Hidden Markov Model Optimization And Its Application In Fault Prognosis For Belt Conveyor

Belt conveyor is a fundamental transport infrastructure in coal enterprise. It playsan important role for smooth and efficient production of coal mine. How to make atimely state judgment and prediction is a valuable issue of safe production in coalenterprise. Hidden Markov Model (HMM) has shown excellent classificationperformance in various problems, such as speech recognition and machine learning.Here we make a thorough research for HMM model optimization and its applicationin fault prognosis for belt conveyor. The dissertation contains the following topics.Discrete Hidden Markov Model (DHMM) is optimized based on variable lengthparticle swarm optimization (VLPSO) algorithm. Based on VLPSO algorithm, a newDHMM training method is proposed in this dissertation. The method is well capableof optimizing the state number so that it makes model parameters convergence to theglobal optimal solution.Multi-objective particle swarm optimization (MOPSO) algorithm is used tooptimize the parameters of HMM.(1) DHMM optimized by MOPSO has a higherclassification accuracy and lower model complexity.(2) HMM optimized by MOPSOhas a higher classification accuracy and lower remaining useful life (RUL) predictiondeviation.Hidden Semi-Markov Model (HSMM) or Mixture of Gaussian Hidden MarkovModel (MG-HMM) is used to predict the RUL.(1) HSMM is built and a faultprognosis algorithm is proposed based on HSMM.(2) MG-HMM is built and thengraph-based algorithm and Chapman-Kolmogorov (C-K) equation based algorithmare used to predict the RUL.A hybrid method of MG-HMM and fixed size least square support vectorregression (FS-LSSVR) is used to make fault diagnosis and fault prognosis. Thepredication system is constructed on three parts.In summary, HMM is suitable for online and indefinite length sampleenvironment. Meanwhile, HMM contains Markov structure which could predict thestate occurrence time in the future. How to give a deeper inspection for faultprognosis of HMM and build applicable system is the future research topic.