Combining redundant inertial sensors to create a virtual sensor output

• US 9,354,058 B2
• Filed: 06/25/2013
• Issued: 05/31/2016
• Est. Priority Date: 11/24/2009
• Status: Active Grant

• Current Assignee: Yost Labs, Inc.
• Original Assignee: Yost Labs, Inc.
• Inventors: Yost, Paul W.
• Primary Examiner(s): Lau, Tung S
• Assistant Examiner(s): SUN, XIUQIN
• Application Number:
  US13/926,581
• Publication Number:
  US 20130311129A1
• Time in Patent Office: 1,071 Days
• Field of Search: 702/41
• US Class Current: 1/1
• CPC Class Codes: G01C 19/56 Turn-sensitive devices using vibrating masses, e.g. vibratory angular rate sensors based on Coriolis forces
• CPC Class Codes: G01C 19/5776 Signal processing not specific to any of the devices covered by groups G01C19/5607 - G01C19/5719
• CPC Class Codes: G01C 21/166 Mechanical, construction or arrangement details of inertial navigation systems
• CPC Class Codes: G01P 15/08 with conversion into electric or magnetic values
• CPC Class Codes: G01P 21/00 Testing or calibrating of apparatus or devices covered by the preceding groups

Abstract

Included are embodiments for determining an inertial quantity. One embodiment of a method includes combining readings from a plurality of inertial sensors to produce an estimate of the value of an inertial quantity in a manner that is fault-tolerant, more accurate than traditional sensor arrangements, and able to handle non-linear and non-Gaussian systems. Embodiments of a method also include utilizing a Monte Carlo estimation-based inference system to adaptively combine the inertial sensor outputs into a fault-tolerant highly-accurate inertial quantity estimate, an axis-reversed-paired physical arrangement of inertial sensors to minimize effects of environmental and process noise, and cross-associating sensors to ensure good sensor associations and reduce the effects of sample impoverishment.

Claims

1. A method of determining an inertial quantity of a single physical system with linear, non-linear, Gaussian, or non-Gaussian characteristics or combinations thereof, the method comprising:

associating a plurality of angular rate sensor devices with a substrate of the single physical system;

acquiring inertial sensor data from the plurality of angular rate sensor devices;

utilizing a computing device for processing the inertial sensor data by application of a Monte Carlo estimation-based inference system configured to produce estimates in systems with linear, non-linear, Gaussian, and non-Gaussian characteristics, wherein the computing device produces a virtual connection between at least two of the plurality of angular rate sensor devices, resulting in a virtual sensor output value, the virtual sensor output value being of a degree of accuracy exceeding that which would otherwise be attributable to one of the single angular rate sensor devices, and wherein the Monte Carlo estimation-based inference system is selected dynamically from a set of Monte Carlo estimation-based inference systems based on a set of operating and performance characteristic criteria including algorithmic execution time or accuracy; and

utilizing the virtual sensor output value as at least part of a tangible indication of the inertial quantity.

2. The method of claim 1, wherein the Monte Carlo estimation-based inference system comprises an Unscented Particle Filter, incorporating an Unscented Kalman Filter to perform predictions of time updates and measurement updates for each particle.

3. The method of claim 1, wherein the computing device comprises a firmware controlled digital processor, a programmed portion of a digital software controlled computer, a logic circuit, or an application-specific integrated circuit for performing the recited function.

4. The method of claim 1, wherein the Monte Carlo estimation-based inference system is an Unscented Particle Filter, configured to perform the following:

read initial inertial sensor data from the plurality of angular rate sensor devices;

establish an initial estimation of the initial inertial sensor data;

create a set of particles and establishing a position for each particle in the set of particles;

initialize a covariance matrix for an Unscented Kalman Filter for each particle;

execute an Unscented Particle Filter update process at least one time, comprising;

read current iteration of inertial sensor data from the plurality of angular rate sensor devices;

scatter particle positions for each particle by adding noise to them according to a probability density function which may be determined experimentally or may be based on noise characteristics of the angular rate sensor devices being used;

use an Unscented Kalman Filter to update the position for each particle further by way of prediction;

perform an importance sampling by assigning a weight to each particle based on the probability of the current iteration of inertial sensor data given that this particle'"'"'s position is treated as a sensor data element;

normalize the importance weights of all particles with respect to the total of all particle importance weights;

apply selective-importance-resampling to the particles to adjust each particle'"'"'s importance weighting by comparing each particle to the set of inertial sensor data obtained from the plurality of angular rate sensor devices;

generate the virtual sensor output value using an importance weighted sum of the particles; and

output the virtual sensor output value as a tangible indication of the inertial quantity.

5. The method of claim 1, in which the Monte Carlo estimation-based inference system is a Particle Filter, Sampling Importance Resampling Filter, Auxiliary Particle Filter, Extended Kalman Particle Filter, Condensation Particle Filter, Ensemble Kalman Filter, or Markov Chain Monte Carlo process.

6. The method of claim 1, wherein each individual angular rate sensor device of the plurality of angular rate sensor devices is a MEMS-based device.

7. The method of claim 1 wherein each individual angular rate sensor device of the plurality of angular rate sensor devices is a non-MEMS-based device.

8. The method of claim 1, wherein the plurality of angular rate sensor devices is configured as one or more axis-reversed pairs of angular rate sensor devices positioned on the substrate.

9. A method of determining an inertial quantity of a single physical system with linear, non-linear, Gaussian, or non-Gaussian characteristics or combinations thereof, the method comprising:

associating a plurality of angular rate sensor devices with a substrate of the single physical system;

acquiring inertial sensor data from the plurality of angular rate sensor devices; and

utilizing a computing device for processing the inertial sensor data by application of a Monte Carlo estimation-based inference system configured to produce estimates in systems with linear, non-linear, Gaussian, and non-Gaussian characteristics, wherein the computing device produces a virtual sensor output value from the inertial sensor data of the plurality of angular rate sensor devices, and wherein the Monte Carlo estimation-based inference system is selected dynamically from a set of Monte Carlo estimation-based inference systems based on a set of operating and performance characteristic criteria including algorithmic execution time or accuracy.

10. The method of claim 9, further comprising utilizing the virtual sensor output value as at least part of a tangible indication of the inertial quantity.

11. The method of claim 9, wherein the Monte Carlo estimation-based inference system comprises an Unscented Particle Filter, incorporating an Unscented Kalman Filter to perform predictions of time updates and measurement updates for each particle.

12. The method of claim 9, wherein the Monte Carlo estimation-based inference system is an Unscented Particle Filter, configured to perform the following:

read initial inertial sensor data from the plurality of angular rate sensor devices;

establish an initial estimation of the initial inertial sensor data;

create a set of particles and establishing a position for each particle in the set of particles;

initialize a covariance matrix for an Unscented Kalman Filter for each particle;

execute an Unscented Particle Filter update process at least one time, comprising;

read current iteration of inertial sensor data from the plurality of angular rate sensor devices;

scatter particle positions for each particle by adding noise to them according to a probability density function which may be determined experimentally or may be based on noise characteristics of the angular rate sensor devices being used;

use an Unscented Kalman Filter to update the position for each particle further by way of prediction;

perform an importance sampling by assigning a weight to each particle based on the probability of the current iteration of inertial sensor data given that this particle'"'"'s position is treated as a sensor data element;

normalize the importance weights of all particles with respect to the total of all particle importance weights;

apply selective-importance-resampling to the particles to adjust each particle'"'"'s importance weighting by comparing each particle to the set of inertial sensor data obtained from the plurality of angular rate sensor devices;

generate the virtual sensor output value using an importance weighted sum of the particles; and

output the virtual sensor output value as a tangible indication of the inertial quantity.

13. The method of claim 9, in which the Monte Carlo estimation-based inference system is a Particle Filter, Sampling Importance Resampling Filter, Auxiliary Particle Filter, Extended Kalman Particle Filter, Condensation Particle Filter, Ensemble Kalman Filter, or Markov Chain Monte Carlo process.

14. The method of claim 9, wherein each individual angular rate sensor device of the plurality of angular rate sensor devices is a MEMS-based device.

15. The method of claim 9 wherein each individual angular rate sensor device of the plurality of angular rate sensor devices is a non-MEMS-based device.

16. The method of claim 9, wherein the plurality of angular rate sensor devices is configured as one or more axis-reversed pairs of angular rate sensor devices positioned on the substrate.