Date of Completion


Embargo Period



Machinery condition prognosis, wavelet transform, computed order tracking, particle filter, expectation maximization algorithm, regression analysis

Major Advisor

Robert X. Gao

Associate Advisor

Jiong Tang

Associate Advisor

Chengyu Cao

Associate Advisor

Horea Ilies

Associate Advisor

Richard Christenson

Field of Study

Mechanical Engineering


Doctor of Philosophy

Open Access

Campus Access


Machine condition prognosis refers to the estimation of the machine’s remaining service life, as well as the risk associated with the failure modes. Prognosis is critical to establishing intelligent maintenance strategies, increasing system reliability and productivity, minimizing costly downtime, and avoiding catastrophic damage. Being predictive in nature, prognosis in highly challenging due a number of reasons: 1) noise and the resulting low signal-to-noise ratio (SNR) in measurement reduce the overall sensitivity of detection techniques to defect identification; 2) structural complexity associated with machines reduces the effectiveness of prognosis models that are typically set up based on assumptions and simplifications; 3) redundancy in data collection reduces the efficiency in computation for prognosis; and 4) uncertainty and its propagation in the degradation process affect the accuracy and confidence of prognosis.

The presented work addresses the above challenges through research on the following three issues:

1) Development of effective feature extraction and representation techniques to reduce the signal’s complexity and improve its SNR. Specially, an optimized envelope order spectrogram technique integrating complex wavelet transform and computed order tracking with wavelet scale selection has been developed to address variability in operating conditions (such as speed) and improve reliability in defect signature extraction and representation in rotating machines and machine components such as rolling bearings; .

2) Investigation of uncertainty propagation in the degradation process based on particle filter, and development of a probabilistic framework for joint parameter estimation and state prediction which are key to machine condition prognosis. System identification, along with state prediction based on particle filter and expectation-maximization algorithm has been investigated to account for the stochastic nature of defect propagation and varying operating conditions;

3) Investigation of regression analysis to introduce predicted measurement and incorporation of the predicted measurement into the prediction stage of particle filter to update model parameters estimation for improved long term prediction accuracy in developed machinery prognosis model.

The study establishes a systematic methodology for machinery defect prognosis, and advances the theory of particle filter from predetermined model to the model with unknown parameters. The prediction accuracy of particle filter is also improved by introducing predicted measurement using regression analysis, especially in long term prediction scenario. Effectiveness of the developed method is demonstrated through numerical simulation and experimental studies, which included a run-to-failure test of a rolling bearing in a wind turbine and a tool wear degradation test of a CNC milling cutter.