Vladimir N. Klyachkin, Doctor of technical sciences, professor, professor of sub-department of applied mathematics and informatics, Ulyanovsk State Technical University (32 Severny Venec street, Ulyanovsk, Russia), E-mail: email@example.com
Irina N. Karpunina, Candidate of technical sciences, associate professor, associate professor of sub-department of general professional disciplines, Ulyanovsk Institute of Civil Aviation (8/8 Mozhaiskiy street, Ulyanovsk, Russia), E-mail: firstname.lastname@example.org
Background. Diagnostics of the technical condition of the hydraulic unit is carried out according to the results of vibration monitoring. The stability of the operation of a hydraulic unit largely depends on the level of its vibrations. Vibration data is sent to the data collection server through the vibration monitoring rack. Then, in real time, in the form of a discrete signal, they arrive at the control unit of the hydraulic unit, with which the load changes (with too large vibrations) or the unit stops (if the vibrations reach critical values). During vibration monitoring of the hydraulic unit, ten indicators were controlled: vibration of the lower generator bearing of the upper pool, the corresponding vibration of the upper generator bearing, the battle of the shaft of the turbine and the shaft of the generator, vibration of the cover of the turbine. Exit of vibrations beyond the permissible limits, as a rule, indicates a violation of the operability of the hydraulic unit.
Methods. Various approaches can be used to analyze vibration stability. It is possible to predict vibrations to a given horizon using time series systems. Another method is the use of machine learning for binary classification of the state of the unit according to the results of the training sample: it is stable or unstable. In the present work, statistical control of processes is used to assess the stability of a hydraulic unit. Shewhart control charts for assessing average level stability and process dispersion are used for independent indicators. Some of the controlled vibration indicators can be independent of others, however, in the general case, significant correlation relationships take place between the indicators. In this situation, multidimensional methods are used, in particular, Hotelling charts are used to control the stability of the average level of the process. Statistical process control is to identify non-random violations; in this case, the control action is applied when the vibrations are still within the tolerance, but some statistical characteristics give reason to assume that there is a nonrandom reason that will lead to an increase in the level of vibrations. The Hotelling chart does not always adequately respond to possible violations of the process: sometimes its reaction is too slow, and sometimes even the violation of the process goes unnoticed. It is shown that the way out of this situation can be the use of the method of principal components with the construction of Shewhart charts.
Results. It is shown that in some cases the charts on the principal components are more effective than the Hotelling charts, which are usually used in multidimensional statistical control. This circumstance, of course, does not exclude the use of the Hotelling charts, which in many situations shows the prompt detection of process violations. For example, a bias in two indicators was simultaneously detected by the Hotelling charts and not detected by maps on the main components. It seems advisable to develop an information system that, according to the training sample of vibration monitoring results for a particular hydraulic unit, taking into account the features of its functioning (that is, assessing which violations are dangerous for this particular object) in automatic mode, would provide operational monitoring of vibration stability, and would also give recommendations to respond to process instabilities diagnosed with control charts.