A BIM-based approach for predicting corrosion under insulation

Automation in Construction 107 (2019) 102923

Contents lists available at ScienceDirect

Automation in Construction journal homepage: www.elsevier.com/locate/autcon

A BIM-based approach for predicting corrosion under insulation a

b

a

a

Yuan-Hao Tsai , Jun Wang , Wei-Ting Chien , Chia-Ying Wei , Xiangyu Wang ⁎⁎ Shang-Hsien Hsieha,

c,d,e,⁎

T

,

a

Department of Civil Engineering, National Taiwan University, Taipei 10617, Taiwan School of Architecture and Built Environment, Deakin University, Geelong, VIC 3220, Australia School of Civil Engineering and Architecture, East China Jiaotong University, Nanchang 330013, China d School of Design and the Built Environment, Curtin University, Perth, WA 6102, Australia e Department of Housing and Interior Design, Kyung Hee University, Seoul, Republic of Korea b c

A B S T R A C T

Corrosion under insulation is one of the most important issues in the petroleum industry. Ordinarily, in order to check the corrosion, inspectors remove the insulation of pipelines to measure the level of corrosion on each section of pipelines. This procedure may take weeks for a site which distinctly affects the financial aspect of oil and gas companies due to the pause production of its high-value products; therefore, in most cases, inspectors spot-check pipeline corrosion based on their experience. However, because the environments on sites are various, experience-based inspection may not be suitable for every site. On the other hand, even though inspectors want to access more data for better understanding of the site before the site trip, historical data sometimes are lost or scattered which leads to a hard situation for preparation of corrosion inspection. This paper utilises passive RFID sensors, which are smart sensing technologies, to collect site data and then integrate them into a Building Information Modeling (BIM) system. A uniform corrosion model is also adapted from the theories of corrosion to leverage both sensor data and BIM elements' properties. They serve as inputs to calculate the corrosion rate which is the key value of corrosion prediction. Then, the corrosion prediction results are colour-coded on a BIM model which helps inspectors intuitively understand the prediction and prepare for the site inspection. In result, the proposed research could provide a novel approach for corrosion management under insulation.

1. Introduction Corrosion is one of the most common damages of pipelines for oil and gas transmission. According to the statistics [1], 25% of the pipeline incidents attributes to corrosion over the last ten years. These damages could cause a waste of gas which leads to a great amount of financial loss. In fact, it is indicated that the expenditures of potential repairs and monitoring for pipeline corrosion cost billions of dollars globally each year [2]. Furthermore, corrosion damages could influence the transmission of oil and gas and cause pipeline failures [3]. According to Cosham et al. [4], when corrosion failures on pipelines occur, it is not a simple ‘corrosion’ failure, but a ‘corrosion control system’ failure which ageing coating, aggressive environment, and rapid corrosion growth may need to be monitored. Therefore, approaches for efficiently managing external corrosion on pipelines are eagerly needed. Pipeline corrosion is an electrochemical reaction and the main component of the corrosion process is the electron transfer which electrons relocate in an aqueous media. Monitoring the voltages and currents of electron transfer is one of the common methods for

corrosion integrity management [5]. In practice, pipelines are inspected periodically by non-destructive test (NDT) tools such as in-line inspection (ILI) tools to identify potential corrosions. The corrosions could be located through different techniques such as ultrasonic transducers and magnetic flux leakages [6,7]. However, monitoring plant sites with the aforementioned inspection tools could cost a lot of manpower and time as the tools interrupt the sites from producing high-value products. In fact, many research studies focus on modeling corrosion defects to highlight potentially corroded elements. Implementing reliability-based assessment of corrosion model is one of the significant approaches. Rather than extrapolating the corrosion depth or corrosion rate mainly from site experience, a reliability-based model could consider several factors such as site condition and sensor values which provides an accurate reference while determining an inspection plan [8]. These corrosion models are various and could be different from assumptions. For instance, deWaard et al. [9] developed a corrosion model considering the effect of temperature and carbon dioxide. In the following research, several correction factors are added to enhance their result. Another model proposed by Nesic et al. [10] focuses on complex effects such as protective scale formation,

⁎

Correspondence to: X. Wang, School of Civil Engineering and Architecture, East China Jiaotong University, Nanchang 330013, China. Correspondence to: S. H. Hsieh, Department of Civil Engineering, National Taiwan University, No. 1, Roosevelt Rd., Sec. 4, 106, Taipei 10617, Taiwan. E-mail addresses: [email protected] (Y.-H. Tsai), [email protected] (J. Wang), [email protected] (W.-T. Chien), [email protected] (C.-Y. Wei), [email protected] (X. Wang), [email protected] (S.-H. Hsieh). ⁎⁎

https://doi.org/10.1016/j.autcon.2019.102923 Received 18 March 2019; Received in revised form 28 June 2019; Accepted 25 July 2019 0926-5805/ © 2019 Elsevier B.V. All rights reserved.

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

2.1. Corrosion prediction

water wetting, and the hydrogen sulfide effect in order to solve the multiphase flow problem. Additionally, Nyborg [11] indicates that there are a considerable amount of corrosion models accounting for oil wetting and the effect of protective corrosion films. In short, while highlighting potential corrosion defects, it is important to choose a model which fits the site environment and the chemical reactions. Smart sensing technology enables the data streaming from sites to the office which makes possible for monitoring sites and providing proper inputs for corrosion models. This technology has been implemented on pipeline monitoring in the industry due to its main features: little processor, small size, wireless, and low-cost [12,13]. Smart sensors are able to collect data on site and send the data back to stations wirelessly with the low installation and maintenance cost [14]. In terms of the communication mechanisms, Wireless Sensor Network (WSN) and Radio Frequency Identification (RFID) are the two common systems used in smart sensing [15]. WSN consists of several sensor nodes which can use Bluetooth, Wi-Fi, or ZigBee to communicate. On the other hand, RFID systems are formed of active or passive RFID tags and RFID readers [16]. Because of different demands of monitoring on sites, such as low cost, long read range or long durability, these systems have been implemented in the industry in different scenarios [17]. In short, smart sensing technology is a reliable method and widely used in industry to steam data from site to office. Building Information Modeling (BIM) technology has been implemented as an integration platform to serve multiple sources of data. Generally, these data could be categorised to graphical and non-graphical data [18]. 2D drawings and 3D models are categorised as graphical data, while documents, sensor values, or materials are categorised as non-graphical data. All the aforementioned information can be stored or referred to BIM models. In fact, many research studies have proved that using BIM techniques to integrate project data is beneficial [18–20]. Sensor data and related maintenance documents from various stakeholders could be integrated and a comprehensive interface could be provided for decision makers while corrosion management. In fact, BIM technology is widely used during construction to visualise multiple sources of data, and it is proved that the efficiency of information delivery could be improved [21]. Sensors can be modeled or highlighted on BIM models to present the location on site. Besides, non-graphical data such as documents or sensor values could be stored or linked to BIM elements. Then, the elements in BIM model could be colour-coded corresponding to non-geometric data. Managers could understand the critical area on the model without reviewing each data on elements. This presents a global view of the data on site. Briefly, BIM technology could serve as an integration platform for data including but not limited to documents, models, and sensors which offers decision makers another way of understanding the data. This paper proposes an approach for predicting corrosion under insulation by utilising smart sensing, mathematical model of corrosion prediction, and BIM technology. Relative research regarding the corrosion prediction and the integration of RFID and BIM technologies will be introduced. Then, a corrosion prediction framework is designed which contains three modules including data collection module, corrosion prediction module, and inspection module. The process of how users interact with the system is illustrated. Besides, a system complied with the proposed framework is developed to validate its feasibility. A case study is adopted to demonstrate the proposed approach. To sum up, a novel approach for predicting corrosion under insulation is proposed and thoroughly introduced in this paper.

Corrosion areas may be predicted based on the great numbers of inputs from sensors and BIM models. Generally, there are two types of approaches, namely data-driven approaches and chemical engineering approach, which are potential to be implemented for corrosion prediction. A data-driven approach is to predict corrosion with a model trained by loads of data. One of the most well-known technologies is machine learning. These trained models would tell the possibility of corrosion based on previous records. This approach is eligible under the proposed mechanism due to the well-collected data. The prediction model can be more accurate as the increasing amount of records across the time as well. However, the model from the data-driven approach may not be easily interpreted because these approaches contain no readable logic. These models would seem like a black box which may not be acceptable in practice. El-Abbasy et al. [22] design an artificial neural network (ANN) models for predicting the condition of offshore pipelines. This research suggests a prediction model which is trained with an ANN proposed by Ward Systems Group in 1996. This research identifies 11 important factors which are relative to the pipeline environment. Then, the inspection data related to those factors are collected from three pipelines in Qatar. After the data is collected, a corrosion prediction model is trained with the ANN. The designed ANN model is compared with the traditional regression models. It is noticed that the performance of both models are close, but the ANN model provides better results. Inspectors could have a better understanding of the condition on site through the prediction model. The inspection or rehabilitation works could be planned and prioritised in advance. Besides, some research studies focus on the corrosion process with specific material and corrosion type. For example, an ANN-based model is designed to predict the status of pitting corrosion of 316 L stainless steel [23]. The model calculates the breakdown potential and determines the status of pitting corrosion automatically along with considering the environmental conditions. The model is evaluated by experimental data and proves that the ANN-based model becomes an efficient tool to predict the status of pitting corrosion. Another type of approaches is based on chemical engineering. The chemical reactions of corrosion are formulated. Besides, each reaction is analysed to calculate its reaction rate which can be converted into corrosion rate (CR) in the unit of millimetres per year. With the safety factor (SF) for engineering purposes, the corrosion can be calculated as shown in Eq. (1) and Fig. 1. As a result, with corrosion rate for each pipe, the corrosion can be predicted with a given time period. Although the accuracy of corrosion rate may highly depend on the chosen theories which may differ from site environments, this approach clearly shows the mechanism of prediction. Therefore, it is more acceptable for industrial applications.

Y (t ) = y0 + (CR × SF ) × t

(1)

Research using chemical methods for corrosion prediction has been proposed since 1970s. One of the first and the most widely used corrosion models for oil and gas industry is proposed by deWaard and

2. Related research studies Two of the major topics are reviewed including approaches of corrosion prediction and research about integration of BIM and sensing technologies. Fig. 1. Terminology for Eq. (1). 2

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

Fig. 2. A BIM-enabled corrosion prediction framework.

By comparing the model with independent loop experiments and empirical models, the corrosion model is evaluated and gives a better picture of the mechanism considering the pH effects, temperature, and solution in electrochemical perspective.

Milliams in 1975 [24], and has been revised in 1995 [25]. They modified the corrosion rate of carbon steel with an equation which is controlled by time, temperature, CO2 fugacity, and CO2 partial pressure. As a result, with the understanding of corrosion on pipelines, this research helps decide necessary actions to protect pipelines from excessive corrosion. Moreover, some specific corrosions are modeled to solve the needs from industry, such as erosion-corrosion phenomena. In erosioncorrosion phenomena, sand is produced during the reactions. Sand particles may entrain in the flow stream and strike the wall of pipeline which increases the corrosion rate at these points. Protective scales and flow velocities are considered to model the erosion-corrosion phenomena [26]. From electrochemical points of view, corrosion models for various materials and environments are designed. For instance, a model for predicting corrosion of mild steel in aqueous carbon dioxide solutions is designed [27]. Several reactions are taken into account such as hydrogen ion (H+) reduction, carbonic acid (H2CO3) reduction, direct water reduction, oxygen reduction, and anodic dissolution of iron.

2.2. Integration of BIM and sensing technologies There are various types of smart sensing technologies that can be implemented on site. In addition, many of them are integrated to BIM models across the building lifecycle. For example, regarding site management during construction, customised sensors are designed and installed in confined spaces to monitor work site and process data. The sensor values are streamed to servers and integrated into BIM models as well. When health and safety hazards happen, health and safety department could take necessary actions with adequate information [28]. During construction, video sensors and laser sensors are adopted and integrated with BIM technology to provide an easy-to-use and useful 3

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

read range and the application scenario. Next, due to long period of corrosion process, it could normally take years to notice the variation of corrosion. Therefore, durability is much concerned in industry. In fact, batteries, the power of sensors, are the major factor for durability. In order to maximise the durability, passive sensors which collect energy from nearby readers are widely utilised in industry. However, the power of passive sensors is relatively small compared to active sensors. Chips on the passive sensors should be energy-efficient which limits the functions of sensing. Regarding cost effectiveness, sensors with long read range and durability could reduce the cost. On the other hand, the more applications benefited from the sensor data could increase the value of sensors which results in increasing the cost effectiveness. Both reducing the implementing cost and increasing the value of sensors are essential while considering the cost effectiveness in industry. In short, read range, durability, and cost effectiveness are three of the most important criteria while selecting sensors in industrial perspective. Next, manually or automatically streaming data from site to system should be considered. Assigning workers to patrol particular area and collect sensor data could be easily implemented. However, data missing or long period of sensing cycle may happen which results in the compromises of quality of the data. An effective method for managing the groups of workers may be needed. On the contrary, automatically collecting sensor data is robust. With an array of readers implemented on site, sensor data could be updated to the system up to every minute. In fact, the frequency of sensing may be crucial in some scenarios because some changes on site happen at a glance. For example, gas leaking on pipelines may only happen few seconds but regularly. If the frequency of sensing is not intense enough, the leaking could not be noticed which could influence predicting results. In terms of data storage, both electrical signal as known as raw data and sensor value should be stored. The reason is that most sensor values are converted from the electrical signal which generated from the sensor on tags. Specifically, when the environment changes, e.g. temperature changes or moisture changes, resistance on the sensor changes which results in electric current changes. Most sensors capture the changes and converted to sensor values. Therefore, because of the indirect sensing method, the electrical signal could provide a better reference for calibrating the sensor values. Nowadays, because the technologies of local databases or cloud database services are mature, all sensor data can be stored in these databases effortlessly. Storing sensor data in local databases or cloud databases may be attributed to a business decision. For the corrosion prediction module, algorithms and their inputs should be considered. Typically, the sources of inputs are from BIM models and sensors which represent the source of static inputs and dynamic inputs respectively. Regarding the static data, BIM models could be a repository for elements' properties because most properties are already embedded in BIM models while modeling such as material and thickness of pipelines. By leveraging the BIM models, properties could be collected as inputs for corrosion model with less efforts. In addition, BIM technologies are implemented in more and more applications for facilities management (FM) [31]. Properties of elements in BIM models are updated regularly by FM systems. Corrosion model could access the latest elements' properties which reflects the real site conditions. For the dynamic data, the various inputs which corrosion models depend on are from sensors. For example, humidity and temperature are the inputs of dynamic data for uniform corrosion model [32]. The inputs are streamed from sensors and updated frequently. In short, static data and dynamic data serves as inputs for corrosion model to predict the status of corrosion for pipelines. As regards to the algorithms of corrosion model, theory-based algorithms or data-driven algorithms could be implemented for calculating corrosion rate which is used for corrosion prediction. As aforementioned studies, the fundamental of theory-based algorithms is chemical engineering. The process of corrosion is modeled through electrochemical or experimental approaches. On the other hand, data-driven algorithms leverage the

navigation system for tower crane operators. Instead of numerical data from collision-detection systems, tower crane operators could understand surroundings in 3D environment better. It is proved by on site implementation that tower crane operators heavily rely on the proposed navigation system compared with the numerical anti-collision system [29]. The integration of sensing and BIM technologies could be implemented in operation and maintenance stage as well. Series of sensors are installed on bridge deck deicing system to monitor the state of bridge deck under different climate condition [20]. In brief, integration of sensing and BIM technologies could be implemented across lifecycle for both buildings and infrastructures. One of the commonly used sensors on site is RFID sensors. Ordinarily, there are two types of RFID sensors, passive and active. Both passive and active RFID sensors can receive signal from readers, and return the encoded information. However, the mechanism of the sensors is different. Passive RFID sensors generate power from reader's signal which means they do not need batteries. Therefore, they can last for a long period. On the other hand, active RFID sensors have batteries. They send out signal powered by themselves. Because of the difference of power source, an active RFID sensor usually provides a longer read range than a passive RFID sensor; however, the lifetime of an active RFID sensor is generally shorter than a passive RFID sensor. Several research studies are conducted to leverage RFID sensors for information integration. A pilot study from Meadati et al. [18] points out that RFID sensors could reduce tedious works of mapping digital objects to elements in real world. In the following research, an integration of RFID sensors and BIM models is implemented to simulate the lighting levels on a physical model in real-time which helps testing the facade system before construction [30]. To sum up, the aforementioned research utilised sensing and BIM technologies to build the bridge between digital environment and real site. This connection could be adopted in this research to collect pipelines' condition for corrosion prediction. 3. A BIM-based approach for predicting corrosion under insulation An approach for predicting corrosion under insulation is proposed in the aspect of the system framework and the interactions between stakeholders and the system. 3.1. A BIM-based framework for corrosion prediction In order to manage corrosion under insulation, a framework leveraging the power of BIM, corrosion model, and sensing technologies is developed. Sensor values and elements' properties could stream to the system as inputs for corrosion model. Then, the result of corrosion prediction could be visualised in a 3D environment powered by BIM technology. These could help identify the potentially corroded area. Fig. 2 illustrates the framework which consists of three modules: data collection with smart sensing technologies, mathematical model for corrosion prediction, and BIM-enabled external pipeline corrosion prediction. For the data collection module, types of sensors, how the data be transferred and stored should be considered. Due to the large scale of plant site, how to install or maintain sensors should be considered while selecting sensors. Read range, durability, and cost effectiveness are the three criteria for choosing sensing technologies. Long read range is normally needed on plant site because it could save lots of manpower while collecting sensor data. On that account, RFID sensors working in ultra-high frequency (UHF) which is normally around 900 (MHz) are preferred. The reason is that read range and working frequency are positive-correlation. Higher working frequency could result in better read range. Besides, the size of antenna on sensors is relative to read range. It is better to select a large antenna for better read range; however, the size and shape of antenna may influence the applicability on site. Therefore, the selection of antenna should balance the needs of 4

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

Fig. 3. Process model of predicting corrosion under insulation. 5

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

tasks could be planned and documented because lots of information are stored or could be referred through elements in BIM models. Most of tasks are well-planned before the site trip. In addition, these tasks could be saved in a database and ready for on-site usage.

emerging technologies of data science. Inputs and maintenance records could serve as training data to develop a machine learning model to help identify potential corrosion. Briefly, either theory-based or datadriven algorithms could provide a predictive model for corrosion prediction. For the inspection module, the visualisation of predictive results and how the results support the inspection works should be considered. Regarding the visualisation and smart inspection planning, sensor data and the results from corrosion prediction module should be visualised through BIM technology. Elements can be colour-coded by the scale of sensor values or corrosion depth. This could help inspectors identify the potentially corroded area. In addition, spot-check lists could be suggested by the system automatically. Routes of inspection could be schemed before site trip as well. These provide a straightforward plan for inspection which eliminates the missing of critical spots on site. This approach is consistent with other research studies which agree that the effectiveness and efficiency of inspection could be improved [33,34].

4. System design and implementation A system has been designed and developed to validate the feasibility of proposed framework. A training site from one of the biggest oil and gas company in Australia is chosen to be the testbed. It could serve as the main area where the sensor data can be collected. Although sensors are not deployed on site at this stage, capability assessments of sensors are conducted. As a result, the read range, read angle, and sensor data from simulated environment could be acquired. These help determine a cost-effective and efficient plan for sensor deployment. By mapping sensor data to pipelines, corrosion rate on each section of pipeline could be calculated through sensor data and elements' properties. Therefore, corrosion severity for the next few years can be predicted.

3.2. Process modeling of the approach for corrosion prediction 4.1. System architecture The process model of proposed approach is illustrated in Fig. 3 by implementing Business Process Modeling Notation (BPMN) tool [35]. Specifically, how users interact with the system and the mechanism of the system are identified. Four groups in the process model including sensor implementation, sensor data streaming, data visualisation, and tasks for inspection are elaborated in the followings:

A system is developed on the basis of proposed framework in order to validate the feasibility of the approach. Essential components are identified in the system architecture shown in Fig. 4. To start with, RFID readers are chosen to read tags on site and communicate with sensor data hub in a local wireless internet environment. The sensor data hub integrates all reader's values and streams sensor values to system through Hypertext Transfer Protocol (HTTP) requests. The web server located at office processes the data streaming through HTTP requests. It acts as a gate to handle requests of inputs and outputs. Therefore, the web UI could get sensor data or analysed results from web server to present the information in a user-friendly way. In addition, three main logics support the data processing and analysing which are the sensor data processing, algorithm for computing corrosion rate, and spot-check suggestion. To support the aforementioned logics, four databases are built: raw signal database, sensor value database, BIM model database, and inspection task database. The logics and related databases are the core of approach which are derived from the proposed framework.

1. Sensor implementation: To start with, proper sensors are selected and the BIM model are prepared. Then, the sensors are codified to map the elements in BIM model. Normally, general unique identifications in elements' property are chosen for codifying sensors because they are unique and available in every element. After preparing sensors and BIM models, the BIM models should be uploaded to the system and stored in specific model databases for the following usage. Meanwhile, sensors could be attached to pipelines on site to finish the hardware implementation. 2. Sensor data streaming: Either manually or automatically transfer sensor data could be adopted while streaming the data from site to office. For manually collecting sensor data, workers are assigned to a specific area and requested to scan all sensors in that area. Afterwards, the workers need to upload the sensor data from reader to the system. On the other hand, if fixed readers are installed, sensors are regularly read and the readers upload sensor values to the system automatically. No matter how sensor data are collected, the system processes the sensor data and save these data into raw signal database and sensor value database which is a readable value converted from raw signal. Then, users could decide their operations for the sensor data. 3. Data visualisation: There are three operations while visualising the sensor data: visualising sensor data, visualising corrosion prediction, and visualising spot-check suggestions. First of all, sensor values are embedded in the corresponding elements in BIM model by accessing BIM model database and sensor value database. Therefore, elements could be colour-coded by the user-defined scale. Secondly, the sensor values and elements' properties are served as inputs for mathematical model of corrosion prediction. Corrosion rate could be calculated based on the inputs. With a given target date which is normally 3 to 5 years from present, BIM models could be colourcoded by the depth of corrosion. Lastly, based on the colour-coding of sensor values and corrosion prediction on BIM models, users can check the highlighted elements for better understanding of the site. Because a 3D environment is provided for visualising the suggested spot-check elements, 3D operations such as zooming, rotating, filtering models could be applied to help identifying tasks for inspection. 4. Tasks for inspection: While deciding tasks for inspection, details of

4.2. Data collection Each sensor has its own advantage and dedicates for specific environment. In general, passive RFID sensors which are battery-free and wireless are essential. Rather than using active sensors which need a power source inside, passive sensors do not need batteries of wiring for power. Therefore, the action of changing battery is eliminated. Because of this elegant mechanism, passive sensors are almost maintenance-free devices and designed for set-n-forget operation. It makes passive RFID sensors cost-effective by the reduction of maintenance fee. In order to collect applicable values from both temperature and moisture sensors on pipelines, a series of passive sensors providing by RFMicron, a sensing solution company, are considered. For moisture sensors, three sensors are taken into consideration: RFM2100, RFM2110, and RFM2120 as shown in Fig. 5. These sensors are dedicated for different working environment. Firstly, RFM2100 is dedicated for sensing environmental and material moisture, for example, monitoring moisture on food products, building and roofing materials, or moisture detection for corrosion prevention. It is configured to be sensitive to moisture and rain which results in relatively accurate to other moisture sensors. However, this sensor is not suitable for direct use on metallic surfaces due to the high interference with radio frequency. Second to all, RFM2110 detects in direct contact or liquid brought to the sensor through wicking tails. Because of this design, RFM2110 is configured to detect moisture when placed on metallic surfaces. Therefore, it could be deployed to the application of 6

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

Fig. 4. System architecture for predicting corrosion under insulation. (For interpretation of the references to colour in this figure, the reader is referred to the web version of this article.)

In terms of temperature sensors, three sensors are taken into consideration as well which are RFM3200, RFM3240, RFM3250 as shown in Fig. 5. As the aforementioned discussion, the ability of working on metallic surfaces is essential. Therefore, only RFM3240 and RFM3250 are applicable for the proposed approach because RFM3200 is not configured to direct use on metal. On the other hand, both RFM3240 and RFM3250 can get exact values for temperature. The differences between these two sensors are the size of sensors, read range, and read angle. In order to decide a better fit to the practical use, detailed comparisons are conducted. First of all, in terms of the size of sensors, RFM3240 has larger size than RFM3250 by two times wider and five times higher as shown in Fig. 6. Despite RFM3250 has a more compact build, both sizes of sensors are suitable for deployment on site.

moisture detection in automotive vehicle assembly, aircraft maintenance or corrosion prevention programs. Lastly, RFM2120 are capable of detecting moisture similar to RFM2100. They are configured to be sensitive to moisture and rain, but only work on non-metallic surfaces. The only difference is that RFM2120 is sensitive to a slight change of moisture which mostly targets the medical applications. In brief, RFM2110 is chosen as the moisture sensor. The reason is that most of pipelines on site are made of metal. It is essential to choose a metal-compatible sensor which could resist the interference of radio frequency. Besides, even though the sensor values may not be sensitive to a slight change of moisture, sensor values from RFM2110 suffice the need of corrosion prediction because only the scale of moisture is needed.

Fig. 5. Moisture and Temperature sensors: (a) RFM2100 - wireless moisture sensor (b) RFM2110 - wireless moisture sensor (c) RFM2120 - wireless moisture sensor (d) RFM3200 - wireless temperature sensor (e) RFM3240 - wireless temperature sensor (f) RFM3250 - wireless temperature sensor. 7

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

Fig. 6. Size of temperature sensors: (a) RFM3240 - wireless temperature sensor (b) RFM3250 - wireless temperature sensor.

Fig. 7. Read range and read angle for RFM3240.

be an intermediary of the reader and server aimed at storing data (Fig. 9). Data hub can assist in converting the raw data into meaningful information and display them onto the customised dashboard as Fig. 10.

Regarding read range and angle, a lab test is conducted. Both sensors are attached on a pipeline to measure the maximum of read range and read angle. The results for these sensors are shown in Figs. 7 and 8. Specifically, RFM3240 has approximately 8 m for its maximal read range and 30 degrees for its read angle; on the other hand, RFM3250 has approximately 1 m for its maximal read range and 10 degrees for its read angle. It is indicated that RFM3240 has a better read range and read angle than RFM3250 which provides a possibility of minimal amount of sensors while deploying on site and reduces the initial cost of installation. In short, RFM3240 is chosen for temperature sensors because of its suitable size, better read range and angle compared to others. Fixed reader AR52 RFM5008 manufactured by Nordic ID has been chosen. The main reason why this reader is chosen is the reader's ability to automatically detect the sensor data over a certain period of time. Using AR52 RFM5008 along with the antenna, data can be updated every 30 s and stored into the database which can help in easy extraction and input into the BIM system. Apart from the reader, a customised data hub has also been used which is a raspberry pi system to

4.3. Streaming sensor values to the BIM model In order to visualise sensor values or prediction results, it is essential to acquire BIM applications which enables users to visualise 3D BIM model and develop user-defined extensions to achieve specific functions. Forge is one of the aforementioned applications developed by Autodesk company. It encompasses all the functions needed in a visualisation tool and the main difference of this platform from others is its web-based characteristic. Through any internet browser, users can view the 3D BIM model colour-coded with sensor values and have the real-time corrosion prediction result for each pipeline displayed in element's property panel. Fig. 11 shows an example of a web-based BIM model viewer. To insert the sensor data into BIM model for visualisation, system 8

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

Fig. 8. Read range and read angle for RFM3250.

Fig. 9. (a) Fixed reader AR52 RFM5008 (b) data hub.

Tracking”, “Temperature Calibration”, “Temperature Tracking”, “Moisture Delta”, “Corrosion Rate”, “Prediction Time Duration” and “Corrosion Prediction” as shown in Fig. 11. Among the embedded properties, corrosion rate is calculated by the algorithm and the prediction time duration is from user inputs on web UI. In short, the example shows the mechanism of streaming sensor data to elements in the BIM model and integrated with the proposed system.

architecture shown in Fig. 4 is elaborated. While RFID readers monitoring RFID sensors (Fig. 4a) on site, the sensor data are continuously collected by sensor data hub (Fig. 4b). Then, HTTP requests with sensor data are sent to web server (Fig. 4d) to process the data input. The data are converted by the logics of sensor data processing (Fig. 4e) and stored in raw signal database (Fig. 4h) and sensor value database (Fig. 4i) for further usages. Because all RFID sensors are pre-coded with the element identification in BIM model, the sensor values could be easily mapped to BIM elements when needed. Once user access the web UI (Fig. 4c), the system retrieves the model from BIM model database (Fig. 4j) and the sensor data from sensor value database (Fig. 4i) to display both static and dynamic data in the system. In brief, it is shown that the data hub acts as a bridge to stream data from numerous readers on site to the cloud system. With these abundance of sensor data, sensor data visualisation and corrosion predictions could be realised. When corrosion prediction function is triggered from the web UI (Fig. 4c), the system retrieves sensor values from sensor value database (Fig. 4i) to get the inputs for the algorithm of corrosion prediction. The predicting process (Fig. 4f) then calculates estimated corrosion depth at a given time period. As a result, the estimation of corroded depth is embedded in element properties in BIM model. Besides, these values could be colour-coded on BIM model to have a better understanding of the corrosion prediction in a larger scale. Whenever clicking on a pipeline element from BIM model, value in element's attributes would update. Apart from some static information received from BIM model itself, some extra sensor information is attempted to be embedded in each BIM element from the system. A property group called “Sensor Data” is created to accommodate our sensor data measured, such as “Moisture Calibration”, “Moisture

4.4. Mathematical model of corrosion prediction As the emergence of oil and gas normally accompanies by water and various amounts of acid gases, such as carbon dioxide, CO2, it affects the integrity of mild steel which is often used construction material in oil and gas industry. To understand the CO2 corrosion of mild steel, this research integrates with an entry level of corrosion simulation and prediction model which mathematically illustrates the electrochemical process of the CO2 corrosion. To be specific, a uniform corrosion is assumed as the corrosion prediction model in this research. The reason of the selection is that there are some factors which affect the reaction during the corrosion, yet not easily integrate with sensors. For example, the dissolution of iron is pH dependent. Namely, the more iron is dissolved, the faster corrosion is reacted which results in higher corrosion rate. However, there may be difficulties for acquiring passive sensors to monitor pH scale due to the lack of power on passive sensors. This situation may be fixed once the sensing technology comes up with energy-efficient sensors. Until then, selecting a simplified corrosion model which assumes some factors as constants could be an alternative. It may influence the accuracy of corrosion prediction, but still be applicable in practical usages. 9

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

Fig. 10. Real-time sensor values on the AXZON dashboard.

overall reaction of the electrochemical process for carbon steel is shown in Eq. (2).

Mild steel is chosen as standard pipe material in the uniform corrosion. The pipelines are corroded by acid gases such as H2CO3 which is generated from carbon dioxide. In these series of reactions, two main elements for corrosion are carbon dioxide, CO2, and water, H2O. The

Fe + CO2 + H2 O → FeCO3 + H2

(2)

Fig. 11. Property Panel of a section of pipeline in web-based BIM model viewer. (For interpretation of the references to colour in this figure, the reader is referred to the web version of this article.) 10

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

Eq. (11) illustrates the calculation of current potential of hydronium ion. In terms of symbols, Erev(H+)is reversible potential for H+ion reduction (V). bc(H+)is cathodic Tafel slope for H+iron reduction (V). Because hydronium ion is reversible as shown in Eq. (5) and (6), Erev(H+) is calculated in Eq. (14). Besides, Eqs. (12), (13), and (14) are affected by temperature.

To look into the details of electrochemical reactions of Eq. (2), the electrochemical dissolution of iron in water is the dominant anodic reaction in corrosion which shown as Eq. (3). This reaction is pH dependent in acidic solutions which means the lower the pH scale is, the faster the iron dissolution happens which results in affecting the overall reaction. However, because of the assumptions from selected corrosion model, pH scale of corrosion is assumed as neutral which means the pH scale will not affect the corrosion.

Fe → Fe 2 + + 2e−

(3)

The CO2 gas in the environment is soluble in water and make carbonic acid, H2CO3. The dissolution of carbon dioxide is expressed as Eq. (4). This reaction is affected by temperature and moisture. Namely, the solubility of carbon dioxide decreases while the temperature increases which increases the reaction rate to acidic gases, H2CO3. This carbonic acid from Eq. (4) further dissociates in water to produce hydronium which shown in Eqs. (5) and (6). Apart from the temperature factor, moisture could affect the reaction rate in Eq. (4) as well. The carbon dioxide could only form carbonic acid, H2CO3, with adequate amounts of moisture, H2O. In other words, if there is not enough moisture, carbon dioxide in Eq. (4) stays the form itself and produce no acidic gas which stops the reaction of corrosion. In short, both temperature and moisture are two important factors when it comes to corrosion rate.

ic (H +) = iO (H +) 10(Ecorr − Erev (H +) )/ bc (H +)

(11)

1 1 ⎛ −ΔH(H +) ⎛ ⎞⎞ iO (H +) = iOref(H +) exp ⎜ − ⎜ ⎟ R T + 273.15 T + 273.15 ⎠ ⎟ c c , ref ⎝ ⎠ ⎝

(12)

bc (H +) =

2.303R (Tc + 273.15) 0.5F

Erev (H +) = −

2.303R (Tc + 273.15) pH F

(13) (14)

Eq. (15) calculates the current potential of carbonic acid. As for the symbols, Erev(H2CO3)is reversible potential for H2CO3 reduction (V). bc(H2CO3)is cathodic Tafel slope for H2CO3 reduction (V). Since the reductions of H2CO3and H+are equivalent, Erev(H2CO3)is same as Erev(H+) shown in Eq. (18). In addition, Eq. (16), (17), and (18) are affected by temperature.

(4)

ic (H2 CO3) = iO (H2 CO3 ) 10(−Ecorr − Erev (H2 CO3) )/ bc (H2 CO3)

H2 CO3 ⇔ H+ +

HCO3−

(5)

HCO3−

CO32 −

(6)

1 1 ⎛ −ΔH(H2 CO3 ) ⎛ ⎞⎞ iO (H2 CO3) = iOref(H2 CO3 ) exp ⎜ − ⎜ ⎟ ⎟ R T + 273.15 T + 273.15 c c ref , ⎝ ⎠⎠ ⎝ (16)

CO2 + H2 O ⇔ H2 CO3

⇔

H+

+

By calculating the reaction rates from each aforementioned dissolution, the corrosion rate for pipelines could be identified. There are two variables taken into consideration: temperature and moisture. Temperature values from sensors serve as inputs (Tc) in the following equations to calculate the reaction rate for each electrochemical process. On the other hand, moisture values serve as the determination whether the electrochemical process could start. As a result, corrosion rate on each pipeline could be calculated independently based on the attached temperature and moisture sensors. These corroded reactions are a series of charge transfer. To specify the charge transfer for the aforementioned reactions, the overall corrosion potential (ic) and current density (ia) can be written as Eq. (7). By using Tafel equation [36], electrochemical kinetics for the electrochemical reactions could be calculated. Each corrosion potential and current density in Eq. (7) can be formulated, as shown in Eqs. (8), (11), (15), and (19).

ic (H +) + ic (H2 CO3) + ic (H2 O) = ia (Fe) = icorr

bc (H2 CO3) =

2.303R (Tc + 273.15) 0.5F

Erev (H2 CO3) = −

2.303R (Tc + 273.15) pH F

(15)

(17) (18)

Eq. (19) identifies the current potential of water. With regard to symbols, Erev(H2O)is reversible potential for H2O reduction (A m−2). bc(H2O)is cathodic Tafel slope for H2O reduction (V). Because the reductions of H2Oand H+are equivalent, Erev(H2O)is same as Erev(H+) shown in Eq. (22). In addition, Eqs. (20), (21), and (22) are affected by temperature.

(7)

ic (H2 O) = iO (H2 O) 10−(Ecorr − Erev (H2 O) )/ bc (H2 O)

(19)

1 1 ⎛ −ΔH(H2 O) ⎛ ⎞⎞ iO (H2 O) = iOref(H2 O) exp ⎜ − ⎜ ⎟ ⎟ R T 273.15 T 273.15 + + c c , ref ⎝ ⎠⎠ ⎝

(20)

2.303R (Tc + 273.15) 0.5F

Eq. (8) shows the calculation of current density of iron. Regarding symbols, iOis the exchange current density of dissolution. Ecorris corrosion potential (V). Erev(Fe)is reversible potential of dissolution, Erev(Fe) = − 0.488 V. ba(Fe)is the anodic Tafel slope for Fe dissolution.

bc (H2 O) =

ia (Fe) = iO (Fe) 10(Ecorr − Erev (Fe) )/ ba (Fe)

By substituting the expressions in Eqs. (7) to (22), the only unknown corrosion potential, Ecorr, can be solved. The value of Ecorris returned to the Eqs. (7). Then, the reaction of corrosion, icorr, can be solved. Once icorr is solved in Eq. (7), the corrosion rate (CR) can be further calculated. Finally, the corrosion rate is computed by Faraday's law [32] as shown in Eq. 23 where M is the molecular mass, ρ is the density, n is the number of electrons and F is the Faraday's constant. If the unit amperes per square meters is used for the corrosion current density, icorr, the corrosion rate expressed in millimeter per year is computed as: CR = 1.155 icorr.

Erev (H2 O) = −

(8)

The exchange current density, iO(Fe), and the anodic Tafel slope, ba(Fe), are functions relative to temperature. Eqs. (9) and (10) explain the functions where iO(Fe)ref is reference exchange current density of Fe oxidation, iO(Fe)ref = 1 Am−2, ΔHFeis activation enthalpy, ΔHFe = 50 kJ mol−1, R is universal gas constant, R = 8.314 J mol−1K−1, Tcis temperature (°C), Tc, refis reference temperature, Tc, ref = 25 ° C, F is Faraday's constant, F = 96485 C mole−1.

1 1 ⎛ −ΔHFe ⎛ ⎞⎞ iO (Fe) = iOref(Fe) exp ⎜ − ⎜ ⎟ R ⎝ Tc + 273.15 Tc, ref + 273.15 ⎠ ⎟ ⎠ ⎝

ba (Fe)

2.303R (Tc + 273.15) = 1.5F

(9)

2.303R (Tc + 273.15) pH F

CR (mm per year) =

icorr MFe ρFe n F

(21) (22)

(23)

Inputs for the calculation of corrosion rate can be extracted from BIM models due to the well-arranged properties within each element.

(10) 11

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

pipeline, the predictive depth of corrosion can be calculated through Eq. (1) which considers prediction time duration, corrosion rate, and safety factor all together. Among the inputs of the calculating corrosion depth, prediction time duration is a user input which specifies the target date of corrosion prediction. As shown in Fig. 14, users can pick up prediction time through the interface. After selecting the time, the attribute “Prediction Time” of each element will automatically update. As the corrosion rate for each element has been calculated in advance, the corrosion prediction depth would also update and embedded in elements. Predictive corrosion depth is colour-coded on pipelines to identify corrosion level on site. The scale of colour coding is customisable to accommodate various site conditions. Generally, values of the scales could be set along with the inspection needs. In this case, as Fig. 15 shown, grey colour is configured as normal which corrosive depth is less than 3 mm; ivory colour is configured as the “may inspect” pipelines which corrosive depth is between 3 and 5 mm; on the other hand, the dark red colour is configured as the “must inspect” pipelines which corrosive depth is over 5 mm. In result, the pipeline with dark red colour means replacement is needed severely whereas the grey pipeline indicates no attention is needed on that particular item. These colour coding intuitively assist users in understanding which pipeline needs detailed inspection. In some cases, target pipelines for inspection may not be easily identified because the corrosion depth of these pipelines are all categorised into same level which results in same colour on the BIM model. In this situation, the prediction time of attributes should be modified to find a suitable value for distinguishing severely corroded pipelines from slightly corroded pipelines. As an alternative, a rough area with corroded pipelines could be identified through the colour-coded BIM model. Then, inspectors could compare among sensor values of the pipelines and locate the possible corroded pipelines by their site experience. In short, traditionally, an inspection takes place while inspector assumes that the pipeline in some particular area is heavily corroded based mainly on site experience or some laboratory experiment. However, with the proposed approach, inspectors can take advantage of the corrosion value embedded in each element and the colour-coded visualisation to make decision as to where to inspect. It is not necessary to find the exact corrosion depth of every pipeline, at least the BIM model with colour coding function can assist in identifying the trend of corrosion which can be auxiliary information for the inspector to

Fig. 12. Inputs for corrosion model from BIM model.

By integrating sensor values and static properties in BIM model, these inputs can easily fit into the corrosion rate calculation as shown in Fig. 12. Although the assumptions may be different from each corrosion model and result in different inputs, Fig. 12 lists the potential properties for corrosion rate calculation. 4.5. Application for predicting corrosion under insulation One of the main goals of this research is to assist users in efficiently identifying which pipeline suffers from heavy corrosion and needs to be replaced. A web-based system powered by Forge is developed to integrate the sensors, mathematical model, and BIM models. To visualise the results in the system, three different thresholds have been set for moisture, temperature and corrosion depth respectively. Once the sensor data already are imported into Forge, a corresponding colour would also be given to each pipeline element in accordance with the thresholds. For example, after clicking on the moisture visualisation button, the black colour would be given to the pipeline with moisture delta value below 0.01 mm, and the blue colour represents pipeline with moisture delta value over 0.01 mm as shown in Fig. 13. As for the corrosion visualisation, the same mechanism is used, yet different colours are chosen. When the corrosion prediction is triggered, the application retrieves the latest sensor values on each pipeline for the algorithm of corrosion prediction. Then, the algorithm calculates the corrosion rate, which unit is millimeter per year, then embedded the value on each element of pipelines. With the corrosion rate of each

Fig. 13. Colour coding for moisture values on pipelines. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) 12

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

Fig. 14. Selecting a time period for corrosion prediction.

Fig. 15. Colour coding for corrosion depth on pipelines. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

there is less time demand for spot-checking the status of corrosion under insulation. To specify the approach, a framework for corrosion prediction is developed and the process of the approach derived from the framework is modeled. A demonstrative case is implemented to validate the practicality of the approach as well. Limitations are identified in this research. To start with, mathematical models of corrosion prediction are various regarding the theories and environments. In this research, a uniform corrosion model which assumes ideal factors such as neural pH scale or sufficient Oxygen is selected. The material of pipeline is assumed as mild steel as well. Therefore, in some cases, the mathematical model should be adjusted to fit the usage. Besides, inputs of the model may be limited by the types of sensing technology. Factors in the corrosion model may

consider. 5. Conclusions, limitations, and future works Corrosion under insulation has been an issue in petroleum industry for decades. An approach for managing corrosion is eagerly needed to save the maintenance cost. This research proposed a novel approach to predict corrosion under insulation by implementing the sensing technology and mathematical model of corrosion prediction into a BIMbased system. The sensor values and analytic results of corrosion prediction could be visualised on BIM models. As a result, inspection areas could be decided mainly by the data rather than experience or instincts. Besides, the period of pausing production could be reduced because 13

Automation in Construction 107 (2019) 102923

Y.-H. Tsai, et al.

need to be collected indirectly or set as constants which may influence the accuracy of prediction. To maximise value of the proposed approach, sensors and readers which are two of the major cost while implementing the approach could be economised in the future. The location and angle of sensors or readers could be deployed in a way of covering most of the site. As the initial cost reducing, the proposed approach could be more beneficial and practical to the industry.

[18]

[19]

Acknowledgements [20]

This work was undertaken with the benefits of a research project sponsored by Kaefer Integrated Services (Project Name: Pipeline External Corrosion Monitoring and Prediction through Passive RFID and BIM Integration).

[21]

References [1] European Gas Pipeline Incident Data Group, Gas pipeline incidents 10th report of the European gas pipeline incident data group (period 1970-2016), Retrieved May 16, 2019, from, 2018. https://www.egig.eu/startpagina/$61/$108. [2] National Association of Corrosion Engineers, Recommended Practice: Control of External Corrosion on Underground or Submerged Metallic Piping Systems, National Association of Corrosion Engineers International, Houston, United States of America, 2005 Retrieved May 16, 2019, from https://www.onepetro.org/ standard/NACE-SP0169-2013. [3] A.C. Benjamin, J.L.F. Freire, R.D. Vieira, D.J. Cunha, Interaction of corrosion defects in pipelines–part 1: fundamentals, Int. J. Press. Vessel. Pip. 144 (2016) 56–62, https://doi.org/10.1016/j.ijpvp.2016.05.007. [4] A. Cosham, P. Hopkins, K.A. Macdonald, Best practice for the assessment of defects in pipelines–corrosion, Eng. Fail. Anal. 14 (7) (2007) 1245–1265, https://doi.org/ 10.1016/j.engfailanal.2006.11.035. [5] H.R. Vanaei, A. Eslami, A. Egbewande, A review on pipeline corrosion, in-line inspection (ILI), and corrosion growth rate models, Int. J. Press. Vessel. Pip. 149 (2017) 43–54, https://doi.org/10.1016/j.ijpvp.2016.11.007. [6] T.A. Bubenik, J.B. Nestlroth, R.J. Eiber, B.F. Saffell, Magnetic flux leakage (MFL) technology for natural gas pipeline inspection, Nondestruct. Test. Eval. Int. 1 (30) (1997) 36, https://doi.org/10.1016/issn.0963-8695. [7] F. Varela, M. Yongjun Tan, M. Forsyth, An overview of major methods for inspecting and monitoring external corrosion of on-shore transportation pipelines, Corros. Eng. Sci. Technol. 50 (3) (2015) 226–235, https://doi.org/10.1179/ 1743278215Y.0000000013. [8] K.A.T. Vu, M.G. Stewart, Structural reliability of concrete bridges including improved chloride-induced corrosion models, Struct. Saf. 22 (4) (2000) 313–333, https://doi.org/10.1016/S0167-4730(00)00018-7. [9] C. De Waard, U. Lotz, D.E. Milliams, Predictive model for CO2 corrosion engineering in wet natural gas pipelines, J. Sci. Eng. 47 (12) (1991) 976–985, https:// doi.org/10.5006/1.3585212. [10] S. Nesic, J. Cai, K.L. Lee, A multiphase flow and internal corrosion prediction model for mild steel pipelines, In the Proceeding of Corrosion 2005, 3rd -7th April, 2005. Texas, United States of America, 2005 Retrieved 16 May, 2019, from https://www. onepetro.org/conference-paper/NACE-05556. [11] R. Nyborg, CO2 corrosion models for oil and gas production systems, In the Proceeding of Corrosion 2010, 14th -18th March, 2010. Texas, United States of America, 2010 Retrieved 16 May, 2019, from https://www.onepetro.org/ conference-paper/NACE-10371. [12] B.F. Spencer Jr., M.E. Ruiz-Sandoval, N. Kurata, Smart sensing technology: opportunities and challenges, Struct. Control. Health Monit. 11 (4) (2004) 349–368, https://doi.org/10.1002/stc.48. [13] M. Dener, C. Bostancıoğlu, Smart technologies with wireless sensor networks, Procedia Soc. Behav. Sci. 195 (2015) 1915–1921, https://doi.org/10.1016/j. sbspro.2015.06.202. [14] M.Y. Aalsalem, W.Z. Khan, W. Gharibi, M.K. Khan, Q. Arshad, Wireless sensor networks in oil and gas industry: recent advances, taxonomy, requirements, and open challenges, J. Netw. Comput. Appl. 113 (2018) 87–97, https://doi.org/10. 1016/j.jnca.2018.04.004. [15] L. Ruiz-Garcia, L. Lunadei, P. Barreiro, I. Robla, A review of wireless sensor technologies and applications in agriculture and food industry: state of the art and current trends, Sensors (Basel, Switzerland) 9 (6) (2009) 4728–4750, https://doi. org/10.3390/s90604728. [16] O.T. Arulogun, A.S. Falohun, N.O. Akande, Radio frequency identification and internet of things: a fruitful synergy, Br. J. Appl. Sci. Technol. 18 (5) (2016) 1–16, https://doi.org/10.9734/BJAST/2016/30737. [17] S. Petersen, P. Doyle, S. Vatland, C.S. Aasland, T.M. Andersen, D. Sjong,

[22]

[23]

[24]

[25]

[26]

[27]

[28]

[29]

[30]

[31]

[32]

[33]

[34]

[35]

[36]

14

Requirements, drivers and analysis of wireless sensor network solutions for the Oil & Gas industry, In the Proceeding of Emerging Technologies and Factory Automation, 25th -28th September, 2007, 2007, https://doi.org/10.1109/EFTA. 2007.4416773 Patras, Greece. P. Meadati, J. Irizarry, A.K. Akhnoukh, BIM and RFID integration: a pilot study, In the Proceeding of the Second International Conference on Construction in Developing Countries (ICCIDC-II), 3rd -5th August, 2010. Cairo, Egypt, 2010, pp. 570–578 Retrieved 16 May, 2019, from https://www.researchgate.net/ publication/228962800_BIM_and_RFID_integration_A_pilot_study. M.Y. Cheng, N. Chang, Radio frequency identification (RFID) integrated with building information model (BIM) for open-building life cycle information management, In the Proceedings of the 28th International Symposium on Automation and Robotics in Construction, 29th June -2nd July, 2011. Seoul, Korea, 2011, pp. 485–490, , https://doi.org/10.22260/ISARC2011/0088. J. Chen, T. Bulbul, J.E. Taylor, G. Olgun, A case study of embedding real-time infrastructure sensor data to BIM, In the Proceeding of Construction Research Congress 2014: Construction in a Global Network, 19th -21st May, 2014, 2014, pp. 269–278, , https://doi.org/10.1061/9780784413517.028 Georgia, United States of America. T.H. Chuang, B.C. Lee, I.C. Wu, Applying cloud computing technology to BIM visualization and manipulation, In the Proceeding of the 28th International Symposium on Automation and Robotics in Construction, 29th June -2nd July, 2011, 2011, pp. 144–149, , https://doi.org/10.22260/ISARC2011/0023 Seoul, Korea. M.S. El-Abbasy, A. Senouci, T. Zayed, F. Mirahadi, L. Parvizsedghy, Artificial neural network models for predicting condition of offshore oil and gas pipelines, Autom. Constr. 45 (2014) 50–65, https://doi.org/10.1016/j.autcon.2014.05.003. M.J. Jiménez-Come, I.J. Turias, J.J. Ruiz-Aguilar, A two-stage model based on artificial neural networks to determine pitting corrosion status of 316L stainless steel, Corros. Rev. 34 (1–2) (2016) 113–125, https://doi.org/10.1515/corrrev-20150048. C. De Waard, D.E. Williams, Prediction of carbonic acid corrosion in natural gas pipelines, In the Proceeding of the 1st International Conference on the Internal and External Protection of Pipes, September, 1975, 28(340) University of Durham, United Kingdom, 1975, pp. 24–26, , https://doi.org/10.1108/issn.0144-0772. C. De Waard, U. Lotz, A. Dugstad, Influence of liquid flow velocity on CO2 corrosion: a semi-empirical model, In the proceeding of National Association of Corrosion Engineers International Annual Conference, 26th -31st March, 1995. Orlando, United States of America, No. 128, 1995 Retrieved 16 May, 2019, from https://www.osti.gov/biblio/106125-influence-liquid-flow-velocity-co-subcorrosion-semi-empirical-model. J.R. Shadley, S.A. Shirazi, E. Dayalan, E.F. Rybicki, Prediction of erosion-corrosion penetration rate in a carbon dioxide environment with sand, Corrosion 54 (12) (1998) 972–978, https://doi.org/10.5006/1.3284819. S. Nesic, J. Postlethwaite, S. Olsen, An electrochemical model for prediction of corrosion of mild steel in aqueous carbon dioxide solutions, Corrosion 52 (4) (1996) 280–294, https://doi.org/10.5006/1.3293640. Z. Riaz, M. Arslan, A.K. Kiani, S. Azhar, CoSMoS: a BIM and wireless sensor based integrated solution for worker safety in confined spaces, Autom. Constr. 45 (2014) 96–106, https://doi.org/10.1016/j.autcon.2014.05.010. G. Lee, J. Cho, S. Ham, T. Lee, G. Lee, S.H. Yun, H.J. Yang, A BIM-and sensor-based tower crane navigation system for blind lifts, Autom. Constr. 26 (2012) 1–10, https://doi.org/10.1016/j.autcon.2012.05.002. K.M. Kensek, Integration of environmental sensors with BIM: case studies using Arduino, dynamo, and the Revit API, Inf. Constr. 66 (536) (2014), https://doi.org/ 10.3989/ic.13.151. I. Motawa, A. Almarshad, A knowledge-based BIM system for building maintenance, Autom. Constr. 29 (2013) 173–182, https://doi.org/10.1016/j.autcon. 2012.09.008. B. Cottis, M.J. Graham, R. Lindsay, S.B. Lyon, T. Richardson, D. Scantlebury, H. Stott, Shreir's Corrosion, Elsevier, Oxford, 2010, pp. 1270–1298 https://doi.org/ 10.978.0444/r527875. Y.H. Tsai, S.H. Hsieh, S.C. Kang, A BIM-enabled approach for construction inspection, In the proceedings of the 2014 Conference on Computing in Civil and Building Engineering, 23th -25th June, 2014, 2014, pp. 721–728, , https://doi.org/10.1061/ 9780784413616.090 Florida, United States of America. Y.H. Tsai, S.H. Hsieh, Process modeling of a BIM-enabled construction inspection approach with BPMN, In the Proceeding of 2016 International Conference on Innovative Production and Construction, 3rd -5th October, 2016, 2016, pp. 80–83 Perth, Australia. Retrieved 16 May, 2019, from https://www.researchgate.net/ publication/333186835_Process_Modeling_of_a_BIM-enabled_Construction_ Inspection_Approach_with_BPMN. Object Management Group, Business Process Model and Notation (BPMN), Retrieved 16 May, 2019, from, 2011. https://www.omg.org/spec/BPMN/2.0.2/ PDF. E. McCafferty, Validation of corrosion rates measured by the Tafel extrapolation method, Corros. Sci. 47 (12) (2005) 3202–3215, https://doi.org/10.1016/j.corsci. 2005.05.046.