Identification of fuel evaporation parameters in spark ignition engines during cold start

Control of the air-fuel ratio in spark ignition engines during throttle transients requires knowledge of the fuel evaporation dynamics. The dynamics are dependent upon temperature and undergo rapid change during cold start, the first few minutes after starting the engine. Identification of fuel evaporation dynamics parameters during this cold start period has been conducted.; Probability distribution functions of the estimates of continuous-time parameters have been derived. The distribution function of the parameter associated with the evaporation rate is a function of many variables including the sample interval, the bias of the discrete-time parameter estimates, and the variance of the discrete-time parameter estimates. Simulations show that estimates closely follow the derived density.; The fuel evaporation model contains correlated noise described by ν( k) = e(k) + a ₁e(k − 1), where a is the error associated with each measurement and a₁ is a model parameter. The a₁ parameter is associated with the fuel evaporation time constant, τ, through the nonlinear relation τ = T/ln(−a₁) where T is the sample interval. Consequently, the model parameters, and thus the noise correlation and variance, are affected by changes in the sample interval, which varies with engine speed. Two new ad hoc estimators based on least-squares estimates have been developed that significantly reduce bias over the expected range of model parameters and over the expected range of sample intervals. The ad hoc compensation algorithms are well suited to on-line estimation of time-varying systems due to their speed and accuracy properties.; The observation period required for good estimates of the fuel evaporation parameters, while supplying frequent updates, was determined to be in the range of four to six seconds. Less than four seconds resulted in large variance and increased bias. Observation periods longer than about six seconds marginally reduce the variance of the estimates but contain more old, i.e., inaccurate, data.; Estimation of model parameters was conducted on data collected during cold start tests of an existing engine. Forward simulations conducted using estimated parameters confirm the dependence on sample interval selection. The most effective sample interval is approximately one-third of the system time constant.