Analysis of Data Streams for Qualification and Certification of Inconel 738LC Airfoils Processed through Electron Beam Melting

Abstract

Qualification and certification of components fabricated through additive manufacturing (AM) are daunting tasks due to the large numbers of process variables and defect structures that can form during the AM process. However, unlike traditional manufacturing processes, AM offers the unprecedented opportunity to observe the interior of parts in real time as they are being grown, in addition to the capacity to record the entire manufacturing process as a function of time and space. As a result, a digital twin can be generated for each part given the capability to interpret and analyze available data streams into meaningful results. However, transforming these data streams into meaningful results is currently the crux of the AM certification challenge. Oak Ridge National Laboratory (ORNL) has completed a program to develop the necessary processing science to enable the fabrication of prototype airfoils for industrial gas turbine engines from the difficult-to-AM-process nickel-base (Ni-base) superalloy Inconel 738LC for hot-fire engine trial evaluation. While scaling the electron beam melting (EBM) process to allow serial production of airfoils in sufficient quantities for blading a disk is straightforward, effectively capturing the subset population of defect-bearing airfoils before entering service presents concerns not only for the structural integrity of the airfoil but also the gas turbine engine. Currently, industry accepted protocols for screening critical AM rotating components do not currently exist. The process and protocols established for screening and identifying defective airfoils within the context of this program are discussed. This includes the development and training of a machine learning algorithm for identifying defects (such as porosity, lack of fusion, and cracking) coupled with high-throughput computed tomography for validating the results of the machine learning algorithm and data analytics used to identify build-to-build and machine-to-machine variability throughout the build program.