In the process of diesel engine working state detection by using the vibration acceleration signals

in order to obtain high identification rate on the premise of only extracting fewer characteristic parameters in time-domain

a state detection method by orthogonal vibration signal for reciprocating machinery is presented.Two vibration acceleration sensors are installed perpendicular to each other at the top and side of the cylinder head of the engine.The vibration signals are collected

and the time-domain characteristic parameters are extracted.When the vibration signal in one direction is considered

and when the vibration signals in both directions are considered

the corresponding characteristic parameters are used respectively as the input eigenvectors of the generalized regression neural network（GRNN）

to identify the working state of the engine under various working conditions.Experiments show that using this pair of orthogonal vibration signals perpendicular to each other is a method of reciprocating machinery state detection with great engineering application value.