Artificial intelligence and machine learning algorithms to detect defects in additive manufacturing by fusing multiple sensor data

While additive manufacturing (AM) has enabled the fabrication of complex geometries, process repeatability, and part quality has been an inhibitor to widespread industry adoption. Parts fabricated using metal AM can have their mechanical properties compromised due to the presence of defects. Presently, quality assurance is achieved by X-ray Computed Tomography (CT), which is costly and time-consuming. In-situ monitoring in AM can help in the democratization of AM by offering real-time detection of anomalies. This research aims to develop artificial intelligence and machine learning (AI/ML) algorithms that use the data generated by the machine sensors in conjunction with inexpensive optical and acoustic sensors to detect defects in Laser Powder Bed Fusion (LPBF) in real-time.   The popular EOS-M290 LPBF machine has 20+ in-built sensors such as O2 concentration, chamber temperature, layer time, recoater speed, etc. This machine data in conjunction with optical and acoustic data will be used to develop AI/ML models for real-time issue detection. A high-resolution optical camera will be mounted on the EOS M290 LPBF machine to collect layer-by-layer images. These images will be analyzed by an artificial intelligence (AI) model trained to detect layer anomalies such as spatter, streaking, hopping, part swelling, etc. An acoustic sensor will be deployed to capture airborne sound waves and analyzed to identify features denoting the evolution and presence of defects. AI/ML algorithms will combine or fuse the data from the different sensors to provide higher accuracy of detection. This project will use the underutilized machine data to achieve high special and temporal information of defects. This project will further build on the current capability of Addiguru’s optical camera-based layerwise real-time monitoring technology which detects issues using computer vision and AI.   Data will be generated by printing samples that will be deliberately designed to include various defects. This data will be analyzed and used for training AI/ML models. The training of the AI/ML models requires labeling of data with the ground truth which will be obtained by performing CT scans. CT scans will determine porosity in the parts. Metallographic analysis will be carried out to uncover issues and train the models.   Combining in-situ machine sensor data along with optical and acoustics signals will provide a significant enhancement in the detection of anomalies. The data fusion will be carried out by developing advanced AI/ML algorithms by Addiguru’s AI/ML and software engineering experts. It is proposed that high spatial and temporal resolution obtained by fusing multiple sensor data will enable precise location of defects and thereby, significantly reduce post-build inspection costs. Successful completion of the project will demonstrate the feasibility to develop AI/ML algorithms for multiple sensory data for real-time issue detection in the LPBF process.