Exhaust system and method of estimating diesel particulate filter soot loading for same using two-tier neural network

Patent metadata

Field	Value
Publication number	US-9328644-B2
Application number	US-201314034615-A
Country	US
Kind code	B2
Filing date	Sep 24, 2013
Priority date	Sep 24, 2013
Publication date	May 3, 2016
Grant date	May 3, 2016

Abstract

A method of estimating soot loading in a diesel particulate filter (DPF) in a vehicle exhaust system includes estimating an engine-out soot rate using a first neural network that has a first set of vehicle operating conditions as inputs. The method further includes estimating DPF soot loading using a second neural network that has the estimated engine-out soot rate from the first neural network and a second set of vehicle operating conditions as inputs. Estimating the engine-out soot rate and estimating the DPF soot loading are performed by an electronic controller that executes the first and the second neural networks. The method also provides for training the first and second neural networks both offline (for initial settings of the neural networks in the vehicle), and online (when the vehicle is being used by a vehicle operator). An exhaust system has a controller that implements the method.

First claim

The invention claimed is:

1. A method of estimating soot loading in a diesel particulate filter (DPF) in a vehicle exhaust system, the method comprising:
estimating an engine-out soot rate using a first neural network that has a first set of vehicle operating conditions as inputs; wherein the engine-out soot rate is for an engine in exhaust flow communication with the DPF; and
estimating DPF soot loading using a second neural network that has the estimated engine-out soot rate from the first neural network and a second set of vehicle operating conditions as inputs; wherein said estimating the engine-out soot rate and said estimating DPF soot loading are performed by an electronic controller that executes the first and the second neural networks; and wherein the first and second neural networks are stored on the electronic controller.

2. The method of claim 1 , further comprising:
training the first and the second neural networks offline by
recording the first and the second sets of vehicle operating conditions in real time at predetermined sample intervals as operating points;
distributing a periodically-determined DPF weight to respective ones of said operating points occurring since an immediately preceding periodically-determined DPF weight; wherein each periodically-determined DPF weight is determined by periodically removing, weighing, and reinstalling the DPF; and
updating nodes of the first neural network and the second neural network using the operating points as inputs to the nodes of the first and second neural networks and the distributed, periodically-determined weights as desired outputs of the second neural network.

3. The method of claim 2 , wherein said updating nodes is by back propagation.

4. The method of claim 1 , further comprising:
monitoring a pressure differential of the exhaust flow across the DPF;
training the two-tier neural network according to a training algorithm using a pressure-based model for DPF soot loading based on the monitored pressure differential; and wherein the training algorithm, and the pressure-based model is are stored algorithms executed by the electronic controller.

5. The method of claim 4 , wherein said training the first and second neural networks occurs in real time by
updating nodes of the first neural network and the second neural network based in part on a difference between the estimated DPF soot loading of the pressure-based model and the estimated DPF soot loading of the second neural network
when the engine operating conditions are within the first set of engine operating conditions; and
updating node values of the first neural network and the second neural network after a return to engine operating conditions within the first set of engine operating conditions after operation in the second set of engine operating conditions, based in part on a saved estimated soot DPF soot loading value from an operating point in the first set of engine operating conditions prior to said operation in the second set of engine operating conditions.

6. The method of claim 5 , further comprising:
measuring time of operation at each operating point during the second set of operating conditions;
calculating a total time between a last engine operating point in the first set of engine operating conditions prior to operation in the second set of engine operating conditions and a first engine operating point in the first set of engine operating conditions after a return from operation in the second set of engine operating conditions;
calculating a first difference between the estimated DPF soot loading based on the pressure-based model and the estimated DPF soot loading based on the second neural network, both measured at the first engine operating point;
calculating a second difference between the estimated DPF soot loading based on the pressure-based model and the estimated DPF soot loading based on the second neural network, both measured at the last engine operating point; wherein the estimated soot loading based on the pressure-based model at the last engine operating point is said saved estimated soot loading value;
subtracting the second difference from the first difference to provide a soot loading increment error;
dividing the soot loading increment error by the total time to provide an average total soot rate error; and
wherein said updating node values of the first neural network and the second neural network after a return to operation within the first set of engine operating conditions is by (i) distributing a respective portion of the average total soot rate error to operating points occurring in the second set of engine operating conditions in proportion to said measured time of operation at each such operating point occurring in the second set of engine operating conditions to said total time, and (ii) performing back propagation of the first and second neural networks for each distributed respective portion at each of said operating points occurring in the second set of engine operating conditions.

7. The method of claim 6 , wherein the measured time of operation at each such operating point occurring in the second set of engine operating conditions is saved in a time lookup table according to operating points within the second set of engine operating conditions, and further comprising:
resetting the time lookup table to clear the measured time following said updating after a return to engine operating conditions within the first set of engine operating conditions.

8. An exhaust system for treating exhaust from an engine on a vehicle, the exhaust system comprising:
a diesel particulate filter (DPF) in exhaust flow communication with the engine; and
a controller in operative communication with the engine and the exhaust system to determine vehicle operating conditions; wherein the controller is configured to execute:
a two-tier neural network that includes (i) a first neural network that estimates an engine-out soot rate and has a first set of vehicle operating conditions as inputs, and (ii) a second neural network that estimates DPF soot loading and has the estimated engine-out soot rate from the first neural network as an input.

9. The exhaust system of claim 8 , further comprising:
a differential pressure measurement device operatively connected to the DPF and operable to provide a signal corresponding with a pressure differential across the DPF; wherein the controller is in operative communication with the differential pressure measurement device to monitor the pressure differential;
wherein the vehicle operating conditions include engine operating conditions;
wherein the controller is further configured to execute a pressure-based model of DPF soot loading based on the pressure differential; and
wherein the controller is configured to execute a learning algorithm that trains the two-tier neural network using a DPF soot loading estimate of the pressure-based model.

10. The exhaust system of claim 9 , wherein the learning algorithm updates nodes of the two-tier neural network based in part on a difference in estimated soot loading between the pressure-based model and the second neural network (i) in real time when the engine operating conditions are within a first set of engine operating conditions, and (ii) after a return to operation within the first set of engine operating conditions after operation in a second set of engine operating conditions; wherein updating after a return to operation within the first set of engine operating conditions is based in part on a saved DPF soot loading estimate of the pressure-based model from an operating point in the first set of engine operating conditions prior to said operation in the second set

Assignees

• Gm Global Tech Operations Inc

Inventors

• Cui John X
• Sun Min
• Will Anthony B

Classifications

• F02D41/1405

  Neural network control · CPC title

• F01N11/00 Primary

  Monitoring or diagnostic devices for exhaust-gas treatment apparatus · CPC title

• F02D41/1467

  with determination means using an estimation · CPC title

• F01N2900/08

  said parameters being related to the engine · CPC title

• F02D41/1448

  the characteristics being an exhaust gas pressure · CPC title