Institute for Technical Acoustics · RWTH Aachen · R&D program · Supervision · Research strategy

Five-year ultrasonic welding research program

A research program built around an uncertain technical question, with parallel workstreams that continuously informed—and sometimes stopped—one another.

RoleResearch Scientist / Research Lead
Duration5 years
Portfolio≈20 connected projects
Supervision≈15 student projects
Technical evidence · AI & ML

Statistical inference and supervised learning

How two processing pipelines built from the same ultrasonic welding measurements supported quality classification and strength regression.

Explore the AI/ML methods →
Program goal

Improve ultrasonic metal welding when the route to improvement was not yet known.

The five-year research program was broader than one thesis question. The job was to learn what mattered in the process, build the measurement and modeling capability to observe it, and repeatedly redirect effort as evidence accumulated.

5 yearsmulti-disciplinary research direction
≈20connected projects coordinated
15student projects within the program
2–5projects typically active in parallel, sometimes up to 10
Workstreams

Several technical routes into the same process.

Structural & acoustic simulationUnderstand machine vibration and sound radiation.
Thermal / air effectsTest whether temperature gradients around the weld could refract radiated sound and distort measurements.
Measurement developmentAcquire reliable process data and build new measurement methods where needed.
Signal processing & statisticsExtract repeatable information from measured vibration signals.
Supervised learningEstimate welding stages and weld strength from process-informed features.
System identificationInfer the state of the weld and mechanical interfaces from frequency-domain behavior.
How the program moved

Results from one project changed the next project.

The program was not a collection of independent studies. The valuable work was in the dependency chain.

Vibration simulations
Identify meaningful measurement locations
→
Measurement problem
Laser path disturbed by welding particles
→
New apparatus
3D-printed suction system enables measurements near the horn tip
→
Better data
Repeatable vibration signatures reveal welding stages
→
New analyses
Stage-specific statistics, supervised models and system identification

One breakthrough was recognizing the welding stages in the vibration measurements. That gave the data a physical structure: instead of treating the signal as one undifferentiated time series, each stage became a meaningful state that could support feature design and model assumptions.

Decision-making under uncertainty

Stop a weak path when the evidence says to stop.

A separate line of work investigated whether airborne sound could provide a simpler non-contact monitoring signal. Numerical sound-radiation studies showed a complex field strongly shaped by machine geometry and reflections. Experimental measurements could not be linked reliably to the welding process.

Program pivot

The microphone-based monitoring route was deprioritized, and effort shifted toward structural vibration measured with laser vibrometers, where the signal-to-process relationship was stronger.

Program outcomes

What the combined work established.

  • Airborne sound was not a reliable route for monitoring the process in the investigated setup.
  • Horn and anvil vibrations could be measured during welding using a purpose-built measurement approach.
  • Anvil vibrations in particular could identify welding stages and support a stop decision at the maximum-strength stage.
  • Process-informed vibration features could also estimate weld strength, providing a second layer of control.
  • The research program created a connected technical basis spanning measurement, signal processing, simulation, statistics and mechanical modeling.