AI-driven Part Printability Recommendation System for Additive Manufacturing

J.A. Steets, D. Mooney, B. O’Briant
Illumination Works, LLC.,
United States

Keywords: machine learning, artificial intelligence, additive manufacturing, decision support system, part printability, artificial neural network, computer vision, 2D engineering drawing, 3D engineering model


Additive manufacturing (AM) coupled with artificial intelligence (AI) is game-changing enabling technology. AM is disrupting manufacturing, repair, and maintenance with many benefits including mass customization, speeding time to market, and reducing costs. Additionally, AM helps mitigate supply chain risks associated with part obsolescence, part demands in remote locations, and long lead times for traditionally manufactured parts, as AM offers an alternative source of supply. At present, determining whether a part is suitable for AM is largely a manual process that involves a heavy time investment by AM engineers. Through a Phase I Small Business Innovative Research (SBIR) effort, Illumination Works, LLC (ILW) demonstrated the feasibility of our Linnea Part Printability Recommendation System, which automatically processes technical/engineering data, identifies relevant data features, applies AI/machine learning (ML) algorithms, and predicts the suitability of parts for AM. ILW successfully determined achievability of an automated pipeline for processing 2D/3D engineering drawings/models to extract and derive relevant data features for input to Linnea’s Classification System. Linnea’s Feature Extraction and Analysis Pipeline leverages traditional data engineering and computer vision approaches as well as AI/ML techniques to appropriately extract key information from technical data and automatically transforms that data into features for input to the Linnea Classification Engine. The Linnea Classification Engine is a two-stepped process for predicting AM suitability. The Classification Engine first applies expert decision rules to automatically exclude parts with features that deem the part not suitable for printing. Next, the Linnea Classification Engine applies well-performing ML classifiers to predict the suitability of the part for AM based on complexity, geometry, and related features. ILW trained and tested three ML algorithms for performance in predicting AM part candidacy based on these key features. The best performing model, an artificial neural network, predicted part suitability for AM with 90% accuracy. Additionally, ILW built out the initial decision rules for Linnea and successfully applied these rules to 2D and 3D engineering data. ILW’s Linnea is an enabler for the Department of Defense (DoD) that will greatly improve the efficiency of the AM selection process and save AM engineers’ time by automating aspects of the part candidacy analysis pipeline. Ultimately, Linnea will output both summary information about a part and a printability recommendation, upon which engineers will make final printability decisions. By infusing AI-driven automation and human-machine teaming into the AM candidacy process, the DoD will be able to take fuller advantage of AM technologies and improve operational readiness. ILW designed Linnea to be flexible, scalable, and open source for rapid extension through Army, across DoD, into Product Lifecyle Management (PLM) tools, and then to commercial industries. Given the projected growth of AM in the next decade, Linnea has exceptional commercialization potential and impact for the manufacturing, repair, and maintenance industry. Linnea’s automation and integration into PLM systems will enable all industries who manufacture parts and/or repair equipment that need specialized components to benefit from speedier and less labor-intensive part printability candidacy analysis.