SUMMARY

Sensors play an important role in smart manufacturing. Different types of sensors have been used in process monitoring to ensure the quality of products. As a result, the life-cycle cost of quality control is rising. The reliability of sensors also affects the reliability of complex systems with a large number of sensors onboard. Another challenge is the available bandwidth in communication channels for transmission of large volumes of data. The original purpose of data cannot be fulfilled if they are not shared and used. In this research, a new approach that uses low-fidelity measurements with limited sensors to provide high-fidelity information in additive manufacturing (AM) process monitoring is investigated. A physics based compressive sensing (PBCS) approach is proposed to reduce the number of sensors and amount of data collection for AM process monitoring. PBCS significantly improves the compression ratio from traditional compressed sensing by incorporating the knowledge of physical phenomena in specific applications. PBCS has been demonstrated to monitor the temperature distribution in fused filament fabrication process and the thermofluid field in selective laser melting for metal AM. PBCS can recover and reconstruct the complete temperature and velocity fields in three dimensions based on limited measurements. The sensing performance is further improved with a physics-constrained dictionary learning approach by optimizing the placement of low-fidelity measurements to obtain high-fidelity information. The dictionary learning approach has been demonstrated with one-dimensional signals where sampling time stamps are optimized. In two-dimensional images, the optimal locations of pixels to sample are determined. When monitoring the surface temperature of the builds in AM processes with infrared thermal imaging systems, low-resolution pixel values at the designed locations can be used to recover high-resolution images. Based on the recovered images, more accurate three-dimensional temperature distributions can be reconstructed with PBCS. The proposed PBCS scheme provides a systematic and rigorous approach to design efficient sensing protocols for future manufacturing systems, where sensors are ubiquitously utilized in monitoring process and quality.