Research On High-value Dimensionless Feature Extraction And Fault Diagnosis Method Of Petrochemical Unit Vibration Signal

The petrochemical industry is a basic industry in the national economy,which aims to improve people’s life,promote economic growth and safeguard national defense security.The modern petrochemical industry equipment is becoming more and more large-scale.In modern oil factories,the malfunction of equipment can lead to performance degradation and result in downtimes as well as financial losses and safety issues.At present,the manufacturing process and internal structure of large-scale equipment are complex,and the signal fault features extracted from their operation process present complex characteristics such as multiple coupling and ambiguity,which bring difficulty in accurate fault diagnosis of petrochemical unit.Therefore,it is of great practical significance to actively carry out the research on the fault diagnosis methods for petrochemical unit,to avoid the occurrence of malignant damage accidents and avoid unexpected downtimes.Also,it is critical to reduce financial losses and improve the reliability of the whole industrial system.In this paper,the petrochemical unit gearbox is taken as the research object.The research takes into account the problems such as many types of faults,many characterization variables,many-to-many coupling correlations of different fault parameters,and the overlapping of the single fault and composite fault vibration signals.This research studies vibration signal analysis,feature extraction and intelligent fault diagnosis methods for the petrochemical unit.The main research contents are as follows:(1)This dissertation analyzes the root causes and failure modes of gear and bearing in petrochemical unit,simulates varied gear and bearing failure states,and finally establishes fault diagnosis models.With time-domain and frequency-domain methods,the fundamental principles of non-linear and non-stationary vibration signals measured from the petrochemical unit studied.(2)Vibration signals collected from petrochemical unit are often contaminated by noise and disturbance,resulting in the difficulty in extracting salient components that are linked to failure modes.To solve this problem,this research proposes a high-value dimensionless feature extraction method for analyzing vibration signals of petrochemical unit.The wavelet de-noising model is established to de-noise vibration signals,and feature representations are extracted from vibration signals of petrochemical units with different states are extracted.The use of dimensionless indexes is suitable to express and extract high-value dimensionless features.It is suitable to solve the problem that it is difficult to distinguish between various machine health conditions because of useful fault information is buried and hidden in the complex data.(3)A fault diagnosis method based on distributed Bayesian model and neural network is proposed to solve the problems of many fault types,many characteristic variables and many-to-many coupling relations of different fault parameters.According to the high-value dimensionless characteristics of the normal state of petrochemical units,the distributed Bayesian model is established to describe the operation state of petrochemical units.Combined with neural network fault diagnosis techniques,the fault location and fault diagnosis of petrochemical units are realized.The experimental results of large-scale petrochemical units test platform demonstrate that the proposed method has improved fault diagnosis performance compared to traditional methods.(4)A composite fault diagnosis method based on the fusion of time,frequency and high-value dimensionless features is proposed for composite fault diagnosis.For multi-faults diagnosis,it is often challenging to detect fault types and severities because of overlapping issues in machine fault modes and multi-coupling,fuzziness and concealment phenomenon during the operation of petrochemical units.In this research,the proposed method fully integrates the advantages of time-domain,frequency-domain and high-value dimensionless features and applies information fusion technology,to achieve improved composite fault diagnosis based on data fusion.