Skip to main content
U.S. flag

An official website of the United States government

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.

Data Analytics for Smart Manufacturing Systems


In simple terms, smart manufacturing systems have the ability to transform data gathered from a variety of manufacturing processes into actionable knowledge for decision making. Today, this ability exists in a large number of data analysis methods, algorithms, and tools – collectively known as data analytics (DA). DA has the potential to identify performance improvements across multiple levels of manufacturing system functionality. A number of manufacturers are applying data analytics tools, but with limited success – often obtaining actionable results too late to have impact. Much time is spent on identifying the target performance objectives, selecting or developing the appropriate DA tool, and managing input data (including data acquisition and data formatting.) For small and medium enterprises (SMEs), DA applications are prohibitively complex and expensive. A typical SME cannot afford to have a DA expert on staff. DA tools are being made available on the internet, which is increasing access and greatly reducing cost, but technical barriers remain that prevent more widespread adoption of DA. This project is addressing two major technical barriers to the adoption of data analytics in manufacturing: 1) selecting the appropriate data analytics tools and 2) integrating data analytics tools with data acquisition and decision support systems.


Objective:  Develop standards, software tools, methodologies, and guidelines to enable small and medium enterprises to apply data analytics services to improve decision-making and performance in smart manufacturing systems.

Technical Idea:  Manufacturers need enhanced decision-support tools that would help make decisions faster and better. Those tools will support the various decisions made by manufacturing supervisors, product designers, factory managers, as well as supply-chain managers. In smart manufacturing systems, decision-makers will use smart systems that include a data feedback loop that models, senses, transmits, analyzes, communicates, and takes action on data. Data analytics (DA) tools are expected to analyze that data and produce actionable intelligence for the decision-makers.

DA tools are complex and difficult to apply effectively. Manufacturers must identify the DA tools that best fit their needs. This requires a formalization of the desired performance requirements and optimization objectives for a specific manufacturing problem. These requirements and objectives must be matched to a prospective DA tool’s capabilities. Improving this process requires guidelines to represent those requirements and capabilities. It also requires measurement methods for evaluating the underlying algorithms that are implemented in DA tools including the expected uncertainty. Best practices for applying data analytics are needed to support SMEs.

DA tools are difficult to integrate into existing manufacturing systems and control processes. Two integration problems must be addressed. The first involves integrating the DA tool into the current decision-making architecture. The challenge is to enable the information flow from the operational technologies at different levels of the factory to their respective DA tools. A framework for integrating operational decisions in the factory with the DA tools that support them is needed along with new standards and tools to implement the framework.

The second integration problem involves integrating acquired data with the DA tool and results from DA tools with appropriate decision support systems. Many DA tools have proprietary interfaces and have specialized data formats for input and output. New information standards for DA systems are needed to solve these integration problems.

Research Plan:  The research plan for the project is divided into four thrusts: information standards, measurement methods, integration framework, and data analytics (DA) testbed.

Information Standards
One of the common parts of both requirements and capabilities involves the algorithms that are used to create the models that DA tools use to make predictions. Because of this, the DA project will focus initially on developing XML-based, information standards for representing those algorithms that support smart-manufacturing decision discussed above. The Predictive Model Markup Language (PMML), standardized by the Data Mining Group (DMG), was chosen to develop additional needed algorithms and models.

Currently, PMML’s regression and neural networks algorithms are used extensively in the finance industry; they are sufficient for the needed financial predictive models. However, manufacturing data is significantly different from finance. The project will extend the PMML standard to capture the needed predictive models for smart manufacturing applications.

Since analytics supports decisions in multiple levels of abstraction, manufacturers need algorithms that can produce the required predictive models. Initially, we plan to propose two new models to extend PMML: Gaussian process regression (GPR) and Bayesian networks (BN). These algorithms are capable of modeling the complexity and variability of smart manufacturing systems. The GPR model can provide confidence bounds around an estimated prediction, and the BN model can provide likelihood distributions for predicted values. Both are needed to provide the foundation for uncertainty quantification analysis.

Measurement Methods
ASME Verification and Validation in Computational Modeling and Simulation Committee (ASME V&V) was chosen as the organization to develop new measurement methods to evaluate DA models. Project staff successfully initiated a new subcommittee within ASME V&V focused on advanced manufacturing – V&V 50. A number of working groups within V&V 50 have been chartered to address standard vocabulary, verification and validation methods, and uncertainty quantification. Our focus will be on the development of practices/guidelines for data-driven models. These guidelines will enable practitioners to quantify uncertainty and ensure the credibility and accuracy of simulations and data-driven models such as DA.

Integration Framework
The project will develop a Multilevel Modeling Framework that will facilitate the integration of multi-level operational technologies with DA tools. The framework will allow users to maintain their operational specifications in a unified, coherent repository and access different DA services for analysis and evaluation. Users will interact with an ecosystem of models and tools through a domain-specific interface
that provides abstractions to easily describe the various operational technologies in the factory. To enable these interactions, and a visual interface for the execution of analytics algorithms and the interpretation of analytics results.

DA Testbed
As foundation for a longer-term effort to build a DA testbed, the project will develop standard implementations of PMML/GPR and PMML/BN using the translators that developed by NIST and their project partners for import and exports GPR and BN files. The project will also develop a variety of simulation models that can generate input data to test DA interfaces and the DA integration framework. The project intends to make the testbed publicly available.

Major Accomplishments


  • Gaussian process regression extension to the Data Mining Group Predictive Model Markup Language (PMML) standard. The extension provides a standard mechanism to represent confidence bounds for predictive estimations in system modeling.
  • Journal article describing the data generator using virtual and domain specific models of manufacturing machines.
  • Journal report on data acquisition, data analytics, optimization, and UQ using existing software.
  • Journal report on Uncertainty Quantification (UQ) and sensitivity analyses for prediction model for time varying data.
Created April 18, 2014, Updated March 17, 2017