Underdetermined Source Estimation And Blind Extraction Method Of Mechanical Failure Signal

Vibration signals and acoustic signals generated by rotating machinery equipment in operation contain a large number of important information for condition monitoring and fault diagnosis. The state of the current running equipment will affect the time signal characteristic parameters at anytime. The main method of condition monitoring and fault diagnosis of equipment is processing the mechanical fault signal picked up by the sensors, and we grasp the current running status of equipment according to the distribution of related characteristic parameters. Therefore, the premise and the key of successful condition monitoring and fault diagnosis are how to extract absolutely useful, objective evaluation and judgment the state of an object from the strong interference of mechanical state signals. However, in real industrial field, there are a lot of background interference noises, a variety of the unknown source signals coupling with each other. The phenomenon that the number of the sensors is less than the number of source signals prevailingly exists. Besides, the transmission process of the signals picked up by the sensor is unknown, so as a result, the observation signals are often multiple-mixed, and the fault source signals to be estimated are often mixed with other interfering signal. Valuable information can hardly be obtained from the observation fault signals. Therefore, in order to be able to accurately and efficiently extract the fault source target signals, firstly, the background noise and other interfering signals must be suppressed or excluded.Blind source separation has become a hot research topic in the field of signal processing disciplines with mechanical signal processing no exception. Especially under the condition of little prior knowledge, the blind signal processing technology can be recover or estimate the source signals from the mixed signal, which demonstrates its superiority and becomes a powerful tool to solve the composite mechanical failure signal blind source separation. However, the traditional blind signal processing method has a lot of shortcomings when they are used to recognize and extract the mechanical failure signals from the actual working condition. Therefore, the present research is supported by the National Natural Science Foundation of China and Science and technology plan projects in Yunnan province, on the basis of blind signal processing, theoretical researches and experiment testing are combined to address the problems of complex machinery vibration signal and noise signal extraction and separation in complicated working conditions, so that the underdetermined blind source estimation and blind extraction model and method are preliminarily established. A new way of thinking is offered in terms of underdetermined estimation and separation of machinery complex failure signals. The main research is as follows:(1) Proceed from engineering actual condition, the selected topic background and research significance are introduced. research status of domestic of Blind signal processing technology and mechanical fault diagnosis of dealing with the application and abroad are reviewed and the current blind signal processing technology application of existing problems in the field of fault diagnosis are summarized.(2) In view of characteristics of machinery fault signal picked up by the sensors from the industrial field, two algorithm were proposed respectively based on morphological filter combined with kernel independent component analysis algorithm and morphological filter combined with genetic simulated annealing fuzzy c-means clustering improved sparse component analysis algorithm. Simulation and experimental results show that a combination of morphological filtering technology of the two algorithms under the condition of complete can solve complex fault blind source separation, and improve the practicability of the algorithm and the reliability of the separation results. But, under the condition of underdetermined, the former is no longer working, while the latter need to be given clustering number in advance.(3) For unknown the source number, underdetermined problem in industrial field and SCA algorithm needs to given source number in advance, the mechanical signal source number estimation theory frameworthe was established. On the basis of this framework, the method based on the overall experience mode and singular value decomposition algorithm of adaptive threshold setting was proposed. Computer simulation and complex fault of bearing vibration signal source number estimation were used to verify the feasibility of this theoretical framework, research shows that the algorithm has good adaptability.(4) For background noise and underdetermined problem in industrial field, the compression perception and underdetermined blind convolution equivalence theory framework were build, based on this framework, the improved morphological filtering and frequency domain underdetermined blind extraction algorithm to reconstruct the compressed perception were put forward. The number of source signals was estimated by using the proposed source estimation algorithm; the morphology filter was used to filter out background noise; the mixing matrix was estimated by genetic simulated annealing optimization of FCM algorithm, and then the sensing matrix was constructed based on the mixed matrix; Finally the source signals was recovered by using compression perception reconstruction algorithm of orthogonal matching algorithm in frequency domain. The results from computer simulation and experiment showed that algorithms correctness and can be well separated composite fault signal.(5) In view of the existing blind solution of convolution algorithm valid on the single fault signal, but it can extract the compound fault impact signal very well, and the compression perception reconstruction algorithm for composite fault acoustic signal was no longer working. Convolution blind solution combined with frequency domain compression reconstruction algorithm was put forward, and the composite fault bearing acoustic signal was separated, and the separation results are very good.