Anomaly detection is critical for ensuring the reliability of industrial cyber-physical systems. Identifying anomalies based on data distribution is regarded as a promising approach. However, inherent noise in data collection and complex dependencies within the underlying structure can lead to class ambiguity. This ambiguity obscures the boundary between normal data and anomalies, thereby degrading the accuracy of distribution modeling. To address this issue, we shift the distribution modeling from the data space to a latent space to mitigate ambiguity and then propose a label free anomaly detection network, named ALDM. In ALDM, a contrastive-based methods is designed to facilitate the construction of a latent space, where the margin between normal data and anomalies has been expanded. Anomalies are then discerned through embeddings using a flow-based process. Recognizing the importance of distance metrics in contrastive-based methods, we propose an adaptive approach to obtain the optimal distance metric during network training instead of presetting a fixed formula. Furthermore, to accommodate anomalies of varying durations, we propose another event-wise performance index for evaluation. Extensive evaluations on three widely used benchmarks and a newly constructed dataset demonstrate that ALDM achieves state-of-the-art detection performance across both conventional metrics and our proposed index.

Multivariate time series anomaly detection has been extensively studied in diverse domains. Within the horizon of unsupervised methods, density estimation is regarded as a promising direction. However, the anomalies that distribute closely with normal data in data space impact the desired density estimation results, thereby significantly affecting the discrimination of anomalies. In this work, to tackle this challenge, we propose Dynamic Soft Contrastive Learning (DiSCo). DiSCo generates a representation that better captures the underlying correlations within the data distribution, providing a more compact, informative, and suitable description for the subsequent density estimation task. Furthermore, DiSCo employs a dynamic and learnable distance metric to measure similarities between inputs. The learned task-specific distance metric contributes to minimizing the intra-class distance and maximizing the inter-class distance, thereby assisting the generation of better representation. The efficacy of DiSCo is proved through extensive experiments conducted on four common anomaly detection datasets.

Most of existing category-level object pose estimation methods devote to learning the object category information from point cloud modality. However, the scale of 3D datasets is limited due to the high cost of 3D data collection and annotation. Consequently, the category features extracted from these limited point cloud samples may not be comprehensive. This motivates us to investigate whether we can draw on knowledge of other modalities to obtain category information. Inspired by this motivation, we propose CLIPose, a novel 6D pose framework that employs the pre-trained vision-language model to develop better learning of object category information, which can fully leverage abundant semantic knowledge in image and text modalities. To make the 3D encoder learn category-specific features more efficiently, we align representations of three modalities in feature space via multi-modal contrastive learning. In addition to exploiting the pre-trained knowledge of the CLIP’s model, we also expect it to be more sensitive with pose parameters. Therefore, we introduce a prompt tuning approach to fine-tune image encoder while we incorporate rotations and translations information in the text descriptions. CLIPose achieves state-of-the-art performance on two mainstream benchmark datasets, REAL275 and CAMERA25, and runs in real-time during inference (40FPS).

In practical systems, due to reasons such as sensor limitations, sensor faults, and packet losses in networks, the observation of events becomes nondeterministic. In this article, we extend strong detectability and weak detectability to the case of nondeterministic observations and denote them as A-($k_{1}$,$k_{2}$)-detectability and O-($k_{1}$,$k_{2}$)-detectability, respectively. A-($k_{1}$,$k_{2}$)-detectability says that, for any string, we can distinguish state pairs in the specification for all possible observations of the string. O-($k_{1}$,$k_{2}$)-detectability says that, for at least one string, we can distinguish state pairs in the specification for all possible observations of the string. For A-($k_{1}$,$k_{2}$)-detectability, we construct a transformed automaton and then translate the A-($k_{1}$,$k_{2}$)-detectability problem into the traditional detectability problem that has been solved. We show that A-($k_{1}$,$k_{2}$)-detectability can be used to solve the deterministic supervisory control problem. For O-($k_{1}$,$k_{2}$)-detectability, we construct an augmented automaton that includes all the information of the given automaton and its state estimates. Based on the augmented automaton, we propose a depth-first search (DFS)-based algorithm to check O-($k_{1}$,$k_{2}$)-detectability.Note to Practitioners—Nowadays, practical engineering systems become more and more complex. In these systems, the observation of events often becomes nondeterministic due to reasons such as sensor limitations, sensor faults, and packet losses in networks. Consider a mobile robot as an example. The availability of a sensor output may depend on the current location of the mobile robot. If the mobile robot is in an area where the wireless network is unreliable, the sensor output may not be received by the supervisor. In this article, we investigate the state estimation problem for practical engineering systems under nondeterministic observations within a discrete-event system framework. The results in this article provide not only insights for engineers in the automatic control field to understand nondeterministic observations in practical systems but also the methodology to estimate the current discrete state that is always an important issue. Therefore, we believe that the engineers in the automatic control field should be interested in this article and can benefit from it.

In this paper, we investigate the robust control problem of discrete event systems in a networked environment. In other words, we use a supervisor to control several possible plants (discrete event systems) to achieve given specifications when there are communication delays and losses in communication networks linking the supervisor and the plants. We translate the robust networked control problem into a conventional networked control problem by constructing an augmented automaton for all possible plants and an augmented specification automaton for the corresponding specification automata. We then solve the robust networked control problem. We consider two cases. The first case is when all the specifications are the same. For this case, we derive a necessary and sufficient condition for the existence of a robust networked supervisor. The second case is when the specifications are different, which is more general compared with the first case. In the second case, we can only obtain a sufficient condition for the existence of a networked supervisor. The results are illustrated by an example of customized products filling line.

Opacity is important in investigating secrecy, privacy, and other properties in general systems that can be modeled as discrete event systems. To ensure opacity, a controller may be used to control information released to the public. For transparency and other reasons, it is often desired that the information released to the public be maximum, as long as opacity is not violated. In this paper, we investigate how to release the maximum information while ensuring opacity. We find a necessary and sufficient condition for an information release policy to ensure opacity. We also develop methods and algorithms to design a controller that releases maximum information. We consider both strong opacity and weak opacity. We apply the results to the dining cryptographers problem. Note to Practitioners-Nowadays, networks and computers make the collection, storage and dissemination of information much easier. This brings us a lot of convenience, but at the same time, also brings us the worry that some private information may be released to the public undesirably. The problem of how to control the release of information thus becomes very important. In this paper, we propose a discrete-event-system approach to investigate the information release problem. The solution can ensure the opacity of private information while releasing the maximum information to the public. We believe that the practitioners both in the field of system and control and in the field of information science will benefit from the results.

Opacity is important in investigating secrecy, privacy, and other important properties in general systems that can be modeled as discrete event systems. To ensure opacity, a controller may be used to control information released to the public. For transparency and other reasons, it is often desired the information released to the public be maximum, as long as opacity is not violated. In this paper, we investigate how to release the maximum information while ensuring opacity. We find a necessary and sufficient condition for a control policy to ensure opacity. We also develop methods and algorithms to design a controller that releases maximum information. We consider both strong opacity and weak opacity.

In this paper, we investigate recoverability of discrete-event systems with faults. For a discrete-event system, recoverability is defined as the ability to fire recovery events after the occurrence of a finite number of events for any event sequence which leads the system to faulty modes. By constructing the observer of the system, we discuss how to check whether the system is recoverable or not. Furthermore, we propose one more efficient approach to check recoverability. Finally, all the results are illustrated by a practical example.

This paper considers fault-tolerant control that ensures the safety of a discrete-event system. We consider multiple faulty modes. Each faulty mode is modeled by an automaton. These automata, together with the automaton modeling normal mode, describe a discrete-event system with faults. Each faulty mode has some illegal states that must be avoided by control so that the fault can be tolerated. We assume that fault-tolerant control takes actions (disablements) only when the occurrence of a fault is certain. We consider cases of both full event observation and partial event observation. We derive necessary and sufficient conditions for the existence of fault-tolerant control. We also provide formulas and algorithms to calculate control actions if the necessary and sufficient conditions are satisfied. Both offline control synthesis (for full event observation) and online control synthesis (for partial event observation) are investigated. We allow multiple faults as well as single faults.

In this paper, we investigate online sensor activation to ensure detectability of discrete event systems. Detectability requires that states of a system can be determined or certain pairs of states can be distinguished by an external observer eventually or periodically. Since minimal sensor activation policies for detectability may not exist, two new concepts are introduced: 1)k-step distinguishability is introduced for strong detectability and 2) information-preserving is introduced for strong periodic detectability. The online sensor activation is then proposed and is based on the best state estimate available at the time of decision making. Three algorithms are developed for online sensor activation. The first two algorithms are for strong detectability. They minimize sensor activation while preservingk-step distinguishability. The third algorithm deals with strong periodic detectability. It minimizes sensor activation while preserving state information.

State estimation has always been important in discrete-event systems. There are two types of state estimation problems in discrete-event systems: one is to determine the initial state of the system and the other is to determine the current state of the system. In this paper, we investigate the initial state estimation problem. We formulate initial state estimation problem as I-detectability. A discrete-event system is strongly I-detectable if we can determine the initial state of the system after a finite number of event observations for all trajectories of the system. It is weakly I-detectable if we can determine the initial state of the system for some trajectories of the system. We construct I-observer to analyze strong and weak I-detectability and construct I-detector to check strong I-detectability. For some applications, strong I-detectability is required but not satisfied; hence we investigated how to control a system to achieve strong I-detectability if needed. If there exists a controllable, observable, and strongly I-detectable sublanguage, then we say the system is closed-loop strongly I-detectable. We derive an effective algorithm to check whether a system is closed-loop strongly I-detectable. The algorithm can also calculate a controllable, observable, and strongly I-detectable sublanguage if the system is closed-loop strongly I-detectable.

This paper considers fault-tolerant control that ensures the safety of a discrete event system. We consider multiple faulty modes. Each faulty mode is modeled by an automaton. These automata, together with the automaton modeling the normal mode, describe a discrete event system with faults. Each faulty mode has some illegal states that must be avoided by control so that the fault can be tolerated. We assume that fault-tolerant control will take action (disablements) only when the occurrence of a fault is certain. We consider both full event observation and partial event observation. We derive necessary and sufficient conditions for the existence of fault-tolerant control. We also provide formula and algorithm to calculate control actions if the necessary and sufficient conditions are satisfied. Both offline control synthesis (for full event observation) and online control synthesis (for partial event observation) are investigated.

Here we discussed recoverability of faulty discrete event systems. A faulty discrete event system includes the healthy dynamics and the faulty dynamics. In faulty modes, the performance of the system often degrades. Hence we need to execute some recovery actions to recover the performance of the system. Recovery actions may be renewing/repairing the faulty device or re-configuring the healthy devices. Generally recovery actions can be fired not for all the time, but for some time. They were modeled as a mapping from the state set of the faulty mode to a binary set in which 1 means recovery actions can be fired and 0 means recovery actions cannot be fired. Then we defined two types of recoverabilities: recoverability and weak recoverability. Recoverability means we periodically have chances to do recovery actions and weak recoverability means we have at least one chance to do recovery actions when the system has a fault. With observer technology, we proposed criteria to check these two types of recoverabilities.