AI closes the loop for composites manufacturing

AI closes the loop for composites manufacturing

Researchers at CFMS and NCC use machine learning to adjust process parameters in automated composites manufacturing, removing the need for human intervention and improving quality.
Composite manufacturing, although state of the art in terms of product performance, remains something of a craft in terms of manufacturing. Even when processes are physically automated, highly skilled engineers are still required to adjust process parameters, correcting for variation in the process and materials. This human intervention increases product variation and cost. At the same time, ever increasing volumes of data are now available through embedded sensors. The composites manufacturing industry is very interested in how this data can be used to automatically make process adjustments in closed loop manufacturing (CLM).

"Many composites manufacturing processes require the intervention of experienced engineers to overcome process and material variability. In this demonstration, the in-process decision-making is being automated, using digital process modelling and advanced machine learning techniques. The National Composites Centre (NCC) and Centre for Modelling & Simulation (CFMS) are helping to bring the composites industry one step closer to intelligently automated production," says NCC Chief Engineer Giuseppe Dell'Anno.

In a collaborative project between the CFMS and the NCC, both in Bristol, England, a closed loop manufacturing process has been demonstrated for the resin transfer molding (RTM) of high-quality composite components. The process involves the following steps:
 
  1. Measure the process parameters during production
  2. Measure the critical-to-quality outputs on the finished product
  3. Use machine learning to model the relationships between the inputs and the outputs
  4. Use the model to set process parameters at optimum levels for quality

RTM involves placing a dry fiber preform within a closed mold. A liquid thermoset resin is then injected at high pressure and allowed to infuse into the fibers. Critical-to-quality outputs are porosity and dry-spot defects, where resin has failed to fully infuse into the fiber preform. During the process sensors monitor the flow of resin, enabling a real-time understanding of the process and improving simulation accuracy. The flow of resin within the mold is controlled by adjusting the flow of inlet and outlet valves at multiple locations.

An issue with machine learning, or artificial intelligence (AI), is that it generally requires very large training data sets to identify patterns reliably. These data sets are often much larger than a human would require to spot similar patterns. The advantage of AI is that, when enough data is available, extremely complex patterns can be understood, which may be beyond the capabilities of a human.

In this project, the computer was trained to recognize the relationships between the valve positions at different times in the RTM process and the flow of resin. A test part was used which was 40 centimeters (cm) x 18 cm x 90 cm. It was designed to include key features of a full-scale aircraft wing, demonstrating that the experiment is scalable, according to CFMS AI Domain Specialist Kiran Krishnamurthy. These features included shallow ramps, a steep ramp close to edge of part, and a fillet and a flange next to a ramp. Training the algorithm to dynamically set valve positions for this part was estimated to require the manufacture of 15,000 components. Simulations, rather than actual component moldings, were used to obtain the initial training data.

“Through the creation of a series of collaborative proof of concept demonstrators, we want to highlight the potential application of AI technology, to reduce time-to-market of novel design and manufacturing processes,” says CFMS Chief Operating Officer, Sam Paice.

The NCC and CFMS are running an event titled Lean Composites Manufacturing through Machine Learning on July 5, 2018 at CFMS. This event will demonstrate how combining machine-learning techniques with virtual manufacturing simulation enables an accelerated manufacturing process learning curve for shorter time to market, while also providing insights into improving process and product quality.

 

Learn more



  Continue reading »
X