Operational data analytics for failure prediction and availability improvement in gas turbine power plants

Abstract

Operational data analytics is crucial in enhancing failure prediction and improving the availability of gas turbine power plants. However, existing research lacks a comprehensive approach that integrates real-time operational data with predictive maintenance models to address high failure rates and prolonged downtime. This study bridges this gap by investigating five GE MS5001 single-shaft, open-cycle gas turbine units, utilizing real-time operational data, historical maintenance records, and performance metrics. Predictive models, including Bayesian simulation and MATLAB-based analysis, were employed to assess failure probabilities and optimize maintenance planning. Key findings reveal a strong correlation between maintenance efficiency and turbine availability, with units exhibiting lower failure rates and shorter mean time to repair (MTTR) demonstrating higher reliability. Conversely, units with frequent failures and extended downtime underscore the limitations of traditional maintenance approaches. The study emphasizes the importance of implementing advanced predictive maintenance strategies to mitigate operational inefficiencies, prevent unexpected failures, and enhance turbine performance. By integrating data analytics with reliability engineering, this research presents a data-driven framework for enhancing the reliability of gas turbine plants. The study contributes to bridging the research gap by demonstrating how predictive analytics can transform maintenance strategies. The study recommends that future research should focus on refining predictive maintenance models through machine learning and AI-based analytics to further improve turbine efficiency and operational resilience. By integrating data analytics with reliability engineering, this study contributes to the advancement of maintenance practices in gas turbine power plants.

Introduction

A gas turbine is a combustion engine that converts liquid fuels, such as natural gas, into mechanical energy, which in turn powers a generator to produce electrical energy. This energy is transmitted to homes and businesses through power lines [1]. Gas turbine power plants are integral to meeting the growing global demand for energy, particularly in the power generation sector. To ensure a stable and efficient supply of electricity, it is crucial to maintain the reliability and availability of these plants. Despite their advanced design, modern gas turbine power plants primarily consist of three main components: the compressor, combustion system, and turbine [2].

Given the complexity of these systems and the multitude of components involved, there is an inherent risk of failure. These failures can lead to significant economic losses, environmental damage, and public safety concerns. Factors such as equipment degradation, maintenance practices, and operational conditions heavily influence the reliability and availability of gas turbine power plants.

Technological advancements, combined with comprehensive training, provide an effective way to mitigate risks and improve maintenance practices [3, 4]. By prioritizing operational and maintenance strategies, operators can enhance gas turbine performance, ensuring reliable and efficient energy production. In this regard, integrating operational data analytics is a powerful approach to proactively predict failures and improve turbine availability.

Operational data analytics employs advanced data processing, machine learning algorithms, and predictive modeling to analyze large volumes of operational data from gas turbine systems. This methodology offers maintenance teams valuable insights into the health and performance of turbines. By leveraging such data, operators can anticipate potential failures, prioritize maintenance tasks, and maximize asset availability [5].

Despite advancements in maintenance strategies, limited studies have integrated real-time operational data analytics with predictive maintenance approaches to enhance gas turbine reliability. Existing literature predominantly focuses on traditional maintenance methodologies, often lacking empirical validation of predictive analytics in gas turbine maintenance. Additionally, few studies have conducted long-term trend analyses using historical data to establish definitive correlations between key reliability metrics, such as MTBF, MTTR, failure rate, and downtime hours, and overall plant availability. Furthermore, a significant research gap exists in the comparative assessment of Bayesian forecasting techniques with conventional reliability analysis models for failure prediction. Addressing these gaps is critical for developing a data-driven, predictive maintenance framework that enhances turbine availability, optimizes maintenance planning, and improves overall operational efficiency in gas turbine power plants.

Aim of the study

This study aims to enhance the availability of a gas turbine power plant consisting of five GE MS5001 single-shaft, open-cycle gas turbines through the application of operational data analytics, predictive maintenance strategies, and advanced analytical tools.

Study objectives, research questions, and research gaps

Research objectives, questions, and gap analysis form essential elements of any study. Research objectives delineate the specific aims and goals, providing direction to the research process. Research questions pinpoint the investigation’s focus, guiding it toward addressing specific inquiries. The research gap identifies current knowledge within the field and underscores areas where additional research is needed to deepen comprehension or resolve unanswered questions. Together, these components establish a framework that shapes the research design, data gathering, analysis, and interpretation. Table 1 highlights these areas.

Table 1. Research objectives, questions, and gaps

S/no	Research objectives	Research questions	Research gaps
1	To investigate the role of operational data analytics in predicting failures and improving the availability of gas turbine power plants	How does operational data analytics contribute to failure prediction and availability improvement in gas turbine power plants?	Limited studies have integrated real-time operational data analytics with predictive maintenance strategies for gas turbine availability enhancement
2	To evaluate the effectiveness of predictive maintenance strategies in ensuring availability and reliability and reducing downtime in gas turbine operations	What impact do predictive maintenance strategies have on reliability metrics such as MTBF, MTTR, and failure rate?	Existing literature primarily focuses on traditional maintenance approaches, with limited empirical validation of predictive analytics in gas turbine maintenance
3	To analyze key maintenance metrics, such as MTTR, MTBF, availability, failure rates, and downtime hours, to assess the performance of gas turbine power plants	What are the relationships between key maintenance metrics and the overall performance of gas turbine power plants?	Few studies have conducted long-term trend analysis using historical data to establish correlations between these reliability metrics and plant availability
4	To apply predictive models and simulations, including MATLAB and Bayesian simulations and to optimize failure analysis and the availability of GE MS5001 gas turbines	How can Bayesian simulations and MATLAB predictive models optimize failure analysis and improve gas turbine availability?	There is a research gap in the comparative assessment of Bayesian forecasting techniques with conventional reliability analysis models for failure prediction

Literature review

While a growing body of research explores operational data analytics for failure prediction and availability optimization, many existing studies provide broad or generalized approaches that do not sufficiently capture the specific operational dynamics of gas turbine power plants. This study builds on previous work by critically evaluating these limitations and aligning new insights with the unique mechanical and thermodynamic challenges inherent in gas turbine systems.

For example, [6] applied ML to predict operational data trends but did not tailor the methodology to the intricacies of gas turbine operations. Similarly, [7] proposed a model for gas turbine performance prediction, yet it focused primarily on performance parameters without integrating real-time failure prediction or linking results to availability metrics, both of which are central to this study.

Further, [8] used ML algorithms to optimize turbine efficiency, providing valuable energy performance insights, but without predicting component-level failures or assessing how efficiency gains impact long-term availability. Although [9] introduced a hybrid HDMR-ANN technique for assessing performance deterioration, its effectiveness is constrained by high computational demand and the need for high-resolution data—conditions that may not be met in gas turbine environments with sensor limitations or intermittent data capture.

In maintenance strategy studies, [10] discussed the role of RCM and data analytics but did not address the operational constraints and frequent start-stop cycles of gas turbines. Likewise, [11] analyzed cost-efficiency improvements without directly connecting these to system availability.

Advanced methods such as AI and big data analytics often assume continuous, high-quality data inputs. However, gas turbines frequently operate with legacy control systems and inconsistent data collection. Studies like [12] and [13] emphasized decision-making and reliability maintenance yet failed to address challenges like data sparsity and delayed fault detection in high-speed rotating equipment.

This study addresses these gaps by proposing a tailored analytics framework focused on improving failure prediction accuracy and system availability in gas turbine power plants.

Methodology

This study adopted a structured methodology to analyze operational data and predict failures in gas turbine power plants. Twelve years of historical operational and maintenance data were collected from the HMI database and ERP system to ensure accuracy and completeness. The data was then cleaned, missing values addressed, and consistency validated across turbine units. Predictive modeling using linear regression and machine learning algorithms followed, focusing on key reliability parameters. Statistical evaluation, including cross-validation and sensitivity analysis, confirmed model robustness. Model predictions were validated against actual performance, enhancing maintenance planning and turbine reliability.

Figure 1 illustrates the methodological flow chart of the study.

[See PDF for image]

Fig. 1

Study methodological flow chart

Data collection

A 12-year dataset (2011–2022) from five GE MS5001 GTUs formed the foundation of a comprehensive reliability and maintenance analysis. The data, sourced from the case study company’s HMI historical database and ERP system, included key metrics such as Rh, F, MTBF, and MTTR. These indicators are crucial for evaluating equipment performance [14], identifying failure trends, and enhancing maintenance strategies.

The HMI system captured real-time turbine operations, including start-up/shutdown logs, alarms, and failure events, enabling continuous monitoring of operational and downtime periods. However, while highly effective for process monitoring, HMI systems often lack structured long-term data essential for historical reliability analysis. The ERP system complemented this by providing detailed maintenance records, covering both scheduled and unscheduled activities.

Integrating HMI and ERP data yielded a comprehensive operational dataset, supporting cross-validation and improving data accuracy by resolving inconsistencies. Reliability metrics were calculated using standard formulas.

Although only five turbine units were analyzed, the longitudinal span of 12 years yielded 60 turbine-year data points. This robust time-series dataset allowed for meaningful trend analysis across varying load conditions and maintenance cycles. Despite the small sample size, the operational diversity and controlled environments of the turbines ensured valid, generalizable insights, supported by cross-validation and Bayesian inference techniques that accounted for variability and uncertainty.

Data processing and cleaning

To ensure the reliability and accuracy of data used in analyzing availability and key reliability parameters, a rigorous preprocessing phase was carried out. This step was essential for eliminating inconsistencies, handling missing values, and verifying data integrity, thereby enhancing the quality of inputs for reliability analysis and predictive modeling.

The first stage involved identifying and correcting inconsistencies arising from data entry errors and irregular recording practices. Common issues included duplicate records, inconsistent timestamps, and implausible values such as negative downtime durations. Outlier detection techniques, particularly the IQR method, were applied to eliminate anomalous data points. To maintain consistency across different turbine units and data sources, normalization and standardization techniques were employed to ensure uniform data formatting.

Addressing missing data was equally important. Randomly missing values were imputed using statistical methods such as mean, median, or mode, while patterned missing data were treated using regression-based or multiple imputation approaches. Records with excessive missing values that could not be reliably estimated were excluded.

To verify data integrity, turbine records were cross-referenced with maintenance and historical reports. Discrepancies, such as mismatches, were flagged and corrected.

Analytical and predictive modeling

Various analytical and predictive modeling techniques were utilized to assess turbine performance. An overview of the key performance metrics is provided below. In addition, Bayesian time-series forecasting, regression analysis, and correlation analysis were applied to predict performance trends, identify failure patterns, and improve operational efficiency through data-driven insights. A summary of these methods and their applications is presented below.

Reliability metrics calculation

Reliability metrics are essential for assessing equipment performance and maintenance effectiveness. Highlighted below are various metrics used for the study, which include: MTTR, MTBF, availability, failure rate, and probability of failure and downtime.

• Mean time to repair (MTTR): The MTTR refers to the average time it takes to fix a gas turbine and restore it to normal operation after a malfunction or failure. This includes the time needed to identify the issue, replace any faulty components, complete the required repairs, and perform the necessary testing and inspections before the turbine can resume its operation. The MTTR is a crucial metric that operators and maintenance personnel should monitor to ensure the overall reliability and availability of gas turbine units, and it is expressed as follows:

  1

  $$MTTR=\frac{\mathrm{DT}}{\mathrm{F}}$$

where DT is the downtime hours and F is the number of failures. The objective is to minimize the MTTR while simultaneously increasing the power plant’s output and profitability. A lower MTTR indicates that the plant has a quick rate of recovery and is well-maintained. Depending on the type and condition of the equipment, world-class MTTR varies. Accurately quantifying MTTR requires the use of CMMS or ERP with suitable processes and timestamps that collect data from multiple sources.

• Mean time between failures(MTBF): The MTBF of a gas turbine refers to the average time interval between two consecutive turbine failures or the expected operating time before the turbine breaks down. A higher MTBF indicates a well-maintained and reliable plant. Several factors can affect the MTBF of a gas turbine, such as the turbine’s design, component quality, operating conditions, maintenance procedures, and external factors. Gas turbines typically have MTBF ratings ranging from 20,000 to 50,000 h or more, implying their long-lasting nature. However, the actual MTBF value may vary significantly depending on the turbine and its operating circumstances. A study by [15] demonstrated that increasing the maintenance interval leads to a decrease in the mean time between failures, highlighting the importance of selecting the maintenance interval carefully. It is important to note that MTBF is merely a statistical estimate and cannot guarantee that a gas turbine will function without failure for a specific period. It is derived from past performance and other variables. Therefore, regular maintenance and inspections are crucial to ensure the reliability and safe operation of gas turbines [16]. MTBF is expressed as follows:

  2

  $$MTBF=\frac{\mathrm{Rh}}{\mathrm{F}}=\frac{\mathrm{TH}-\mathrm{DT}}{\mathrm{F}}=\frac{8760-\mathrm{DT}}{\mathrm{F}}$$

where TH is the total hours for a specified period and Rh is the run hours within that period. When calculating a GT’s MTBF, historical data on the failures of similar gas turbines can be used. Specifically, the MTBF of a GT is determined by dividing the total run hours of all the gas turbines in the dataset by the number of failures that occurred during that time (in this study, 8760 yearly hours are considered). World-class MTBF varies and depends on equipment type, age, and duty cycle.

• Availability (A): Gas turbine availability indicates the percentage of time the unit is operational and ready for use, typically measured annually. A high availability rate suggests effective operation and maintenance, while a low rate signals potential issues. Availability is a key metric for evaluating performance and maintenance efficiency, reflecting both reliability and downtime. Although related, availability and reliability differ; reliability focuses on the frequency of failures, while availability considers total downtime. GT availability is commonly calculated as the ratio of actual operating time to the total scheduled operating time as follows:

  3

  $$A=\frac{\mathrm{Rh}}{\mathrm{Rh}+\mathrm{DT}}=\frac{\mathrm{TH}-\mathrm{DT}}{\mathrm{TH}}=\frac{8760-\mathrm{DT}}{\mathrm{TH}}=\frac{8760-\mathrm{DT}}{8760}=\frac{\mathrm{Rh}}{8760}$$

The formula for availability in terms of MTBF and MTTR is given by the following:

4

$$A=\frac{\mathrm{MTBF}}{\mathrm{MTBF}-\mathrm{MTTR}}\times 100$$

Relatively, for average availability ($A_{\mathit{ave}}$) of several years, y₁, y₂, y₃, y₄, y₅, …… y_n are as follows:

5

$$A_{\mathrm{ave}}=\frac{{\mathrm{Y}}_1+{\mathrm{Y}}_2+{\mathrm{Y}}_3\cdots \cdots \cdots \cdots \cdots \cdots {\mathrm{Y}}_{\mathrm{n}}}{\mathrm{n}}$$

Availability is the measure of how often a system or resource remains fully operational without interruptions. It depends on uptime, downtime, MTBF, and MTTR. High availability results from good system design, regular maintenance, skilled personnel, and favorable conditions. Enhancing availability through redundancy, monitoring, and training reduces downtime costs, boosts efficiency, and ensures consistent service. This leads to higher customer satisfaction, better productivity, and improved business competitiveness and reputation.

• Failure rate (λ): The failure rate of an asset is the frequency with which it performs below expected levels or is not used to its full potential. It is frequently expressed as the number of failures per unit of time or use and is a helpful metric for assessing the reliability and effectiveness of a GTU. A study conducted by [17] revealed that a higher failure rate results in reduced system reliability and availability. It is essential to have effective maintenance management to mitigate the negative impact of equipment failures. This can be achieved by accurately predicting failures and planning appropriate actions. A low failure rate indicates improved quality and reliability. The failure rate is expressed as follows:

  6

  $$\lambda =\frac{1}{\mathrm{MTBF}}$$

Equation (6) is a simplified relationship that assumes a constant failure rate over time.

Although real-world failure patterns often follow the “bathtub curve,” with high failure rates during early (infant mortality) and late (wear-out) phases, this study focuses on the middle phase, where failures occur randomly, and the failure rate is relatively constant. This assumption, widely adopted in reliability engineering, is practical when time-dependent data is limited and supports effective reliability estimation during normal operations. Failures may arise from human error, poor maintenance, wear and tear, or lack of system redundancy. High failure rates signal reliability issues and increased risk. Analyzing these rates is essential for identifying critical components, improving maintenance planning, reducing downtime, and optimizing system performance.

During the infant mortality period, failure rates are high due to manufacturing defects or early malfunctions, often referred to as the “burn-in” phase. This stage is short-lived, as most manufacturers implement screening processes to detect and eliminate these flaws. Once resolved, the system enters its “useful life” or “operational life” phase, where the failure rate significantly decreases and remains low due to prior testing and validation. Over time, however, failures increase again during the “wear-out” phase, caused by aging, fatigue, corrosion, or thermal stress. This progression of failure rates over time is typically represented by the bathtub curve in Fig. 2, which illustrates the system’s failure rate (λ) across its lifecycle. The equation to find λ is as follows:

7

$$\lambda =\frac{\mathrm{F}}{\mathrm{Rh}}$$

[See PDF for image]

Fig. 2

Bathtub curve (Troyler [18])

Plotting failure rates over time reveals the three distinct stages of the bathtub curve: an initial rise due to early failures (infant mortality), a steady period of low failure rates, and an eventual increase during the wear-out phase. Understanding this curve is vital for reliability engineering and maintenance planning. By addressing early failures, manufacturers can enhance system reliability. Proactive maintenance during the operational phase minimizes downtime and costs. Using historical data, engineers can predict wear-out failures and extend component life. Overall, the bathtub curve helps identify improvement areas, optimize maintenance strategies, and ensure higher reliability of complex systems.

• Probability of failure (P[f]): The likelihood of a system, process, or outcome failing is what is meant by the probability of failure. It reflects the likelihood of an event or situation happening, with potential implications for individuals, entities, and society. In this study, we model the probability of failure, P(f), assuming a constant failure rate, which indicates time independence. This assumption simplifies the analysis and is suitable for systems operating during their normal (useful life) phase, where failures occur randomly and the risk remains relatively constant over time [19]. However, we recognize that actual failure behavior often follows the bathtub curve, making P(f) time dependent. To enhance reliability assessments, future work may integrate time-dependent models, such as the Weibull distribution, to account for early-life and wear-out failures. The likelihood of failure is typically represented by a number from 0 to 1, where 0 means no possibility of failure and 1 means complete certainty of failure. This is shown mathematically as follows:

  8

  $$P\left(\mathrm{f}\right)=1-{\mathrm{A}}_{\mathrm{ave}}$$

The likelihood of failure is vital in decision-making and risk assessment across engineering, medicine, finance, and technology. Engineers assess risks like equipment breakdowns, while medical and financial professionals evaluate treatment success or investment risks. Failure can stem from design flaws, human error, or environmental factors. Due to system complexity, predicting failure can be difficult, but statistical models and data analysis help forecast risks and guide mitigation strategies. Understanding failure probability supports informed decisions and effective risk management, reducing negative outcomes and improving success in critical operations and strategic planning.

• Downtime hours (DT): Equipment downtime is the time spent on maintenance, repairs, malfunctions, or other issues when a machine or other piece of equipment is not functioning as planned or is not usable. Three types of equipment downtime may be distinguished: planned, unplanned, and unexpected. Whereas planned downtime is prearranged and scheduled, unplanned downtime occurs abruptly and without warning. While unplanned downtime is not entirely unanticipated, it is also not scheduled. Equipment failure, supply chain issues, age, obsolescence, human error, maintenance and repair, and environmental considerations are just a few of the many elements that can be held accountable for equipment downtime. The annual downtime hours of equipment can be mathematically expressed as follows:

  9

  $$DT=8760-\mathrm{Rh}$$

Equipment downtime can lead to reduced productivity, increased costs, issues with inventory management and quality control, and damage to reputation. Important elements of equipment downtime measurement include tracking % downtime, MTBF, MTTR, and total downtime hours. Equipment downtime may be reduced by employing a variety of strategies, including predictive maintenance, equipment improvements, operator training, effective inventory and spares management, and routine maintenance. Manufacturing companies that understand the causes of equipment downtime and implement effective programs to reduce downtime hours may improve performance, optimize costs, and enhance productivity.

Bayesian time-series forecasting

The Encyclopaedia Britannica defines Bayes’ theorem as a process for “revising predictions in light of relevant evidence.” Moreover, it combines newly acquired evidence with preexisting beliefs to generate a new or posterior belief [21]. This idea is demonstrated by Fagan nomogram, which updates its forecast by integrating the likelihood ratio (new data) with the pre-test probability [22]. The graph in Fig. 3 illustrates this idea visually by showing how, after the prior belief is updated with new evidence, the posterior belief for an intervention indicates a lower relative risk. Daniel Kahneman says in Thinking, Fast and Slow that “Bayes’ rule specifies how prior beliefs should be combined with the diagnosticity of the evidence,” which is a helpful explanation, the extent to which it prefers the theory to the alternative [23]. The formulation for expressing Bayes’ theorem is as follows:

10

$$\mathrm{P}\left(\mathrm{A},/,\mathrm{B}\right)=\frac{\left[\mathrm{P},\left(\mathrm{B},/,\mathrm{A}\right),.,\mathrm{P},\left(\mathrm{A}\right)\right]}{\mathrm{P}\left(\mathrm{B}\right)}$$

[See PDF for image]

Fig. 3

Bayes theorem representation [20]

where P (A/B) is the likelihood of seeing A if B is true. The chance of observing B if A is true is represented by P (B/A), which is the new posterior belief. P (A) is the prior, or the chance that the hypothesis will hold true, before the data are collected, and P(B) is the marginal or the probability that the data will be collected under all conceivable hypotheses. This is sometimes referred to as the likelihood.

Modeling time-series data is essential for generating accurate insights and reliable forecasts. The Bayesian approach enhances this by integrating prior knowledge with observed data, allowing for uncertainty in both model structure and parameters [24]. This is particularly valuable when data are limited or conditions are changing. The process begins with selecting a suitable model, such as AR, MA, ARIMA, or state space, to represent the data. Priors, either informative or non-informative, are assigned to parameters, and a likelihood function is defined. Bayes’ theorem is then used to combine the prior and likelihood, producing a posterior distribution [25]. Forecasts are generated by simulating from the posterior predictive distribution, enabling adaptive, data-informed predictions. Despite computational demands, Bayesian forecasting supports robust planning and failure prediction.

Linear regression analysis

Linear regression analysis is a statistical technique used to establish the connection between a dependent variable and one or more independent variables. The primary goal of linear regression is to create a straight line that best represents the relationship observed in the data. This line, known as the “regression line,” aids in forecasting outcomes for new data based on the identified relationship. The fundamental components of this method include the dependent variable (Y), which is the outcome being predicted, and the independent variable(s) (X), which are utilized in making the prediction. There are two main types of linear regression: simple and multiple. Simple linear regression involves a single independent variable, while multiple linear regression deals with two or more independent variables. The equation for a regression line can be expressed as follows:

11

$$Y=\beta 0+\beta 1\mathrm{X}+\in$$

where Y is the dependent variable, X is the independent variable, β0 is the intercept, β1 is the slope, and ϵ is the error term. The error term signifies the disparity between the observed data and the values anticipated by the regression line. Linear regression begins with data collection and visualization to determine if a linear relationship exists between variables. A regression line is then fitted using the least-squares method, minimizing the sum of squared differences between observed and predicted values. The slope and intercept coefficients are calculated, and the model’s performance is evaluated using metrics such as R-squared and p-values to assess goodness-of-fit and statistical significance. Key assumptions include linearity, independence, homoscedasticity, and normally distributed residuals. In multiple regression, avoiding multicollinearity among predictors is essential. While linear regression is a powerful and simple method, its reliability depends on meeting assumptions and minimizing the impact of outliers.

Regression and correlation analysis

Linear regression models are effective tools for analyzing how key reliability parameters, such as MTBF, MTTR, failure rate, and downtime hours, influence equipment availability. Availability measures the proportion of time an asset remains operational and capable of performing its intended function, making it a crucial metric in reliability engineering.

MTBF, representing the average time between failures, generally shows a positive correlation with availability; a higher MTBF suggests fewer interruptions and increased uptime. Conversely, MTTR negatively affects availability, as longer repair durations extend downtime. Failure rate, being the inverse of MTBF, also exhibits a negative relationship, indicating that more frequent failures reduce availability. Similarly, downtime hours have a strong inverse impact, as extended nonoperational periods diminish system efficiency.

To evaluate these relationships, the coefficient of determination (R²) is used. A high R² value indicates that the selected parameters strongly explain variations in availability. This analysis enables reliability engineers to identify critical factors, optimize maintenance strategies, reduce unplanned failures, and improve resource allocation, ultimately enhancing performance, lowering costs, and increasing operational efficiency.

Data visualization and interpretation (MATLAB)

This research employs MATLAB as the primary analytical tool due to its specialized toolboxes that enhance core functionality, enabling advanced analysis across various domains [26, 27]. MATLAB is used to visualize and analyze operational and reliability data for five GTUs, to produce interpretable plots that support reliability assessment, failure trend analysis, and predictive maintenance planning. Key outputs include failure probability distributions, Bayesian reliability intervals, and linear regression plots linking metrics like MTBF, MTTR, failure rate, and availability. Comparative bar charts illustrate yearly availability trends, while scatter and line plots highlight the impact of downtime on performance.

MATLAB’s powerful statistical tools, including Bayesian inference, allow for the generation of posterior intervals that quantify uncertainty around reliability metrics. These visualizations aid in identifying underperforming units and validating analytical models. In this study, MATLAB functions not as a general-purpose tool but as a domain-specific platform that transforms complex datasets into actionable insights for optimizing turbine performance and maintenance strategies.

Validation and sensitivity analysis

To ensure the accuracy, robustness, and reliability of linear regression models analyzing availability and key reliability parameters, cross-validation techniques are employed. A commonly used method is k-fold cross-validation, where the dataset is divided into k subsets. The model is trained on k-1 folds and tested on the remaining one, repeating this process k times to evaluate performance across different data partitions. This reduces the risk of over-fitting and ensures the model generalizes well to unseen data. Leave-one-out cross-validation (LOOCV), where each data point serves once as a test set, is also utilized for smaller datasets.

Sensitivity analysis further enhances model robustness by examining how changes in input variables, such as MTBF, MTTR, failure rate, and downtime hours, affect availability predictions. Techniques like one-at-a-time (OAT) analysis assess each variable’s influence, while global sensitivity analysis captures interactions among multiple inputs [28]. This process helps identify which parameters most significantly impact model outcomes, supporting more informed maintenance strategies.

Model predictions are then compared against industry benchmarks and findings from existing literature to validate their credibility. Discrepancies are examined for possible data issues, flawed assumptions, or overlooked variables. This comprehensive approach, combining cross-validation, sensitivity analysis, and benchmarking, ensures the models are statistically valid and practically relevant, enabling data-driven decisions for reliability improvement and maintenance optimization.

Results and discussion

Obtained data from the company’s HMI and ERP system for units 2540 (A–E)

In the period 2011–2012 (Table 2), units 2540D and 2540E exhibited the highest run hours, each surpassing 8000 h annually. Despite this high utilization, both units recorded the lowest number of failures (32 and 27 in 2011 and 37 and 28 in 2012, respectively), translating to high MTBF values, 274 and 324 h in 2011 and 237 and 313 h in 2012. These figures suggest superior reliability of these two units. In stark contrast, unit 2540 A had the highest failure counts during this period, 122 failures in 2011 and 127 in 2012, and the lowest MTBF values of 72 and 69 h. Notably, unit 2540D achieved a dramatic reduction in MTTR from 22 h in 2011 to only 5 h in 2012, reflecting significant improvement in maintainability practices.

Table 2. Metrics values obtained from the company’s ERP system between 2011 and 2012

Year	2011					2012
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	5606	7306	5755	8059	8208	5361	7192	6211	8480	8436
Number of failures—F, f	122	51	86	32	27	127	54	89	37	28
MTBF, h	72	172	102	274	324	69	162	98	237	313
MTTR, h	23	26	35	22	20	27	29	29	5	16

As shown in Table 3, covering the 2013–2014 period, similar trends persisted. Units 2540D and 2540E continued to demonstrate strong reliability performance, with MTBF values for 2540D at 265 and 219 h and for 2540E at 214 and 195 h in 2013 and 2014, respectively. These high MTBF values correlate with relatively low failure counts (e.g., 2540D had 33 and 40 failures). Their MTTRs were also among the lowest recorded. Meanwhile, unit 2540 C began to decline in reliability, with an MTBF falling from 101 h in 2013 to 75 h in 2014, while the number of failures rose to 117 in 2014, the highest among all units that year.

Table 3. Metrics values obtained from the company’s ERP system between 2013 and 2014

Year	2013					2014
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	5773	7665	5983	8410	8156	5370	7569	5860	8331	8296
Number of failures—F, f	119	50	87	33	41	132	49	117	40	45
MTBF, h	74	175	101	265	214	66	179	75	219	195
MTTR, h	25	22	34	11	15	26	24	25	11	12

The 2015–2016 data (Table 4) reinforce the earlier observations. Units 2540D and 2540E continued to operate reliably with MTBFs in the range of 204 to 302 h and moderate failure rates. Unit 2540D had 39 and 43 failures across the 2 years, maintaining solid performance. Meanwhile, unit 2540 C became increasingly problematic. Its failure count jumped from 79 in 2015 to 121 in 2016, the highest across all units and years, while its MTBF plunged from 111 to just 72 h. The MTTR for 2540 C also remained high, averaging 30 h, further emphasizing its poor maintainability.

Table 4. Metrics values obtained from the company’s ERP system between 2015 and 2016

Year	2015					2016
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	5738	6903	5887	8208	8392	5536	7788	5764	8559	8348
Number of failures—F, f	130	53	79	39	29	132	52	121	43	30
MTBF, h	67	165	111	225	302	66	168	72	204	292
MTTR, h	32	35	36	14	13	24	19	25	5	14

In the following period, 2017–2018 (Table 5), 2540D recorded an outstanding MTBF of 324 h in 2017, the highest value across all tables, accompanied by just 27 failures. Although this value declined to 243 h in 2018, it remained high relative to other units. The MTTR for 2540D was also notably low (18 and 17 h). Units 2540 A and 2540 C again demonstrated weak performance. For instance, 2540 A recorded 137 and 139 failures in 2017 and 2018, with MTBFs of just 64 and 63 h. unit 2540C’s reliability also remained subpar, with 94 and 126 failures, and MTBFs were below 100 h.

Table 5. Metrics values obtained from the company’s ERP system between 2017 and 2018

Year	2017					2018
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	6132	7761	5992	8261	8296	5641	7481	5913	8147	8157
Number of failures—F, f	137	57	94	27	41	139	59	126	36	42
MTBF, h	64	153	93	324	214	63	148	70	243	208
MTTR, h	19	18	29	18	11	22	22	23	17	10

According to Table 6, which presents the 2019–2020 data, unit 2540D continued to lead in reliability, achieving the highest MTBF values of 337 h in 2019 and 231 h in 2020. Its failure counts remained low, and MTTR values were among the best, at 15 and 11 h, indicating both robust reliability and efficient maintainability. On the other hand, Units 2540 A and 2540 C sustained their unfavorable trends. For instance, 2540 A recorded 134 and 133 failures, respectively, while 2540 C showed 118 and 90 failures with MTBFs of only 74 and 97 h. Their MTTRs, especially for 2540 C (23 and 29 h), suggest ongoing difficulties in restoring functionality after breakdowns.

Table 6. Metrics values obtained from the company’s ERP system between 2019 and 2020

Year	2019					2020
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	5755	7393	6053	8366	8103	5466	7761	6158	8357	8453
Number of failures—F, f	134	50	118	26	31	133	61	90	38	32
MTBF, h	65	175	74	337	282	66	144	97	231	274
MTTR, h	22	27	23	15	21	25	16	29	11	10

Finally, in the 2021–2022 period (Table 7), 2540D and 2540E once again confirmed their strong operational reliability. unit 2540D maintained high MTBFs of 258 and 250 h and failure counts of only 34 and 35, with MTTRs still efficient (15 and 20 h). Unit 2540E followed suit with consistent MTBF values of 214 and 204 h. In contrast, unit 2540 A showed 135 and 139 failures, yielding MTBFs of just 65 and 63 h, while unit 2540C’s MTBFs remained low at 88 and 85 h, alongside high MTTRs of 28 and 29 h, respectively.

Table 7. Metrics values obtained from the company’s ERP system between 2021 and 2022

Year	2021					2022
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	6342	7849	5948	8252	8156	5098	7753	5782	8059	8155
Number of failures—F, f	135	60	99	34	41	139	63	103	35	43
MTBF, h	65	146	88	258	214	63	139	85	250	204
MTTR, h	18	15	28	15	15	26	16	29	20	14

Source: [29]

Across all tables (Tables 2, 3, 4, 5, 6, and 7), Units 2540D and 2540E consistently demonstrate superior performance in both reliability (high MTBF, low failure rates) and maintainability (low MTTR). Their strong performance over a 12-year period indicates effective maintenance planning, better spare part management, skilled personnel, and likely fewer operational anomalies. In contrast, Units 2540 A and 2540 C exhibit persistent reliability and maintainability issues, with failure rates frequently exceeding those of other units and MTBF values consistently falling below 100 h in most years. These trends suggest systemic problems, such as aging equipment, suboptimal maintenance routines, or operational overstress.

Moreover, the consistently high MTTRs for unit 2540 C indicate either complex repair procedures or inefficiencies in the execution of corrective maintenance. To improve the reliability of Units 2540 A and 2540 C, implementing a comprehensive preventive maintenance schedule is essential. These machines require regular inspections, potential component upgrades, and adjusted maintenance timelines to prevent further degradation. Additionally, incorporating predictive maintenance based on machine data could help forecast failures for unit 2540 C, thereby mitigating the spikes in failures observed after 2018.

For Units 2540D and 2540E, the current maintenance strategies are effective and should be continued. Conducting root cause analyses on GTUs with high failure rates, particularly 2540 A, could help identify underlying issues contributing to frequent breakdowns. Finally, ensuring that spare parts are readily available for Units 2540 A and 2540 C could reduce MTTR and operational costs by expediting repairs. Addressing these key areas will enhance overall equipment reliability, minimize downtime, and improve operational efficiency.

Calculated DT, λ, and A data from the obtained data (Rh, F, MTBF, and MTTR) between 2011 and 2022

Table 8 presents reliability and availability metrics for five gas GTUs between 2011 and 2012. Across both years, GTUs 2540 A and 2540 C experienced the highest number of failures, resulting in lower MTBF values of 72–69 h and 102–98 h, respectively. These units also showed higher failure rates (λ), with GTU 2540 A peaking at 0.0145 f/h in 2012. Conversely, GTUs 2540D and 2540E consistently performed better, with high MTBF values ranging from 237 to 324 h and low failure rates between 0.0030 and 0.0042 f/h. Availability for GTU 2540D reached a peak of 96.8% in 2012, while GTU 2540 A was the least available, with values dropping from 64.0% to 61.2%. These patterns reflect early indications of persistent reliability issues with GTU 2540 A, while GTUs 2540D and 2540E displayed operational resilience (Table 8).

Table 8. Calculated DT, λ, and A between 2011 and 2022 for the GTUs

Year	2011					2012
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	5606	7306	5755	8059	8208	5361	7192	6211	8480	8436
Number of failures—F, f	122	51	86	32	27	127	54	89	37	28
MTBF, h	72	172	102	274	324	69	162	98	237	313
MTTR, h	23	26	35	22	20	27	29	29	5	16
Downtime hours—DT (8760—Rh)h	3154	1454	3005	701	552	3399	1568	2,549	280	324
Failure rate—λ (1 ÷ MTBF)f/h	0.0139	0.0058	0.0098	0.0036	0.0030	0.0145	0.0062	0.0102	0.0042	0.0032
Availability—A (Rh ÷ 8760)%	64.0	83.4	65.7	92.0	93.7	61.2	82.1	70.9	96.8	96.3

In Table 9, the reliability metrics in 2013 and 2014 followed similar trends to the previous period. GTU 2540 A continued to experience high failure occurrences (119 and 132 times, respectively), resulting in consistently low MTBFs of 74 and 66 h and elevated failure rates of 0.0135 and 0.0152 f/h. Despite the improved performance of GTU 2540B with high MTBFs (175–179 h) and availability values reaching up to 87.5%, GTU 2540 C deteriorated in 2014 with an MTBF of just 75 h and a failure rate of 0.0133 f/h. GTU 2540D maintained strong performance, recording high availability above 95% across both years, with significantly low failure rates and short MTTR values. The data highlights persistent maintenance challenges with GTUs 2540 A and 2540 C, while GTU 2540D remained reliable and efficient (Table 9).

Table 9. Calculated DT, λ, and A between 2013 and 2014 for the GTUs

Year	2013					2014
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	5773	7665	5983	8410	8156	5370	7569	5860	8331	8296
Number of failures—F, f	119	50	87	33	41	132	49	117	40	45
MTBF, h	74	175	101	265	214	66	179	75	219	195
MTTR, h	25	22	34	11	15	26	24	25	11	12
Downtime—DT (8760–Rh)h	2987	1095	2777	350	604	3390	1191	2900	429	464
Failure rate—λ, (1 ÷ MTBF)f/h	0.0135	0.0057	0.0099	0.0038	0.0047	0.0152	0.0056	0.0133	0.0046	0.0051
Availability—A (Rh ÷ 8760)%	65.9	87.5	68.3	96.0	93.1	61.3	86.4	66.9	95.1	94.7

Table 10 reveals that GTU 2540A’s reliability metrics remained stagnant, with MTBFs of 67–66 h and the highest failure rates recorded during these years at 0.0149 and 0.0152 f/h. Downtime for this unit was significantly high, exceeding 3000 h each year, and availability dropped to 65.5% and 63.2%, respectively. GTU 2540 C worsened in 2016, recording a low MTBF of 72 h and a failure rate of 0.0139 f/h, suggesting maintenance practices were insufficient to curb increasing failure events. In contrast, GTUs 2540D and 2540E continued to demonstrate excellent performance. GTU 2540D, in particular, achieved the highest availability of 97.7% in 2016 with the lowest downtime of just 201 h. This table reinforces the trend of chronic underperformance from 2540 A and highlights the sustainability of GTU 2540D’s maintenance regime.

Table 10. Calculated DT, λ, and A between 2015 and 2016 for the GTUs

Year	2015					2016
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	5738	6903	5887	8208	8392	5536	7788	5764	8559	8348
Number of failures—F, f	130	53	79	39	29	132	52	121	43	30
Downtime hours—DT, (8760–Rh)h	3022	1857	2873	552	368	3224	972	2996	201	412
MTBF, h	67	165	111	225	302	66	168	72	204	292
MTTR, h	32	35	36	14	13	24	19	25	5	14
Failure rate—λ (1 ÷ MTBF)f/h	0.0149	0.0061	0.0090	0.0044	0.0033	0.0152	0.0060	0.0139	0.0049	0.0034
Availability—A (Rh ÷ 8760)%	65.5	78.8	67.2	93.7	95.8	63.2	88.9	65.8	97.7	95.3

Table 11 shows continued challenges for GTUs 2540 A and 2540C. GTU 2540 A reached its highest failure rate of 0.0159 f/h in 2018, with downtime spiking to 3119 h. Similarly, GTU 2540 C recorded its worst MTBF of 70 h in 2018 with a corresponding availability of just 67.5%. These figures imply worsening reliability for these units. Meanwhile, GTU 2540D maintained a solid performance, achieving MTBF values above 240 h and availability exceeding 93% across both years. GTU 2540B also showed consistency with MTBFs of 153 and 148 h and corresponding availability of 88.6% and 85.4%. These results illustrate how certain units benefit from robust maintenance planning, while others continue to lag.

Table 11. Calculated DT, λ, and A between 2017 and 2018 for the GTUs

Year	2017					2018
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	6132	7761	5992	8261	8296	5641	7481	5913	8147	8157
Number of failures—F, f	137	57	94	27	41	139	59	126	36	42
MTBF, h	64	153	93	324	214	63	148	70	243	208
MTTR, h	19	18	29	18	11	22	22	23	17	10
Downtime hours—DT (8760–Rh)h	2628	999	2768	499	464	3119	1279	2847	613	603
Failure rate—λ (1 ÷ MTBF)f/h	0.0156	0.0065	0.0108	0.0031	0.0047	0.0159	0.0068	0.0143	0.0041	0.0048
Availability—A (Rh ÷ 8760)%	70.0	88.6	68.4	94.3	94.7	64.4	85.4	67.5	93.6	93.1

In Table 12, performance trends remained similar. GTU 2540 A continued with high failure rates around 0.0154–0.0152 f/h, and low availability (65.7% and 62.4%), showing no sign of performance recovery. GTU 2540 C remained unstable, with 118 and 90 failures over the 2 years and availability hovering near 70%. In stark contrast, GTU 2540D consistently recorded the highest availability values (95.5% and 95.4%) and the lowest failure rates (0.0030–0.0043 f/h). These numbers highlight the robustness of its operational and maintenance framework. GTU 2540E also showed improvement, with availability rising from 92.5% to 96.5% in 2020.

Table 12. Calculated DT, λ, and A between 2019 and 2020 for the GTUs

Year	2019					2020
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	5755	7393	6053	8366	8103	5466	7761	6158	8357	8453
Number of failures—F, f	134	50	118	26	31	133	61	90	38	32
MTBF, h	65	175	74	337	282	66	144	97	231	274
MTTR, h	22	27	23	15	21	25	16	29	11	10
Downtime -DT (8760–Rh)h	3005	1367	2705	394	657	3294	999	2602	403	307
Failure rate—λ, (1 ÷ MTBF)f/h	0.0154	0.0057	0.0135	0.0030	0.0035	0.0152	0.0069	0.0103	0.0043	0.0036
Availability—A (Rh ÷ 8760)%	65.7	84.4	69.1	95.5	92.5	62.4	88.6	70.3	95.4	96.5

Table 13 indicates that the final 2 years of observation mirrored the earlier trends. GTU 2540 A still experienced elevated failure rates (0.0154–0.0159 f/h) and significant downtimes (2418 and 3662 h), resulting in the lowest availability figures of 72.4% and 58.2%. GTU 2540 C, while slightly improved, maintained failure rates above 0.0114 and 0.0118 f/h, with low availability of 67.9% and 66.0%. On the other hand, GTU 2540B improved, reaching nearly 90% availability. GTU 2540D remained the most reliable throughout, recording high availability (94.2% and 92.0%) and low failure rates (0.0038–0.0040 f/h), proving the consistency and effectiveness of its operational management (Table 13).

Table 13. Calculated DT, λ, and A between 2021 and 2022 for the GTUs

Year	2021					2022
Metrics	2540A	2540B	2540C	2540D	2540E	2540A	2540B	2540C	2540D	2540E
Run hours—Rh, h	6342	7849	5948	8252	8156	5098	7753	5782	8059	8155
Number of failures—F, f	135	60	99	34	41	139	63	103	35	43
MTBF, h	65	146	88	258	214	63	139	85	250	204
MTTR, h	18	15	28	15	15	26	16	29	20	14
Downtime hours—DT (8760–Rh)h	2418	911	2812	508	604	3662	1007	2980	701	605
Failure rate—λ, (1 ÷ MTBF)f/h	0.0154	0.0068	0.0114	0.0038	0.0047	0.0159	0.0072	0.0118	0.0040	0.0049
Availability—A (Rh ÷ 8760)%	72.4	89.6	67.9	94.2	93.1	58.2	88.5	66.0	92.0	93.1

The 12-year data across Tables 8, 9, 10, 11, 12, and 13 provides compelling insights into the operational stability of the five GTUs. GTU 2540 A consistently underperformed, showing persistently high failure rates (ranging from 0.0135 to 0.0159 f/h), low MTBFs, and the lowest availability values among all units. The unit’s high frequency of breakdowns and extended downtimes suggests systemic issues—either design flaws, suboptimal maintenance strategy, or operational stress. GTU 2540 C also demonstrated irregular performance, especially from 2014 onwards, fluctuating in MTBF and λ values and recording availability below 70% in many years.

Conversely, GTU 2540D emerged as the most reliable and efficient unit across the entire assessment period. It consistently recorded the highest MTBF (frequently above 230 h), lowest failure rates (as low as 0.0030 f/h), and peak availability nearing or exceeding 95% in nearly every year. GTU 2540E also maintained high availability and low λ values, though it exhibited more variability than GTU 2540D. GTU 2540B held the middle ground, with moderate failure frequencies and relatively stable availability.

This long-term assessment indicates that some units have benefited from better inherent design, load management, or maintenance planning. In contrast, units like 2540 A and 2540 C likely require detailed root cause analysis and targeted reliability-centered maintenance interventions to enhance their lifecycle performance. The significant differences among identical units operating in similar conditions highlight the importance of individualized maintenance strategies and real-time reliability analytics in modern gas turbine fleet management.

To improve the performance and reliability of the gas turbine units (GTUs), we can adopt several strategies. First, implementing advanced predictive maintenance techniques will help forecast potential failures and minimize unexpected downtimes. For units such as 2540 A and 2540 C, which exhibit notably high failure rates, developing predictive models to monitor the health of critical components and proactively schedule maintenance could be beneficial.

Second, targeted replacement of components in units that consistently demonstrate high failure rates should be considered. Upgrading or replacing critical components that frequently malfunction can significantly reduce failure rates, improving both availability and reliability. Additionally, optimizing maintenance schedules to minimize impact on operational hours is crucial. This involves focusing on reducing mean time to repair (MTTR) to shorten downtimes.

Conducting thorough root cause analyses for units experiencing frequent failures and low reliability, such as 2540 A and 2540 C, is essential. Understanding the underlying causes will enable the implementation of effective solutions that address core issues rather than just treating surface symptoms. Finally, ensuring that maintenance personnel are well-trained and adequately equipped is vital for efficient task performance. Providing access to necessary tools and resources for timely maintenance execution, along with regular training and appropriate resource allocation, can contribute to reduced repair times and enhanced overall maintenance efficiency.

Units’ metrics averages calculated between 2011 and 2022

Table 14 presents average values for availability and other metrics aggregated from data collected from 2011 to 2022, for units 2540 A through 2540E, revealing varying performance levels. Unit 2540 A has the lowest availability at 65%, primarily due to a high failure rate (0.1822 f/h) and a relatively high mean time to repair of 290 h, resulting in a significant average downtime of 37,302 h. Unit 2540B performs better with an availability of 86%, benefiting from a lower failure rate (0.07451 f/h) and a moderate MTTR of 197 h, leading to an average downtime of 14,699 h, due to a high MTTR of 345 h and a significant failure rate of 0.1396 f/h. Unit 2540 C has a 68% availability rate, and this results in a high average downtime of 33,814 h. With extremely low failure rates (0.0488 f/h for 2540D and 0.0480 for 2540E) and low MTTR (175 h for 2540D and 171 h for 2540E), units 2540D and 2540E have the highest level of availability (94%). As a result, their average downtimes are significantly lower at 5631 h and 5964 h, respectively. To increase availability, it is advised to focus on decreasing MTTR by enhancing training programs for maintenance staff, improving spare parts management, and optimizing maintenance and repair processes. Improving MTBF is possible by implementing effective preventive maintenance practices, utilizing advanced condition monitoring technologies, and making design enhancements to increase equipment lifecycle and availability. Utilize data analysis to delve deep into failure analysis and monitor performance data to make decisions based on data and strive for ongoing improvement. Ultimately, by sharing best practices from top-performing units and utilizing them as benchmarks can aid in establishing performance goals and driving enhancements in other units. Utilizing these strategies can improve the overall availability and reliability, resulting in decreased downtime and effective operation of gas turbine power plant performance through a comprehensive analysis of operational data.

Table 14. Units’ metrics averages calculated from Tables 8, 9, 10, 11, 12, and 13

GT units	${\mathit{R}\mathit{h}}_{\mathit{a}\mathit{v}\mathit{e}}$ (h)	${\mathit{M}\mathit{T}\mathit{T}\mathit{R}}_{\mathit{a}\mathit{v}\mathit{e}}$ (h)	${\mathit{M}\mathit{T}\mathit{B}\mathit{F}}_{\mathit{a}\mathit{v}\mathit{e}}$ (h)	${\mathit{F}}_{\mathit{a}\mathit{v}\mathit{e}}$ (f)	${\mathit{D}\mathit{T}}_{\mathit{a}\mathit{v}\mathit{e}}$ (h)	${\mathit{\lambda }}_{\mathit{a}\mathit{v}\mathit{e}}$ (f/h)	${\mathit{A}}_{\mathit{a}\mathit{v}\mathit{e}}$ (%)
2540A	67,818	290	800	1579	37,302	0.1822	65
2540B	88,739	197	1923	656	14,699	0.0745	86
2540C	71,289	345	1077	1215	33,814	0.1396	68
2540D	99,469	175	3067	420	5631	0.0488	94
2540E	99,156	171	3034	412	5964	0.0480	94

Results of failure probability evaluation

The failure probability of the GT units, which was obtained from the data in Tables 2, 3, 4, 5, 6, and 7), is visualized in Fig. 4. Between 2011 and 2022, the likelihood of differential unit failure decreased due to regular maintenance procedures on the GTUs. The probability of failure for unit 2540 A varied slightly, with an increase in 2019 compared to previous years. Unit 2540B’s probability was relatively constant but increased slightly between 2020, 2021, and 2022, suggesting a largely unstable chance of failure. Unit 2540 C significantly declined in 2012, possibly due to a higher failure risk. Unit 2540D had a low failure probability with small variations over the years, indicating high reliability. Unit 2540E’s probability showed oscillation in 2022, indicating fluctuations in the chance of failure over the years and its availability likened to 2540D.

[See PDF for image]

Fig. 4

Failure probabilities of gas turbine units

According to the MATLAB graphical results of the failure probability evaluation, units 2540B, 2540D, and 2540E exhibit consistent performance with few variations and low failure probability suggesting optimal maintenance procedures, whereas unit 2540 A indicated a greater chance of failure in 2019; unit 2540 C had a higher risk of failure from 2015 to 2018. It is advised to optimize maintenance strategies for units 2540 A and 2540 C by using more frequent inspections and predictive maintenance techniques to solve these problems. Further investigation into the root causes of increased failure probabilities during specific years should be conducted to identify and mitigate potential issues. Regular reviews and updates of maintenance schedules, considering performance trends and failure probabilities, will help maintain the overall availability of the units.

Results of time-series Bayesian numerical simulation

Figure 5 illustrates the responses of 2540 A intervals to the 95% Bayesian posterior interval, the prior distribution sample of potential curves, the posterior distribution utilizing discretization, and the prior distribution. In general, assessing probability is a challenging and time-consuming operation. The availability range for the unit was 0.612 to 0.724 between 2011 and 2022. 2021 and 2012 saw the greatest and lowest efficiency levels, respectively. Despite these variations, the turbine’s availability stayed within a narrow range over the years, indicating lower availability levels and a moderate level of stability in performance.

[See PDF for image]

Fig. 5

2540A reliability interval responses by applying Bayesian

Figure 6 shows the prior distribution sample of potential curves, the posterior distribution sample of discretization, the prior distribution sample of possible curves, and the 95% Bayesian posterior interval. While it is often possible to estimate the prior probability from historical data, obtaining appropriate estimates of the conditional probabilities can be challenging at times. When these intervals can be approximated using a formal casual interval, this process is greatly simplified. Between 2011 and 2022, 2540B exhibited variability, with values ranging from 0.788 to 0.896 between 2011 and 2022. Surprisingly, 2015 had the lowest availability (0.788), while 2021 recorded the highest availability (0.896). Notwithstanding these fluctuations, it consistently maintained relatively moderate-to-high reliability levels, with the majority of values exceeding 0.8.

[See PDF for image]

Fig. 6

2540B reliability interval responses by applying Bayesian

Figure 7 shows the 95% Bayesian posterior interval, the prior distribution sample of feasible curves, the posterior exploitation discretization, and the 2540 C intervals availability estimate of the prior distribution. A lesser degree of reliability is shown by some of the curves that are not regularly distributed, indicating a lower level of availability. Availability levels for 2540 C also show minor variations over the years ranging from 0.65 to 0.71. It has comparatively minimal variations in availability when compared to some of the other units. The years 2012 and 2020 have the highest availability values (0.709), while 2016 had the lowest reliability (0.658). The steady reliability levels imply that 2540 C operates steadily over an extended period, with minimal variations that might be within an allowable variability range.

[See PDF for image]

Fig. 7

2540C reliability interval responses by applying Bayesian

Figure 8 shows the posterior using discretization, the prior distribution sample of potential curves, the posterior using discretization, and the 95% Bayesian posterior interval. It can be seen that some of the curves, which represent strong availability, are not normally distributed. It can be seen that some of the curves, which represent strong availability, are not normally distributed. From 2011 to 2022, 2540D showed availability levels ranging from 0.92 to 0.968, having less fluctuation in its availability trending when compared to other units; most of its values are consistently higher than 0.92. The years 2012 and 2013 recorded the highest availability values of 0.968, while 2022 recorded the lowest reliability of 0.92. Its steady performance suggests that it is given proper maintenance, which helps to ensure that its availability is sustained over time.

[See PDF for image]

Fig. 8

2540D reliability interval responses by applying Bayesian

Figure 9 shows a 2540E interval availability estimate of the posterior exploitation discretization, the prior distribution sample of feasible curves, the posterior exploitation discretization, and the 95% Bayesian posterior interval. With a maximum degree of confidence, some of the curves are not normally distributed, availability levels range from 0.925 to 0.965 over the years that are being examined, and having good reliability like its 2540D counterpart, it exhibited minor variations, most values being greater than 0.92. 2018 has the lowest at 0.931, while the highest values, at 0.965, are recorded in 2012 and 2020. Its consistent performance over the assessment period points to appreciable adherence to maintenance suggestions that support its availability.

[See PDF for image]

Fig. 9

2540E reliability interval responses by applying Bayesian

Based on the analysis, units 2540 A and 2540 C are predicted to have the highest failure probabilities and are likely to experience lower availability in the future. Unit 2540 A demonstrated an availability range of between 0.612 and 0.724 from 2011 to 2022, indicating lower overall availability and moderate stability. The unit showed the greatest efficiency in 2021 and the lowest in 2012, but despite these variations, its performance remained within a narrow range, suggesting potential challenges in maintaining consistent availability. Similarly, unit 2540 C exhibited availability range from 0.65 to 0.71, reflecting lower availability compared to the other units. Although the variations in availability were minor, with the highest values recorded in 2012 and 2020 (0.709) and the lowest in 2016 (0.658), the unit’s performance was consistently lower than its counterparts. The steady availability levels imply that unit 2540 C operates with minimal variations, but these lower availability figures indicate the potential for higher failure rates in the future. In contrast, units 2540D and 2540E showed higher and more consistent availability levels, with unit 2540D ranging from 0.92 to 0.968 and unit 2540E from 0.925 to 0.965. These units exhibited less fluctuation in availability and maintained values consistently above 0.92, suggesting effective maintenance practice. Therefore, units 2540 A and 2540 C, with their lower availability ranges and higher variability in performance, are anticipated to have higher failure probabilities and lower availability in the future compared to units 2540D and 2540E.

Results of availability analysis of gas turbine units

Figure 10 visualizes data of availability for gas turbine units: 2540 A, 2540B, 2540 C, 2540D, and 2540E, and the bar lengths indicate significant variations in their performance. As shown in the figure, with the longest bar length, units 2540D and 2540E consistently exhibit the highest availability rates at 94%, while unit 2540 A shows the lowest performance at 65%. Units 2540B and 2540 C fall in between with 86% and 68% availability, respectively. Targeted maintenance interventions are necessary for units 2540 A and 2540 C as they are underperforming compared to the others. Implementing predictive or preventive maintenance strategies could greatly enhance the availability of these units. The superior availability of units 2540D and 2540E suggests that they experience fewer failures or have shorter downtimes possibly due to more effective maintenance practices, better operational conditions, or higher-quality components. Conversely, the lower availability of units 2540 A and 2540 C, and the moderate performance of unit 2540B, indicates underlying issues such as frequent breakdowns, extended repair durations, or suboptimal maintenance procedures. Addressing these deficiencies could significantly improve their operational efficiency. A comprehensive review of their maintenance history and operational performance is recommended to enhance the availability of units 2540 A, 2540 C, and 2540B. Identifying the root causes of their reduced availability, like maintenance inefficiencies or operational challenges, can facilitate targeted improvements. Techniques such as condition-based monitoring and advanced diagnostics can help detect failures early and minimize downtime. Analyzing the performance and failure trends of these units in comparison to the more reliable units 2540D and 2540E can provide valuable insights for enhancement. Upgrading critical components, optimizing maintenance schedules, and ensuring the availability of spare parts and repair teams can all contribute to improved reliability and availability. In summary, addressing maintenance and operational efficiency gaps offers opportunities to elevate the performance of units 2540 A, 2540 C, and 2540B, leading to more consistent and higher availability across all gas turbine units.

[See PDF for image]

Fig. 10

A comparative bar chart of availability percentages of the different GTUs

Figure 11 illustrates the relationship between the availability and MTBF of the GTUs (2540A, 2540B, 2540 C, 2540D, and 2540E). The linear regression model equation is y = 0.0129x + 55.824, where y represents availability and x represents MTBF. The positive slope of 0.0129 indicates that availability improves by approximately 0.0129% for every hour increase in MTBF. A value of 0.9544 reveals a strong correlation between MTBF and availability, with 95.44% of the variability in availability explained by changes in MTBF. The positive correlation suggests that systems with longer intervals between failures tend to have higher availability. This is expected because systems with fewer failures require less repair downtime, resulting in better performance. For example, when MTBF is around 3000 h, availability is close to 90%. In contrast, lower MTBF values like 1000 h decrease availability to around 70%. This trend emphasizes the importance of maximizing MTBF to ensure the system remains operational for longer periods and experiences fewer breakdowns. To improve system performance, it is essential to focus on increasing MTBF. This can be achieved by implementing robust preventive maintenance programs that address wear and tear before equipment fails. Regular inspection and maintenance can extend the time between failures, leading to higher MTBF values and increased availability. Advanced predictive maintenance techniques like condition monitoring and data-driven maintenance can help anticipate potential failures, further improving MTBF. Improving system design and reliability by selecting high-quality materials and components resistant to failure can also contribute to a higher MTBF. By implementing these design improvements, systems can withstand more operational stress and perform reliably over longer periods, resulting in less frequent failures and higher availability. In conclusion, the graph highlights the critical importance of increasing MTBF to improve system availability. Implementing preventive and predictive maintenance strategies, along with enhancing system design, is essential to maximizing MTBF. These approaches lead to longer intervals between failures, reduced downtime, and ultimately higher availability, resulting in improved operational efficiency and system performance.

[See PDF for image]

Fig. 11

Linear regression graph correlating availability and MTBF of the GTUs

Figure 12 illustrates the relationship between availability and MTTR using a linear regression model. The equation y =  − 0.1695x + 121.34 shows a negative correlation between these factors, with y representing availability and x representing MTTR. The negative slope (− 0.1695) indicates that for every additional hour spent on repairs, availability decreases by 0.1695%. The strong correlation value of 0.8858 means that about 88.58% of the variance in availability is due to changes in MTTR. This shows that MTTR significantly impacts availability, and reducing MTTR will improve system performance. As MTTR increases, availability decreases. For example, when MTTR reaches 300 h, availability drops below 70%, indicating a significant performance issue. This suggests that consistently high repair times will result in unacceptable levels of availability, affecting operational efficiency. To address this, strategies should focus on reducing MTTR. One approach is to improve the spare parts management system to ensure parts are readily available. Additionally, enhancing the skills of the maintenance workforce and optimizing repair processes can reduce repair times. Preventive and predictive maintenance strategies can also help prevent failures that lead to extended MTTR. By addressing potential issues before they cause breakdowns, availability can be maintained at higher levels. Continuous monitoring of key performance indicators, like MTTR and availability, is essential. This allows organizations to identify trends early and prevent further degradation in system performance. In conclusion, the data analysis clearly shows the critical impact of MTTR on availability. Reducing MTTR through improved processes, skilled personnel, and predictive maintenance strategies is vital for maintaining high availability levels and efficient operations.

[See PDF for image]

Fig. 12

Linear regression graph correlating availability and MTTR of the GTUs

Figure 13 presents a graphical representation that depicts the correlation between system availability and the failure rate of the GTUs. The chart demonstrates a clear negative linear relationship between system availability and the failure rate. The regression equation shown on the chart is y =  − 230.66x + 104, with an R² value of 0.9659. This indicates that approximately 96.6% of the variability in availability can be explained by changes in the failure rate. The notably steep negative slope implies that with each increase in the failure rate, there is a significant decrease in system availability. At the outset, when the failure rate is low, availability stands at about 100%. However, as the failure rate approaches 0.2 failures per hour, availability dwindles to reduce to 50%. The high R^2 value of 0.9659 suggests that the linear model of the slope of the regression equation, which is − 230.66, reveals that for each unit increase in the failure rate, the availability decreases by approximately 230.66%. This emphasizes the substantial impact of failure rates on system availability. While the practicality of this value might be limited, given that availability cannot fall below zero, it stresses the importance of minimizing failure rates to maintain high availability levels. The intercept of 104% indicates the system’s availability when the failure rate approaches zero, implying that under optimal conditions, the system could operate with an availability close to 104%. However, availability deteriorates swiftly as the failure rate rises, highlighting the system’s sensitivity to component failures. The strong linear relationship emphasizes the significant role of the failure rate in influencing system availability. To address the insights uncovered, prioritizing key strategies can effectively boost system availability and reduce failure rates. Firstly, incorporating predictive maintenance methods ensures constant monitoring of equipment, enabling proactive measures to prevent failures and enhancing uptime significantly. Secondly, embracing condition-based maintenance practices aids in the early identification of potential failures, thereby strengthening system reliability. Thirdly, efficient management of critical components is paramount. Targeting components with high failure rates through upgrades or enhanced materials helps minimize downtime. Lastly, continuous monitoring and adjustment of maintenance schedules play a vital role. Robust data collection systems empower teams to spot trends and make informed decisions, optimizing maintenance strategies over time. By implementing these strategies, organizations can mitigate the adverse effects of failures, uphold elevated availability levels, and guarantee optimal performance and reliability throughout their systems.

[See PDF for image]

Fig. 13

Linear regression graph correlating availability and failure rate of the GTUs

Figure 14 visualizes data of availability over run hours for the GTUs. Analyzing the availability trends of gas turbines based on run hours offers valuable insights into their performance and areas for enhancement. The visualization indicates a direct correlation between increased run hours and higher bar heights. Particularly noteworthy are units 2540 A and 2540 C, which consistently exhibit shorter bars, reflecting lower availability compared to the other units. This suggests ongoing suboptimal performance trends that warrant further attention and potential improvement efforts. Implementing real-time data monitoring for these units can facilitate adaptive maintenance scheduling, potentially enhancing their availability. On the other hand, units 2540D and 2540E, maintaining consistently high availability levels, highlight the effectiveness of proactive and timely maintenance strategies. These successful maintenance approaches should be upheld and extended to underperforming turbines where applicable. Examining gas turbine availability about their running hours reveals significant patterns. Units like 2540D and 2540E, boasting higher average run hours (99,469 and 99,156 h), demonstrate superior availability at 94%, indicating efficient and consistent maintenance practices. In contrast, unit 2540 A, with the lowest run hours (67,818), records the lowest availability of 65%, suggesting a need for enhanced maintenance methodologies or a decline in performance. Units such as 2540B and 2540 C exhibit moderate performance levels, with run hours of 88,739 and 71,289 and corresponding availabilities of 86% and 68%. While these turbines perform relatively well, there is room for optimization. Generally, availability improves with increasing run hours, signifying that turbines with more operational experience likely benefit from established maintenance schedules that minimize downtime and sustain performance. Monitoring availability periodically enables the tracking of performance trends, aiding in identifying improvements or deteriorations. For turbines like 2540 A and 2540 C with lower availability, continuous monitoring is crucial to assess progress and adjust maintenance strategies accordingly. It is recommended to optimize maintenance approaches for units with lower availability, particularly 2540 A, and 2540 C, by enhancing preventive and predictive maintenance practices to reduce downtime and enhance availability. Implementing predictive analytics for turbines with higher run hours, like 2540B, 2540D, and 2540E, can help sustain their high availability by mitigating unexpected failures. Regular monitoring of run hours and availability data is essential to spot performance issues early and make informed decisions regarding maintenance scheduling. Consideration of component upgrades for units with consistently lower availability can further enhance reliability, decrease repair frequency, and ultimately boost long-term performance.

[See PDF for image]

Fig. 14

A comparative bar chart of GTUs and their corresponding run hours

Figure 15 presents a graphical representation that showcases a clear negative linear relationship between the system’s availability and downtime. The equation y =  − 0.0009x + 99.4, supported by a high coefficient of determination (R² = 0.9999), signifies a robust fit of the linear model to the data. With a slope of − 0.0009, it is evident that a mere additional hour of downtime results in approximately a 0.09% decrease in availability. This sensitivity underscores the critical need to minimize interruptions for optimal system performance. Conversely, the intercept of 99.4% indicates that under ideal conditions with minimal downtime, availability approaches near-perfect operational efficiency. The high R² value of 0.9999 confirms that nearly all variability in availability relates to downtime, emphasizing the importance of reducing downtime for enhancing system reliability. In light of these findings, several recommendations can be made. Firstly, prioritizing downtime reduction strategies is key for maintaining higher availability. Implementing predictive maintenance techniques, optimizing repair strategies, and ensuring prompt responses to emerging issues can significantly reduce downtime occurrences. Secondly, proactive maintenance can be performed on components before failure by implementing maintenance optimization techniques like condition-based or predictive maintenance, minimizing unexpected downtimes, and extending critical component lifespans. Moreover, identifying and addressing key bottlenecks contributing to downtime, such as frequent component failures or lengthy repair times, can enhance overall system efficiency. Allocating resources strategically to address these bottlenecks helps minimize disruptions and optimize operational continuity. Lastly, continuous monitoring and adjustment of downtime metrics and availability trends are vital. This iterative process allows for refining maintenance strategies over time, ensuring sustained high availability and operational reliability. By implementing these strategies, organizations can effectively maximize system availability while mitigating the adverse impacts of extended downtimes, ultimately optimizing overall operational performance and reliability.

[See PDF for image]

Fig. 15

Linear regression graph correlating availability and downtime of the GTUs. All illustration data were derived from Table 8

Furthermore, while this statistical correlation is consistent with engineering expectations, it is important to acknowledge that correlation alone does not imply causation. Multiple factors, such as failure severity, maintenance response time, or operational scheduling, may also contribute to availability loss. However, within the domain of gas turbine operations, there exists strong theoretical and empirical justification to interpret this relationship as at least partially causal. In RCM frameworks, availability is mathematically defined by uptime and downtime (A = run hours/[run hours + downtime]), directly linking any increase in downtime to a proportional decline in availability. Moreover, sensitivity analysis conducted in the study confirms that small increases in downtime consistently result in observable reductions in availability across all five turbine units, even after accounting for variations in MTTR and failure rates. This consistency, coupled with the physical operational logic of turbine systems, where extended downtimes directly reduce asset productivity, supports a causal interpretation.

It should be noted that while availability cannot practically exceed 100%, areas where the intercepts is above 100% suggest that the system is theoretically close to full availability when the failure rate is very low.

Factors affecting gas turbine availability and improvement measures

The analysis of factors influencing gas turbine availability reveals several key insights. Units with lower MTTR, such as 2540D and 2540E, exhibit higher availability, suggesting that shorter repair times contribute to better operational efficiency. Conversely, units with higher MTTR, like 2540 A and 2540 C, experience lower availability, indicating that prolonged repair times reduce operational uptime. Additionally, units with higher MTBF also show higher availability, as longer intervals between failures result in fewer disruptions and greater operational time. In contrast, units with lower MTBF, such as 2540 A and 2540 C, face more frequent failures, leading to decreased availability.

The number of failures is another critical factor; units with fewer failures, like 2540D and 2540E, have higher availability, reflecting that reduced failure frequency supports greater uptime. Units with more frequent failures, such as 2540 A and 2540 C, experience increased downtime and lower availability. Similarly, units with less downtime, such as 2540D and 2540E, demonstrate higher availability, as reduced downtime translates to more time in operation. High downtime, as seen in units like 2540 A and 2540 C, adversely impacts availability.

Finally, a lower failure rate is associated with higher availability. Units like 2540D and 2540E, with lower failure rates, have better availability compared to units with higher failure rates, such as 2540 A and 2540 C, which face more frequent disruptions. To enhance availability, it is crucial to focus on reducing MTTR, increasing MTBF, decreasing the number of failures, minimizing downtime, and lowering the failure rate. Implementing effective maintenance strategies and optimizing repair processes can significantly improve overall operational efficiency and availability.

Future research direction

Based on current research findings, future directions could explore enhanced predictive maintenance strategies using AI and machine learning, integrating real-time data analytics for proactive equipment management, and assessing the environmental impact of maintenance practices on sustainable operations in industrial settings. These areas aim to advance reliability and efficiency while minimizing the environmental footprint. These areas are as follows:

• Integration of machine learning for failure prediction: Explore the application of advanced machine learning models, such as deep learning, decision trees, and support vector machines, to enhance the accuracy of failure predictions and optimize maintenance schedules. A comparative analysis of different AI-driven models could provide insights into the most effective predictive maintenance approach for gas turbine reliability.

• Real-time data analytics and digital twin implementation: Investigating the use of real-time data analytics combined with digital twin technology can provide a more dynamic and proactive maintenance framework. Future research could develop and validate digital twin models for gas turbines, simulating operational conditions to predict failures and optimize performance.

• Lifecycle cost analysis of maintenance strategies: A comprehensive study comparing the lifecycle costs of different maintenance strategies—corrective, preventive, and predictive—can provide a data-driven approach to selecting the most cost-effective maintenance model. This research should consider long-term financial implications, including downtime costs, repair expenses, and return on investment for predictive maintenance technologies.

Conclusions

This research offers a comprehensive analysis of data-driven methods for failure prediction and availability optimization in gas turbine power plants, utilizing long-term operational data from five GE MS5001 GTUs collected between 2011 and 2022. By integrating historical maintenance records with advanced analytics, such as Bayesian inference, MATLAB-based prognostic modeling, and linear regression analysis, the study identified critical reliability trends and recurring failure patterns. The results highlight significant performance variations among turbines, directly linked to the effectiveness of maintenance practices.

Units 2540 A and 2540 C, which recorded the lowest availability, experienced higher failure rates and longer MTTR, contributing to extended downtimes. In contrast, units 2540D and 2540E showed consistently high availability due to reduced failure incidences and quicker repair responses. Bayesian forecasting provided credible future availability projections, underscoring the importance of predictive maintenance. These findings support earlier works (e.g., [6]– [9]) that advocate data-driven strategies to optimize turbine reliability and availability.

Unlike many prior studies that focused on conventional maintenance, this research closes a key gap by combining real-time operational analytics with predictive methodologies, offering more accurate and actionable insights. Trend analyses confirmed strong correlations between key metrics, MTBF, MTTR, failure rate, and downtime and plant availability. Cross-validation and sensitivity tests confirmed the reliability and real-world applicability of the predictive models.

The study advocates future research into machine learning models such as deep learning, decision trees, and support vector machines to further enhance prediction accuracy. Integrating digital twin technologies with real-time analytics is also recommended for proactive maintenance. Additionally, a lifecycle cost analysis comparing corrective, preventive, and predictive strategies would clarify long-term financial impacts.

Operational data analytics significantly strengthens gas turbine maintenance planning, offering measurable gains in reliability, cost efficiency, and asset longevity and key priorities in modern maintenance engineering.

Acknowledgements

Many thanks to the maintenance team for their cooperation and contributions during the data collection phase.

Authors’ contributions

Methodology, AJO. Validation, AJO and EGS. Formal analysis, AJO and MIG. Writing—original draft preparation, AJO. Writing—review and editing, AJO, EGS, MIE, and DE.

Funding

This study was not supported by any grants from funding bodies in the public, private, or not-for-profit sectors. All research expenses were self-funded by the authors.

Data availability

The corresponding author will provide the data supporting the study’s findings upon request.

Declarations

Competing interests

The authors declare that they have no competing interests.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. GE Vernova (2023) What is a gas turbine, and how does it work?. Available: https://www.ge.com/gas-power/resources/education/what-is-a-gas-turbine

2. USDE (2023) How gas turbine power plants work. Office of Fossil Energy and Carbon Management. Available: https://www.energy.gov/fecm/how-gas-turbine-power-plants-work

3. Okirie, AJ. Impact of reliability variations on overall power station performance. International Journal of Scientific Research and Engineering Development; 2024; 7, pp. 1024-1030. [DOI: https://dx.doi.org/10.5281/zenodo.13164105]

4. Okirie, AJ; Barnabas, M; Ejomarie, E; Asomie, AS. Understanding deviations from original equipment manufacturers’ maintenance recommendations: reasons, barriers, and benefits. Journal of Engineering and Applied Sciences; 2024; 71, 219. [DOI: https://dx.doi.org/10.1186/s44147-024-00563-y]

5. Arinze CA, Izionworu VO, Isong D, Daudu CD, Adefemi A (2024) Predictive maintenance in oil and gas facilities: leveraging AI for asset integrity management. https://doi.org/10.53294/ijfetr.2024.6.1.0026

6. Goyal V, Xu M, Kapat J, Vesely L (2020) Prediction of gas turbine performance using machine learning methods. In: Turbo Expo: Power for Land, Sea, and Air, vol. 84157, p. V006T09A004. https://doi.org/10.1115/GT2020-15232

7. Liu, Z; Karimi, IA. Gas turbine performance prediction via machine learning. Energy; 2020; 192, [DOI: https://dx.doi.org/10.1016/j.energy.2019.116627] 116627.

8. Jayachandran AVT, Omar HH, Tkachenko AY (2020) Machine learning predictor for micro gas turbine performance evaluation. Aeron Aero Open Access Journal, 4(4):172–180. https://doi.org/10.15406/aaoaj.2020.04.00120

9. Kiakojoori, S; Khorasani, K. Dynamic neural networks for gas turbine engine degradation prediction, health monitoring, and prognosis. Neural Comput Appl; 2016; 27, 8 pp. 2157-2192. [DOI: https://dx.doi.org/10.1007/s00521-015-1990-0]

10. Ziegel, E. R. (1996). Review of Reliability-Centered Maintenance, by J. Moubray. Technometrics 38(1):87-7. https://doi.org/10.2307/1268924

11. Badran, OO. Gas-turbine performance improvements. Appl Energy; 1999; 64, 1–4 pp. 263-273. [DOI: https://dx.doi.org/10.1016/S0306-2619(99)00088-4]

12. Ghasemaghaei, M; Ebrahimi, S; Hassanein, K. Data analytics competency for improving firm decision-making performance. J Strateg Inf Syst; 2018; 27, 1 pp. 101-113. [DOI: https://dx.doi.org/10.1016/j.jsis.2017.10.001]

13. Djeddi, AZ; Hafaifa, A; Hadroug, N; Iratni, A. Gas turbine availability improvement based on long short-term memory networks using deep learning of failure data analysis. Process Saf Environ Prot; 2022; 159, pp. 1-25. [DOI: https://dx.doi.org/10.1016/j.psep.2021.12.050]

14. De Souza, GFM. Thermal power plant performance analysis; 2012; London, Springer:

15. Ahsan, S; Lemma, TA; Gebremariam, MA. Reliability analysis of gas turbine engine by means of bathtub-shaped failure rate distribution. Process Saf Prog; 2020; 39, [DOI: https://dx.doi.org/10.1002/prs.12115] e12115.

16. Aneke, A; Odukwe, A; Onah, T. Effect of failure rates on system reliability of a gas turbine power plant. International Journal of Multidiscip Research and Development; 2017; 4, 7 pp. 68-73. [DOI: https://dx.doi.org/10.1016/j.ijpvp.2004.07.010]

17. Roesch, WJ. Using a new bathtub curve to correlate quality and reliability. Microelectron Reliab; 2012; 52, 12 pp. 2864-2869.

18. Troyler D (2013) Reliability engineering principles for the plant engineer. Noria Corporation. Available: https://supplychainminded.com/reliability-engineering-principles-for-the-plant-engineer/

19. Giribone, R; Valette, B. Principles of failure probability assessment (PoF). Int J Press Vessels Pip; 2004; 81, 10–11 pp. 797-806. [DOI: https://dx.doi.org/10.1016/j.ijpvp.2004.07.010]

20. Steel MF (2010) Bayesian time series analysis. In: Macroeconometrics and Time Series Analysis. London: Palgrave Macmillan UK, 35–45. https://doi.org/10.1057/9780230280830_4

21. Koop, G; Korogilis, D. Bayesian multivariate time series method for empirical microeconomics. Foundations and Trends in Econometrics; 2010; 3, 4 pp. 267-358. [DOI: https://dx.doi.org/10.1561/0800000013]

22. Walker G (2020) Bayesian Statistics. Available: https://www.thebottomline.org.uk/blog/ebm/bayesian-statistics/

23. Routledge R (2024) Bayes’s theorem. Encyclopedia Britannica. Available: https://www.britannica.com/topic/Bayess-theorem

24. Barber D, Cemgil AT, Chiappa S, Bayesian time series models. Cambridge University Press

25. Qiu, J; Jammalamadaka, SR; Ning, N. Multivariate time series analysis from a Bayesian machine learning perspective. Ann Math Artif Intell; 2020; 88, 10 pp. 1061-1082.4156189 [DOI: https://dx.doi.org/10.1007/s10472-020-09710-6] 1465.62101

26. Niazai, S; Rahimzai, AA; Atifnigar, H. Applications of MATLAB in natural sciences: a comprehensive review. Eur J Theoret Appl Sci; 2023; 1, 5 pp. 1006-1015. [DOI: https://dx.doi.org/10.59324/ejtas.2023.1(5).87]

27. Turk, I. Practical MATLAB; 2019; Berkeley, CA, USA, Apress:

28. Stano, PM; Spirzewski, M. Bayesian integrated global uncertainty & sensitivity analysis of system’s availability. In 2019 Annual Reliability and Maintainability Symposium (RAMS). IEEE, FL, USA; 2019; 28, pp. 1-7. [DOI: https://dx.doi.org/10.1109/RAMS.2019.8768905]

29. Okirie AJ, Saturday EG, Mathew IG, Ewe D (2024) Acquisition of reliability and availability metrics data. (Unpublished)