Research Of The Fault Diagnosis Of Mould And Breakout Prediction System In Slab Continuous Casting

In modern steel enterprise, casting plays a connecting role in the entire production chain.Mould is one of the core components of the caster. Once the mould breakdown, it will haveserious implications for continuous casting process and lead to huge economic losses.Breakout is a major incident in the continuous casting process. To keep the occurrence of thecasting accident on minimum, equipment condition monitoring and fault diagnosis technologyis needed. Therefore, making fault diagnose for the working mould, to achieve breakoutprediction, has a great significance to the high-efficiency casting. The main contents of thisthesis are as follows:1. Breakout mechanism analysis and mould friction (MDF) monitoring principleintroductionTo find out the causes of the breakout, the mechanisms of the sticking breakout andlongitudinal breakout are analyzed. The MDF monitoring, using power method is introduced.The feature of the MDF waveform in time domain, on normal condition and fault conditionstatus, is analyzed. Then, establish the correspondence different between mould workingconditions and MDF signals2. MDF signal de-noising processing and feature extraction algorithm research.To reduce the impact of noise on the MDF signal, using adaptive wavelet thresholdalgorithm based on wavelet threshold algorithm, in MDF signal de-noising. Meanwhile, Onthe basis of the analysis of the MDF feature in time domain, using wavelet analysistechnology to extract the feature of the one-dimensional MDF waveform, on different workstatus, and pretreatment for computer intelligent recognition.3. The MDF waveform recognition algorithm researchSupport vector machine (SVM) is a machine learning method, based on statisticallearning theory, which can deal with limited samples, the pattern recognition and regressionproblems. Support vector machines using structural risk minimization principle, caneffectively solve over-learning, local minima, and dimensions curse problems. Therefore,apply SVM to recognize the MDF extracted feature.Finally, some simulation experiments are made with Matlab7.0for MDF signalde-noising, using adaptive wavelet threshold algorithm and its feature extraction algorithms,the results shows the feasibility of algorithms; at the same time, for achieving MDF waveformidentification, Establish a MDF recognition model with LibSVM, and eventually verifiedbased on the friction of the mould breakout prediction system is feasible.