Welding quality intelligent evaluation method based on current data

Abstract

The invention belongs to the technical field of welding, and discloses an intelligent welding quality evaluation method and system based on current data. The system comprises a current signal acquisition module, a signal preprocessing module, a time-frequency feature extraction module, an improved DenseNet convolutional neural network module, a welding quality prediction module and a real-time monitoring module. According to the method for extracting the time-frequency characteristics of the welding current signals through continuous wavelet transformation, the time-frequency characteristics are analyzed and trained through an improved DenseNet convolutional neural network model, so that a welding quality prediction model is established, and online evaluation is conducted on the welding quality through a real-time monitoring module. The method is suitable for the welding quality control stage, unified and standardized management is conducted on the welding quality evaluation process, and the defect rate in the welding process is reduced; the feature extraction and evaluation method is high in accuracy, and the welding quality is remarkably improved; and through real-time and visual monitoring, the technical level and experience requirements of operators on welding quality evaluation are reduced, and the detection time is saved.

Description

Technical Field

[0001] The present invention relates to the field of welding technology, and in particular to an intelligent welding quality evaluation method based on current data. Background Art

[0002] Traditional welding quality detection methods mainly rely on manual inspection, including visual inspection, X-ray, ultrasonic, cupping test, etc. These methods have problems such as low efficiency, poor accuracy, and strong subjectivity. In recent years, welding quality prediction methods based on welding parameters have gradually become a research hotspot. The welding parameters usually used are gas flow, welding temperature, wire feeding speed, welding current, etc. At present, most methods can only deal with problems with simple structures and obvious features. There are also some welding quality assessment methods based on machine learning and deep learning. These methods still have problems such as insufficient feature extraction and limited model accuracy. With the development of industrial automation, the quality control requirements in the welding process are getting higher and higher, and there is an urgent need for an efficient, accurate and objective welding quality assessment method. Therefore, the present invention proposes an intelligent welding quality assessment method based on an improved DenseNet model. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent welding quality evaluation method based on current data, and to improve the accuracy and efficiency of welding quality prediction through an improved DenseNet model.

[0004] Step 1: Data Collection

[0005] During the welding process, the welding current signal is collected in real time through the data acquisition module with a collection frequency of 50Hz to 1000Hz to ensure the integrity of the signal and capture details.

[0006] Step 2: Data preprocessing

[0007] The collected current signal is preprocessed, and some invalid data collected before and after welding are eliminated. At the same time, a bandpass filter is used to remove high-frequency and low-frequency noise to eliminate environmental interference and equipment noise, ensuring the purity of the signal so as not to affect subsequent deep learning training.

[0008] Step 3: Feature extraction

[0009] The collected welding current signal is subjected to continuous wavelet transform (CWT) to extract the time-frequency characteristics of the current signal during the welding process. Continuous wavelet transform is a time-frequency analysis tool that decomposes the signal into convolutions of different frequencies and wavelet functions. It can effectively describe the changes in the signal in time and frequency, thereby capturing the dynamic characteristics of the welding process.

[0010] The continuous wavelet transform is defined as follows:

[0011]

[0012] Where x(t) is the input signal, is the wavelet function, a is the scale parameter, and b is the translation parameter

[0013] Step 4: Create a dataset

[0014] The data after continuous wavelet transformation is used to generate a 16*1 time-frequency diagram and then cropped into 8 2*1 images, so as to distinguish which segments are qualified and which are unqualified during welding. The original image is mirrored left and right. This operation not only expands the scale of the data set, but also helps to improve the generalization ability of the model. Then the generated images are normalized to ensure that the data have the same scale. Finally, the data set is divided into training set and test set according to 8:2.

[0015] Step 5: Model construction

[0016] The attention mechanism is embedded in the DenseNet model. Each dense block consists of multiple convolutional layers, and each convolutional layer receives the output of all previous convolutional layers. An attention module is added after each dense block to perform weighted processing on the feature map to generate enhanced feature representation.

[0017] Step 6: Model training

[0018] The training set data and test set data are input into the DenseNet model for training, and the model parameters are adjusted to enable it to accurately predict the welding quality. The cross-validation method is used to evaluate the model performance during the training process, and the prediction accuracy of the model is continuously improved through the optimization algorithm.

[0019] Step 7: Online Assessment

[0020] In practical applications, the real-time collected welding current signal is input into the trained DenseNet model to predict the welding quality online. The welding quality score output by the model includes the defect type and defect location. It can be used to monitor the welding process in real time and detect and correct welding defects in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0022] Figure 1It is a schematic diagram of the continuous wavelet transform feature extraction process.

[0023] Figure 2 This is a schematic diagram of the initial DenseNet model structure.

[0024] Figure 3 This is a schematic diagram of the DenseNet model structure that introduces the attention mechanism.

[0025] Figure 4 It is a flowchart of model training and prediction. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Example 1

[0028] Step 1: Obtain welding current signals in real time through high-precision sensors and data acquisition systems. The sensors are installed on the welding equipment to accurately capture the changes in current during welding. The sampling frequency of the data acquisition system is set to 50Hz to 1000Hz to ensure signal integrity and detail capture. The key to this step is to ensure high fidelity and stability of the data for subsequent analysis and processing.

[0029] Step 2: Preprocess the collected current signal. First, remove invalid data before and after welding to ensure that the object of analysis is the actual welding process signal. Use a bandpass filter to remove high-frequency and low-frequency noise to eliminate environmental interference and equipment noise to ensure the purity of the signal. The purpose of the preprocessing step is to remove redundant information and noise, improve the quality of the signal, and thus ensure the accuracy of model training.

[0030] Step 3: Perform continuous wavelet transform (CWT) on the preprocessed current signal to extract time-frequency features. CWT is an effective time-frequency analysis tool that can decompose the signal into convolutions of different frequencies and wavelet functions, and can describe the changes in the signal in time and frequency. Through CWT, the dynamic characteristics of the welding process can be captured, providing high-quality feature data for subsequent model training.

[0031] Step 4: Generate a 16×1 time-frequency image from the time-frequency features extracted by CWT, and then crop it into 8 2×1 images to distinguish which segments are qualified and which are unqualified during the welding process. In this way, more fine-grained feature data can be obtained. Next, the original image is mirrored left and right to expand the scale of the dataset and improve the generalization ability of the model. Finally, the generated images are normalized to ensure that the data have the same scale. The dataset is divided into a training set and a test set in a ratio of 8:2 to provide basic data for model training and evaluation.

[0032] Step 5: Embed the attention mechanism into the DenseNet convolutional neural network model to enhance the model's ability to weight feature maps. The DenseNet model contains multiple dense blocks, each of which consists of multiple convolutional layers, and each convolutional layer receives the output of all previous convolutional layers. Add an attention module (such as the SE module) after each dense block to perform weighted processing on the feature map and generate an enhanced feature representation. The attention mechanism can adaptively adjust the weight of the feature map, highlight important features, and improve the recognition ability and robustness of the model.

[0033] Step 6: Input the training set data into the improved DenseNet model for training. During the training process, the cross-validation method is used to evaluate the model performance, and the prediction accuracy of the model is continuously improved through the optimization algorithm. The specific training steps are as follows:

[0034] Initialize model parameters and set hyperparameters such as learning rate, batch size, and number of iterations.

[0035] Use the training set data for forward propagation and calculate the prediction results for each sample.

[0036] The loss function between the predicted results and the actual labels is calculated, and the model parameters are adjusted through the back-propagation algorithm.

[0037] After each iteration, the validation set data is used to evaluate the performance of the model and the hyperparameters are adjusted based on the validation results.

[0038] Repeat the above process until the model converges and achieves the best prediction effect.

[0039] Step 7: In actual application, the real-time collected welding current signal is input into the trained DenseNet model for online evaluation. The model outputs the welding quality score in real time, including the type and location of welding defects. The evaluation results can be used to monitor the welding process in real time, detect and correct welding defects in time, and improve the efficiency and accuracy of welding quality control.

[0040] Experimental results and analysis

[0041] Through the above steps, a complete intelligent welding quality assessment system was established. In the actual welding project, the system can monitor the current changes in the welding process in real time and accurately assess the welding quality. The experimental results show that this method can effectively predict the welding quality, and the assessment accuracy rate reaches 96.55%. Compared with the traditional welding quality detection method, the method described in this paper has higher accuracy and efficiency.

[0042] It is worth noting that the contents not described in detail in the embodiments of the present invention belong to the prior art known to professionals in the field;

[0043] The above descriptions are only some embodiments of the present application, and certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A welding quality intelligent evaluation method based on current data, characterized in that: The following steps are involved: Step 1: During the welding process, a high-precision current sensor is used to collect the signal of the current change over time during welding in real time to ensure the accuracy and completeness of data collection. The acquisition frequency range is 50Hz to 1000Hz to capture the details of the current change during the welding process and record the type of welding defects. Step 2: Preprocess the collected current signal. Step 3: Use cgau6 wavelet to perform continuous wavelet transform on the preprocessed current signal to obtain high-resolution time-frequency characteristics of the current signal. Step 4: Create a data set based on the time-frequency feature data after continuous wavelet transform. Step 5: Use the improved DenseNet convolutional neural network model to train and analyze the extracted time-frequency features and establish a welding quality prediction model; Step 6: Input the real-time collected welding current signal into the trained welding quality prediction model, and output the welding quality evaluation results in real time through the online evaluation function of the model. The evaluation results include the type and location of welding defects, helping operators to promptly discover and deal with problems in the welding process, and improve welding quality and production efficiency.

2. The intelligent welding quality assessment method based on current data according to claim 1 is characterized in that: Some invalid data collected before and after welding are eliminated. At the same time, a bandpass filter is used to pre-process the collected current signal to remove high-frequency and low-frequency noise.

3. The welding quality intelligent evaluation method based on current data according to claim 1 is characterized in that: In step 4, a data set is first prepared by using continuous wavelet transform to obtain a three-dimensional time-frequency diagram of the time domain, frequency domain, and wavelet coefficients after the time-frequency model of the welding current, and the type of welding defect is marked at the same time. The time-frequency diagram is divided into several segments at the same interval, and the data set is increased so that it can be distinguished whether each segment in the welding is normal in the future.

4. The welding quality intelligent evaluation method based on current data according to claim 1 is characterized in that: In step 5, an improved DenseNet convolutional neural network model is constructed, and the model includes multiple dense blocks, each of which is composed of multiple convolutional layers, and each convolutional layer receives the output of all previous convolutional layers. At the same time, an attention mechanism is introduced, and an attention module is added to the DenseNet model to adaptively recalibrate the feature channels to enhance the effectiveness of feature extraction.

5. The welding quality intelligent evaluation method based on current data according to claim 1 is characterized in that: In step 6, the welding quality assessment result includes the type and location of the welding defects.

Images