Online industrial lubrication oil health condition monitoring, diagnosis and prognostics

In this dissertation, lubrication oil degradation basic degradation features have been investigated. Lubrication oil degradation is classified into three categories: particle contamination, water contamination and oxidation which are defined as three basic degradation features. A comprehensive review of current state of the art lubrication oil condition monitoring techniques and solution has been conducted. Viscosity and dielectric constant are selected as the performance parameters to model the degradation of lubricant based on the result of the literature review. Physics models have been developed to quantify the relationship between lubricant degradation level and the performance parameters. Commercially available viscosity and dielectric sensors have been acquired and installed in a temperature controlled chamber to validate the developed performance parameter based lubrication oil deterioration physics models. Water and particle contamination are the most common oil deterioration features. Therefore, it is essential to keep monitoring the water and particle content of the lubricant. Particle filtering techniques are introduced and adapted to predict the remaining useful life of lubrication oil based on the developed physics models. In the particle filtering algorithm, state transition function was constructed to estimate the fault progression. Observation function was assembled based on the output of the sensors (physics model based on state transition function) which are viscosity and dielectric constant, respectively.;The developed prognostic methodology has been implemented into two case studies to test the effectiveness and the robustness of the developed RUL prediction algorithm. The first study is an industrial scenario simulation with progressing water contamination. The second case study is an industrial simulation with progressing iron contamination. Temperature compensation module has been integrated to smooth the prediction result. The impact of the number of observations (number of sensors implemented), particle populations have been investigated and compared.;The contributions of the research described in this dissertation are summarized as follows:;1) A comprehensive investigation and evaluation on current state of the art oil condition monitoring techniques and solutions have been conducted. The results of the investigation have showed that viscosity and dielectric constant sensors are capable of performing online oil condition analysis. This investigation is the first publication that systematically summarized and evaluated current oil condition monitoring solutions in the industry and academia, commercially available and under development.;2) Physics based models for lubrication oil performance degradation evaluation have been developed. The two most common basic degradation features: water contamination and particle contamination have been both successfully modeled and validated. Commercial available dielectric constant sensor and viscometer have been acquired and utilized in lab based simulation tests to validate the developed physics models. Most oil degradation models reported are data driven, this research is the first one that developed physics based models to describe the degradation of the lubricant and also the first one to use physics based model to perform lubrication oil remaining useful life prediction.;3) With the help of particle filtering technique, the remaining useful life prediction of lubrication oil has been successfully performed. The developed physics models have been integrated into the particle filtering framework as observation functions. The state transition function can be correlated based on previous experience and data of the system dynamics. Also within the particle filtering algorithm, an l-step ahead state parameter prediction and RUL estimator have been developed to enable this technique to perform l-step ahead prediction while most of other papers published just show one-step prediction. Therefore the developed RUL prediction technique is capable of providing practical and feasible solution to the current condition based maintenance systems. This is the first time particle filtering technique was successfully implemented to predict the remaining useful life of the lubrication oil.