Back to results
US 11,144,046 B2Granted
Fault signal recovery apparatus and method
AssigneeDoosan Heavy Industries Construction Co., Ltd
InventorsJee Hun Park, Hyun Sik Kim
Life of this patent
Estimating the term…
Abstract
A fault signal recovery apparatus and method for collecting signals obtained in a plant and recovering normal signals from fault signals contained in the measured signals through a machine learning method includes receiving an input X including only normal signals for a plurality of tags, an input U including fault signals for a first group of tags among the plurality of tags and normal signals for a second group of tags, and an input S having information on the first group of tags including fault signals, and estimating and recovering normal signals for the first group of tags including fault signals based on feature information F, recovery model information P, and ensemble learning.
Claims10
A fault signal recovery apparatus comprising:
- a plant comprising a plurality of sensors respectively generating a plurality of tags over time, each of the plurality of tags indicative of a signal value acquired through measuring a state of the plant, the plurality of sensors generating signal values over a period of time as a signal group X composed of normal signals for the plurality of tags and including sensors for respectively detecting a difference pressure of an inlet filter of a power generator, a turbine exhaust pressure of the power generator, and a turbine exhaust temperature of the power generator,
- wherein the signal group X includes learning data used to create an existing fault prediction model,
- wherein the plurality of sensors further generate a signal group U composed of signals for the plurality of tags including a first group of tags having fault signals and a second group of tags having normal signals,
- wherein the fault signals of the first group include a signal group S composed of signals indicative a fault in the plant and signals resulting from a sensor fault or network problem,
- the fault signal recovery apparatus further comprising:
- a data processing unit configured to generate
- a signal group U* from the signal group U by removing signals of the first group of tags from the signal group U, such that the signal group U* is composed of all the signals of the signal group U except for the signals of the signal group S,
- a signal group Xs from the signal group X, the signal group Xs composed of only the normal signals for the first group of tags, and
- a signal group X* from the signal group X, the signal group X* composed of only the normal signals for the second group of tags;
- a modeling unit configured to receive the signal group X* and the signal group Xs and to extract feature information F from the signal group X*, the modeling unit creating plural pieces of recovery model information P for fault signal recovery based on the extracted feature information F, the received the signal group X*, and the received signal group Xs;
- and
- a recovery unit configured to estimate and recover normal signals for the first group of tags based on the signal group U*, the signal group Xs, the feature information F, and the plural pieces of recovery model information P,
- wherein the feature information F is extracted from the signal group X* by a first feature extraction unit included in the modeling unit, the first feature extraction unit using at least one specific tag included in the normal signals for the second group of tags,
- wherein the recovery unit comprises:
- a second feature extraction unit configured to extract feature information F* from the signal group U*, wherein the second feature extraction unit is provided separately from the first feature extraction unit and is included in the recovery unit;
- and
- a recovery model selection unit configured to select recovery models to be used for fault signal recovery from among the plural pieces of recovery model information P based on a comparison result between the feature information F and the feature information F*, and
- wherein the extracted feature information F indicates whether the state of the power generator is started up, whether the power generator is operated in a normal state, and in which of four seasons data were collected by the plurality of sensors.
15 more elements
A fault signal recovery method comprising:
- generating a plurality of tags over time, each of the plurality of tags indicative of a signal value acquired through measuring a state of a plant comprising a plurality of sensors generating signal values over a period of time as a signal group X composed of normal signals for the plurality of tags and including sensors for respectively detecting a difference pressure of an inlet filter of a power generator, a turbine exhaust pressure of the power generator, and a turbine exhaust temperature of the power generator,
- wherein the signal group X includes learning data used to create an existing fault prediction model,
- wherein the plurality of sensors further generate a signal group U composed of signals for the plurality of tags including a first group of tags having fault signals and a second group of tags having normal signals,
- wherein the fault signals of the first group include a signal group S composed of signals indicative a fault in the plant and signals resulting from a sensor fault or network problem,
- the fault signal recovery method further comprising:
- a data processing step of generating
- a signal group U* from the signal group U by removing signals of the first group of tags from the signal group U, such that the signal group U* is composed of all the signals of the signal group U except for the signals of the signal group S,
- a signal group Xs from the signal group X, the signal group Xs composed of only the normal signals for the first group of tags, and
- a signal group X* from the signal group X, the signal group X* composed of only the normal signals for the second group of tags;
- a recovery model creation step of receiving the signal group X* and the signal group Xs and extracting feature information F from the signal group X*, the recovery model creation step creating plural pieces of recovery model information P for fault signal recovery based on the extracted feature information F, the received the signal group X*, and the received signal group Xs;
- and
- a recovery step of estimating and recovering normal signals for the first group of tags based on the signal group U*, the signal group Xs, the feature information F and the plural pieces of recovery model information P,
- wherein the feature information F is extracted from the signal group X* by a first feature extraction step included in the recovery model creation step, the first feature extraction step using at least one specific tag included in the normal signals for the second group of tags,
- wherein the recovery step comprises:
- a second feature extraction step of extracting feature information F* from the signal group U*, wherein the second feature extraction step is provided separately from the first feature extraction unit and is included in the recovery step;
- and
- a recovery model selection step of selecting recovery models to be used for fault signal recovery from among the plural pieces of recovery model information P based on a comparison result between the feature information F and the feature information F*, and
- wherein the extracted feature information F indicates whether the state of the power generator is started up, whether the power generator is operated in a normal state, and in which of four seasons data were collected by the plurality of sensors.
15 more elements
Dependencies are inferred from the claim text.