State observer based analysis of crankshaft speed measurements with application to misfire detection

In this paper, a state observer based approach for monitoring fluctuations of the crankshaft angular speed and acceleration of an internal combustion engine is discussed. With this approach, irregularities in the speed or acceleration signal can be detected, which has application potential for misfire detection. The state observer used in this approach is a stationary Kalman filter that reconstructs the crankshaft speed from noise-corrupted measurements and generates estimates of the crankshaft angular acceleration. This is superior to generating acceleration from a coarse numerical differentiation of the speed signal, which would be more sensitive to high-frequency noise. Whereas in earlier Kalman-filtering approaches the underlying accelerating engine torque has been modeled as low-pass filtered white noise, a model that accounts for the periodic character of the torque is used here. Irregularities in the engine speed or acceleration signal can then be detected in the acceleration correction term of the state observer. This approach can elegantly handle missing measurements, which occur due to the instrumentation used to measure the engine speed in the car (tooth wheel with a gap). Experimental results show that even in a difficult case (large number of cylinders, high engine speed, low load) large peaks in the signal result when cylinders misfire or when cylinders resume firing.