Generalized Hidden Markov Model And Its Application In Chatter Recognition

In engineering practice, two types of uncertainty are ubiquitous when modeling andobserving the physical processes. In most cases, traditional methods can only characterizevariability and randomness. However, quantify variability and epistemic uncertaintyseparately can increase the reliability of observation and modeling. In this thesis, theGeneralized Hidden Markov Model (GHMM) is proposed based on the Generalized IntervalProbability theory and the Hidden Markov Model (HMM) algorithms. Furthermore, thetoolbox of the model is established and then tested in the prediction of temperature rise offeed system bearing. A new approach in recognizing cutting states based on GeneralizedHidden Markov Model is proposed.First, research on the uncertainty theory and applications of HMM is carried out. Asuncertainty is ubiquitous in states recognition and prediction, the basic algorithms of GHMMis proposed based on the Generalized Interval Probability theory and HMM to solve theproblem.Matlab toolbox of the GHMM is established based on the basic algorithms of GHMM. Afurther study in prediction of bearing temperature rise by using the GHMM toolbox isaccomplished which verifies the effectiveness and feasibility of the algorithm and the toolbox.A milling experiment under variable conditions is carried out and the vibration signal ofthe milling spindle obtained during the experiment is used to identify the milling state, usingthe wavelet packet energy as the feature vector. The result shows that the Generalized HiddenMarkov Model is efficient in cutting state recognition under small sample data, which formsthe foundation of the prediction of cutting state transformation.