Real-Time Transient Soot and NOx Virtual Sensors for Diesel Engine Using Neuro-Fuzzy Model Tree and Orthogonal Least Squares

Abstract

Diesel engine combustion and emission formation is highly nonlinear and thus creates a challenge related to engine diagnostics and engine control with emission feedback. This paper presents a novel methodology to address the challenge and develop virtual sensing models for engine exhaust emission. These models are capable of predicting transient emissions accurately and are computationally efficient for control and optimization studies. The emission models developed in this paper belong to the family of hierarchical models, namely the “neuro-fuzzy model tree.” The approach is based on divide-and-conquer strategy, i.e., to divide a complex problem into multiple simpler subproblems, which can then be identified using a simpler class of models. Advanced experimental setup incorporating a medium duty diesel engine is used to generate training data. Fast emission analyzers for soot and NOx provide instantaneous engine-out emissions. Finally, the engine-in-the-loop is used to validate the models for predicting transient particulate mass and NOx.

Dynamic NOx emission modelling using local model networks

“…In equation (9), N defines the number of training data. Note that the particular choice of equation (9) aims at a locally WLS optimisation of the consequent parameters so that local interpretability is conserved.…”

Real-Time Simulation of Torque and Nitrogen Oxide Emissions in an 11.0 L Heavy-Duty Diesel Engine for Model-Based Combustion Control

“…Finally, in ref. [25] the authors propose hierarchical models, namely "neuro-fuzzy model trees", to predict transient particulate mass and NO x .…”

Virtual sensors for transient diesel soot and NO<inf>x</inf> emissions: Neuro-fuzzy model tree with automatic relevance determination

“…Advantage of using a deterministic signal is that it can be applied to the system multiple times and any corrupted data can be easily discarded. A Galois field (GF), defined by 11 elements and a primitive polynomial 23 () 1 4 fx x x (11) is used to create the 11-level sequence which is then mapped to 0-100. The signal is further scaled between maximum and minimum throttle command possible to the engine for a given speed.…”

“…Design of identification test signal needs some priori information about the process such as its dominant time constant, bandwidth, limitations of actuators and disturbance characteristics. This information is gathered through engine operation experience as well as preliminary tests like step, stair case excitation [23], [24], [25]. The switching time for signal is short enough to prevent capturing predominantly steady state data, but long enough to allow engine transients to fully develop before the next instance of signal is sent.…”