Research On Fault Diagnosis And Health Assessment Of Rotaing Machinery Based On Meta-Learning

During the long-term operation of mechanical equipment,it is very easy to have various faults.The slight damage of a key component may affect the safe operation of the whole system,causing economic losses at the light level and endangering human life safety at the heavy level.Therefore,the research on fault diagnosis and health assessment of rotating machinery is of great significance to the safety of life and property.The traditional deep learning method requires a large number of tagged data for training.In actual industrial production,it is extremely difficult to obtain a large number of tagged fault samples.In the case of small samples,the deep learning model is prone to over-fitting problems,resulting in the reduction of model accuracy.With the rapid development of small sample learning,meta-learning method has gradually become one of the hot research directions in the field of fault diagnosis.Metalearning is widely used in computer vision,natural language processing and other fields because of its advantages of fast learning new tasks with a small number of training samples.Aiming at the problems of difficulty in obtaining fault samples and poor generalization of deep learning models in fault diagnosis and health assessment of rotating machinery,this paper uses meta learning methods to solve the problem of fault diagnosis and health assessment.Firstly,improvements are made to metric based meta learning models such as prototype networks and relational networks to obtain more accurate class prototype representations,thereby improving the classification accuracy of small sample bearing fault diagnosis.Then,meta learning methods are further utilized to solve the problem of remaining service life prediction.By utilizing the metric based characteristics of relational networks,a health assessment model is constructed,Realize the regression problem of predicting the remaining useful life of bearings.The main research content of this article is as follows:(1)Research on bearing fault diagnosis based on improved prototype networks.Propose an improved fault diagnosis method based on the prototype network,introducing auxiliary classification tasks on the basis of the prototype network to improve the discriminative ability of extracted features,and thereby enhance the representation ability of the class prototype for fault bearings.To verify the effectiveness of the proposed method,fault diagnosis experiments were conducted on a rolling bearing dataset.The experimental results show that the improved prototype network has better distinguishability,accuracy,and higher accuracy compared to the prototype network and relational network.(2)Research on bearing fault diagnosis based on Conv GRU relational network.A fault diagnosis model based on Conv GRU relational network is proposed.Firstly,the embedding module is used to extract bearing fault features.Then,Conv GRU is used as a learnable class prototype generator to generate class prototypes for each class.Finally,the relationship module is used to measure the similarity between class prototypes and query set sample features,achieving fault diagnosis.The experimental results show that compared with prototype networks and relational networks,Conv GRU relational networks can generate more accurate class prototypes,which has a certain effect on improving diagnostic accuracy.(3)Research on predicting the remaining service life of bearings based on relationship networks.A bearing residual life prediction method based on relational network is designed and implemented.First,the embedded module is used to extract the bearing state characteristics,and the relationship module is used to measure the similarity between the bearing state characteristics.Based on the similarity,the bearing health indicators are constructed.Then,the Savitzky Golay filter is used to smooth the health indicators to reduce the impact of oscillation on the prediction results.Finally,the linear function is used to fit the health indicators,Obtain the remaining service life of the bearing.Compared with Conv LSTM,Transformer,RNN,CNN+LSTM,Attention Mechanics and other methods,the relationship network based method can achieve better prediction results on a small number of training samples through experiments on the PHM 2012 bearing measurement dataset,and has certain application value.