Industrial processes are becoming increasingly complex, requiring advanced modelling techniques to understand their behaviour and improve their performance. In this context, deep learning algorithms have proven to be effective tools for modelling dynamic systems, with Recurrent Neural Networks (RNNs) being particularly suitable for time-series data. However, the computational complexity of deep learning models can be a limitation in industrial environments, where real-time responses are required.This work proposes the use of Deep Echo State Networks to model an industrial system. The aim is to evaluate its performance in real-time industrial applications when running on embedded devices. The approach is validated on a process composed of four interconnected water tanks, which exhibits typical nonlinear industrial dynamics. Among several candidate architectures (including vanilla RNNs or LSTMs), Deep ESNs were selected for their balance of accuracy and computational efficiency. Different input-output setups and number of Deep ESN layers are tested, and results are compared with LSTMs in terms of accuracy and execution time. Finally, the best Deep ESN models are implemented on industrial embedded devices to evaluate the possibility of running these models in real time.The proposed approach achieved up to a 33% reduction in RMSE and a 14% improvement in R 2 compared to traditional reservoir computing, highlighting its superior predictive performance. The results show that Deep ESN models can effectively model the industrial system, with the best configurations achieving high accuracy and low execution times, demonstrating the feasibility of running these models in real time in industrial environments.

DNA methylation is a key epigenetic modification with diagnostic and prognostic relevance across a wide range of diseases, particularly cancer. Modern array-based technologies enable high-throughput quantification of methylation states at hundreds of thousands of CpG sites, yielding high-dimensional datasets that pose significant challenges for exploratory analysis and feature prioritization. Existing visualization tools often lack interactivity, integration with machine learning methods, or flexible mechanisms for dynamic dimensionality reduction and biological interpretation. This work presents an interactive analytical framework that extends the Self-Organizing Map approach for epigenomic data exploration. Our method introduces meta sites—representative prototypes of CpG site clusters—enabling interpretable, real-time visualization and machine learning over reduced feature spaces. Through conditional sample projections (e.g., via PCA, t-SNE, or UMAP), user-driven region selection, and the integration of sparsity-controlled logistic regression, we generate metasite relevance maps that reveal discriminative epigenetic patterns and guide downstream analysis. The proposed approach supports iterative, visually driven discovery of co regulated modules and disease-associated methylation signatures, offering a powerful and intuitive interface for multidimensional exploration of complex methylation landscapes. Its utility is demonstrated through the analysis of DNA methylation in pheochromocytomas and paragangliomas, focusing on SDHB mutation status and the role of protocadherine gene clusters.

Background To identify sex-specific, actionable gaps in metabolic control and multifactorial target attainment in adults with type 2 diabetes Methods Cross-sectional analysis of 40,211 adults with type 2 diabetes from a primary care registry (46,3% women). Outcomes were HbA1c, LDL-cholesterol (LDL-C), systolic/diastolic blood pressure (SBP/DBP), body mass index (BMI), estimated glomerular filtration rate (eGFR), and urine albumin-to-creatinine ratio (UACR). Targets were HbA1c < 7%, LDL-C < 100 mg/dL, BP < 140/90 mmHg, and a composite triple target. Sex differences were assessed with multivariable linear regression (HbA1c) and robust Poisson regression (target attainment), adjusted for age, diabetes duration, and BMI, and including sex-by-age and sex-by-duration interaction terms. Results Women were older than men (73.9 ± 11.7 vs 70.2 ± 11.3 years) and had higher mean LDL-C (103.0 ± 32.8 vs 95.7 ± 32.4 mg/dL) and BMI (31.1 ± 6.9 vs 30.3 ± 6.6 kg/m²) (all p < 0.001). Female sex was independently associated with higher HbA1c (adjusted β + 0.075, 95% CI + 0.050 to + 0.100; p < 0.001), and the sex gap in HbA1c widened at older ages and longer diabetes duration (both interaction p < 0.001). Men more frequently had albuminuria (UACR ≥ 30 mg/g: 24.5% vs 17.3%), whereas women more often had reduced renal function (eGFR < 60 mL/min/1.73 m²: 24.9% vs 17.3%) (both p < 0.001). Overall, only 17.9% achieved the triple target (men 19.5% vs women 16.2%, p < 0.001). Compared with women, men were 3% more likely to achieve the HbA1c target (adjusted prevalence ratio (aPR) 1.03, 95% CI 1.01–1.04), 22% more likely to reach the LDL-C target (aPR 1.22, 1.20–1.24), and 21% more likely to attain the composite triple target (aPR 1.21, 1.16–1.26), but 4% less likely to achieve the BP target (aPR 0.96, 0.95–0.98). Equalizing women’s composite attainment to men’s would translate into ≈ 614 additional women meeting all three targets. Conclusions In contemporary primary care,women with type 2 diabetes have consistent shortfalls in LDL-C control and composite multifactorial target attainment, and their glycemic disadvantage increases with age and longer diabetes duration. These actionable patterns support sex-aware implementation strategies—prioritizing lipid-lowering optimization in women across ages and tailoring glycemic intensification for older and long-duration women—to close treatment gaps and improve risk factor control.

Model Predictive Control (MPC) is a powerful strategy for industrial process control, but its reliance on accurate dynamic models often poses computational challenges in data-driven implementations. Echo State Networks (ESNs), a type of Recurrent Neural Network (RNN), offer a promising solution due to their ability to capture nonlinear dynamics with simplified training. However, the high dimensionality of ESNs poses an obstacle for real-time implementations. Although recent works have studied the Proper Orthogonal Decomposition (POD) and Discrete Empirical Interpolation Method (DEIM) capabilities to reduce the order of ESNs, a systematic evaluation of these reduced models integrated into an MPC framework remains underexplored. In this paper, a comprehensive study of reduced-order ESN-MPC is presented, showing that the proposed approach maintains control accuracy while significantly reducing optimization time. The effectiveness of the approach is validated not only over simulated systems, but also in a real-world implementation of a quadruple-tank system. Additionally, some of the main problems associated with the application of DEIM are addressed.

La automatización de procesos de producción conlleva la eliminación de la supervisión humana directa, lo que exige el desarrollo de sistemas capaces de analizar la calidad de manera autónoma y en tiempo real. En este contexto, se propone el diseño de un sistema inteligente de inspección para la dispensación de adhesivo en un proceso de montaje de luminarias. El sistema parte de la observación de que los fluidos no newtonianos, como los utilizados en estos procesos, presentan una dinámica compleja influenciada por la reología, la gravedad y la geometría de aplicación, pudiendo manifestar comportamientos inestables que resultan en una deposición irregular y, por lo tanto, en ineficiencias. La solución propuesta se basa en un sistema de perfilometría láser 3D integrado sobre una célula robotizada, en la que un brazo manipulador desplaza el chasis de la luminaria bajo una válvula de dispensación de adhesivo. A partir de los perfiles generados durante el proceso, se extraen características representativas mediante técnicas de ingeniería de características y/o reducción de dimensionalidad. Estas se utilizan como entrada para un clasificador supervisado, cuyo objetivo es evaluar en tiempo real si la dispensación se está realizando correctamente. En conjunto, se presenta una solución viable y escalable para entornos de fabricación flexible, orientada a la inspección autónoma de calidad en procesos de montaje robotizado.

Digital twins have emerged as a powerful tool for industrial process monitoring and optimization, requiring accurate and computationally efficient models to capture system dynamics. This paper presents the implementation of Echo State Networks (ESNs) for modelling complex industrial systems with non-linear behavior. The study is conducted on a pilot plant representing an industrial process, where four variables (level, pressure, flow, and temperature) are controlled and estimated using ESN models. To assess the impact of different architectures, each model is tested with various configurations, including deep networks, feedback integration, and a combination of both. The results demonstrate that the effectiveness of these architectures depends on the dynamic characteristics of each variable. The most notable improvement is observed in process temperature estimation, where feedback significantly enhances performance. The results highlight that the combination of both, feedback mechanisms and deeper architectures, can improve the prediction of variables with slower dynamics and higher inertia.

Modeling real industrial systems is a challenging task due to the complex nature of process data. Data-driven models commonly employ machine learning algorithms, but they often lack the ability to adapt to changes in the system over time. This paper proposes a method that uses Echo State Networks (ESN), a simplified version of Recurrent Neural Networks (RNN), to model an industrial plant. The ESN model incorporates online adaptation to system changes and enables the visualisation of deviations or errors in the operation of the plant. This adaptive model acknowledges acceptable changes within the original system while identifying potential problems or errors. The advantage of this approach is that the same model can be applied to other systems with the same design, eliminating the need for algorithm retraining. Firstly, its successful offline application in visualising anomalies applied to the reference plant is assessed. Secondly, the model is tested for online adaptation to changes in another plant with an identical design but slight differences, while still observing the generated faults. Residual colour maps are used for the visualisation of anomalies.

The automation of production processes entails the elimination of direct human supervision, requiring the development of systems capable of autonomously assessing quality in real time. In this context, an intelligent inspection system is proposed for monitoring adhesive dispensing in a luminaire assembly process. The system is based on the observation that non-Newtonian fluids, such as those used in these processes, exhibit complex dynamics influenced by rheology, gravity, and application geometry, potentially leading to unstable behaviors and irregular deposition that compromise efficiency. The proposed solution employs a 3D laser profilometry system integrated into a robotic cell, where a manipulator arm moves the luminaire chassis beneath an adhesive dispensing valve. From the profiles acquired during the process, representative features are extracted through feature engineering and/or dimensionality reduction techniques. These are used as input to a supervised classifier designed to evaluate, in real time, whether the dispensing is being performed correctly. Overall, the work presents a scalable and practical solution for flexible manufacturing environments, aimed at autonomous quality inspection in robotic assembly processes.

The use of Recurrent Neural Networks (RNNs) is interesting to obtain prediction models for control. Among them, Echo State Networks (ESNs) display advantages such as simplified training and implementation. However, its high dimensionality might become an obstacle for optimization and interpretability. For that reason, this paper presents a comparative study of Model Predictive Control (MPC) implementations using Echo State Networks and their reduced models. The original ESN model and a reduced model obtained through Proper Orthogonal Decomposition (POD) are evaluated in terms of model precision, control performance and computational cost, in the task of controlling a quadruple-tank process. The results indicate that, despite a slight decrease in precision due to model reduction, the reduced model provides a significant reduction in execution time with almost the same control performance, making it a viable alternative for real-time applications.

Energy management requires reliable tools to support decisions aimed at optimising consumption. Advances in data-driven models provide techniques like Non-Intrusive Load Monitoring (NILM), which estimates the energy demand of appliances from total consumption. Common single-target NILM approaches perform energy disaggregation by using separate learned models for each device. However, the use of single-target systems in real scenarios is computationally expensive and can obscure the interpretation of the resulting feedback. This study assesses a conditioned deep neural network built upon a Fully Convolutional Denoising AutoEncoder (FCNdAE) as multi-target NILM model. The network performs multiple disaggregations using a conditioning input that allows the specification of the target appliance. Experiments compare this approach with several single-target and multi-target models using public residential data from households and non-residential data from a hospital facility. Results show that the multi-target FCNdAE model enhances the disaggregation accuracy compared to previous models, particularly in non-residential data, and improves computational efficiency by reducing the number of trainable weights below 2 million and inference time below 0.25 s for several sequence lengths. Furthermore, the conditioning input helps the user to interpret the model and gain insight into its internal behaviour when predicting the energy demand of different appliances.

Monitoring and fault detection can be critical for efficient, safe and reliable operation of electric drive systems. Unfortunately, developing accurate physics-based models for these tasks is difficult due to unknown machine parameters and incomplete knowledge of the physical phenomena occurring within the system. Machine Learning (ML) methods can learn the system’s behavior from data without requiring explicit models. However, expert knowledge of the system is still crucial to extract useful features before applying ML models. This paper presents an overview of the use of ML and data visualization methods for condition monitoring of inverter fed induction motors. More specifically, stator winding temperature estimation and insulation degradation are considered. The analyzed methods make use of the signals normally available in electric drives. Time and frequency-based approaches are considered. The developed methods are assessed on an experimental test bench. The paper is intended to bridge ML and electric drive domains. The desired outcome of this work is to provide useful guidelines for researchers in the electric drives field who aim to apply modern ML and data visualization techniques for monitoring and fault detection.

In this paper, we propose a machine learning approach for detecting superficial defects in metal surfaces using point cloud data. We compare the performance of two popular deep learning architectures, multilayer perceptron networks (MLPs) and fully convolutional networks (FCNs), with varying feature sets. Our results show that FCNs (F1=0.94) outperformed MLPs (F1=0.52) in terms of precision, recall, and F1-score. We found that transfer learning with pre-trained models can improve performance when the amount of available data is limited. Our study highlights the importance of considering the amount and quality of training data in developing machine learning models for defect detection in industrial settings with 3D images.

The use of the common mode current for stator winding insulation condition assessment has been extensively studied. Two main approaches have been followed. The first models the electric behavior of ground-wall insulation as an equivalent RC circuit; these methods have been successfully applied to high-voltage high-power machines. The second uses the high frequency of the common mode current which results from the voltage pulses applied by the inverter. This approach has mainly been studied for the case of low-voltage, inverter-fed machines, and has not yet reached the level of maturity of the first. One fact noticed after a literature review is that in most cases, the faults being detected were induced by connecting external elements between winding and stator magnetic core. This paper presents a case study on the use of the high-frequency common mode current to monitor the stator insulation condition. Insulation degradation occurred progressively with the machine operating normally; no exogenous elements were added. Signal processing able to detect the degradation at early stages will be discussed.

In the industrial field, the modelling of complex systems is a relevant task to understand their evolution, to infer their most representative characteristics, or to detect anomalous situations. Nevertheless, this modelling is notably challenging within the industrial environment, with large amounts of data to be processed but several difficulties in extracting knowledge from these data. In this paper, we work with an industrial plant with four water tanks, focusing on estimating the levels of two sequentially connected tanks. For this purpose, Deep Echo State Networks (Deep ESNs), within the framework of Reservoir Computing (RC), are used, representing an increasingly popular methodology for efficient learning to modelling systems with diverse time-scale dynamics. Specifically, we have designed a learning system that makes use of a dedicated Deep ESN module for the prediction of the level of each tank. We conducted numerical experiments to examine how the performance of the predictions is affected by the number of layers. Our findings indicate that increasing the number of recurrent layers leads to better predictions, and also highlight noteworthy differences in the dynamics of the upper and lower tanks.

In Reservoir Computing, signals or sequences are fed into a set of interconnected non-linear units (neurons) with capabilities for storing information (reservoir). The reservoir generates an expanded representation of the input, which is subsequently mapped onto the desired output using a trained output layer (readout). However, despite their success in various experimental tasks, the dynamics of the reservoir are not yet well understood. In this paper we introduce a new technique, based on the well known Singular Value Decomposition (SVD), to obtain the main dynamic modes of the reservoir when excited with an input signal. We conduct experiments using Echo State Networks (ESN) to demonstrate the technique’s potential and its ability to decompose input signals into Principal Component Modes as expanded by the reservoir. We expect that this approach will open new possibilities in its application to the field of visual analytics in process state visualisation, determination of attribute vectors, and detection of novelties. Furthermore, this technique could serve as a foundation for a better understanding of the reservoir’s dynamic state that could help in other areas of research, such as domain shift or continual learning.

In the field of industrial engineering, it is extremely interesting and useful to develop models that faithfully represent the behaviour of real physical systems, which allow to understand their dynamics, to optimise them or to detect anomalous or faulty situations. This paper proposes the modelling of the dynamics of water tanks within an industrial plant with real instrumentation, using Echo State Networks. The developed models have been used to explore the detection of anomalies in the system by analysing the differences between the estimated system state and the actual state after the introduction of faults.