Research On Gearbox Condition Estimation And Trend Prognosis Based On Virtual Sensing

As a key component in the oil and gas industry,gearbox often suffers from complex operating conditions such as variable speed and variable load,which make its degradation trend be characterized by multiple modes,nonlinearity,and uncertainty.However,traditional physical sensing methods have two serious limitations including poor dynamics and low precision,which makes it difficult to take into account the requirements of on-line monitoring and accurate estimation of mechanical degradation condition.But fortunately,virtual sensing based on intelligent inference model provides a new idea for real-time monitoring of mechanical operating status by integrating the advantages of direct sensing and indirect sensing technologies.Secondly,mechanical degradation trend is usually non-linear and non-Gaussian in its running process,which poses a great challenge for predicting RUL of the machine accurately through trend prognostic methods.Therefore,starting from two aspects of monitoring methods and trend prediction models of gearbox,research on adaptive predictive model based on virtual sensing has important theoretical significance and practical value for its condition monitoring and RUL prognosis.This article studies how to apply virtual sensing technology and PF method to mechanical condition monitoring and trend forecasting as shown below:(1)By integrating direct sensing and indirect sensing technologies,virtual sensing method based on intelligent inference models is used to construct the complex relationship between easy-to-measure on-line auxiliary variables and difficult-to-measure key variables,thereby realizing real-time monitoring the degradation condition of gearbox.Considering the characteristics of small degradation samples in the operation of gearbox,this paper proposes a virtual sensing monitoring framework based on ELM algorithm.By comparing with other data-driven models and gearbox oil observation data,the experimental results prove that the proposed sensing model can not only ensure the accuracy of condition monitoring,but also greatly reduce the time for on-line calculation,thus improving the accuracy and efficiency of gearbox early warning.(2)Traditional Bayesian inference-based trend prognosis model often assumes a linear relationship between the hidden state of the system and original measured signal in the construction of observation equation,which results in the lack of physical basis for inference information and the difficulty in judging the severity of failure.In order to solve the above problems,this paper proposes an augmented prognosis method integrating PF and virtual sensing technology.In addition,through comparison with traditional PF method based on linear measurement equation,it is validated that the augmented PF method can effectively represent the non-linear relationship between mechanical degradation states and fused observation features in a more generalized analytical form,thereby improving the accuracy of gearbox state prediction.(3)In the traditional trend prediction methods,model parameters are often pre-set according to experience knowledge.There are few studies on on-line identification of model parameters,and existing methods seldom further analyze the influencing factors and quantitative methods of the uncertainty in long-term prediction,which makes the RUL prediction less robust and accurate.Therefore,this paper proposes a nonlinear online identification method based on EM algorithm,which can update and optimize the model parameters in prognosis methods according to the latest available observations.In addition,this paper further studies the influence of parameters distribution and different predictive steps on the uncertainty in long-period prognosis,and quantifies uncertainty through confidence representation and statistical methods,thus providing supports for RUL prediction and maintenance decision analysis.