Acoustic process monitoring during projection welding using airborne sound analysis and machine learning

Abstract Resistance projection welding is predominantly performed using capacitor discharge machines, known for their short welding times, rapid current rise times, and high currents compared to medium-frequency inverter technology. The resulting joints are covered up during resistance welding, so that either destructive or non-destructive testing is required to evaluate the quality. Process monitoring is therefore essential in resistance projection welding. The requirement for this is process data that can be acquired and integrated into the process monitoring easily, cost-effectively, and contactlessly. This study investigates the use of low-cost condenser microphones to utilize the airborne sound generated during welding for process monitoring. It is shown that, acoustic data processed by the fast Fourier transform can be used to evaluate the quality of the connection. Only a minor influence of the microphone position could be determined. A machine learning model was also used to detect the batch of the welding nut. The machine parameters, welding nut geometry and material were kept constant. The results show a batch prediction of more than 90% using airborne sound.