Research On Bearing Fault Early Warning Based On Gaussian Mixture Model

With the intellectualization,diversification of functions,and complexity of structures of mechanical equipment,there are higher requirements for the reliability of mechanical equipment.Therefore,it is necessary to conduct fault prediction and health management for key components in mechanical equipment,which can significantly improve the safety and reliability of mechanical equipment.The bearing is a key component in rotating machinery.Once the bearing fails,it will have an impact on the stability of the transmission and even the entire machine.Therefore,this article takes bearings as the research object,and conducts research from three aspects: the extraction of fault characteristics,the construction of performance degradation evaluation models,and the fault warning of bearings under varying operating conditions,with the aim of better judging the performance of bearings and timely and accurately detecting early failures of bearings.The main work of this thesis is as follows:(1)Extract fault features that can reflect the comprehensive performance of bearings.The traditional feature extraction of vibration signals,including multi-domain fault features such as time domain,frequency domain and time-frequency domain,is used as the initial input to establish a probability distribution model,and the probability distribution of multi-dimensional fault features in space is obtained.The vibration signal is directly taken as the input to establish the probability distribution model,and the probability distribution of the vibration signal value in space is obtained.The probability distribution model is the fault feature.The change of probability distribution with time can represent the change of bearing performance.(2)Establishment of performance degradation evaluation model.First of all,we use divergence to quantify the change of probability distribution that is difficult to judge intuitively,compare the commonly used divergence,and select KL divergence which is the most sensitive to change.Then the KL divergence is standardized to make it not only reflect the monotony of bearing degradation,but also be consistent with the actual performance of the bearing.The results of simulation and experimental data sets show that the health indicators constructed in this thesis can not only accurately judge the initial failure time of the bearing,but also better judge the degradation stage of the bearing,so as to facilitate the formulation of reasonable maintenance strategies after the failure.(3)In order to better provide fault warning for bearings that constantly change operating conditions in actual operation,a new variable operating condition bearing fault warning method has been proposed.Combining the global background model with the performance indicators proposed in this article,the performance indicator values in the healthy state are used as the background model to judge the state of the input indicators.Through the test of bearing under variable operating conditions,it is verified that the fault early warning method proposed in this thesis can timely and effectively early warning the faults generated by bearings.