Artificial intelligence based health management of host system

Patent metadata

Field	Value
Publication number	US-10089203-B2
Application number	US-201615074549-A
Country	US
Kind code	B2
Filing date	Mar 18, 2016
Priority date	May 27, 2015
Publication date	Oct 2, 2018
Grant date	Oct 2, 2018

Abstract

This disclosure relates generally to health management, and more particularly to a method and system for artificial intelligence based diagnostic and prognostic health management of host systems. In an embodiment, the system includes a memory to store instructions, and a neural network controller coupled to the memory. The neural network controller is configured by the instructions to monitor a plurality of unique patterns generated in real-time. The plurality of system parameters is indicative of a system-level performance of the host system. The neural network controller is configured by the instructions to compare the plurality of unique patterns with a plurality of predetermined patterns corresponding to the plurality of system parameters to detect potential anomalies in the host system and one or more subsystems of the plurality of subsystems, where the one or more subsystems are responsible for contributing to the potential anomalies in the host system.

First claim

What is claimed is:

1. A system for artificial intelligence based diagnosis and prognosis of a host system having multiple subsystems, the system comprising:
a memory to store instructions and a plurality of predetermined patterns; and
a neural network controller coupled to the memory, wherein the neural network controller is configured by the instructions to:
monitor a plurality of unique patterns generated in real-time corresponding to a subset of multiple system parameters of the host system, wherein the system parameters comprise at least one of one or more of input parameters, control parameters, feedback parameters, and output parameters, wherein the plurality of unique patterns are indicative of the system-level performance of the host system in real-time, and each of the unique patterns is unique for a respective subsystem of the host system and only the subset of system parameters are capable of determining one or more potential anomalies in the host system, and wherein the unique patterns enable identification of faults associated with one or more subsystems without depending on physical sensors for detecting fault in the subsystem, and wherein the neural network controller is configured to
preconfigure the plurality of predetermined patterns by acquiring training data comprising the system-level performance of the host system under normal and abnormal working conditions of the one or more subsystems,
extract a plurality of feature vectors from the training data, wherein the plurality of feature vectors exhibiting the plurality of predetermined patterns indicative of the one or more potential anomalies in the host system;
detect the plurality of unique patterns generated in real-time based on the host system responses for the normal and abnormal working conditions due to the one or more subsystem failure, wherein the data acquisition enables detection and identification of abnormal response by enabling distinguishing the normal and abnormal performance of the host system based on domain experience and training associated with operational conditions to further identify a failed component/subsystem of the host system that leads to abnormality in the system response; and
compare the plurality of unique patterns with the plurality of predetermined patterns corresponding to the subset of multiple system parameters to detect one or more potential anomalies in the host system and at least one faulty subsystem of the plurality of subsystems based on the comparison, the at least one faulty subsystem is responsible for contributing to the one or more potential anomalies in the host system.

2. The system of claim 1 , wherein the neural network controller is further configured by the instructions to train a neural network model based on the plurality of feature vectors to classify the one or more potential anomalies with the at least one faulty subsystem responsible for contributing to the one or more potential anomalies.

3. The system of claim 1 , wherein to acquire the training data, the neural network controller is further configured by the instructions to:
simulate the normal working condition and the plurality of abnormal working conditions of the plurality of subsystems; and
generate the system-level performance under the normal and the plurality of abnormal working conditions.

4. The system of claim 3 , wherein to simulate the normal and the plurality of abnormal working conditions, the neural network controller is further configured by the instructions to:
capture the system-level performance from a reference model and a fault introduced model continuously in a specified window of time scale, wherein the reference model comprises modeling of the normal working condition and the fault introduced model comprises modeling of the plurality of abnormal working conditions of the plurality of subsystems; and
preprocess the system-level responses to remove trends in training data.

5. The system of claim 1 , wherein the host system comprises one of an aircraft system, an automotive system, a turbine system and an engine system.

6. A processor-implemented method for artificial intelligence based diagnosis and prognosis of a host system having multiple subsystems, the method comprising:
monitoring, by a neural network controller, a plurality of unique patterns generated in real-time corresponding to a subset of multiple system parameters of the host system, wherein the system parameters comprise at least one of one or more of input parameters, control parameters, feedback parameters, and output parameters, the plurality of unique patterns are indicative of the system-level performance of the host system in real-time, and each of the unique patterns is unique for a respective subsystem of the host system and only the subset of system parameters are capable of determining one or more potential anomalies in the host system, and wherein the unique patterns enable identification of faults associated with one or more subsystems without depending on physical sensors for detecting fault in the subsystem, and wherein the neural network controller is configured to
preconfigure the plurality of predetermined patterns by acquiring training data comprising the system-level performance of the host system under normal and abnormal working conditions of the one or more subsystems; and
extract a plurality of feature vectors from the training data, wherein the plurality of feature vectors exhibiting the plurality of predetermined patterns indicative of the one or more potential anomalies in the host system, and wherein the data acquisition enables detection and identification of abnormal working condition by enabling distinguishing the normal and abnormal performance of the host system based on domain experience and training associated with operational conditions to further identify a failed component/subsystem of the host system that leads to abnormality in the system response;
detect the plurality of unique patterns generated in real-time based on host system responses for various normal and abnormal performance scenarios due to one or more sub-system failures; and
comparing, by the neural network controller, the plurality of unique patterns with the plurality of predetermined patterns corresponding to the subset of multiple system parameters for detecting, by the neural network controller, one or more potential anomalies in the host system and at least one faulty subsystem of the plurality of subsystems based on the comparison, the at least one faulty subsystem responsible for contributing to the one or more potential anomalies in the host system.

7. The method of claim 6 , further comprising training a neural network model based on the plurality of feature vectors to classify the one or more potential anomalies with the at least one faulty subsystem responsible for contributing to the one or more potential anomalies.

8. The method of claim 6 , wherein acquiring the training data comprises:
simulating the normal working condition and the plurality of abnormal working conditions of the plurality of subsystems; and
generating the system-level performance under the normal and the plurality of abnormal working conditions.

9. The method of claim 8 , wherein simulating the normal working condition of the plurality of sub-systems and the plurality of abnormal working conditions of the one or more subsystems comprises:
capturing the system-level performance from a reference model and a fault introduced model continuously in a specified window of time scale, wherein the reference model comprises modeling of the normal working condition and the fault introduced model comprises modeling of the plurality of abnormal working conditions of the plurality of sub

Assignees

• Tata Consultancy Services Ltd

Inventors

• Mohan Sreeram
• Rao Pallanti Srinivasa
• Chindam Venkateshwar

Classifications

• G06N99/005

  Physics · mapped topic

• G05B23/0254 Primary

  based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks · CPC title

• G06N3/088

  Non-supervised learning, e.g. competitive learning · CPC title

• G06F11/2257 Primary

  using expert systems · CPC title

• G06N20/00

  Machine learning · CPC title