In recent years, the integration of Digital Twins (DT) for the adoption of smarter maintenance strategies has grown exponentially in different industrial sectors. New IoT and edge computing systems are being developed for this purpose, however, there are still some open issues and challenges to be solved. Firstly, this paper presents new approaches to the initial dependencies of the studied solution and make a new proposal to improve the interoperability of the presented system. Secondly, this paper provides a methodology applicable to similar developments of edge-based AI (Artificial Intelligence) solution, which comprises of four phases: the presentation of the multi-objective problem and the pre-selection of AI-based models, the description of the evaluation architecture, the profiling of the different models for the selection of the most suitable one and explainable AI strategies for getting insights of the selected model. Finally, it presents a use case of an edge-solution for the railway catenary geometry diagnostic (stagger amplitude of the overhead wire), saving the interoperability of the message exchange with other systems is provided.

Maintenance is one of the major concerns of the industrial sector. Acquiring better levels of maintenance maturity is one of the objectives to be achieved. Therefore, prescriptive maintenance is one of the areas of recent research. Current works in literature are focused on specifics of maintenance strategies (from preventive to prescriptive), usually related to a fixed asset. No previous work has been identified regarding the methodology and guidelines to be followed to be able to evolve within the different strategies from a generic perspective. To address the lack of a methodology that shows a more evolutionary path between maintenance strategies, this paper presents Maintenance Maturity Model M3: a maturity model that identifies three areas of action, four levels of maturity, and the activities to be carried out in each of them to make progress in the maturity level of maintenance strategies. The implementation of prescriptive maintenance should be done in a gradual way, starting at the lowest levels. M3 approaches the problem from a broader perspective, analyzing the 18 different domains and the different levels of prior maturity to be considered for prescriptive maintenance. A study has also been carried out on the different maintenance actions and the applicability of the proposed M3 maturity model to the railway infrastructure maintenance is discussed. In addition, this paper also highlights future research lines and open issues.

The monitoring of overhead contact lines (OCL) is a key part of railway infrastructure maintenance. This paper proposes a methodology to assess the lateral geometry of contact wire, the so-called stagger, by using the dynamic response of a pantograph. The methodology is tested in a validated virtual environment that resembles the behaviour of the pantograph when it interacts with the OCL. A signal processing is developed to define features relating the lateral position of the contact wire with the vertical acceleration of the contact strip. It is demonstrated that these features have a clear and close connection with the lateral position of the contact wire. Subsequently, model-driven machine learning algorithms are defined using these features to address the OCL stagger prediction and the detection of out-of-range lateral displacement due to a faulty steady-arm. The methodology shows a good prediction performance in the estimation of the stagger amplitude/central position and the steady-arms diagnosis. The prediction of the stagger amplitude is performed with a root-mean-square error of 4.7(10) mm. In addition, the area under the Precision-Recall curve is 0.952 CI95 [0.940, 0.962] for the steady-arms diagnosis.

Rolling contact fatigue (RCF) is a common cause of rail failure due to repeated stresses at the wheel-rail contact. This phenomenon is a real problem that greatly affects the safety of train operation. Preventive and corrective maintenance tasks have a big impact on the Life Cycle Cost (LCC) of railway assets, and therefore cutting-edge strategies based on predictive functionalities are needed to reduce it. A methodology based on physical models is proposed to predict the degradation of railway tracks due to RCF. This work merges a crack initiation and a crack growth model along with a fully nonlinear multibody model. From a multibody assessment of the vehicle-track interaction, an energy dissipation method is used to identify points where cracks are expected to appear. At these points, crack propagation is calculated considering the contact conditions as a function of crack depth. The proposed methodology has been validated with field measurements, conducted using Eddy Currents provided by the infrastructure manager Network Rail. Validation results show that RCF behavior can be predicted for track sections with different characteristics without the necessity of previous on-track measurements.

The interoperability between on track balises and the on board Balise Transmission Module systems depends on both sides' susceptibility and allowed emissions. For that assessment, the document that governs the testing methodology, tools and procedures (Subset 116) needs to be completed prior to its publication. The present paper proposes an advance beyond the state of the art for the rolling stock emission assessment in terms of the test setup and of the post-processing procedure. The documentation commonly used in ERTMS-related issues has been analyzed and the common tools and procedures have been taken into consideration for the proposal presented by the authors. (C) 2015 Elsevier Ltd. All rights reserved.

Power consumption is one of the most critical issues when designing low-cost electronic devices, such as sensing nodes in wireless sensor networks. To support their operation, such systems usually contain a battery; however, when the battery has consumed all its energy, the node (e.g. the sensor) must be retrieved and the battery replaced. If the node is located in a remote and non-accessible placement, battery replacement can become an expensive (and even impossible) task. This way, energy harvesting has emerged as a suitable alternative to supply low-power electronic systems, by converting ambient energy into electric power. Scavenged energy can be used to directly supply the circuits, or stored to be used when needed. This paper summarises the power needs of a general wireless sensor node and describes the main principles of most representative energy harvesting technologies. Copyright (c) 2012 John Wiley & Sons, Ltd.

Low-frequency (flicker) noise is one of the most important issues in the design of direct-conversion zero-IF front-ends. Within the front-end building blocks, the direct-conversion mixer is critical in terms of flicker noise, since it performs the signal down-conversion to baseband. This paper analyzes the main sources of low-frequency noise in Gilbert-cell-based direct-conversion mixers, and several issues for minimizing the flicker noise while keeping a good mixer performance in terms of gain, noise figure and power consumption are introduced in a quantitative manner. In order to verify these issues, a CMOS Gilbert-cell-based zero-IF mixer has been fabricated and measured. A flicker noise as low as 10.4 dB is achieved (NF at 10 kHz) with a power consumption of only 2 mA from a 2.7 V power supply. More than 14.6 dB conversion gain and noise figure lower than 9 dB (DSB) are obtained from DC to 2.5 GHz with an LO power of -10 dBm, which makes this mixer suitable for a multi-standard low-power zero-IF front-end. Copyright

This paper presents a tutorial on RF and mixed signal circuits design for a digital video broadcasting-handheld tuner. A detailed description of the wideband low noise amplifier, the mixer, the synthesizer and the ADC, which are the most challenging components of a receiver, are carried out. Requirements relative to frequency range, sensitivity, noise figure, linearity, phase noise gain and dynamic range are discussed. The LNA uses a cascode configuration, combining a resistive loaded LNA with a conventional resistive Shunt-feedback, in order to achieve a low power, low noise and wide bandwidth. The mixer uses a classical Gilbert cell configuration. The VCO employs techniques like emitter degeneration, capacitor divider, and optimum bias for minimum noise to improve phase noise requirements and oscillation amplitude. There are two ADC structures, one of which is a delta sigma ADC. The blocks are implemented in a AMS 0.35 mu m BiCMOS process.