Skip to main content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Knowledge Extraction and Application for Manufacturing Operations

Summary

To develop and deploy advances in standards, measurement science, and software tools using actionable, computable, domain knowledge stemming from informal text-based data to augment a manufacturers’ ability to perform model-based and data-driven analyses. 

Description

Human expertise remains a major component of many procedures in manufacturing. However, the prevailing paradigm underlying smart manufacturing technologies often stops short of explicit integration of human knowledge and human-machine interactions in operational analysis. Unfortunately, this means human-generated knowledge is captured in text-based documents that are often riddled with inconsistencies, relatively few-in-number, and filled with domain-specific jargon. The problems with these documents leads to difficulties in consistently using them when making operational decisions on the manufacturing floor as they are difficult to analyze in their current form. This project will explore hybridized Artificial Intelligence (AI) and expert-driven methodologies for quantifying human knowledge, in which Natural Language Processing and graph-theoretic methods are introduced. The project aims to assist in labelling and analyzing text-based documents to enable decision making and continuous improvement. This project will contribute to domain-specific standards and associated open-source toolkits for extracting knowledge to improve manufacturing decision making.

WHAT IS THE TECHNICAL IDEA?
Decision making within manufacturing is still largely driven by human expertise. While developments in smart manufacturing have enabled more automated decision making, human experts still contain a wealth of tacit information that is informally captured and often explicitly underutilized. For example, if an experienced operator only needs to listen to a machine to “know” it’s health, then can the same knowledge be captured computationally? Also, when an experienced worker leaves the organization, what happens to his/her knowledge? Is the knowledge captured and reused to support newer, less experienced employees or does the knowledge leave with the employee? 

Often, this tacit knowledge is implicitly represented in text-based documents. These documents are filled with jargon, abbreviations, and domain-specific shorthand making them difficult to analyze with commercially available solutions. However, text-based documents often yield important contextual information that sensor data alone cannot provide. For example, during a machine failure sensor data might provide spindle speed patterns leading up to the failure, but a human technician can describe the cause and resolution of the problem at length. This type of contextual information is useful when these same patterns occur again to inform the proper control actions.

The technical idea of this project is to create methods, guidelines, and toolkits to study and analyze these informal, text-based documents to support operational decisions.  The project will research methodologies for transforming these documents into a computable format to augment manufacturer’s ability to perform analysis. These methods will be realized in standards and best practice guidelines for analyzing manufacturing text-based documents. The goal of this project is to enable manufacturers to transform and analyze these text-based documents themselves with aid from methodologies realized in open source toolkits and guidelines developed at NIST. 

WHAT IS THE RESEARCH PLAN?
Industry needs guidelines and associated toolkits for efficiently extracting tacit knowledge from manufacturing-related, text-based documents. The goal of this project is to explore methods for processing and analyzing these documents to support operational decisions. The project will create recommended practices for analyzing data generated from documents. Further, the project will develop guidelines and open source toolkits to aid manufacturers in analyzing documents. The research plan includes three main thrusts: 

  • Manufacturing Text-Based Document Analysis 
    • Improve and further distribute a NIST developed tool for cleaning and labeling text-based manufacturing historical logs (e.g. maintenance work orders, engineering change requests)
    • Develop and standardize methods using NLP techniques to clean text-based data for manufacturing
    • Create a publicly-available reference manufacturing text-based dataset
    • Explore text-based documents outside of text-based manufacturing historical logs (e.g. quality reports, requirements documents, maintenance manuals) 
    • Research methods for capturing tacit knowledge from other sources than text-based documents. 
  • Integrating text-based data with operational analysis 
    • Develop reusable data analysis workflows with text-based documents 
    • Create open source toolkits and visualizations for text-based data analysis 
  • Facilitate decision making in specific domain applications using text-based data
    • Explore utilization of text-based data for shop-floor scheduling 
    • Research methods for merging text-based data with streaming sensor data for production control 
    • Develop methods for analyzing the role of human knowledge in manufacturing capability modeling 

The research plan will be carried out by exploring and adapting methods in natural language processing for use in technical, manufacturing-related, text-based documents. The result will enable knowledge extraction from documents, used in conjunction with streaming sensor data, in support of operational decisions on the shop floor, as explored by other projects within the MBE program.
 

Created December 3, 2018, Updated August 26, 2019