Supervised deep learning for real-time quality monitoring of laser welding with X-ray radiographic guidance

Abstract

Laser welding is a key technology for many industrial applications. However, its online quality monitoring is an open issue due to the highly complex nature of the process. this work aims at enriching existing approaches in this field. We propose a method for real-time detection of process instabilities that can lead to defects. Hard X-ray radiography is used for the ground truth observations of the sub-surface events that are critical for the quality. A deep artificial neural network is applied to reveal the unique signatures of those events in wavelet spectrograms from the laser back-reflection and acoustic emission signals. The autonomous classification of the revealed signatures is tested on reallife data, while the real-time performance is reached by means of parallel computing. The confidence of the quality classification ranges between 71% and 99%, with a temporal resolution down to 2 ms and a computation time per classification task as low as 2 ms. This approach is a new paradigm in the digitization of industrial processes and can be exploited to provide feedbacks in a closed-loop quality control system. The introduction of laser technology in metal welding of metals is dated back to the late 1960s 1,2 when it immediately showed advantages as compared to traditional arc welding 3. The attractions of this technique are in the non-contact processing, the absence of tool wear, high aspect ratio of the melt pool, better material fusion, possibility to process refractory materials, low running costs and high processing speed 3,4. Today, laser welding is a key technology in many fields e.g. automotive 5 and aerospace 3,6 industries, naval and heavy machinery production 7 , medicine and micromechanics 3. Unfortunately, the potential of this technology is not fully exploited, particularly in applications that require the guarantee of high weld quality. The reason is the non-linear nature of light-matter interactions, which complicates the reproducibility of the weld quality in mass production 8-10. The complex dynamics of the process, especially in keyhole welding regime, and its instabilities can cause various defects at the joint 3,10-12. A defect type of particular interest is porosity, which is a hidden threat for the mechanical properties of the workpieces 3,9-11. Obviously, an adequate, robust and low cost quality monitoring system is of great desire. The major challenge in developing such technique is in the difficulties to inspect directly the sub-surface behavior of the process zone in real-life conditions 13. Multiple approaches have been proposed, which are mostly based on mathematical modeling aiming to reconstruct the under surface dynamics using inspections of the surface via measurements of temperature 11,12,14,15 , optical 16,17 and/or acoustic 18,19 emissions (AE). However, those approaches face three main problems. Firstly, modeling often suffers inaccuracies originating from the deviations of the model assumptions from the real parameters' values. More complicated assumptions can be used to imp...

Machine learning monitoring for laser osteotomy

“…WPT spectrograms, formed as the relative energies of the frequency bands, provide simplified time‐frequency representations of the original signals (ie including a lower number of elements and, at the same time, conserving the features of the signal evolution). In the works of Shevchik et al [15, 17, 18], it was shown that such representation of the input data allows reducing the training sets, still providing a high classification accuracy. In the present contribution, these previously reported results were proved, while feeding the raw signals to both algorithms from the training set and showing lower classification accuracy.…”

“…Laser‐induced AE is a well‐known phenomenon [7, 13] and the modulations of AE are strongly dependent from the chemical composition, mechanical, and optical properties of the tissues that are exposed to laser irradiation [7, 13]. The same approach has been well established in industrial applications such as laser welding and additive manufacturing [15–19]. Based on these experiences, we believe that this approach will potentially provide information about the ablated zone in biomedical applications as well.…”

Adaptive Laser Welding Control: A Reinforcement Learning Approach

“…For this reason, the full setup depicted in Fig. 2 This unit is based on our previous work [13], where the AE and OE signals from the PZ were used to identify quality critic momentary events. In this contribution, the output of the classifier is made up of labels that correspond to predefined welding qualities in terms of penetration depth and pore content.…”

“…Consequently, it is difficult to provide an effective feedback from the process to the control system, since it requires the correlation of the surface measurements with the sub-surface events (e.g., pore formation), which is not a trivial task [12]. Nevertheless, some pilot works in LW monitoring report successes in identifying quality critic momentary events from the corresponding AE and OE signals from the processed zone [13], [14]. The present study starts from the aforementioned preliminary results of process monitoring and focuses on the use of Reinforcement Learning (RL) towards keyhole LW closedloop control.…”

Multi-Output Monitoring of High-Speed Laser Welding State Based on Deep Learning

“…Bacioiu et al [ 18 , 19 ] used CNN models to detect defects in tungsten inert gas welding. Shevchik et al [ 20 ] proposed a method for real-time detection of laser welding instabilities by the application of CNN.…”