Considering situations such as sensor failures and communication losses, limited data samples are a common challenge in actual industrial processes, making the traditional integrated architecture of fault detection (FD) and fault-tolerant control (FTC) based on subspace identification difficult to be applicable. Regarding this problem, this article proposes a nuclear norm-based subspace identification method for FD and FTC. This method leverages key structural matrix properties in the input and output data model, alleviating reliance on data samples. The parameter matrices required to construct the fault detector and fault-tolerant controller can be directly identified within the nuclear norm optimization framework, enabling the design of an integrated FD and FTC architecture. Two case studies demonstrate that the developed method enhances detection and control performance compared with traditional subspace identification methods, particularly in the case of limited data samples.

Considering the influence of subsystem state interconnection in large-scale systems, the existing integrated design methods of data-driven fault detection (FD) and fault-tolerant control (FTC) that follow centralized architecture cannot be applied in distributed scenarios. To address this problem, this article proposes a subspace predictor-assisted framework to perform the data-driven integrated design of FD and FTC for large-scale systems. FD and FTC are organically combined through a subspace predictor framework. For the subspace predictor designed for each subsystem, no global input and output (I/O) data information is required but only the I/O data of the local and neighboring subsystems is used, thus realizing a distributed design. In addition, the integrated architecture of FD and FTC does not need any large-scale system mechanism information, and is completely driven by process I/O data. Two case studies including a numerical simulation example and cascaded continuously stirred-tank reactor verify the feasibility and effectiveness of the proposed data-driven distributed FD and FTC method.

This article investigates the problem of data-driven distributed optimal fault detection for large-scale interconnected systems with the unmeasurable interaction term of neighboring system information. For large-scale systems, the computational and storage burdens hinder the application of centralized fault detection methods, while the existence of the unknown interaction term in residual generators brings challenges to distributed fault detection problems. To solve the above problems, the unknown interaction term is implicitly included in each subsystem through an algebraic equivalent transformation, so that the residual generator constructed by the distributed method will not lose the fault information propagated along the network topology. Furthermore, an optimization scheme is designed to measure the effect of the residual signal on noise and faults in all dimensions of the parity space, making the residual generator sufficiently sensitive to even weak faults. Numerical examples and a real hot strip rolling case verify the effectiveness and superiority of the proposed method.

Traditional data-driven fault detection methods based on subspace identification encounter difficulties when performing distributed fault detection on large-scale systems, mainly due to the presence of the unknown interaction term in the residual generator constructed for each subsystem. To tackle these problems, this article proposes a robust distributed fault detection method based on subspace identification for large-scale systems. First, the initial identification error of the residual generator constructed by each subsystem is obtained by using the input and output data information of the local and neighbors, and then a one-step correction theorem is introduced to minimize the error further. In addition, a robust residual generator that is sensitive to faults and robust to noise is constructed by studying all dimensions of the parity space. The effectiveness and superiority of the proposed method are illustrated by comparing the existing methods in two case studies of numerical simulation and real finishing mill system in hot strip rolling.

Unknown interaction items in the construction of distributed residual generators for large-scale systems will lead to the failure of existing data-driven fault detection (FD) methods based on subspace identification. To solve this problem, a subspace-assisted distributed FD scheme under the cloud-edge-end collaboration framework is proposed. For the residual generator constructed for each subsystem, the unknown input item is proved to be represented by the global system's input and output (I/O) data. In addition, based on the represented unknown input term, a data-driven form of the residual generator required for each subsystem is designed. Meanwhile, to eliminate the computing and storage burden caused by the global I/O data required by the represented unknown input item, a cloud-edge-end collaboration architecture is proposed and the corresponding tasks are deployed on the three sides of the cloud-edge-end, respectively. The effectiveness of the proposed method is analyzed and verified by numerical simulation and a real hot rolling case.

Considering that the interaction information between neighbor subsystems is unmeasurable, this article investigates the problem of distributed fault detection (FD) for individual subsystems in a large-scale system. Unmeasurable interaction terms as unknown inputs to subsystems pose a challenge for distributed FD. To cope with this problem, in this article, a distributed FD scheme for large-scale systems is proposed, which utilizes only local and neighbor input and output data information to achieve the estimation of the unknown input term by subspace intersection. In addition, a data-driven distributed residual generator construction method is designed based on the estimated unknown input term. Meanwhile, the rank conditions that need to be satisfied by the distributed method are provided. Finally, the effectiveness of the proposed method is verified and discussed in a simulation example and a real manufacturing case.