Ultrasonic process parameter regulation and control system and method, storage medium and program product

Abstract

The invention relates to the technical field of ultrasonic regulation and control, in particular to an ultrasonic technological parameter regulation and control system and method, a storage medium and a program product, and the system comprises detection equipment which is used for detecting stress data and deformation data of a welding part; the processing circuit is provided with a processing module and a reasoning module, and the processing module processes the stress data and the deformation data and generates a control signal based on the adjusted ultrasonic process parameters; the reasoning module determines the adjusted ultrasonic process parameters by using a back propagation neural network model and the processed stress data and deformation data; and the controller is used for adjusting ultrasonic process parameters of the ultrasonic welding system based on the control signal. Therefore, the problems of insufficient regulation and control precision and dynamic adaptability of ultrasonic process parameters in related technologies and the like are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ultrasonic regulation, and in particular to an ultrasonic process parameter regulation system and method, a storage medium, and a program product. BACKGROUND

[0002] High-energy ultrasonic regulation of residual stress technology has important applications in the fields of aerospace precision components and new energy vehicle battery welding. It can achieve precise regulation of subsurface stress through non-contact energy transmission mechanism, inhibit fatigue cracks, and maintain material performance stability. However, its practical application is still limited by the experience-driven parameter configuration system.

[0003] Related technologies for ultrasonic process parameter configuration rely on equipment supplier recommendations or manual experience, lack systematic and collaborative optimization methods, resulting in insufficient regulation accuracy and low energy efficiency. Static parameter settings are difficult to cope with material property fluctuations or environmental changes, easily leading to local temperature overheating, uneven stress distribution, and effect fluctuations, severely restricting the stability, repeatability, and adaptability of the technology in large-scale industrial scenarios. SUMMARY

[0004] The present application provides an ultrasonic process parameter regulation system and method, a storage medium, and a program product to solve the problems of insufficient ultrasonic process parameter regulation accuracy and dynamic adaptability in related technologies.

[0005] The first aspect of the present application provides an ultrasonic process parameter regulation system, comprising: a detection device for detecting stress data and deformation data of a welded part; at least one processing circuit, wherein the processing circuit is provided with a processing module and an inference module, the processing module processes the stress data and deformation data, and generates a control signal based on the adjusted ultrasonic process parameters, the inference module determines the adjusted ultrasonic process parameters using a backpropagation neural network model and the processed stress data and deformation data; a controller for adjusting the ultrasonic process parameters of the ultrasonic welding system based on the control signal.

[0006] Optionally, the inference module is deployed in the programmable logic of the processing circuit; the processing module is deployed in the programmable logic or on-chip processing system of the processing circuit.

[0007] Optionally, the back propagation neural network model is pre-trained, and the training process of the back propagation neural network model includes: obtaining a training data set and an unlabeled data set, wherein the training data set includes multiple sets of ultrasonic process parameters and corresponding stress data and deformation data, and the unlabeled data set includes multiple sets of stress data and deformation data; using the training data set to train the back propagation neural network model, and using the trained back propagation neural network model to predict the probability distribution of each set of data in the unlabeled data set; determining the corresponding confidence score based on the probability distribution of each set of data, and sorting the uncertainty of each set of data in the unlabeled data set based on the confidence score; determining multiple sets of target data based on the uncertainty sorting results, labeling the multiple sets of target data with corresponding ultrasonic process parameters, and adding the labeled multiple sets of target data to the training data set, and retraining the back propagation neural network model based on the added training data set.

[0008] Optionally, the controller includes: a power controller and an ultrasonic amplifier, wherein the power controller is used to drive the ultrasonic amplifier based on a control signal; and the ultrasonic amplifier is used to adjust ultrasonic process parameters of the ultrasonic welding system based on the control signal.

[0009] Optionally, the back propagation neural network model includes a first propagation layer, a second propagation layer and a loss function, wherein the first propagation layer generates a first prediction result based on the input data of the back propagation neural network model, wherein the input data includes stress data and deformation data in the training data set; the second propagation layer generates a second prediction result based on the first prediction result; the loss function calculates the error based on the second prediction result and the ultrasonic process parameters of the training data set, passes the error back to the input of the back propagation neural network model, and updates the model parameters of the neural network model based on the error until the loss function converges.

[0010] Optionally, the first propagation layer and the second propagation layer have the same propagation layer structure, wherein the propagation layer structure includes: multiple hidden layers, nonlinear activation functions and fully connected layers, wherein the multiple hidden layers are used to extract and transform features of the input data of the back propagation neural network model; the nonlinear activation function is used to introduce nonlinear characteristics to the output data of the multiple hidden layers; and the fully connected layer is used to output the prediction results.

[0011] Optionally, updating the model parameters of the back-propagation neural network model according to the error includes: calculating the model gradient of the back-propagation neural network model based on the error; and updating the model parameters of the back-propagation neural network model based on the model gradient and a preset learning rate.

[0012] The second aspect of the present application provides an ultrasonic process parameter control method, which is implemented based on the ultrasonic process parameter control system of the above embodiment, and includes the following steps: acquiring stress data and deformation data of the welded part; processing the stress data and deformation data, and using the back propagation neural network model, the processed stress data and deformation data to determine the adjusted ultrasonic process parameters; generating a control signal based on the adjusted ultrasonic process parameters, and sending the control signal to a controller, and the controller adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal.

[0013] The third aspect of the present application provides a computer-readable storage medium on which a computer program or instruction is stored. The computer program or instruction is executed by a processor to perform the ultrasonic process parameter control method as described in the above embodiment.

[0014] The fourth aspect of the present application provides a computer program product, including a computer program or instructions. When the computer program or instructions are executed, the ultrasonic process parameter control method as described in the above embodiment is implemented.

[0015] Therefore, this application has at least the following beneficial effects:

[0016] The embodiment of the present application constructs an ultrasonic process parameter control system, in which the detection equipment transmits the stress data and deformation data of the detected welded parts to the processing circuit. The processing circuit is provided with a processing module and an inference module. The processing module is used to process the stress data and deformation data and generate a control signal based on the adjusted ultrasonic process parameters. The inference module uses a back-propagation neural network model and the processed stress data and deformation data to determine the adjusted ultrasonic process parameters. The adjustment and determination of the ultrasonic process parameters are achieved through the back-propagation neural network model, thereby improving the control accuracy. The control signal is transmitted to the controller, and the controller adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal, thereby achieving closed-loop control of the ultrasonic process parameter adjustment. The ultrasonic process parameters can be adjusted in time according to the stress data and deformation data after the adjustment of the ultrasonic process parameters, thereby improving dynamic adaptability. There is no need to redesign the circuit or replace components, resulting in a short development cycle and low maintenance cost. Thus, the system solves the technical problems that the related technologies cannot achieve dynamic adjustment of ultrasonic process parameters.

[0017] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0019] Figure 1Schematic diagram of an ultrasonic process parameter control system provided according to an embodiment of the present application;

[0020] Figure 2 A schematic diagram of an active learning training back-propagation neural network model provided according to an embodiment of the present application;

[0021] Figure 3 A schematic diagram of an ultrasonic process parameter control system provided according to a specific embodiment of the present application;

[0022] Figure 4 The present invention provides a flowchart of an ultrasonic process parameter control method according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0024] Before describing the solution of this application, the related technologies of this application are first introduced to assist in understanding the solution of this application.

[0025] The ultrasonic process parameter control system, method, storage medium and program product of the embodiment of the present application are described below with reference to the accompanying drawings. In view of the fact that the related technologies mentioned in the above background technology mostly rely on the equipment supplier's suggestions or manual experience for the configuration of ultrasonic process parameters, and lack a systematic collaborative optimization method, resulting in insufficient control accuracy and low energy efficiency, and static parameter settings are difficult to cope with fluctuations in material properties or environmental changes, which can easily lead to local excessive temperature, uneven stress distribution and effect fluctuations, seriously restricting the stability, repeatability and adaptability of the technology in large-scale industrial scenarios, the present application provides an ultrasonic process parameter control system, in which the detection equipment transmits the stress data and deformation data of the detected weld to the processing circuit, and the processing circuit is provided with a processing module and an inference module, and the processing module is used to process stress data The ultrasonic process parameters are adjusted based on the stress and deformation data, and a control signal is generated based on the adjusted ultrasonic process parameters. The reasoning module uses the back-propagation neural network model and the processed stress and deformation data to determine the adjusted ultrasonic process parameters. The adjustment and determination of the ultrasonic process parameters are achieved through the back-propagation neural network model, which improves the control accuracy. The control signal is transmitted to the controller, which adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal, realizing closed-loop control of the ultrasonic process parameter adjustment. The ultrasonic process parameters can be adjusted in time according to the stress and deformation data after the adjustment of the ultrasonic process parameters, thereby improving the dynamic adaptability. There is no need to redesign the circuit or replace components, resulting in a short development cycle and low maintenance cost. This solves the problem that the related technology cannot achieve dynamic adjustment of ultrasonic process parameters.

[0026] Specifically, Figure 1 A schematic diagram of an ultrasonic process parameter control system provided in an embodiment of the present application.

[0027] like Figure 1 As shown, the ultrasonic process parameter control system 10 includes: a detection device 11, at least one processing circuit 12 and a controller 13.

[0028] Among them, the detection equipment 11 is used to detect the stress data and deformation data of the welded parts; at least one processing circuit 12 is provided with a processing module and an inference module, the processing module processes the stress data and deformation data, and generates a control signal based on the adjusted ultrasonic process parameters, and the inference module uses the back propagation neural network model and the processed stress data and deformation data to determine the adjusted ultrasonic process parameters; the controller 13 is used to adjust the ultrasonic process parameters of the ultrasonic welding system based on the control signal.

[0029] The processing circuit 12 may be an FPGA, which may be deployed on a Xilinx Artix-7 or Zynq-7000 series platform.

[0030] It can be understood that the embodiment of the present application constructs an ultrasonic process parameter control system 10, and the detection equipment 11 transmits the stress data and deformation data of the detected weldment to the processing circuit 12. The processing circuit 12 is provided with a processing module and an inference module. The processing module is used to process the stress data and deformation data, and generate a control signal based on the adjusted ultrasonic process parameters. The inference module uses the back propagation neural network model and the processed stress data and deformation data to determine the adjusted ultrasonic process parameters. The adjustment and determination of the ultrasonic process parameters are realized through the back propagation neural network model, thereby improving the control accuracy. The control signal is transmitted to the controller 13, and the controller 13 adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal, thereby realizing closed-loop control of the ultrasonic process parameter adjustment. The ultrasonic process parameters can be adjusted in time according to the stress data and deformation data after the adjustment of the ultrasonic process parameters, thereby improving dynamic adaptability, and there is no need to redesign the circuit or replace components, with a short development cycle and low maintenance cost.

[0031] In the embodiment of the present application, the inference module is deployed in the programmable logic of the processing circuit 12; the processing module is deployed in the programmable logic or on-chip processing system of the processing circuit 12.

[0032] It can be understood that the inference module of the embodiment of the present application can be set in the PL (programmable logic) of the processing circuit to obtain maximum parallelism and minimum delay. The processing module can be set on the PL or on-chip CPU (PS) side.

[0033] In an embodiment of the present application, the controller 13 includes: a power controller and an ultrasonic amplifier, wherein the power controller is used to drive the ultrasonic amplifier based on a control signal; the ultrasonic amplifier is used to adjust the ultrasonic process parameters of the ultrasonic welding system based on the control signal.

[0034] It can be understood that the controller 13 of the embodiment of the present application includes a power controller and an ultrasonic amplifier, wherein the power controller is used to receive a control signal, act on the ultrasonic amplifier based on the control signal, and the ultrasonic amplifier adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal.

[0035] In an embodiment of the present application, the back propagation neural network model is pre-trained, and the training process of the back propagation neural network model includes: obtaining a training data set and an unlabeled data set, wherein the training data set includes multiple sets of ultrasonic process parameters and corresponding stress data and deformation data, and the unlabeled data set includes multiple sets of stress data and deformation data; using the training data set to train the back propagation neural network model, and using the trained back propagation neural network model to predict the probability distribution of each set of data in the unlabeled data set; determining the corresponding confidence score based on the probability distribution of each set of data, and sorting the uncertainty of each set of data in the unlabeled data set based on the confidence score; determining multiple sets of target data based on the uncertainty sorting results, labeling the multiple sets of target data with corresponding ultrasonic process parameters, and adding the labeled multiple sets of target data to the training data set, and retraining the back propagation neural network model based on the added training data set.

[0036] Among them, the formula with the lowest screening confidence is: Given an input x, choose the predicted probability The smallest sample x i,c Here It may be based on a certain model (such as a neural network) to predict the probability that sample x belongs to a certain category. By selecting the sample with the lowest confidence, it can be used as a sample selection strategy in active learning to obtain the samples that are most helpful for model training.

[0037] It can be understood that the embodiment of the present application trains the back propagation neural network model through active learning, which can dynamically screen high-uncertainty samples, eliminate ultrasonic process parameters with poor control effects, achieve multi-parameter collaborative optimization, and improve the model's adaptability to the control process, thereby significantly reducing costs.

[0038] Active learning algorithms can select samples from a pool of unlabeled data where the model is uncertain and request labeling, significantly reducing cost and time. Existing machine learning methods require significantly more data and time to achieve comparable performance. By identifying the most uncertain data points, active learning can more quickly improve model accuracy and generalization. Active learning converges faster because it prioritizes samples that minimize model uncertainty. Furthermore, active learning is more adaptable to dynamic datasets, dynamically selecting new samples based on the current state of the model and adapting to shifts in data distribution. Existing methods typically rely on static datasets. Existing machine learning methods are sensitive to data quality, and outliers or low-quality data can lead to performance degradation. By prioritizing high-quality, informative data points, active learning improves the model's generalization to new data and robustness to noisy data. For example, it selects samples near the decision boundary, which are most critical for model improvement, thereby mitigating the impact of low-quality data and enhancing the model's stability in complex environments.

[0039] The specific training of the back propagation neural network model is as follows Figure 2 Shown, including:

[0040] Use the data in the training dataset to train the back propagation neural network model, and use the trained back propagation neural network model to predict the probability distribution of each group of data in the unlabeled dataset;

[0041] Determine the corresponding confidence score based on the probability distribution of each set of data, and rank the uncertainty of each set of data in the unlabeled dataset based on the confidence score;

[0042] Determine multiple groups of target data based on the uncertainty sorting results, such as selecting the top 10% or 15% of the sorted unlabeled data set, and manually label the corresponding ultrasonic process parameters for the selected target data;

[0043] The labeled multiple sets of target data are added to the training dataset, and the back propagation neural network model is retrained based on the added training dataset.

[0044] In an embodiment of the present application, the back propagation neural network model includes a first propagation layer, a second propagation layer and a loss function, wherein the first propagation layer generates a first prediction result based on the input data of the back propagation neural network model, wherein the input data includes stress data and deformation data in the training data set; the second propagation layer generates a second prediction result based on the first prediction result; the loss function calculates the error based on the second prediction result and the ultrasonic process parameters of the training data set, passes the error back to the input of the back propagation neural network model, and updates the model parameters of the neural network model based on the error until the loss function converges.

[0045] Among them, stress data includes residual stress value and residual stress uniformity (variance); deformation data includes angular deformation value and shrinkage deformation value; the loss function can be an MSE (mean square error) function; ultrasonic process parameters include control time, ultrasonic frequency, ultrasonic amplitude, number of sound sources, etc.

[0046] It can be understood that the back propagation neural network model of the embodiment of the present application includes a first propagation layer, a second propagation layer and a loss function, wherein:

[0047] The first propagation layer generates a first prediction result based on input data of the back propagation neural network model, where the input data includes stress data and deformation data in the training data set;

[0048] The second propagation layer generates a second prediction result based on the first prediction result;

[0049] The loss function calculates the error based on the second prediction result and the actual ultrasonic process parameters, passes the error back to the input of the back propagation neural network model, and updates the model parameters of the neural network model according to the error until the loss function converges.

[0050] In an embodiment of the present application, the propagation layer structure of the first propagation layer and the second propagation layer is the same, wherein the propagation layer structure includes: multiple hidden layers, nonlinear activation functions and fully connected layers, wherein the multiple hidden layers are used to extract and transform features of the input data of the back propagation neural network model; the nonlinear activation function is used to introduce nonlinear characteristics to the output data of the multiple hidden layers; and the fully connected layer is used to output the prediction results.

[0051] The number of hidden layers may be 3, and the nonlinear activation function may be a Sigmoid activation function.

[0052] It can be understood that the structures of the first propagation layer and the second propagation layer of the back propagation neural network model in the embodiment of the present application are the same, wherein the multiple hidden layers are used to extract and transform the input data of the back propagation neural network model;

[0053] The nonlinear activation function is used to introduce nonlinear characteristics to the output data of the multi-layer hidden layer; the fully connected layer is used to output the prediction results.

[0054] In an embodiment of the present application, updating the model parameters of the back propagation neural network model according to the error includes: calculating the model gradient of the back propagation neural network model based on the error; and updating the model parameters of the back propagation neural network model based on the model gradient and a preset learning rate.

[0055] The preset learning rate can be set according to specific circumstances and is not specifically limited thereto, for example, 0.1.

[0056] It is understandable that the embodiments of the present application can calculate the model gradient of the back-propagation neural network model based on the error, and update the model parameters of the back-propagation neural network model based on the model gradient and the preset learning rate.

[0057] Specifically, the construction and training process of the back propagation neural network model in the embodiment of the present application is as follows:

[0058] 1. Data collection.

[0059] A large amount of experimental data on residual stress control was collected during the production and high-energy ultrasonic control process, including control time, high-energy ultrasonic frequency, high-energy ultrasonic amplitude, number of sound wave sources, angular deformation value, shrinkage deformation value, residual stress value, residual stress uniformity (variance) and other parameters to construct a data set.

[0060] 2. Construction of error back propagation neural network model (bp algorithm).

[0061] An error back propagation neural network model is constructed with angular deformation value, shrinkage deformation value, residual stress value, and residual stress uniformity (variance) as input and output parameters such as control time, ultrasonic frequency, ultrasonic amplitude, and number of sound sources.

[0062] In the first step, initial weights are randomly assigned. The input layer receives values ​​for angular deformation, shrinkage deformation, residual stress, and residual stress uniformity (variance) and transmits them to the hidden layer. Sigmoid activation functions in the three hidden layers perform a nonlinear transformation on the inputs. The fully connected layer then outputs the predicted results, including control time, ultrasonic frequency, ultrasonic amplitude, and the number of sound sources. This completes the forward propagation. In the second step, the mean square error (MSE) function is used to calculate the error between the predicted and true values ​​of deformation and residual stress. This error is then propagated back from the output layer to the input layer. The gradient of each layer is calculated, and the weights and biases of the optimization network are updated using gradient descent.

[0063] In the third step, repeat the forward propagation, error calculation and back propagation, and weight update of the first two steps, and continue iterating until the loss function converges.

[0064] 3. Model training and improvement.

[0065] A backpropagation neural network model was trained using the preprocessed data. The model had three hidden layers, with 10 neurons in the first layer, 10 neurons in the second layer, and 10 neurons in the third layer. The dataset consisted of 880 samples, and the data set was divided into training, test, and validation sets in an 8:1:1 ratio. The model was iterated through the training set for 1000 epochs with a learning rate of 0.1. After training, the model made predictions on a new set of unlabeled samples, calculating the confidence score of the predicted probability distribution for each sample and identifying the samples for which the model had the least uncertainty. The samples with the highest uncertainty in the new set were ranked according to uncertainty, with the top 10% selected. These samples were relabeled and added to the training set.

[0066] In addition, it should be noted that this application can use fuzzy logic / neuro-fuzzy control to deal with nonlinear and uncertain problems without relying on BP neural network and FPGA neural network accelerator. This solution maps data inputs such as angular deformation and residual stress to control outputs such as ultrasonic frequency and amplitude by defining several language variables and rule bases, and can use ANFIS to adjust the membership function online to achieve adaptive optimization; or use model predictive control (MPC) to modularize the constrained MPC algorithm on FPGA, and achieve optimal adjustment of welding parameters by solving the finite time domain optimization problem online, which has both Dynamic prediction capabilities for future control systems; or the use of sliding mode control (SMC), whose structure includes switching surface design and equivalent control laws, can provide robust compensation for system model uncertainties and external disturbances, and can parallelize the switching of FPGA logic and the signal feedback processing of monitoring devices, significantly reducing response delays and improving robustness. In terms of parameter optimization and online labeling, evolutionary algorithms (such as genetic algorithms) can also be introduced to search for the optimal ultrasonic control parameter combination offline or online, or combined with fuzzy controllers to form a GA-FLC hybrid strategy, which improves control accuracy by optimizing membership and rule bases, achieving more precise control. Reinforcement learning (RL) is also a viable alternative. By defining states (angular deformation, shrinkage, etc.), actions (adjusting frequency / amplitude), and rewards (weld quality indicators), Deep Q-Network or Actor-Critic algorithms are used to update strategies online to achieve adaptive control of dynamic data distribution.

[0067] Furthermore, a minimum variance estimator can be embedded within the adaptive Kalman filter framework to correct sensor noise and system parameter estimates in real time, providing high-quality state feedback for subsequent controllers (such as PID or MPC). FPGA implementation can accelerate filtering and estimation operations in parallel. These solutions can all be implemented on FPGA platforms such as Xilinx Artix-7 / Zynq-7000, achieving low-latency, low-power closed-loop control and online optimization through pipeline parallelization and resource reuse.

[0068] The ultrasonic process parameter control system of the present application is described below through a specific embodiment. Figure 3 As shown, it mainly includes: ultrasonic stress detection and deformation monitoring device, FPGA module, and high-energy ultrasonic power supply controller.

[0069] The system timing can be described as follows: first, the receiver receives the detection data of the ultrasonic stress detection device and the deformation detection device, and then transmits the data to the receiver through the I 2The C interface sends data to the FPGA / SoC. The neural network module then performs forward reasoning on the state data and outputs control signals (control time, frequency, amplitude, and number of control sources). These control signals drive the power controller of the high-energy ultrasonic control system through a DAC, PWM, or dedicated interface. The power controller then acts on the high-energy ultrasonic amplifier, changing its parameters and regulating the residual stress again. After the control begins, the detection equipment remeasures the new data and feeds it back into the next cycle. In the FPGA design, the neural network reasoning module can be placed in the programmable logic (PL) to achieve maximum parallelism and minimum latency. In addition, other control logic (state machine, data scheduling, etc.) is placed on the PL or on-chip CPU (PS) side.

[0070] Through this system, the FPGA reduces inference latency to below microseconds through pipeline and hierarchical parallelism. Control and learning tasks are accomplished collaboratively between the programmable logic and the on-chip ARM CPU (PS) via a high-bandwidth AXI bus, ensuring real-time data exchange and bandwidth requirements. This enables dynamic closed-loop optimization of welding deformation and residual stress control, improving the stability and adaptability of the high-energy ultrasonic control process. Ultimately, high-quality welds are achieved, meeting the requirements of several critical and complex components.

[0071] In summary, this application introduces active learning and FPGA-based real-time closed-loop control technology. First, through active learning, key samples (such as high-uncertainty data) are dynamically screened and combined with virtual simulation to replace physical experiments to quickly generate the optimal stress control parameter combination, which greatly reduces the cost and time of the experiment. At the same time, the hardware parallel computing capability based on FPGA quickly realizes real-time data processing and neural network reasoning, and dynamically adjusts the parameters of high-energy ultrasonic control to respond to environmental changes, which can avoid the lag and uncertainty caused by traditional static control. Compared with the method of setting parameters only in the initial stage, the combination of the two not only breaks through the traditional method's dependence on experience and static data, but also adapts to the needs of multiple scenarios and complex welds by building a hardware architecture, greatly improving the accuracy, stability and industrial applicability of the control, and providing an efficient and adaptive solution for the large-scale application of high-energy ultrasonic residual stress technology.

[0072] In general, the ultrasonic process parameter control system of this embodiment can achieve the following effects:

[0073] 1、Real-time performance is strong, FPGA hardware and parallel processing capabilities make the control system have very low delay, can timely process the key data monitored in the regulation process and carry out optimization and control, deploy the back propagation neural network model on the FPGA platform, use its hardware parallelization characteristics to realize fast response and feedback, can support high-frequency real-time data processing and dynamic process parameter adjustment, realize the closed-loop control of "parameter-result-FPGA-optimized parameter".

[0074] 2、High parallel processing capability: FPGA can process multiple control loops simultaneously, suitable for complex welding part regulation and control, improving the overall performance of the system.

[0075] 3、Flexibility and reconfigurability: The programmable nature of FPGA allows dynamic adjustment of control signals according to demand, to adapt to different working conditions and equipment regulation and control, can adapt to multiple scene requirements, without changing hardware to complete function upgrade, reduce the cost and time of hardware replacement in high-energy ultrasonic stress regulation system.

[0076] 4、High reliability and safety: In safety-critical applications, FPGA provides higher system reliability, reduces the risk of software failure, and can achieve more stable regulation and control, homogenization and significantly reduce residual stress.

[0077] 5、Economy and scalability: Although the initial development cost of FPGA may be higher, its reusability and flexibility reduce the cost of optimization analysis in the long-term regulation process, and it is easy to expand, can adapt to the growth of future performance requirements, through software and hardware co-design to integrate control logic and learning module, reduce the debugging of complex code, shorten the development cycle, improve the system scalability and industrial applicability.

[0078] 4、In the process of training the back propagation neural network model, the mechanism of active learning is introduced, which can select the samples uncertain to the model from the unlabeled data pool and request labeling, eliminate the parameters with poor regulation effect, thereby greatly reducing the cost and time.

[0079] According to the ultrasonic process parameter regulation system provided in the embodiment of the present application, the detection equipment transmits the stress data and deformation data of the welded part detected to the processing circuit, the processing circuit is provided with a processing module and an inference module, the processing module is used for processing the stress data and deformation data, and a control signal is generated based on the adjusted ultrasonic process parameters, the inference module determines the adjusted ultrasonic process parameters by using a back propagation neural network model and the processed stress data and deformation data, the adjustment of the ultrasonic process parameters is realized through the back propagation neural network model, the regulation accuracy is improved, the control signal is transmitted to the controller, the controller adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal, the closed-loop control of the adjustment of the ultrasonic process parameters is realized, the ultrasonic process parameters can be adjusted in time according to the stress data and deformation data after the adjustment of the ultrasonic process parameters, the dynamic adaptability is improved, and the circuit does not need to be redesigned or the components need to be replaced, the development cycle is short, and the maintenance cost is low.

[0080] Secondly, the ultrasonic process parameter regulation method provided in the embodiment of the present application is described with reference to the accompanying drawings.

[0081] Figure 4 The flowchart of the ultrasonic process parameter regulation method in the embodiment of the present application is shown in the figure.

[0082] As shown in the figure, Figure 4 The ultrasonic process parameter regulation method is realized based on the ultrasonic process parameter regulation system, and includes the following steps.

[0083] In step S101, the stress data and deformation data of the welded part are obtained.

[0084] The stress data includes residual stress value and residual stress uniformity (variance), and the deformation data includes angular deformation value and shrinkage deformation value.

[0085] In step S102, the stress data and deformation data are processed, and the adjusted ultrasonic process parameters are determined by using a back propagation neural network model and the processed stress data and deformation data.

[0086] The ultrasonic process parameters include regulation time, ultrasonic frequency, ultrasonic amplitude, and sound source quantity.

[0087] It can be understood that the stress data and deformation data can be processed, the processed stress data and deformation data can be input into the back propagation neural network model, and the adjusted ultrasonic process parameters can be determined by the back propagation neural network model.

[0088] In step S103, a control signal is generated based on the adjusted ultrasonic process parameters, and the control signal is sent to the controller, and the controller adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal.

[0089] It can be understood that the embodiment of the present application can generate a control signal based on the adjusted ultrasonic process parameters, and send the control signal to the controller, and the controller adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal, so as to realize closed-loop adjustment control of the ultrasonic process parameters, and the ultrasonic process parameters can be adjusted in time according to the stress data and deformation data after the ultrasonic process parameters are adjusted, accurate regulation and control in the welding process are realized, and the effect and stability of the residual stress regulation process are ensured.

[0090] It should be noted that the foregoing explanation and description of the embodiment of the ultrasonic process parameter regulation system also apply to the ultrasonic process parameter regulation method of the embodiment, which will not be described here.

[0091] According to the ultrasonic process parameter regulation method provided in the embodiment of the present application, the adjusted ultrasonic process parameters can be determined by using the back propagation neural network model, the stress data and the deformation data according to the stress data and the deformation data of the welded part, a control signal is generated based on the adjusted ultrasonic process parameters, and the control signal is sent to the controller, and the controller adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal, so as to realize closed-loop adjustment control of the ultrasonic process parameters, and the ultrasonic process parameters can be adjusted in time according to the stress data and deformation data after the ultrasonic process parameters are adjusted, accurate regulation and control in the welding process are realized, and the effect and stability of the residual stress regulation process are ensured.

[0092] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program or instructions, and the computer program or instructions are executed by a processor to realize the ultrasonic process parameter regulation method as above.

[0093] The embodiment of the present application also provides a computer program product, which includes a computer program or instructions, and the computer program or instructions are executed to realize the ultrasonic process parameter regulation method as above.

[0094] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that the specific features, structures, materials or characteristics described in combination with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, the different embodiments or examples described in the present specification and the features of the different embodiments or examples can be combined and combined by those skilled in the art without contradiction.

[0095] Furthermore, the terms "first", "second", etc. are used herein only to describe one implementation, and do not imply either an actual temporal sequence or an order of importance, unless explicitly stated otherwise. Thus, a feature defined with "first", "second" etc. can include one or more of the features implicitly or explicitly.

[0096] Any process or method descriptions or blocks in flow charts described herein and elsewhere can be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps in the process. Alternate implementations are included within the scope of the preferred embodiments of the present application in which additional functionality can be added or one or more steps can be modified, eliminated, or added. Modifications are also included within the scope of the present application where appropriate, e.g., permutations of the attributes discussed or other steps added for purposes of implementing functionality. The word "including" and "comprising" as used herein is used in the open-ended way, i.e. meaning "including, but not limited to", and thus should be interpreted to cover the above aspects as well as other aspects of the application.

[0097] It should be understood that aspects of the application can be implemented in software, firmware, hardware, or combinations thereof. In the embodiments described above, the steps or methods can be implemented in software or firmware to be executed by a special purpose computer or general purpose computer. As such, the steps of the methods described above can be stored in a computer readable medium, which can include random access memory (RAM), read only memory (ROM), magnetic media, optical media, or any other suitable medium, etc. As will be appreciated by persons skilled in the art, the steps of the methods described above can be implemented in hardware, software, firmware, or combinations thereof, using any of a variety of techniques known in the art. For example, the steps can be implemented using object oriented techniques, or using any other techniques that are well known in the art.

[0098] Those of skill in the art will appreciate that the steps of the methods described above can be carried out by program instructions stored in a computer readable medium, which can be executed by a special purpose computer or general purpose computer. The computer readable medium can include a floppy disk, RAM, ROM, removable media, or any other suitable medium.

Claims

1. An ultrasonic process parameter control system, characterized in that: include: Testing equipment, used to detect stress data and deformation data of welded parts; at least one processing circuit, wherein the processing circuit is provided with a processing module and a reasoning module, the processing module processes the stress data and the deformation data and generates a control signal based on the adjusted ultrasonic process parameters, and the reasoning module determines the adjusted ultrasonic process parameters using a back propagation neural network model and the processed stress data and deformation data; A controller is used to adjust ultrasonic process parameters of the ultrasonic welding system based on the control signal.

2. The ultrasonic process parameter control system according to claim 1, characterized in that: The reasoning module is deployed in the programmable logic of the processing circuit; the processing module is deployed in the programmable logic or on-chip processing system of the processing circuit.

3. The ultrasonic process parameter control system according to claim 1, characterized in that: The back propagation neural network model is pre-trained, and the training process of the back propagation neural network model includes: Obtaining a training data set and an unlabeled data set, wherein the training data set includes multiple sets of ultrasonic process parameters and corresponding stress data and deformation data, and the unlabeled data set includes multiple sets of stress data and deformation data; Using the training data set to train the back propagation neural network model, and using the trained back propagation neural network model to predict the probability distribution of each group of data in the unlabeled data set; Determining a corresponding confidence score based on the probability distribution of each set of data, and ranking the uncertainty of each set of data in the unlabeled dataset based on the confidence score; Based on the uncertainty sorting results, multiple groups of target data are determined, corresponding ultrasonic process parameters are labeled for the multiple groups of target data, and the labeled multiple groups of target data are added to the training data set, and the back propagation neural network model is retrained based on the added training data set.

4. The ultrasonic process parameter control system according to claim 1, characterized in that: The controller includes: a power supply controller and an ultrasonic amplifier, wherein: The power controller is configured to drive the ultrasonic amplifier based on the control signal; The ultrasonic amplifier is used to adjust the ultrasonic process parameters of the ultrasonic welding system based on the control signal.

5. The ultrasonic process parameter control system according to claim 2, characterized in that: The back propagation neural network model includes a first propagation layer, a second propagation layer and a loss function, wherein, The first propagation layer generates a first prediction result according to input data of the back propagation neural network model, wherein the input data includes stress data and deformation data in a training data set; The second propagation layer generates a second prediction result according to the first prediction result; The loss function calculates an error based on the second prediction result and the ultrasonic process parameters of the training data set, passes the error back to the input of the back propagation neural network model, and updates the model parameters of the neural network model based on the error until the loss function converges.

6. The ultrasonic process parameter control system according to claim 5, characterized in that: The first propagation layer and the second propagation layer have the same propagation layer structure, wherein the propagation layer structure includes: multiple hidden layers, nonlinear activation functions and fully connected layers, wherein, The multi-layer hidden layer is used to extract and transform the input data of the back propagation neural network model; The nonlinear activation function is used to introduce nonlinear characteristics to the output data of the multi-layer hidden layer; The fully connected layer is used to output the prediction result.

7. The ultrasonic process parameter control system according to claim 5, characterized in that: The updating of the model parameters of the back propagation neural network model according to the error comprises: Calculating a model gradient of the back propagation neural network model based on the error; The model parameters of the back propagation neural network model are updated based on the model gradient and a preset learning rate.

8. A method for controlling ultrasonic process parameters, characterized in that: The method is implemented based on the ultrasonic process parameter control system according to any one of claims 1 to 7, and comprises the following steps: Obtain stress and deformation data of welded parts; Processing the stress data and the deformation data, and determining adjusted ultrasonic process parameters using a back propagation neural network model, the processed stress data and the deformation data; A control signal is generated based on the adjusted ultrasonic process parameters and sent to a controller, which adjusts the ultrasonic process parameters of the ultrasonic welding system based on the control signal.

9. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: The computer program or instruction is executed by a processor to implement the ultrasonic process parameter control method according to claim 8.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed, the ultrasonic process parameter control method according to claim 8 is implemented.

Images