AI/ML for Additive Manufacturing Defect Detection

OUSD (R&E) MODERNIZATION PRIORITY: Artificial Intelligence (AI)/Machine Learning (ML) TECHNOLOGY AREA(S): Materials / Processes OBJECTIVE: Develop Artificial Intelligence/Machine Learning (AI/ML) based software tools to help identify additive manufacturing (AM) defects from in-situ sensor-based data. Capture sufficient process control and monitoring data in real-time to later on, through AI/ML analysis, help improve the reliability, speed, and cost of post processing inspections by knowing where and what to look for ahead of time. DESCRIPTION: There is continued advancement in the use of in-situ sensing in metal AM processes. This includes the use of in-situ sensor data to help develop stable AM process windows and more recently the use of sensors to help control the AM process through feed forward control or other real-time adaptive control methodologies. Advanced sensing capabilities for metal AM includes cameras and sensor arrays with increased temporal and spatial resolution, and cameras with adaptable fields of view and broader thermal sensing range. Advances are taking place not just in the specification of the sensor arrays used, but also on the types of sensing modalities incorporated into the AM process chamber. Aside from the more traditional infrared (IR) and visual infrared (VIS) cameras mentioned previously, other sensor types include optical emission spectrometers, acoustic and vibration spectral sensors, laser profilometers, and others. Additionally, sensors within the AM system may include power monitoring, galvo locations, oxygen monitoring, etc. Despite all the progress achieved in process monitoring and control to improve the quality of metal AM parts, very little progress has been accomplished in intelligently fusing all the data collected during the AM process to help reduce the cost and increase the reliability of post-fabrication nondestructive evaluation (NDE) techniques. In particular, X-Ray Tomography remains the gold standard for AM part inspections, though it can be costly and ill-suited for large components. This SBIR topic explores the use of AI/ML tools to help identify the location and type of potential defects (with statistical margins of error and confidence intervals). Even though the objective of the topic is to use existing process monitoring and control data to develop AI/ML algorithms, the Navy is open to new and creative hardware enhancements that can improve the reliability of AI/ML predictions. Enhancements such as replacing a sensor by an array of sensors, adding a new sensing modality or advanced data processing hardware card. PHASE I: Define, design, and develop the AI/ML methodology for defect type identification and localization (with statistical bounds). Identify the metal powder bed fusion system that the proposer plans to upgrade with AI/ML tools. Provide a list of all the sensors and control parameters (including ones already available in the system and additional ones) to fuse via the AI/ML framework. This will include the rationale for the selections \. Indicate if there will be modification(s) or addition(s) of new sensing modalities/other hardware for added defect identification reliability. As part of the Phase I AI/ML algorithm development effort, simple sample coupons with embedded defects (e.g., porosity, hot cracking, keyholing, etc.) should be fabricated. Define the ground truth methodology to be used (i.e., coupon sectioning, x-ray tomography) for AI/ML training purposes. Provide a Phase II plan. PHASE II: Focus on increased validation of AI/ML tools with aggregated large data sets from multiple sensors. This may also include aspects of transfer learning. Validation and comparison to NDE/I techniques will also be emphasized for Phase II. Phase II will also focus on key performance property impacts based on defect population. PHASE III DUAL USE APPLICATIONS: Validate AI/ML tools for a different metal alloy to test AI/ML tools. Engagement with an OEM is highly encouraged. Commercial applications of additive manufacturing can be found in a wide range of commercial sectors such as: aerospace, shipping, transportation, rail, automotive, medical, etc. This technology would be applicable to identifying defects in critical metallic applications across all the sectors. REFERENCES: Petrich, J.; Snow, Z.; Corbin, D. and Reutzel, E.W. Multi-modal sensor fusion with machine learning for data-driven process monitoring for additive manufacturing. - Additive Manufacturing, Volume 48, Part B, December 2021, 102364. https://www.sciencedirect.com/science/article/abs/pii/S2214860421005182 Qi, X.; Chen, G.; Li, Y.; Cheng, X. and Li, C. Applying neural-network-based machine learning to additive manufacturing: current applications, challenges, and future perspectives. Engineering, Volume 5, Issue 4, August 2019, pp/ 721-729. https://www.sciencedirect.com/science/article/pii/S2095809918307732 Westphal, Erick and Seitz, Hermann. A machine learning method for defect detection and visualization in selective laser sintering based on convolutional neural networks. Additive Manufacturing, Volume 41, May 2021, 101965. https://www.sciencedirect.com/science/article/pii/S2214860421001305 KEYWORDS: additive manufacturing; AM; artificial intelligence/machine learning; AI/ML; nondestructive evaluation; defects; discontinuities