Method and apparatus for predicting object properties and events using similarity-based information retrieval and modeling

• US 8,775,427 B2
• Filed: 12/04/2012
• Issued: 07/08/2014
• Est. Priority Date: 06/25/2009
• Status: Active Grant

• Current Assignee: University of Tennessee Research Foundation (University of Tennessee)
• Original Assignee: University of Tennessee Research Foundation (University of Tennessee)
• Inventors: Birdwell, J. Douglas, Wang, Tse-Wei, Icove, David J., Horn, Sally P.
• Primary Examiner(s): MORRISON, JAY A
• Application Number:
  US13/693,864
• Publication Number:
  US 20130124526A1
• Time in Patent Office: 581 Days
• Field of Search: None
• US Class Current: 707/737
• CPC Class Codes: G06F 16/2228 Indexing structures
• CPC Class Codes: G06F 16/245 Query processing
• CPC Class Codes: G06F 16/2453 Query optimisation
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06F 16/3331 Query processing
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification
• CPC Class Codes: G06F 18/2323 based on graph theory, e.g. minimum spanning trees [MST] or graph cuts
• CPC Class Codes: G06F 18/40 Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor
• CPC Class Codes: G06Q 30/0185 Product, service or business identity fraud
• CPC Class Codes: G06Q 50/265 Personal security, identity or safety
• CPC Class Codes: G06V 10/7635 based on graphs, e.g. graph cuts or spectral clustering
• CPC Class Codes: G06V 10/945 User interactive design; Environments; Toolboxes

Abstract

Method and apparatus for predicting properties of a target object comprise application of a search manager for analyzing parameters of a plurality of databases for a plurality of objects, the databases comprising an electrical, electromagnetic, acoustic spectral database (ESD), a micro-body assemblage database (MAD) and a database of image data whereby the databases store data objects containing identifying features, source information and information on site properties and context including time and frequency varying data. The method comprises application of multivariate statistical analysis and principal component analysis in combination with content-based image retrieval for providing two-dimensional attributes of three dimensional objects, for example, via preferential image segmentation using a tree of shapes and to predict further properties of objects by means of k-means clustering and related methods. By way of example, one of a machine component or process failure event, an intrusion event and a fire event and residual objects may be predicted and located and qualified such that, for example, properties of the residual objects may be qualified, for example, via black body radiation and micro-body databases including charcoal assemblages.

Claims

1. A method of determining or localizing a worn component in one of a machine and a process by periodically recording measured data in a machine or process database that supports one of a similarity-based or of a nearest neighbor search during the operation of the machine or process to form a historical record of the operation of the machine or process, receiving an input query comprising recorded measurement data for a current machine or process operating period to the machine or process database, and utilizing results returned from the machine or process database in response to the query to model component wear status of a component of the machine or process and to predict future behavior of the machine or process, the machine or process database being coupled to processor search manager apparatus, the processor search manager apparatus comprising a processor and memory, the method comprising:

receiving the input query comprising information representing recorded measurement data from a current machine or process operating period of the machine or process database;

utilizing a tree-structured index determined using multivariate statistical analysis of the historical record to retrieve stored information from the historical record of the machine or process database that is most similar to said input query information using one of a nearest neighbor and a similarity-based search;

obtaining component information of the one of the machine and the process associated with said retrieved stored information from the machine or process database; and

predicting the wear status of at least one component of the machine or process by using a model for processing said retrieved information comprising the retrieved stored information from the historical record and obtained component information associated with said retrieved stored information from the machine or process database.

2. The method of claim 1 wherein said measured data comprise one of electrical, electromagnetic, thermal, and acoustic spectra.

3. The method of claim 1 wherein said measured data comprise one of time series and frequency series data and said model is obtained by solving one of a quadratic programming optimization problem and a convex optimization problem.

4. The method of claim 1 wherein said component information comprises a record of one of a component failure and of component wear measurement data.

5. The method of claim 2 further comprising using principal component analysis to transform said spectra by projecting the spectral information onto a lower dimension subspace and to determine said tree-structured index using multivariate statistical analysis of the spectral data.

6. The method of claim 1 wherein said measured data are elements of a metric space and said tree-structured index utilizes reference points and a triangle inequality.

7. The method of claim 1 further comprising acquiring new information to add to the historical record, storing said new information to the machine or process database, and updating said tree-structured index using multivariate statistical analysis to permit retrieval of information of said new information similar to information in subsequent queries.

8. The method of claim 7 wherein said updating of said tree-structured index comprises dynamic machine learning.

9. The method of claim 1, the machine or process database further comprising a spectral property database for said machine or process, the spectral property database comprising at least one of thermal property data, acoustic property data and electromagnetic property data of one of components of the machine and steps of the process.

10. The method of claim 2 comprising determining a worn or failed component responsive to predicting the wear status of the worn or failed component.

11. The method of claim 2 wherein said machine or process database comprises first and second databases, the first database containing different machine or process properties comprising spectral properties from the second database comprising image data and retrieving stored information from both the first and second databases, retrieving from the second image database data comprising content-based image recognition.

12. The method of claim 1 further comprising predicting a future behavior of one of the machine and the process.

13. A method of predicting one of a machine failure and a process failure event via receiving an input query to search manager apparatus coupled to a plurality of databases, the processor search manager apparatus comprising a processor and memory, the method comprising:

generating at least one database of property data for a plurality of objects, an object being described by a vector of attributes;

analyzing via said processor of said search manager apparatus the attributes utilizing multivariate statistical analysis to determine a subspace of an attributes space, the subspace having at least two dimensions;

clustering via said processor the objects into groups of objects where a measure of similarity between objects comprises projecting attribute vectors onto the attributes subspace, the measure being an inverse of the Euclidean distance between points in the attributes subspace;

determining projections via said processor onto a subspace for clustering and indexing objects in said groups of objects in relation to the target object using a tree-structured index determined by multivariate statistical analysis; and

predicting the property of the target object responsive to said clustering and indexing by distance, the method further comprising predicting one of a machine failure and a process failure event.

14. The method of claim 13, the plurality of databases comprising a database of image data for said plurality of objects and using content-based image recognition to recognize a component of the machine or of the process.

15. The method of claim 13 wherein said property data comprises one of time series and frequency series data for an object.

16. The method of claim 13 wherein the attributes subspace is the object clustering and indexing subspace.

17. The method of claim 13 wherein generating at least one database of property data for a plurality of objects comprises generating temperature data over time for a machine using one of a passive microwave sensor and an infrared sensor.

18. The method of claim 13 wherein the plurality of databases comprises first and second databases, the first database comprising spectral property data for a machine or a process and the second database comprising one of manufacturer, vendor and location data for said plurality of objects, the predicting one of a machine failure and a process failure event being responsive to retrieving stored data from the first database and from the second database.

19. The method of claim 13 wherein the plurality of databases comprises a first database of spectral property data, the spectral property data comprising one of thermal property data, acoustic property data and electromagnetic property data.

20. A method of predicting one of a machine failure and a process failure event via receiving a query to search manager apparatus coupled to a plurality of databases comprising a core database of property data for objects, the processor search manager apparatus comprising a processor and memory, the method comprising:

determining a set of objects most similar to a target object via said processor search manager apparatus using a tree-structured index determined by multivariate statistical analysis,

applying at least one model to predict a property of the target object,

comparing hypothetical relationships involving the target object and a subset of a set of similar objects to the target object using one of likelihoods, probabilities and beliefs,

determining a probable relationship and a property value of the target object in relation to the set of objects, and

predicting an occurrence of one of a machine failure and process failure event responsive to said determined probable relationship and property value of the target object.

21. The method of claim 20 further comprising inferring a property value for a newly acquired object by modeling how a property value varies with other attributes of the newly acquired object.

22. The method of claim 20 wherein the plurality of databases further comprises a database of image data for objects and content-based image recognition for recognizing an object and an attribute of the object.

23. The method of claim 20 wherein said at least one model comprises a Bayesian model.

24. A method of predicting one of a machine failure and a process failure event via receiving a query to search manager apparatus coupled to a database, the processor search manager apparatus comprising a processor and memory, the method comprising:

generating at least one database of one of time series data and frequency series data comprising a historical record for each of a plurality of objects, an object being described by a vector of attributes;

analyzing via said processor of said search manager apparatus the attributes utilizing a tree-structured index determined by multivariate statistical analysis of the historical record to determine a subspace of an attributes space, the subspace having at least two dimensions;

clustering via said processor the objects into groups of objects where a measure of similarity between objects comprises projecting attribute vectors onto the attributes subspace, the measure being an inverse of the Euclidean distance between points in the attributes subspace;

determining projections via said processor onto a subspace for clustering and indexing objects in said groups of objects; and

predicting a target object property from said clustering and indexing of groups of objects where the target object is one of a machine and a process and the historical record comprises one of time series data and frequency series data for a process of the machine, the method further comprising a projection of data corresponding to a failure of a component of the machine or of a step of the machine process.

25. The method of claim 24 further comprising:

generating an impulse response model from the one of time series and frequency series data.

26. The method of claim 24 wherein said impulse response model generation comprises solving a convex optimization problem.