Operational Artificial Intelligence

The use of artificial intelligence (AI) in energy applications has a game-changing potential in automating expensive and manual human activities in various types of industries. In the energy industry, power plants (especially nuclear) rely on staff performing several types of manual activities on a regular basis. Future energy plants, including nuclear advanced reactors, are designed to reduce the dependence on people for the operations, maintenance, and support activities of a plant. A light water nuclear power plant is typically full of analog gauges and manual actuators. By comparing the nuclear power plant control room to a modern airplane cockpit where the plane can fly itself and the pilot’s role can be reduced to simply monitoring the airplane, it is obvious that a significant technology gap exists that needs to be closed. Human intelligence needs to be replaced by machine intelligence in various forms of AI if this vision is to be realized.

Idaho National Laboratory (INL) has several efforts currently in progress to apply various forms of AI methods into energy applications. These methods are often available in a ready-to-use form (i.e., the method algorithms are available). However, the applicability of these methods on the various applications in the plants often requires the selection of compatible methods, customizing the methods, combining multiple methods, and integrating the methods with other science processes (e.g., modeling tools) for improved applicability and performance. Through the matching of methods and applications, gaps could be identified that are used to drive the national operational AI research focus.

Collaborations & Projects:

Light Water Reactor Sustainability Program

The U.S. Department of Energy (DOE) Light Water Reactor Sustainability (LWRS) Program has been a pioneer in introducing AI methods to automate human activities for the nuclear energy sector, as the program has recognized the potential of AI for enabling great cost-savings for nuclear plants. The program works directly with the nuclear power industry. Several reports discussing the application of AI can be found at the LWRS website (https://lwrs.inl.gov). Here are two examples of the efforts currently ongoing within the LWRS program:

Anomalies detection using sensors data inference: Anomalies in a plant represent changes in a process that could be considered as early indicators of equipment failure. Escalation of an anomaly to a functional failure can result in severe economic or safety consequences to a plant, which would demand a responsive effort with costs that often far exceed what would have been invested to mitigate the issue. Several machine learning methods were used to monitor a dry well cooling fan in a nuclear power plant, which provided an applied perspective into the method’s performance. Details of this study can be found in: https://lwrs.inl.gov/Advanced%20IIC%20System%20Technologies/Subtle_Process-Anomalies_Detection_Using_Machine-Learning_Methods.pdf.

Automating fire watch using object recognition methods to identity fire in a video stream: The U.S. Nuclear Regulatory Commission (NRC) Regulatory Guide 1.189 defines the fire watch as “Individuals responsible for providing additional (e.g., during hot work) or compensatory (e.g., for system impairments) coverage of plant activities or areas to detect fires or to identify activities and conditions that present a potential fire hazard.” A person conducting a fire watch could spend hours watching certain equipment, a room, or an environment. Multiple methods were evaluated for automated visual recognition of fire; neural networks coupled to a color-based feature-extraction method to extract fire regions resulted in accurate fire recognition results. Details can be found in: https://lwrs.inl.gov/Advanced%20IIC%20System%20Technologies/Automating_Fire_Watch_in_Industrial_Environments_through_Machine.pdf.

MOOSE Simulation Environment

Modeling and simulation has now become standard practice in nearly every branch of science. Building a useful simulation capability has traditionally been a daunting task because it required a team of software developers working for years with scientists to describe a given phenomenon.

Idaho National Laboratory’s MOOSE (Multiphysics Object Oriented Simulation Environment) now makes modeling and simulation more accessible to a broad array of scientists. MOOSE enables simulation tools to be developed in a fraction of the time previously required. The tool has revolutionized predictive modeling, especially in the field of nuclear engineering — allowing nuclear fuels and materials scientists to develop numerous applications that predict the behavior of fuels and materials under operating and accident conditions.

Scientists who don’t have in-depth knowledge of computer science can now develop an application that they can “plug and play” into the MOOSE simulation platform. In essence, MOOSE solves the mathematical equations embodied by the model.

Such a tool means scientists seeking a new simulation capability don’t need to recruit a team of computational experts versed in, for example, parallel code development. Researchers can focus their efforts on the mathematical models for their field, and MOOSE does the rest. The simplicity has bred a herd of modeling applications describing phenomena in multi-scale nuclear fuels (BISON, Marmot), reactor physics (MAMMOTH, Rattlesnake), geology (FALCON), geochemistry (RAT), nuclear power plant systems/safety analysis (RELAP-7), and reactor engineering (Pronghorn).

Publications: