Method for adaptive detection of engine misfire

• US 6,292,738 B1
• Filed: 01/19/2000
• Issued: 09/18/2001
• Est. Priority Date: 01/19/2000
• Status: Expired due to Fees

• Current Assignee: Ford Global Technologies, LLC (Ford Motor Company)
• Original Assignee: Ford Global Tech
• Inventors: Feldkamp, Timothy Mark, Marko, Kenneth Andrew, Prokhorov, Danll Valentinovich, Feldkamp, Lee Albert
• Primary Examiner(s): WOLFE JR, WILLIS RAY
• Application Number:
  US09/488,221
• Time in Patent Office: 608 Days
• Field of Search: 701/102, 701/103, 701/106, 701/110, 701/111, 701/114, 701/115
• US Class Current: 701/106
• CPC Class Codes: G01M 15/11 by detecting misfire

Abstract

An engine misfire detection system (10) for detecting engine misfire. System (10) includes a conventional controller (12) having a memory unit (14) and a plurality of sensors (16). Controller (12) includes a plurality of neural networks, which are trained by system (10), and which determine whether a firing event is a misfire based upon events occurring before the firing event and events occurring after the firing event. The neural networks are adaptively trained to compensate for the effects of engine variability and aging.

Claims

1. A system for determining whether a first firing event within an engine is a normal event or a misfire, said apparatus comprising:

at least one sensor which measures first engine operating data which is associated with said first firing event, and second engine operating data which is associated with at least one second firing event which occurs before said first firing event; and

a controller which is communicatively coupled to said at least one sensor and which receives said first and second engine operating data, said controller including a plurality of neural networks which are employed by said controller to determine whether said first firing event is a misfire or a normal event based upon said first engine operating data and said second engine operating data.

2. The system of claim 1 wherein said at least one second firing event comprises a plurality of firing events.

3. The system of claim 1 wherein said controller is further effective to adaptively train said plurality of neural networks based upon said determination.

4. The system of claim 1 wherein said at least one sensor further measures third engine operating data which is associated with at least one third firing event which occurs after said first firing event, and wherein said controller determines whether said firing event is a misfire or a normal event, based further upon said third engine operating data.

5. The system of claim 4 wherein said at least one third firing event comprises a plurality of firing events.

6. The system of claim 1 wherein said engine includes a plurality of cylinders, and wherein said plurality of neural networks are arranged within a plurality of sets, each of said plurality of sets corresponding to a unique one of said plurality of said cylinders.

7. The system of claim 6 wherein each of said plurality of sets comprises a plurality of neural networks which each correspond to a unique misfire state of said engine.

8. A method for determining whether an event occurring within an engine is a misfire event, said method comprising the steps of:

identifying a plurality of potential misfire states for said engine;

providing a plurality of neural networks, each of said neural networks corresponding to a unique one of said potential misfire states;

determining a first probability that each of said plurality of neural networks corresponds to a current misfire state of said engine;

determining a second probability that each of said plurality of neural networks corresponds to a misfire state of said engine which occurs before said event;

determining a third probability that each of said potential neural networks corresponds to a misfire state of said engine which occurs after said event; and

adaptively training said plurality of neural networks based upon said first, said second, and said third probabilities.

9. The method of claim 8 wherein each of said plurality of neural networks is adapted to receive an engine speed value, an engine load value, and a processed acceleration value and is further adapted to output an expected acceleration value based upon said engine speed value, said engine load value and said processed acceleration value.

10. The method of claim 8 wherein each of said plurality of neural networks is pre-trained by use of labeled data.

11. The method of claim 8 further comprising the steps of:

determining whether said event is a misf ire event based upon said first, said second, and said third probabilities.

12. The method of claim 11 wherein said determination is made by use of a misfire state transition probability matrix.

13. A method for determining if a firing event within an engine is a misfire, said method comprising the steps of:

providing a plurality of neural networks which each correspond to a potential misfire state of said engine;

calculating a plurality of first values, each of said plurality of first values being equal to the probability that a unique one of said plurality of neural networks is appropriate to describe a current misfire state of said engine;

calculating a plurality of second values, each of said plurality of second values being equal to the probability that a unique one of said plurality of neural networks is appropriate to describe a sequence of misfire states of said engine occurring before said firing event;

calculating a plurality of third values, each of said plurality of third values being equal to the probability that a unique one of said plurality of neural networks is appropriate to describe a sequence of misfire states of said engine occurring after said firing event; and

determining whether said firing event is a misfire based upon said plurality of first values, said plurality of second values and said plurality of third values.

14. The method of claim 13 wherein each of said plurality of neural networks comprises a static neural network.

15. The method of claim 14 further comprising the steps of:

acquiring values for a first, second, third and fourth engine operating condition and inputting said values for said first, second, and third engine operating condition into each of said plurality of neural networks, effective to cause each of said plurality of neural networks to generate an output value; and

comparing each of said generated output values to said value of said fourth engine operating condition.

16. The method of claim 15 further comprising the step of:

adaptively training said plurality of neural networks based upon said values of said first, second, third, and fourth engine operating condition, and said generated output values.

17. The method of claim 16, wherein said first engine operating condition is engine load, wherein said second engine operating condition is engine speed, wherein said third engine operating condition is processed acceleration, and where said fourth engine operating condition is measured acceleration.