Sensor fault detection and diagnosis for autonomous systems

• US 9,728,014 B2
• Filed: 04/21/2014
• Issued: 08/08/2017
• Est. Priority Date: 04/23/2013
• Status: Active Grant

• Current Assignee: B.G. Negev Technologies And Applications Ltd (Ben Gurion University of The Negev)
• Original Assignee: B.G. Negev Technologies And Applications Ltd (Ben Gurion University of The Negev)
• Inventors: Khalaschi, Elihau, Rokach, Lior, Kalech, Meir
• Primary Examiner(s): Lee, Tyler J
• Application Number:
  US14/257,130
• Publication Number:
  US 20160217627A1
• Time in Patent Office: 1,205 Days
• Field of Search: 701 297, 701 305
• CPC Class Codes: G05B 2219/34416 Examine, analyse sensor data for co-exclusion sets, memorize, correlate actions
• CPC Class Codes: G05B 23/0232 based on qualitative trend analysis, e.g. system evolution
• CPC Class Codes: G07C 5/02 Registering or indicating driving, working, idle, or waiting time only apparatus forming part of taximeters G07B13/00
• CPC Class Codes: G07C 5/0808 Diagnosing performance data testing of vehicles G01M17/00; testing of electrical installation on vehicles G01R31/005

Abstract

A method for detecting and diagnosing sensor faults in an autonomous system that includes sensors and hardware components, according to which sensors are related to hardware components and correlations between data readings are recognized online and correlation between sensors is determined. Predefined suspicious patterns are identified by online and continuously tracking the data readings from each sensor and detecting correlation breaks over time. The readings from sensors that match at least one of the patterns are marked as uncertain. For each online reading of the sensors, whenever sensors that used to be correlated show a different behavior, reporting that the reading indicates a fault. Upon identifying fault detection, diagnosing which of the internal components or sensors caused the fault, based on a function that returns the state of the sensor which is associated with the fault detection.

Claims

1. A method for detecting and diagnosing sensor faults in autonomous systems including sensors and hardware components, comprising:

a) relating sensors to hardware components, using a structural model;

b) consuming, on the fly, data readings from sensors;

c) recognizing, on the fly, correlations between data readings and determining correlation between sensors;

d) identifying predefined suspicious patterns, associated with a sensor state, by continuously tracking the data readings from each sensor and detecting correlation breaks over time;

e) marking the readings from sensors that match at least one of said patterns, as uncertain;

f) for each uncertain marked reading of said sensors, identifying and reporting that said reading indicates a fault whenever sensors that used to be correlated show a different behavior; and

g) upon identifying a fault, diagnosing which of the internal components and/or sensors caused said fault, by;

i) reporting a sensor that caused said fault as a faulty sensor;

ii) extracting from said structural model, the components that said faulty sensor depends on; and

iii) for each component, determining a probability of being faulty according to the number of dependent sensors of said component that are uncertain, as a ratio between a number of dependent sensors of said component that are suspected, and a total number of sensors which depend on said component.

2. The method of claim 1, further comprising determining that all of a component'"'"'s dependent sensors report faulty data whenever said component is identified as faulty.

3. The method of claim 1, wherein fault detection is determined according to:

abrupt changes in data readings;

drift changes in data readings;

stuck data readings;

scale changes in data readings.

4. The method of claim 1, further comprising:

a) storing online consumed data readings in a sliding window represented by a reading matrix;

b) upon receiving each incoming input, updating said reading matrix, while keeping the current data of the last time steps for each sensor;

c) using the data of said reading matrix to check which sensors are correlated and which sensors display predefined suspicious patterns;

d) seeking another correlated sensor that do not share component dependency but has the same state;

e) seeking an implicating sensor being a correlated sensor that does not share component dependency and has a different state; and

f) if such an implicating sensor is found, determining a failure of the uncertain sensor.

5. The method of claim 1, wherein correlation detection is performed using Pearson Correlation Coefficient calculation with respect to every two sensors.

6. The method of claim 3, wherein indication regarding a slope of a drift is provided by using linear regression.

7. The method of claim 1, wherein the autonomous system is selected from the group of:

A robot;

A flight simulator;

An unmanned vehicle.