Industry 4.0 has facilitated the access to sensor and actuator data from manufacturing systems, leading to studies on data-driven anomaly detection, but limited attention has been paid to finding root causes and automating this process using formalized expert knowledge. This is crucial due to the scarcity of qualified engineers and the time-consuming nature of diagnosing issues in large production systems. To address this gap, we present a framework that combines data-driven anomaly detection with a knowledge graph that provides domain knowledge by leveraging typical explanations of such models (i.e.,data streams potentially caused the detection) for further diagnosis. The framework’s usefulness to infer affected components or data set labels has been evaluated using two deep anomaly detection approaches. For knowledge-based diagnosis, three query strategies that utilize various knowledge graph relationships are implemented through three Artificial Intelligence (AI) techniques. The proposed anomaly detection approach, informed by integrating expert knowledge via the graph structure of the knowledge graph and node embeddings for encoding time series, outperforms baselines and a deep autoencoder in detecting anomalies and in identifying anomalous data streams. In subsequent diagnosis, it achieves the best performance on a complete knowledge graph in combination with a graph pattern matching query by identifying the label or affected component in 60% of detected anomalies by providing 4.1 labels or 2.3 components until the correct one is identified. In case of a corrupted one, Symbolic-Driven Neural Reasoning (SDNR) and Case-Based Reasoning (CBR) with knowledge graph embeddings demonstrate advantages by halving the number of incorrect labels and unaffected components. • Applying three AI techniques (SPARQL - a Query Language for Resource Description Framework (RDF), Case-Based Reasoning (CBR), and Symbolic-Driven Neural Reasoning (SDNR)) for knowledge-based Root Cause Analysis (RCA) by leveraging typical explanations (i.e.,causative data streams) provided by data-driven anomaly detection models. • Proposing an informed deep self-supervised one-class anomaly detection approach that integrates domain knowledge in the form of time series relationships derived from the knowledge graph and knowledge graph embeddings. • Presentation of a general Failure Mode, Effects & Analysis (FMEA) ontology to model expert knowledge about faults and failures that is also instantiated for the used data.

To build intelligent manufacturing systems that react flexibly in case of failures or unexpected circumstances, manufacturing capabilities of production systems must be utilized as much as possible. Artificial Intelligence (AI) and, in particular, automated planning can contribute to this by enabling flexible production processes. To efficiently leverage automated planning, an almost complete planning domain description of the real-world is necessary. However, creating such planning descriptions is a demanding and error-prone task that requires high manual efforts even for domain experts. In addition, maintaining the encoded knowledge is laborious and, thus, can lead to outdated domain descriptions. To reduce the high efforts, already existing knowledge can be reused and transformed automatically into planning descriptions to benefit from organization-wide knowledge engineering activities. This paper presents a novel approach that reduces the described efforts by reusing existing knowledge for planning and scheduling in Industry 4.0 (I4.0). For this purpose, requirements for developing a converter that transforms existing knowledge are derived from literature. Based on these requirements, the SWS2PDDL converter is developed that transforms the knowledge into formal Planning Domain Definition Language (PDDL) descriptions. The approach’s usefulness is verified by a practical evaluation with a near real-world application scenario by generating failures in a physical smart factory and evaluating the generated re-planned production processes. When comparing the resulting plan quality to those achieved by using a manually modeled planning domain by a domain expert, the automatic transformation by SWS2PDDL leads to comparable or even better results without requiring the otherwise high manual modeling efforts.

Abstract Monitoring the state of currently running processes and reacting to ad-hoc situations during runtime is a key challenge in Business Process Management (BPM). This is especially the case in cyber-physical environments that are characterized by high context sensitivity. MAPE-K control loops are widely used for self-management in these environments and describe four phases for approaching this challenge: Monitor, Analyze, Plan, and Execute. In this paper, we present an architectural solution as well as implementation proposals for using MAPE-K control loops for adaptive workflow management in smart factories. We use Complex Event Processing (CEP) techniques and the process execution states of a Workflow Management System (WfMS) in the monitoring phase. In addition, we apply automated planning techniques to resolve detected exceptional situations and to continue process execution. The experimental evaluation with a physical smart factory shows the potential of the developed approach that is able to detect failures by using IoT sensor data and to resolve them autonomously in near real time with considerable results.

Ongoing initiatives such as Made in China 2025 or Industry 4.0 are transforming manufacturing environments into complex cyber-physical production systems (CPPS) consisting of multiple interacting subsystems with a huge number of sensors and actuators. To monitor such an environment, it is necessary to centrally process the high-dimensional time series generated during the operation of the whole CPPS. Characteristically for this scenario is that a particular data stream is usually only causally related to a very small number of other data streams, and only a relatively small subset of the total data streams is relevant for the detection of a particular failure mode. For this purpose, we propose a siamese neural network that employs 2D convolutions for extracting temporal features data stream-wise, followed by graph convolutions for extracting spatial features. Especially, we focus on the integration of expert knowledge for masking irrelevant data streams. We evaluate our approach against state-of-the-art similarity measures for time series such as dynamic time warping, NeuralWarp, and a learned similarity metric based on ROCKET representations. Our approach delivers at least equivalent, if not better, results when compared to the best approach that does not integrate expert knowledge while requiring only 1/20 of learnable parameters, which indicates its practicality for integrating expert knowledge.

Abstract The concept developed during the SEMAFLEX project combines knowledge‐based document management with flexible workflow management to achieve ideal support for small‐ and medium‐sized enterprises. This paper focusses on the workflow approach that implements flexibility by deviation on the basis of constraint satisfaction problem solving. Workflow deviations due to unpredictable circumstances or upcoming events are tolerated at runtime, but still, support is maintained and possible succeeding work items are proposed. Procedurally modelled workflows are transformed to declarative constraints, which serve as database for the workflow engine. The described approach is fully implemented, and our experiments demonstrate sufficient runtime performance as well as improvements for practical use.

This paper targets the automated extraction of components of argumentative information and their relations from natural language text.Moreover, we address a current lack of systems to provide complete argumentative structure from arbitrary natural language text for general usage.We present an argument mining pipeline as a universally applicable approach for transforming German and English language texts to graph-based argument representations.We also introduce new methods for evaluating the results based on existing benchmark argument structures.Our results show that the generated argument graphs can be beneficial to detect new connections between different statements of an argumentative text.Our pipeline implementation is publicly available on GitHub.

Manufacturing systems naturally contain plenty of sensors which produce data primarily used by the control software to detect relevant status information of the actuators. In addition, sensors are included in order to monitor the health status of specific components, which enable to detect certain known, frequently occurring faults or undesired states of the system. While the identification of a failure by using the data of a sensor dedicated explicitly to its detection is a rather straightforward machine learning application, the detection of failures which only have an indirect effect on the data produced by a couple of other sensors is much more challenging. Therefore, a combination of different methods from Artificial Intelligence, in particular, machine learning and knowledge-based (semantic) approaches is required to identify relevant patterns (or failure modes). However, there are currently no appropriate research environments and data sets available that can be used for this kind of research. In this paper, we propose an approach for the generation of predictive maintenance data by using a physical Fischertechnik model factory equipped with several sensors. Different ways of reproducing real failures using this model are presented as well as a general procedure for data generation.

Process-oriented case-based reasoning (POCBR) supports workflow modeling by retrieving and adapting workflows that have proved useful in the past. Current approaches typically require users to specify detailed queries, which can be a demanding task. Conversational case-based reasoning (CCBR) particularly addresses this problem by proposing methods that incrementally elicit the relevant features of the target problem in an interactive dialog. However, no CCBR approaches exist that are applicable for workflow cases that go beyond attribute-value representations such as labeled graphs. This paper closes this gap and presents a conversational POCBR approach (C-POCBR) in which questions related to structural properties of the workflow cases are generated automatically. An evaluation with cooking workflows indicates that C-POCBR can reduce the communication effort for users during retrieval.

In case-based reasoning, improving the performance of the retrieval phase is still an important research issue for complex case representations and computationally expensive similarity measures. This holds particularly for the of retrieval workflows, which is a recent topic in process-oriented case-based reasoning. While most index-based retrieval methods are restricted to attribute-value representations, the application of a MAC/FAC retrieval approach introduces significant additional domain-specific development effort due to design the MAC phase. In this paper, we present a new index-based retrieval algorithm, which is applicable beyond attribute-value representations without introducing additional domain-specific development effort. It consists of a new clustering algorithm that constructs a cluster-based index structure based on case similarity, which helps finding the most similar cases more efficiently. The approach is developed and analyzed for the retrieval of semantic workflows. It significantly improves the retrieval time compared to a linear retriever, while maintaining a high retrieval quality. Further, it achieves a similar performance than the MAC/FAC retriever if the case base has a cluster structure, i.e., if it contains groups of similar cases.

The Collaborative Agile Knowledge Engine (CAKE) is a prototypical generic software system for integrated process and knowledge management. CAKE integrates recent research results on agile workflows, process-oriented case-based reasoning, and web technologies into a common platform that can be configured to different application domains and needs. We describe the main concepts and the architecture of CAKE and sketch three example applications.

The market for health care systems supporting physicians and improving their daily routine is dynamically growing. The development of these systems makes great demands on the handling of medical knowledge, like anatomical and process knowledge. In this paper, a special approach for the radiological domain is presented. Our framework includes mechanisms to store medical knowledge in different knowledge containers, whose importance varies from application to application, and to support the execution of the processes. Three applications are introduced by way of example. We analyze these application scenarios to find the knowledge-intensive tasks that can be supported by an assistant system. The implemented solutions are integrated in the daily work of our radiological partner hospitals.

In this paper, we propose the PARIS approach for improving complex problem solving by learning from previous cases. In this approach, abstract planning cases are learned from given concrete cases. For this purpose, we have developed a new abstraction methodology that allows to completely change the representation language of a planning case, when the concrete and abstract languages are given by the user. Furthermore, we present a learning algorithm which is correct and complete with respect to the introduced model. An empirical study in the domain of process planning in mechanical engineering shows significant improvements in planning efficiency through learning abstract cases while an explanation-based learning method only causes a very slight improvement.

Abstraction is one of the most promising approaches to improve the performance of problem solvers. In several domains abstraction by dropping sentences of a domain description -- as used in most hierarchical planners -- has proven useful. In this paper we present examples which illustrate significant drawbacks of abstraction by dropping sentences. To overcome these drawbacks, we propose a more general view of abstraction involving the change of representation language. We have developed a new abstraction methodology and a related sound and complete learning algorithm that allows the complete change of representation language of planning cases from concrete to abstract. However, to achieve a powerful change of the representation language, the abstract language itself as well as rules which describe admissible ways of abstracting states must be provided in the domain model. This new abstraction approach is the core of Paris (Plan Abstraction and Refinement in an Integrated System), a system in which abstract planning cases are automatically learned from given concrete cases. An empirical study in the domain of process planning in mechanical engineering shows significant advantages of the proposed reasoning from abstract cases over classical hierarchical planning.