Visual monitoring and acoustic signal detection play crucial roles in quality control during laser welding processes

Visual monitoring and acoustic signal detection play crucial roles in quality control during laser welding processes. Visual monitoring utilizes high-speed cameras and image processing techniques to capture real-time images of the molten pool morphology, weld seam quality, and potential defects such as cracks or porosities. This information is critical for promptly adjusting welding parameters or issuing early warnings about welding quality issues.

1. Visual Monitoring of Laser Welding

Visual monitoring of the laser welding process can be categorized based on the acquisition angle of imaging light signals into off-axis and coaxial types. Off-axis monitoring involves capturing signals reflecting the welding process from an angle inclined relative to the laser beam, typically from above or from the side of the welding pool. Coaxial monitoring, on the other hand, captures imaging signals from directly above the welding pool and keyhole along the same axis as the laser beam.

Depending on the presence of illumination sources, visual sensors in laser welding processes can further be classified into active and passive types. Active monitoring utilizes auxiliary illumination sources to provide off-axis or coaxial lighting on the welding pool and keyhole. Passive monitoring, however, uses the radiation light from the plasma or the radiative light from the molten metal in the welding pool as the imaging light signal.

These monitoring techniques enable real-time observation and analysis of the welding process, facilitating adjustments to welding parameters and early detection of welding defects such as cracks or porosities.

2. Acoustic Signals

In laser welding, especially in deep penetration laser welding processes, the laser penetrates into the material through a small keyhole for heating. Due to the small size of the keyhole (with a diameter of a few millimeters) and the strong plasma or metal vapor above and inside it, direct observation of the interior of the keyhole and the weld pool is challenging. Therefore, current real-time diagnostic methods for laser welding quality mainly rely on collecting various indirect signals such as sound, light, and electrical signals that reflect the welding process. Information extracted from these signals enables monitoring and diagnostics of laser welding.

In laser welding under keyhole mode, a specific audible sound in a certain frequency range is emitted. When deep penetration laser welding cannot proceed due to certain reasons or during conduction welding, this characteristic audible signal decreases or disappears. It is generally believed that this sound signal is caused by pressure waves generated when metal vapor or plasma jets out from the keyhole. This signal is closely related to the behavior of the plasma, keyhole, and weld pool, reflecting changes in the welding process.

Currently, numerous studies have utilized the detection of audible sound signals to monitor the laser welding process and its quality, yielding beneficial results. However, there is a lack of deeper exploration into the nature of the welding process reflected by these sound signals. For example, a narrow-band peak signal appearing at 4.5 kHz indicates good welding quality, while the absence of this peak and overall low spectral intensity indicate poor welding quality. Using a microphone to measure sound signals during the welding process, researchers have conducted fast Fourier transform (FFT) analysis of the signal, studying the distribution characteristics of the spectrum. It has been found that the sum of all frequency components in the spectrum correlates with laser energy, welding speed, and focal offset.

The drawback of using audible sound as a detection parameter is its susceptibility to nozzle airflow and environmental noise interference. However, its advantages lie in easy signal acquisition without burdening or adversely affecting the welding head and optical path. Additionally, it is not sensitive to sensor mounting direction or distance.

Microphones are used to capture acoustic signals. Spectral subtraction is employed to reduce noise in the sound signal, and the Welch-Bartlett power spectral density estimation method is used to analyze the frequency characteristics of the sound signal. The results show that well-fused joints (FP) can be clearly distinguished from partially fused joints (PP) based on their sound pressure differences in the time domain. Algorithms based on different frequency characteristics between FP and PP from 500 to 1500 Hz have been developed to identify penetration states. These algorithms effectively differentiate between FP and PP. Furthermore, the mechanisms behind different sound signal characteristics produced by various penetration depths and modes are analyzed and discussed. This indicates that utilizing sound signals for online monitoring of penetration status during laser welding of high-strength steel DP980 in noisy environments is feasible with appropriate digital signal processing methods.