Machine learning for misfire detection in a dynamic firing level modulation controlled engine of a vehicle

Patent metadata

Field	Value
Publication number	US-2021003088-A1
Application number	US-202017026706-A
Country	US
Kind code	A1
Filing date	Sep 21, 2020
Priority date	Nov 14, 2017
Publication date	Jan 7, 2021
Grant date	—

Abstract

Using machine learning for cylinder misfire detection in a dynamic firing level modulation controlled internal combustion engine is described. In a classification embodiment, cylinder misfires are differentiated from intentional skips based on a measured exhaust manifold pressure. In a regressive model embodiment, the measured exhaust manifold pressure is compared to a predicted exhaust manifold pressure generated by neural network in response to one or more inputs indicative of the operation of the vehicle. Based on the comparison, a prediction is made if a misfire has occurred or not. In yet other alternative embodiment, angular crank acceleration is used as well for misfire detection.

First claim

What is claimed is:

1 . A vehicle, comprising:
an internal combustion engine having a plurality of pistons operating within a plurality of cylinders respectively; a dynamic firing level modulation module arranged to operate the internal combustion engine in a dynamic firing level modulation mode; an exhaust manifold fluidly coupled to outputs of the plurality of pistons and arrange to provide exhaust gases from the plurality of cylinders to an aftertreatment system; and a machine learning module arranged to: (a) receive a measured exhaust manifold pressure signal indicative of the pressure in the exhaust manifold; and (b) detect a misfire of one of the cylinders while operating in the dynamic firing level modulation mode, the machine learning module arranged to detect the misfire of the one cylinder by learning to differentiate between the intentional skipping or modulation of the one cylinder versus an actual misfire of the one cylinder at least partially based on the received measured exhaust manifold pressure signal indicative of the pressure in the exhaust manifold.

2 . The vehicle of claim 1 , wherein the machine learning module includes a neural network arranged to rely on a first distribution model for exhaust manifold pressure readings for successful cylinder firings and a second distribution model for exhaust manifold pressure readings for successful cylinder skips.

3 . The vehicle of claim 1 , wherein the neural network includes a plurality of hidden layers, each of the hidden layers includes one or more processors.

4 . The vehicle of claim 1 , wherein the machine learning module includes a neural network arranged to generate a misfire flag in response to receipt of a measured exhaust manifold pressure that falls outside a distribution range for either a successful fire or a successful skip.

5 . The vehicle of claim 1 , wherein the firing level modulation mode is a Dynamic Skip Fire (DSF) mode, wherein for a given reduced effective displacement that is less than full displacement of the internal combustion engine, a select cylinder is fired, skipped and selectively either fired or skipped in successive working cycles.

6 . The vehicle of claim 1 , wherein the machine learning module includes a neural network which is arranged to receive one or more inputs indicative of operation of the vehicle, the one or more inputs selected from the group including: spark timing; fuel mass per cylinder; fire skip status; fire enable flag; cylinder skip number; order skip number; mass air per cylinder; cam phase timing; charge air temperature; engine speed; manifold absolute pressure; transmission gear; Deceleration Cylinder Cut-Off (DCCO) exit; vehicle speed; torque request; pedal position; fuel pressure; and turbocharger waste gate position.

7 . The vehicle of claim 1 , further comprising a misfire counter arranged to count a plurality of misfires as determined by the machine learning module and to generate a notification when the plurality of misfires exceeds a threshold value.

8 . The vehicle of claim 1 , wherein the firing level modulation mode is a dynamic multi-charge level mode where all cylinders of the internal combustion engine are fired, but individual working cycles are operated at different output levels by using different air charge and/or fueling levels.

9 . The vehicle of claim 1 , wherein the firing level modulation mode is a Dynamic Skip Fire (DSF) mode wherein the plurality of cylinders are fired and skipped in a predefined rolling pattern.

10 . A method for controlling an internal combustion engine, the method comprising:
operating cylinders of the internal combustion engine in a skip fire mode such that first firing opportunities of the cylinders are command to be fired while second firing opportunities of the cylinders are commanded to be not fired and intentionally skipped; measuring an exhaust manifold pressure of an exhaust manifold fluidly coupled to the cylinders of the internal combustion engine; and using artificial intelligence to differentiate between (a) misfires of the plurality of cylinders that are commanded to be fired and (b) the firing of the cylinders commanded to be not fired, the differentiation for (a) and (b) at least partially based on a comparison of the measured exhaust manifold pressure for cylinder events with a fire distribution model that defines exhaust manifold pressure distribution ranges for successful and not successful fires and a skip distribution model that defines exhaust manifold pressure distribution ranges for successful and not successful skips.

11 . The method of claim 10 , wherein using artificial intelligence further comprises:
receiving at the neural network one or more inputs indicative of operation of the vehicle while operating in the skip fire mode; using the neural network to predict an exhaust manifold pressure for a cylinder event in response to the one or more inputs; and determining when if a fault occurred with the cylinder event by comparing the measured exhaust manifold pressure with the predicted manifold pressure for the cylinder event.

12 . The method of claim 10 , further comprises generating a misfire flag for unsuccessful cylinder events.

13 . A system, comprising:
an internal combustion engine having a plurality of cylinders; a skip fire engine controller arranged to operate the cylinders of the internal combustion engine in a skip fire manner, the skip fire operation involving firing the cylinders during some working cycles and skipping the cylinders during other working cycles; a storage unit arranged to store:
a first model of exhaust pressures indicative of successful firings of the cylinders of the internal combustion engine; and
a second model of exhaust pressures indicative of successful skips of the cylinders of the internal combustion engine; and
a neural network arranged to generate fault signals for working cycles of the cylinders that were either unsuccessfully fired or unsuccessfully skipped by comparing a measured exhaust pressure with (a) the first model for fire commands and (b) the second model for skip commands.

14 . The system of claim 13 , wherein the first model includes:
a first distribution range of exhaust pressures for successful firings; a second distribution range of exhaust pressures for unsuccessful firings; and a threshold exhaust pressure between the first distribution range and the second distribution range.

15 . The system of claim 14 , wherein the neural network makes a decision to generate a fault signal for a working cycle of a cylinder that unsuccessfully fired if the measured exhaust pressure for the working cycle falls below the threshold.

16 . The system of claim 13 , wherein the second model includes:
a first distribution range of exhaust pressures for successful skips; a second distribution range of exhaust pressures for unsuccessful skips; and a threshold exhaust pressure between the first distribution range and the second distribution range.

17 . The system of claim 16 , wherein the fault detection system makes a decision to generate a fault signal for a working cycle of a cylinder that unsuccessfully skipped if the measured exhaust pressure for the working cycle is above the threshold.

18 . The system of claim 13 , wherein the first model and the second model are maintained in storage locations accessible by the neural network.

19 . The system of claim 13 , wherein the first model and the second model are constructed from empirical data collected from

Assignees

• Tula Technology Inc

Inventors

• Chen Shikui Kevin
• Mandal Aditya
• Chien Li-Chun
• Ortiz-Soto Elliott

Classifications

• G06N3/048

  Activation functions · CPC title

• G06N3/084

  Backpropagation, e.g. using gradient descent · CPC title

• F02D2200/1015

  Engines misfires · CPC title

• F02D2200/1012

  Engine speed gradient · CPC title

• F02D41/1498

  measuring engine roughness · CPC title