Machine learning (ML) has gained widespread adoption in a variety of fields, including computer vision and natural language processing. However, ML models are vulnerable to membership inference attacks (MIAs), which can infer whether access data was used in training a target model, thus compromising the privacy of training data. This has led researchers to focus on protecting the privacy of ML. To date, although there have been extensive efforts to defend against MIAs, we still lack a comprehensive understanding of the progress made in this area, which can often impede our ability to design the most effective defense strategies. In this article, we aim to fill this critical knowledge gap by providing a systematic analysis of membership inference defense. Specifically, we classify and summarize the existing membership inference defense schemes, focusing on optimization phase and objective, basic intuition, and key technology, and we discuss possible research directions of membership inference defense in the future.

Smart railway maintenance is crucial to the safety and efficiency of railway operations. Successful deployment of technologies such as condition-based monitoring and predictive maintenance will enable railway companies to conduct proactive maintenance before defects and failures take place to improve operation safety and efficiency. In this paper, we first propose to develop a classification-based method to detect rail defects such as localized surface collapse, rail end batter, or rail components—such as joints, turning points, crossings, etc.—by using acceleration data. In order to improve the performance of the classification-based models and enhance their applicability in practice, we further propose a deep learning-based approach for the detection of rail joints or defects by deploying convolutional neural networks (CNN). CNN-based models can work directly with raw data to reduce the heavy preprocessing of feature engineering and directly detect joints located on either the left or the right rail. Two convolutional networks, ResNet and fully convolutional networks (FCN), are investigated and evaluated with the collected acceleration data. The experimental results show both deep neural networks obtain good performance, which demonstrate that the deep learning-based methods are effective for detecting rail joints or defects with the expected performance.

Digital transformation has been prioritized in the railway industry to bring automation to railway operations. Digital Twin (DT) technology has recently gained attention in the railway industry to fulfill this goal. Contemporary researchers argue that DT can be advantageous in Railway manufacturing logistics to planning and scheduling. Although underlying technologies of DT, e.g., modelling, computer vision, and the Internet of Things, have been studied for various railway industry applications, the DT has been least explored in the context of railways. Thus, in this paper, we aim to understand the state-of-the-art of DT for railway (DTR), for advanced railway systems. Besides, this survey clarifies how DT can serve the railway twin system designers and developers. As DTR is still in its early adoption stage, there is hardly any clear direction to identify the technologies for specific DTR applications. Therefore, based on our findings we present a taxonomy for DTR for designers and developers. Finally, we describe potential challenges, pitfalls, and opportunities in DTR for future researchers.

Pipelines are economical and efficient modes of transporting oil and gas. Pipelines will inevitably confront various risk factors through their lifespan, which could lead to defects. Defects in pipelines can compromise the integrity of the pipeline systems and may result in catastrophic accidents. Thus, it is vital to conduct the integrity assessment of pipelines so that the safe operation of the pipelines can be ensured. Up to the present, widely used approaches for pipeline integrity assessment include defect characterization, growth rate prediction, and failure pressure evaluation. Although the theoretical developments of pipeline integrity assessment methods have yielded fruitful achievements and significantly benefit the industry practices, there is still a lack of a systematic review covering the whole process from data collection to model establishment of the pipeline integrity assessment. Therefore, a comprehensive review is conducted in this paper on the pipeline defect integrity assessment from the data modeling perspective. Firstly, the description of data required to construct pipeline defect integrity assessment models is presented, where the required data for modeling can be obtained from pipeline inspection measurements, monitoring sensors, testing experiments, etc. Then, different modeling techniques applied to pipeline integrity assessment are reviewed, which are classified into physics-based models, data-driven models, and multi-model fusion. Also, the advantages and limitations of these techniques are discussed. Finally, the possibility of applying the existing models to a digital twin of pipeline defect is explored. This paper aims to provide a guideline for researchers to select optimal models according to data availability and research requirements, which can benefit the research community, as well as, the industry.

Fault detection, diagnostics, and prognostics (FDD&P) ensure the operation efficiency and safety of engineering systems. In the building domain, they can help significantly reduce energy consumption and improve occupant comfort. Specifically, prognostics are becoming increasingly important as a pro-active fault prevention strategy through continuously monitoring the health of energy systems. In this article, we develop a machine learning-based method for building systems. The proposed method can help develop predictive models from historical operation and maintenance data. After the detailed description of the proposed machine learning-based prognostic method, a case study involving prognostics on central heating and cooling plant (CHCP) equipment is provided. To this end, a year's worth of sensor and actuator data from four boilers and five chillers of a CHCP in Ottawa, Canada are collected. The plant operators are interviewed to understand how they handle failure events, and their logbooks are reviewed to extract the date and time of the recorded failure events. The sensor and actuator data up to two weeks prior to each of these failure events are used to develop regression tree models that predict time to failure (TTF). The results indicate that about half of the modeled failure events could be accurately predicted by looking at the data available in the distributed control system. Finally, the future work is outlined.