NIST is working to meet an industry need to measure and improve the operational performance of commercial buildings by leveraging previously untapped capabilities within modern automation and control systems. These improvements will require developing a new measurement science based on state-of-the-art techniques of data analysis to automatically detect and diagnose unwanted operating conditions (“faults”) in the complex mechanical systems typical of large commercial buildings. Such systems include the equipment, sensors, and controllers of building mechanical heating, ventilation, and air-conditioning systems. The resulting automated fault detection and diagnosis (AFDD) software will autonomously acquire and in real time analyze data from control hardware and instrumentation products typically already in large commercial buildings. The outcome is a detailed yet comprehensive “24/7” fault surveillance that no maintenance staff of reasonable size could possibly conduct on its own. The AFDD software will apply artificial intelligence, machine learning, and statistical techniques to the data it acquires, automatically detecting faults in the equipment and systems placed under it’s surveillance. Upon detecting a suspected fault, the software will further provide expert and timely diagnosis of the underlying cause, helping maintenance staff ensure that buildings perform effectively and efficiently.
Objective - To develop a new and practical measurement science using data analytics and artificial intelligence to detect and diagnose faulty conditions in the mechanical systems (i.e., heating, ventilating, and air-conditioning) of commercial buildings, and to transfer that science to private–sector partners in the form of automated fault detection and diagnostic (AFDD) software components those partners can merge into their own products and services.
What is the new technical idea?
Building on its past work in fault detection and diagnostics, NIST has established cooperative research and development agreements with private-sector partners to develop novel AFDD software able to leverage the untapped capabilities of modern automation and control systems to improve the performance of buildings. Existing building automation technology now offers the prospect of data–rich performance surveillance that can be applied on the system-wide scales necessary to optimize overall building performance goals. The new technical idea applies a cross–disciplinary mix of expertise in artificial intelligence, the thermodynamics and fluid mechanics of the heating, ventilating, and air-conditioning processes under surveillance, and the skills to conceptualize and implement that blended expertise in the form of extensible, maintainable software applications able to run on multiple computing platforms. NIST's research will help its industry partners accelerate their development and marketing of new, highly-capable products and services, as already demanded by building owners, by offering its AFDD expertise principally in the form of an application programming interface (API).
What is the research plan?
NIST will enhance the relevancy and transferability of its proven expertise in applying and automating sophisticated artificial intelligence methods to AFDD. That effort began within NIST by building a new competency in formulating and expressing AFDD methods via an API made up of modular, serviceable, technically extensible object–oriented program (OOP) software components. The API is designed to join seamlessly with the software design practices used by industry to make marketable products and services.
The following goals include developing a prototype user interface able to test the new API using varied computing platforms and data sources owned by our private-sector partners. The tests will verify and demonstrate scalability and extensibility needed to meet the expected broad industry requirements. In collaboration with CRADA partners, prototype AFDD applications will be built from the API and deployed at multiple field locations to verify that artificial intelligence expert systems can meet the practical, real–world needs of building systems.
Testing with CRADA partners will include both web-based and cloud-based environments. Field test results will be used to improve the performance of the underlying artificial intelligence methods and identify any needed additional capabilities or features.
Expected future work include reliability enhancements aimed at robustness to spurious data, prioritizing for the user the economic impact of faults found, and enhancing autonomy by replacing some queries of the user for diagnostic information with direct references to new knowledge base elements.
 Turner, C. and M. Frankel. 2008. "Energy Performance of LEED for New Construction Buildings. U.S. Green Building Council.
Some recent accomplishments for the Automated Fault Detection and Diagnostics for the Mechanical Services in Commercial Buildings: