Discriminative hidden kalman filters for classification of streaming sensor data in condition monitoring

Patent metadata

Field	Value
Publication number	US-10565080-B2
Application number	US-201314406606-A
Country	US
Kind code	B2
Filing date	Jun 11, 2013
Priority date	Jun 12, 2012
Publication date	Feb 18, 2020
Grant date	Feb 18, 2020

Abstract

A method for monitoring a condition of a system or process includes acquiring sensor data from a plurality of sensors disposed within the system (S 41 and S 44 ). The acquired sensor data is streamed in real-time to a computer system (S 42 and S 44 ). A discriminative framework is applied to the streaming sensor data using the computer system (S 43 and S 45 ). The discriminative framework provides a probability value representing a probability that the sensor data is indicative of an anomaly within the system. The discriminative framework is an integration of a Kalman filter with a logistical function (S 41 ).

First claim

What is claimed is:

1. A method for monitoring a condition of a system or process, comprising:
acquiring sensor data from a plurality of sensors disposed within the system;
streaming the acquired sensor data in real-time to a computer system;
at the computer system, collecting the acquired sensor data corresponding to a current time value and a plurality of earlier time values to yield a plurality of sensor data observations;
applying, by the computer system, a discriminative function to the sensor data, the discriminative function including a merged Kalman filter and logistic function that are trained simultaneously using prior sensor data;
determining for each of the plurality of sensor data observations, by the discriminative function, a probability of malfunction or failure within the system;
if the probability indicates a system malfunction or failure, automatically initiating a remedial action comprising one or more of generating an alert, partially or fully suspending operation of the system, generating a service order for system maintenance, or generating a purchase order for replacement parts for the system.

2. The method of claim 1 , additionally including reporting on a status of the system based on the probability of a malfunction or failure within the system.

3. The method of claim 1 , wherein monitoring of the condition of the system or process is continuous.

4. The method of claim 1 , wherein the logistic function comprises the equation:

P
⁡

(

y
t

❘

u
t

,
w

)

=

1

1
+

exp
⁡

(

-

y
t

⁢

w
T

⁢

u
t

)

where y t is a class label at a given time t, where y is equal to one of two discrete values indicating failure or normal operation of the system, u t represents the plurality of hidden sensor state variables at the given time t, where the plurality of hidden sensor state variable values are real numbers having a Gaussian distribution, w is a parameter of the logistic function learned by maximizing a log likelihood function with respect to w using a set of training data, and w T is the transpose of w.

5. A method for monitoring a condition of a system or process, comprising:
receiving a stream of sensor data acquired from a plurality of sensors disposed within a system;
at the computer system, collecting sensor data corresponding to a current time value and a plurality of earlier time values from the stream of sensor data to yield a plurality of sensor data observations;
applying, by the computer system, a discriminative function to determine a plurality of hidden sensor state variables based on the plurality of sensor data observations, the discriminative function including a merged Kalman filter and logistic function that are trained simultaneously using prior sensor data;
determining a probability of a malfunction or failure within the system for each of the plurality of sensor data observations using the logistic function parameterized by the plurality of hidden sensor state variables, wherein the logistic function comprising the equation:

P
⁡

(

y
t

❘

u
t

,
w

)

=

1

1
+

exp
⁡

(

-

y
t

⁢

w
T

⁢

u
t

)

where y t is a class label at a given time t, where y is equal to one of two discrete values indicating failure or normal operation of the system, u t represents the plurality of hidden sensor state variables at the given time t, w is a parameter of a logistic function learned by maximizing a log likelihood function with respect to w using a set of training data, and w T is the transpose of w;
if the probability indicates a system malfunction or failure, automatically initiating a remedial action comprising one or more of generating an alert, partially or fully suspending operation of the system, generating a service order for system maintenance, or generating a purchase order for replacement parts for the system.

6. The method of claim 5 , wherein the plurality of hidden sensor state variable values are real numbers having a Gaussian distribution.

7. The method of claim 5 , additionally including reporting on a status of the system based on the probability of the malfunction or failure within the system.

8. A computer system comprising:
a processor; and
a non-transitory, tangible, program storage medium, readable by the computer system, embodying a program of instructions executable by the processor to perform method steps for monitoring a condition of a system or process, the method comprising:
receiving a stream of sensor data acquired from a plurality of sensors disposed within the system;
collecting sensor data corresponding to a current time value and a plurality of earlier time values from the stream of sensor data to yield a plurality of sensor data observations;
applying a discriminative function to the plurality of sensor data observations, the discriminative function including a merged Kalman filter and logistic function that are trained simultaneously using prior sensor data;
determining a probability of a malfunction or failure within the system for

Assignees

• Siemens Ag

Inventors

• Yuan Chao
• Chakraborty Amit
• Hackstein Holger
• Wiebking Leif

Classifications

• G06F11/3089 Primary

  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) · CPC title

• G06F17/18

  for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)} · CPC title

• G06F11/008 Primary

  Reliability or availability analysis · CPC title