Training recurrent neural networks using a hybrid algorithm

This paper proposes a new hybrid approach for recurrent neural networks (RNN). The basic idea of this approach is to train an input layer by unsupervised learning and an output layer by supervised learning. In this method, the Kohonen algorithm is used for unsupervised learning, and dynamic gradient descent method is used for supervised learning. The performances of the proposed algorithm are compared with backpropagation through time (BPTT) on three benchmark problems. Simulation results show that the performances of the new proposed algorithm exceed the standard backpropagation through time in the reduction of the total number of iterations and in the learning time required in the training process.     

Preventive maintenance scheduling for a multi-component system with non-negligible replacement time

When maintenance models are developed for complex continuous operating systems, with large production losses, it is important to take into account the maintenance duration, in order to provide an effective maintenance program. In this work, a preventive maintenance model is presented in order to coordinate the component replacements in a multi-component system. The model is based on the partial periodic renewal policy. At each partial renewal, a group of components to be replaced is defined. When all components are replaced simultaneously, we consider that the system undergoes an overhaul which resets it to the new state. Moreover, the model takes into account for non-negligible replacement times, where different assumptions are considered. The relevant effect of maintenance times on the optimal policy is clearly shown by the numerical results, provided by the application in an oil refinery.     

Incorporating of layered double hydroxide/sepiolite to improve the performance of sulfonated poly(ether ether ketone) composite membranes for proton exchange membrane fuel cells

Sulfonated poly(ether ether ketone) (SPEEK) with a low sulfonation degree was blended using nanofiller layered double hydroxide (LDH, Mg₂AlCl)/sepiolite nanostructured material as additive to use as an electrolyte membrane for fuel cell application. Characterization assessments, that is, mechanical stability, thermal gravimetric analysis, ion exchange capability, swelling properties, water uptake capacities, electrochemical impedance spectroscopy analysis, and Fourier transform infrared spectroscopy of the composite membranes were conducted. The presence of LDH/sepiolite nanoarchitectured material within SPEEK was found to have the highest water retention and proton conductivity value at high temperature rather than LDH/SPEEK and pristine SPEEK membranes.     

A Hybrid Training Algorithm for Feedforward Neural Networks

Feedforward neural networks (FNN) have been proposed to solve complex problems in pattern recognition, classification and function approximation. Despite the general success of learning methods for FNN, such as the backpropagation (BP) algorithm, second-order algorithms, long learning time for convergence remains a problem to be overcome. In this paper, we propose a new hybrid algorithm for a FNN that combines unsupervised training for the hidden neurons (Kohonen algorithm) and supervised training for the output neurons (gradient descent method). Simulation results show the effectiveness of the proposed algorithm compared with other well-known learning methods.     

Imperfect Preventive Maintenance Policy for Complex Systems Based on Bayesian Networks

The main challenge in maintenance planning lies in the realistic modeling of the maintenance policy. This paper is focused on the maintenance optimization of complex repairable systems using Bayesian networks. A new policy is developed for periodic imperfect preventive maintenance policy with minimal repair at failure; this policy allows us to take into consideration several types of preventive maintenance with different efficiency levels. The Bayesian networks are used for complex system modeling, allowing the evaluation of the model parameters. The Weibull parameters and the maintenance efficiency are evaluated thanks to the proposed methodology using Bayesian inference. The approach developed in this paper is applied on a real system, to determine the optimal maintenance plan for a turbo‐pump in oil industry. Copyright © 2016 John Wiley & Sons, Ltd.     

Enhancement of photoelectrochemical performance of CdSe sensitized seeded TiO<Subscript>2</Subscript> films

In this work, we highlight the effect of TiO₂ seed layer (SL) on the photoelectrochemical performances of CdSe/TiO₂ photoanodes (PAs). TiO₂ thin films were prepared by spin coating starting from a sol gel solution containing TiO₂ nanopowder, then sensitized with electrodeposited CdSe nanoparticles. Structural, optical and photoelectrochemical properties of the CdSe/TiO₂ PAs with and without the SL were investigated. Charge accumulation processes and charge transfer characteristics were identified by electrochemical impedance spectroscopy. The introduction of the compact TiO₂ SL was found to significantly increase the electron transport. The photocurrent density produced by the CdSe/TiO₂/SL PA reached 0.95 mA/cm², about two times higher than that performed by the CdSe/TiO₂ PAs. This enhancement might be attributed to a substantial decrease of the leakage current induced by a better crystallization of TiO₂ thin films as well as a higher sensitizing effect of the CdSe nanoparticles.     

System identification of a class of Wiener systems with hysteretic nonlinearities

Existing works on Wiener system identification have essentially been focused on the case where the output nonlinearity is memoryless. When memory nonlinearities have been considered, the focus has been restricted to backlash like nonlinearities. In this paper, we are considering Wiener systems where the output nonlinearity is a general hysteresis operator captured by the well‐known Bouc–Wen model. The Wiener system identification problem is addressed by making use of a steady‐state property, obtained in periodic regime, referred to as ‘hysteretic loop assumption’. The complexity of this problem comes from the system nonlinearity as well as its unknown parameters that enter in a non‐affine way in the model. It is shown that the linear part of the system is accurately identified using a frequency method. Then, the nonlinear hysteretic subsystem is identified, on the basis of a parameterized representation, using a prediction‐error approach. Copyright © 2016 John Wiley & Sons, Ltd.     

Economic analysis of renewable heat and electricity production by sewage sludge digestion—a case study

In this paper, we assess the total cost of energy recovery from sewage sludge through anaerobic digestion with biogas utilization in combined heat and power (CHP) system. The important advantage of anaerobic digestion process is the production of biogas, which can be used to generate electricity and heat as a source of renewable energy. From this study, it can be retained that the generated thermal energy from the anaerobic digestion process meets the needs of the wastewater treatment plant (WWTP) and guarantees its self‐sufficiency in heat. The surplus of renewable heat produced by CHP is not a primary factor to improve the economic viability of the process. Moreover, the sales of electricity output represent about 76% of the operating costs of anaerobic digestion process. Renewable energy production is not economically viable by its own, without considering the wastewater treatment function and the associated incomes. Nevertheless, sludge digestion helps to reduce the wastewater treatment costs mainly by giving a good source of revenue via electricity production. Copyright © 2014 John Wiley & Sons, Ltd.     

Utilization of olive-residue oil in the formulation of lubricating calcium greases

Calcium greases were manufactured from olive-residue oil in a closed reactor by substituting olive-residue oil for animal tallow. The results of the performance tests applied to the obtained greases showed that their physicochemical and mechanical characteristics are comparable to those of calcium grease manufactured industrially by using animal tallow. The substitution of part of the animal tallow with olive-residue oil increased the thermal and mechanical stability of the grease. Olive-residue oil could constitute, alone or in combination with animal tallow, a raw material in the formulation of lubricating calcium grease.     

A two-stage-priority-rule-based algorithm for robust resource-constrained project scheduling

Traditionally, the resource-constrained project scheduling problem (RCPSP) is modeled as a static and deterministic problem and is solved with the objective of makespan minimization. However, many uncertainties, such as unpredictable increases in processing times caused by rework or supplier delays, random transportation and/or setup, may render the proposed solution obsolete. In this paper, we present a two-stage algorithm for robust resource-constrained project scheduling. The first stage of the algorithm solves the RCPSP for minimizing the makespan only using a priority-rule-based heuristic, namely an enhanced multi-pass random-biased serial schedule generation scheme. The problem is then similarly solved for maximizing the schedule robustness while considering the makespan obtained in the first stage as an acceptance threshold. Selection of the best schedule in this phase is based on one out of 12 alternative robustness predictive indicators formulated for the maximization purpose. Extensive simulation testing of the generated schedules provides strong evidence of the benefits of considering robustness of the schedules in addition to their makespans. For illustration purposes, for 10 problems from the well-known standard set J30, both robust and non-robust schedules are executed with a 10% duration increase that is applied to the same randomly picked 20% of the project activities. Over 1000 iterations per instance problem, the robust schedules display a shorter makespan in 55% of the times while the non-robust schedules are shown to be the best performing ones in only 6% of the times.