Research On The Source Separation And De-noising Of The Train Bearing Wayside Acoustic Fault Signal

Diagnosis first came up in the medical field. With the developing of machine maintenance technique, equipment fault diagnosis arises in different areas of mechanical engineering. Rotating machine (bearing, gear. etc) plays a significant role in the mechanical field. While they can go out of order more easily than the other machine constructions. So it is much important to implement monitoring and diagnosis on such components. In China, railway transportation shows its indispensable role in national economy and modernization construction. The monitoring and diagnosis of train bearings also attracts much attention, as it’s the main cause of the train fault and even the derail accident. To monitor the train bearing conditions, we need to collect and analyze the vibration or acoustic signal, which can imply the inner fault information of the bearings. Among all the diagnosis strategies, TADS (Trackside Acoustic Detector System) owns the best prospect and superiority for its feature of non-contact measurement. In this dissertation, we focus on the bearings of active passenger train with type of NJ(P)3226X1 and signal processing strategy of TADS. The key points are the source separation under Doppler distortion and weak signal detection with heavy background noise for the train bearings wayside acoustic monitoring and fault diagnosis.First, the structure of antifriction bearing, fault characteristic frequency and main failure mode are introduced. The vibration mechanism is also expounded. Then we analyze the geometric model of the TADS. A simulative experimental platform is designed for the wayside acoustic train bearing fault signal. This plat is composed of two sub-plats. One is the statical bearing signal collecting system. In this system, the artificial fault is set on the bearing by wire-electrode cutting for the corresponding acoustic fault signal. The other one is the dynamic bearing signal collecting system. Both the single bearing or signal fault dynamic experiment and multi-source dynamic experiment can be implemented. The signals with similar features to the actual wayside train bearing fault signal are finally acquired for the further analysis.Subsequently, we study the multi-source separation of the wayside train bearing fault signal. As the speciality of this kind of signal, this dissertation takes advantage of the feature of Doppler distortion, which makes the signals of different sources separated from each other in the time-frequency domain. Two separation methods are proposed. The first one is based on a pseudo time-frequency analysis (PTFA). In its a pseudo time- frequency distribution (PTFD), the time and frequency centers can be extracted for different sources. Then the Dopplerlet filters (DF) are utilized to separate each signal one by one. The second method refers to a time-frequency data fusion (TFDF) strategy and a Doppler feature matching search (DFMS) algorithm for the signal parameters extracting. With the determined time centers, time-frequency filters (TFF) are designed to separate the acoustic signals in the time-frequency domain. The methods are verified with both simulated and experimental signals for the signal separation of different sources.Then the Doppler distortion and acoustic theory are deep studied by analyzing the wayside acoustic signal form the experiment. Two different techniques are proposed again to extract the Doppler parameters and remove the Doppler distortion based on the signal resampling in time-domain, which are PTFD and DFMS respectively. The two technique are applied to estimate the time center and characteristic frequency for the Doppler distortion elimination with signal resampling. Both the simulating and experimental results show the effectiveness and availability of the proposed method in the correction of Doppler distortion.Finally, weak signal detection and enhancement are realized referring to the stochastic resonance (SR). We propose three different SR systems based on the optimization of potential model for the de-noising and signal-to-noise ratio (SNR) amplification of bearing fault signal. They are 1. SR based on a Joint Woods-Saxon and Gaussian potential.2. SR in an underdamped system with pinning potential.3. Step-varying asymmetric SR. The weak periodic signal is extracted from background noise for the bearing fault information by the newly-proposed or improved methods and algorithms. They improve the accuracy and reliability of diagnosis. All the models are validated with both simulated and experimental signals.In general, this dissertation proposes the whole method for the wayside acoustic fault diagnosis system. Theory and algorithm are both studied in source separation, distortion removal and signal de-noising. The experimental results indicate the feasibility and availability of the technical route.