1 8 A D V A N C E D M A T E R I A L S & P R O C E S S E S | A P R I L 2 0 2 3 • In situ process sensors and feed-forward, adaptive process controls • Standards (process, control, sensor, data fusion, and informatics) • Testing (statistically substantiated data) While there appears to be community convergence on the tools to be used, there are countless approaches to their unified integration being explored. The application of these tools to accelerate AM mechanical property allowables requires a paradigm shift in our thinking. The conventional approach to mechanical property allowables development is to a) freeze the materials process, b) produce multiple lots and heats, and c) conduct extensive mechanical property tests to generate statistically substantiated properties[6]. ML learning permits the concomitant assessment of AM key process parameters on a range of mechanical properties of interest. It thus provides both a means of converging on a set of optimal process parameters and establishing the robustness (sensitivity of properties to changes in process variables) ofmechanical properties (Fig. 4). Within the “defect free” quality process envelope, process regions (A and B) can be identified to achieve a desired set of mechanical properties. After ML has converged on a processing window to achieve the desired set of mechanical properties, ICME tools could be employed to further refine the processing window prior to investing in the development of statistically substantiated mechanical properties. The consistent and reproducible production of quality materials could then be ensured by employing adaptive, feedforward process controls. This would require real time data and analysis of information obtained from in situ sensors to be used in conjunction with predictive ICME tools. For AM, this methodology has several advantages to “freezing” a process. Operating within the quality processing envelope helps ensure a defect-free material. Machine-to- machine or manufacturer-to-manufacturer qualification is made easier as one needs only to ensure that the process is operating in the predetermined quality envelope. The identification of optimal processing parameters required to achieve a new set of customer materials property requirements is made easier as the entire process space has been previously mapped to mechanical performance. SUMMARY Within this article, the authors provided a brief introduction to machine learning. ML allows engineers and scientists to tease out causal relationships from complex data sets. To develop a functional model, data scientists must first thoroughly scrub the data to ensure pedigree, provenance, quality, and form. Appropriate algor- ithms can then be applied to the data sets to develop a useful ML model, i.e., a representation of reality. The potential of ML to accelerate the process of mechanical property allowables development was demonstrated in two recently completed, narrowly scoped projects. ML mechanical property allowables for laser powder bed fusion (LPBF) of a metal (17-4PH stainless steel) and a polymer (Nylon) were shown to be comparable to a conventional statistical based approach. The ML approach was shown to reduce the cost and time of an AM product deployment. The AMMP project demonstrated that development of two ML Fig. 3 — Cost comparison of machine learning and conventional allowables development. Fig. 4 — Quality AM processing envelope in n-dimensional space for a hypothetical alloy.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==