RTM-flow front detection for GFRP components with embedded MEMS-acceleration sensors
Topic(s) : Special Sessions
Co-authors :
Markus MÜNCH (GERMANY), Pritish BODKE (GERMANY), Gernot REPPHUN (GERMANY), Peter MIDDENDORFAbstract :
Resin Transfer Moulding (RTM) is a manufacturing method for fiber-reinforced polymer components with serial production capability. Thereby, various parameters influence the resin injection and thus component quality. Hence, knowledge about resin flow can help to obtain information on the quality of produced components. However, RTM-moulds are often opaque such that resin flow is not observable without the use of additional sensors. Conventional techniques use pressure or capacitive sensors, for example, integrated into the mould to track filling and curing status. In contrast, we use acceleration sensor networks embedded in preforms with the overall goal to determine porosity in components. These sensors can afterwards be used for further monitoring functions. As a first step towards this, we investigate a corresponding offline resin flow front detection system.
The system applies a changepoint detection (CPD) algorithm to recorded sensor data. The sensor networks consist of micro electro-mechanical systems BMA456 sensors mounted on flexible circuit boards. These boards are woven into textiles of glass fiber preforms. The data processing is developed from a production of plate components. Considering the overall goal, RTM-processes with and without deliberately introduced porosity are performed. All resin infiltrations are recorded with the embedded sensors. Visual inspection of the acceleration curves leads to characteristic “V”-shapes being expected to result from an arriving flow front. Timestamps are manually assigned to the “V”-shapes for calibration and testing purposes of an offline CPD-algorithm. Each subset includes data from both kinds of plates. Thus, the system is expected to be robust to manufacturing deviations regarding variation in porosity. To further check the system’s plausibility, a simulation of infiltrations without artificial porosity is performed and compared to the outputs on the test set.
Moreover, the transferability of the system to other RTM-processes is investigated. Therefore, components of higher geometrical complexity and with fewer embedded sensors are produced. Again, visual inspection shows similar “V”-shapes in the acceleration data. Hence, transferability to further RTM-processes beyond the present two can be expected.
Here, a new flow front detection system for RTM-processes is investigated. By introducing a sliding window approach, the present offline flow front detection can be transformed to an online detection. In comparison to commonly used pressure or capacitive sensors, the BMA456 sensors have a high measurement resolution and frequency. Therefore, the present method can be developed into an online infiltration monitoring system. Incorporating more measured data after the “V”-shapes and sophisticated feature engineering, porosity prediction seems feasible. From serial production perspective, this leads to quality controls being performed during production itself instead of additional control stages.
The system applies a changepoint detection (CPD) algorithm to recorded sensor data. The sensor networks consist of micro electro-mechanical systems BMA456 sensors mounted on flexible circuit boards. These boards are woven into textiles of glass fiber preforms. The data processing is developed from a production of plate components. Considering the overall goal, RTM-processes with and without deliberately introduced porosity are performed. All resin infiltrations are recorded with the embedded sensors. Visual inspection of the acceleration curves leads to characteristic “V”-shapes being expected to result from an arriving flow front. Timestamps are manually assigned to the “V”-shapes for calibration and testing purposes of an offline CPD-algorithm. Each subset includes data from both kinds of plates. Thus, the system is expected to be robust to manufacturing deviations regarding variation in porosity. To further check the system’s plausibility, a simulation of infiltrations without artificial porosity is performed and compared to the outputs on the test set.
Moreover, the transferability of the system to other RTM-processes is investigated. Therefore, components of higher geometrical complexity and with fewer embedded sensors are produced. Again, visual inspection shows similar “V”-shapes in the acceleration data. Hence, transferability to further RTM-processes beyond the present two can be expected.
Here, a new flow front detection system for RTM-processes is investigated. By introducing a sliding window approach, the present offline flow front detection can be transformed to an online detection. In comparison to commonly used pressure or capacitive sensors, the BMA456 sensors have a high measurement resolution and frequency. Therefore, the present method can be developed into an online infiltration monitoring system. Incorporating more measured data after the “V”-shapes and sophisticated feature engineering, porosity prediction seems feasible. From serial production perspective, this leads to quality controls being performed during production itself instead of additional control stages.