For the reduction of emissions and combustion noise in an internal combustion diesel engine, multiple injections are normally used. A pilot injection reduces the ignition delay of the main injection and hence the combustion noise. However, normal variations of the operating conditions, component tolerances, and aging may result in the lack of combustion i.e. pilot misfire. The result is a lower indicated thermal efficiency, higher emissions, and louder combustion noise. Closed-loop combustion control techniques aim to monitor in real-time these variations and act accordingly to counteract their effect. To ensure the in-cycle controllability of the main injection, the misfire diagnosis must be performed before the start of the main injection. This paper focuses on the development and evaluation of in-cycle algorithms for the pilot misfire detection.
Based on in-cylinder pressure measurements, different approaches to the design of the detectors are compared. For non-adaptive methods, a constant threshold, direct misfire probability, and posterior misfire probability detectors are investigated. For adaptive methods, an adaptive threshold update is suggested, an adaptation of the predictive stochastic models and a sensor fusion of them is proposed to increase the detection performance.
A Scania D13 engine is used to perform the experiments under different operating conditions. The effectiveness of the algorithms is tested for different engine speeds, rail pressures, injection durations, starts of injection, EGR levels, and fuels. The results show that the observability of in-cycle pilot misfire depends on the operating conditions, and its detection can be performed successfully before the start of the main injection. With a maximum in-cycle pilot misfire observability of 98.5%, a maximum successful detection ratio of about 96% can be reached with the proposed in-cycle pilot misfire detectors. The algorithms are therefore suitable for in-cycle closed-loop combustion control feedback. By including cycle-to-cycle adaptation, the detection performance and robustness are improved significantly. The limitations are directly related to the signal-to-noise ratio of each operating condition.