Research On Key Technology Of Big Data Driven Mechanical Equipment Health Monitoring

Mechanical equipment is the key element to realize intelligent manufacturing.However,due to the complex working environment and the influence of various environmental forces,mechanical equipment is prone to component damage,health deterioration and other problems,which lead to the anomaly of mechanical equipment,reduced service life and serious faults,thus causing huge losses to enterprises.With the development of informatization and intelligence of mechanical equipment,by analyzing the monitoring data of mechanical equipment and extracting useful features quickly and accurately,it is possible to perceive the abnormal state of the equipment in advance,estimate the remaining useful life(RUL),and identify the fault type in real time.In order to ensure the safe and reliable operation of mechanical equipment,this paper systematically studies the key technologies of data-driven mechanical equipment health monitoring.The problems of early weak anomaly perception,RUL prediction under various operating conditions and compound fault diagnosis of mechanical equipment are studied respectively,and the corresponding solutions are proposed.The main contents of this dissertation are summarized as follows:(1)Considering the characteristics of mechanical equipment early anomaly,such as the symptom is not obvious and the feature is weak,an adaptive kernel spectral clustering(AKSC)model is proposed,and an early weak anomaly perception method based on aksc model is introduced for mechanical equipment.Compared with the traditional anomaly perception method,this method can effectively perceive the early weak anomaly of mechanical equipment.AKSC model has strong adaptability,and does not need to set a large number of hyper-parameters.After the initial model is trained,it can still use the subsequent monitoring data to calibrate the model and realize the adaptive adjustment of model parameters.Finally,the rolling bearing data set is used to verify the effectiveness of the method.(2)Aiming at the problem that RUL is difficult to effectively predict due to the changing operating conditions of mechanical equipment,a multi-dimensional recurrent neural network(MDRNN)model is proposed,and a RUL prediction method based on MDRNN model is introduced.Compared with the traditional RUL prediction method,this method can be applied to RUL prediction under single working condition and variable operating condition.MDRNN can not only analyze the monitoring signal data,but also mine the operating condition data at the same time.MDRNN has a parallel network layer structure,which can realize data feature mining from different dimensions.Finally,the prediction performance of the method under single and off design conditions is verified by using the data set of aviation turbofan engine.(3)Aiming at the characteristics of mechanical equipment components which are interrelated and prone to compound faults,a local binary convolution neural network(LBCNN)model is designed,and a compound fault diagnosis method of mechanical equipment based on LBCNN model is proposed.Compared with the traditional fault diagnosis method,this method can not only realize the diagnosis of single fault,but also be suitable for the diagnosis of composite fault.By replacing the traditional convolution kernel with the local binary convolution kernel,LBCNN model has faster training efficiency.By designing the multi label classification strategy to label the composite fault samples,LBCNN model can effectively identify the composite fault samples Feature capture and online identification.Finally,the bearing fault data set and gearbox fault data set are used to verify the performance of the proposed method in single fault and composite fault scenarios.(4)The engineering examples of agricultural tractor gearbox and automobile engine are carried out,and the timeliness and availability of the proposed mechanical equipment anomaly perception method and fault diagnosis method in the engineering field are further analyzed.In the early abnormal perception of agricultural tractors,the accurate abnormal perception of CVT transmissions of four tractors is realized.In the fault diagnosis of automotive engine cylinder head tightening,according to the principle and process of engine cylinder head tightening,the fault analysis and data acquisition are implemented,and the fault discrimination model and fault identification model are constructed by using the collected data.On the verification data,the average fault identification accuracy and the average fault daignostic accuracy reach 98.07% and 92.43% respectively.