Multi-source unlabeled data machine learning method for complex equipment performance evaluation and prediction

Abstract

The invention discloses a complex device performance evaluating and predicting method of multi-source no-label data machine learning. Standard pretreatment is conducted on multi-source no-label degradation data of the complex device at different rotating speeds and different initial degradation degrees, and a mapping relation between the multi-source no-label degradation data and system comprehensive degradation features is established; the nonlinear mapping relation between the comprehensive degradation features and the performance state of the complex device is established, model hyperparameter optimization is conducted, a non-homogeneous implicit semi-Markov model with optimized structure and parameters is obtained; aero-engine multi-source no-label degradation test data is analyzed based on the established model to obtain the initial value, the change time and the duration of each state of the aero-engine implicit performance state, and finally the evaluating of performance and thepredicting of the rest of service life are achieved. The feasible method is provided for evaluating and predicting the performance state by considering the nonlinear factors of the complex device under multi-source no-label data.

Description

technical field [0001] The invention belongs to the field of health assessment and prediction of mechanical equipment, and in particular relates to a complex equipment performance assessment and prediction method based on multi-source unlabeled data machine learning. Background technique [0002] In the past ten years, mechanical equipment health assessment and prediction technology has become a key technology for long-life, high-reliability mechanical equipment operation management, and complex equipment performance degradation assessment and prediction is an important prerequisite for equipment health management. Compared with the traditional equipment operation management technology, the health assessment and prediction technology can detect early system performance decline, can give the current and future operation health status of the equipment, and can develop the current regular equipment maintenance to condition-based maintenance, which can Therefore, research on per...

AI Technical Summary

Problems solved by technology

The supervised method relies on degradation-related prior knowledge in the construction of performance characteristics, and requires equipment health status labels in state classification, which makes this method difficult for complex equipment that lacks early degradation knowledge and health status labels in practice.

The unsupervised method based on unlabeled degraded data overcomes the dependence on degraded prior knowledge and health status labels, and has been widely used in practice. Insufficient consideration of linear dominant factors, model parameter optimization is often carried out on the premise of assuming the model structure during state identification, ignoring the optimization of the model structure, resulting in a large difference between the results of performance state evaluation and prediction and the actual situation of complex equipment

Images