Alarm floods can overload control room operators during abnormal operation of process plants. Online ”alarm floodclassification” (AFC) assigns an unfolding alarm-event stream to a predefined alarm flood class while only a prefix of theepisode is observable. In deployment, the observed prefix can deviate from curated historical logs due to two distinctuncertainty sources: (i) alarm-log degradations within extracted episodes, including missing events, spurious events,and timing uncertainty that can alter recovered event ordering; and (ii) pipeline-induced uncertainty, including delayedalarm flood detection. We propose AFC-RobustBench, a perturbation-based statistical robustness benchmarkingprotocol for prefix-based online AFC that models both these sources as perturbation functions acting on alarm-eventstreams and supports reproducible mixed-perturbation composition. The method-agnostic protocol reports degradationprofiles over severity and observation progress, scalar robustness scores, and uncertainty estimates across Monte-Carlodraws and repeated folds. Applying this benchmarking protocol, we benchmarked seven relevant AFC methods on twoprocess-industry datasets. Results indicate that nominal performance on clean test data can misrepresent behavior underdegraded alarm logs and detection delays, and that clean-performance rankings can differ substantially from robustnessrankings under online, prefix-based observation. These findings demonstrate the usefulness of AFC-RobustBench forassessing AFC methods under realistic degraded and delayed observation conditions.

Cyber-physical system (CPS) forecasting models depend on sensor streams with noisy, biased, missing, or temporally misaligned readings, yet standard forecasting evaluation often selects models by nominal error without showing whether they remain robust under such faults. We introduce SensorFault-Bench, a shared CPS-grounded sensor-fault stress-test protocol for evaluating forecasting architectures and robustness-improvement methods, and an operational taxonomy organizing the method comparison. Across four real-world datasets and eight scored scenarios governed by a standardized severity model, it reports worst-scenario degradation, clean mean squared error (MSE), and worst-scenario fault-time MSE, separating relative robustness from absolute error. A disjoint fault-transfer split lets explicit fault-training methods train on adjacent fault families while evaluation uses separate benchmark scenarios. Empirically, forecasting architectures favored by clean MSE can degrade sharply under faults, and clean-MSE rankings can disagree with worst-scenario fault-time error rankings. Chronos-2, the evaluated zero-shot foundation-model representative, matches or trails the last-value naive forecaster in clean MSE on the two single-target datasets and has the largest worst-scenario degradation on ETTh1 and Traffic, where all channels are forecast targets. For the evaluated robustness-improvement method set, paired deltas show selective degradation reductions: projected gradient descent adversarial training and randomized training lead where value faults dominate observed degradation, while fault augmentation leads where availability faults dominate. SensorFault-Bench provides open-source code, documented data access, and reproduction and extension guides, so new datasets, architectures, and robustness-improvement methods can be evaluated under the same CPS sensor-fault robustness protocol.

The Asset Administration Shell (AAS), as a core concept of Industry 4.0, provides a standardized digital representation of assets, enabling interoperability, modularization, and integration across industrial environments. This paper extends the application of the AAS framework to alarm lifecycle management by introducing a dedicated submodel for representing process alarm in AAS format. Furthermore, an event-driven notification mechanism is implemented to support standardized, cross-organizational exchange of alarm information between diverse industrial systems. The integration of this mechanism with the Asset Interface Description (AID) submodel ensures semantic consistency, scalability, and interoperability in alarm communication. The proposed AAS-based approach enhances the transparency and traceability of alarm data across the entire lifecycle, contributing to more reliable and interoperable alarm management within Industry 4.0 ecosystems.

Alarm floods remain a significant challenge in industrial processes. However, benchmark datasets for evaluating alarm flood analysis methods remain scarce. This paper introduces a novel dataset derived from a fluid catalytic cracking unit simulation model. We adapt the original simulation by incorporating an alarm system and defining realistic disturbances. The dataset's utility is demonstrated through an evaluation of three relevant alarm flood classification methods from the literature. Copyright (c) 2025 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)

Alarm floods remain a challenge in industrial operations, potentially overwhelming human operators with excessive alarm notifications during abnormal situations. To address this, alarm flood classification (AFC) methods utilize historical data to classify recurring alarm patterns automatically. However, the practical utility of these methods may be limited by potential degradation in alarm data quality, resulting from sensor faults, communication errors, or detection delays, which can substantially compromise their classification accuracy and reliability. This paper proposes a novel methodology to systematically analyze the robustness of AFC methods against realistic alarm data quality issues. We introduce four distinct perturbations: missing alarms, false alarms, delayed alarm flood detection, and alarm reordering, to replicate real-world alarm data degradation. We evaluate our methodology using a novel alarm dataset derived from the Tennessee-Eastman process, while examining the robustness of six relevant AFC methods from the literature. The results demonstrate significant variations in robustness across different AFC methods and perturbation types, providing insights into their practical reliability under various realistic scenarios.

Alarm floods present challenges to human operators in industrial process plants, potentially impairing their situational awareness. Alarm flood classification addresses this issue by assigning observed alarm sequences to known alarm flood classes. However, the dynamic nature of evolving alarm sequences can lead to temporary class ambiguities, where insufficient information at early stages makes it challenging to distinguish between alarm flood classes. Recent advancements integrated conformal prediction and alarm flood classification to address this challenge, providing dynamically updated sets of plausible alarm flood classes. Despite their promise, such methods require extensive calibration datasets, which may be scarce in industrial contexts. We propose a novel method leveraging data-efficient crossconformal prediction to reduce calibration data requirements. Additionally, a novel postprocessing step is introduced to resolve erroneously empty prediction sets. Evaluated on a synthetic alarm dataset, our approach is compared to an existing inductive conformal prediction method across three alarm flood classification methods from the literature. The results demonstrate our method's effectiveness in maintaining a high coverage and a relatively small set of plausible alarm flood classes, while reducing calibration demands compared to the existing method.

This paper introduces a contextual, continuous risk metric for predictive safety validation in unmanned autonomous vehicles. The proposed method employs an adaptive ovoidal envelope around unmanned autonomous vehicles, dynamically adjusting based on vehicle kinematics, environmental disturbances, sensor uncertainties, and hardware reliability. The risk metric smoothly quantifies safety violations, serving as a prerequisite for generating counterfactual explanations, which could be used to avoid safety-critical situations. A preliminary simulation demonstrates the metric’s context sensitivity, continuity, computational efficiency, and predictive capability.

Supervised machine learning has presented impressive results in the industrial domain - when it can depend on reliable ground truth labels. However, the concept of golden labels often represents only an idealization of reality. Instead, data labeling can be highly subjective and hence susceptible to corruption, such as unintentional errors, resulting in socalled label noise. Since label noise can significantly impair the performance of supervised machine learning models, it can be helpful to assess their robustness to label noise when evaluating existing limitations. Despite the large number of publications offering best practices and techniques for making models more robust to label noise, literature lacks generic tools that would allow data scientists and industrial practitioners to test and compare which technique works best for their given model. To address this gap, we propose a model-agnostic, reusable framework for measuring the robustness of supervised machine learning models to label noise.

Alarm floods in industrial plants overwhelm operators by triggering numerous alarms within short time intervals, significantly complicating effective root-cause analysis. Existing causal analysis methods can support operators, but typically either neglect conditional causal relationships inherent in control loops or require manual, time-consuming extraction. This paper introduces a novel automated methodology leveraging Large Language Models (LLMs) and the logic programming language Prolog to systematically extract conditional causal relationships directly from readily available textual control narratives, textual engineering documents that describe the control system in natural language. Our logic-first approach prioritizes a thorough logical analysis of control system behavior before Prolog rule generation. Evaluations on a synthetic control system confirm accurate representation of cascade and selector control logic, demonstrating the method’s capability to reliably automate causal rule extraction and effectively support root-cause analysis during alarm floods.

Online alarm flood classification (AFC) methods support operators by assigning alarm sequences to known alarm flood classes. However, temporary class ambiguities can arise due to limited data. To address this, we extend conformal prediction-based AFC with a method to predict bifurcation points-moments when the predicted set of plausible classes becomes more specific. We use random forest regression with jackknife+ for quantile regression and introduce a delay timer to stabilize predictions. Evaluation on two datasets demonstrates that our approach has the potential to effectively forecast upcoming bifurcation points and improve prediction stability, offering decision support under uncertainty during industrial alarm floods. Copyright (c) 2025 The Authors. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/)

Supervised machine learning (ML) has achieved significant outcomes in the industrial domain, dependent on the availability and accuracy of ground truth labels. However, the usual assumption of a ground truth to exist in the data often represents only an idealization of real-world conditions, as data labeling can be subjective and prone to errors, leading to so-called label noise. Such noise can significantly degrade model performance. Although extensive research has identified methods to improve model robustness against label noise, there is a notable absence of generic, reusable frameworks that allow industrial practitioners to systematically assess model robustness. To address this gap, we extend our previous work and propose a model-agnostic framework designed specifically for evaluating robustness against label noise. Our framework incorporates two distinct label noise perturbation mechanisms: an instance-independent symmetric perturber and an instance-dependent one. We demonstrate the utility of our extended framework through empirical evaluations on two industrial datasets using six relevant time series classification methods from the literature. The results highlight the significant vulnerability of supervised ML models to both noise types and underscore the value of our framework in uncovering these robustness limitations.

Alarm management systems in the process industry use root-cause analysis methods to reduce alarm logs. To enable the application of these methods in different plant types, the alarm characteristics of a continuous, two discrete, and a hybrid plant are examined. The main contribution is threefold. First, root-cause analysis requirements, posed by different plant types, are revealed. Next, existing approaches are assessed against the requirements. Since the root-cause is not necessarily the first alarm in time, its justification requires further plant knowledge. Thus, engineering documents and the necessary formalized knowledge to justify root-causes are identified.

Alarm flood classification (AFC) methods are crucial in assisting human operators to identify and mitigate the overwhelming occurrences of alarm floods in industrial process plants, a challenge exacerbated by the complexity and data-intensive nature of modern process control systems. These alarm floods can significantly impair situational awareness and hinder decision-making. Existing AFC methods face difficulties in dealing with the inherent ambiguity in alarm sequences and the task of identifying novel, previously unobserved alarm floods. As a result, they often fail to accurately classify alarm floods. Addressing these significant limitations, this paper introduces a novel three-tier AFC method that uses alarm time series as input. In the transformation stage, alarm floods are subjected to an ensemble of convolutional kernel-based transformations (MultiRocket) to extract their characteristic dynamic properties, which are then fed into the classification stage, where a linear ridge regression classifier ensemble is used to identify recurring alarm floods. In the final novelty detection stage, the local outlier probability (LoOP) is used to determine a confidence measure of whether the classified alarm flood truly belongs to a known or previously unobserved class. Our method has been thoroughly validated using a publicly available dataset based on the Tennessee-Eastman process. The results show that our method outperforms two naive baselines and four existing AFC methods from the literature in terms of overall classification performance as well as the ability to optimize the balance between accurately identifying alarm floods from known classes and detecting alarm flood classes that have not been observed before. Impact Statement We introduce the convolutional kernel-based alarm subsequence identification method (CASIM), which improves industrial alarm flood classification. CASIM extracts a wide range of alarm dynamics, unlike previous approaches that use a limited set of alarm characteristics. This helps CASIM identify more relevant features, improving its ability to classify complex alarm floods. Moreover, expanding windows in CASIM's online application, inspired by early time series classification, allows alarm flood classification over time. Our evaluation shows that this can provide faster and more accurate insights than existing methods. We believe that our proposed method CASIM can improve operational decision-making and reduce operator effort. By making the implementation of our method publicly available, we aim to encourage wider adoption and research in the field.

Alarm flood management is essential for industrial process plant safety and efficiency. Online “alarm flood classification” (AFC) assigns an observed sequence of alarms to one (of many) alarm flood classes known from the past. Nevertheless, accurately differentiating between alarm flood classes as they evolve over time remains a challenge due to classification uncertainty. Furthermore, AFC usually assigns to just one alarm flood class, preventing the operator from recognizing that multiple classes should be considered given the limited information at hand. This paper proposes a novel approach by integrating “conformal prediction” with “early time series classification” to address these challenges, offering a probabilistic understanding of alarm flood classes to be considered, and enhancing decision-support for operators. Using a method-agnostic framework, our method can be used with existing AFC models, making them adaptable to evolving alarm dynamics. We apply our approach to five AFC methods from literature using a novel and openly accessible synthetic alarm dataset we designed. Our evaluation shows how our method can improve coverage of the existing models, particularly in the early stages of alarm floods, ensuring with high confidence that the true class is captured while preserving a reasonably small set of possible alarm flood classes.

This paper introduces a novel alarm dataset specifically designed for the evaluation of alarm analysis methods within process plants. The complexity of industrial systems and the demands for operational safety and efficiency underscore the critical need for advanced diagnostic tools capable of handling alarm floods-situations where numerous alarms are triggered simultaneously. To bridge the gap identified in existing research regarding the availability of alarm datasets, we have developed a novel publicly available dataset derived from simulated data of a nuclear power plant. This dataset allows for a detailed analysis of alarm dynamics and enables a comprehensive evaluation of alarm analysis methods. We present a systematic methodology for generating alarm data, which involves setting alarm thresholds based on the trade-off between “false alarm rates” (FAR) and “missed alarm rates” (MAR). The dataset is employed to evaluate three existing “alarm flood classification” (AFC) methods, showcasing the practical implications and benefits of our approach. We demonstrate that AFC methods exhibit varying performances based on the implemented alarm thresholds and the quantity of available alarm data.

In this paper, we propose a novel method for enhancing the explainability of alarm flood classification results using and adapting concepts from the field of explainable artificial intelligence. Alarm flood classification methods are helpful in managing complex industrial processes; however, their predictions can be challenging to understand and justify, especially for operators without expertise in machine learning. Our proposed model-agnostic method generates counterfactual alarm floods to provide explanations for classification results obtained from any alarm flood classification model. By examining the differences between the original alarm flood and counterfactuals, we provide actionable insights for plant operators for decision-making and understanding the underlying dynamics of alarm floods. We demonstrate the effectiveness of our approach by experiments on three state-of-the-art alarm floods classification methods and an openly accessible dataset based on the “Tennes-see-Eastman” process, showcasing the added value of our method in improving the explainability and trustworthiness of alarm flood classification results.

We present a novel counterfactual-based dashboard for explainable artificial intelligence (XAI) in process industries, aimed at enhancing the understanding and adoption of machine learning (ML) models by providing transparency, explainability, and performance evaluation. Our dashboard comprises two modules: a statistical analysis module for data visualization and model performance assessment, and an XAI module for exploring counterfactual explanations at varying levels of abstraction. Through a case study of an industrial batch process, we demonstrate the dashboard’s applicability and potential to increase trust in ML models among stakeholders, paving the way for confident deployment in process industries.

Industrial ML models are primarily data-driven. Therefore, one of the main focus for monitoring the model should be towards identifying the drifts in the data that might affect the performance of the model. The traditional drift detecting methods are usually based on some assumptions related to the underlying data such as no inter-dependence. However industrial sensor data typically consists of time series data, which is collected at regular intervals. Therefore, detecting drift in dependent data where the current readings depend on the previously registered readings demands a different approach. Existing solutions require either the ground truth, a fixed size, or the underlying model details. We propose an Unsupervised Drift Detection method for industrial Time series data or UDDT, a generic approach with no such pre-requisites. In our approach, we can check whether two series belong to the same model. Apart from detecting the drift in the two series, it can also provide the rationale behind the observed drift, i.e., whether the drift is due to a difference in stationarity, correlation structures, or noise distributions. We evaluate the UDDT on two datasets to demonstrate its correctness and the trust regions under various circumstances.

The performance of machine learning models can be significantly impacted by variations in data quality. Typically, conventional model testing does not examine how robust the model would be in the face of potential data quality deterioration. In an industrial use case, however, data quality is a pertinent issue, as sensors are susceptible to a variety of technical and external issues that may result in poor data quality over time. In order to develop robust machine learning models, industrial data scientists must understand the sensitivity of their models against data quality issues, through the application of an appropriate and comprehensive testing solution. In this work, we propose a generic framework for systematically analyzing the impact of data quality issues on the performance of machine learning models by intentionally applying gradual perturbations to the original time series data. The evaluation is performed using a benchmark industrial process consisting of multivariate time series from sensors in a complex chemical process.

Managing industrial alarm floods is a challenge that demands analytical tools that assist human operators online. While existing methods offer some insight into causality analysis, they often fall short of addressing dynamic causal dependencies, particularly during the plant-wide propagation of abnormal situations. Additionally, evolving alarm floods, indicative of escalating disturbances, have received limited attention. This paper introduces a novel approach for analyzing causality in alarm floods. By combining a data-driven nonlinear causality estimator with supplementary process information and operator knowledge, the proposed method effectively assesses dynamic causal dependencies among relevant process variables. Moreover, we present an operator-centric tool that visualizes these dependencies, enabling operators to comprehend and respond to evolving situations more effectively. The potential of our approach is exemplified using the "Tennessee-Eastman" process, which highlights its capability in analyzing complex and time-varying causal relationships.