Invalid Blind Source Separation Algorithm And Its Application In Mechanical Fault Diagnosis

The vibration signal of a mechanical device reflects the important information which represents its operating status, so the signal’s effective acquisition method becomes the key steps in the process of machine’s identification and fault diagnosis. While, the industrial production environment is severe in practice, the machinery is always working with interference, noise and several faults which makes the fault characteristic signals and noise confuse with each other in the band. For these reasons, it is very difficult to identify the machinery’s fault characteristic signal and achieve the diagnosis of fault. In addition, the feature of mechanical vibration signal is characterized by non-stationary and non-Gaussian which leads to the traditional time-frequency analysis method is not applicable. The theory of Blind Source Separation is an effective way to solve that problem, which can separate the source signals from the finite observation signals in accordance with the assumption of independence. However, the classical blind source separation method do not work for the underdetermined problem. So, for the sake of engineering practice, studying effective method of underdetermined blind source separation becomes an important direction in this filed.In this paper, we study from the basic theory of blind source separation algorithm and compare several algorithms of independent component analysis. The simulation shows that the RobustICA algorithm performance better than others in many ways. Then, for the sake of engineering practice, we introduce the concept of mode decomposition which realizes the underdetermined problem transforming into well-posed or over-determined blind source separation. In the meantime, we present the noise-reduction method to improve the signal to noise ratio which solves the problem of blind source separation with noise. Finally, we combine the processed signal with RobustICA to provide an effective solution to the problems of underdetermined blind separation with noise.Aiming at the usual fault modal of rolling bearing, we apply the algorithm of underdetermined blind source separation based on modal decomposition and noise reduction to the compound fault diagnosis of roller bearing. The result shows that the method can extract the feature signal effectively and successfully. That means the fault diagnosis is realized and verifies the method’s validity. So the referenced algorithm proposed in this paper have important significance in practice.