In this paper, the use of Large Language Models (LLMs) for the configuration of hybrid TSN/5G networks is investigated. We discuss promising use cases where LLMs offer significant potential to simplify complex configuration tasks. Particularly, we consider two important scenarios: In the first scenario, the LLM functions as an engineering component enhancing traditional network control entities such as the Centralized User Configurations (CUCs) and Centralized Network Configurations (CNCs) of TSN networks and interacting with the 5G control plane. In the second scenario, the LLM serves as an interactive assistance tool for users performing manual configuration tasks. For these use cases, an LLM-based architecture for network configuration is proposed, which consists of a Retrieval-Augmented Generation (RAG) system, a verification component, and an orchestration layer. Within the framework of this architecture, we introduce LLM-based methods to enhance the reliability of configuration in complex real-time networks, leveraging strategies such as divide and conquer, prompt engineering, and verification.

Artificial intelligence (AI) is becoming increasingly popular in the context of industrial manufacturing. However, in industrial manufacturing in particular, it is important to ensure the trustworthiness of AI. In this article, we give an overview of different aspects of trustworthy AI in this context. At first, we divide the topic into three different components, namely data, algorithm, and IT infrastructure. We identify several aspects of these components that are required for the trustworthy use of AI. Measures to achieve trustworthy AI are then derived and illustrated on the basis of a specific use case. It is further intended in the ongoing work to evaluate the impact of the individual measures.

Automated storage and retrieval systems (ASRS) are important in distribution centers and warehouses. To decrease cost or CO2emissions it is natural to optimize various aspects of an ASRS. In this work, we provide a concept for a two-phase optimization combining two important optimization tasks in ASRS: Given multiple rearrangement jobs, we first sequence these jobs to minimize the total travelling distance of the cranes. We continue the optimization by computing optimal trajectories for the sequence to guarantee energy efficient driving of the cranes. We describe our algorithms for a complex ASRS architecture with two cranes on parallel rails in one aisle. Additionally, we describe how to use our results for parallelization of crane movements in the considered warehouse architecture.

This paper presents a novel fault detection approach for industrial batch processes. The batch processes under consideration are characterized by the interaction between discrete system modes and non-stationary continuous dynamics. Therefore, a stochastic hybrid process model (SHPM) is introduced, where process variables are modeled as time-variant Gaussian distributions, which depend on hidden system modes. Transitions between the system modes are assumed to be either autonomous or to be triggered by observable events such as on/off signals. The model parameters are determined from training data using expectation-maximization techniques. A new fault detection algorithm is proposed, which assesses the likelihoods of sensor signals on the basis of the stochastic hybrid process model. Evaluation of the proposed fault detection system has been conducted for a penicillin production process, with the results showing a significant improvement over the existing baseline methods.Note to Practitioners—Automatic fault detection makes it possible to limit the effects of faults by taking countermeasures at an early stage. In this work, a data-driven fault detection method for industrial batch processes is proposed, in which the underlying process model is learned from training data. The proposed fault detection system can be used for various industrial batch processes without the need for complex and error-prone manual configuration. In contrast to many other data-driven approaches such as neural networks, only a few process cycles are required to create a robust process model. It should be noted that in data-driven fault detection methods, the training data should cover a large part of the process states that occur during error-free process cycles. The developed method is therefore particularly suitable for cyclical processes, which, however, can have alternative process paths and variability between the process cycles.

AI methods for fault handling in Cyber-Physical Production Systems (CPPS) such as production plants and tank systems are an emerging research topic. In the last years many methods for the detection of anomalies and faults, the diagnosis of the root cause and the automated repair have been developed. However, most of the methods are barely evaluated using a wide range of systems but applicability is shown using single use cases. In this paper, an ensemble of simulated benchmark systems is presented, which allows for a broad evaluation of AI methods for fault handling. The ensemble consists of seven different tank systems from process engineering with varying sizes and complexities and is made publicly available on Github. The suitability of the ensemble is shown using AI methods for fault handling such as anomaly detection, diagnosis and reconfiguration.

Manufacturing industry is facing major challenges to meet customer requirements, which are constantly changing. Therefore, products have to be manufactured with efficient processes, minimal interruptions, and low resource consumptions. To achieve this goal, huge amounts of data generated by industrial equipment needs to be managed and analyzed by modern technologies. Since the big data era in manufacturing industry is still at an early stage, there is a need for a reference architecture that incorporates big data and machine learning technologies and aligns with the Industrie 4.0 standards and requirements. In this paper, requirements for designing a scalable analytics platform for industrial data are derived from Industrie 4.0 standards and literature. Based on these requirements, a reference big data architecture for industrial machine learning applications is proposed and compared to related works. Finally, the proposed architecture has been implemented in the Lab Big Data at the SmartFactoryOWL and its scalability and performance have been evaluated on parallel computation of an industrial PCA model. The results show that the proposed architecture is linearly scalable and adaptable to machine learning use cases and will help to improve the industrial automation processes in production systems.

The complexity of industrial production systems is steadily growing. Hence, the plant stuff has to search in an increasing number of documents within the daily work routine, e.g. in manuals, commissioning instructions, service notes, shift books, process data, repair instructions, data sheets, R/I flow charts, CAD drawings etc. To support the plant stuff, an intelligent search engine for industrial production environments is proposed in this paper. Characteristics of the developed search engine with respect to the domain of industrial production environments, e.g. tailored synonym replacements and document classifications, are outlined. Particularly, two methods for document classifications, a k-nearest-neighbor classifier and a Naive Bayes classifier, are evaluated with documents from industrial production environments.

A large part of the programmable logic controls (PLCs) used in industrial automation systems is based on automata, which are employed to model the different stages of the automated processes and to determine the discrete control signals. Complex PLCs are typically composed of several parallel automata, which are related to a subset of the IO signals, respectively. In this paper, a novel model learning approach is proposed, which allows to learn the parallel automata from the discrete IO signals during normal operation of the PLC. Learning the parallel automata is accomplished by means of a synchronous side-by-side decomposition of the overall system model. The side-by-side decomposition is based on the clustering of the correlation matrix computed between the individual IO signals. The learnt automata can be employed for automatic fault detection and visualization of the normal operation of the PLC. Evaluations are conducted for both a baseline method, where a single automaton is learned as model for the complete system, and the proposed learning algorithm for parallel automata. Experimental results show that the computed parallel automata are superior to a single automaton with respect to compactness, accuracy and fault detection capabilities.

Wireless communication systems such as bluetooth, WLAN and RFID gain more and more importance in industrial production systems. However, high requirements with respect to availability and determinism have to be met in industrial environments with several sources of interference such as frequency converters and welding machines. Finding the optimal parameter configuration, which allows for low power consumption and reliable operation, is in many cases timeconsuming and error-prone. In this paper, a self-optimizing RFID system is presented, which adapts both the transmission frequency and the transmission power of the RFID reader to the system environment. The proposed method allows for a robust communication with low energy consumption. Evaluation has been conducted for an application scenario in the SmartFactory OWL. In the investigated settings, transmission power of the RFID reader could be reduced from 27dBm to 15.2dBm on average.

In the present work, fault detection in industrial automation processes is investigated. A fault detection method for observable process variables is extended for application cases, where the observations of process variables are noisy. The principle of this method consists in building a probability distribution model and evaluating the likelihood of observations under that model. The probability distribution model is based on a hybrid automaton which takes into account several system modes, i.e. phases with continuous system behaviour. Transitions between the modes are attributed to discrete control events such as on/off signals. The discrete event system composed of system modes and transitions is modeled as finite state machine. Continuous process behaviour in the particular system modes is modeled with stochastic state space models, which incorporate neural networks. Fault detection is accomplished by evaluation of the underlying probability distribution model with a particle filter. In doing so both the hybrid system model and a linear observation model for noisy observations are taken into account. Experimental results show superior fault detection performance compared to the baseline method for observable process variables. The runtime of the proposed fault detection method has been significantly reduced by parallel implementation on a GPU.

In this paper, an FPGA based FIFO with efficient memory management is proposed, which allows fast forwarding of real-time Ethernet frames. There are two main drawbacks of the existing FIFO implementations with respect to the buffering of Ethernet frames. Currentness of data is not guaranteed in case of buffer overflow because the new frames are dropped in this case. Furthermore, exhaustive resources are required for traffic priorization because an individual FIFO is required for each priority level. The proposed FIFO incorporates efficient strategies for both frame dropping and traffic priorization. The approach is based on a small ring buffer for meta data of individual frames and a page table that maps the frames to pages in RAM where the data bits of the frame are stored. The FIFO has been implemented on a low-cost Xilinx Spartan 6 FPGA. The solution requires little overhead for page table and ring buffer. Compared to an implementation with standard FIFOs that incorporates traffic priorization and frame dropping, RAM size is decreased from 44 kbytes to 2.7 kbytes.

In this paper, fault detection in piecewise stationary industrial processes is investigated. Such processes can be modeled as sequences of distinct system modes in which the respective expectation values and variances of process variables do not change. In particular, piecewise stationary processes with autonomous transitions between system modes are considered in this work, i.e. processes without observable trigger events such as on/off signals. A Hidden Markov Model (HMM) is employed as underlying system model for such processes. System modes are modeled as hidden state variables with given transition probabilities. Continuous process variables are assumed to be Gaussian distributed with constant second order statistics in each system mode. A novel HMM-based fault detection method is proposed which incorporates the Viterbi algorithm into a fault detection method for hybrid industrial processes. Experimental results for the proposed fault detection method are presented for a module of the Lemgo Smart Factory.

This paper addresses the efficient generation of models for cyber-physical systems from large historical data sets. A cyber-physical system (CPS) is a system composed of physical subsystems together with computing and networking. CPS models are required for monitoring and control of the physical processes. Such models are in general hybrid models that take into account both discrete control signals and continuous system behaviour. Model-learning is the key to a new generation of intelligent automation systems: Automatic generation of models from system observations allows to model complex CPS in cases where manual model creation is time-consuming, expensive or not even possible. In general, the quality of the generated models increases with the size of training data. However, model learning from large historical data sets is in many cases time-consuming. For this reason, MapReduce algorithms are proposed in the present work that allow for efficient model learning.

In this paper, stochastic models for fault detection in industrial automation processes are investigated. Thereby, nonlinear, time-variant systems are considered. The basic idea consists in building a probability distribution model and evaluating the likelihood of observations under that model. In contrast to the existing methods, this paper considers the practically important case in which measurement noise is negligible and all process variables are observable. This assumption allows the direct evaluation of a probability distribution for fault detection without approximations such as second order statistics or particles. The main part of this paper deals with adequate models for this probability distribution such as Gaussian and Hidden Markov models. Such models require predictions of the expectation values of the respective probability distributions. Regression models such as (multivariate) linear regression models and neural networks are investigated for this purpose. Evaluations are conducted with respect to prediction accuracies and fault detection capabilities of the employed models. Evaluations show superior results of the novel approach compared to existing fault detection methods, which are based on approximations such as second order statistics.

Model-learning is the key to the new generation of intelligent automation systems: Without the automatic generation of models from system observations, models of the plant's behavior will not be available for most systems. And without such models, no intelligent capabilities such as self-diagnosis or self-optimization can be implemented. This paper therefore presents a novel classification schema for systems, models and model learning algorithms. Based on this analysis of open research questions, the new learning algorithm HyBUTLA is presented. In this paper, this solution approach is applied to the analysis and diagnosis of the energy consumption of production plants. To the best of the authors' knowledge, this is the first learning and adaptable energy anomaly detection solution for complex hybrid production systems.

In the presented work, the detection of anomalous energy consumption in hybrid industrial production systems is investigated. A model-based approach with a timed hybrid automaton as overall system model is employed for anomaly detection. The approach is based on the assumption of several system modes, i.e. phases with continuous system behavior. Transitions between the modes are attributed to discrete control events such as on/off signals. The underlying discrete event system which comprises both system modes and transitions is modeled as finite state machine. The focus of this paper is set on the modeling of the energy consumption in the particular system modes. Sequences of stochastic state space models are employed for this purpose. Model learning and anomaly detection for this approach are considered. The proposed approach is further evaluated in a small model factory. The experimental results show significant improvements compared to existing approaches to anomaly detection in hybrid industrial systems.

In automatic speech recognition, hidden Markov models (HMMs) are commonly used for speech decoding, while switching linear dynamic models (SLDMs) can be employed for a preceding model-based speech feature enhancement. In this paper, these model types are combined in order to obtain a novel iterative speech feature enhancement and recognition architecture. It is shown that speech feature enhancement with SLDMs can be improved by feeding back information from the HMM to the enhancement stage. Two different feedback structures are derived. In the first, the posteriors of the HMM states are used to control the model probabilities of the SLDMs, while in the second they are employed to directly influence the estimate of the speech feature distribution. Both approaches lead to improvements in recognition accuracy both on the AURORA2 and AURORA4 databases compared to non-iterative speech feature enhancement with SLDMs. It is also shown that a combination with uncertainty decoding further enhances performance.

In this paper, parameter estimation of a state-space model of noise or noisy speech cepstra is investigated. A blockwise EM algorithm is derived for the estimation of the state and observation noise covariance from noise-only input data. It is supposed to be used during the offline training mode of a speech recognizer. Further a sequential online EM algorithm is developed to adapt the observation noise covariance on noisy speech cepstra at its input. The estimated parameters are then used in model-based speech feature enhancement for noise-robust automatic speech recognition. Experiments on the AURORA4 database lead to improved recognition results with a linear state model compared to the assumption of stationary noise.

In this paper a switching linear dynamical model (SLDM) approach for speech feature enhancement is improved by employing more accurate models for the dynamics of speech and noise. The model of the clean speech feature trajectory is improved by augmenting the state vector to capture information derived from the delta features. Further a hidden noise state variable is introduced to obtain a more elaborated model for the noise dynamics. Approximate Bayesian inference in the SLDM is carried out by a bank of extended Kalman filters, whose outputs are combined according to the a posteriori probability of the individual state models. Experimental results on the AURORA2 database show improved recognition accuracy.

A marginalized particle filter is proposed for performing single channel speech enhancement with a non-linear dynamic state model. The system consists of a particle filter for tracking line spectral pair (LSP) parameters and a Kalman filter per particle for speech enhancement. The state model for the LSPs has been learnt on clean speech training data. In our approach parameters and speech samples are processed at different time scales by assuming the parameters to be constant for small blocks of data. Further enhancement is obtained by an iteration which can be applied on these small blocks. The experiments show that similar SNR gains are obtained as with the Kalman-LM-iterative algorithm. However better values of the noise level and the log-spectral distance are achieved