9 August 2018 in IEEE Transactions on Industrial Electronics
Incipient faults in electrical drives can corrupt overall performance of high-speed trains; however, they are difficult to be discovered because of their slight fault symptoms. By sufficiently exploiting the distribution information of incipient faults, this paper presents the reason why incipient faults cannot be detected by the existing fault detection and diagnosis (FDD) methods. Under principal component analysis (PCA) framework, we propose a new data-driven FDD method, named probability-relevant PCA (PRPCA), for electrical drives in high-speed trains. The salient strengths of the PRPCA-based FDD method are: 1) It can greatly improve the fault detectability. 2) It is suitable for non-Gaussian electrical drives. 3) Based on the improved fault detectability, it can achieve accurate fault diagnosis via support vector machine (SVM). 4) It can be easily applied to electrical drives even if neither physical models or parameters nor expert knowledge of drive systems is given. 5) It is of highly computational efficiency that can meet requirements on the real-time FDD. A set of experiments on a dSPACE platform based traction system of the CRH2A-type high-speed train are carried out to demonstrate the effectiveness of the proposed method.
19 June 2019 in IEEE Transactions on Industrial Electronics
11 November 2019 in IEEE Transactions on Industrial Informatics
Dynamics and nonlinearity may exist in time and batch directions for batch processes, thereby complicating the monitoring of these processes. This paper proposes a two-dimensional deep correlated representation learning (2D-DCRL) method to achieve efficient fault detection and isolation of nonlinear batch processes. Three-way historical data are first unfolded as two-way time-slice data. Second, a stacked autoencoder-based deep neural network is constructed to characterize the correlation among process variables. Considering that the time and batch directions may be dynamic, for each time-slice measurement, a constructed 2D measurement containing samples from previous time instants and batches is then obtained. Subsequently, DCRL is performed between the current running-batch measurements and the constructed 2D measurements to characterize the two-dimensional dynamics and nonlinearity. 2D-DCRL-based monitoring examines the status of a sample by considering the 2D nonlinear and dynamic information, providing improved monitoring performance. Applications on two typical batch processes demonstrate the effectiveness of the proposed 2D-DCRL monitoring scheme.
26 February 2019 in IEEE Transactions on Systems, Man, and Cybernetics: Systems
1 August 2018 in IEEE Transactions on Industrial Electronics
Successive batch processes generally involve within-batch and batch-to-batch correlations, and monitoring of such batch processes is imperative. This study proposes a multiobjective two-dimensional canonical correlation analysis (M2D-CCA)-based fault detection scheme to achieve efficient monitoring of successive batch processes. First, three-way historical batch process data are unfolded into two-way time-slice data. Second, for each time-slice measurement, CCA is performed between the current measurement and previous measurements from both time and batch directions, which takes the within-batch and batch-to-batch correlations into account. To determine the involved measurements and eliminate the influence of unrelated variables, multiobjective evolutionary optimization is performed, which tries to maximize the preserved canonical correlation coefficients and minimize the number of involved variables. Finally, based on the established M2D-CCA model, an optimal fault detection residual is generated for each time-slice measurement. The M2D-CCA fault detection scheme performs fault detection using the current measurement and the information provided by its previous samples and batches, and therefore exhibits a superior monitoring performance. The M2D-CCA fault detection approach is tested on a numerical example and an industrial injection molding process. Monitoring results verify the feasibility and superiority of the proposed monitoring scheme.
1 July 2022 in IEEE Transactions on Instrumentation and Measurement
Hot spots (HSs) in the early stage can corrupt the generation efficiency of photovoltaic (PV) systems, whose evolution may cause fire hazards as time goes on. Whilst, they are difficult to detect because of the slight anomaly symptoms. In this paper, we propose a novel data-driven detection method of HSs, named as slow manifold analysis (SMA), for PV systems. SMA sufficiently extracts the nonlinear information hidden in monitoring data from PV modules to detect HSs. The salient strengths of the SMA-based detection method are: 1) the algorithm is of high computational efficiency, which can meet requirements of real-time detection; 2) it can fully mine the information of operation status changes caused by HSs in the early stage; 3) the proposed method without using physical models nor expert knowledge can be directly applied to PV systems. Finally, the effectiveness of the designed scheme are verified via nine sets of HSs on PV experimental platforms.