# US 11,443,850 B2 · Max-margin temporal transduction for automatic prognostics, diagnosis and change point detection

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

- Number: US 11,443,850 B2
- Document ID: US11443850B2
- Status: Granted
- Kind: B2
- Application number: 15/634,042
- Filed: 2017-06-27
- Published: 2018-12-27
- Granted: 2022-09-13
- Earliest claimed: 2017-06-27
- Assignees: General Electric Company
- Applicants: General Electric Company
- Inventors: Abhay Harpale
- Web page: https://patentgrep.com/patent/US11443850B2

## CPC

- G16H 50/20: for computer-aided diagnosis, e.g. based on medical expert systems
- G05B 23/0283: {Predictive maintenance, e.g. involving the monitoring of a system and, based on the monitoring results, taking decisions on the maintenance schedule of the monitored system; Estimating remaining useful life [RUL] (preventive maintenance, i.e. planning maintenance according to the available resources without monitoring the system G06Q10/06)}
- G06N 7/08: using chaos models or non-linear system models
- G06N 20/00: Machine learning
- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- G16H 50/00: ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H 50/50: for simulation or modelling of medical disorders
- A61B 5/0022: {Monitoring a patient using a global network, e.g. telephone networks, internet}
- A61B 5/02055: {Simultaneously evaluating both cardiovascular condition and temperature}
- A61B 5/7267: {involving training the classification device}
- A61B 5/7275: {Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor}
- G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)

## Abstract

A method of detecting status changes, and a corresponding point-in-time, in monitored entities, includes receiving one or more elements of time-series data from one or more sensors, the elements of time-series data representing an operational state of the monitored entity, creating a predictive model from the time-series data in a datastore memory, applying a transduction classifier to the predictive model, the transduction classifier detecting a change from healthy to unhealthy in the time-series data, and the corresponding point-in-time when the change occurred, and providing an identification of the change in the time-series data and the corresponding point-in-time. In some embodiments the transduction classifier can be a maximum margin classifier having a support vector machine component and a temporal transductive component. A system and a non-transitory computer readable medium are also disclosed.

## Claims

### Claim 1

A method of detecting status changes, and a corresponding point-in-time, in one or more monitored entities, the method comprising:
receiving, via one or more sensors, one or more signals being representative of an operational state of the one or more monitored entities;
transferring, via the one or more sensors, one or more elements of time-series data derived from the one or more signals, to a processor being communicatively coupled to the one or more sensors, the time-series data comprising both labeled and unlabeled data;
wherein the labeled data includes: the initial measurements from all of the monitored entities and the final measurements from monitored entities with known unhealthy states;
creating a predictive model from the time-series data in a datastore memory;
applying a transduction classifier to the predictive model, the transduction classifier detecting a change from healthy to unhealthy in the time-series data, and the corresponding point-in-time when the change occurred;
classifying, by the transduction classifier, the unlabeled data, generating classified data to (i) classify a plurality of final states of monitored entities with unknown final diagnoses to identify those with disorders, and (ii) classify intermediate observations to identify the corresponding point-in-time when the change occurred;
providing an identification of the change in the time-series data and the corresponding point-in-time; and
training the predictive model for optimized performance based on the unlabeled data, the labeled data, and the classified data.

### Claim 2 (depends on 1)

The method of claim 1, wherein the transduction classifier is a maximum margin classifier having a support vector machine component and a temporal transductive component.

### Claim 3 (depends on 2)

The method of claim 2, wherein the maximum margin classifier incorporates a constraint that penalizes a violation in a chronology of events, such that the one or more monitored entities cannot go from unhealthy to healthy without an intervention.

### Claim 4 (depends on 2)

The method of claim 2, wherein the maximum margin classifier implements a stochastic gradient descent.

### Claim 5 (depends on 2)

The method of claim 2, wherein the maximum margin classifier detects the change in the time-series data by analyzing one element of the time-series data and its next subsequent element.

### Claim 6 (depends on 2)

The method of claim 2, wherein applying the maximum margin classifier to the predictive model further includes:
setting a weight vector to a value of zero;
for each element of the one or more elements of the time-series data:
selecting an element of the one or more elements of the time-series data;
updating a learning rate;
updating a gradient to account for a regularization loss;
if the weight vector is chronologically incorrect, setting a first gradient component to zero;
if the weight vector is chronologically correct, generating an update value for the first gradient component;
if a health status indicated by a corresponding predictive model member matches a known health status of the selected element, setting a second gradient component to a value of zero;
if the health status indicated by the corresponding predictive model member does not match the known health status of the selected element, generating an update value for the second gradient component;
if a true condition of the health status of the selected element is known, setting a third gradient component to a value of zero;
if a true health status of the selected element is not known, generating an update value for the third gradient component; and
updating a weight vector for the selected element based on values of the first gradient component, the second gradient component, the third gradient component and the regularization loss component.

### Claim 7

A non-transitory computer readable medium containing
computer-readable instructions stored therein for causing a control processor to perform a method of detecting status changes, and a corresponding point-in-time, in one or more monitored entities, the method comprising:
receiving, via one or more sensors, one or more signals being representative of an operational state of the one or more monitored entities;
transferring, via the one or more sensors, one or more elements of time-series data derived from the one or more signals, to the processor being communicatively coupled to the one or more sensors, the time-series data comprising both labeled and unlabeled data;
wherein the labeled data includes: initial measurements from all of the monitored entities and final measurements from monitored entities with known unhealthy states;
creating a predictive model from the time-series data in a datastore memory;
applying a transduction classifier to the predictive model;
the transduction classifier detecting a change from healthy to unhealthy in the time-series data, and the corresponding point-in-time when the change occurred;
classifying, by the transduction classifier, the unlabeled data, generating classified data to (i) classify a plurality of final states of monitored entities with unknown final diagnoses to identify those with disorders, and (ii) classify intermediate observations to identify the corresponding point-in-time when the change occurred;
providing an identification of the change in the time-series data and the corresponding point-in-time; and
training the predictive model for optimized performance based on the unlabeled data, the labeled data, and the classified data.

### Claim 8 (depends on 7)

The non-transitory computer readable medium of claim 7, further comprising computer-readable instructions stored therein to cause the control processor to apply the transduction classifier as a maximum margin classifier having a support vector machine component and a temporal transductive component.

### Claim 9 (depends on 8)

The non-transitory computer readable medium of claim 8, further comprising computer-readable instructions stored therein to instruct the control processor to perform the method, including incorporating a constraint, via the maximum margin classifier, that penalizes a violation in a chronology of events, such that the one or more monitored entities cannot go from an unhealthy state to a healthy state without an intervention.

### Claim 10 (depends on 8)

The non-transitory computer readable medium of claim 8, further comprising computer-readable instructions stored therein to cause the control processor to perform the method, including implementing a stochastic gradient descent via the maximum margin classifier.

### Claim 11 (depends on 8)

The non-transitory computer readable medium of claim 8, further comprising computer-readable instructions stored therein to cause the control processor to perform the method, including detecting the change in the time-series data by analyzing one element of the time-series data and its next subsequent element via the maximum margin classifier.

### Claim 12 (depends on 7)

The non-transitory computer readable medium of claim 7, further comprising computer-readable instructions stored therein to cause the control processor to perform the method, including:
setting a weight vector to a value of zero;
for each element of the one or more elements of the time-series data:
selecting an element of the one or more elements of the time-series data;
updating a learning rate;
updating a gradient to account for a regularization loss;
if the weight vector is chronologically incorrect, setting a first gradient component to a value of zero;
if the weight vector is chronologically correct, generating an update value for the first gradient component;
if a health status indicated by a corresponding predictive model member matches a known health status of the selected element, setting a second gradient component to a value of zero;
if the health status indicated by the corresponding predictive model member does not match the known health status of the selected element, generating an update value for the second gradient component;
if a true condition of the health status of the selected element is known, setting a third gradient component to a value of zero;
if a true health status of the selected element is not known, generating an update value for the third gradient component; and
updating a weight vector for the selected element based on values of the first gradient component, the second gradient component, the third gradient component and the regularization loss component.

### Claim 13

A method of detecting status changes, and a corresponding point-in-time, in one or more monitored entities, the method comprising:
receiving, via one or more sensors, one or more signals being representative of an operational state of the one or more monitored entities;
transferring, via the one or more sensors, one or more elements of time-series data derived from the one or more signals, to a processor being communicatively coupled to the one or more sensors the time-series data comprising both labeled and unlabeled data;
wherein the labeled data includes: initial measurements from all of the monitored entities and final measurements from monitored entities with known unhealthy states;
creating a predictive model from the time-series data in a datastore memory;
applying a transduction classifier to the predictive model; the transduction classifier detecting a change from healthy to unhealthy in the time-series data, and the corresponding point-in-time when the change occurred;
classifying, by the transduction classifier, the unlabeled data, generating classified data to (i) classify a plurality of final states of monitored entities with unknown final diagnoses to identify those with disorders, and (ii) classify intermediate observations to identify the corresponding point-in-time when the change occurred;
incrementing a chronological counter when a prediction based on a current weight vector for a selected element is chronologically incorrect;
incrementing a label counter when (i) a true health status of the selected element is known and (ii) a predicted health status does not match the true health status;
incrementing an unlabel counter when (i) the true health status of the selected element is unknown and (ii) a confidence in the predicted health status is low;
updating the weight vector using the chronological counter, the label counter, and the unlabel counter; and
training the predictive model for optimized performance based on the unlabeled data, the labeled data, and the classified data.

### Claim 14

A system for detecting status changes, and a corresponding point-in-time, in one or more monitored entities, the system comprising:
a server having a control processor and a transductive learning unit, the transductive learning unit including a max-margin temporal classifier unit;
the server in communication with at least one computing device across an electronic network, the at least one computing device having a memory with time-series data; the server configured to receive the time-series data;
the control processor configured to execute executable instructions that cause the system to perform a method of:
receiving, via one or more sensors, one or more signals being representative of an operational state of the one or more monitored entities;
transferring, via the one or more sensors, one or more elements of time-series data derived from the one or more signals, to a processor being communicatively coupled to the one or more sensors, the time-series data comprising both labeled and unlabeled data;
wherein the labeled data includes: initial measurements from all of the monitored entities and final measurements from monitored entities with known unhealthy states;
creating a predictive model from the time-series data in a datastore memory; applying a transductive classifier to the predictive model, the transductive classifier detecting a change from healthy to unhealthy in the time-series data, and the corresponding point-in-time when the change occurred;
classifying, by the transduction classifier, the unlabeled data, generating classified data to (i) classify a plurality of final states of monitored entities with unknown final diagnoses to identify those with disorders, and (ii) classify intermediate observations to identify the corresponding point-in-time when the change occurred;
providing an identification of the change in the time-series data and the corresponding point-in-time; and
training the predictive model for optimized performance based on the unlabeled data, the labeled data, and the classified data.

### Claim 15 (depends on 14)

The system of claim 14, wherein the executable instructions include instructions to cause the control processor to apply the transduction classifier as a maximum margin classifier having a support vector machine component and a temporal transductive component.

### Claim 16 (depends on 15)

The system of claim 15, wherein the executable instructions include instructions to cause the control processor to perform the method including incorporating a constraint, via the maximum margin classifier, that penalizes a violation in a chronology of events, such that the one or more monitored entities cannot go from an unhealthy state to a healthy state without an intervention.

### Claim 17 (depends on 15)

The system of claim 15, wherein the executable instructions include instructions to cause the control processor to perform the method including the maximum margin classifier implementing a stochastic gradient descent.

### Claim 18 (depends on 16)

The system of claim 16, wherein the executable instructions include instructions to cause the control processor to perform the method including the maximum margin classifier detecting the change in the time-series data by analyzing one element of the time-series data and its next subsequent element.

### Claim 19 (depends on 14)

The system of claim 14, wherein the executable instructions include instructions to cause the control processor to perform the method including:
setting a weight vector a value of zero;
for each element of the one or more elements of the time-series data:
selecting of the time-series data;
updating a learning rate;
updating a gradient to account for a regularization loss;
if the weight vector is chronologically incorrect, setting a first gradient component to zero;
if the weight vector is chronologically correct, generating an update value for the first gradient component;
if a health status indicated by a corresponding predictive model member matches a known health status of the selected element, setting a second gradient component to a value of zero;
if the health status indicated by the corresponding predictive model member does not match the known health status of the selected element, generating an update value for the second gradient component;
if a true condition of the health status of the selected element is known, setting a third gradient component to a value of zero;
if a true health status of the selected element is not known, generating an update value for the third gradient component; and
updating a weight vector for the selected element based on values of the first gradient component, the second gradient component, the third gradient component and the regularization loss component.

### Claim 20 (depends on 15)

The system of claim 15, wherein the executable instructions include flagging an error if a prediction based on a subsequent element of the time-series data indicating a healthy state after a preceding element indicated an unhealthy state.