System and methods for estimating emissions

1. A method for a vehicle, comprising:

acquiring measurements from at least one exhaust emission sensor of the vehicle, the at least one exhaust emission sensor positioned to measure one or more of NO x , NH 3 , and O 2 levels in exhaust gas of the vehicle, wherein the acquiring includes obtaining a signal representing an ammonia concentration in the exhaust gas and subtracting the signal representing the ammonia concentration from a total linear oxygen signal of the at least one exhaust emission sensor, and wherein subtracting the signal representing the ammonia concentration from the total linear oxygen signal includes generating a first value representing a first sum of a CO concentration, a hydrogen concentration, and a hydrocarbon concentration in the exhaust gas;

inputting the measurements from the at least one exhaust emission sensor into a machine learning model trained to output a predicted real-time amount of at least one exhaust gas constituent, the at least one exhaust gas constituent including CO and/or hydrocarbons; and

adjusting and assessing operations of an emissions aftertreatment system based on the predicted real-time amount of the at least one exhaust gas constituent.

2. The method of claim 1 , wherein the machine learning model is trained with training data, the training data including a plurality of training measurements from the at least one exhaust emission sensor and a plurality of corresponding training measurements of the at least one exhaust gas constituent, and wherein the plurality of training measurements from the at least one exhaust emission sensor include a signal representing a sum of NO x and NH 3 concentrations and/or oxygen concentration in the exhaust gas.

3. The method of claim 2 , wherein the training data is collected from exhaust gas of one or more training vehicles operated at rich, lean and stoichiometric combustion conditions.

4. The method of claim 2 , further comprising applying one or more of coefficients, weights, nodes, and hyperparameters, via the machine learning model, to the measurements input from the at least one exhaust emission sensor to output the predicted real-time amount of the at least one exhaust gas constituent.

5. The method of claim 4 , wherein training of the machine learning model includes generating a features time trace based on the plurality of training measurements from the at least one exhaust emission sensor, training measurements from additional exhaust emission sensors, and training measurements from sensors monitoring engine operating parameters of a training vehicle, and generating a target time trace based on the plurality of training measurements of the at least one exhaust gas constituent, and wherein the features time trace and the target time trace are acquired over a common duration of time.

6. The method of claim 5 , wherein the training of the machine learning model includes aligning the features time trace with the target time trace over the common duration of time and correlating variations in the plurality of training measurements of the at least one exhaust gas constituent from the target time trace to variations in the plurality of training measurements from the at least one exhaust emission sensor, the additional exhaust emission sensors, and the sensors monitoring the engine operating parameters of the training vehicle.

7. The method of claim 1 , wherein adjusting the operations of the emissions aftertreatment system based on the predicted real-time amount of the at least one exhaust gas constituent includes adjusting the operations without relying on a sensor measuring an actual amount of the at least one exhaust gas constituent of the exhaust gas.

8. The method of claim 1 , further comprising inputting measurements from sensors monitoring engine operating parameters of the vehicle to the machine learning model in addition to the measurements from the at least one exhaust emission sensor to output the predicted real-time CO concentration, and wherein the at least one exhaust emission sensor includes one or more of a UEGO sensor, a HEGO sensor, and a NO x sensor.

9. The method of claim 1 , wherein assessing the operations of the emissions aftertreatment system includes diagnosing a condition of the emissions aftertreatment system and activating a malfunction indicator light if the emissions aftertreatment system is deemed degraded.

10. The method of claim 1 , further comprising using the machine learning model to predict emissions of the at least one exhaust gas constituent in real-time across a fleet of vehicles, and wherein using the machine learning model across the fleet of vehicles includes communicatively linking the fleet of vehicles to monitor fleet-scale emissions of the at least one exhaust gas constituent.

11. An exhaust aftertreatment system of an engine, comprising:

an exhaust tailpipe;

one or more exhaust aftertreatment devices arranged along the exhaust tailpipe;

one or more exhaust sensors positioned in the exhaust tailpipe, the one or more exhaust sensors including a NO x sensor; and

a controller, configured with executable instruction stored in non-transitory memory that, when executed, cause the controller to:

acquire measurements from the NO x sensor;

apply one or more of coefficients, weights, nodes, and hyperparameters to the measurements to determine a value proportional to a sum of a concentration of at least one exhaust gas constituent concentration in exhaust gas by entering the measurements into the machine learning model trained to generate predicted real-time emissions of the at least one exhaust gas constituent; and

adjust one or more engine operating parameters based on the predicted real-time concentration of the exhaust gas constituent to maintain an actual exhaust gas constituent concentration of the exhaust gas below a threshold concentration.

12. The exhaust aftertreatment system of claim 11 , further comprising executable instructions stored in the non-transitory memory of the controller that, when executed, cause the controller to:

acquire measurements from one or more engine sensors configured to monitor engine operating parameters and from the one or more exhaust sensors to validate a training of the machine learning model.

13. The exhaust aftertreatment system of claim 11 , wherein the one or more engine operating parameters include one or more of engine temperature, engine rpm, air mass charge, fuel mass, and engine load.

14. The exhaust aftertreatment system of claim 11 , wherein the measurements from the NO x sensor includes a first data set corresponding to a sum of concentrations of NO x and NH 3 , a second data set corresponding to a binary O 2 signal, and a third data set corresponding to a linear O 2 signal, and wherein the second data set is correlated to a signal from a HEGO sensor of the one or more exhaust sensors and the third data set is correlated to a signal from a UEGO sensor of the one or more exhaust sensors.

15. A method for predicting emissions in real-time, comprising:

obtaining measurements from one or more exhaust sensors, the one or more exhaust sensors including a UEGO sensor, a HEGO sensor, and a NOx sensor;

applying one or more of coefficients, weights, nodes, and hyperparameters to the measurements to determine a value proportional to a sum of a concentration of at least one exhaust gas constituent and hydrogen concentration in exhaust gas by entering the measurements into a machine learning model trained to generate predicted real-time emissions of the at least one exhaust gas constituent; and

adjusting engine operation based on the predicted real-time emissions of the at least one exhaust gas constituent to maintain an actual concentration of the at least one exhaust gas constituent in the exhaust gas below a threshold level.

16. The method of claim 15 , further comprising collecting a data output from one or more engine sensors, the one or more engine sensors monitoring engine operating parameters, and entering the data output to the machine learning model to obtain the predicted real-time emissions of the at least one exhaust gas constituent.

17. The method of claim 16 , wherein the machine learning model is trained by generating a features time trace, wherein generating the features time trace including applying feature engineering to generate the features time trace, and wherein applying the feature engineering includes one or more of normalizing data, adding time-lagged features, adding a flag for deceleration fuel shutoff events, and applying a catalyst oxygen storage model.

18. The method of claim 17 , wherein adjusting the engine operation to maintain the actual concentration of the at least one exhaust gas constituent in the exhaust gas below the threshold level includes maintaining the predicted real-time emissions of the at least one exhaust gas constituent below 600 mg/km.