Fault diagnosis method of fan rotor based on hierarchical classification algorithm

Abstract

The invention discloses a fan rotor fault diagnosis method based on a hierarchical classification algorithm, and the method comprises the following steps: obtaining original operation data of fan rotor unit equipment, and carrying out the N-A marking of the original operation data, i.e., the classification marking of a data set: employing a stratified sampling mode, selecting 80% of the original operation data of the unit as a training set, and carrying out the N-A marking of the training set; 20% of unit original operation data is used as a test set; carrying out R-H marking on the failed fan rotor unit equipment based on the duration from the observation time point of the failed fan rotor unit equipment to the failure time; and training an N-A classification model by using 80% of the N-A data set, and training an R-H classification model by using 80% of the degraded data set of the failed unit. According to the method, different weights are given to the data based on the time distance from the fault occurrence moment, the prediction accuracy of the model is evaluated, and in addition, in consideration of online continuous monitoring and prediction conditions, indexes are constructed to evaluate the prediction robustness of the model in the intervals when the equipment is in different states.

Description

technical field [0001] The invention relates to the technical field of fan rotor fault prediction, in particular to a fan rotor fault diagnosis method based on a hierarchical classification algorithm. Background technique [0002] Large rotating machinery, such as compressors, steam turbines, and gas turbines, is critical equipment in many process industries such as energy, chemical and power generation. Due to the high rotational speed and high momentum of the rotor, centrifugal force may cause the rotor parts to fly away, posing a great threat to operational safety. Early detection and prediction of potential failures can prevent catastrophic plant shutdowns and financial losses. [0003] In the prior art, the classification model evaluation indicators, such as confusion matrix and true-false accuracy, are static, and most of the model evaluation indicators are offline evaluation of the model, such as: numerical indicators, error rate, sensitivity, specificity, Accuracy ...

AI Technical Summary

Problems solved by technology

[0003] However, the classification model evaluation indicators in the prior art, such as confusion matrix, true and false accuracy, are static, and most of the model evaluation indicators are to evaluate the model offline, such as: numerical indicators, error rate, sensitivity, specificity, Accuracy, etc. and graph-based evaluation indicators and cost curves are all based on classification algorithms such as decision trees and neural networks. They learn and model from large historical data sets, and then count the performance of different models in offline test sets. Based on The "optimal" fault prediction model is selected based on the principle of error minimization. These evaluation indicators do not take into account that after the model is deployed online, it is impossible to continuously monitor and evaluate the accuracy and stability of the model prediction.

Images