The dependable guarantee of very high seam quality requirements in laser welding of demanding material combinations and highly stressed structures, such as powertrain components, is becoming increasingly important. The combination of sensor-based inline process monitoring and real-time data analysis using machine learning shows enormous potential for ensuring this. The subject of this paper is the assessment of process monitoring based on acoustic and optical sensor data by means of machine learning during laser welding on rotationally symmetric test specimens. The results show that typical welding defects caused by process variations can be detected with an accuracy of approx. 96 %, almost in real-time. Furthermore, approaches for predictive maintenance of system components and predictive modeling of component properties, supported by numerical simulations, are presented.