# US 11,308,408 B2 · Fault signal recovery system and method

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 11,308,408 B2
- Document ID: US11308408B2
- Status: Granted
- Kind: B2
- Application number: 15/949,364
- Filed: 2018-04-10
- Published: 2018-11-15
- Granted: 2022-04-19
- Earliest claimed: 2017-05-15
- Assignees: Doosan Heavy Industries & Constructon Co., LTD
- Applicants: Doosan Heavy Industries & Construction Co., LTD
- Inventors: Jee Hun Park; Hyun Sik Kim
- Web page: https://patentgrep.com/patent/US11308408B2

## CPC

- G06N 5/047: Pattern matching networks; Rete networks
- G01M 99/00: Subject matter not provided for in other groups of this subclass
- G06F 11/00: Error detection; Error correction; Monitoring (error detection, correction or monitoring in information storage based on relative movement between record carrier and transducer G11B20/18; monitoring, i.e. supervising the progress of recording or reproducing G11B27/36; in static stores G11C29/00)
- G06F 11/0751: {Error or fault detection not based on redundancy (power supply failures G06F1/30; network fault management H04L41/06)}
- G06F 11/3013: {where the computing system is an embedded system, i.e. a combination of hardware and software dedicated to perform a certain function in mobile devices, printers, automotive or aircraft systems (testing or monitoring of control systems or parts thereof G05B23/02)}
- G06F 11/3055: {Monitoring arrangements for monitoring the status of the computing system or of the computing system component, e.g. monitoring if the computing system is on, off, available, not available (error or fault processing without redundancy G06F11/0703; error detection or correction by redundancy in data representation G06F11/08; error detection or correction of the data by redundancy in operations G06F11/14; error detection or correction by redundancy in hardware G06F11/16)}
- G06F 11/3058: {Monitoring arrangements for monitoring environmental properties or parameters of the computing system or of the computing system component, e.g. monitoring of power, currents, temperature, humidity, position, vibrations (thermal management in cooling arrangements of a computing system G06F1/206)}
- G06F 11/3089: {Monitoring arrangements determined by the means or processing involved in sensing the monitored data, e.g. interfaces, connectors, sensors, probes, agents (software debugging using additional hardware using a specific debug interface G06F11/3656; performance evaluation by tracing or monitoring G06F11/3466)}
- G06N 20/00: Machine learning
- G06N 20/20: Ensemble learning
- G06F 2201/805: Real-time
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G01M: TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Disclosed is a fault signal recovery system including a data processor configured to generate a signal subset U* by removing, from a signal set U for a plurality of tags, some tags including a fault signal, and a first learning signal subset X* by removing tags disposed at positions corresponding to the some tags from a learning signal set X containing only tags of normal signals, a modeling unit configured to generate feature information F extractable from the first learning signal subset X* and recovery information P on a plurality of recovery models usable for restoring the fault signal, and a recovery unit configured to select an optimum recovery model by matching the feature of the learning signal set X with the recovery models generated through the recovery in formation P to estimate and recover normal signals for the some tags.

## Claims

### Claim 1

A fault signal recovery system comprising:
a data processor configured to receive a signal set U from sensors in a plant, where U includes one or more tags that are not normal signals and include a fault signal that is caused by one of an actual fault, a sensor error, and a transmission error, and wherein the one or more tags are to be restored to normal data when the fault signal is caused by a sensor error or a transmission error, an Input S comprising information on fault data in the one or more tags to be restored, and a learning signal set X comprising learning data comprising a plurality of tags when there are all normal signals acquired from the sensors and used to construct a failure prediction model;
wherein the data processor is configured to generate a signal subset U* by removing, from the signal set U the one or more tags that include a fault signal identified by the Input S including information on the one or more tags that contain the fault data to be restored to normal data, and to generate a first learning signal subset X* by removing from & the learning signal set X tags of the normal signals tags disposed at positions corresponding to the one or more tags that are not normal in signal set U, and wherein the data processor is further configured to generate a second learning signal subset Xs containing only signals for the one or more tags that are not normal and the first learning signal subset X* containing only signals for the remaining tags other than the one or more tags that are not normal by processing the learning signal set X;
a modeling device configured to receive the first learning signal subset X* and the second learning signal Xs, and wherein the modeling device is configured to use machine learning to generate feature information F extractable from the first learning signal subset X* and to generate recovery information P using X*, Xs, and F on a plurality of recovery models usable for restoring the fault signal; and
a recovery device configured to receive the signal subset U*, the first learning signal subset X*, and the second learning signal subset Xs, and wherein the recovery device is configured to analyze a feature of the learning signal set X based on the feature information F and wherein the recovery device is configured to use machine learning to select an optimum recovery model Y by matching the feature of the learning signal set X with the recovery models generated through the recovery information P to estimate and recover normal signals for the one or more tags that are not normal and to use the normal signals and the first learning signal subset X* as the learning data, and
thereby the system provides a fault recovery system with improved accuracy to recover the fault signal when the one or more tags that are not normal are sensor errors or transmission errors and wherein the learning signal set X can be updated with the signal set U and the recovered normal signals.

### Claim 2 (depends on 1)

The fault signal recovery system according to claim 1, wherein the recovery device comprises:
a pattern analyzer configured to analyze the feature of the learning signal set X based on the feature information F and extract information on the feature;
a model selector configured to select the optimum recovery model matching the feature of the learning signal set X from among the recovery models generated through the recovery information P; and
an ensemble learning device configured to generate an optimum recovery algorithm by generating a recovery value for the one or more tags by selecting an algorithm matching the optimum recovery model.

### Claim 3 (depends on 2)

The fault signal recovery system according to claim 2, wherein the optimum recovery model comprises at least one model selected in each of a parametric model group and a nonparametric model group.

### Claim 4 (depends on 3)

The fault signal recovery system according to claim 3, wherein the model selector selects a plurality of recovery models in each of the parametric model group or the nonparametric group, and
wherein the model selector selects the optimum recovery model by setting weights for the plurality of recovery models based on the feature of the learning signal set X.

### Claim 5 (depends on 4)

The fault signal recovery system according to claim 4, wherein the ensemble learning device selects an algorithm to be applied to the optimum recovery model based on the weights assigned to a plurality of recovery models, and
wherein the algorithm is matched with a recovery model that is assigned a greater one of the weights assigned for the recovery models in generating the optimum recovery model.

### Claim 6 (depends on 2)

The fault signal recovery system according to claim 2, wherein the learning signal set X comprises a plurality of features, and
wherein the model selector selects a plurality of the recovery models matched with the features of the learning signal set X, respectively.

### Claim 7 (depends on 1)

The fault signal recovery system according to claim 1, wherein the modeling device comprises:
a feature extractor configured to extract the feature information F extractable from the first learning signal subset X* received from the data processor; and
a model generator configured to generate the recovery information P on the plurality of recovery models usable for restoring the fault signal based on the first learning signal subset X* and the second learning signal subset Xs received from the data processor and the feature information F.

### Claim 8

A fault signal recovery method comprising:
from receiving a signal set U comprising data from sensors in a plant for a plurality of tags, where U includes one or more tags that are not normal signals and include a fault signal that is caused by one or an actual fault, a sensor error, and a transmission error, and wherein the one or more tags are to be restored to normal data when the fault signal is caused by a sensor error or a transmission error,
generating a signal subset U* by removing from the signal set U one or more tags that include a fault signal identified by an Input S comprising information on fault data in the one or more tags to be restored including information on the one or more tags that contain the fault data to be restored, and to generate a first learning signal subset X* by removing from a learning signal set X tags disposed at positions corresponding to the one or more tags and wherein the learning signal set X comprises learning data comprising a plurality of tags when there are all normal signals acquired from the sensors and used to construct a failure prediction model;
generating feature information F extractable from the first learning signal subset X* and to generate recovery information P using X*, Xs, and F on a plurality of recovery models usable for restoring the fault signal;
analyzing a feature of the learning signal set X based on the feature information F and selecting an optimum recovery model by matching the feature of the learning signal set X with the recovery models generated through the recovery information P;
estimating and recovering a normal signal for the some tags by applying an algorithm matching the optimum recovery model and using the normal signal and the first learning signal subset X* as learning data, and
generating a second learning signal subset Xs containing only signals for the some tags and the first learning signal subset X* containing only signals for the remaining tags other than the some tags by processing the learning signal set X; and
thereby the system provides a fault recovery method with improved accuracy to recover the fault signal when the one or more tags that are not normal are sensor errors or transmission errors and wherein the learning signal set X can be updated with the signal set U and the recovered normal signals.

### Claim 9 (depends on 8)

The method according to claim 8, wherein the selecting of the optimum recovery model comprises:
selecting at least one recovery model in each of a parametric model group and a nonparametric model group and matching the same with the learning signal set X, the parametric model group and the nonparametric model group constituting the recovery information.

### Claim 10

The method according to 8, wherein the selecting of the optimum recovery model comprises:
selecting a plurality of recovery models in each of the parametric model group or the nonparametric model group; and
selecting the optimum recovery model by setting weights for the plurality of recovery models based on the feature of the learning signal set X.

### Claim 11 (depends on 10)

The method according to claim 10, further comprising:
selecting an algorithm to be applied to the optimum recovery model based on the weights assigned to a plurality of recovery models, and
wherein the algorithm is matched with a recovery model that is assigned a greater one of the weights assigned for the recovery models in generating the optimum recovery model.

### Claim 12

A method of using data containing one or more fault signals as learning data comprising:
receiving a plurality data signals from a plurality of sensors in a plant once per a time period, wherein each data signal from each sensor is a tag;
adding the plurality of data signals to a learning signal set X when each of the tags is a normal signal;
receiving a signal set U from the sensors in the plant wherein U includes one or more tags that is not a normal signal and include a fault signal that is caused by one of an actual fault, a sensor error, and a transmission error, and wherein the one or more tags are to be restored to normal data when the fault signal is caused by a sensor error or a transmission error;
generating a signal set U* by removing from the signal set U the one or more tags that contain the fault data;
generating a first learning signal subset X* by removing from the learning signal set X tags of the normal signal tags corresponding to the one or more tags that are not normal in signal set U;
generating a second learning subset Xs containing only signals from the one or more signal tags that are not normal;
using a modeling device to receive X* and Xs to generate recovery information for restoring each fault signal of the one or more tags that are not normal;
thereby the method recovers normal data for the one or more tags that are not normal and updating the learning signal set X with the signal set U and the recovered normal signals.