Hyperspectral imaging provides high-dimensional spectral information for material characterization and remote sensing analytics. However, deployment is constrained by hardware cost, inter-band redundancy, and the limited suitability of predominantly spatially oriented models for one-dimensional (1D) spectral signatures. This paper presents an integrated sensor-learning framework that combines data-driven wavelength selection with sequence-aware spectral classification to support compact multispectral sensing. sparse principal component analysis with strong $L_{1}$ regularization is used to identify informative wavelengths and derive reduced-band multispectral configurations that mitigate redundancy and the curse of dimensionality. A 1D recurrent multi-head convolutional network is then introduced, which integrates bidirectional recurrent encoding, multi-head self-attention, and hierarchical 1D convolutions to capture local spectral structure and long-range inter-band dependencies. In contrast to convolutional neural network (CNN), recurrent neural network (RNN) and Transformer variants designed for spatial (2D) feature learning or tokenized hyperspectral cubes, the proposed architecture targets reduced-band 1D spectral sequences and is suitable for computationally constrained operation. Experiments on Indian Pines and Houston2013 show that the proposed approach outperforms representative CNN-, RNN-, and transformer-based baselines and requires fewer spectral bands. Overall accuracy reaches 82.3% on Indian Pines and 88.6% on Houston2013.

This paper compares the performance of model-free controllers on a nonlinear system under cyberattacks, including false data injection and denial-of-service attacks. Four RL reward types are analyzed for accuracy, cost, and resilience. Results show that the Lyapunov reward offers the best resilience with low tracking error. Exponential mode also provides good trade-offs with acceptable resilience under moderate training conditions. Progressive and linear rewards converge faster but are less robust. RL-MPCs show strong steady-state resilience but require longer training times; RL-PID controllers are faster with significantly less training time. Proximal Policy Optimization outperforms Deep Deterministic Policy Gradient with a significant reduction in KPI variance. This study serves to highlight how well-designed RL rewards can improve performance and resilience against cyber threats.

Alarm floods are disruptive events in process industries, where a large number of alarms are triggered within a short period, overwhelming operators and compromising safe operation. Traditional approaches for alarm flood analysis, ranging from sequence alignment to deep learning models, are often computationally demanding, sensitive to chronological variations, or unsuitable for real-time deployment. This study proposes a novel framework for industrial-scale online classification of alarm floods based on deterministic fingerprinting with combinatorial hashing. In offline phase, the alarm activity of historical alarm floods is encoded using an Alarm Evolution Matrix (AEM), which captures short-horizon activation rates of relevant tags via a sliding-window strategy. The salient local features of alarm floods are converted into compact and reproducible fingerprints through a Combinatorial Temporal Fingerprint Hashing (CTFH) scheme. Category-wise Consensus Fingerprint Profiles (CFP) are constructed by majority voting across fingerprints, while a variability index was proposed to quantify the internal consistency of floods within each fault class. During online classification, fingerprints are continuously generated from the incoming alarm data and matched against the offline database, enabling early and robust classification decisions. By combining compact temporal representations, deterministic fingerprinting, and variability-aware confidence measures, the framework offers a practical and scalable solution for real-time alarm flood classification in industrial environments. The effectiveness of the method is validated using the Tennessee Eastman Process (TEP) benchmark. Results indicate high overall classification accuracy with reliable early decision times, while maintaining scalability in both runtime and storage. The method further demonstrates robustness under four perturbation scenarios that emulate practical data quality issues in industrial alarm systems.

Efficient water management in agriculture is essential for addressing the growing freshwater scarcity crisis. Multi-Agent Reinforcement Learning (MARL) has emerged as a promising method for solving daily irrigation scheduling problems in spatially variable fields, where management zones are employed to account for field variability. To enhance the application of MARL to address daily irrigation scheduling in large-scale fields with significant spatial variation, this study proposes a Semi-Centralized MARL (SCMARL) framework. The SCMARL framework adopts a hierarchical structure, decomposing the daily irrigation scheduling problem into two levels of decision-making. At the top level, a centralized coordinator agent determines irrigation timing, which is modeled as a discrete variable, based on field-wide soil moisture data, crop conditions, and weather forecasts. At the lower level, decentralized local agents use local soil moisture, crop, and weather information to determine the appropriate irrigation amounts for each management zone. To address the issue of non-stationarity in this framework, a state augmentation technique is employed, wherein the coordinator’s decision is incorporated into the decision-making process of the local agents. The SCMARL framework, which leverages the Proximal Policy Optimization algorithm for training the agents, is evaluated on a large-scale field in Lethbridge, Canada, and compared with an existing MARL irrigation scheduling approach. The results demonstrate improved performance, achieving a 4.0% reduction in water use and a 6.3% increase in irrigation water use efficiency.

Accurate soil moisture estimation is essential for advancing closed-loop irrigation. Central to this task are soil hydraulic parameters, which are rarely known precisely and must be inferred from moisture measurements. Inferring these parameters for large-scale agricultural fields presents practical difficulties due to the sparse and noisy nature of moisture measurements. To address this challenge, a framework is developed that combines sensitivity analysis and orthogonal projection to identify parameters that are most reliably estimable from the measurements. The selected parameters, together with soil moisture states, are estimated by assimilating remotely sensed soil moisture observations into the Richards equation using an extended Kalman filter. Numerical simulations and field experiments conducted on a large-scale site in Lethbridge, Alberta, Canada, demonstrate improvements of 24%-43% in soil moisture estimation accuracy and a 50% enhancement in predictive performance. Furthermore, the estimated parameters, particularly saturated hydraulic conductivity, show good agreement with experimental measurements.

Data-driven methods for fault detection increasingly rely on large historical datasets, yet annotations are costly and time-consuming. As a result, learning approaches that minimize the need for extensive labeling, such as self-supervised learning (SSL), are becoming more popular. Contrastive learning, a subset of SSL, has shown promise in fields like computer vision and natural language processing (NLP), yet its application in fault detection is not fully explored. In this paper, we introduce Cross-Domain Predictive and Contextual Contrasting (CDPCC), a novel contrastive learning framework that integrates temporal and spectral information to capture informative time-frequency features from time series data. CDPCC consists of two key components: cross-domain predictive contrasting, which predicts future embeddings across time and frequency domains, and cross-domain contextual contrasting, which aligns time- and frequency-based representations in a shared latent space. We evaluate CDPCC on fault detection tasks using both simulated and industrial datasets. Our results show that a linear classifier trained on features learned by CDPCC performs comparably to fully supervised models. Moreover, CDPCC proves highly effective in scenarios with limited labeled data, achieving superior performance with only 50% of the labeled data compared to fully supervised training on the entire dataset. The source code is publicly available at https://github.com/iy641/CDPCC.git.

This paper examines the suitability of unsupervised machine learning methods for image analysis, within the innovative visual analytics framework for process monitoring, and proposes a set of performance metrics that evaluate accuracy for visual analytics. The effectiveness of the proposed method is demonstrated via a case study using real industrial data from a steam boiler. Copyright (c) 2024 The Authors.

The efficient fusion of information from different data silos in a process plant is essential for a smart data analytics platform. Researchers have developed a GUI-based framework, namely, AMtool, to integrate alarm data and sensor (process) data complemented with process connectivity information. This state-of-the-art smart data analytics framework integrates advanced techniques for data analysis and visualization, performance evaluation, and alarm configuration. This framework also supports alarm flood analysis along with documentation to aid alarm rationalization, enabling industrial processes to become compliant with industry standards. However, an earlier version of the AMtool faced limitations in meeting the automation and real-time data exchange requirements emphasized by the Industry 4.0 paradigm. In this paper, we discuss the recent advances and improvements to make the tool compliant with Industry 4.0 standards, namely, the data extraction interface and automated reporting feature. The data extraction interface facilitates secure and efficient data collection via SQL databases and OPC UA platforms. Additionally, the report generation is more automated for an efficient and more user-friendly option for the regular generation of alarm analysis reports.

Amid concerns about freshwater scarcity, the agricultural sector faces challenges in water conservation and optimizing crop yields, highlighting the limitations of traditional irrigation scheduling methods. To overcome these challenges, this paper introduces a unified, learning-based predictive irrigation scheduler that integrates machine learning and Model Predictive Control (MPC), while also incorporating multi-agent principles. The proposed framework incorporates a three-stage management zone delineation process, utilizing k-means clustering and hydraulic parameters estimates for optimized agro-hydrological modeling. Long Short-Term Memory (LSTM) networks are employed for accurate and computationally efficient root zone soil moisture modeling. The scheduler, formulated as a mixed-integer MPC with zone control, utilizes the identified LSTM networks to maximize root water uptake while minimizing overall water consumption and fixed irrigation costs. Additionally, the learning-based scheduler adopts a multi-agent MPC paradigm, where decentralized hybrid actor–critic agents and the concept of a limiting irrigation management zone are employed to enhance computational efficiency. Evaluating the performance on a 26.4-hectare field in Lethbridge for the 2015 and 2022 growing seasons demonstrates the superiority of the proposed scheduler over the widely-used triggered scheduling approach in terms of Irrigation Water Use Efficiency (IWUE) and total prescribed irrigation. Notably, the proposed approach achieves water savings between 7 to 23%, coupled with IWUE increases ranging from 10 to 35%.

Safe and efficient operation of complex industrial processes is highly dependent on alarm systems. There is a phenomenon called alarm flood, which significantly impairs the performance of alarm systems. An alarm flood refers to a situation when a plant operator is overloaded with many alarms triggered sequentially within a short time period. As a result, the operator can be distracted from taking the required safety actions. It is, therefore, imperative to deal with the problem of alarm floods, particularly for online applications. By taking advantage of historical alarm and event (A&E) logs, efficient alarm flood monitoring methods based on similarity analysis can be developed in the area of machine learning (ML). In this article, we develop a probabilistic framework for transforming alarm flood data into suitable inputs for convolutional neural networks (CNNs). The proposed alarm flood representation is used to establish an ML-based alarm flood analysis framework, where a CNN is trained to predict upcoming fault scenarios by observing online alarm floods. This method is capable of incorporating the activated alarm tags and their triggering times, as well as tolerating the effect of irrelevant alarms and alarm order ambiguities. Two alarm datasets from the Tennessee Eastman (TE) benchmark and a Vinyl Acetate Monomer (VAM) process are used to evaluate the proposed approach.

Efficient water management in agriculture is important for mitigating the growing freshwater scarcity crisis. Mixed-integer Model Predictive Control (MPC) has emerged as an effective approach for addressing the complex scheduling problem in agricultural irrigation. However, the computational complexity of mixed-integer MPC still poses a significant challenge, particularly in large-scale applications. This study proposes an approach to enhance the computational efficiency of mixed-integer MPC-based irrigation schedulers by employing Rectified Linear Unit (ReLU) surrogate models to describe the soil moisture dynamics of the agricultural field. By leveraging the mixed-integer linear representation of the ReLU operator, the proposed approach transforms the mixed-integer MPC-based scheduler with a quadratic cost function into a mixed-integer quadratic program, which is the simplest class of mixed-integer nonlinear programming problems that can be efficiently solved using global optimization solvers. The effectiveness of this approach is demonstrated through comparative studies conducted on a large-scale agricultural field across two growing seasons, involving other machine learning surrogate models, specifically Long Short-Term Memory (LSTM) networks, and the triggered irrigation scheduling method. The ReLU-based approach significantly reduces solution times — by up to 99.5% — while achieving comparable performance to the LSTM approach in terms of water savings and Irrigation Water Use Efficiency (IWUE). Moreover, the ReLU-based approach achieves enhanced performance in terms of irrigation water savings and IWUE compared to the triggered approach.

Alarm systems are essential for the safe and efficient operation of process industries. However, complex plant connectivity and process interactions could cause many correlated alarms in practice and thus compromise alarm system performance. To address correlated alarms, it is desired that alarm correlations are discovered from historical Alarm and Event (A&E) logs, so the obtained results could help improve alarm configurations or design suppression strategies. Motivated by this problem, a systematic method to extract alarm correlation is proposed in this work and the contributions are: (1) Correlated alarms and their occurrence orders are captured as correlation patterns through pattern mining, and such patterns are characterized by statistical features. (2) Alarm correlations and their statistical features are visualized as network graphs to indicate process interactions and identify alarms for prioritized analysis. To demonstrate the effectiveness of the proposed method, case studies are provided using an industrial simulation benchmark Vinyl Acetate Monomer (VAM) plant model.

This study proposes a Semi-centralized Multi-agent Reinforcement Learning (SC-MARL) approach for irrigation scheduling in agricultural fields, which are characterized by spatial variability and therefore delineated into management zones. The SCMARL framework is hierarchical in nature, with a coordinator agent at the top level and local agents at the second/lower level. The coordinator agent makes daily 'yes/no' irrigation decisions based on field-wide observations from all the management zones, which are then communicated to local agents. These local agents are tasked with determining the optimal daily irrigation depths for specific management zones, utilizing both the coordinator agent's decision and local observations. A comparison between the SCMARL method and a Fully Decentralized Multi-agent Reinforcement Learning approach is presented, highlighting the superior performance of the SCMARL approach in terms of water savings and improved irrigation water-use efficiency. Copyright (C)2024 The Authors. This is an open access article under the CC BY-NC-ND license (htips://creativecommons.org/licenses/by-nc-nd/4.0/)

The agricultural sector currently faces significant challenges in water resource conservation and crop yield optimization, primarily due to concerns over freshwater scarcity. Traditional irrigation scheduling methods often prove inadequate in meeting the needs of large-scale irrigation systems. To address this issue, this paper proposes a predictive irrigation scheduler that leverages the three paradigms of machine learning to optimize irrigation schedules. The proposed scheduler employs the k-means clustering approach to divide the field into distinct irrigation management zones based on soil hydraulic parameters and topology information. Furthermore, a long short-term memory network is employed to develop dynamic models for each management zone, enabling accurate predictions of soil moisture dynamics. Formulated as a mixed-integer model predictive control problem, the scheduler aims to maximize water uptake while minimizing overall water consumption and irrigation costs. To tackle the mixed-integer optimization challenge, the proximal policy optimization algorithm is utilized to train a reinforcement learning agent responsible for making daily irrigation decisions. To evaluate the performance of the proposed scheduler, a 26.4-hectare field in Lethbridge, Canada, was chosen as a case study for the 2015 and 2022 growing seasons. The results demonstrate the superiority of the proposed scheduler compared to a traditional irrigation scheduling method in terms of water use efficiency and crop yield improvement for both growing seasons. Notably, the proposed scheduler achieved water savings ranging from 6.4% to 22.8%, along with yield increases ranging from 2.3% to 4.3%.

The Internet of Things (IoT) systems provide the technological support to perform automatic indoor air quality monitoring. However, most of the data IoT systems needs processing to obtain valuable information. In this work, we analyze the suitability of imaging time-series in terms of the amount of information based on images entropy, the accuracy to detect abnormal indoor air quality conditions by supervised classification with artificial intelligence (AI) methods, and its computational cost.

Improving the accuracy of soil moisture estimation is desirable from the perspectives of irrigation management and water conservation. To this end, this study proposes a systematic approach to select a subset of soil hydraulic parameters for estimation in large-scale agrohydrological systems to enhance soil moisture estimation accuracy. The proposed method involves simultaneous estimation of the selected parameters and the entire soil moisture distribution of the field, taking into account soil heterogeneity and using soil moisture observations obtained through microwave radiometers mounted on a center pivot irrigation system. At its core, the proposed method models the field with the cylindrical coordinate version of the Richards equation and addresses the issue of parameter estimability (quantitative parameter identifiability) through the sensitivity analysis and orthogonal projection approaches. Additionally, the study assimilates remotely sensed soil moisture observations into the field model using the extended Kalman filtering technique. The effectiveness of the proposed methodology is demonstrated through numerical simulations and a real field experiment, with cross-validation results showing a 24-43% improvement in soil moisture estimation accuracy. Overall, the study highlights the potential of this method to enhance soil moisture estimation in large-scale agricultural fields.

Rapid development in data-driven process monitoring has provided a rich selection of models and data preprocessing strategies for applications such as fault detection and diagnosis. However, the development, comparison, and selection of process monitoring algorithms can become complicated and unnecessarily onerous. As a result, numerous publicly available benchmark datasets have emerged in the literature. Unfortunately, benchmark literature often suffers from problems such as low fidelity, inconsistent usage, and lack of transparency. This paper presents a benchmark challenge based on a large-scale industrial dataset that aims to enhance the evaluation and comparison of learning algorithms and overall data preprocessing workflows. We introduce the arc loss challenge, a machine learning benchmark with data from a large-scale mining and pyrometallurgy operation. By providing a supervised learning challenge based on large quantities of raw industrial process data with transparent and consistent evaluation procedures, the arc loss challenge is a unique contribution to fault detection benchmarking.