Vladimir Mahalec

Component failures in integrated heating and cooling networks impact reliable energy supply. This work studies reliability of heating and cooling systems constrained either by the annualized capital cost (ACC) or by the annualized total cost (ATC). A novel approach for assessing the supply reliability, representing energy networks as stochastic graph networks, is introduced, thereby enabling imposition of operational uncertainties and constraints as well as budgetary limitations as the graph networks' features. The reliability index is defined as the average percentage of the demands that are met. Single or multiple component failures may occur anytime; the reliability evaluation method is not restricted to one failure at a time. Examples of two different energy supply networks, maximizing the reliability under ACC or ATC constraints, are presented to illustrate the methodology. It is shown that, under ATC or ACC constraints, the 100% reliability and maximum consecutive loss of load hours (MCLOLH) equal to zero can be attained via a wide range of equipment sizes provided the thermal storage is sized correctly. Maps representing relative costs of heating, cooling, and storage, for the given system structure, help decision makers to determine equipment sizes if scaling up/down the system, while maintaining a high level of reliability.

Designing reliable fault detection and diagnosis (FDD) systems remains difficult when only limited fault-free data are available. Kolmogorov–Arnold Networks (KANs) have recently been proposed as parameter-efficient alternatives to multilayer perceptrons, yet their effectiveness for unsupervised FDD has not been systematically established. This study presents a statistically grounded comparison of Kolmogorov–Arnold Autoencoders (KAN-AEs) against an orthogonal autoencoder and a PCA baseline using the Tennessee Eastman Process benchmark. Four KAN-AE variants (EfficientKAN-AE, FastKAN-AE, FourierKAN-AE, and WavKAN-AE) were trained on fault-free data subsets ranging from 625 to 250,000 samples and evaluated over 30 independent runs. Detection performance was assessed using Bayesian signed-rank tests to estimate posterior probabilities of model superiority across fault scenarios. The results show that WavKAN-AE and EfficientKAN-AE achieve approximately 90–92% fault detection rate with only 2500 samples. In contrast, the orthogonal autoencoder requires over 30,000 samples to reach comparable performance, while PCA remains markedly below this level regardless of data size. Under data-rich conditions, Bayesian tests show that the orthogonal autoencoder matches or slightly outperforms the KAN-AEs on the more challenging fault scenarios, while remaining computationally more efficient. These findings position KAN-AEs as compact, data-efficient tools for industrial fault detection when historical fault-free data are scarce.

Kolmogorov-Arnold Networks (KANs) have recently emerged as a flexible and parameter-efficient alternative to conventional neural networks. Unlike standard architectures that use fixed node-based activations, KANs place learnable functions on edges, parameterized by different function families. While they have shown promise in supervised settings, their utility in unsupervised fault detection remains largely unexplored. This study presents a comparative evaluation of KAN-based autoencoders (KAN-AEs) for unsupervised fault detection in chemical processes. We investigate four KAN-AE variants, each based on a different KAN implementation (EfficientKAN, FastKAN, FourierKAN, and WavKAN), and benchmark them against an Orthogonal Autoencoder (OAE) on the Tennessee Eastman Process. Models are trained on normal operating data across 13 training set sizes and evaluated on 21 fault types, using Fault Detection Rate (FDR) as the performance metric. WavKAN-AE achieves the highest overall FDR ($\geq$92\%) using just 4,000 training samples and remains the top performer, even as other variants are trained on larger datasets. EfficientKAN-AE reaches $\geq$90\% FDR with only 500 samples, demonstrating robustness in low-data settings. FastKAN-AE becomes competitive at larger scales ($\geq$50,000 samples), while FourierKAN-AE consistently underperforms. The OAE baseline improves gradually but requires substantially more data to match top KAN-AE performance. These results highlight the ability of KAN-AEs to combine data efficiency with strong fault detection performance. Their use of structured basis functions suggests potential for improved model transparency, making them promising candidates for deployment in data-constrained industrial settings.

This work presents a data‐driven model of a two‐product distillation tower that is suitable for real‐time optimization (RTO) of distillation columns. The proposed model accurately predicts product mass fractions using operating variables and tray temperatures by integrating a linear data‐driven inferential composition model (based on two tray temperatures in each section of the tower, reflux/distillate ratio, and reboiler duty/bottoms flow ratio) with a neural network (NN) model that predicts tray temperatures from the value of the manipulated variables. RTO is carried out via an iterative procedure where the sensitivity matrix is initially calculated from the model and updated using plant measurements from subsequent values. A butane splitter column is presented as a case study. Our results show that the implementation of the data‐driven model‐based RTO results in a solution that is within 0.1% of the optimization solution based on the rigorous tray‐to‐tray distillation simulation.

This paper is an invitation to the process systems engineering community to change the paradigm for process plants. The goal is to achieve much easier convergence while retaining accuracy on par with the rigorous models. Accurate plant models of existing plants can be linear or much less nonlinear if they are based on mass component flows and stream properties per unit mass properties instead of molar flows and mole fractions. Accurate stream properties per unit mass can be calculated at stream specific conditions by linear approximations which in many instances eliminates mole fraction-based flash calculations. Hybrid data-driven node models fit naturally in this paradigm, since they used measured data, which is either in mass or in volumetric units, but never in moles. Instantiation of models at all levels of abstraction (planning, scheduling, optimization, and control models) from the same plant topology representation will ensure inheritance of solutions from mass-only to mass-and-energy to mass-and-energy-and-stream-properties, thereby ensuring consistency of solutions between these models. None of the existing software provides inheritance between different levels of plant abstraction (i.e. inheritance between models for different business applications) or different levels of abstractions per plant sections or per time periods, which motivates this exposition.