Research On Weak Signal Detection In Incipient Fault Prognosis Of Mechanical Equipment

Incipient fault of mechanical equipment contains two meanings: one means early fault, faint fault or latent fault, another means that a kind of fault is the early stage of the other kind. It is helpful for the equipment's reliably working if we detect the fault in its early stage. Because the feature of the incipient fault is weak and usually submerged in heavy noise, it is difficult to be extracted. Aiming at mechanical equipment, this dissertation focuses on weak signal detection and practical diagnosis techniques in incipient fault prognosis.Traditional adiabatic elimination stochastic resonance (SR) in small parameters is not adapt to engineering weak signal detection in large parameters, so a new numerical method called the step-changed SR is proposed. The properties of approximate entropy (ApEn) in signal complexity measure is analysed, and a novel adaptive SR method based on ApEn measurement is presented. It can solve the problem of parameter adjustment in SR. The successful application of vibration analysis of metal cutting and fault diagnosis of rolling bearings show the methods'efficiency.Weak periodic signals can be detected by identifying the transformation of the chaotic oscillator from the chaotic state to the large-scale periodic state when the external signal is applied. We usually judge the change of chaotic oscillator only by our eyes, and there is not an objective criterion. Two-dimensional ApEn proposed in this dissertation has been proved to be an effective measure of the states of chaotic oscillator. A new weak signal detection method based on chaotic oscillator and two-dimensional ApEn is presented. Satisfactory results have been achieved when using this method to the rotating machinery condition monitoring and rolling bearings fault diagnosis.Useless components in engineering signal usually lead weak feature extraction to be difficult. Independent component analysis (ICA) is an effective weak signal detection method, and it can separate the source component which is statistically independent from the mixed signals. But the capacity of ICA is usually effected by the phase difference and noise of the mixed signals. For this reason, an improved method called frequency domain blind source separation (FDBSS) is proposed. Successful applications of FDBSS are achieved in the detection of eddy-current sensor failure and the diagnosis of incipient impact-rub fault. The results show that FDBSS has widely prospect for application in the condition monitoring and fault diagnosis of mechanical equipment.Support vector data description (SVDD) is a new one-class classification method. It can build a classifier with only one class data (or normal samples). So the application of SVDD to the machine fault diagnosis is expected to solve the problem of the shortage of incipient fault data in intelligent diagnosis. A hybrid intelligent prognosis method based on empirical mode decomposition and SVDD is proposed, and it is applied to fault diagnosis of rolling bearings and gearbox. The results show that the presented method is efficient to extract the fault feature, reduce the dimension of the signals and improve the veracity of one-class classification in intelligent diagnosis significantly.As the carrier of the key technology, some practical techniques of the development of remote monitoring and diagnosis system based on LabVIEW are summarized. A simple effective modification algorithm for the vibration signals integration in frequency domain is presented. This algorithm provides the technical support for the integrality and accuracy of equipment dynamic information.