Large Language Models (LLMs) offer powerful reasoning capabilities for industrial process control, yet their non-deterministic nature, susceptibility to hallucination, and lack of intrinsic physical understanding make direct deployment in safety-critical environments unacceptable. This paper addresses five research questions on safely integrating LLM-based reasoning into industrial process automation through the Autonomous Action Execution (AAE) framework. For safe architectural integration (RQ1), we present a four-layer multi-agent architecture that confines LLM inference to an observation-only Monitor layer while safety-critical decisions are made by deterministic Verification and Execution agents. For structuring heterogeneous plant data (RQ2), we introduce a text-level aggregation framework with pluggable analyzers that transforms SCADA states, time-series measurements, Piping and Instrumentation Diagrams (P IDs), and Standard Operating Procedures (SOPs) into contextually rich documents for LLM consumption. For automated validation (RQ3), a P ID-grounded method uses graph traversal over the P ID topology to verify physical consistency of LLM-generated proposals, checking tag existence, actuatability, fail-state consistency, and downstream impact. For quantifiable context enrichment (RQ4), a graduated baseline comparison (B0–B3) demonstrates the incremental value of each pipeline component. For cross-domain generalisability (RQ5), evaluation across five industrial scenarios—three derived from the Tennessee Eastman Process (TEP) benchmark (Downs Vogel, 1993) providing community-standard validation, plus two retained scenarios (PolyReactor, Dryer) that establish performance boundaries from best-case (zero hallucination) to worst-case (70

Traditional information retrieval systems rank documents by semantic similarity, yet this fails to capture document utility for autonomous agents in multi-step tasks. We introduce AgentRank, a ranking framework that learns from agent execution trajectories to identify documents leading to task completion. Using dual PageRank on success/failure trajectory graphs, Thompson Sampling, and terminal path analysis, AgentRank reduces semantic dead-ends by 79% compared to the strongest baseline (LinUCB contextual bandit) while maintaining perfect task completion rates in our controlled evaluation. Ablation analysis identifies terminal path bonus as the most critical component, and sensitivity analysis characterizes the operating regime where trajectory learning provides significant benefit.

Vision Foundation Models (VFMs) have demonstrated remarkable performance on a wide range of visual tasks involving everyday objects, largely due to their large-scale pre-training on natural image datasets. However, their effectiveness in industrial domains remains limited. Industrial applications such as anomaly detection and quality inspection pose unique challenges that differ significantly from those encountered in natural image settings. In this paper, we present evaluation of several state-of-the-art pretrained VFMs on industrial datasets, focusing on the task of multi-label image segmentation. Our findings reveal limitations of directly applying VFMs to industrial scenarios. We further discuss the implications of these results and propose key considerations and best practices for adapting VFMs to industrial use cases. These insights aim to guide future research toward the development of industrial-grade vision foundation models that are robust, efficient, and tailored to domain-specific requirements.

Vibration-based condition monitoring systems are receiving increasing attention due to their ability to accurately identify different conditions by capturing dynamic features over a broad frequency range. However, there is little research on clustering approaches in vibration data and the resulting solutions are often optimized for a single data set. In this work, we present an extensive comparison of the clustering algorithms K-means clustering, OPTICS, and Gaussian mixture model clustering (GMM) applied to statistical features extracted from the time and frequency domains of vibration data sets. Furthermore, we investigate the influence of feature combinations, feature selection using principal component analysis (PCA), and the specified number of clusters on the performance of the clustering algorithms. We conducted this comparison in terms of a grid search using three different benchmark data sets. Our work showed that averaging (Mean, Median) and variance-based features (Standard Deviation, Interquartile Range) performed significantly better than shape-based features (Skewness, Kurtosis). In addition, K-means outperformed GMM slightly for these data sets, whereas OPTICS performed significantly worse. We were also able to show that feature combinations as well as PCA feature selection did not result in any significant performance improvements. With an increase in the specified number of clusters, clustering algorithms performed better, although there were some specific algorithmic restrictions.

Federated Learning (FL) has gained prominence as a method for facilitating collaborative and privacy-preserving model training across multiple heterogeneous devices in recent years. In most approaches, the clients are closely deployed to the data source. However, as FL systems are implemented in the industry, multiple platform options can be considered in the design phase.In this paper, we present a novel approach for deploying FL clients to multiple locations considering a multi-platform strategy with cloud, fog, and edge resources. We provide a FL architecture that integrates mechanisms for building cohorts of similar clients and a client selection algorithm for optimizing the performance of all clients with respect to energy consumption, model performance, and FL completion time.We evaluate seven deployment strategies in three scenarios given a real-world use case from the electronics industry and heterogeneous hardware capabilities. Our results show that our approach can improve model performance by up to 15%, while energy consumption and completion time converge to the relatively best deployment.

Federated Learning (FL) holds immense potential for transforming the industrial landscape by leveraging distributed data to solve Artificial Intelligence (AI) use cases with collectively trained models in a privacy-preserving way. In this regard, Industrial FL (IFL) arose as a collaborative approach for training AI models between multiple industry partners and devices without the need to share the actual training data. However, despite its promising prospects, the transition and successful implementation of FL in practice is currently lagging behind, posing challenges for industrial companies. To address this, it is of crucial relevance to analyze different business types and involved stakeholders to be able to design FL-based solutions tailored to the industries needs. This paper presents the results of 13 semi-structured interviews conducted in Austrian industries, involving 11 companies from different domains. We identify AI applications, pain points, and attitudes towards AI and FL. Based on the interviews, three industry personas are derived, namely, service business, production optimization, and complex product and project business. To address the needs of these personas, three collaborative FL solution blueprints are proposed. The blueprints include system architectures, implementation steps, and collaboration modes for the involved parties. The blueprints are discussed based on dimensions such as FL paradigm, collaboration mode, key benefits, main addressed needs, and challenges.

Monitoring of batch production processes is complex and existing solutions do not offer good performance in providing real-time feedback about the state of the process. Therefore, we introduce an AI system that monitors a fermentation process and detects deviations from the normal process execution directly on the edge and provides real-time feedback to the operator, allowing intervention before the process gets out of control. We analyze the accuracy of the novel AI-based approach by carrying out several experiments and compare the outcome with statistical methods as a baseline. The experiments show that the AI-based approach performs significantly better at detecting anomalies in a fermentation process than the statistical methods.

In industrial settings, traditional centralized ap-proaches for training AI models can be insufficient due to limited training data. Industrial Federated Learning (IFL) offers a promising solution by enabling collaborative training across multiple industrial devices, while keeping data on-premises. In this paper, we propose a novel approach for supporting the development, deployment, integration and execution of IFL solutions. The proposed method provides a lifecycle management of FL artifacts and supports FL as a Service (FlaaS). This enables the extensibility and customizability of FL-based edge applications in industrial settings. Additionally, we introduce a federated clustering algorithm that we have integrated into a condition monitoring app running on client locations to evaluate the proposed lifecycle management. We run two scenarios with four and 33 clients using real-world time series data from industrial pumps. Our results show the applicability of the implemented lifecycle management and demonstrates that privacy-preserving approaches compete well with privacy-disclosing ones.

We propose a new sampling strategy, called smart active sapling, for quality inspections outside the production line. Based on the principles of active learning a machine learning model decides which samples are sent to quality inspection. On the one hand, this minimizes the production of scrap parts due to earlier detection of quality violations. On the other hand, quality inspection costs are reduced for smooth operation.

As the adoption of federated learning (FL) in the manufacturing industry grows and systems get increasingly complex, a need to inspect their behavior arises. Stakeholders of the FL process want a more transparent system to understand the current state and analyze how its performance changed over time. However, current representation approaches are often not designed for industrial applications and do not cover the entire FL model lifecycle. We propose the lifecycle dashboard, which considers the different requirements and perspectives of industrial stakeholders by visualizing information from the FL server. In addition, our representation approach is generic enough to be applied to different use cases and industries. We evaluate the lifecycle dashboard in a semistructured expert interview, show improvements in the understandability of FL systems, and discuss possible use cases in the industry.

Data-driven machine learning is playing a crucial role in the advancements of Industry 4.0, specifically in enhancing predictive maintenance and quality inspection. Federated learning (FL) enables multiple participants to develop a machine learning model without compromising the privacy and confidentiality of their data. In this paper, we evaluate the performance of different FL aggregation methods and compare them to central and local training approaches. Our study is based on four datasets with varying data distributions. The results indicate that the performance of FL is highly dependent on the data and its distribution among clients. In some scenarios, FL can be an effective alternative to traditional central or local training methods. Additionally, we introduce a new federated learning dataset from a real-world quality inspection setting.

Artificial Intelligence (AI) is increasingly explored in various domains and industries. Many companies experiment with AI, but too often those experiments are one-off analyses based on outdated data and the resulting models never make it into production. This paper proposes a framework for building and operating AI models at the industrial edge. The center of this framework is the model artifact, a model-generating entity. We analyze three AI model use-cases and user roles involved in industrial AI applications to illustrate the challenges in deploying and operating AI applications in industrial edge scenarios. We propose to structure the AI models into predefined artifacts that enable deployments with only a few clicks. The edge device links sensor data with the model input and returns the model output as feedback back into the industrial process. Model training, deployment, and management can be carried out in a scalable manner even by a non-expert. Several models can be managed in parallel, and data can be linked to the respective sensor or machine.

Machine Learning (ML) is increasingly applied in industrial manufacturing, but often performance is limited due to insufficient training data. While ML models can benefit from collaboration, due to privacy concerns, individual manufacturers cannot share data directly. Federated Learning (FL) enables collaborative training of ML models without revealing raw data. However, current FL approaches fail to take the characteristics and requirements of industrial clients into account. In this work, we propose a FL system consisting of a process description and a software architecture to provide \acrfull{flaas} to industrial clients deployed to edge devices. Our approach deals with skewed data by organizing clients into cohorts with similar data distributions. We evaluated the system on two industrial datasets. We show how the FLaaS approach provides FL to client processes by considering their requests submitted to the Industrial Federated Learning (IFL) Services API. Experiments on both industrial datasets and different FL algorithms show that the proposed cohort building can increase the ML model performance notably.

Federated Learning (FL) decouples model training from the need for direct access to the data and allows organizations to collaborate with industry partners to reach a satisfying level of performance without sharing vulnerable business information. The performance of a machine learning algorithm is highly sensitive to the choice of its hyperparameters. In an FL setting, hyperparameter optimization poses new challenges. In this work, we investigated the impact of different hyperparameter optimization approaches in an FL system. In an effort to reduce communication costs, a critical bottleneck in FL, we investigated a local hyperparameter optimization approach that – in contrast to a global hyperparameter optimization approach – allows every client to have its own hyperparameter configuration. We implemented these approaches based on grid search and Bayesian optimization and evaluated the algorithms on the MNIST data set using an i.i.d. partition and on an Internet of Things (IoT) sensor based industrial data set using a non-i.i.d. partition.

The increasing deployment of low-cost industrial IoT (IIoT) sensor platforms on industrial assets enables great opportunities for anomaly classification in industrial plants. The performance of such a classification model depends highly on the available training data. Models perform well when the training data comes from the same machine. However, as soon as the machine is changed, repaired, or put into operation in a different environment, the prediction often fails. For this reason, we investigate whether it is feasible to have a robust and transferable method for AI based anomaly classification using different models and pre-processing steps on centrifugal pumps which are dismantled and put back into operation in the same as well as in different environments. Further, we investigate the model performance on different pumps from the same type compared to those from the training data.

Anomaly detection in large industrial cooling systems is very challenging due to the high data dimensionality, inconsistent sensor recordings, and lack of labels. The state of the art for automated anomaly detection in these systems typically relies on expert knowledge and thresholds. However, data is viewed isolated and complex, multivariate relationships are neglected. In this work, we present an autoencoder based end-to-end workflow for anomaly detection suitable for multivariate time series data in large industrial cooling systems, including explained fault localization and root cause analysis based on expert knowledge. We identify system failures using a threshold on the total reconstruction error (autoencoder reconstruction error including all sensor signals). For fault localization, we compute the individual reconstruction error (autoencoder reconstruction error for each sensor signal) allowing us to identify the signals that contribute most to the total reconstruction error. Expert knowledge is provided via look-up table enabling root-cause analysis and assignment to the affected subsystem. We demonstrated our findings in a cooling system unit including 34 sensors over a 8-months’ time period using 4-fold cross validation approaches and automatically created labels based on thresholds provided by domain experts. Using 4-fold cross validation, we reached a F1-score of 0.56, whereas the autoencoder results showed a higher consistency score (CS of 0.92) compared to the automatically created labels (CS of 0.62) – indicating that theanomaly is recognized in a very stable manner. The automatically created labels, however, detected anomaly earlier. The main anomaly was found by the autoencoder and automatically created labels, and was also recorded in the log files. Further, the explained fault localization highlighted the most affected component for the main anomaly in a very consistent manner.

The increasing deployment of low-cost IoT sensor platforms in industry boosts the demand for anomaly detection solutions that fulfill two key requirements: minimal configuration effort and easy transferability across equipment. Recent advances in deep learning, especially long-short-term memory (LSTM) and autoencoders, offer promising methods for detecting anomalies in sensor data recordings. We compared autoencoders with various architectures such as deep neural networks (DNN), LSTMs and convolutional neural networks (CNN) using a simple benchmark dataset, which we generated by operating a peristaltic pump under various operating conditions and inducing anomalies manually. Our preliminary results indicate that a single model can detect anomalies under various operating conditions on a four-dimensional data set without any specific feature engineering for each operating condition. We consider this work as being the first step towards a generic anomaly detection method, which is applicable for a wide range of industrial equipment.

An increasing number of industrial assets are equipped with IoT sensor platforms and the industry now expects data-driven maintenance strategies with minimal deployment costs. However, gathering labeled training data for supervised tasks such as anomaly detection is costly and often difficult to implement in operational environments. Therefore, this work aims to design and implement a solution that reduces the required amount of data for training anomaly classification models on time series sensor data and thereby brings down the overall deployment effort of IoT anomaly detection sensors. We set up several in-lab experiments using three peristaltic pumps and investigated approaches for transferring trained anomaly detection models across assets of the same type. Our experiments achieved promising effectiveness and provide initial evidence that transfer learning could be a suitable strategy for using pretrained anomaly classification models across industrial assets of the same type with minimal prior labeling and training effort. This work could serve as a starting point for more general, pretrained sensor data embeddings, applicable to a wide range of assets.

An important prerequisite for determining whether a certain product is producible in any given production facility is an accurate assessment of which production lines and/or the machines are able to execute the necessary production steps. Not only the static information about the capabilities of the machines, but also the conditions of machines and tools are significant for this analysis. Because of the deviation of machine capabilities with increasing deterioration and weary of the equipment, it is also necessary to continuously monitor the status of the machine and analyze the machine conditions. In this paper, we present an approach for generating production plans across multiple factories, considering both static information and dynamic data analysis. Edge devices constantly monitor high frequency machine data and report condensed machine states to an Industrial IoT platform (IIoT). A marketplace within the cloud-application MindSphere enables us to integrate the requirements of the products and the capabilities of the production sites. Customers are be able to evaluate these production plans based on duration, energy consumption, CO 2 footprint etc.