Research on Reconstruction of Motor Early Fault Model Based on Large Data Lazy Learning

Abstract

In order to solve the problem of complex working environment and difficult judgment of motor fault state in the process of motor operation of rotary steering system in petroleum industry, this study uses online detection method to monitor and analyze the daily working state of rotary motor used in drill string, analyze the law of load variation, establish dynamic mathematical model, and use its transient parameters as starting point. The concept of large data is used to establish the database of influence factors of motor historical parameters, and the historical parameters are added or replaced by on-line self-learning method combined with inert learning method of adjacent rules. At the same time, the correlation degree of parameters is constantly calculated as the basis of fault threshold judgment. The concept of time axis is introduced in this process. It not only crosswise the data, but also longitudinally compares it, and realizes the data model structure of one machine and one mode by means of online continuous updating, thus solving the difference of fault judgment process formed by different working environments. Secondly, the parameters of the database are not only the prior data input in advance, but also the real-time data and processed data. The parameters of fault diagnosis are introduced into the process of load change, that is, the power spectrum of each harmonic content and other parameters. Finally, the validity and feasibility of the proposed fault detection system are proved by experiments.