Method and apparatus for real-time learning-based augmented irradiation control and optimization
AssigneeOregon State University
InventorsSteven R. Reese, Ophir Frieder, Mario Enrique Gomez Fernandez
Life of this patent
Estimating the term…
Abstract
A machine-learning tool learns from sensors' data of a nuclear reactor at steady state and maps them to controls of the nuclear reactor. The tool learns all given ranges of normal operation and responses for corrective measures. The tool may train another learning tool (or the same tool) that forecasts the behavior of the reactor based on real-time changes (e.g., every 10 seconds). The tool implements an optimization technique for differing half-life materials to be placed in the reactor. The tool maximizes isotope production based on optimal controls of the reactor.
Claims10
A machine-readable storage medium having machine-readable instructions that, when executed, cause one or more machines to perform a method to improve an operation of a nuclear reactor by monitoring an alarm threshold to prevent unplanned shutdown of the nuclear reactor to increase an isotope production, the method comprising:
- collecting data from sensors of the nuclear reactor at a steady state condition;
- training an optimization model with multiple objectives using the collected data, wherein the optimization model uses one or more machine-learning algorithms configured to:
- learn patterns by mapping the collected data to a set of controls of the nuclear reactor, wherein the set of controls are associated with a control panel of the nuclear reactor, wherein the set of controls include one or more controls to:
- adjust movement of a control rod, modify operating modes, change primary cooling flow rate, change secondary cooling flow rate, adjust cooling fan speed, adjust ventilation rate, or adjust differential pressure;
- and
- learn a setting of the set of controls that increases the isotope production from the predicted learned patterns;
- reporting whether an output from the sensors of the nuclear reactor is above or below the alarm threshold;
- and
- predicting, by processing the collected data via the trained optimization model, operational issues with the nuclear reactor to avoid the unplanned shutdown of the nuclear reactor to increase the isotope production, wherein training the optimization model comprises applying one or more supervised machine-learning algorithms by analyzing and learning the collected data to:
- determine a location for placement of a sample in the nuclear reactor;
- schedule for the isotope production from the sample;
- and
- adjust the set of controls, via application of the trained optimization model by a computer, to reduce disruption of the schedule which in turn minimizes the unplanned shutdowns of the nuclear reactor to increase the isotope production.
A method to improve an operation of a nuclear reactor by monitoring an alarm threshold to prevent unplanned shutdown of the nuclear reactor to increase an isotope production, the method comprising:
- collecting data from sensors of the nuclear reactor at a steady state condition;
- training an optimization model with multiple objectives using the collected data, wherein the optimization model uses one or more machine-learning algorithms configured to:
- learn patterns by mapping the collected data to a set of controls variables of the nuclear reactor, wherein the set of controls is associated with a control panel of the nuclear reactor, wherein the set of controls include one or more controls to:
- adjust movement of a control rod, modify operating modes, change primary cooling flow rate, change secondary cooling flow rate, adjust cooling fan speed, adjust ventilation rate, or adjust differential pressure;
- and
- learn the set of controls that increases the isotope production from the learned patterns;
- reporting whether an output from the sensors of the nuclear reactor is above or below the alarm threshold;
- and
- predicting, by processing the collected data via the trained optimization model, operational issues with the nuclear reactor to avoid the unplanned shutdown of the nuclear reactor to increase the isotope production, wherein training the optimization model comprises applying one or more supervised machine-learning algorithms by analyzing and learning the collected data to:
- determine a location for placement of a sample in the nuclear reactor;
- schedule for the isotope production from the sample;
- and
- adjust the set of controls, via application of the trained optimization model by a computer, to reduce disruption of the schedule which in turn reduces the unplanned shutdowns of the nuclear reactor to increase the isotope production.
Dependencies are inferred from the claim text.