Surface reconstruction with data-driven exemplar priors

In this paper, we propose a framework to reconstruct 3D models from raw scanned points by learning the prior knowledge of a specific class of objects. Unlike previous work that heuristically specifies particular regularities and defines parametric models, our shape priors are learned directly from existing 3D models under a framework based on affinity propagation. Given a database of 3D models within the same class of objects, we build a comprehensive library of 3D local shape priors. We then formulate the problem to select as-few-as-possible priors from the library, referred to as exemplar priors. These priors are sufficient to represent the 3D shapes of the whole class of objects from where they are generated. By manipulating these priors, we are able to reconstruct geometrically faithful models with the same class of objects from raw point clouds. Our framework can be easily generalized to reconstruct various categories of 3D objects that have more geometrically or topologically complex structures. Comprehensive experiments exhibit the power of our exemplar priors for gracefully solving several problems in 3D shape reconstruction such as preserving sharp features, recovering fine details and so on.     

Leak localization in water distribution networks using a mixed model-based/data-driven approach

This paper proposes a new method for leak localization in water distribution networks (WDNs). In a first stage, residuals are obtained by comparing pressure measurements with the estimations provided by a WDN model. In a second stage, a classifier is applied to the residuals with the aim of determining the leak location. The classifier is trained with data generated by simulation of the WDN under different leak scenarios and uncertainty conditions. The proposed method is tested both by using synthetic and experimental data with real WDNs of different sizes. The comparison with the current existing approaches shows a performance improvement.     

A hybrid model combining a support vector machine with an empirical equation for predicting polarization curves of PEM fuel cells

A hybrid model was proposed by combining a support vector machine (SVM) model with an empirical equation for more accurate prediction of the polarization curves of a PEM (polymer electrolyte membrane) fuel cell under various operating conditions. Operational data were obtained from designed experiments for a PEM fuel cell for training, testing, and validating the hybrid model, and a model training procedure was presented for determining the model coefficients and hyper-parameters of the hybrid model. The predictive performance of the hybrid model was compared with that of a SVM model. The SVM model showed somewhat poor performance, especially yielding large prediction errors in the high voltage ranges of the polarization curves as reported in the literature. In contrast, the hybrid model exhibited almost perfect matches between the predicted and measured polarization curves, resulting in significantly lower root-mean-square errors of 1.7–4.4 mV which correspond to only 14–21% of those obtained from the SVM model.     

Parameterized data-driven fuzzy model based optimal control of a semi-batch reactor

A parameterized data-driven fuzzy (PDDF) model structure is proposed for semi-batch processes, and its application for optimal control is illustrated. The orthonormally parameterized input trajectories, initial states and process parameters are the inputs to the model, which predicts the output trajectories in terms of Fourier coefficients. Fuzzy rules are formulated based on the signs of a linear data-driven model, while the defuzzification step incorporates a linear regression model to shift the domain from input to output domain. The fuzzy model is employed to formulate an optimal control problem for single rate as well as multi-rate systems. Simulation study on a multivariable semi-batch reactor system reveals that the proposed PDDF modeling approach is capable of capturing the nonlinear and time-varying behavior inherent in the semi-batch system fairly accurately, and the results of operating trajectory optimization using the proposed model are found to be comparable to the results obtained using the exact first principles model, and are also found to be comparable to or better than parameterized data-driven artificial neural network model based optimization results.     

Comparing a knowledge-based and a data-driven method in querying data streams for system fault detection: A hydraulic drive system application

The field of fault detection and diagnosis has been the subject of considerable interest in industry. Fault detection may increase the availability of products, thereby improving their quality. Fault detection and diagnosis methods can be classified in three categories: data-driven, analytically based, and knowledge-based methods.     

ASDEX Upgrade Discharge Control System—A real-time plasma control framework

ASDEX Upgrade is a fusion experiment with a size and complexity to allow extrapolation of technical and physical conditions and requirements to devices like ITER and even beyond. In addressing advanced physics topics it makes extensive use of sophisticated real-time control methods. It comprises real-time diagnostic integration, dynamically adaptable multivariable feedback schemes, actuator management including load distribution schemes and a powerful monitoring and pulse supervision concept based on segment scheduling and exception handling. The Discharge Control System (DCS) supplies all this functionality on base of a modular software framework architecture designed for real-time operation. It provides system-wide services like workflow management, logging and archiving, self-monitoring and inter-process communication on Linux, VxWorks and Solaris operating systems. By default DCS supports distributed computing, and a communication layer allows multi-directional signal transfer and data-driven process synchronisation over shared memory as well as over a number of real-time networks. The entire system is built following the same common design concept combining a rich set of re-usable generic but highly customisable components with a configuration-driven component deployment method.We will give an overview on the architectural concepts as well as on the outstanding capabilities of DCS in the domains of inter-process communication, generic feedback control and pulse supervision. In each of these domains, DCS has contributed important ideas and methods to the on-going design of the ITER plasma control system. We will identify and describe these essential features and illustrate them with examples from ASDEX Upgrade operation.     

Uncertainty quantification of the premixed combustion characteristics of NH3/H2/N2 fuel blends

Green ammonia is a candidate fuel to decarbonise shipping and other industries. However, ammonia features a lower reactivity compared to conventional fuels and is therefore difficult to burn. To resolve this issue, thermo-catalytic cracking of ammonia using waste heat is often employed to produce NH₃/H₂/N₂ blends as fuel. However, on-site operational variations in this process can become sources of uncertainty in the fuel composition, causing randomness of the flame's physicochemical properties and challenging flame stability. In the present work, a surrogate model is built using the polynomial chaos expansion (PCE) method to investigate the impact of fuel composition variability on combustion characteristics at different operating conditions. Impacts of 1.5% deviation in the fuel composition on the flame properties for different initial pressures (P_i) and unburnt fuel temperatures (T_u) are investigated for a wide range of equivalence ratios covering lean and rich mixtures. The uncertainty effects defined by the coefficient of variation (COV) fluctuate for equivalence ratios greater than 1.1, while no fluctuation is observed in COV for near stoichiometric combustion conditions. It is shown that H₂ variation in the fuel blend has the strongest effect (over 80%) on the uncertainty of all investigated physicochemical properties of the flame. The least affected property is the adiabatic flame temperature with variations of about 2.5% in richer fuel conditions. The results further show that preheating of the reactants can significantly reduce the COV of laminar flame speed. The consequences of these uncertainties upon different combustion technologies are then discussed and it is argued that moderate and intense low oxygen dilution (MILD) and colourless distributed combustion (CDC) technology may remain resilient.     

A novel online degradation model for proton exchange membrane fuel cell based on online transfer learning

A novel online degradation prediction model is proposed to prognosticate the future degradation trend (FDT) of proton exchange membrane fuel cell (PEMFC) stack in this paper. In order to overcome the fact that existing FDT prediction methods of PEMFC stack based on data-driven model rely on the assumption that the operating conditions of the training data and testing data need to be consistent, an end-to-end prediction algorithm based on the combination of transfer learning and transformer neural network, referred to as TLTNN, is proposed to predict the FDT of PEMFC stack. Besides, in order to demonstrate the effectiveness and superiority of the proposed method in prognostics tasks of PEMFC stack FDT, the prediction performance is validated on the PEMFC test system. The results show that the RMSE, MAE and MAPE values of the predicted degradation voltage are 0.00598 V, 0.004842 V and 0.1518%, respectively, which indicates that the proposed TLTNN strategy based on transfer online learning can be used to predict the degradation voltage of PEMFC stack and the superiority of the proposed method is better, thus solving the problem that the distribution of training and test data must be the same in traditional machine learning models.     

An unconstrained optimization approach to empirical mode decomposition

Empirical mode decomposition (EMD) is an adaptive (data-driven) method to decompose non-linear and non-stationary signals into AM-FM components. Despite its well-known usefulness, one of the major EMD drawbacks is its lack of mathematical foundation, being defined as an algorithm output. In this paper we present an alternative formulation for the EMD method, based on unconstrained optimization. Unlike previous optimization-based efforts, our approach is simple, with an analytic solution, and its algorithm can be easily implemented. By making no explicit use of envelopes to find the local mean, possible inherent problems of the original EMD formulation (such as the under- and overshoot) are avoided. Classical EMD experiments with artificial signals overlapped in both time and frequency are revisited, and comparisons with other optimization-based approaches to EMD are made, showing advantages for our proposal both in recovering known components and computational times. A voice signal is decomposed by our method evidencing some advantages in comparison with traditional EMD and noise-assisted versions. The new method here introduced catches most flavors of the original EMD but with a more solid mathematical framework, which could lead to explore analytical properties of this technique.     

Optimal behaviour prediction using a primitive-based data-driven model-free iterative learning control approach

This paper suggests an optimal behaviour prediction mechanism for Multi Input-Multi Output control systems in a hierarchical control system structure, using previously learned solutions to simple tasks called primitives. The optimality of the behaviour is formulated as a reference trajectory tracking problem. The primitives are stored in a library of pairs of reference input/controlled output signals. The reference input primitives are optimized at the higher hierarchical level in a model-free iterative learning control (MFILC) framework without using knowledge of the controlled process. Learning of the reference input primitives is performed in a reduced subspace using radial basis functions for approximations. The convergence of the MFILC learning scheme is achieved via a Virtual Reference Feedback Tuning design of the feedback controllers in the lower level feedback control loops. The new complex trajectories to be tracked are decomposed into the output primitives regarded as basis functions. Next, the optimal reference input fed to the control system in order to track the desired new trajectory is then recomposed from the reference input primitives. The efficiency of this approach is demonstrated on a case study concerning the control of a two-axis positioning mechanism, and the experimental validation is offered.