Application And Research On Large-Scale Complex Electronmechanical Equipments Fault Diagnosis Based On Support Vector Regression Technology

Large-scale and Complex Mechanical equipments become more and more important in modern industry development. Condition monitoring and fault diagnosis technology of large-scale complex electromechanical system play a key role in insuring safe, reliable and healthy operation of these equipment. Correctly analyzing signal and extracting fault characteristic is the base of possible fault analysis, identification and prevention. So signal analysis technology directly effect on result of fault diagnosis. Basic proof is provided by fault model, fault diagnosis based on model is robust and broad adaptable. Credible fault diagnosis conclusion is obtained by correct model.Randomicity, many kind of physical effect coupling and simultaneous happen are characteristic of large-scale complex electromechanical equipment. Non-linear and non-stationary properties are characteristic large-scale complex electromechanical system. Because HHT is very applicable to analyze nonlinear and non-stationary signal which frequency is variable with time. HHT becomes hotspot and research direction of large-scale complex electromechanical equipment. However, end effect of HHT is an important issue in order to insure analysis precision and result. The end effect problem is outstanding in the condition of short data length of real time fault diagnosis. At present, a great deal of research is carried out in domestic and overseas. But defect lie on, Operation simpleness is the merit of wave prediction method, but this method can't completely eliminate end effect. Time series prediction method is based on traditional statistic and neuron network. The purpose of HHT is to analyze non-stationary, non-linear signal, AR or ARMA based on stationary hypothesis can't obtain excellent result. Because of Badconvergence, over fitting and generalization, end effect elimination making use of neuron network prediction is limitary.In recent years, excellent prediction precision and generalization for non-linear, non-stationary and even chaos time series is achieved by support vectors regression technology based on newly Statistical Learning Theory. It is better than neuron network time series prediction based on empirical risk minimization. The new idea of end effect elimination based on support vectors regression prediction is put forward in dissertation. End effect elimination idea based on support vectors regression prediction is put forward. Improving operation speed, time series prediction based on least square support vectors machine is suggested. LS-SVM algorithm is introduced. Simulation shows that end effect elimination can realize by LS-SVM for ordinary, short and practical vibrating signal. End effect elimination based on support vectors regression supply technique support of characteristic extraction based on HHT in large-scale complex electromechanical equipment fault diagnosis. In addition, research on sampling of HHT is carried out by author. Quantitative engineering application conclusion is acquired.Author's research fruit is applied to fault diagnosis of large-scale rolling mill abnormal vibrating. Aiming at the characteristic that abnormal vibrating take place in biting steel strip, transient non-stationary signal is analyzed by HHT time-frequency analysis. Signal characteristic difference between abnormal and normal operation is obtained by HHT time-frequency analysis technology. Proof is supplied by HHT analysis for engineering problem solving.The fault diagnosis effect Based on model of large-scale complex electromechanical equipment depends on Veracity and precision of model. Effective diagnosis result is obtained by model that reflects actually dynamic behavior of large-scale complex electromechanical equipment, simplication of large-scale complex equipment is necessary by mechanism modeling. This kind of simplication may not accord with practical condition. This is the reason that fault diagnosis according to mechanism modeling of large complex electromechanical equipment is often disabled. System identification modeling based on input and output data play more and more role in large complex device fault diagnosis. However, defect exists in many kinds identification methods. For example, least square identification methods orneuron network identification method etc. Optimization result of these methods can be obtained when trained samples are infinite. This is the reason that traditional identification method can not successful in small sample condition. It is difficult or costly to acquire Fault sample of large-scale complex electromechanical equipment. Fault modeling in the condition of limited sample is puzzle for large-scale complex electromechanical equipment. At the same time, traditional linear system identification method is not adaptive for non-linear property of realistic system. The defect that typical sample is difficult to acquire and bad model generalization is obtained from Non-linear modelmg method. For example neuron network identification.The practical problem that is provided with small sample, non-linear, high dimension, local minimum and generalization is solved by support vectors regression based on Statistical Learning Theory. Support vectors regression modeling technology is better than neuron network etc non-linear modeling method based on empirical risk minimization. Theory and technology support of large complex electromechanical equipment fault diagnosis can obtain from support vectors regression. The new idea of fault modelmg based on support vectors regression of large-scale complex electromechanical is put forward in dissertation. The key idea and contents of Statistical Learning Theory and support vectors regression algorithm based on Statistical Learning Theory are introduced. Relation between support vectors regression and system identification is put forward. System identification Algorithm based on support vectors regression and formula of parameter identification of linear system is obtained. Identification effect based on support vector regression identification using simulating data lying on Outlier and noise is verified by computer simulation. The simulation result shows that system identification based on support vectors regression is an effective method. Identification modeling based on support vectors regression in large-scale complex electromechanical equipment fault diagnosis is new idea.Combining engineering dynamic, author applies research fruit to self-exciting vibrating of rolling mill fault. Because limited sample of Fault modeling in field is utilized. System identification modeling based on support vectors regression is suggested. The generalization of model based on support vector regression identification is more excelled than modeling based on traditional least square identification method. Model based on support vectorsregression is adopted as precise fault model for fault diagnosis of hot roll mill. This model can be utilized in hydraulic screw-down system fault diagnosis. Model structure is compared in normal and abnormal vibrating rolling process. Time delay is obvious increased in vibrating rolling. System stability analysis shows that time delay increase in control system model is cause of abnormal vibration of hydraulic screw-down system. Operational deflection shape analysis between sensor shell and sensor joint pole is carried out. The conclusion that Non-Reasonable movement between sensor shell and sensor joint pole is cause of abnormal vibrating in rolling is obtained. Modeling Verification for real large-scale complex electromechanical equipment making use of support vectors regression prove that the algorithm of support vectors regression identification is promising methods in fault diagnosis modeling domain. Application foreground of support vectors regression identification modeling in large complex electromechanical equipment is expansive.Sensor shell dynamic modification is carried and scheme is adopted in field, abnormal vibration of hot rolling mill is eliminated self-vibrating of hot rolling mill is solved completely. Guarantee of Product quality and equipment and personnel safety is obtained. Perfect economic and social benefit is gained by corporation. At last summarization of research content is supplied and prospect of further research in the future is provided.Fault diagnosis orientation and elimination is realized by based on support vector regression linear system identification, non-linear system identification based on support vector regression is generic significance because of complexity of large-scale electromechanical equipment. In thesis, compute simulation of non-linear system identification is carried out. But verification is not put in practice. In addition, farther research about map between identification result and fault need develop. Farther research of parameter choice research of support vector regression in vibrating signal need develop.