters mainly depends on the accumulated experience of operators. Adjusting parameters through repeated trial and error is not only time-consuming and labor-intensive, but the optimization results are also greatly affected by the skill level of the personnel, making it impossible to achieve precise parameter matching. At the same time, different operators have different optimization approaches, leading to poor repeatability of experimental results, making it difficult to form a standardized optimal parameter scheme, and easily converging to local optima instead of achieving global optima. Third, the utilization of big data is insufficient, resulting in low optimization efficiency. There is a complex nonlinear relationship between ultrasonic welding parameters and welding quality of lithium batteries, such as joint strength, penetration depth, and contact resistance. It is necessary to accumulate and analyze a large amount of experimental data to find the optimal parameter combination. However, existing equipment does not have a dedicated big data processing module, and it is impossible to clean, screen, and deeply analyze experimental data from multiple stations and batches. This makes it impossible to fully utilize the value of experimental data, resulting in long parameter optimization cycles, poor results, and difficulty in adapting to the welding requirements of different specifications of lithium batteries, such as copper foils and tabs of different thicknesses. Fourth, there is a lack of a closed-loop optimization mechanism. The parameter adjustment and data recording of existing equipment are independent of each other. It is impossible to automatically feed back and optimize subsequent welding parameters based on the real-time collected welding parameters and welding quality data to form a closed loop of "collection-analysis-optimization-application". This results in low experimental efficiency, difficulty in quickly iterating to find the optimal welding process parameters, and inability to cope with parameter deviations caused by problems such as dust and penetration depth fluctuations during the welding process.

[0003] In summary, in view of the above-mentioned shortcomings of the existing technology, there is an urgent need for a lithium battery ultrasonic welding experimental machine and an automatic optimization method for welding parameters. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a lithium battery ultrasonic welding experimental machine and an automatic optimization method for welding parameters. Compared with existing technologies, this invention provides more comprehensive and accurate data recording, more precise and efficient parameter optimization, and eliminates reliance on manual labor; it also achieves closed-loop optimization, resulting in more stable welding quality.

[0005] The technical solution adopted in this invention is as follows: An ultrasonic welding experimental machine for lithium batteries includes a large turntable multi-station welding mechanism, which is configured to carry multiple welding stations and sequentially complete workpiece clamping and ultrasonic welding. The parameter acquisition module includes multiple sensors installed at each welding station to collect welding parameters and welding quality data at each station in real time. It also includes an AI backend processing module electrically connected to the parameter acquisition module, the AI ​​backend processing module comprising: The data storage unit is used to classify and store welding parameter data, welding quality inspection data, and optimized parameter schemes for all workstations and batches. The data cleaning and filtering unit is used to remove abnormal data, remove duplicate data, fill in missing data, and filter valid data according to preset conditions. The big data analysis unit is used to perform statistical and correlation analysis on effective data and establish a correlation model between welding parameters and welding quality. The AI ​​optimization algorithm unit is used to automatically iteratively optimize welding parameters based on the correlation model and output the optimal combination of welding parameters. The parameter output execution module is electrically connected to the AI ​​background processing module and the large turntable multi-station welding mechanism. It is used to receive the optimal welding parameters and control each welding station to perform welding according to the parameters. The human-computer interaction module is electrically connected to the AI ​​backend processing module and is used to realize the functions of setting experimental parameters, viewing data, alarm prompts and manual intervention.

[0006] Furthermore, the abnormal data removed by the data cleaning and screening unit includes abrupt changes caused by sensor malfunctions and invalid data generated by sudden malfunctions during the welding process; the method for filling missing data adopts interpolation and is reasonably filled based on adjacent data of the same batch and the same workstation; at the same time, effective experimental data is screened out according to preset screening conditions, including welding qualification rate and parameter fluctuation range.

[0007] Furthermore, the big data analysis unit includes mining the nonlinear correlation between welding pressure, amplitude, energy, and welding time parameters and welding quality, and establishing a parameter-quality correlation model; the welding quality includes joint strength, contact resistance, penetration depth, and joint width.

[0008] The big data analysis unit is also configured to compare and analyze data from multiple workstations and batches to identify the optimal parameter ranges for different workstations and workpiece specifications, providing data support for AI optimization algorithms.

[0009] Furthermore, the parameter output execution module is used to receive the optimal welding parameters output by the AI ​​backend and transmit the parameter instructions to the ultrasonic welding execution unit of each welding station to control each station to perform welding according to the optimized parameters; at the same time, it provides real-time feedback on the deviation between the actual welding parameters of each station and the AI ​​optimized parameters. If the deviation exceeds the preset range, it automatically adjusts the parameters of the execution unit to ensure the accurate execution of the welding parameters.

[0010] Furthermore, the human-machine interaction module includes a touch screen and operation buttons for displaying real-time welding parameters, welding quality data, AI optimization parameter schemes, and historical experimental data. It also supports setting the number of workstations, welding batches, data filtering conditions, and fault alarm functions.

[0011] This invention also provides an automatic optimization method for welding parameters based on a lithium battery ultrasonic welding experimental machine, which includes the following steps: Step 1: Initialization settings. Set the experimental parameters through the human-computer interaction module, including the rotation speed of the large turntable, the number of welding stations, the workpiece specifications, the initial welding parameters, the quality judgment criteria, and the data filtering conditions. Step 2: Multi-station welding and parameter acquisition. Start the experimental machine. The parameter acquisition module collects the welding parameters and welding quality data of each station in real time and transmits them to the AI ​​backend processing module. Step 3: Data storage and cleaning / screening. The AI ​​backend processing module classifies and stores the raw data, removes anomalies, fills in missing data, and filters valid data. Step 4: Big Data Analysis and AI Optimization. The big data analysis unit establishes a correlation model between welding parameters and welding quality. The AI ​​optimization algorithm unit automatically iterates and optimizes based on this model, outputting the optimal combination of welding parameters. Step 5: Parameter execution and feedback. The parameter output execution module sends the optimal welding parameters to each welding station for execution and provides real-time feedback of actual parameters and quality data to the AI ​​backend processing module. Step 6: Iterative optimization. The AI ​​backend processing module compares the actual welding quality with the preset standard. If the requirements are not met, steps 4 to 5 are repeated until the optimal parameter solution that meets the requirements is output. Step 7: The experiment ends. Output the experiment report and store the optimal parameter scheme.

[0012] Furthermore, in step 3, the data storage unit of the AI ​​backend processing module categorizes and stores the received raw data, and the data cleaning and filtering unit automatically removes abnormal and duplicate data, fills in missing data, and filters out valid data that meets the conditions. Specifically, the data storage unit of the AI ​​backend processing module categorizes and stores the received raw data according to workstation, batch, and workpiece specification. The data cleaning and filtering unit uses a dual judgment mechanism of 3σ statistical criteria combined with equipment process hard threshold to automatically identify abnormal data. By calculating the mean and standard deviation of the dataset under the same working condition, it removes statistically abnormal samples that exceed the normal fluctuation range, and filters out hardware abnormal data such as exceeding the equipment range, parameter mutations, and signal failures. For missing data due to instantaneous disconnection or single-point gaps during the acquisition process, it uses adjacent time-series linear interpolation and adaptive algorithm of mean of samples under the same working condition to intelligently fill in the missing data to ensure data continuity and authenticity.

[0013] Furthermore, in step 4, the AI ​​optimization algorithm unit uses a BP neural network and MOPSO joint algorithm based on the association model to automatically iterate and optimize, and outputs the optimal welding parameter combination that is adapted to the current workpiece specifications and each station; wherein, the BP neural network adopts a fully connected feedforward network structure of input layer-double hidden layer-output layer.

[0014] Furthermore, the AI ​​backend processing module compares the actual welding quality with the preset standard. If the welding qualification rate does not meet the preset requirements, it performs big data analysis and AI optimization again based on the feedback data, adjusts the welding parameters, and outputs the optimal parameter scheme. If the welding qualification rate meets the preset requirements, the optimal parameter scheme is stored in the database as a reference parameter for welding workpieces of the same specifications in the future.

[0015] The advantages of this invention over the prior art are as follows: (i) More comprehensive, accurate, and efficient data recording: This invention enables real-time, automatic, and comprehensive recording of welding parameters at multiple stations on a large turntable, covering all key parameters such as welding pressure, amplitude, energy, time, maximum power, and welding depth. It eliminates errors and omissions in manual recording, significantly improves data recording efficiency, and forms a complete experimental data archive, providing reliable support for subsequent experimental analysis and parameter optimization, while avoiding optimization deviations caused by missing data. (ii) More precise and efficient parameter optimization, eliminating reliance on manual labor: This invention uses big data refinement and joint optimization algorithms in the AI ​​backend to automatically discover the correlation between parameters and welding quality, output the optimal parameter combination, shorten the parameter optimization cycle by more than 50%, and increase the welding qualification rate from about 85% of the existing technology to more than 99.5%. Moreover, the optimization results have good repeatability, forming a standardized parameter scheme, eliminating the need to rely on the experience of operators, reducing labor costs, and avoiding the problem of local optima, achieving global optima. (iii) Achieving closed-loop optimization and stronger adaptability: This invention constructs a closed-loop mechanism of "acquisition-processing-optimization-execution-feedback", which can dynamically adjust parameters according to real-time welding data to adapt to the welding requirements of different specifications of lithium batteries (such as copper foil of different thicknesses, tabs, and dissimilar metal welding). At the same time, it optimizes the welding head fixing method, improves amplitude uniformity, reduces parameter deviation between workstations, expands the applicability of experimental equipment, and can meet the welding experimental requirements of various workpieces such as 0.1mm copper protective sheet, 50 layers of 6μm copper foil, and 0.8mm copper connecting piece; (iv) Improve experimental efficiency and reduce experimental costs: The multi-station simultaneous welding of this invention combined with AI automatic optimization greatly reduces the number of experimental trials and errors, reduces workpiece wear and experimental time costs; at the same time, it reduces the workload of operators, who only need to complete the initialization settings and workpiece clamping, without having to manually record data and adjust parameters, further improving experimental efficiency and saving more than 30% of experimental costs. (v) More stable welding quality and higher safety: This invention reduces the generation of welding defects (such as incomplete welding, over-welding, and spatter) through precise parameter control and quality feedback. The contact resistance of the weld joint is ≤3mΩ and the penetration depth fluctuation rate is <3%, which far exceeds the industry standard. This improves the safety and reliability of lithium battery welding, reduces the risk of battery recall due to welding defects, and reduces the adverse effects of welding dust from the source. Attached Figure Description

[0016] Figure 1 This is a flowchart of the automatic optimization method for welding parameters in this invention. Detailed Implementation

[0017] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0018] Example 1 This embodiment provides a lithium battery ultrasonic welding experimental machine, mainly including a large turntable multi-station welding mechanism, a parameter acquisition module, an AI backend processing module, a parameter output execution module, and a human-machine interaction module. These modules are electrically connected to each other, forming a complete closed-loop control system. The AI ​​backend processing module is the core module, responsible for data recording, filtering, big data refinement, and parameter optimization, achieving automated and intelligent optimization of welding parameters.

[0019] In one embodiment of the present invention, a large turntable multi-station welding mechanism is configured to carry multiple welding stations and sequentially complete workpiece clamping and ultrasonic welding; the large turntable rotates at a constant speed under the drive of a drive motor, and each welding station sequentially completes workpiece clamping and welding operations.

[0020] In one embodiment of the invention, multiple sensors installed at each welding station are also included to collect welding parameters and welding quality data at each station in real time. Specifically, each sensor in the parameter acquisition module collects welding pressure, amplitude, energy, welding time, maximum power, welding depth, and welding quality data at each station in real time. After noise reduction and analog-to-digital conversion by the data acquisition unit, the data is transmitted in real time to the AI ​​backend processing module.

[0021] In one embodiment of the present invention, the AI ​​backend processing module is mainly implemented by backend software on the host computer, including a data storage unit, a data cleaning and filtering unit, a big data analysis unit, and an AI optimization algorithm unit. These units work together to achieve full-process data processing and parameter optimization. Specific functions are as follows: The data storage unit employs a large-capacity database to store welding parameter data, welding quality inspection data, and AI-optimized parameter schemes for all workstations and batches transmitted by the parameter acquisition module. The database supports data classification storage (by workstation, batch, and workpiece specification), enabling rapid data query and retrieval. It also supports data backup and recovery to ensure the security of experimental data, and its storage capacity can meet the storage needs of at least 10,000 batches of experimental data.

[0022] The data cleaning and filtering unit automatically processes the stored raw data, removing abnormal data (such as abrupt changes caused by sensor malfunctions or invalid data generated by sudden failures during welding) and duplicate data, and filling in missing data (using interpolation to fill in missing data based on adjacent data from the same batch and workstation). Simultaneously, based on preset filtering conditions (such as welding pass rate ≥90% and parameter fluctuation range ≤5%), it filters out valid experimental data, providing a high-quality data foundation for subsequent big data analysis and AI optimization, ensuring the accuracy of the analysis results.

[0023] The big data analysis unit performs in-depth mining and analysis of the cleaned effective data, using methods such as data statistics and correlation analysis to uncover the nonlinear correlation between parameters such as welding pressure, amplitude, energy, and welding time and welding quality (joint strength, contact resistance, penetration depth, and joint width), and establishes a parameter-quality correlation model. Simultaneously, it compares and analyzes data from multiple workstations and batches to identify the optimal parameter ranges for different workstations and workpiece specifications, providing data support for AI optimization algorithms. This unit can process more than 360 sets of random sample data, establishing a mapping relationship between parameters and welding performance with an error controlled within 10%.

[0024] In particular, the AI ​​optimization algorithm unit employs a combined BP neural network and multi-objective particle swarm optimization (MOPSO) algorithm. It optimizes welding pass rate, joint strength, and welding efficiency as objectives, and welding parameters (pressure, amplitude, energy, and time) as variables. Based on a correlation model established by the big data analysis unit, it automatically iterates and optimizes to output the optimal combination of welding parameters. The BP neural network predicts the welding quality corresponding to different parameter combinations, while the MOPSO algorithm performs global optimization to avoid convergence to local optima. The optimized parameters must meet the following requirements: welding pressure 0.18–0.23 MPa, energy 300–475 J, and amplitude 65%–78% (dynamically adjustable according to workpiece specifications), ensuring optimal welding quality, highest efficiency, and lowest cost. Simultaneously, the algorithm supports self-learning, continuously optimizing the parameter model and improving optimization accuracy as experimental data accumulates. The iteration count can be set to 50 times to further enhance optimization efficiency.

[0025] In one embodiment of the present invention, the parameter output execution module is electrically connected to the AI ​​backend processing module and the large turntable multi-station welding mechanism. It receives the optimal welding parameters output by the AI ​​backend and transmits the parameter instructions to the ultrasonic welding execution units at each welding station, controlling each station to weld according to the optimized parameters. Simultaneously, it provides real-time feedback on the deviation between the actual welding parameters at each station and the AI-optimized parameters. If the deviation exceeds a preset range (≤3%), the parameters of the execution units are automatically adjusted to ensure accurate execution of the welding parameters. Furthermore, this module supports manual parameter adjustment, allowing operators to manually modify the optimized parameters according to experimental needs, flexibly adapting to different experimental scenarios.

[0026] In one embodiment of the present invention, the human-machine interaction module includes a touch screen and operation buttons for realizing human-machine interaction. Operators can view real-time welding parameters, welding quality data, AI-optimized parameter schemes, and historical experimental data for each workstation through the touch screen. At the same time, they can set experimental parameters (such as the number of workstations, welding batches, and screening conditions), start / stop experiments, and manually trigger AI optimization functions through the operation buttons. In addition, the human-machine interaction module supports a fault alarm function. When there is a sensor failure, abnormal data transmission, or excessive parameter deviation, an alarm signal is automatically issued to remind the operator to handle the situation in a timely manner.

[0027] Example 2: like Figure 1 As shown, this embodiment provides an automatic optimization method for welding parameters based on the lithium battery ultrasonic welding experimental machine described in Embodiment 1, including the following steps: Step 1, Initialization settings: The operator sets the experimental parameters through the human-computer interaction module, including the rotation speed of the large turntable, the number of welding stations, the specifications of the lithium battery workpiece to be welded, the initial welding parameters (pressure, amplitude, energy, time), the welding quality judgment criteria (such as contact resistance ≤5mΩ, penetration depth 0.02-0.05mm), and the data filtering conditions. Step 2, Multi-station welding and parameter acquisition: Start the experimental machine, and the large turntable rotates at a constant speed under the drive motor. Each welding station completes the workpiece clamping and welding operation in sequence. The sensors of the parameter acquisition module collect the welding pressure, amplitude, energy, welding time, maximum power, welding depth and welding quality data of each station in real time. After noise reduction and analog-to-digital conversion by the data acquisition unit, the data is transmitted to the AI ​​backend processing module in real time. Step 3, Data Storage and Cleaning: The data storage unit of the AI ​​backend processing module categorizes and stores the received raw data. The data cleaning and filtering unit automatically removes abnormal and duplicate data, fills in missing data, and filters out valid data that meets the criteria. Specifically, the data storage unit of the AI ​​backend processing module categorizes and stores the received raw data by workstation, batch, and workpiece specification. The data cleaning and filtering unit uses a dual judgment mechanism of 3σ statistical criteria and equipment process hard threshold to automatically identify abnormal data. By calculating the mean and standard deviation of the dataset under the same working conditions, it removes statistically abnormal samples that exceed the normal fluctuation range, and filters out hardware abnormal data such as exceeding the equipment range, parameter mutations, and signal failures. For missing data due to instantaneous disconnection or single-point gaps during the acquisition process, it uses adjacent time-series linear interpolation and adaptive algorithms based on the mean of samples under the same working conditions to intelligently fill in the missing data, ensuring data continuity and authenticity. Step 4, Big Data Analysis and AI Optimization: The big data analysis unit deeply mines effective data to establish a correlation model between welding parameters and welding quality. Based on this model, the AI ​​optimization algorithm unit uses a BP neural network and the MOPSO joint algorithm to automatically iterate and optimize, outputting the optimal combination of welding parameters adapted to the current workpiece specifications and each workstation. Among them, the BP neural network adopts a fully connected feedforward network structure of "input layer - double hidden layer - output layer". The specific data architecture and training mechanism are as follows: The network input layer is set with 4 input nodes, corresponding to the four core controllable welding process parameters: welding pressure, ultrasonic amplitude, welding energy, and welding holding time. The network is set with two hidden layers. The first hidden layer has 18 neurons and the second hidden layer has 12 neurons. The complex nonlinear relationship between parameters and quality is fitted through multi-dimensional feature mapping. The output layer is set with 4 output nodes, corresponding to key welding quality indicators: joint contact resistance, weld penetration, joint tensile strength, and welding pass rate, realizing multi-dimensional quality synchronous prediction. This model uses the LM (Leuwenberg-Marquardt) training algorithm for iterative training. During training, the original parameter data is automatically normalized and preprocessed, and the training error threshold is set to 10. -4 The maximum number of iterations is 50. The network weights and biases are continuously corrected through backpropagation to avoid gradient vanishing and overfitting problems and improve the model prediction accuracy. The high-precision neural network after training convergence is used as a quality prediction proxy model. Then, the MOPSO algorithm is used to globally optimize with multiple objectives of "low contact resistance, high penetration consistency, high welding pass rate, and high welding efficiency". Iteratively solves and outputs the optimal welding parameter combination that adapts to the current workpiece specifications and the consistency deviation of each station.

[0028] Step 5, Parameter Execution and Feedback: The parameter output and execution module transmits the AI-optimized parameter instructions to each welding station, controlling each station to weld according to the optimized parameters; at the same time, it collects the actual welding parameters and welding quality data of each station in real time and feeds them back to the AI ​​backend processing module. Step 6, Iterative Optimization: The AI ​​backend processing module compares the actual welding quality with the preset standard. If the welding qualification rate does not meet the preset requirements (e.g., ≥99.5%), it will perform big data analysis and AI optimization again based on the feedback data, adjust the welding parameters, and output the optimal parameter scheme. If the welding qualification rate meets the preset requirements, the optimal parameter scheme will be stored in the database as a reference parameter for welding workpieces of the same specifications in the future. Step 7, Experiment End: After completing the preset batch of welding experiments, the operator can view the experiment report through the human-computer interaction module, including the welding parameters of each station, welding quality data, AI optimization process and optimal parameter scheme, and can export the experimental data for subsequent analysis.

[0029] The above description is an explanation of the present invention and not a limitation thereof. The scope of the present invention is defined by the claims. Within the scope of protection of the present invention, any form of modification may be made.

Claims

1. A lithium battery ultrasonic welding experimental machine, characterized in that: This includes a large rotary table multi-station welding mechanism, which is configured to support multiple welding stations and sequentially complete workpiece clamping and ultrasonic welding. The parameter acquisition module includes multiple sensors installed at each welding station to collect welding parameters and welding quality data at each station in real time. It also includes an AI backend processing module electrically connected to the parameter acquisition module, the AI ​​backend processing module comprising: The data storage unit is used to classify and store welding parameter data, welding quality inspection data, and optimized parameter schemes for all workstations and batches. The data cleaning and filtering unit is used to remove abnormal data, remove duplicate data, fill in missing data, and filter valid data according to preset conditions. The big data analysis unit is used to perform statistical and correlation analysis on effective data and establish a correlation model between welding parameters and welding quality. The AI ​​optimization algorithm unit is used to automatically iteratively optimize welding parameters based on the correlation model and output the optimal combination of welding parameters. The parameter output execution module is electrically connected to the AI ​​background processing module and the large turntable multi-station welding mechanism. It is used to receive the optimal welding parameters and control each welding station to perform welding according to the parameters. The human-computer interaction module is electrically connected to the AI ​​backend processing module and is used to realize the functions of setting experimental parameters, viewing data, alarm prompts and manual intervention.

2. The lithium battery ultrasonic welding experimental machine as described in claim 1, characterized in that: The abnormal data removed by the data cleaning and screening unit includes abrupt changes caused by sensor malfunctions and invalid data generated by sudden failures during the welding process. The method for filling missing data is interpolation, and reasonable filling is performed based on adjacent data from the same batch and the same workstation. At the same time, valid experimental data is screened out according to preset screening conditions, including welding pass rate and parameter fluctuation range.

3. The lithium battery ultrasonic welding experimental machine as described in claim 1, characterized in that: The big data analysis unit includes mining the nonlinear correlation between welding pressure, amplitude, energy, and welding time parameters and welding quality, and establishing a parameter-quality correlation model; the welding quality includes joint strength, contact resistance, penetration depth, and joint width. The big data analysis unit is also configured to compare and analyze data from multiple workstations and batches to identify the optimal parameter ranges for different workstations and workpiece specifications, providing data support for AI optimization algorithms.

4. The lithium battery ultrasonic welding experimental machine as described in claim 1, characterized in that: The parameter output execution module is used to receive the optimal welding parameters output by the AI ​​backend and transmit the parameter instructions to the ultrasonic welding execution unit of each welding station to control each station to perform welding according to the optimized parameters; at the same time, it provides real-time feedback on the deviation between the actual welding parameters of each station and the AI ​​optimized parameters. If the deviation exceeds the preset range, it automatically adjusts the parameters of the execution unit to ensure the accurate execution of the welding parameters.

5. The lithium battery ultrasonic welding experimental machine as described in claim 1, characterized in that: The human-machine interaction module includes a touch screen and operation buttons, which are used to display real-time welding parameters, welding quality data, AI optimization parameter schemes and historical experimental data, and support the setting of the number of workstations, welding batches, data filtering conditions, and fault alarm functions.

6. A method for automatically optimizing welding parameters based on the ultrasonic welding experimental machine for lithium batteries according to any one of claims 1-5, characterized in that, Includes the following steps: Step 1: Initialization settings. Set the experimental parameters through the human-computer interaction module, including the rotation speed of the large turntable, the number of welding stations, the workpiece specifications, the initial welding parameters, the quality judgment criteria, and the data filtering conditions. Step 2: Multi-station welding and parameter acquisition. Start the experimental machine. The parameter acquisition module collects the welding parameters and welding quality data of each station in real time and transmits them to the AI ​​backend processing module. Step 3: Data storage and cleaning / screening. The AI ​​backend processing module classifies and stores the raw data, removes anomalies, fills in missing data, and filters valid data. Step 4: Big Data Analysis and AI Optimization. The big data analysis unit establishes a correlation model between welding parameters and welding quality. The AI ​​optimization algorithm unit automatically iterates and optimizes based on this model, outputting the optimal combination of welding parameters. Step 5: Parameter execution and feedback. The parameter output execution module sends the optimal welding parameters to each welding station for execution and provides real-time feedback of actual parameters and quality data to the AI ​​backend processing module. Step 6: Iterative optimization. The AI ​​backend processing module compares the actual welding quality with the preset standard. If the requirements are not met, steps 4 to 5 are repeated until the optimal parameter solution that meets the requirements is output. Step 7: The experiment ends. Output the experiment report and store the optimal parameter scheme.

7. The automatic optimization method for welding parameters as described in claim 6, characterized in that: In step 3, the data storage unit of the AI ​​backend processing module categorizes and stores the received raw data, while the data cleaning and filtering unit automatically removes abnormal and duplicate data, fills in missing data, and filters out valid data that meets the criteria. Specifically, the data storage unit of the AI ​​backend processing module categorizes and stores the received raw data by workstation, batch, and workpiece specification. The data cleaning and filtering unit uses a dual judgment mechanism combining the 3σ statistical criterion and the equipment process hard threshold to automatically identify abnormal data. By calculating the mean and standard deviation of the dataset under the same working conditions, it removes statistically abnormal samples that exceed the normal fluctuation range, and filters out hardware abnormal data such as those exceeding the equipment range, parameter mutations, and signal failures. For missing data due to instantaneous disconnection or single-point gaps during the acquisition process, it uses adjacent time-series linear interpolation and adaptive algorithms based on the mean of samples under the same working conditions to intelligently fill in the missing data to ensure data continuity and authenticity.

8. The automatic optimization method for welding parameters as described in claim 6, characterized in that: In step 4, the AI ​​optimization algorithm unit uses a BP neural network and MOPSO joint algorithm based on the correlation model to automatically iterate and optimize, and outputs the optimal welding parameter combination that is suitable for the current workpiece specifications and each station; wherein, the BP neural network adopts a fully connected feedforward network structure of input layer-double hidden layer-output layer.

9. The automatic optimization method for welding parameters as described in claim 6, characterized in that: The AI ​​backend processing module compares the actual welding quality with the preset standard. If the welding pass rate does not meet the preset requirements, it performs big data analysis and AI optimization again based on the feedback data, adjusts the welding parameters, and outputs the optimal parameter scheme. If the welding pass rate meets the preset requirements, the optimal parameter scheme is stored in the database as a reference parameter for welding workpieces of the same specifications in the future.

Images