AM data is essential for establishing part traceability, understanding AM processes and making decisions during the product development lifecycle. The curation, integration, fusion, sharing and analysis of this data are considerable challenges because AM embodies all the 4 V's characteristics of Big Data - volume, velocity, variety, and veracity. This project will develop models, methods and best practices for data lifecycle management, data integration, and data fusion in additive manufacturing to facilitate the effective and efficient curation, sharing, processing and use of AM data and enable AM knowledge discovery for process improvement. More specifically, the project will focus on four major research threads: 1) Development of a common information model for AM lifecycle data management. 2) Development of best practices and guidelines for AM data generation, collection, sanitization, anonymizing, curation, validation and storing. 3) Development of data structure and format to represent predictive models for AM applications. 4) Development of AM data federation and information fusion methods for integrating diverse datasets. The research results will be transferred to standard development organizations to improve their existing standards or initiate new ones to fill data related AM standards gaps.
Data plays the most critical role in linking AM lifecycle and value chain activities and streamlining the AM development process. While the AM data from a single build is essential for establishing part traceability, when methodically collected, the full processing history of thousands of components can be mined to advance our understanding of AM processes. Hence, this full body of data must be captured, stored, shared and properly managed for easy query and analysis. The curation, integration, fusion, sharing and analysis of this data are considerable challenges because AM embodies all the 4 V's characteristics of Big Data - volume, velocity, variety, and veracity. Previous works have investigated the characteristics and needs of AM representations at different stages of the AM lifecycle, including material/machine data, design data, process data and test/experiments data. While meta-level understandings of AM data are essential to investigating and defining the flow of AM data across a product lifecycle, such approaches only begin to address the data integration challenges, such as disconnected AM activities and a lack of standard data exchange formats, currently faced by the AM community.
This project will continue to develop models, methods and best practices for data lifecycle management, data integration, and data fusion in additive manufacturing to facilitate the effective and efficient curation, sharing, processing and use of AM data and enable AM knowledge discovery for process improvement. More specifically, the project focuses on addressing the data modeling and integration needs to close several high and medium priority gaps identified in the latest Standardization Roadmap for Additive Manufacturing from America Makes & ANSI Additive Manufacturing Standardization Collaborative (AMSC). There are four major research threads defined within the project: 1) Enhancement of the common AM data model to incorporate in-situ monitoring data, post NDE inspection data, AM design feature characteristics, microstructure characteristics and material/machine capability. 2) Development of best practices and guidelines for AM data generation, collection, sanitization, anonymizing, curation, validation and storing, through sharing exemplary data generated from NIST MSAM research activities. 3) Development of data structure and format to represent predictive models for AM applications and demonstrating the use through a prototype toolset 4) Development of AM data federation and information fusion methods for integrating diverse datasets from various AM stakeholders.
What is the Research Plan?
The efforts to enhance the existing AM information model will serve the AM community a common structure to understand, curate and exchange AM data generated through product development lifecycle and material and machine value chains. Representing and capturing the full history of AM processing data of thousands of samples and parts will help establish AM material process-structure-property relationships, which is essential to close two of AMSC high priority gaps: Gap FMP1: Material Properties and Gap FMP4: Design Allowable. Gap FMP1 recommends the identification of the means to establish minimum mechanical properties and Gap FMP4 recommended R&D of generation of a set of initial seed data and subsequent statistical analysis. This research effort also helps address GAP FMP5, which recommends developing a standard for characterization and acceptance of AM microstructure, as well as the gaps in the Design Guides/Tools section by characterizing and modeling AM difficult-to-build features, AM process/material/machine capabilities. The project team will collaborate with the 3Ds project and other AM projects project in understanding the data representation and integration needs from AM in-situ monitoring, AM post inspection, AM microstructure characterization, AM design and AM predictive model modeling, and will actively participate in the joint group ISO/TC 261-ASTM F 42 JG73: Digital Product Definition and Data management and propose new work items on standardizing AM information models.
The effort to develop best practices for AM Data creation, collection, sanitization, anonymizing, curation, validation and storing is to provide guidelines for the AM community to generate and share high quality data in support of their research and production activities. The best practices will be developed from the processes of collecting, curating and sharing the data generated from MSAM research activities. The Additive Manufacturing Material Database will serve for the project as an infrastructure to capture this exemplary data. During this effort, the AM Data Integration and Management project will work with other projects in the MSAM program, to identify their critical data curation challenges, discuss potential ways of overcoming them, and develop a solution. Such challenges may come in the form of identifying data generation needs, labeling raw datasets especially for video and image registration, cleaning the raw data to eliminate errors in data handling, defining naming conventions to protect vendor’s privacy and data quality validation. The research results can be transferred to ISO/TC 261-ASTM F 42 JG73 and SAE AMS-AM Data Management subcommittee to publish AM data handling guidelines.
The effort to develop data structure and format to represent predictive models for AM applications will focus on developing standards needed to solve that interoperability problem to enhance the use ML algorithms in AM product, process and material development lifecycle. Many AI machine-learning-related tools have proprietary interfaces and have specialized data formats for inputs and outputs. Predictive Model Markup Language(PMML) can be the foundation for developing the input data and data mining model needed by ML services, either through web-based or cloud-based calls. The project team will extend the PMML standard to capture the needed predictive models for additive manufacturing applications. Initially, convolutional neural network (CNN) structure, one type of DL, will be proposed to the DMG. To demonstrate the potential manufacturing benefits of DL and PMML/CNN, the project will collaborate with AM’s other projects (Real-time Monitoring and 3Ds) to analyze image data. Image data, in the form of either infrared melt-pool images or optical layer images, is particularly important for controlling AM processes. These images are collected in real-time during the deposition of each layer. The collected images will be labeled for further analysis. Analyzing these images with the CNN algorithms could help manufacturers to (1) monitor process parameters and machine condition, (2) (re)set process parameters if necessary, (3) detect voids and faults in the product, and (4) make predictions about the properties of as-built parts. A prototype toolset will be created during the practice. To achieve the goals, the AM data management project will work closely with the 3Ds project, getting the requirements from them on the representation needs of ML algorithms and providing them a prototype toolset to test their algorithms.
The effort in AM data federation and information fusion will develop methods to integrate heterogeneous datasets from various data repositories and to derive meaningful interpretation by fusing data generated from multiple sensor systems and at separate time phases and locations. Current data integration relies heavily on reverse engineering and repopulation of legacy/existing databases, which is not only costly but also error-prone. This project will investigate not only common information model based data integration, but also data integration based on metadata registry federation. Algebraic Query Language (AQL), which was developed under a NIST SBIR grant, is currently being used in the Smart grid project to integrate Manufacturing Service databases Once the data is integrated, AQL allows for extensions and incorporation of new and additional analysis tools including machine learning tools. Information fusion is to derive the accurate meaning of data from multiple sources and at different AM lifecycle phases. The project team will focus on fusing the in-situ monitoring data from multiple sensor systems and the NDE post inspection data, in a coordinate system defined with the design CAD models. This effort will help address the data characterization, modeling and fusion challenges identified by AMSC Gap D22: In-Process Monitoring for design documentation, Gap PC16: In-Process Monitoring for process control and Gap NED5: Data Fusion for nondestructive evaluation. The results from this research effort can be incorporated into an ASTM draft guide standard (WK62181) led by E7.10. The project team will work with the AM Part Qualification project on investigating heterogeneous NDE post inspection data fusion for part surface qualification.