Systems and methods for estimating a time of an engine event

Patent metadata

Field	Value
Publication number	US-10054043-B2
Application number	US-201514680863-A
Country	US
Kind code	B2
Filing date	Apr 7, 2015
Priority date	Apr 7, 2015
Publication date	Aug 21, 2018
Grant date	Aug 21, 2018

Abstract

Systems and methods for estimating when an engine event occurs is described. The system includes a controller configured to receive a first signal from at least one knock sensor coupled to a combustion engine, receive a second signal from at least one engine crankshaft sensor coupled to the combustion engine, transform the first and second signals into a plurality of feature vectors using a multivariate transformation algorithm, determine an expected window of an engine event with a statistical model, center a segment of the plurality of feature vectors around the expected window, estimate, using the statistical algorithm, a time in the expected window corresponding to when the engine event occurred, and adjust operation of the combustion engine based on the time.

First claim

The invention claimed is:

1. A system for estimating when an engine event occurs, comprising:
at least one knock sensor coupled to a combustion engine;
at least one engine crankshaft sensor coupled to the combustion engine;
an engine control unit configured to:
receive a first signal from the at least one knock sensor;
receive a second signal from the at least one engine crankshaft sensor;
transform the first signal from the at least one knock sensor into a plurality of feature vectors using a multivariate transformation algorithm by converting the first signal into a spectrogram and transforming the spectrogram into the plurality of feature vectors;
sort, using the second signal from the at least one engine crankshaft sensor, the plurality of feature vectors into a first state and a second state, wherein the first state corresponds to a time before the engine event and the second state corresponds to a time after the engine event; and
form a first statistical model using the first state;
form a second statistical model using the second state;
combine the first statistical model and the second statistical model into a third statistical model;
determine, using the third statistical model, an expected timing window of the engine event;
center a segment of the plurality of feature vectors around the expected timing window by associating a frequency pattern indicative of the engine event with the first signal from the at least one knock sensor;
estimate, using the third statistical model, an estimated time in the expected timing window corresponding to the engine event; and
sending a signal to take corrective action based on the estimated time of the engine event.

2. The system of claim 1 , wherein the engine event lasts 100 milliseconds (“ms”) or less.

3. The system of claim 1 , wherein the engine event is a valve closure.

4. The system of claim 1 , wherein the engine event is a mechanical failure of the combustion engine.

5. The system of claim 1 , wherein the third statistical model is a Gaussian Mixture Model.

6. The system of claim 1 , wherein the third statistical model is a machine learning algorithm.

7. The system of claim 1 , wherein the multivariate transformation algorithm is a short time Fourier Transform.

8. The system of claim 1 , wherein the third statistical model estimates the time of the engine event by determining a sequence of feature vectors that comprises a maximum likelihood of the engine event occurring.

9. The system of claim 1 , wherein the third statistical model is trained offline to estimate the time in the expected window corresponding to when the engine event occurred.

10. The system of claim 1 , wherein the estimate of the time in the expected window corresponding to when the engine event occurs is within 30 degrees of the true event.

11. A method for training a controller to estimate the timing of an engine event, comprising:
receiving a signal from at least one knock sensor, wherein the signal comprises data corresponding to an engine event;
transforming the signal into a plurality of feature vectors using a multivariate transformation algorithm;
sorting the plurality of feature vectors into a first state and a second state, wherein the first state corresponds to a time before the engine event and the second state corresponds to a time after the engine event; and
forming a first statistical model using the first state;
forming a second statistical model using the second state;
combining the first statistical model and the second statistical model into a third statistical model, wherein the third statistical model is configured to predict when the engine event occurs during normal engine operation.

12. The method of claim 11 , wherein the signal is a one-dimensional, frequency signal.

13. The method of claim 11 , wherein the first statistical model is a first Gaussian Mixture Model and the second statistical model is second Gaussian Mixture Model.

14. The method of claim 13 , wherein the number of Gaussian mixtures of the first Gaussian Mixture Model and the second Gaussian Mixture Model is between 0 and 10.

15. The method of claim 13 , wherein the first state and the second state comprise between 1 and 20 feature vectors.

16. A method for estimating a time of an engine event, comprising:
receiving a first signal from at least one knock sensor coupled to a combustion engine;
receiving a second signal from at least one engine crankshaft sensor coupled to the combustion engine;
transforming the first signal from the at least one knock sensor into a plurality of feature vectors;
sorting, using the second signal from the at least one engine crankshaft sensor, the plurality of feature vectors into a first state and a second state, wherein the first state corresponds to a time fore the engine event and the second state corresponds to a time after the engine event; and
forming a first statistical algorithm using the first state;
forming a second statistical algorithm using the second state;
combining the first statistical algorithm and the second statistical algorithm into a third statistical algorithm, wherein the third statistical algorithm is configured to predict when the engine event occurs during normal engine operation;
determining, using the third statistical algorithm, an expected window of an engine event;
centering, using the third statistical algorithm, a segment of the plurality of feature vectors around the expected window of the engine event by associating a frequency pattern indicative of the engine event with the first signal from the at least one knock sensor;
estimating, using the third statistical algorithm, a time of the engine event; and
outputting a control action for at least the combustion engine based on the time of the engine event.

17. The method of claim 16 , wherein the engine event is a valve closure.

18. The method of claim 16 , wherein the engine event is a mechanical failure of the combustion engine.

19. The method of claim 16 , wherein the statistical algorithm is a Gaussian Mixture Model.

20. The method of claim 16 , wherein the control action comprises shutting down the combustion engine.

Assignees

• Gen Electric

Inventors

• Matthews Brett Alexander
• Bizub Jeffrey Jacob
• Batal Iyad
• Wheeler Frederick Wilson

Classifications

• Y02T10/40

  Engine management systems · CPC title

• F02D41/2464

  Characteristics of actuators · CPC title

• F01L1/34403

  using helically teethed sleeve or gear moving axially between crankshaft and camshaft · CPC title

• G01L23/221

  for detecting or indicating knocks in internal combustion engines · CPC title

• F02D2200/025

  Engine noise, e.g. determined by using an acoustic sensor · CPC title