RESUMO
Driven by industrial big data and intelligent manufacturing, deep learning approaches have flourished and yielded impressive achievements in the community of machine fault diagnosis. Nevertheless, current diagnosis models trained on a specific dataset seldom work well on other datasets due to the domain discrepancy. Recently, adversarial domain adaptation-based approaches have become an emerging and compelling tool to address this issue. Nonetheless, existing methods still have a shortcoming since they cannot guarantee sufficient feature similarity between the source domain and the target domain after adaptation, resulting in unguaranteed performance. To this end, a Cycle-consistent Adversarial Adaptation Network (CAAN) is advanced to realize more effective fault diagnosis of machinery. In CAAN, specifically, an adversarial game formed by the feature extractor and the domain discriminator is constructed to induce transferable feature learning. Meanwhile, the feature translators and discriminators between source and target domains are further designed to build a more comprehensive cycle-consistent generative adversarial constrain, which can more reliably ensure domain-invariant and class-separate characteristics of learned features. Extensive experiments constructed on three datasets from different mechanical systems demonstrate the effectiveness and superiority of CAAN.
Assuntos
Aprendizado Profundo , Indústria ManufatureiraRESUMO
K-singular value decomposition (K-SVD), as an extension of sparse coding, has attracted great attention for fault feature extraction of rolling element bearings (REBs) in recent years. However, the performance of original K-SVD algorithm is flawed since its atoms in the dictionary are invariably updated according to the principle component, which reduces its pertinence to periodic impulses under complex interferences. To cope with this deficiency, this paper proposes an information-based K-SVD (IK-SVD) method considering the intrinsic properties of fault response. In this framework, an information-based atom selection strategy (IASS) is designed to seek a group of fault-related atoms for dictionary updating, in which a new index named harmonic ratio (HR) is employed to provide the evaluation criterion. On this basis, the fault signatures can be specifically extracted from degraded vibration signals with proposed method. Moreover, a residual estimation method is presented to compute the threshold value used in the sparse coding stage. The superiority of IK-SVD is verified on simulated signal and practical signals from a locomotive bearing test rig. The analysis results demonstrate the good performance of proposed method for the fault detection of REBs.
RESUMO
Rolling element bearings (REBs) play an essential role in modern machinery and their condition monitoring is significant in predictive maintenance. Due to the harsh operating conditions, multi-fault may co-exist in one bearing and vibration signal always exhibits low signal-to-noise ratio (SNR), which causes difficulties in detecting fault. In the previous studies, maximum correlated kurtosis deconvolution (MCKD) has been validated as an efficient method to extract fault feature in the fault signals. Nonetheless, there are still some challenges when MCKD is applied to fault detection owing to the rigorous requirements of multiple input parameters. To overcome limitation, a multi-objective iterative optimization algorithm (MOIOA) for multi-fault diagnosis is proposed. In this method, correlated kurtosis (CK) is taken as a criterion to select optimal Morlet wavelet filter using the whale optimization algorithm (WOA). Meanwhile, to further eliminate the effect of the inaccurate period on CK, the update process of period is incorporated. After that, the simulated and experimental signals are utilized to testify the validity and superiority of the MOIOA for multiple faults detection by the comparison with MCKD. The results indicate that MOIOA is efficient to extract weak fault features even with heavy noise and harmonic interferences.