MONITORING AND FORECAST OF RESIN INFUSION PROCESSES WITHOUT CONTACT TO PART

Kurzfassung

Liquid composite moulding (LCM) processes are very sensitive to deviations in raw materials and process parameters. Small deviations can lead to different fill patterns with the risk of incomplete impregnation. Providing information about ongoing processes is therefore subject of major effort in research and industry. Ultrasonic (US) sensors are particularly well suited for process monitoring of LCM, as they are able to measure all of the most crucial parameters of the filling and also cure state. Furthermore they do not require direct contact to the part, thus the part surface and the mould vacuum integrity are not affected. The DLR has developed small and low cost ultrasound sensors allowing the integration of a dense sensor network. These sensors are relatively easy to integrate into the mould and provide a large spectrum of information in comparison to conventional transducers. The sensors are applied on the outer side of the mould, from where the sound waves propagate through the wall and then interact with the part. When the sound waves reach the mould-part-interface, most of the signal is reflected back while the remaining sound waves are transmitted into the part. Prior to flow front arrival the sound signal gets reflected completely. When the interface gets wetted the reflection amplitude starts to decrease until wetting is completed. Also the through-transmission signal can be used for resin arrival detection. Both methods work very robust with almost every mould material. After calibration the velocity of the flow front and its direction can be obtained through enhanced signal analysis. This information, which is provided by every single sensor, is very valuable for reconstructing the flow pattern. The method was evaluated in a transparent mould, where the impregnation progress was documented by video. From the video the flow front position, speed and direction in reference to the sensors was later analysed by image processing. The comparison of the sensor and video signal shows strong correlation. The flow front speed can be measured with at least 3.5 % accuracy. The precision of the angle measurement depends largely on the mould thickness, where at small thicknesses the precision is best. This is due to the fact that the non-rotational-symmetric character of the sound pressure distribution decreases with increasing mould thickness. Unfortunately only specialists with a deep understanding of LCM processes are able to rate a running process on basis of the raw data. Therefore a monitoring and forecast system is under development at DLR. The procedure is as follows: First training data is generated by variation of simulation parameter which is then used for supervised machine learning to infer a function between sensor data and process parameters. During manufacturing data is gathered online with US-Sensors which are feed to the inferred function in order to calculate process parameter and material properties. With these identified parameters the current degree of filling and a forecast of further process evolution are simulated.