A diagnostic method, which uses the two-sided directional power spectr
a of complex-valued engine vibration signals, is presented and tested
with four-cylinder compression and spark ignition engines for the diag
nosis of cylinder power faults. As spectral estimators, the maximum li
kelihood and FFT methods are compared, and the multi-layer neural netw
ork is employed for pattern recognition. Experimental results show tha
t the success rate for identifying the misfired cylinder is much highe
r with the use of two-sided directional power spectra than conventiona
l one-sided power spectra.