Research On Fault Diagnosis And Condition Monitoring Of Digital Patternless Sand Mold Process Machine Tool

With the rapid development of China’s aerospace industry,the requirements of high quality and high reliability of complex thin-walled light alloy components are becoming increasingly urgent.The stable,safe and efficient operation of aerospace parts processing equipment for a long time is a necessary condition for processing high-quality and highly reliable aircraft parts.The digital patternless sand mold process machine tool,using the computer control technology and sand processing technology,through the program can achieve space complex thin-walled light alloy component rapid processing of sand mold and sand core.In order to produce high quality and reliable parts,on the one hand,it is necessary to timely identify and repair the faults in the operation of the digital patternless sand mold process machine tool,on the other hand,it is necessary to carry out real-time monitoring on the performance of the equipment during operation to ensure the processing quality of the equipment during operation.So,this paper will carry out relevant research on fault diagnosis and health condition monitoring for the digital patternless sand mold process machine tool,and the main contents are as follows:(1)Fault Tree Analysis was conducted on the digital patternless sand mold process machine tool.Firstly,by analyzing the structure and function of the digital patternless sand mold process machine tool,the fault tree of each subsystem of the digital patternless sand mold process machine tool is established.Then carry out qualitative analysis to find out the minimum cutting set of the digital patternless sand mold process machine tool,and then quantitative analysis,to get the reliability of digital patternless sand mold process machine tool and the fault tree bottom event probability importance degree and key importance.The weak reliability parts of the digital patternless sand mold process machine tool are identified,which will be the focus for the follow-up research on fault diagnosis and reliability assessment of the equipment.(2)A multi-ensemble fault diagnosis method based on stack auto-encoder is proposed for weak reliability parts.In this method,features of different types are extracted from multiple SAEs constructed by different activation functions to form feature pools.Then the features in the feature pool are evaluated and selected,and the selected features are used to divide different sample sets to train different classifiers.The final results are obtained by majority voting through these classifiers.Finally,an example is given to verify the effectiveness and practicability of the method.(3)Proposed a fault diagnosis method based on the generative adversarial networks under the condition of variable operating conditions and incomplete data.This method is proposed mainly because the multi-ensemble fault diagnosis method based on SAE needs to meet two basic conditions: 1)The training sample data used for model training and the test sample data used for model testing must follow the same distribution and be independent from each other.2)A good diagnostic model can only be trained if there is enough tagging data for model training.However,in the actual industrial fault diagnosis scenario,these two conditions are basically untenable.In addition,the conventional transfer learning method cannot be implemented in the case of incomplete tagging data of the training set.Finally,this method is tested in different test scenarios.(4)A condition monitoring method based on multi-source data is proposed for weak system components.First,features are extracted from different data sources using SAE.Then these features were selected by the evaluation indexes based on monotonicity and correlation performance degradation characteristics,and the state parameter method based on the state space distance was used to construct the health index.Finally,the proposed monitoring method is verified through several test sets,and the results show that the method has a good performance on multiple test sets and a good generalization ability.