Application Cases

[Overview]

In this study, an Aigtek ATA-5000 series preamplifier (low-noise signal amplifier) was used to build a welding experiment platform. A neural network model was developed to achieve high-precision, intelligent online monitoring of welding quality, meeting industrial application requirements.

Experiment Name: Acoustic Emission Monitoring Experiment for Femtosecond Laser Glass–Copper Micro-Welding

Research Directions: Femtosecond laser; glass–copper dissimilar-material micro-welding; acoustic emission monitoring; intelligent monitoring; convolutional neural network; process optimization; non-destructive testing.

Experiment Objective: This experiment aims to determine the optimal process parameters for femtosecond laser glass–copper micro-welding and identify the dominant forming body of the welded connection layer. It verifies the feasibility of using acoustic emission signals for monitoring the welding process, designs effective features to distinguish welding success from failure, and builds a neural network model to realize high-precision, intelligent online monitoring of welding quality for industrial applications.

Test Equipment:

1. FemtoYL40 all-fiber femtosecond laser

2. High-precision motion platform

3. FUJICERA 1045S acoustic emission sensor

4. Preamplifier

5. NI PXIe 5105 data acquisition card

6. KEYENCE VHX-7000 ultra-depth-of-field 3D microscopy system

7. Hitachi S3400N scanning electron microscope (SEM)

8. EDAX energy dispersive spectrometer (EDS)

9. LabVIEW acquisition system

Figure 1. Schematic of the experimental platform

Experiment Procedure: The area method was used to measure the femtosecond laser ablation thresholds of quartz glass and copper. Welding experiments were carried out by varying laser power, scanning speed, and defocus amount. Shear strength was used as the optimization metric, and repeated welding and gradient-gap experiments were performed to determine the connection主体. An acoustic emission monitoring system was built to collect signals during welding. After filtering, FFT, and wavelet transform processing, traditional features were compared and a "relative root mean square sum" feature was designed. One-dimensional acoustic signals were converted into two-dimensional time–frequency maps, and a VGG-like convolutional neural network was constructed. The model was optimized by enlarging the receptive field, fusing multi-dimensional data, and using a customized physics-informed loss function, then completed training, validation, and testing.

Experiment Results:

1. The measured femtosecond laser ablation thresholds of copper and quartz glass were 0.1620 J/cm² and 1.1692 J/cm², respectively. The optimal process parameters for glass–copper micro-welding were determined, confirming that the welded connection layer is dominated by molten glass.

2. Acoustic emission monitoring effectively characterizes the welding process, and the "relative root mean square sum" feature accurately judges welding quality.

3. The optimized neural network model achieved 95% accuracy, with precision and recall of 0.923 and 0.988 respectively, satisfying the requirements for industrial online monitoring.

Product Advantages of Aigtek Amplifiers in This Application:

1. Gain amplification — Increases signal amplitude.

2. Suppresses environmental noise — Improves signal-to-noise ratio (SNR).

3. Stably and completely captures valid signals — Ensures monitoring accuracy.

The experimental materials in this article were compiled and published by Xi'an Aigtek Electronics. Aigtek has become an instrumentation supplier with a broad product portfolio and considerable scale in the industry; demonstration units are available with free trial support. Xi'an Aigtek Electronics is a high-tech enterprise specializing in the R&D, manufacturing, and sales of electronic test instruments including power amplifiers, high-voltage amplifiers, power signal sources, preamplifiers (low-noise signal amplifiers), high-precision voltage sources, and high-precision current sources.