Mechanical monitoring label data quality guarantee method based on natural neighbor class

Abstract

The invention relates to the technical field of big data processing and mechanical fault diagnosis, in particular to a mechanical monitoring label data quality assurance method based on a natural neighbor class, which comprises the following steps: establishing a relationship between different samples of label data based on a natural neighbor graph constructed by a nonparametric algorithm, and searching for different classes through the relationship; the method comprises the following steps: cleaning label data, calculating class local outlier factors to evaluate abnormal degrees of different classes, detecting the classes with CLOF greater than a predetermined threshold value as low-quality data, performing natural neighbor graph reconstruction on the cleaned label data, detecting wrongly marked data based on the natural neighbor graph, identifying labels of unmarked data, and searching for label data with new types; according to the method, low-quality data in different label data can be effectively detected, so that the label data can be automatically identified and marked based on natural neighbor, the quality of the label data is further improved, and intelligent fault diagnosis modeling and improvement of mechanical monitoring data quality are facilitated.

Description

technical field [0001] The invention relates to the technical field of big data processing and mechanical fault diagnosis, in particular to a method for ensuring the quality of mechanical monitoring label data based on natural neighbors. Background technique [0002] Machinery condition monitoring has entered the era of big data, which also brings great opportunities and challenges. Existing technologies utilize big data for processing and mining to accurately identify machinery status or diagnose faults. However, the special properties of big data make it a challenge to process these big data for fault diagnosis. Many traditional fault diagnosis methods are constructed based on signal processing technology and are usually performed based on expert experience, so they are not suitable for processing monitoring big data. In order to solve this problem, the intelligent fault diagnosis technology based on deep learning method has been paid more and more attention by researche...

AI Technical Summary

Problems solved by technology

Existing low-quality data detection methods cannot be used for quality assurance of labeled data for the following reasons. First, there are discrepancies between data with different labels, and data of one label may be wrongly detected as low-quality data from data containing another label. quality data

Moreover, these methods cannot perform fault identification by detecting low-quality data when there are many low-quality data with similar characteristics

In addition, in actual engineering, common data with unknown labels cannot be directly used for intelligent fault diagnosis modeling, which reduces the data quality

Aiming at the shortcomings of existing low-quality data detection methods and spending a lot of effort on manually marking these data, the present invention proposes a method for automatically marking data

Images