Equipment failure early-warning system based on model lifecycle management

Abstract

The invention relates to an equipment failure early-warning system based on model lifecycle management. The system comprises a data preparation module, a real-time failure early-warning module, a model risk management module, a model self-learning module and a model base, wherein the data preparation module reads in and preprocesses external real-time data and transmits the processed external real-time data to the real-time failure early-warning module and the model risk management module for analysis; the real-time failure early-warning module predicts failure risks and generates early-warning information and maintenance suggests; the model risk management module evaluates reliability of model results; the model self-learning module reads in accumulated annotation samples and retrains models in the real-time failure early-warning module. The system can realize failure early-warning of equipment, online monitoring of model lifecycle and dynamic update of models, and the continuity reliability of model results is guaranteed; besides, running data and operation data are introduced simultaneously, so that early-warning errors are smaller.

Description

technical field [0001] This application relates to an equipment failure early warning system based on model-based full life cycle management. Background technique [0002] In recent years, with the popularization of the Internet and artificial intelligence technology in the field of wind power, the health status monitoring and operation and maintenance of high-value industrial equipment such as wind turbines, steam turbines, and CNC machine tools are also developing towards intelligence. Taking wind turbine equipment as an example, its fault early warning system uses equipment operation data such as SCADA (Data Acquisition and Supervisory Control System) data widely connected to wind farms to perform fault early warning and diagnosis on the health status of key components to guide predictive Improve equipment maintenance, reduce downtime accidents, and reduce operation and maintenance costs. [0003] Existing equipment failure warning systems based on SCADA data can only be...

AI Technical Summary

Problems solved by technology

[0003] Existing equipment failure warning systems based on SCADA data can only be modeled based on very limited equipment operation data and a very small number of failure labels, and the modeling is too dependent on machine learning modeling methods, resulting in models that are only available in Only within a certain range of adaptation can a correct early warning be carried out

However, the existing system lacks a quantitative assessment of the range of adaptation; and after the model goes online, it lacks the monitoring of the reliability of the early warning results of the model and the adaptive update of the model parameters

Due to the lack of these mechanisms, the accuracy of the model will decline rapidly over time after the model is launched, and it is difficult to achieve continuous and accurate early warning during the entire life cycle of the equipment

In the past, when such a problem occurred, it could only be completed by manual offline remodeling training

[0004] Some existing patents mainly focus on the modeling and analysis method of equipment failure early warning itself, but lack the design of the model's full life cycle management method and system, and the risk of the model in the system cannot be monitored online

Some existing patents have proposed methods for model self-training or automatic parameter adjustment of some sub-modules in the fault warning system, but the training model cannot be dynamically updated, and the continuous reliability of the model results cannot be guaranteed.

In addition, in the prior art, only operating parameters are included in the read equipment parameters, without taking into account the management and operating parameters, resulting in many invalid or distorted operating parameters being used as training models to generate early warning errors

Images