Large-scale multi-detector predictive modeling

• US 9,561,810 B2
• Filed: 04/30/2013
• Issued: 02/07/2017
• Est. Priority Date: 01/11/2013
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Hampapur, Arun, He, Qing, Li, Hongfei, Li, Zhiguo, Parikh, Dhaivat P.
• Primary Examiner(s): Craig, Dwin M
• Application Number:
  US13/873,859
• Publication Number:
  US 20140200869A1
• Time in Patent Office: 1,379 Days
• Field of Search: 703/6
• US Class Current: 1/1
• CPC Class Codes: B61K 9/08 Measuring installations for surveying permanent way applications of measuring apparatus or devices for track building purposes E01B35/00; measuring techniques G01
• CPC Class Codes: B61L 23/042 Track changes detection
• CPC Class Codes: B61L 23/044 Broken rails
• CPC Class Codes: B61L 23/045 Rail wear
• CPC Class Codes: B61L 23/047 Track or rail movements
• CPC Class Codes: B61L 27/53 for trackside elements or systems, e.g. trackside supervision of trackside control system conditions
• CPC Class Codes: B61L 27/60 Testing or simulation
• CPC Class Codes: G06F 30/20 Design optimisation, verification or simulation optimisation, verification or simulation of circuit designs G06F30/30
• CPC Class Codes: G06Q 10/06 Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
• CPC Class Codes: G06Q 10/0635 Risk analysis of enterprise or organisation activities

Abstract

Predicting operational changes in a multi-detector environment includes generating, via a computer processing device, a factor matrix for each univariate time series data in a set of sparse time series data collected from data sources, identifying a subset of the time series data as a feature selection based on application of a loss function, and generating a predictive model from the subset of the time series data.

Claims

1. A system for predictive modeling, the system comprising:

a computer processing device communicatively coupled to data sources operating in a railroad environment; and

logic executable by the computer processing device, the logic configured to implement a method, the method including;

generating a factor matrix for each univariate time series data in a set of sparse time series data collected from the data sources, the time series data including at least one of temperature, optical geometry, load, and acoustic data;

identifying a subset of the time series data as a feature selection based on application of a loss function;

generating a predictive model from the subset of the time series data; and

generating, from the predictive model, a prediction indicating an amount of time before a failure occurs with respect to a component of the railroad environment.

2. The system of claim 1, wherein the logic is further configured to implement:

receiving new data from the data sources;

determining a change in a failure rate based on a one-sample weighted rank test; and

upon determining the change exceeds a defined threshold value, updating the predictive model based on the change.

3. The system of claim 2, wherein the predictive model is updated using a Bayesian inference.

4. The system of claim 2, wherein the defined threshold value is selected by a user of the computer processing device.

5. The system of claim 1, wherein the data sources are detectors including at least one of:

a machine vision detector;

a wheel impact load detector;

an optical geometry detector;

a truck performance detector;

an acoustic bay detector;

a hot box detector;

a warm bearing detector;

a hot wheel detector; and

a cold wheel detector.

6. The system of claim 1, wherein the data sources are configured to capture time, physical location, and object location regarding corresponding subjects of measurement.

7. The system of claim 1, wherein the factor matrix is generated using a supervised matrix factorization technique.