Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

TUGAS PERBAIKAN

Fakultas Teknologi Industri


Jurusan : Teknik Mesin, Semester VIII
Mata Kuliah : Analisa Kegagalan Material
Hari, Tanggal : Senin, 5 Agustus 2019
Nama Dosen : Antonius Yunianto, ST, MT
Dikumpulkan : Selasa, 6 Agustus 2019, Pukul 11.59 WITA
By Email antonius.yunianto@gmail.com

---------------------------------------------------------------------------------------------------------------------

1. Di bidang korosi, ada istilah yang disebut dengan Risk Based Inspection (RBI) berdasarkan
API580/581
a. Jelaskan konsep dari RBI
b. Jelaskan ruang lingkup dari RBI
c. Jelaskan kegunaannya RBI
d. berilah contoh di lapangan berikut resikonya!

Reference : API 580 & 581 & DNV GL RP-G101

2. Summary the paper :


a. Risk Based Approach to Integrated Asset Corrosion Management in the Oil and Gas
Industry
b. Artificial Neural Network Model for Risk-Based Inspection Screening Assessment of
Oil and Gas Production System
Proceedings of the Twenty-eighth (2018) International Ocean and Polar Engineering Conference www.isope.org
Sapporo, Japan, June 10-15, 2018
Copyright © 2018 by the International Society of Offshore and Polar Engineers (ISOPE)
ISBN 978-1-880653-87-6; ISSN 1098-6189

Artificial Neural Network Model for Risk-Based Inspection Screening Assessment of Oil and Gas
Production System

Rachman, Andika; Ratnayake, R.M. Chandima


Department of Mechanical and Structural Engineering and Material Science, University of Stavanger
Stavanger, Rogaland, Norway

ABSTRACT Zheng, and He, 2011). RBI allows organizations to focus their
inspection resources on equipment with a high risk of failure, thus
Screening assessment is a part of risk-based inspection (RBI) preventing over-inspection of lower-risk equipment and under-
methodology, which enables the filtering of equipment that makes a inspection of higher-risk equipment (Chang, Chang, Shu, and Lin,
significant contribution to the overall risk of the system. The screening 2005).
assessment directs detailed assessment to focus on higher-risk items;
thus, resources (e.g., time and labor) can be allocated more effectively
and efficiently. However, performing RBI screening assessment
requires considerable time and resources, and its results are prone to
high variability, due to its qualitative nature. To mitigate the
aforementioned challenges and to make the knowledge work involved
lean (i.e. to minimize waste), an artificial neural network (ANN) model
is suggested for performing the RBI screening assessment. The
development of the ANN model is demonstrated by using a dataset
containing the data and information from an RBI assessment conducted
for onshore and offshore hydrocarbon production and process systems.
It is revealed that the suggested model is capable of achieving
respectable performance with 90.65% accuracy, 82.76% precision, and
76.59% recall.

KEY WORDS: Risk-based inspection; artificial neural network;


machine learning; oil and gas; maintenance; integrity management; lean

INTRODUCTION Fig. 1. Generic RBI planning process (adapted from (American


Petroleum Institute, 2016b))
The advancement of inspection and maintenance strategies from a
conventional calendar-based approach to more sophisticated plans, such A generic RBI planning process is presented in Fig. 1. For detailed
as condition-monitoring, reliability-centered maintenance, and risk- information and an explanation regarding RBI methodology, refer to
based approach, has been driven by the increase in the number, size, the API RP 580 (2016b) and DNV GL RP-G101 (2017). Normally,
and complexity of physical assets (Khan and Haddara, 2003). These there are two types of risk assessment in an RBI planning process: (1)
advanced strategies help organizations to maximize the availability and screening assessment and (2) detailed assessment. Screening and
reliability of the equipment, while still ensuring that operations are safe detailed assessment differ in terms of the level of detail in estimating
for the personnel, society, and environment. One of these strategies is the equipment’s risk level. Screening assessment typically utilizes
risk-based inspection (RBI), which shifts the inspection planning qualitative analysis, which has a simpler procedure, fewer data
paradigm from time-based inspection planning to more proactive requirements, and faster analysis time than the quantitative approach.
inspection planning, by using a risk-based approach (Tan, Li, Wu, Meanwhile, detailed assessment typically uses a quantitative approach

270
that requires more data and has a more rigorous computation and the intelligent system can provide instantaneous RBI screening
analysis procedure than qualitative analysis. Computer software is assessment results, which is the basis for reducing the time and effort
commonly used to generate outputs related to detailed assessment. involved in carrying out RBI analysis. The schematic of the suggested
approach is displayed in Fig. 2.
In most processing systems, a significant portion of the system’s total
risk is accumulated in a relatively small percentage of the items of The paper is structured as follows. The second section provides an
equipment (American Petroleum Institute, 2016b). Therefore, a overview of the ANN concept. The third section elaborates the
screening assessment is conducted prior to the detailed assessment, to methodology to develop the intelligent system, while the fourth section
identify equipment that makes a significant contribution to the system’s summarizes and analyzes the performance of the intelligent system.
overall risk (DNV GL, 2017). This allows the detailed assessment to Finally, the fifth section concludes the paper.
focus on higher-risk items; thus, resources (e.g., time and labor) can be
allocated more effectively and efficiently (American Petroleum ARTIFICIAL NEURAL NETWORK
Institute, 2016b).
ANN is a machine learning algorithm that was originally developed to
However, there are two fundamental challenges in performing imitate the neural systems of biological organisms (Karunanithi,
screening assessment. First, because of its qualitative nature, screening Grenney, Whitley, and Bovee, 1994; Zhang, Eddy Patuwo, and Y. Hu,
assessment depends more on the knowledge, expertise, and judgement 1998). Various models of neural network have been proposed. Among
of the engineering personnel, which makes it prone to human biases them, multi-layer perceptrons (MLP) and recurrent neural networks are
and errors (Drucker, 1999). Moreover, qualitative assessment, based on two widely-known neural network models that are normally used for
subjective judgement and limited data and information, causes forecasting tools (Karunanithi et al., 1994). In this paper, we focus only
considerable variability in outputs (Geary, 2002). Additionally, human on the utilization of MLP.
error is inevitable, due to the limitation of the human mind in
processing and storing information (Turban, Aronson, and Liang, An MLP typically comprises several distinctive layers, with each layer
2005). The variability of outputs and errors can undermine the integrity consisting of neurons or nodes. The first layer is called the input layer,
of the physical assets, by causing incongruity between the equipment where the input data are received. The last layer is the output layer,
risk level and the inspection quality and characteristics (e.g., inspection where the results are generated. The input layer and output layer are
frequency, method, and level of detail) (Geary, 2002). Second, separated by one or more intermediate layers, where the data are
considerable time and resources are needed to perform qualitative processed (Zhang et al., 1998). These intermediate layers are also
analysis (Ratnayake, 2014). Organizations may need to gather called hidden layers because their outputs are not explicitly shown
personnel from different departments (e.g., process, safety, asset (Goodfellow, Bengio, and Courville, 2016). The nodes in the lower
integrity, inspection, etc.) to decide whether certain equipment needs to layer are connected to the nodes of the adjacent higher layer by acyclic
be included in the detailed assessment (Seneviratne and Ratnayake, arcs (Zhang et al., 1998). The direction of the arcs implies the direction
2015; Stadnicka, Antosz, and Ratnayake, 2014). It is possible to avoid of information, which flows forward from the input layer to the output
the screening assessment entirely and include all equipment in the RBI layer through hidden layers to generate output (i.e., forward-
detailed assessment, but this option can be costly, considering the propagation). This is why the MLP is also called a feedforward neural
laborious and rigorous nature of the detailed assessment process. network (Gardner and Dorling, 1998). Fig. 3 shows an example of an
MLP with two hidden layers.

Fig. 2. A schematic of the RBI screening assessment based on machine


learning approach

This paper attempts to address the aforementioned challenges by


developing an intelligent system, based on the artificial neural network
(ANN) concept. ANN is one of the machine learning techniques that
enables transfer and reuse of past RBI detailed assessments, by Fig. 3. Illustration of an MLP
mechanizing knowledge acquisition and converting existing data and
information into actionable knowledge (Ford, 1989; Langley and The predictive power of an MLP depends significantly on the network
Simon, 1995). In other words, the intelligent system enables past RBI architecture, i.e., the number of input nodes, the number of output
detailed assessments’ data and information to be utilized as the nodes, and the number of hidden layers and hidden nodes. The design
references for conducting the new RBI screening assessment of the network architecture depends on the problem, i.e., there is no
(Hoppmann, Rebentisch, Dombrowski, and Zahn, 2011). This approach defined formula that can guarantee to find the optimal design for all
is expected to reduce the inherent output variability in qualitative types of problems (Zhang et al., 1998). The number of input nodes
analysis, as the machine learning algorithm is less susceptible to human corresponds to the number of input features (i.e., independent variables)
errors and biases (Frey and Osborne, 2017) than human workers that are taken into consideration. For classification problems, the
(Acemoglu and Autor, 2011; Frey and Osborne, 2017). Furthermore, number of output nodes is dependent on the number of classes or

271
category memberships. One hidden layer is considered adequate for layer after the data have flowed forward through the network
many datasets (Dreiseitl and Ohno-Machado, 2002). Two hidden layers (Karunanithi et al., 1994; Singh et al., 2009). The distribution of error
may be necessary if a single hidden layer requires an extensive number back through the nodes allows the adjustment of the weights of the
of hidden nodes, which is not preferred, due to the longer training time connections between nodes, in order to reduce the error. Several
and poorer generalization ability (Barron, 1994). More hidden layers iterations of forward-propagation and back-propagation are required to
may be necessary if the datasets are highly complex. Determining the find the optimum weights that minimize the output error (Singh et al.,
number of hidden nodes is challenging. A network with a small number 2009). Learning rate influences the back-propagation algorithm in
of hidden nodes may be able to prevent overfitting and have better updating the weights of the connections between nodes. A low learning
generalization ability, but it may have insufficient degrees of freedom rate will slow down the rate of convergence in the MLP models
to learn from the data (Zhang et al., 1998). On the contrary, a network because numerous iterations are needed to find the minimum error
with a high number of hidden nodes requires a longer training time, and (Singh et al., 2009). A higher learning rate can accelerate the training
the risk of overfitting is greater (Karunanithi et al., 1994). process, but too high a learning rate may cause undue network
oscillations in the weight space (Zhang et al., 1998).
Each connection between nodes has a weight that, together with the
transfer (activation) function residing in the nodes, modifies the signal METHODOLOGY
passing through the nodes (Gardner and Dorling, 1998). The activation
function is typically characterized as non-linear, which allows the MLP RBI assessment can be considered as a ‘data classification task’, where
to model complex non-linear relationships between the considered the equipment to be assessed constitutes the instances that need to be
variables (Singh, Basant, Malik, and Jain, 2009). It is possible to define classified into a particular class. For instance, in RBI detailed
the activation function as linear, but then the MLP can only infer linear assessment, there are normally four distinct classes that represent the
relationships between variables. The signals arriving at a particular equipment’s level of risk (i.e., low, medium, medium-high, and high)
node are multiplied by the weights of the corresponding arcs. The (American Petroleum Institute, 2016a). A risk matrix is commonly used
results of these multiplications are then summed and passed through the to visualize the distribution of risk. An example of a risk matrix for
activation function, to generate output for that node (Agatonovic- detailed assessment, with four classes of risk, is displayed in Fig. 4a.
Kustrin and Beresford, 2000). This process occurs in each node in an
MLP. While it is possible for a single node to process all incoming In this paper, RBI screening assessment based on DNV GL (2017) is
information, the power of the MLP comes from the numerous used as the reference. For RBI screening assessment, three classes can
interconnected nodes in the network (Agatonovic-Kustrin and be used to represent the equipment’s risk level (DNV GL, 2017). The
Beresford, 2000). risk matrix for the screening assessment is the simplification of the risk
matrix for detailed assessment (see Fig. 4b). In screening analysis, only
The learning process in an MLP encompasses the adjustment of the equipment that belongs to 1A low-risk category can skip the detailed
weights of nodes’ interconnections, to minimize the output error assessment. Thus, the number of classes for the screening assessment
(Karunanithi et al., 1994). The frequently used learning algorithm for can be reduced to two: low risk and medium/high risk. Therefore, the
an MLP is the back-propagation algorithm. This algorithm is a gradient screening assessment can be treated as a binary classification task.
descent method that computes the error contribution of nodes in each

Fig. 4. An example of (a) a risk matrix for RBI detailed assessment (American Petroleum Institute, 2016a) and (b) a simplified risk matrix for RBI
screening assessment (DNV GL, 2017)

272
Fig. 5. The workflow for developing an RBI screening assessment classifier

A classifier based on an MLP is constructed to support the RBI and informative features that can yield adequate model performance.
screening assessment. The classifier task is to categorize equipment in As the domain knowledge of RBI assessment is well established, expert
either the low-risk or the medium/high-risk class. The workflow to knowledge and engineering standards and codes (e.g., API RP 580
build the classifier is shown in Fig. 5. Standard MacBook Pro (16GB (2016b) and API RP 581 (2016a)) are used as the main method and
2133 MHz LPDDR3 memory, 2.8 GHz Intel Core i7 processor, 256 references for selecting features to be incorporated in the model
GB of flash storage, and Mac OS X as operating system) with Python (Müller and Guido, 2016). The list of selected input features is shown
3.6 (complete with the necessary packages such as NumPy, Pandas, and in Table 1.
Scikit-learn (Pedregosa et al., 2011)) and Jupyter Notebook are used.
Each step is discussed individually in the following subsections. Data Pre-Processing

Data Collection and Feature Selection Data pre-processing is required to convert raw data into machine-
readable data. The following data pre-processing steps are performed:
The dataset used in this paper was obtained from an RBI assessment • Data cleaning and integration: Minor data cleaning and
project conducted for three onshore and four offshore production units. integration is required, as the dataset is sourced from seven
The scope of the project was to perform an RBI detailed assessment for separate spreadsheets.
these units, comprising over 200 pressure vessels and 3000 piping tags, • Data generalization: Generalization is carried out to avoid having
based on API 581 methodology. In total, 1581 instances are contained too many unique values in a particular feature. For instance, the
in the dataset. The number of instances in the dataset is fewer than the material of construction feature has more than 10 unique values
original number of pressure vessels and piping tags because each that are transformed into only four corresponding values. The
instance may represent a group of piping and pressure vessels with input features that undergo generalization are fluid containment,
similar characteristics (e.g., similar operating pressure/temperature, material of construction, and insulation type.
fluid containment, dimensions, etc.). Henceforward, the term ‘instance’ • Data conversion: An MLP is not suitable for processing features
is used to refer to the piping and pressure vessels assessed in the with categorical values (Brouwer, 2002). Therefore, input features
dataset. with categorical values (e.g., material of construction, equipment
type, post-weld heat treatment, insulation type, and fluid
To decide the class membership (i.e., the risk level) of certain containment) are converted into one or more new features that
instances, a set of input features (i.e., independent variables) is used. In possess binary numbers, to enable the MLP to process them. This
the acquired dataset, more than 30 input features were used to process is also known as one hot encoding. The features created
determine the risk level of each instance. This number is normal, by the one hot encoding process are called dummy variables
considering the rigorousness of RBI detailed assessment computation (Müller and Guido, 2016). A total of 32 input features are created
and data requirements. For the purpose of building a machine learning after one hot encoding process.
model for supporting the RBI screening assessment, not all the features • Data standardization: It is important to standardize all the input
are included, for the following reasons: features’ values before passing them into the MLP, to enable the
• The number of input features should be kept low, to preserve the optimization algorithm converge faster (Dawson and Wilby,
simplistic characteristic of the screening analysis. Having too many 1998). There are no fixed rules regarding standardizing input data
input features potentially increases the time and resources required in the machine learning subject. In this paper, standardization is
to collect data. performed such that the values of all input features will have the
• High-dimensional feature space is found to reduce the performance properties of standard normal distribution (i.e., the values are
of machine learning models (Verleysen, Francois, Simon, and centered around 0 with a standard deviation of 1).
Wertz, 2003). This phenomenon is known as the curse of
dimensionality. Training and Test Set Definition
Table 1. Selected input features for developing the model Training, in machine learning, can be understood as the process of
finding and extracting patterns in the data, to generate rules for
Input features generalizing the data (Langley and Simon, 1995). It can also be defined
• Material of construction • Insulation type as the process of learning the relationship between input features and
output features, to make a predictive model based on the inferred
• Equipment type • Operating pressure relationship (Kelleher, Mac Namee, and D'Arcy, 2015). Thus, it is
• Equipment diameter • Operation temperature necessary for the training set to have known output and input features
• Commissioning date • Fluid containment to be fed to the learning algorithm. Meanwhile, the test set is used to
evaluate the trained model generalizability to unseen data. From the
• Post-weld heat treatment • Corrosion allowance total of 1581 instances from the original dataset, 75% of them (or 1185
instances) are used for training purposes, while the remaining 25% (or
Having too few input features is also not desirable, as it will undermine 396 instances) are utilized for testing only.
the generalization capability of the model (Kavzoglu and Mather,
2002). Therefore, feature selection is performed to identify important

273
Model Optimization precision by evaluating the ability of the classifier to discover instances
with low risk. Recall is able to measure the ‘missed opportunity’ to
The MLP model is optimized by tuning its corresponding classify equipment into the low-risk class. Meanwhile, accuracy
hyperparameters, i.e., the parameters that cannot be derived from provides observation regarding the overall performance of the
training and thus shall be set prior to the learning process. The MLP classifier.
has numerous hyperparameters that can be set to yield better
performance. In this paper, randomized search cross-validation (RSCV)
is used as the method for hyperparameter optimization. RSCV
iteratively picks a combination of values from the available space of
hyperparameter values, performs cross-validation for each of the
randomly picked combination of hyperparameter values, and decides
the combination that generates the best model performance. The
number of iterations is set to 200, while the number of folds for cross-
validation is set to 5. The training set is used for the purpose of cross-
validation.
Fig. 6. (a) Confusion matrix for RBI screening analysis and (b) its
corresponding evaluation metrics
In this paper, the combination of hyperparameter values that gives the
greatest precision is retained and used. The reasoning behind selecting
precision as the metric for choosing the best combination of RESULTS AND DISCUSSION
hyperparameter values is given in the subsequent section (performance
evaluation section). The following hyperparameters are assessed: The result of model optimization shows that the model with 14 hidden
nodes for the first hidden layer and 9 hidden nodes for the second
• The number of hidden layers and hidden nodes: The number of
hidden layer, ReLU activation function for all nodes, Adam
hidden layers is fixed to 2. Permutations of the value range [1, 15]
optimization algorithm, and adaptive learning rate decay gives the best
are used as the set of the number of hidden nodes to be assessed.
performance. The Adam optimization algorithm is known to perform
For instance, with 2 as the number of hidden layers, the set of the
number of hidden nodes for the first and second hidden layer is [(1, well in large datasets (Pedregosa et al., 2011). The complete set of
hyperparameters and the network architecture for the model are shown
1), (1, 2), (1, 3), …, (15, 15)].
in Fig. 7.
• The type of activation function: Logistic function, hyperbolic
tangent function, and rectified linear unit (ReLU) function are
The performance of the classifier and its corresponding confusion
selected as the set of activation functions to be evaluated. These
matrix examined by the test set is presented in Fig. 8. The classifier can
functions are deemed to be suitable for solving complex problems
achieve 90.65% accuracy, 82.76% precision, and 76.59% recall. This
that require non-linear computations (Buduma and Locascio, 2017).
shows that, even with a small number of input features and simple
Sigmoid function (e.g., logistic and hyperbolic tangent function) is
architecture, the MLP model is promising as a classifier for RBI
deemed to suit binary classification tasks (Zhang et al., 1998).
screening assessment.
ReLU has been used for various tasks, especially in computer
vision (Buduma and Locascio, 2017).
• Optimization algorithm: Stochastic gradient descent, adaptive
moment estimation (Adam), and Limited-memory Broyden–
Fletcher–Goldfarb–Shanno (L-BFGS) are the set of optimization
algorithms that are assessed. These algorithms are used to perform
learning (i.e., adjustment of the weights of nodes’
interconnections), by utilizing the gradient of loss function
calculated by the back-propagation algorithm (Goodfellow et al.,
2016).
• Learning rate: The learning rate should be set below 0.035, to
prevent unstable learning (Hagan, Demuth, and Beale, 1996; Singh Fig. 7. The selected hyperparameters and network architecture for the
et al., 2009). Thus, the initial learning rate is set to 0.001. The MLP model
learning rate is set to either constant or decayed. Learning rate
decay can help the model to achieve minimum error (Goodfellow et
al., 2016). There are two methods of learning rate decay: adaptive
and inv-scaling.

Performance Evaluation

Accuracy, precision, and recall are set as the metrics for evaluating the
performance of the MLP model. The definition of these metrics is
shown in Fig. 6. In this paper, the low-risk class can be referred to as
the positive class, while the medium/high-risk class can be denoted as
the negative class.
Fig. 8. (a) Confusion matrix and (b) performance of the classifier
examined by the dataset
Precision is the primary metric, due to the high cost of making a false
positive decision (i.e., predicting a particular instance to have low risk,
In reality, the classifier outputs are the probability estimates of a certain
while its actual risk is medium/high). Recall is used to complement
instance being classified into a particular membership class. The class

274
membership is decided by setting a decision boundary, which is the eliminate human biases and errors entirely, but it intuitively reduces the
cutting point between the positive and negative classes. Let be the i- subjective judgement involved in qualitative analysis, through the
th instance in the dataset, the function that generates the provision of machine intelligence insights.
probability estimate for being in the positive class, the class
prediction for , and the decision boundary, thus: Future works should be focused on improving the performance of the
ANN model by investigating broader range of hyperparameters. For
instance, in this paper, the number of hidden layers is fixed to 2 and
(1) more hidden layers should be investigated.

CONCLUSION
with = 1 representing the positive class and = -1 representing the
negative class. The value of the decision boundary that yields the This paper aims to develop an RBI screening assessment tool based on
performance shown in Fig. 8 is 0.5. the ANN concept, in order to tackle the resource intensiveness and
output variability inherent in the RBI screening assessment process. A
Considering the high cost of misclassifying the medium/high-risk dataset containing the data and information from an RBI assessment for
instance into the low-risk class, it is imperative to maximize the onshore and offshore hydrocarbon production systems is used to
precision of the classifier. It is possible to do so by adjusting the develop the ANN model. An ANN model with two hidden layers (14
decision boundary but with a trade-off with the recall score of the hidden nodes for the first layer and 9 hidden nodes for the second
classifier. To examine the trade-off between precision and recall, a layer), ReLU activation function for all nodes, Adam optimization
precision-recall (PR) curve is used to illustrate precision and recall algorithm, and adaptive learning rate decay (with 0.001 initial learning
scores at different decision boundaries (see Fig. 9). A subtle precision rate) is constructed and trained. Of the dataset, 75% is used for training,
improvement to approximately 85% is achieved by adjusting the while the remaining 25% is used for testing (i.e., evaluating the
decision boundary to a little over 0.55, with only a marginal reduction performance of the model). The result shows that the constructed ANN
of recall to roughly 75% (marked as a blue circle in Fig. 9). A more model can achieve respectable performance with 90.65% accuracy,
extreme precision score improvement can be made by shifting the 82.76% precision, and 76.59% recall. Considering the high cost of
boundary beyond 0.85, to make a precision score of over 90% but at the misclassification, full dependency and automation of RBI screening
expense of reducing recall to a value below 42% (marked as an orange assessment, based on the ANN model, is not feasible. The involvement
circle in Fig. 9). of human intelligence is still required, in order to refine the insights
provided by machine intelligence, through the incorporation of the tacit
knowledge of human workers.

ACKNOWLEDGEMENTS

This work has been carried out as part of a Ph.D. research project at the
University of Stavanger, funded by the Norwegian Ministry of
Education and Research. This work is also aligned with the ILA-Lean
project: Innovative Learning Approaches for Implementation of Lean
Thinking to Enhance Office and Knowledge Work Productivity, project
number 2016-1-PL01-KA203-026293. This project has been co-funded
with support from the European Commission.

REFERENCES

Acemoglu, D. and Autor, D. (2011). “Skills, tasks and technologies:


implications for employment and earnings,” Handbook of Labor
Economics, 4, 1043-1171.
Agatonovic-Kustrin, S. and Beresford, R. (2000). “Basic concepts of
artificial neural network (ANN) modeling and its application in
Fig. 9. Precision-recall curve of the classifier pharmaceutical research,” Journal of Pharmaceutical and
Biomedical Analysis, 22(5), 717-727.
The constructed MLP model provides instantaneous RBI screening American Petroleum Institute. (2016a). Risk-Based Inspection
assessment, which can be the foundation for reducing the time and Methodology: API Recommended Practice 581 (3rd ed.), American
effort involved in carrying out RBI analysis. However, despite its Petroleum Institute.
satisfactory performance and results, full dependency and automation American Petroleum Institute. (2016b). Risk-Based Inspection: API
of screening analysis, based on the model, may still not be practical, Recommended Practice 580 (3rd ed.), American Petroleum
considering the high risk and cost of making a misclassification. Hence, Institute.
it is essential to continue involving human intelligence as a means to Barron, A. R. (1994). “[Neural networks: a review from statistical
reduce the risk of misclassification, by refining the machine perspective]: Comment,” Statistical Science, 9(1), 33-35.
intelligence outputs through the incorporation of the experience, Brouwer, R. K. (2002). “A feed-forward network for input that is both
expertise, and intuition of human workers (i.e., tacit knowledge) categorical and quantitative,” Neural Networks, 15(7), 881-890.
(Sharma, Mithas, and Kankanhalli, 2014). Tacit knowledge is Buduma, N. and Locascio, N. (2017). Fundamentals of Deep Learning:
imperative in making knowledge productive, although its codification Designing Next-Generation Machine Intelligence Algorithms,
into machine algorithms still remains a challenge (Hendriks and Vriens, O'Reilly Media.
1999). Bringing together human and machine intelligence will not Chang, M.-K., Chang, R.-R., Shu, C.-M. and Lin, K.-N. (2005).

275
“Application of risk based inspection in refinery and processing Khan, F. I. and Haddara, M. M. (2003). “Risk-based maintenance
piping,” Journal of Loss Prevention in the Process Industries, (RBM): a quantitative approach for maintenance/inspection
18(4), 397-402. scheduling and planning,” Journal of Loss Prevention in the
Dawson, C. W. and Wilby, R. (1998). “An artificial neural network Process Industries, 16(6), 561-573.
approach to rainfall-runoff modelling,” Hydrological Sciences Langley, P. and Simon, H. A. (1995). “Applications of machine
Journal, 43(1), 47-66. learning and rule induction,” Communications of the ACM, 38(11),
DNV GL. (2017). DNV GL-RP-G101: Risk Based Inspection of 54-64.
Offshore Topsides Static Mechanical Equipment. Müller, A. C. and Guido, S. (2016). Introduction to Machine Learning
Dreiseitl, S. and Ohno-Machado, L. (2002). “Logistic regression and with Python: A Guide for Data Scientists, O’Reilly Media.
artificial neural network classification models: a methodology Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B.,
review,” Journal of Biomedical Informatics, 35(5), 352-359. Grisel, O., . . . Dubourg, V. (2011). “Scikit-learn: machine learning
Drucker, P. (1999). “Knowledge-worker productivity: the biggest in python,” Journal of Machine Learning Research, 12, 2825-2830.
challenge,” California Management Review, 41(2), 79-94. Ratnayake, R. M. C. (2014). “Application of a fuzzy inference system
Ford, N. (1989). “From information-to knowledge-management: the for functional failure risk rank estimation: RBM of rotating
role of rule induction and neural net machine learning techniques in equipment and instrumentation,” Journal of Loss Prevention in the
knowledge generation,” Journal of Information Science, 15(4-5), Process Industries, 29, 216-224.
299-304. Seneviratne, A. and Ratnayake, R. M. C. (2015). “Classification of
Frey, C. B. and Osborne, M. A. (2017). “The future of employment: static mechanical equipment using a fuzzy inference system: a case
how susceptible are jobs to computerisation?,” Technological study from an offshore installation,” International Journal of
Forecasting and Social Change, 114, 254-280. Performability Engineering, 11(1), 53-60.
Gardner, M. W. and Dorling, S. R. (1998). “Artificial neural networks Sharma, R., Mithas, S. and Kankanhalli, A. (2014). “Transforming
(the multilayer perceptron)—a review of applications in the decision-making processes: a research agenda for understanding the
atmospheric sciences,” Atmospheric Environment, 32(14), 2627- impact of business analytics on organisations,” European Journal
2636. of Information Systems, 23(4), 433-441.
Geary, W. (2002). Risk Based Inspection: A Case Study Evaluation of Singh, K. P., Basant, A., Malik, A. and Jain, G. (2009). “Artificial
Offshore Process Plant, Health and Safety Laboratory. neural network modeling of the river water quality - a case study,”
Goodfellow, I., Bengio, Y. and Courville, A. (2016). Deep Learning, Ecological Modelling, 220(6), 888-895.
MIT Press. Stadnicka, D., Antosz, K. and Ratnayake, R. M. C. (2014).
Hagan, M. T., Demuth, H. B. and Beale, M. H. (1996). Neural Network “Development of an empirical formula for machine classification:
Design, PWS Publishing. prioritization of maintenance tasks,” Safety Science, 63, 34-41.
Hendriks, P. H. and Vriens, D. J. (1999). “Knowledge-based systems Tan, Z., Li, J., Wu, Z., Zheng, J. and He, W. (2011). “An evaluation of
and knowledge management: friends or foes?,” Information & maintenance strategy using risk based inspection,” Safety Science,
Management, 35(2), 113-125. 49(6), 852-860.
Hoppmann, J., Rebentisch, E., Dombrowski, U. and Zahn, T. (2011). Turban, E., Aronson, J. E. and Liang, T.-P. (2005). Decision Support
“A framework for organizing lean product development,” Systems and Intelligent Systems (7th ed.), Prentice-Hall.
Engineering Management Journal, 23(1), 3-15. Verleysen, M., Francois, D., Simon, G. and Wertz, V. (2003). “On the
Karunanithi, N., Grenney, W. J., Whitley, D. and Bovee, K. (1994). effects of dimensionality on data analysis with neural networks”,
“Neural networks for River Flow prediction,” Journal of Proc 7th International Work-Conference on Artificial and Natural
Computing in Civil Engineering, 8(2), 201-220. Neural Networks, Berlin, Springer, 105-112.
Kavzoglu, T. and Mather, P. M. (2002). “The role of feature selection Zhang, G., Eddy Patuwo, B. and Y. Hu, M. (1998). “Forecasting with
in artificial neural network applications,” International Journal of artificial neural networks:: The state of the art,” International
Remote Sensing, 23(15), 2919-2937. Journal of Forecasting, 14(1), 35-62.
Kelleher, J. D., Mac Namee, B. and D'Arcy, A. (2015). Fundamentals
of Machine Learning for Predictive Data Analytics: Algorithms,
Worked Examples, and Case Studies, MIT Press.

276
Paper No.

9767

Risk Based Approach to Integrated Asset Corrosion Management in the Oil and Gas Industry

Joseph Akanni Oladipo Alonge


Shell Nigeria Exploration and Production Shell Petroleum Development Company
Company Port Harcourt
Port Harcourt Nigeria
Nigeria

Edmund Oseahon
Shell Petroleum Development Company
Port Harcourt
Nigeria

ABSTRACT

Oil and gas facilities are subject to several degradation mechanisms depending on the process fluids,
operating conditions and environmental factors. There is need to address these degradation threats
and mitigate their undesired consequences. This would entail developing a Corrosion Management
System (CMS) for all equipment; the application of which would reduce the corrosion risks associated
with operating the facility.

The Corrosion Management System has key elements; some of which are the corrosion risk
assessment, planning, implementation and analysis. Implementation covers both corrosion mitigation
activities that provide barriers to corrosion and also inspection which verifies that the barriers are
effective. The selection of inspection routines can be fixed time-based or risk based. The risk based
inspection approach is an optimal and fit for purpose inspection routine. This would require a detailed
analysis of the risk of operating the facilities based on economic, health and safety consideration.

An integrated corrosion monitoring process can be achieved with proper implementation of the risk
based inspection approach, where all likely degradation threats have been identified, susceptibility and
impact assessed. With a defined risk ranking systems the choice of inspection methods can be ranked
to determine business, economic benefit and ultimately assure the technical integrity on the facility.

This paper defines the critical elements for consideration in risk management while also detailing a
systematic approach of risk based inspection based on API 580. The process, benefits, possible pitfalls
and opportunity for improvements are clearly detailed based on practical application of RBI and
corrosion management of oil and gas facilities

©2017 by NACE International.


Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
1
INTRODUCTION

The management of corrosion threats to facilities can effectively be achieved through a robust
Corrosion Management System (CMS). The CMS is that part of an asset overall management system,
which is concerned with the development, implementation and review of an organization’s corrosion
policy. A corrosion management policy sets a clear direction for an organization to reduce the likelihood
of corrosion related failures, assure asset integrity and improve operational reliability.

Some of the key elements of the CMS include organizational structure and responsibilities; corrosion
risk assessment and planning; implementation and analysis and system performance review. The
corrosion risk assessment and planning are necessary to adequately manage the threat of degradation
and achieve the intended design life at minimum life cycle cost. It entails definition of corrosion systems
and loops; identifying threats and barriers; assessing inspection and corrosion monitoring frequencies
and determining corrosion management plans.

Implementation of the corrosion management plans and also analysis cover both mitigation activities
that provide barriers to corrosion and inspection and verification that the barriers are effective. These
involve implementing planned corrosion management activities, data gathering, data analysis and
identifying and following through with corrective actions.

A systems performance is measured both proactively and reactively against set criteria on a continuous
basis. The proactive measurement helps to demonstrate that barriers to corrosion are maintained while
the reactive measurement shows that the corrosion management system is effective. Based on the
measurement results, the system performance is reviewed to learn from relevant experience, apply the
lessons and achieve continuous improvement.

The Risk Based Inspection (RBI) approach helps to determine inspection and corrosion monitoring
scope from an understanding of corrosion threats, extent, locations and optimised inspection cycle for
static equipment based on susceptibility to failure, the health, safety, environmental and economic risks
of its operation and remnant life of the equipment. Thus, RBI is a process which comprises risk analysis
to develop effective/optimized inspection plan/requirements for static process equipment in a plant or
production facility by considering all credible damage mechanisms that could be expected to affect an
equipment and identification of other risk-reducing actions. Risk levels are prioritized in a systematic
manner so that an inspection programme can be planned that focuses resources on higher risk
equipment while reducing lower value inspection activities.

The RBI plan produced combined with a set of integrity operating windows for process units in a facility
help to achieve effective corrosion and asset integrity management in oil and gas facilities. Technical
integrity is maintained by executing inspections and process monitoring at the right locations using
appropriate techniques with the appropriate coverage and at the right frequencies.

RBI implementation is a proactive approach to replacing time based inspection with flexible periodic
inspection programmes. It is applicable to static equipment such as process piping, pressure vessels,
heat exchangers, above ground storage tanks and pressure relief devices, boilers and heaters. It takes
into account only degradation mechanisms related to the operation of an installation (corrosion, fatigue,
etc.). Failures resulting from excursions outside the defined Integrity Operating Window (IOW) are not
considered.

In ensuring optimal corrosion control and mitigating impact of degradation; there are other obvious
challenges and areas for consideration. These include optimal utilization of existing and new inspection
©2017 by NACE International.
Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
2
methods, production deferment from extended turnaround for unnecessary inspection, optimal
utilisation and integrity database and introducing non intrusive inspection methodology.

MANAGING RISK WITH INSPECTION


The figure below illustrates how the inherent risk in operation of facilities is reduced by conducting
inspection. The typical inspection program reduces the risk but after a while has the tendency to
increase risk from conducting unnecessary inspection tasks. The risk is introduced from the continuous
opening and closing of the equipment at fixed interval. On the other hand, the risk based inspection
approach gives a lower risk level even with the same level of inspection. This is achieved by focusing
inspection on high risk areas and reduced inspection on low risk areas. However, there is still a residual
risk that cannot be affected by RBI or the typical inspection program which is influenced by natural
disasters, human errors, inspection equipment limitations, design error and some other unintended
actions or facility configuration.

Figure 1: Managing risk with RBI (ref: API580)

RISK BASED INSPECTION WORKFLOW

The implementation of RBI starts with a clear objective and scope of equipment to be assessed within
agreed boundaries in the facility. Typically, a workshop is organized for the assessment which would
require a multidisciplinary team in attendance. The core team would consist of the following:

• Inspector;
• Asset operator;
• Maintenance engineer
• Process engineer/Technologist
• Materials & Corrosion engineer
• Mechanical static Engineer
• Representatives of local legislative body

The RBI team may be supplemented by additional expertise including production chemist, data analyst
and IT support engineer on an ad hoc basis.

©2017 by NACE International.


Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
3
It is essential to bring together all relevant information including asset integrity data in preparation for
the RBI assessment. After completion of the preparation, the workshop which is preferably off-site can
be organised. It is best practice to schedule the RBI workshop as a full week or 2 week session
depending on the asset and equipment to be assessed.

The RBI assessment workflow is represented in Figure 2 with details stated below:

• Review of asset integrity data


• Identification of corrosion loops
• Marking of the loops on PEFS
• Generating the loop descriptions (loop, process and corrosion)
• Reviewing loop description and development of integrity operating window
• Assessing criticality of failure, confidence rating and inspection interval
• Determine the maximum inspection intervals
• Recommend appropriate inspection strategy for non-age related failure
• Inspection and process monitoring planning

Asset
Integrity Corrosion
Database Loop
Analysis Assessment
Feedback &
Review

Task Remnant Life


Execution Calculation

Criticality
Inspection Assessment
Interval &
Tasks Confidence
Rating

Figure 2: The Risk Based Inspection Assessment Workflow

Asset Integrity Data

The RBI process starts with a precise definition of its boundaries and identifying all static equipment to
be included. The next step is to compile and check all relevant asset integrity data such as:

• Process data and drawings


• Engineering design data
• Description and evaluation of degradation mechanisms
• Compilation of inspection history

The RBI is a live process and must include update of the asset integrity database with results of
inspections, monitoring and process/asset changes. This will progressively eliminate uncertainties in
assessments and calculations, and result in an effective inspection programme.

Corrosion Loop Definition

All facilities are subjected to several potential degradation mechanisms, which need to be carefully
controlled to avoid unsafe situations and /or high costs. Corrosion loops are a means to describe and
©2017 by NACE International.
Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
4
control degradation mechanisms in sections of a facility. They contribute immensely to improving the
application of correct inspection techniques. A corrosion loop is a section of a facility that:

• Consists of similar materials of construction: this means that pipings are made of similar piping
classes and equipment are made of comparable materials.
• Operates under similar process conditions: this means that for each corrosion loop, the
corrosive species shall be the same throughout the loop and temperatures shall be such that no
stepwise changes in corrosion phenomena occur.
• Exposed to similar degradation/corrosion mechanisms and rates.

The RBI assessments consider all possible degradation mechanisms that the equipment is exposed to.
The mechanisms are considered as age related and non-age related. Age related degradation rate can
be measured and its trend established over time while non-age related failures tend to occur suddenly
and it is usually not possible to estimate a degradation rate or remnant life, as they are dependent more
on process conditions and generally associated with excursions outside the IOW.

Some of the commonly addressed degradation mechanisms include CO2 corrosion, Microbial Induced
Corrosion (MIC), Oxygen corrosion, under deposit corrosion, erosion corrosion, fatigue and Corrosion
Under Insulation (CUI) etc.

Criticality Assessment

Criticality is the combination of two parameters; namely, the susceptibility to failure (StF) and the
consequences of failure. Criticality is failure mode related and for each important failure mode, the
aforementioned parameters have to be determined.

The susceptibility/likelihood to failure is assessed using questionnaires based on relevant information


regarding process conditions, degradation mechanisms and also the materials, corrosion and
inspection knowledge.

The major consequence categories considered are: economic, health/safety and environment. In recent
times, there has been the need to consider reputation because equipment failure such as leak can
impact a company’s reputation. The worst case consequence determined under the different categories
defines the overall consequence class used for the assessment.

The economic consequence considers the financial effect of a failure on assets and production.
Damage to assets typically includes material and labour costs of repair and/or replacement of
equipment. Costs associated with product value such as lost revenues due to production losses are
also determined. Production loss is deferred income plus product wasted.

Health and safety consequences consider impact effects of equipment failure on humans and also
mitigating factors. Exposure of people in a hazardous zone and the possibility of averting a hazardous
event are also considered.

The environmental consequences of an incident consider the effects of liquid release on soil and water
and also the release of gas which result in pollution of the atmosphere.

©2017 by NACE International.


Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
5
Confidence Rating

This reflects the confidence that the inspector and the corrosion engineer have in the assessment of
Susceptibility to Failure and for age-related failure failures, the remnant life. It depends on their
judgement on the stability and control of degradation mechanisms; the number and quality of
inspections carried out; and whether reliable process (Integrity Operating Window) monitoring is carried
out.

Confidence rating is one of the parameters that determine inspection intervals or process monitoring
extent. The higher the confidence rating the longer the inspection interval obtained

Remnant Life Determination

The assessment of remnant life of equipment with regard to degradation mechanism and failure
characteristics is essential to the RBI process.

For non-age related (random) failure characteristics, the remnant life cannot be assessed because the
onset and rate of degradation cannot be predicted. These failure characteristics require a separate
approach based on control of internal and/or external operating conditions. A proper monitoring scheme
for the integrity operating window should be agreed for such cases.

Age related (trendable) degradation mechanisms allow a quantitative determination of the remnant life.
For degradation mechanisms that cause wall thinning, the remnant life can be assessed by considering
the remaining corrosion allowance and the corrosion rate.

Remnant life = Remaining corrosion allowance/Selected corrosion rate.

When an item’s failure mechanism is a result of simultaneous degradation, calculation of the remnant
life should be based on the worst case corrosion rate, e.g. addition of internal and external corrosion
rates. Alternatively, the remaining corrosion allowance may be split over the competing degradation
mechanisms.

Inspection and Process Monitoring Plan

For age related degradation, the criticality values together with the confidence rating are used to
determine the maximum inspection interval as a fraction of remnant life. This fraction, inspection
interval factor is established by combining the confidence rating and criticality. The remnant life is
multiplied by the interval factor to derive a maximum inspection interval.

Maximum inspection interval = Interval factor X Remnant life

A recommended maximum inspection interval defined by local legislation may only be exceeded after
approval by the statutory body.

For non-age related degradation, it is not possible to estimate a remnant life as this form of degradation
is more dependent on process conditions and it is very difficult to establish a time-dependent
degradation rate. Very high degradation rates and rapid changes are possible if the process conditions
are not properly controlled. For this reason, it should be recognised that time based inspections cannot
help in preventing the failure of these equipment. Process monitoring (Integrity Operating Window)
monitoring is most essential here.

An inspection task is defined by the following parameters:

©2017 by NACE International.


Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
6
Location: This refers to the location representative of the condition of the subject equipment or system.
Representative locations (affected area) can usually be derived from the corrosion description/analysis
for the corrosion loop. Examples are welds, nozzles and dead legs.

Technique: An inspection technique matching the credible degradation mechanisms must be selected.
For example, to monitor pitting in small-bore piping, radiography is more appropriate as opposed to the
use of ultrasonic wall thickness measurements. However, the latter technique is more suitable for
monitoring general wall loss. Similarly, appropriate process monitoring to reliably monitor the integrity
operating window must be selected.

Coverage: The inspection coverage describes the amount of affected area to be inspected to
confidently monitor the condition of an item and where possible, predict its remaining life.

Frequency: This is the frequency determined by the criticality, confidence rating and remaining life of
equipment. This frequency may vary as the confidence rating and remaining life can change over time.

Update/Review

After the inspection tasks have been executed, it is important to update the asset database. Information
like inspection and test results, condition monitoring data and actual condition revealed during the
maintenance tasks shall be fed back into the system. It is necessary to review RBI analyses regularly.
Events that typically prompt a review are plant shutdown, major integrity operating window excursion
and major plant change including changes in process conditions.

In order to realise the full benefits of an RBI study, it is necessary to link the output of RBI to the
maintenance management system or processes. This will enable the workflow to be managed,
equipment condition to be recorded and performance trends to be analysed.

RISK BASED INSPECTION TEAM RESPONSIBILITIES

The section gives a typical description of the responsibilities in the RBI process. The Table 1 below
delineates process steps and also who is responsible. “RBI” in the table stands for the complete RBI
team, “M” for the materials and corrosion engineer, “I” for the asset inspector, “PO” for the
process/operations engineer and “ME” for the maintenance engineer.

Table 1
Typical RBI Responsibility Chart
Responsible
S/N Section Contents Party
1 Asset integrity database Acceptance of data in the asset integrity I
database
2 Production loss equation Define production loss rate (dollars per hour) PO
in case of business interruption for
subsequent time intervals
3 Repair cost and time Calculate all costs and time required to ME
restore a failed equipment to its required
performance
4 Labour cost Calculate the labour costs related to a ME
specific failure
5 Corrosion loop definition Define corrosion loops on process diagrams RBI

©2017 by NACE International.


Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
7
6 Corrosion loop description Prepare descriptions of all credible corrosion O&M
mechanisms, their morphologies, areas of
susceptibility and materials of construction
7 Process description Prepare a corrosion oriented process PO
description for each loop
8 Criticality rating Answer probability and consequence RBI
questions
9 Equipment history Briefly outline historical inspection results I
and calculate actual corrosion rates
10 Confidence rating Assign confidence rating RBI
11 Remnant life Determine remnant life RBI
12 Inspection plan Inspection scope, strategy and specific I
recommendations

INTEGRITY OPERATING WINDOW (IOW)

The integrity operating window (IOW) is a key deliverable for the RBI assessment. It represents a
further limitation within the broader plant Operating Window, and refers to the recommended range of
values for relevant operating (corrosion and process) parameters to be monitored in the process fluid.
The broader plant Operating Window is established from the process and equipment design and is
safeguarded by alarms, trips, pressure relief valves and other delimiting devices, but it may not have
fully addressed all possible degradation issues.

The IOW is established from only measurable and controllable parameters and is usually determined
for a definite corrosion loop based on design, operational considerations and economic considerations
to assure asset integrity. The IOW is the basis for risk assessment and operating outside of the defined
limits will have negative impact on the integrity of the asset/equipment and inspection strategies derived
from the analysis.

The parameters used to define IOW should have one or more of the following characteristics:
• IOW parameters are either single, independently controlled process variables such as
temperature, pressure, pH, chemical injection rate, concentration, etc. or they are calculated
variables such as flow ratios, dew point conditions etc.
• The IOW can be based on a physical condition of the equipment, such as degree of vibration.

In order to achieve a controllable IOW, the parameters establishing the IOW must meet the following
criteria:

• All IOW parameters must be quantifiable and the measurement of each parameter must
become an established procedure
• All IOW parameters must be either controlled by the operating or inspection staff or they should
themselves control the maintenance and inspection schedules.

It is important to monitor the operating parameters identified as part of the IOW during the RBI study.
This monitoring of the integrity operating window is a condition of assuring the validity of the maximum
inspection intervals recommended in the RBI study. The monitored parameters are also to be used in
trending of relevant degradation mechanisms and fitness for purpose assessment of facilities as they
approach end of design life. Hence, quantification and monitoring of IOW parameters results in building
higher confidence in the carious degradation modes, increasing remnant life of equipment and
improving equipment integrity.

©2017 by NACE International.


Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
8
RBI BENEFITS AND CHALLENGES

Benefits

The Benefits of implementing RBI on a facility are numerous with positive impacts on the business,
process safety, people and the overall technical integrity of assets as listed below:
Business
• Reduced maintenance cost: cost saving from optimized inspection cycle and reduced scheduled
deferment
• Effective use of resources with less unplanned inspection
• Increased equipment availability and reduced shutdown scope: RBI provides basis for
lengthening inspection intervals and optimising maintenance and inspection scope for less
critical items. Also, equipment can be monitored on-stream in lieu of the traditional out-of-
service methods.
• Basis for assurance of asset technical integrity to stakeholders (Regulators/Audits)

Process safety
• Identification of highly critical static equipment and required inspection
• Identification of safe integrity operating window
• Improved knowledge of the facility risk and degradation mechanisms
• Enhance assessment for impact of process change on asset integrity
• Compliance with changing Health, Safety and Environment (HSE) and legal requirements

People
• Collaboration among relevant disciplines responsible for asset integrity
• Cross-discipline competence development: improved corrosion/degradation knowledge for plant
operators and plant operations knowledge for inspectors
• RBI awareness for workshop participants across all disciplines
• Enhance inspector’s competence in risk based monitoring approach

Industry
• Identifying areas for application of new technologies in facilities

RBI Implementation Critical Success Factors

There are certain factors and conditions that contribute to a successful implementation of RBI
approach. These critical success factors are:

• Company leadership commitment for implementation


• Commitment and competence of members of RBI team
• Timely availability of resources
• Quality of asset integrity data available
• Ownership of RBI implementation process by plant/asset team
• Participation of local regulatory bodies – capturing concerns and demonstrating compliance

©2017 by NACE International.


Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
9
Challenges

Typical Challenges encountered in implementing RBI are:


• Unavailability of adequate Inspection history which makes it difficult to establish actual corrosion
rates
• Data quality and assurance
• Limited inspection tools/methods in place
• Insufficient information to fully assess some non- age related failures

RECOMMENDATIONS

Recommendations for improvement of the process and to obtain maximum value for work done are:

• Deploy enhanced inspection method – improved techniques and other non-intrusive methods
• Ensure effective monitoring of Integrity Operating Window (IOW) to assure the validity of
inspection tasks recommended in the RBI study
• Conduct Fitness for Service (FFS) assessment of critical equipment. An assessment of fitness
for continued service is a key part of RBI implementation as it further helps to accomplish
effective degradation risk reduction.
• Ensure update of asset integrity database and review RBI analysis either on a period or ad-hoc
basis to ensure that most recent inspection and maintenance information are included.

REFERENCES

1. Guidance for Corrosion Management in Oil and Gas Production and Processing, (Energy
Institute, London, May 2008), p. 2
2. API RP 580 (latest revision), “Risk Based Inspection” (Washington, DC: API Publishing
Services).
3. Akanni Joseph, Oseahon Edmund, “Managing Corrosion with Risk Based Inspection,” Offshore
West Africa (OWA) Conference, 2015.

©2017 by NACE International.


Requests for permission to publish this manuscript in any form, in part or in whole, must be in writing to
NACE International, Publications Division, 15835 Park Ten Place, Houston, Texas 77084.
The material presented and the views expressed in this paper are solely those of the author(s) and are not necessarily endorsed by the Association.
10

You might also like