Multilabel Decoupling-driven Intelligent Diagnosis Method For Compound Fault Of Machines

Fault diagnosis technology is a significant tool for improving the security,stability and reliability of Industrial Equipment(IE),which can not only ensure the safety of people's lives and property,as well as the national defense,but also impact strategic significance on the transformation and upgrading of manufacturing industry.One of the major challenges facing fault diagnosis is that compound fault is a primary failure of IE that often emerges and evolves when multiple key components are sequentially or simultaneously degraded or even damaged.The monitoring signals become more complex since the fault characteristics of each component are coupled and exerted influence reciprocally,which dramatically increases the difficulties of fault diagnosis.Considering the demand of IE on the intelligent diagnostic and prognostic,this dissertation focuses on the task of intelligent compound fault decoupling and diagnosis,which can make the Intelligent Fault Diagnosis(IFD)methods more reliable,robust,and applicable.Therefore,aiming at solving the problems when develops an IFD methods in practical industry application,a multilabel decoupling-driven intelligent diagnosis method for compound fault of IE is proposed and investigated in the scenario of working condition transfer,working condition generalization and multisource data fusion,which can provide effective solutions to promote the IFD widely adopted in practical manufacturing industry systems.The main contribution of this dissertation can be summarized as follows:First,two multilabel decoupling-driven intelligent diagnosis models are proposed for compound fault of IE.Traditional IFD algorithms ignore the relationships between the compound fault and its corresponding single faults and simply regard compound fault as an independent fault category to make a classification.To overcome such limitation and endow the IFD method with the ability to decouple the compound fault in an intelligent manner,the supervised multilabel classifier and the unsupervised decoupling classifier is designed for the two scenarios where the labeled compound fault data is available or not,respectively,which can imitate the learning ability of humans.The proposed two models can intelligently decouple the compound fault of IE via the mechanism of multilabel output,which significantly improve the effectiveness and applicability of IFD methods.Second,an intelligent compound fault decoupling method is developed for IE under varying working conditions,combining the feature-based transfer learning and the unsupervised decoupling classifier.A Transferable Capsule Network-based compound fault decoupling model is constructed to reduce the discrepancy between the deep features learned from the source and target domains,which can transfer the trained diagnosis model from one working condition to another,improve the generalization performance of diagnosis model,and provide an effective solution for the intelligent compound fault diagnosis of IE under varying working conditions.Third,two intelligent diagnosis methods are investigated for further improving the generalization performance of diagnosis model while the testing data cannot be collected in advance for some extreme working conditions.Inspired by the weight-shared mechanisim of Convolutional Neural Networks(CNNs),a robust weight-shared capsule network is introduced for intelligent machinery fault diagnosis,which provides a feasible solution that the diagnosis model can be trained with the labeled data collected under a certain working condition,and applied to monitor the health status of IE under some extreme working conditions.With respect of the compound fault diagnosis,a novel Deep Adversarial Capsule Network(DACN)is proposed to embed multidomain generalization into the intelligent compound fault diagnosis for machinery,so that the DACN not only can intelligently decouple the compound fault into multiple single faults for IE but also can be generalized from certain working conditions to another new.Fourth,a multisensory data fusion-based compound fault decoupling method is proposed to solve the problem where the aforemention model cannot work well when the number of compound faults is increased since only individual sensor data are used to train the diagnosis model.Considering that combining the ensemble learning and deep learning can take full advantages both of deep learning and ensemble learning,a novel intelligent compound fault decoupling method based on capsule network and ensemble learning is introduced to address the problems mentioned above for the compound fault diagnosis,which provides a feasible way for the development and application of intelligent compound fault decoupling method in multisensory data industrial scenarios.The effectiveness and superiority of the proposed methods are validated on a dataset collected from a five-speed automobile transmission.It is worth mentioning that,compared with traditional IFD methods,the proposed methods have made some successful breakthroughs to intelligent compound fault diagnosis,however,some problems and challenges detatiled in the outlook of this dissertation should be focused and further investigated in the future.