Equipment state identification method based on Markov model and probability network

Abstract

The invention relates to an equipment state identification method based on a Markov model and a probability network. The state of equipment is identified using a hierarchical hidden Markov model (HHMM), and the identification result can be calculated more accurately in the form of probability. In view of the problem that the model parameters exponentially increase with the increase of equipment states, a dynamic Bayesian network is introduced to reduce the computational complexity of the model and shorten the inference time. HHMM is expressed as a dynamic Bayesian network. The health state of equipment is identified with the help of a preprocessed vibration signal. In view of the limitation of the existing state classification method, a state number optimization method based on a K-means algorithm and a cross verification method is presented. Through the state number optimization method, the stages in the process of fault development can be divided more accurately to lay a foundation for the accurate identification of equipment states. The change of the health state can be detected before functional failure, and the remaining life of equipment can be predicted using a trained model according to the change observed in the current behavior.

Description

Technical field [0001] The invention relates to an equipment early warning technology, in particular to an equipment state recognition method that integrates a Markov model and a probabilistic network. Background technique [0002] Among the annual operating costs, equipment maintenance costs usually account for more than 15%. Equipment failure causes huge losses to manufacturing companies around the world every year. Due to the increasing complexity of related equipment, the failure rate of related equipment also increases. There are many traditional fault diagnosis methods, including neural networks, expert systems, signal analysis and processing, etc. Each method has its own characteristics and has advantages and disadvantages. It will be very difficult to make a judgment through the traditional single fault diagnosis method. [0003] As the power plant has a lot of real-time and historical data, and people combine the data model with the failure mechanism through long-term e...

AI Technical Summary

Problems solved by technology

[0004] The state identification method of the traditional fault early warning model is not accurate, and the state numbers of the model are directly given by subjective experience, which lacks scientificity and versatility, resulting in passive maintenance that can only be repaired and repaired when a problem occurs

Moreover, the model parameters increase exponentially with the increase of the device state, resulting in a significant increase in model complexity and calculation time

Images