Quality monitoring of resistance spot welding based on digital twin

DOI: https://doi.org/10.21203/rs.3.rs-2428723/v1

Abstract

As an important means to realize intelligent manufacturing, digital twin is a digital expression of physical entities, which realizes virtual-real interaction and iterative optimization of product design and manufacturing by constructing a bridge of information mapping between the physical world and the virtual world. Resistance spot welding technology is widely used in automotive manufacturing, aerospace and other fields as a spot linking process for the manufacture of thin sheet structures. Resistance spot welding is a highly nonlinear coupled process, and physical models make it difficult to accurately monitor its quality. This paper takes 2219/5A06 aluminum plates with different thicknesses as the object, and applies digital twin technology to the welding process monitoring of aluminum plates to effectively improve the quality and efficiency of aluminum plate welding. In order to break through the key technologies such as information interaction in the digital twin system, a data acquisition system for resistance spot welding process is established and a real-time data processing technology based on wavelet threshold analysis is proposed. Based on the real-time data, the processed process parameters are tested in Digital-twin space to verify the feasibility of the solution. Feedback process parameters to physical entities to enable online quality monitoring of resistance spot welding and promote product quality stability during production.

1. Introduction

Resistance spot welding technology is a kind of point connection technology widely used in the manufacture of thin plate structure. Its principle is to apply a certain pressure between two electrodes and the workpiece to be welded, and use the resistance heat generated when the current passes through the workpiece to melt the local metal and form a welding spot process. Resistance spot welding is widely used in automotive, aerospace, home appliance and other manufacturing fields due to its advantages of low cost and high production efficiency[1].

With the development of industry, the quality of resistance spot welding heads is becoming more and more demanding in various applications. However, the resistance spot welding process involves force, heat, electricity, magnetism and flow, which is a highly nonlinear coupling process, which makes it difficult to accurately monitor the quality of the physical model[2]. At the same time, affected by interference factors such as assembly gaps and electrode wear, welding quality problems such as spatter and cracks appear, which increases the uncertainty of the welding process. Therefore, an effective resistance spot welding monitoring technology is needed to realize real-time monitoring of process elements in the welding process and ensure product welding quality.

Since resistance spot welding is a fully closed process, direct observation of the nucleation process is not possible, so the nucleation process can only be inferred by monitoring the physical phenomena accompanying the spot-welding process. The former process signals that can be applied to monitor spot weld quality in real time include dynamic resistance[3, 4],electrode pressure[5] and electrode displacement[6]. Dynamic resistance is one of the most widely used process signals in the field of resistance spot welding quality monitoring. In 1950, Roberts first experimentally discovered the change in dynamic resistance during resistance spot welding of mild steel [7]. The dynamic resistance signal is obtained by measuring the welding current and the voltage signal across the electrode and calculating it using Ohm's law. In the welding high current operating environment, the time-varying characteristics of the current lead to large induced noise in the voltage signal, and the traditional current-voltage ratio algorithm produces serious calculation errors. Gong[8] proposed the current-over-zero derivative ratio method to achieve the calculation of dynamic resistance by analyzing the voltage and current signals of the primary circuit of resistance spot welding and retrograde modeling, which requires calibration of the inductance value of the secondary circuit of the welding equipment before use. Su[9] et al. performed dynamic resistance measurements by a recursive least squares algorithm based on a forgetting factor, and optimized the genetic factor using sensitivity analysis, which can effectively eliminate inductive noise. Ji[10] et al. found a significant trend in electrode displacement during resistance spot welding of thin aluminum alloys. Panza[11] et al. used the analysis of the electrode unique signal from a non-contact sensor installed in the welding machine to monitor the electrode degradation during the welding process and thus analyze the quality of the welding process. Wang[12] established the HMM model for spot welding quality judgment by using the time series model to monitor the waveform curve of the electrode pressure during the resistance spot welding process. Tang’s[13] experimental studies have shown that different electrode pressure signals are obtained when spot welding with different welding equipment is applied to the same material, so the study of electrode pressure signals needs to be differentiated according to the type of welding equipment. Sensing and detection technology for different signals of resistance spot welding process has been relatively mature, however, achieving accurate measurement of dynamic resistance under variable current conditions has not been practically solved and failed to break through the electrode displacement measurement technology of the weld clamp mechanism.

At present, the application of intelligent technology in the field of welding from process optimization, welding quality monitoring and other aspects greatly promote the progress of welding technology. Wang[14] proposes a framework for spot weld quality using signal processing and artificial intelligence techniques that uses a non-invasive Rogowski coil to extract electrical signal features, an RNN to evaluate the size of the hot zone of the welding process, and a new self-organizing mapping-type classifier to detect the time of occurrence. Zhang[15] et al. proposed a new method of converting electrode displacement signals into binary images in order to improve the efficiency of acquiring resistance spot welding monitoring features, and while selecting displacement signals as monitoring welding process features, probabilistic neural networks were used to classify the quality of resistance spot welding welds, which retains as much information as possible about the quality of the weld and avoids the extraction of complex features of the welding process and the selection of intelligent algorithms. Dai[16] et al. proposed a GAN-based data enhancement technique to improve the quality of spot welding process. Firstly, BAGAN and gradient penalty (GP) were used to generate the welding process images, and secondly, a migration learning approach was used to construct an image classifier, and the images generated by BAGAN-GP were added to the training set to improve the classification performance of the classifier. Most of the above methods rely on historical data analysis in the process of welding monitoring, making it difficult to guide timely quality monitoring and subsequent process decisions.

The concept of the digital twin was originally developed by Grieves[17] and is used in the military and aerospace sectors. The digital twin technology has gained widespread attention and practical application in various fields of industry due to its features of virtual-real integration and real-time interaction, iterative operation and optimization, and full-factor data drive. Tao[18] et al. describe the digital twin as a virtual model of a physical entity created digitally to simulate the behavior of the physical entity in the real environment with the help of data, providing a more real-time, efficient and intelligent service oriented to the whole product lifecycle process.

In summary, most of the above-mentioned studies have contributed to the rapid development of resistance spot welding quality monitoring technology. However, there are limitations in the practical application of the current research results, and there is a mismatch between the actual welding conditions and the experimental conditions. To this end, this paper proposes a digital twin-based approach aimed at monitoring the resistance spot welding process in real time, establishing a high-precision quality evaluation model capable of adapting to a variety of situations, achieving dynamic control of welding quality, and ensuring product stability.

The main contributions of this paper are as follows. Establishing a multi-scale twin model based on a real resistance spot welding environment, synchronizing the twin model with the physical model in real time, and realizing resistance spot welding process optimization and quality monitoring on the basis of data collected in the field. Based on historical data and future data, the optimized process parameters are tested in twin space to verify the feasibility of the optimized solution afterwards. Feedback of process parameters to physical entities for online quality control of resistance spot welding to promote stability of product quality during production.

2. Resistance Spot Welding System And Digital Twin Environment

With the change of manufacturing industry to flexible and intelligent, the traditional welding process monitoring cannot meet the modern demand of fine and high quality production, and it is difficult to realize intelligent control of welding process. Therefore, this paper proposes a digital twin model for resistance spot welding process. The model consists of five modules: physical entity layer, model layer, data layer, communication layer, and application layer. As shown in Fig. 1.

Physical entity layer. The physical entity layer is the physical foundation of the digital twin technology and is also the underlying source for acquiring data. The physical entity layer is composed of elements such as resistance spot welding equipment, work environment, workpieces and operators, which contain a large amount of dynamic and static data that affects the quality of the weld.

Model layer. The model layer is a key vehicle for weld quality prediction by using the data obtained from the physical layer to build a high-fidelity simulation model. Its main function is to model the system on a multi-physical and multi-scale level.

Data layer. The data layer is the basis for the operation of the digital twin system. The constituent elements contain welding site sensor data, service data, and knowledge data, such as welding current, welding voltage, electrode pressure, and workpiece geometry information. The data layer is mainly responsible for data representation, classification, storage, maintenance and pre-processing, which drives the fusion of physical entities and twin models to realize their intelligent services.

Communication layer. The communication layer is responsible for aggregating and transmitting all kinds of data collected, and realizing two-way communication between the physical entity layer, model layer and data layer.

Application layer. The application layer contains two modules for real-time data processing and welding process monitoring. First, the data pre-processing module is the technical support for realizing the real-time monitoring of the welding process, and the main function is to monitor and analyze the real-time collected data and extract the data features. Second, the welding process monitoring module is the technical embodiment of resistance spot welding process optimization. Its main function is to realize the dynamic formation process of resistance spot welding nucleus based on twin model.

3. Welding Quality Monitor Of Spot Welding Based On Digital Twin

3.1 System working framework and process

The flowchart of the digital twin-based system proposed in this paper is shown below.

3.2 Data acquisition system for resistance spot welding process

The digital twin system as a real-time dynamic mapping of physical entities, the real-time data collection and update is of vital importance for the digital twin system. The system expresses the welding status through distributed sensors collecting various types of physical quantity information from the system, while using the built transmission network to transmit the welding status information to the twin model in an efficient, real-time and accurate manner to establish data support for realizing real-time monitoring of the resistance spot welding process. As shown in Fig. 3, the data acquisition system for resistance spot welding process includes four layers: physical entity, data transmission, data processing and data sensing.

3.3 Real-time monitoring data processing and storage

The monitoring data involved in the resistance spot welding process is detected and dynamically updated in real time using a data acquisition system, and the data is transmitted to the virtual space through technologies such as the Internet of Things. Real-time data collected by intelligent sensing devices have massive and redundant characteristics and cannot be used directly for welding process monitoring, which requires data pre-processing to propose redundant and abnormal data, using data-model mapping technology to drive a virtual space digital twin model to run simulations.

Resistance spot welding process will release a large number of acoustic, optical and thermal signals, while these signals are susceptible to interference from external conditions, the collected welding electrical signals need to be denoised to reduce the impact of interference. Common denoising methods include low-pass filtering, smoothing filtering and wavelet filtering in the frequency domain, among which wavelet filtering has good time-frequency localization characteristics to remove noise while retaining the detailed information of the original signal.

3.3.1 Theory of wavelet analysis

Since the time-domain sliding window processing of the conventional short-time Fourier transform is equivalent to the filtering of a frequency-domain filter bank, the frequency characteristics of each filter are the same, and the center frequencies are distributed at equal intervals along the analyzed frequency band, the time-domain equal-width method of analyzing non-smooth signals is not applicable. The wavelet transform is a multi-resolution signal analysis method by using a grid to divide the time-frequency surface so that different time-frequency locations have different resolutions.

The theory of wavelet analysis is as follows.

Let \(x(t)\) be a square-integrable function, denoted as\(x(t) \in {L^2}(R)\),\(\psi (t)\) fundamental wavelet function. Then

$$W{T_x}(\alpha ,\tau )=\frac{1}{{\sqrt \alpha }}\int\limits_{{ - \infty }}^{{+\infty }} {x(t)\psi (\frac{{t - \tau }}{\alpha })} dt$$
1

where \(\alpha\) is the scale facto, . \(\tau\) indicates displacement. The above equation is called continuous wavelet transform. Its equivalent frequency domain is expressed as

$$W{T_x}(\alpha ,\tau )=\frac{{\sqrt \alpha }}{{2\pi }}\int\limits_{{ - \infty }}^{{+\infty }} {X(\omega )} \psi (\alpha ,\omega ){e^{+jw\tau }}d\omega$$
2

By performing a denoising linear transformation of a signal using wavelet analysis, the wavelet transform of a signal can be viewed as a wavelet transform of the original signal and the noise.

3.3.2 Wavelet threshold noise cancellation analysis of welding process signals

Since the electrical signals of the resistance spot welding process collected by the welding data acquisition system are interspersed with environmental noise, switching noise and other unavoidable interference, the collected signals need to be processed for noise reduction in order to eliminate the influence of external noise on the later data analysis results.

The wavelet threshold noise cancellation method is formulated as follows: assume that a one-dimensional signal model containing Gaussian noise is expressed in the following form.

$${y_i}={x_i}+e \cdot {z_i}(i=0,1,2,...,n - 1)$$
3

where\({x_i}\) is the true signal, and \({z_i}\) is the standard Gaussian white noise \({z_{i - iid}}N(0,1)\), e is the noise level.\({y_i}\) indicates the signal with noise. Wavelet multi-resolution analysis can perform multi-resolution decomposition of the signal at different scales, decomposing the original signal into components of different frequency bands. In the actual welding process, the useful signal usually behaves as a low-frequency signal, while the noise component has high-frequency characteristics. By processing the noise part of the signal through wavelet thresholding, the original signal \({y_i}\) is recovered from the noise-containing signal \({x_i}\) to achieve the purpose of noise cancellation. The steps of wavelet threshold noise cancellation are as follows.

(1) Perform orthogonal wavelet transform on the noise-containing signal, select the appropriate wavelet and wavelet decomposition level j, and obtain the corresponding wavelet decomposition coefficients.

(2) Thresholding of wavelet coefficients at different scales is shown below.

The hard threshold method is shown below

where y is the original signal, T is the threshold value, and \({T_h}(y,T)\) is the signal after quantization by threshold.

The soft threshold method is shown below.


The soft thresholding approach first makes the elements with absolute values less than the threshold zero, and second shrinks the remaining non-zero elements toward zero. hard threshold is discontinuous at y = ± T, while soft threshold is continuous at y = ± T.

(3) Wavelet reconstruction. The signal is reconstructed by the low-frequency coefficients of the jth layer of wavelet decomposition and the high-frequency coefficients of the 1st to jth layers after quantization to obtain a signal with noise components eliminated.

4. Case Study

4.1 Case background

In order to verify the effectiveness of the proposed digital twin-based resistance spot welding process quality monitoring method, simulations were performed with 2219 and 5A06 aluminum alloy resistance spot welding process data. The validation process of the digital twin-based resistance spot welding process quality monitoring method consists of three aspects. (1) Digital twin system. A physical-virtual welding process system is required to assist the process personnel in welding process monitoring, parameter tuning and evaluation of process instructions. (2) Real-time data pre-processing of the welding process. Sensor devices are used to obtain parameters such as welding current and electrode force in the resistance spot welding process, which are processed for noise reduction to obtain the input data needed for simulation. (3) Digital twin-based monitoring of the resistance spot welding process.

4.2 Welding conditions and materials

The following diagram shows the size of the specimens used in the experiment and the working conditions of the margins. The specimen size is 30*300*7 (mm) and 30*300*2 (mm) respectively.

Two kinds of plates with different thicknesses and strengths were selected for the experiments: 2219 and 5A06, and their chemical composition, mechanical and thermal properties are shown below.

Table 1

Chemical composition of 2219 and 5A06

Material

Si

Fe

Cu

Mg

V

Mn

Zr

Zn

Ti

Ag

Li

Al

2219

0.06

0.17

6.3

0.02

0.1

0.31

0.15

0.02

0.07

-

-

Bal.

5A06

0.06

0.13

0.03

6.4

  

0.6

  

0.02

0.05

-

-

Bal.


4.3 Finite element modeling

Build the finite element model of resistance spot welding based on Simufact.Welding software. As shown in the figure below, it contains 14428 nodes and 25850 nodes, and the minimum size of the welding area is about 8mm. According to ISO 5821 standard, "A0-16-20-100" and "A0-13-18-100" are selected as the upper and lower electrodes for resistance spot welding respectively.

It is assumed that the welding current is uniformly distributed on the upper surface of the upper electrode and is allowed to pass through the contact area of the electrode-workpiece and workpiece-workpiece interfaces, eventually reaching the lower surface of the lower electrode. The bottom of the lower electrode is set to zero voltage. The convective heat transfer coefficients of air to electrode and cooling water to electrode were 19.4 and 3800 \(W{m^{ - 2}}{K^{ - 1}}\). The temperature of cooling water and air was determined to be 20°C.

Set the contact conductivity between the electrodes and 5A06 and 2219, respectively, as shown in the figure below.

4.4 Real-time data collection and processing of the welding process

The welding process data acquisition system is produced by HKS Technology GmbH, Germany, with a sampling rate of 8000 HZ. The welding signal is collected from 0.012s before power on, and the collected signal contains welding current, welding voltage and electrode force. The figure below shows the original curve collected during the resistance spot welding experiment, and the pre-processed curve after noise elimination is obtained by using wavelet threshold noise elimination analysis with the following specific parameters, the wavelet type is DB4, the number of noise elimination is 6, and the threshold value is 50% each time.

The data from the wavelet threshold noise elimination analysis is used as input to the resistance spot weld simulation to implement a finite element simulation based on real-time data acquisition. The finite element simulation welding parameters are shown below.

Results And Discussion

The use of digital twin model for 5A06/2219 spot welding process temperature field and current density changes with time simulation. At the beginning of spot welding, the temperature field after 100ms of welding current is shown in Fig. 10(a), it can be seen from the figure that the temperature of the contact area between the spherical electrode and the aluminum plate is significantly higher than other parts, at this time the contact resistance is the dominant factor affecting the heat generation, and the heat is mainly generated at the contact of the interface.

Spot welding time to 150ms (Fig. 10 (b)), time has been time in the preheating phase of the second heating pulse, at this time with the preheating of the initial temperature field compared to a larger change, when the heat production gradually from the contact resistance to the body resistance dominated. Spot welding time of 200ms when the temperature field as shown in Fig. 10(c), time has entered the welding phase, the highest temperature of the workpiece reached 910K, began to melt, the overall source of heat at this time is the body resistance dominated by the resistance heat effect, making the rapid growth of the molten nucleus. At 400ms, the temperature field as shown in Fig. 10(d), the temperature field distribution trend does not change significantly, the welding process tends to stabilize, at which point the welding process is complete, followed by the spot welding maintenance phase, 1.25s spot welding process ends.

Conclusion

Digital twin technology is a key technology to realize the fusion of physical and virtual models. In this paper, we propose a simulation calculation of the temperature field of resistance spot welding process of heterogeneous materials using digital twin model. A data acquisition system for the resistance spot welding process is established, and the data obtained is processed as an input source for the welding simulation using wavelet threshold noise elimination analysis. A case study of resistance spot welding of 5A06/2219 aluminum alloy is analyzed. The results show that the demonstrated method has good reliability and stability. This allows the synchronization of the numerical simulation based on sensor data and the physical resistance spot welding process, and proposes a feasible physical fusion method for the digital twin modeling of the resistance spot welding process.

Declarations

Funding: This work was supported by the National Natural Science Foundation of China (Nos.52075378 and U21B2079).

Author contribution: All authors contributed to the study conception and design. Conceptualization, Jianwei Dong, Jianming Hu; Experimental test and analysis, Zhen Luo. All authors read and approved the final manuscript.

Institutional Review Board Statement: Not applicable.

Informed Consent Statement: Not applicable.

Data Availability Statement: Data sharing is not applicable to this article.

Conflicts of Interest: The authors declare no conflict of interest.

References

  1. ZHOU K, YAO P. Overview of recent advances of process analysis and quality control in resistance spot welding [J]. Mechanical Systems and Signal Processing, 2019, 124: 170-98. https://doi.org/10.1016/j.ymssp.2019.01.041.
  2. GHATEI-KALASHAMI A, ZHANG S, SHOJAEE M, et al. Failure behavior of resistance spot welded advanced high strength steel: The role of surface condition and initial microstructure [J]. Journal of Materials Processing Technology, 2022, 299: 117370. https://doi.org/10.1016/j.jmatprotec.2021.117370.
  3. FAN Q, XU G, WANG T. The influence of electrode tip radius on dynamic resistance in spot welding [J]. The International Journal of Advanced Manufacturing Technology, 2018, 95(9): 3899-904. https://doi.org/10.1007/s00170-017-1513-5.
  4. PéREZ DE LA PARTE M, ESPINEL HERNáNDEZ A, SáNCHEZ OROZCO M C, et al. Effect of zinc coating on delay nugget formation in dissimilar DP600-AISI304 welded joints obtained by the resistance spot welding process [J]. The International Journal of Advanced Manufacturing Technology, 2022, 120(3): 1877-87. https://doi.org/10.1007/s00170-022-08849-2.
  5. CHOI D-Y, SHARMA A, UHM S-H, et al. Liquid metal embrittlement of resistance spot welded 1180 TRIP steel: effect of electrode force on cracking behavior [J]. Metals and Materials International, 2019, 25(1): 219-28. https://doi.org/10.1007/s12540-018-0180-x.
  6. PANZA L, DE MADDIS M, SPENA P R. Use of electrode displacement signals for electrode degradation assessment in resistance spot welding [J]. Journal of Manufacturing Processes, 2022, 76: 93-105. https://doi.org/10.1016/j.jmapro.2022.01.060.
  7. PANDYA K S, GROLLEAU V, ROTH C C, et al. Fracture response of resistance spot welded dual phase steel sheets: Experiments and modeling [J]. International Journal of Mechanical Sciences, 2020, 187: 105869. https://doi.org/10.1016/j.ijmecsci.2020.105869.
  8. KO W-H, GU J-C, LEE W-J. Energy efficiency improvement of a single-phase ac spot welding machine by using an advanced thyristor switched detuning capacitor bank [J]. IEEE Transactions on Industry Applications, 2018, 54(3): 1958-65. https://doi.org/10.1109/TIA.2018.2796060.
  9. SU Z-W, XIA Y-J, SHEN Y, et al. A novel real-time measurement method for dynamic resistance signal in medium-frequency DC resistance spot welding [J]. Measurement Science and Technology, 2020, 31(5): 055011. https://doi.org/10.1088/1361-6501/ab6673.
  10. JI C, ZHOU Y. Dynamic electrode force and displacement in resistance spot welding of aluminum [J]. Journal of Manufacturing Science and Engineering-Transactions of the Asme, 2004, 126(3): 605-10. https://doi.org/10.1115/1.1765140.
  11. ZENG J, CAO B, TIAN R. Heat generation and transfer in micro resistance spot welding of enameled wire to pad [J]. Journal of Manufacturing Processes, 2022, 82: 113-23. https://doi.org/10.1016/j.jmapro.2022.07.046.
  12. WANG B. A study on spot welding quality judgment based on hidden Markov model [J]. Proceedings of the Institution of Mechanical Engineers, Part E: Journal of Process Mechanical Engineering, 2021, 235(2): 208-18. https://doi.org/10.1177/0954408920953952.
  13. TANG H, HOU W, HU S, et al. Force characteristics of resistance spot welding of steels [J]. WELDING JOURNAL-NEW YORK-, 2000, 79(7): 175-s.
  14. WANG X-J, ZHOU J-H, YAN H-C, et al. Quality monitoring of spot welding with advanced signal processing and data-driven techniques [J]. Transactions of the Institute of Measurement and Control, 2018, 40(7): 2291-302. https://doi.org/10.1177/0142331217700703.
  15. ZHANG H, WANG F, GAO W, et al. Quality assessment for resistance spot welding based on binary image of electrode displacement signal and probabilistic neural network [J]. Science and Technology of Welding and Joining, 2014, 19(3): 242-9. https://doi.org/10.1179/1362171813Y.0000000187.
  16. DAI W, LI D, TANG D, et al. Deep learning approach for defective spot welds classification using small and class-imbalanced datasets [J]. Neurocomputing, 2022, 477: 46-60. https://doi.org/10.1016/j.neucom.2022.01.004.
  17. LIU M, FANG S, DONG H, et al. Review of digital twin about concepts, technologies, and industrial applications [J]. Journal of Manufacturing Systems, 2020. https://doi.org/10.1016/j.jmsy.2020.06.017.
  18. TAO F, QI Q. Make more digital twins [J]. Nature, 2019, 573(7775): 490-1. https://doi.org/10.1038/d41586-019-02849-1.