| 摘要 |
FIELD: engines and pumps. ^ SUBSTANCE: reference characteristics of controlled parameters are determined on the tested engine in its serviceable condition, for example during delivery-acceptance tests in the form of mathematical relations; reference N time interval is set; subsequent recording of each parameter is performed with the chosen interval within the limits of reference N time interval. Parameter recording for the next diagnostic control is performed beyond the limits of this time interval, and after the deviations of each of those parameters from reference characteristic are determined, values of deviations are smoothed. Instruction selection for Kohonen neuron net is formed from data of reference N time interval; for that purpose, there formed are sets of reference sequences of m series smoothed values of parameter deviations in each one; at that, the first sequence of each set is started to be formed from the first smoothed value of deviation, and the next ones - by means of shift to one value; statistical characteristics are determined for each of the formed sequences; parameters of reference sequence are normalised relative to statistical characteristics for each set of reference sequences, and additional sequences are formed. Vector of input parameters for neuron net is formed from reference and additional sequences, as well as statistical characteristics of reference sequence, and after processing of the obtained data, during which the instruction and clusterisation of Kohonen neuron net is being performed, a list of neurons, which characterises classes of the engine state, is composed. Similarly, set of sequences is formed of deviations of parameter values for diagnostic control from reference characteristics; after cluster analysis of set of sequences and after selection of neuron with maximum value of output signal there compared is this neuron with a list of neurons, which characterises engine state classes. Conclusion on the absence of changes in the technical state of the engine is made at availability of the chosen neuron in the composed list. ^ EFFECT: enlarging technological capabilities of the method; detecting failures at early stages and dynamic monitoring of technical state of the engine. ^ 6 dwg |
