Model based Online Fault Diagnosis of Automotive Engines using Joint State and Parameter Estimation

Nadeer E P, Amit Patra, and Siddhartha Mukhopadhyay
Submission Type: 
Full Paper
phmc_15_006.pdf801.14 KBAugust 4, 2015 - 10:36pm

A four-stroke automotive engine system, due to its reciprocating nature, is a hybrid nonlinear system. An instantaneous physics-based model capturing the within cycle dynamics of such a system, as contrasted with the mean value models, has the advantage of being able to detect, isolate and identify small faults within shorter time duration. This, however, comes at greater computational effort. In this paper, an Extended Kalman Filter (EKF) based tunable diagnoser, which uses a minimal hybrid nonlinear state space model of the automotive engine, is used for the detection and isolation of a variety of engine system faults including intake manifold leak, injector fault and exhaust manifold leak. The state estimates and innovation sequences from the EKF based estimator are shown to be sufficient for the detection and isolation of the faults under consideration. Once a fault is detected and isolated, the diagnoser could be tuned online to perform fault identification by redefining a model/fault parameter as an additional state to be estimated, and then performing a joint state and parameter estimation. This process is demonstrated for the identification of intake manifold leak and the leak area is found for small throttle inputs. The engine model that the EKF estimator uses is a continuous-discrete one, i.e., the state transition equations derived from mass and energy balance are continuous and the measurement model is discrete. The model and diagnoser are implemented in SimulinkTM and is validated against an AMESimTM model of the engine. The parameters of the model are properties of air, burnt gas and fuel as well as the geometrical parameters of the engine like the volumes of storage elements (intake manifold, cylinder and exhaust manifold) and areas of valves. The model takes the fuel control and exhaust gas recirculation (EGR) control signals from the engine control unit (ECU) as inputs. In addition, the measurements of throttle position and crankshaft position โ€“ from which engine speed and crank angle can be extracted โ€“ are also considered as model inputs. The engine states are pressures (or temperatures) inside the individual storage elements and masses of individual gas species (air, burnt gases and fuel). The sensor measurements used for estimation are mass air flow (MAF), intake manifold pressure, intake manifold temperature and exhaust manifold pressure. For the nominal engine model, the EKF estimator performance is compared with two other computationally more expensive nonlinear estimators, namely the Unscented Kalman Filter (UKF) and Rao-Blackwell Particle Filter (RBPF).

Publication Year: 
Publication Volume: 
Publication Control Number: 
Submission Topic Areas: 
Model-based methods for fault detection, diagnostics, and prognosis
Submitted by: 

follow us

PHM Society on Facebook Follow PHM Society on Twitter PHM Society on LinkedIn PHM Society RSS News Feed