Analysis of the vibration effects on combined cycle power plants' mechanical parts /Оценка воздействия вибрации на механическое оборудование электростанций комбинированного цикла тема диссертации и автореферата по ВАК РФ 00.00.00, кандидат наук Аль-Текрити Ватбан Халид Фахми

• Аль-Текрити Ватбан Халид Фахми
• кандидат наук
• 2025

Оглавление диссертации кандидат наук Аль-Текрити Ватбан Халид Фахми

TABLE OF CONTENTS

ABSTRACT

CHAPTER 1. INTRODUCTION

1.1 Background

1.2 Significance of the Study - Motivations

1.3 Research Objectives

1.4 Research Questions/Hypotheses

1.5 Methodology

1.6 Scope and Limitations of the Study

1.7 Introducing the thesis structure

1.8 Achievements of this research

CHAPTER 2. LITERATURE REVIEW

2.1 Introduction

2.2 How does a Combined-Cycle Gas Steam Turbine Power Plant Works?

2.3 Vibration Induced by Mechanical Failure

2.3.1. Unbalancing & Misalignment

2.3.2. Critical speed

2.3.3. Rubbing

2.3.4. Steam flow fluctuation

2.3.5. Shorted-Turn

2.4 Summary

CHAPTER 3. VIBRATION SIGNALS STRUCTURAL ANALYSIS AND MONITORING SYSTEMS

3.1 Introduction

3.2 Vibration Signals

3.1.1 Vibration Kinematics

3.2.2 Vibration Analysis by Using Fourier Transform

3.3 Instrumentation and Measurement

3.3.1 Eddy Current Proximity Transducers

3.3.2 Accelerometer Sensors

3.3.3 Strain Gauge

3.3.4 Blade tip timing

3.3.5 Laser Doppler vibrometer

3.3.6 Velocity sensor

3.4. Summary

CHAPTER 4. DATA COLLECTION AND PREPARATION

4.1 Power plant information

4.2 Turbine gas information

4.3 Monitoring & Data Collection Systems

4.4 Summary

CHAPTER 5. ADVANCED ANOMALY DETECTION BY USING DEEP LEARNING TECHNIQUES

5.1 Introduction

5.2 Data Preprocessing

5.2.1 Data Augmentation

5.3 Design DLSTM-Autoencoder model

5.3.1 Reconstruction Autoencoder Designed Structure

5.3.2 Deep LSTM Architecture as novel proposed methodology

5.4 Check the efficiency of proposed new methodology

5.5 Summary

CHAPTER 6. CONCLUSIONS AND FUTURE WORK

6.1 Conclusions

6.2 Future Work

LIST OF ABBREVIATIONS AND SYMBOLS

REFERENCES

LIST OF FIGURES

APPENDIX I

APPENDIX

APPENDIX

180

Введение диссертации (часть автореферата) на тему «Analysis of the vibration effects on combined cycle power plants' mechanical parts /Оценка воздействия вибрации на механическое оборудование электростанций комбинированного цикла»

ABSTRACT

The fast-growing field of industrial maintenance is experiencing a paradigm shift with the integration of advanced Machine Learning (ML) techniques, particularly for anomaly detection in complex machinery like gas turbines. This thesis presents a novel approach by developing and implementing a Deep Long Short-Term Memory (DLSTM) Autoencoder model specifically designed for fault detection in gas turbines, utilizing vibration analysis as the primary diagnostic tool. The research addresses the critical need for predictive maintenance in power plants, emphasizing the significance of early anomaly detection to prevent catastrophic failures and optimize operational efficiency. This method effectively distills essential features from complex time-series vibration data, enabling the precise reconstruction of both general and sudden patterns. A significant contribution of this research is the use of a DLSTM model. This model is instrumental in predicting fault occurrences, marking a substantial advancement in anomaly detection methodologies. The thesis demonstrates the model's efficacy in handling high-dimensional time-series data, a common challenge in industrial settings. To address the scarcity of labeled data, particularly anomalous data, the research employs data augmentation techniques like symmetrical flipping and jittering. This strategy enriches the dataset, enhancing the model's learning and predictive capabilities. The study also explores the optimal configuration of the Autoencoder and DLSTM model, utilizing various machine learning metrics to evaluate performance. The research findings underscore the effectiveness of the DLSTM-Autoencoder model in anomaly detection, show improved accuracy and reliability. The model's success in identifying onset anomalies aligns with real-world operational scenarios, validating its practical applicability in industrial maintenance.

Keywords: Industrial maintenance, Machine learning (ML), Anomaly detection, Gas turbines, Deep Long Short-Term Memory (DLSTM) Autoencoder model.

CHAPTER 1. INTRODUCTION 1.1 Background

The introduction of Combined Cycle Power Plants (CCPPs) has significantly transformed the landscape of power generation, ushering in an era of enhanced efficiency, and reduced environmental impact [1]. These plants leverage the integration of gas and steam turbine technologies to achieve impressive thermal efficiencies, surpassing those of traditional fossil fuel-based power generation methods. In addition to meeting the growing global energy demands, CCPPs adhere to stringent emission standards, contributing to a more sustainable approach to energy production [2].

However, as CCPPs take center stage in energy production efficiency, they also face a unique set of challenges. One of the primary concerns is the mechanical stress experienced by their components, particularly mechanical vibrations. Left unaddressed, these vibrations can result in wear and fatigue, jeopardizing both the safety and efficiency of the plant and leading to costly downtime [3].

Conventional vibration monitoring methods in power plants rely on routine maintenance schedules and the deployment of vibration sensors designed to detect abnormal conditions. Nonetheless, with the advent of advanced computational techniques, Machine Learning (ML) offers a transformative approach to vibration analysis, potentially revolutionizing how we handle mechanical stress in CCPPs [4]. ML methods, specifically those tailored for anomaly detection, have gained recognition for their ability to discern patterns and predict equipment failures before they manifest. Anomaly detection algorithms excel at identifying unusual patterns that may signal potential issues by drawing insights from historical and real-time operational data [5]. The integration of these methods into CCPPs promises to augment traditional monitoring systems by providing early warnings, enabling proactive measures to prevent failures.

This thesis embarks on an exploration of the application of ML algorithms for the detection and prediction of abnormal vibration patterns within the mechanical components

of CCPPs. By fusing data-driven models with conventional vibration analysis, this research aims to establish a predictive maintenance framework that enhances operational reliability and efficiency. Through a comprehensive analysis of historical vibration data and real-time monitoring inputs, ML models can potentially identify subtle changes in vibration signatures that precede mechanical failures, offering a fresh perspective on predictive maintenance.

As we delve into the intersection of ML and mechanical vibration analysis, this work will provide insights into the process of developing and validating ML models tailored for anomaly detection in CCPPs. It will explore the selection of relevant features, data preprocessing techniques, the choice of algorithmic approaches suitable for time-series data, and the challenges associated with deploying these models in operational settings. Ultimately, the study aspires to affirm the viability of ML as an indispensable tool in the operational management of CCPPs, safeguarding the integrity and resilience of these critical components within our energy infrastructure.

1.2 Significance of the Study - Motivations

The significance of this study encompasses multiple dimensions, encompassing advancements in engineering practices, contributions to the field of predictive maintenance, and the enhancement of safety measures in energy production, all of which converge at the intersection of machine learning and the analysis and prediction of mechanical vibrations within combined cycle power plants.

Traditional approaches to vibration analysis within CCPPs have predominantly followed a reactive or, at best, a periodic inspection-based paradigm, which often falls short in capturing the transient dynamics of mechanical failures [6]. In contrast, this research seeks to harness the power of ML to propel the industry towards more proactive and predictive strategies. The significance of this transition extends beyond operational efficiency; it signifies a profound shift in methodology, ushering in a data-centric approach to vibration analysis that has the potential to set a new industry standard. The incorporation of ML

algorithms into predictive maintenance programs represents a significant departure from traditional time-based maintenance schedules. By detecting anomalies in vibration data, ML models can forecast potential failures, enabling timely interventions [6]. This study aims to provide empirical evidence and a practical framework to facilitate this transition, which promises substantial cost savings, optimization of maintenance operations, and increased plant availability.

Operational reliability is paramount for CCPPs, given their pivotal role in meeting the baseload of electricity supply. Failures induced by vibrations can result in unplanned outages, impacting the stability of the power grid. By integrating ML models into vibration analysis, this study directly enhances the reliability and efficiency of CCPPs, ensuring uninterrupted power generation to meet the demand. Excessive vibrations pose safety risks to plant personnel and the environment, potentially leading to catastrophic failures. The application of ML for early detection of potential issues is thus a vital safety measure. Furthermore, by averting failures and optimizing maintenance, this study contributes to environmental conservation by preventing emergency situations that could result in environmental hazards. The economic implications of improving vibration analysis are substantial. Proactive detection and maintenance strategies enabled by ML can significantly reduce the costs associated with downtime, repairs, and replacements of critical mechanical components in CCPPs. The outcomes of this study have the potential to drive down operational costs, rendering power generation more cost-effective and sustainable. On an academic level, this research fills a critical void in scholarly literature by offering a comprehensive analysis of the applicability of anomaly detection techniques within the context of CCPPs. It extends the body of knowledge in both the domains of mechanical engineering and computational intelligence, providing insights and methodologies that can be harnessed by future researchers and industry professionals.

In summation, this study promises substantial contributions to existing vibration analysis and maintenance practices in CCPPs. By showcasing the effectiveness of ML

approaches, it underscores the potential of advanced analytics in enhancing the performance and safety of power plants, with far-reaching implications for the energy sector.

1.3 Research Objectives

The central objective of this research is to gain a comprehensive understanding of the practical application of machine learning techniques for the detection and prediction of mechanical vibrations within combined cycle power plants. Figure 1.1 illustrates gas turbine's blade assembly during maintenance at the Kirkuk power station. This endeavor seeks to contribute to the refinement of maintenance practices and the enhancement of plant reliability, operating under a modest framework. The study is guided by a series of interconnected objectives:

Objective 1 (Characterization of Vibration signs in CCPPs): In this phase, the primary focus will be on the identification and documentation of typical vibration patterns associated with the operational states of mechanical components within CCPPs.

Objective 2 (Evaluation of Existing ML-Based Anomaly Detection Methods): To inform the research, a thorough examination of the existing landscape of ML-based anomaly detection methods, specifically those relevant to vibration analysis, will be conducted.

Figure 1.1. The blade assembly of a gas turbine is undergoing maintenance at the Kirkuk

Power Station.

Objective 3 (Development of a Predictive Model for Vibration Anomalies): This phase entails the design and implementation of ML algorithms that have the capacity to identify patterns indicative of potential mechanical faults.

Objective 4 (Design of a Real-Time Vibration Monitoring System): The culmination of this research involves the integration of the ML models, crafted in Objective 3, into recorded data from monitoring system. This system will be engineered to effectively process the continuous influx of streaming vibration data.

By diligently pursuing these interconnected objectives, this research endeavors to establish a comprehensive understanding of how ML techniques can be effectively applied in the realm of vibration analysis within CCPPs. The goal is to create a sophisticated framework for predictive maintenance and operational efficiency improvement, thereby bolstering the reliability and sustainability of CCPPs in a modest and pragmatic manner.

1.4 Research Questions/Hypotheses

The exploration of machine learning methods for vibration analysis in CCPPs is guided by a set of research inquiries and hypotheses. These have been meticulously crafted to direct the research toward its stated objectives and to provide a structured framework for data collection, analysis, and interpretation.

Research Questions: RQ1(The research seeks to unveil the inherent vibration signatures within mechanical components of CCPPs during typical and atypical operational conditions), RQ2(The study endeavors to tailor ML algorithms for the precise detection and prediction of anomalies in vibration data specific to the nuanced context of CCPPs), RQ3(The research acknowledges the potential constraints and complexities associated with applying existing anomaly detection methods to vibration data within the specialized environment of CCPPs), RQ4(The research investigates the potential impact of real-time vibration monitoring systems, empowered by ML algorithms, on the enhancement of predictive maintenance strategies within CCPPs).

Hypotheses: H1(The study posits that distinctive vibration signs linked to mechanical faults in CCPPs exist and can be discerned and classified through the utilization of ML algorithms), H2(The research hypothesis suggests that ML models, when trained on extensive historical vibration data, have the capacity to forecast potential mechanical failures within CCPPs with a level of precision surpassing that of traditional monitoring methodologies), H3(The research contends that the integration of ML algorithms into realtime monitoring systems will lead to a notable reduction in the frequency and duration of unplanned outages, thereby facilitating timely maintenance interventions), H4 (The research hypothesis maintains that the incorporation of ML methodologies for vibration analysis in CCPPs will result in measurable reductions in maintenance expenditures and extend the operational lifespan of mechanical component).

These research questions and hypotheses collectively shape the research's direction and provide a coherent framework for inquiry, with empirical investigations involving the

collection and analysis of vibration data, the application of ML models, and the interpretation of real-world operational data. The validation of these hypotheses will offer valuable insights into the pragmatic advantages, associated challenges, and practical implications of embracing ML in the realm of power generation and vibration analysis.

1.5 Methodology

The methodology adopted for this study is meticulously designed to provide a systematic and comprehensive approach to investigating the impact of vibration on the mechanical components of combined cycle power plants through the application of machine learning techniques. This preliminary overview offers a glimpse into the research approach and the techniques employed to realize the study's objectives.

Data Collection: To underpin the research, vibration data emanating from crucial mechanical components within operational CCPPs will be systematically gathered. This data compilation comprises both historical and real-time measurements, ensuring a holistic foundation for analysis.

Machine Learning Algorithms: A meticulous selection and evaluation of ML algorithms will be undertaken to discern patterns within the vibration data that may serve as indicators of potential mechanical anomalies. The array of algorithms encompasses classification trees, neural networks, and support vector machines, chosen for their track record in adeptly recognizing patterns within extensive datasets.

Anomaly Detection Framework: The research will craft an anomaly detection framework that aligns with the specific characteristics of vibration data derived from CCPPs. This endeavor encompasses feature extraction, data preprocessing, and the application of unsupervised learning techniques. The goal is to identify outliers and irregular data points that may signify incipient issues.

Real-Time Monitoring System Development: The ML algorithms, integral to this research, will be seamlessly integrated into a prototype real-time monitoring system,

thoughtfully engineered for practical deployment within CCPPs. This system will enable continuous and proactive evaluation of the condition of mechanical components.

Model Validation and Testing: The performance of the ML models will be rigorously validated through a combination of retrospective data analysis and prospective testing within a controlled environment. Emphasis will be placed on key metrics such as accuracy, precision, recall, and the F1 score to ascertain the effectiveness of the models.

The cumulative results of these methodological steps hold the promise of significantly advancing predictive maintenance practices, thereby contributing to the optimization of CCPP operations, making them more efficient, reliable, and sustainable.

1.6 Scope and Limitations of the Study

This study is specifically tailored to address the application of machine learning techniques for the analysis of vibration effects on mechanical parts within combined cycle power plants. The scope encompasses several key areas:

• Data Collection: The study will involve collecting vibration data from multiple mechanical components within CCPPs, including but not limited to turbines, compressors, and pumps.

• ML Model Development: A range of ML algorithms will be explored and developed to detect and predict anomalies in the collected vibration data.

• Real-Time Monitoring: The implementation scope includes the development of a realtime vibration monitoring system that employs the developed ML models.

• Performance Evaluation: An evaluation of the performance of the ML model in terms of accuracy and reliability is within scope.

While aiming to be comprehensive, this study acknowledges certain limitations that may influence its breadth and depth:

• Data Availability: The study is contingent on the availability of comprehensive and high-quality vibration data. Limited access to proprietary or sensitive data may restrict the analysis.

• Complexity of Plant Operations: CCPP operations are complex, and the study may not account for every operational variable that affects mechanical vibrations.

• ML Model Generalizability: The developed ML models may not be universally applicable to all CCPPs, as plant-specific conditions can vary significantly.

• Technological Constraints: The real-time application of ML models is subject to the limitations of existing hardware and software within the CCPP infrastructure.

• External Factors: The study will not cover external factors such as regulatory changes, market dynamics, or environmental conditions that could affect CCPP operations.

The recognition of these limitations is crucial for setting realistic expectations for the study's outcomes. By acknowledging these constraints, the research maintains a focused approach that aims to deliver actionable insights within the defined scope, while providing a candid perspective on the challenges encountered in the process.

In addition to the comprehensive scope outlined in this section, it is imperative to acknowledge the significant time and effort invested in data collection and skill development, which are integral components of the research endeavor. The collection of vibration data necessitated several dedicated visits over the course of several months from Kirkuk power station. For instance, the acquisition of data for subsequent papers involved an expenditure of approximately 4 to 5 months. Moreover, in parallel with data collection efforts, attendance in various machine learning courses was undertaken to augment proficiency in code development and machine learning techniques. These additional efforts underscore the commitment to ensuring the robustness and reliability of the study's findings. Through data collection and continuous skill enhancement, the research endeavors to harness the full potential of machine learning techniques for the analysis of vibration effects on mechanical parts within CCPPs.

1.7 Introducing the thesis structure

This thesis is structured to provide a comprehensive examination of the application of machine learning techniques for the analysis of vibration effects on mechanical parts within combined cycle power plants. Each chapter contributes to a deeper understanding of the complexities involved in this field and offers innovative solutions to address them effectively.

Chapter 2 (Literature Review): This chapter provides a foundational understanding of vibration analysis in rotary machinery within power plants. We explore key factors contributing to vibrations and their potential consequences, highlighting the importance of predictive maintenance strategies. Fundamental research questions guiding our study are also formulated.

Chapter 3 (Vibration Signals Analysis) Here, we focus on the practical aspects of vibration analysis and monitoring systems. Investigating the effectiveness of vibration analysis in diagnosing rotary machinery issues, we emphasize its role in preventive maintenance. Various technologies for monitoring vibrations across power plant sections are also explored.

Chapter 4 (Power Plant Overview): This chapter offers a detailed overview of the Kirkuk Gas Power Plant, covering its location, capacity, construction, and operational history. Specific details include turbine gas information, historical vibration-related failures, monitoring protocols, maintenance procedures, and details of experimental data collection.

Chapter 5 (Anomaly Detection in Vibration Analysis): This pivotal chapter examines time series data analysis techniques for anomaly detection, focusing on vibrations. We discuss the importance of utilizing machine learning algorithms to detect deviations in vibration patterns, signaling potential failures.

Chapter 6 (Conclusions and Future Directions): In the final chapter, we draw conclusions from our research and outline future avenues. We emphasize the importance of

proactive maintenance, discuss the effectiveness of our anomaly detection approach, and propose directions for further advancements in the field.

By following this structured approach, the thesis aims to contribute valuable insights into the optimization of maintenance practices in combined cycle power plants through the integration of machine learning techniques and vibration analysis.

1.8 Achievements of this research

In the following, the published and under review papers which are prepared based on this research are presented.

№ Title of the scientific paper Type of work Output data Numb er of pages Authors

Scopus-indexed Journal

1. Advancements in gas turbine fault detection: a machine learning approach based on the temporal convolutional network-Autoencoder model Electro nic Applied Sciences. - 2024. - V 14(11). - P. 4551 doi.org/10.3390/ap p14114551 Scopus 17 Reza Kashyzadeh K., Ghorbani S.

2. Fault detection in the gas turbine of the Kirkuk Power Plant: an anomaly detection approach using DLSTM-Autoencoder] Electro nic Engineering Failure Analysis. -2024. - V 160(3). -P.108213 doi.org /10.1016/j.engfaila nal.2024.108213 Scopus 17 Reza Kashyzadeh K., Ghorbani S.

3. Smart maintenance strategies in combined cycle power plant] Electro nic International Journal of Applied Research in Mechanical Engineering (JCARME). -2024. - V 14(1). - P. 35-46. doi.org/10.22061/j carme.2024.10797. 2415 Scopus 12 Reza Kashyzadeh K., Ghorbani S.

4. Enhanced autoregressive integrated moving average model for anomaly detection in power plant operations Electro nic International Journal of Engineering. -2024. - V 37(08). - P. 1691-1699 doi.org/10.5829/ije .2024.37.08b.19 Scopus 9 Reza Kashyzadeh K., Ghorbani S.

5. A comprehensive review on mechanical failures cause vibration in the gas turbine of combined cycle power plants Electro nic Engineering Failure Analysis. -2022. - V. 134. - P. 106094. doi.org/10.1016/j.e ngfailanal .2022.10 6094 Scopus 15 Reza Kashyzadeh K., Ghorbani S.

BAK Journal

6. Comparative Performance of Machine Learning Classifiers in Detecting Vibration Anomalies in Industrial Power Systems Electro nic RUDN Journal of Engineering Research. 2025. -V 26(3). - P. 274288 15 Reza Kashyzadeh K., Ghorbani S.

Patents

7. An improved computer code-based ARIMA model for early detection of power plant anomalies Certificate of State Registration of a Computer Program № 2025614457 Reza Kashyzadeh K.

Published 21.02.2025 Copyright holder: RUDN University

8. Improving the Machine Learning Framework Using a Temporal Convolutional Network Autoencoder Model: To Improve Fault Detection in Industrial Equipment Certificate of State Registration of a Computer Program № 2025614226 Published 20.02.2025 Copyright holder: RUDN University Reza Kashyzadeh K.

9. Intelligent, High-Precision Anomaly Detection in an Online Turbine Monitoring System Power Plants Certificate of State Registration of a Computer Program № 2024663703 Published 10.06.2024 Copyright holder: RUDN University Reza Kashyzadeh K., Ghorbani S.

10. Development of a computer program for detecting vibration anomalies in rotating equipment based on various practical algorithms Certificate of State Registration of a Computer Program № 2024663540 Published 07.06.2024 Copyright holder: RUDN University Reza Kashyzadeh K., Ghorbani S.

Other-indexed Journal

11. Accuracy assessment of the One-Class SVM technique in identifying abnormalities in the vibration monitoring data of a gas turbine Printing Journal of Mechanical Engineering and Vibration. 2022. -V 13(2). - P. 7-14 8 Reza Kashyzadeh K., Ghorbani S.

Conference

12. Industrial vibration detection techniques for enhanced monitoring and maintenance of combined cycle power plants Electro nic Engineering Systems, RUDN university, Moscow, 2023, p. 167-178. 10 Reza Kashyzadeh K., Ghorbani S.

13. Enhancing fault detection in gas turbines using machine learning models: A case study on Kirkuk gas power plant Electro nic The 4th International Conference on Artificial Intelligence and its Future Prospects in electrical, computer, mechanical and telecommunication engineering sciences, Iran, December 2024 6 Reza Kashyzadeh K., Ghorbani S.

14. Anomaly detection in gas turbines using DLSTM-Autoencoder with data augmentation Electro nic 3rd International Conference for Mechanical and Aerospace Engineers Students, Iran, December 2024 7 Reza Kashyzadeh K., Ghorbani S.

Заключение диссертации по теме «Другие cпециальности», Аль-Текрити Ватбан Халид Фахми

CHAPTER 6. CONCLUSIONS AND FUTURE WORK

6.1 Conclusions

The research underscores the importance of developing proactive maintenance schedules for different parts of a power plant. Such measures are crucial in preventing serious system damage and effectively reducing maintenance time and costs. This proactive approach is vital for the sustainable and efficient operation of power plants. Furthermore, the study provides a deeper understanding of the system's health state, highlighting the gradual nature of problem occurrence or anomalous behavior in power plants. This insight is critical for timely intervention and prevention of larger-scale operational failures.

A key contribution of this research is the introduction of a method focused on reducing the volume of input data by leveraging the latent layer output of an Autoencoder. This method is effective in distilling essential features from complex time series data. The Autoencoder model is adept at reconstructing both general and sudden patterns in time series data, achieved by retaining relevant features and learning the encoded representation of the time series.

Additionally, the research employs an anomaly detection approach using the encoder's latent layer output to train a Deep Long Short-Term Memory (DLSTM) model. This model, comprising three hidden layers, is instrumental in predicting fault occurrences, enhancing the precision of fault assessments, especially in gas turbines.

This research makes substantial progress in applying deep learning in industrial environments. This is especially relevant in contexts where data limitations are a known constraint. By utilizing detailed time series data, the research demonstrates the capability of sophisticated deep learning models to identify subtle yet critical patterns that are essential for distinguishing between normal operations and potential anomalies. This aspect is crucial for maintaining operational efficiency and safety in industrial machinery.

A pivotal aspect of the research is the application of vibration analysis in conjunction with a DLSTM-Autoencoder model for gas turbine fault detection. Vibration analysis is a

widely recognized method in industrial settings for the detection and diagnosis of defects in rotary machines. It is effective for identifying, locating, and diagnosing common problems in various parts of rotary machinery. The research shows that using this technique, along with the innovative DLSTM-Autoencoder model, addresses a significant portion of issues related to rotary machines. Moreover, the advancements brought forth in this thesis are key in enhancing predictive maintenance strategies. They contribute to more accurate anomaly detection, which in turn supports the development of predictive maintenance strategies. These strategies enhance the reliability, safety, and efficiency of industrial machinery operations. The thesis highlights how these advancements in machine learning can transform traditional maintenance approaches into more proactive, data-driven strategies.

The anomaly detection approach represents a comprehensive strategy for employing deep learning to effectively detect anomalies, particularly in the context of the Kirkuk Gas Power Plant. The research has made significant strides in training a Deep Learning Long Short-Term Memory (DLSTM) model with three hidden layers. This model, utilizing the output from the Autoencoder's latent layer, is innovatively designed as a predictive tool to detect anomalies within the dataset. The model's architecture is pivotal in enhancing the predictive maintenance capabilities at the power plant, enabling a more proactive and efficient approach to fault detection and maintenance scheduling. A critical component of this approach is the selection of the appropriate deep neural network technique for training the model in anomaly detection. This process begins with a comprehensive understanding of the input data's structure. The structure of the input data is typically categorized into two types: sequential and non-sequential. Sequential data is characterized by the importance of the order of the dataset's elements. This understanding is crucial as it dictates the choice of neural network architecture and influences the overall approach to modeling.

The anomaly detection model developed in this research is a synergistic combination of an Autoencoder for feature reduction and a Deep Long Short-Term Memory (DLSTM) model for prediction. This integrated approach is meticulously designed to effectively

identify and predict anomalies in time series data, with a specific focus on vibration data from industrial machinery.

The process begins with the utilization of a three-layer Autoencoder for automatic feature selection and encoding. The Autoencoder is particularly adept at learning from multiple time series, each exhibiting different patterns. Its key function in the research is to compress and encode the input data, effectively reducing its complexity while retaining the most significant features for anomaly detection.

The encoded data from the Autoencoder is then fed into the DLSTM model, which is responsible for the predictive aspect of anomaly detection. The DLSTM model's architecture, comprising multiple LSTM layers, is tailored to process the encoded features and predict future states of the time series data. This prediction is based on the learned patterns and correlations within the data, allowing the model to anticipate potential anomalies.

The effectiveness of the combined Autoencoder and Deep Long Short-Term Memory (DLSTM) model is thoroughly examined. This model was developed to enhance predictive maintenance in power plant systems, focusing on identifying and predicting anomalous behavior that could indicate potential system issues.

The model configuration played a crucial role in its performance. The Autoencoder was effectively used for feature reduction, compressing complex time series data from the Kirkuk gas turbine power plant into a more manageable form. The DLSTM, with its three hidden layers, leveraged these reduced features to accurately predict potential faults. The incorporation of synthetic data alongside the original dataset significantly improved the training accuracy of the model, demonstrating the value of a comprehensive dataset in enhancing learning capabilities.

The model's ability to detect anomalies was clearly demonstrated through various visual representations in the thesis. These depictions showed the model's effectiveness in identifying the onset of abnormal vibrational patterns, highlighting its practical application in real-world scenarios. One notable instance of successful anomaly detection coincided with

a period when the gas turbine was shut down, underscoring the model's accuracy and applicability.

Optimization of the model's anomaly detection rates was a key aspect of the research, ensuring a balance between sensitivity to actual anomalies and minimizing false positives. This optimization was crucial for the model's reliability in practical applications. Additionally, the validation process, where model predictions were compared with real events at the power plant, provided a solid foundation for its accuracy.

The conclusion of the research underlines the importance of developing maintenance schedules for different parts of a power plant, emphasizing how such proactive measures can prevent serious system damage and effectively reduce maintenance time and costs. Moreover, the approach proposed in this study provides a deeper understanding of the system's health state, highlighting the gradual nature of problem occurrence, or anomalous behavior, in power plants. This research introduces a method focused on reducing the volume of input data by leveraging the latent layer output of an Autoencoder. This method is effective in distilling the essential features from complex time series data. The Autoencoder model is adept at reconstructing both general and sudden patterns in time series data, achieved by retaining relevant features and learning the encoded representation of the time series. Additionally, the research utilizes an anomaly detection approach, employing the encoder's latent layer output to train a deep Long Short-Term Memory (LSTM) model. This model, comprising three hidden layers, is instrumental in predicting fault occurrences. By inputting the prediction model output into a threshold approach, the possibility of faults, especially in gas turbines, can be assessed with higher precision. The research used data provided by the Kirkuk gas turbine power plant, with the CA 202 piezoelectric accelerometer as the primary sensor for data collection. The results indicated that the best reconstruction of multi-features in the Autoencoder was achieved using the Adam optimizer, configured with 512 neurons in the first hidden layer, 256 in the second, and a dropout rate of 0.1. The developed DLSTM model, configured using the Adam optimizer with 200, 100, and 200 neurons for the first, second, and third hidden layers respectively, along with a dropout rate

of 0.1 and a linear activation function, showed promising results. According to the output of the prediction model, the proposed approach demonstrated its potential in predicting anomalous behavior in the power plant.

6.2 Future Work

This research has laid a solid foundation for predictive maintenance in power plants, but there are several avenues for future work that can further enhance the field. The following areas offer promising directions for future research and development:

1. Integration of Real-Time Data: While this research leveraged historical data, the incorporation of real-time data streams can provide more dynamic insights into the health of power plant systems. Future work can explore methods for seamlessly integrating real-time sensor data into predictive maintenance models to enable timely responses to changing conditions.

2. Enhanced Synthetic Data Generation: The use of synthetic data for anomaly detection was a crucial aspect of this research. Future endeavors can focus on refining and expanding synthetic data generation techniques to create more diverse and challenging datasets for model training, thus improving the robustness of predictive models.

3. Automated Threshold Determination: The research discussed the use of a threshold approach for anomaly detection. Future work can explore the development of automated threshold determination algorithms that adapt to changing system conditions, reducing the need for manual threshold setting.

4. Multimodal Data Fusion: Power plants generate diverse data types, including vibration, temperature, pressure, and more. Future research can investigate techniques for fusing information from multiple data sources, enabling a holistic view of system health and improving anomaly detection accuracy.

7. Human-Machine Collaboration: Power plant maintenance involves human expertise. Future work can investigate how predictive maintenance models can be integrated into

decision support systems that facilitate collaboration between machine predictions and human domain knowledge.

8. Generalization to Other Industries: While this research primarily focused on power plants, the methodologies developed here can be adapted to other industries with complex machinery, such as manufacturing, aviation, and transportation. Future research can explore the applicability of these techniques in diverse industrial contexts.

Список литературы диссертационного исследования кандидат наук Аль-Текрити Ватбан Халид Фахми, 2025 год

REFERENCES

1. Boyce M. P. Combined cycle power plants // Combined cycle systems for near-zero emission power generation. - Oxford : Woodhead Publishing, 2012. - P. 1-43.

2. Bierdel T. Life Cycle Value for combined cycle power plants. - Erlangen : Siemens AG, 2013.

3. Meher-Homji C. B., Chaker M. A., Motiwala H. M. Gas Turbine Performance Deterioration // Proceedings of the 30th turbomachinery symposium. - 2001. - Texas A&M University. Turbomachinery Laboratories.

4. Hundi P., Shahsavari R. Comparative studies among machine learning models for performance estimation and health monitoring of thermal power plants // Applied Energy. - 2020. - Vol. 265. - P. 114775.

5. Schmidl S., Wenig P., Papenbrock T. Anomaly detection in time series: a comprehensive evaluation // Proceedings of the VLDB Endowment. - 2022. - Vol. 15, № 9. - P. 1779-1797.

6. Zhao Y., Volovoi V., Waters M., Mavris D. A sequential approach for gas turbine power plant preventative maintenance scheduling // Journal of Aircraft. - 2006. - Vol. 43, № 3. - P. 796-805.

7. Ma J., Jiang J. Applications of fault detection and diagnosis methods in nuclear power plants: A review // Progress in nuclear energy. - 2011. - Vol. 53, № 3. - P. 255-266.

8. Bany Ata A., Alobaid F., Heinze C., Almoslh A., Sanfeliu A., Epple B. Comparison and validation of three process simulation programs during warm start-up procedure of a combined cycle power plant // Energy Convers. Manag. - 2020. - Vol. 207. - P. 112547.

9. Singh A. Past, Present and Future of Combined Cycle Power Plants: From an A/E's Perspective // Proceedings of the ASME Turbo Expo 2001: Power for Land, Sea, and Air. Heat Transfer; Electric Power; Industrial and Cogeneration. New Orleans, Louisiana, USA. June 4-7, 2001. - Vol. 3. - P. V003T02A021.

10.Kanoglu M., Isik S. K., Abusoglu A. Performance characteristics of a diesel engine power plant // Energy Convers. Manag. - 2005. - Vol. 46, № 11-12. - P. 1692-1702.

11.Hossain S., Bari S. Additional Power Generation from the Exhaust Gas of Diesel Engine by Bottoming Rankine Cycle // SAE Tech. Pap. - 2013. - № 2013-01-1639.

12.Weaver C. S., Turner S. H. Dual fuel natural gas/diesel engines: technology, performance, and emissions // SAE Tech. Pap. - 1994. - № 940548.

13.Reinhardt A. K., Kadambi J. R., Quinn R. D. Laser vibrometry measurements of rotating blade vibrations // ASME. J. Eng. Gas Turbines Power. - 1995. - Vol. 117, № 3. - P. 484-488.

14.Halvorsen B. J. Power Plant with CO2 Capture based on Absorption Part-load Performance : guc. ... - Trondheim : Institutt for energi- og prosessteknikk, 2012.

15.Mathioudakis K., Loukis E., Papailiou K. D. Casing vibration and gas turbine operating conditions // ASME J. Eng. Gas Turbines Power. - 1990. - Vol. 112, № 4. - p. 478-485.

16.He W. F., Xu L. N., Han D., Gao L., Yue C., Pu W. H. Thermodynamic investigation of waste heat driven desalination unit based on humidification dehumidification (HDH) processes // Appl. Therm. Eng. - 2016. - Vol. 100. - P. 315-324.

17.Kehlhofer R., Rukes B., Hannemann F., Stirnimann F. Combined-cycle gas and steam turbine power plants. - Tulsa : PennWell Books, LLC, 2009.

18.Cavalcanti E. J. C., Lima M. S. R., DeSouza G. F. Comparison of carbon capture system and concentrated solar power in natural gas combined cycle: exergetic and exergo environmental analyses // Renew. Energy. - 2020. - Vol. 156. - P. 1336-1347.

19.Islam M. M., Hasanuzzaman M., Pandey A. K., Rahim N. A. Modern energy conversion technologies // Energy for Sustainable Development / ed. by M. D. Hasanuzzaman, A. R. Nasrudin. - Academic Press, 2020. - P. 19-39.

20.Borzsony P., Sherikar S. V. Advances in desuperheating technology for reliable performance of combined cycle power plants (CCPP) // ASME Power Conference. Chicago, Illinois, USA. April 5-7, 2005. - P. 53-61.

21.Oriani L., Lombardi C., Paramonov D. Natural gas turbine topping for the iris reactor // Proceedings of the 9th International Conference on Nuclear Engineering, Nice, France, 2001. - P. 1-12.

22.Cho S. K., Kim M., Baik S., Ahn Y., Lee J. I. Investigation of the bottoming cycle for high efficiency combined cycle gas turbine system with supercritical carbon dioxide power cycle // Proceedings of the ASME Turbo Expo 2015: Turbine Technical Conference and Exposition. Volume 9: Oil and Gas Applications; Supercritical CO2 Power Cycles; Wind Energy. Montreal, Quebec, Canada. June 15-19, 2015. - P. V009T36A011.

23.Ibrahim S. M. A., Ibrahim M. M. A., Attia S. I. The impact of climate changes on the thermal performance of a proposed pressurized water reactor: nuclear-power plant // Int. J. Nucl. Energy. - 2014. - Vol. 2014. - P. 1-7.

24.Tumanovskii A. G. Prospects for the development of coal-steam plants in Russia // Therm. Eng. - 2017. - Vol. 64, № 6. - P. 399-407.

25.Kadak A. C. A future for nuclear energy: pebble bed reactors // Int. J. Crit. Infrastruct. - 2005. - Vol. 1, № 4. - P. 330-345.

26.Suppes G. J., Storvick T. S. Production of electricity // Sustainable Nuclear Power / ed. by J. S. Galen, S. S. Truman. - Academic Press, 2007. - P. 185-200.

27.Storm K. Combined cycle power plant (1x1) labor estimate // Industrial Construction Estimating Manual / ed. by K. Storm. - Gulf Professional Publishing, 2020. - P. 95159.

28.Choi B. S., Ahn J. H., Lee J. H., Kim T. S. A comparative performance analysis of a solid oxide fuel cell and gas turbine combined cycles with carbon capture technologies // J. Mech. Sci. Technol. - 2019. - Vol. 33, № 3. - P. 1463-1475.

29.Kang C. W., Golay M. W. An integrated method for comprehensive sensor network development in complex power plant systems // Reliab. Eng. Syst. - 2000. - Vol. 67, № 1. - P. 17-27.

30.Jianfu H., Bryon J. W., Ross A. A. An investigation of fatigue failures of turbine blades in a gas turbine engine by mechanical analysis // Eng. Fail. Anal. - 2002. -Vol. 9, № 2. - P. 201-211.

31.Engvik L. Gas turbines. - 2021. - URL: https://energyfaculty.com/gas-turbines-and-other-heat-machines/ (дата обращения: 20.05.2024).

32.Edwards S. F., Lees A. W., Friswell M. I. The influence of torsion on rotor/stator contact in rotating machinery // J. Sound Vib. - 1999. - Vol. 225, № 4. - P. 767-778.

33.Eshleman R. L., Eubanks R. A. On the critical speeds of a continuous rotor // ASME J. Eng. Ind. - 1969. - Vol. 91, № 4. - P. 180-188.

34.Mitchell L. D., Mellen D. M. Torsional-Lateral Coupling in a Geared, High Speed Rotor System // Design Engineering Technical Conference, Washington D.C., September 17-19, Design Engineering Division, ASME Paper 75-DET-75, 1975.

35.Shatokhin V. F. Studying the development of asynchronous rolling of the rotor over the stator with the turbine unit protection systems having different response times // Therm. Eng. - 2014. - Vol. 61, № 7. - P. 514-523.

36.Sinha J. K. Vibration analysis, Instruments, and Signal Processing. - 1st ed. - Boca Raton : CRC Press, 2015.

37.Azeem N., Xiaoqing Y. Experimental study on the condition monitoring of shaft unbalance by using vibrations spectrum and phase analysis // Cond. Monit. Diagn. -2018. - P. 1-6.

38.Kumar B., Diwakar G., Satynarayana M. R. S. Determination of unbalance in rotating machine using vibration signature analysis // IJMER. - 2012. - Vol. 2, № 5. - P. 34153421.

39.Kim D., Kim Y., Jung J. H., Sohn S. Fault diagnosis method based on feature residual values for industrial rotor machines // KEPCO J. Electric Power Energy. - 2018. -Vol. 4, № 2. - P. 89-99.

40.Smalley A. J. The dynamic response of rotors to rubs during startup // SME J. Vib. Acoust. Stress Reliab. - 1989. - Vol. 111, № 3. - P. 226-233.

41.Rajendran R. Gas turbine coatings - an overview // Eng. Fail. Anal. - 2012. - Vol. 26. - P. 355-369.

42.Sudhakar G. N. D. S., Sekhar A. S. Coupling misalignment in rotating machines: modelling, effects and monitoring // Noise Vib. Worldw. - 2009. - Vol. 40, № 1. - P. 17-39.

43.Gaberson H. A. Rotating machinery energy loss due to misalignment // Proceedings of the 31st Intersociety Energy Conversion Engineering Conference. - 1996. - Vol. 3. - P. 1809-1812.

44.Sekhar A. S., Prabhu B. S. Effects of coupling misalignment on vibrations of rotating machinery // J. Sound Vib. - 1995. - Vol. 185, № 4. - P. 655-671.

45.Bloch H. P. Less costly turbo equipment uprates through optimized coupling selection // Texas A&M University. Gas Turbine Laboratories. - 1975. - P. 149-152.

46.Gibbons C. B. Coupling misalignment forces // Proceedings of the 5th Turbomachinery Symposium. - 1976. - P. 111-116.

47.Saavedra P. N., Ramirez D. E. Vibration analysis of rotors for the identification of shaft misalignment - Part I: Theoretical analysis // Proc. Inst. Mech Eng. Part C: J. Mech. Eng. Sci. - 2004. - Vol. 218. - P. 971-985.

48.Saavedra P. N., Ramirez D. E. Vibration analysis of rotors for the identification of shaft misalignment - Part II: Experimental validation // Proc. Inst. Mech. Eng. Part C: J. Mech. Eng. Sci. - 2004. - Vol. 218. - P. 987-1099.

49.Wang N., Jiang D. Vibration response characteristics of a dual-rotor with unbalance-misalignment coupling faults: Theoretical analysis and experimental study // Mechanism and Machine Theory. - 2018. - Vol. 125. - P. 207-219.

50.Patel T. H., Darpe A. K. Experimental investigations on vibration response of misaligned rotors // Mech. Syst. Sig. Process. - 2009. - Vol. 23, №2 7. - P. 2236-2252.

51.Lees A. W. Misalignment in rigidly coupled rotors // J. Sound Vibr. - 2007. - Vol. 305, № 1-2. - P. 261-271.

52.Jang J. Y., Khonsari M. M. On the behavior of misaligned journal bearings based on mass-conservative thermo hydrodynamic analysis // ASME J. Tribol. - 2010. - Vol. 132, № 1. - P. 011702.

53.Jang J. Y., Khonsari M. M. On the characteristics of misaligned journal bearings // Lubricants. - 2015. - Vol. 3, № 1. - P. 27-53.

54.Pigott R. J. S. Bearings and lubrication-bearing troubles traceable to design can be avoided by engineering study // Mech. Eng. (Am. Soc. Mech. Eng.). - 1942. - Vol. 64. - P. 259-269.

55.Reducing Cycling Damage to Combined Cycle Steam Turbines. - 2020.

56.Lee Y. J., Ju Y. H. An assessment of insulation condition for generator rotor windings // International Conference on Condition Monitoring and Diagnosis. - 2008. - P. 543545.

57.Pennacchi P., Vania A. Shaft crack detection in a steam turbine: experimental evidences and model-based simulations // Proc. ISMA. - 2010. - P. 1055-1070.

58.Rao J. S., Chang J. R., Shiau T. N. Coupled bending-torsion vibration of geared rotors // ASME Design Eng. Tech. Conf. - 1995. - Vol. 84, № 2. - P. 977-989.

59.Saraci J. The Effects of high vibration on the steam turbo-generator machine of the unit B1 339 MW in thermal power plant "Kosovo B" : gnc. ... - Rochester : Rochester Institute of Technology, 2009.

60.Sinha J. K., Hahn W., Elbhbah K., Tasker G., Ullah I. Vibration investigation for low pressure turbine last stage blade failure in steam turbines of a power plant // Proceedings of the ASME Turbo Expo 2012: Turbine Technical Conference and Exposition. Volume 7: Structures and Dynamics, Parts A and B. Copenhagen, Denmark. June 11-15, 2012. - P. 363-371.

61.Bachschmid N., Pennacchi P., Vania A. Thermally induced vibrations due to rub in real rotors // J. Sound Vib. - 2007. - Vol. 299, № 4-5. - P. 683-719.

62.Todorovic P., Jeremic B., Macuzic I., Brkovic A., Proso U. Vibration analysis of cracked rotor during run-up // Tribol. Ind. - 2008. - Vol. 30, № 1-2. - P. 55-62.

63.Medrea C., Papageorgiou D. G., Bravos H., Chicinas I. Failure analysis of a fan blade holding assemblement, installed on the cooling tower of a power plant // Eng. Fail. Anal. - 2021. - Vol. 127. - P. 105505. 64.Sanjeev S., Pandey J. P., Ranjit S. S., Gaurav K. G., Modi O. P. Coupled mechanical, metallurgical and FEM based failure investigation of steam turbine blade // Eng. Fail. Anal. - 2015. - Vol. 52. - P. 35-44.

65.Wanga K. S., Guo D., Heyns P. S. The application of order tracking for vibration analysis of a varying speed rotor with a propagating transverse crack // Eng. Fail. Anal.

- 2012. - Vol. 21. - P. 91-101.

66.Loewy R. G., Piarulli V. J. Dynamics of rotating shafts. - Vol. 4. - Shock and Vibration Information Center, 1970.

67.Zheng G. T. Vibration of a rotor system with a switching crack and detection of the crack // J. Eng. Gas Turbines Power. - 1998. - Vol. 120, № 1. - P. 149-154.

68.Lee A. S., Jin W. H., Hoon C. D. Coupled lateral and torsional vibration characteristics of a speed-increasing geared rotor-bearing system // J. Sound Vib. - 2003. - Vol. 263, № 4. - P. 725-742.

69.Gupta K., Gupta K. D., Athre K. Unbalance response of a dual rotor system: theory and experiment // J. Vib. Acoust. - 1993. - Vol. 115, № 4. - P. 427-435.

70.Hussein I. M., Mohsin A. A., Amjad A. H. Experimental analysis of cracked turbine rotor shaft using vibration measurements // JMERD. - 2020. - Vol. 43, № 2. - P. 294304.

71.Hyoung L., Young C. K., Jong S. K. Development of 29MW gas turbine coupling with maximum torque of 85 KNM // IJARET. - 2020. - Vol. 11, № 5. - P. 287-297.

72.Kaneko Y., Kanki H., Kawashita R. Steam turbine rotor design and rotor dynamics analysis // Advances in Steam Turbines for Modern Power Plants / ed. by T. Tadashi.

- Woodhead Publishing, 2017. - P. 127-151.

73.Gilberto F. M. S., Fernando J. G. C., Leonan S. G., Carmen E. P. R. Combined-cycle gas and steam turbine power plant reliability analysis // Thermal Power Plant Performance Analysis / ed. by G. de Souza. - Springer, London, 2012. - P. 221-247.

74.Kannan P., Amirthagadeswaran K. S., Christopher T., Nageswara B. R. Failures of high-temperature critical components in combined cycle power plants // J. Fail Anal. Preven. - 2013. - Vol. 13, № 4. - P. 409-419.

75.Thomas A., Peter F., Seigou F., Hakan H., Stefan K. Maintenance and overhaul of steam turbines // International Association of Engineering Insurers 38th Annual Conference, Moscow, vol. 29, 2005. - P. 1-46.

76.Historical loss experiences in the global power industry. - Marsh risk, Management research, 2014.

77.Pennacchi P., Vania A., Steven C., Roberto R. Spiral vibrations in power units: modeling and experimental evidences // Proceedings of ISMA-2010, International Conference on Noise and Vibration Engineering. Leuven, Belgium, 2010. - P. 29052519.

78.Saeed B., Mohsen N., Jalal F. S., Amirhassan A. Annihilation of non-stationary vibration of a gas turbine rotor system under rub-impact effect using a nonlinear absorber // Mech. Mach. Theory. - 2019. - Vol. 139. - P. 379-406.

79.Al Yahyai M., Mba D. Rotor dynamic response of a centrifugal compressor due to liquid carry over: a case study // Eng. Fail. Anal. - 2014. - Vol. 45. - P. 436-448.

80.Feiyun C., Jin C., Guangming D., Kun H. Experimental validation of impact energy model for the rub-impact assessment in a rotor system // Mech. Syst. Sig. Process. -2011. - Vol. 25, № 7. - P. 2549-3258.

81.Petrov E. P. Multiharmonic analysis of nonlinear whole engine dynamics with bladed disc-casing rubbing contacts // Proceedings of the ASME Turbo Expo 2012: Turbine Technical Conference and Exposition. Volume 7: Structures and Dynamics, Parts A and B. Copenhagen, Denmark. June 11-15, 2012. - P. 181-191.

82.Childs D. W., Siddiqui N. A. Rotordynamics involving axial rubbing against a disk // Proceedings of the ASME 2001 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference. Volume 6A: 18th Biennial Conference on Mechanical Vibration and Noise. Pittsburgh, Pennsylvania, USA. September 9-12, 2001. - P. 705-710.

83.Villanueva J. A. B., Aguilar F. J. J., Trujillo E. C., Ramirez R. C., Garcia M. T. Analysis of Steam Turbine Instabilities of a 100 MW Combined Cycle Power Plant // ASME International Mechanical Engineering Congress and Exposition. Dynamic Systems and Control, Parts A and B. Vancouver, British Columbia, Canada, vol. 8, November 12-18, 2010. - P. 11-19.

84.Ehrich F. F. Self-excited vibration // Shock and Vibration Handbook / ed. by Harris C. M. - New York : McGraw-Hill, 1976.

85.Peng W., Yingzheng L. Unsteady flow behavior of a steam turbine control valve in the choked condition: field measurement, detached eddy simulation and acoustic modal analysis // Appl. Therm. Eng. - 2017. - Vol. 117. - P. 725-739.

86.Cun W., Dayi Z., Yanhong M., Zhichao L., Jie H. Theoretical and experimental investigation on the sudden unbalance and rub-impact in rotor system caused by blade off // Mech. Syst. Sig. Process. - 2016. - Vol. 76. - P. 111-135.

87.Hui M., Qianbin Z., Xueyan Z., Qingkai H., Bangchun W. Dynamic characteristics analysis of a rotor-stator system under different rubbing forms // Appl. Math. Model. - 2015. - Vol. 39, № 8. - P. 2392-2408.

88.Jacquet R. G., Torkhani M., Cartraud P., Thouverez F., Nouri B. T., Herran M., Gibert C., Baguet S., Almeida P., Peletan L. Rotor to stator contacts in turbomachines. Review and application // Mech. Syst. Sig. Process. - 2013. - Vol. 40, № 2. - P. 401420.

89.Esmaeil P., Hosein B. Fatigue crack growth simulation in a first stage of compressor blade // Eng. Fail. Anal. - 2014. - Vol. 45. - P. 314-325.

90.Claire M. S. Gas turbines in simple cycle and combined cycle applications // The Gas Turbine Handbook / ed. by S. Lance, K. Hasan, E. Shahrokh, C. P. William. -Morgantown : NETL, 1998. - P. 1-74.

91.Poursaeidi E., Salavatian M. Failure analysis of generator rotor fan blades // Eng. Fail. Anal. - 2007. - Vol. 14, № 5. - P. 851-860.

92.Jim H., Frank K., Stephen K. Elimination of flow-induced instability from steam turbine control valves // Proceedings of the 32nd Turbomachinery Symposium, Texas A&M University. Turbomachinery Laboratories, 2003.

93.Ryo M., Fumio I., Michitsugu M., Kenichi T., Yoshinobu T. CFD simulations and experiments of flow fluctuations around a steam control valve // J. Fluids Eng. - 2007. - Vol. 129, № 1. - P. 48-54.

94.Widel K. E. Governing Valve Vibrations in a Large Steam Turbine // Practical Experiences with Flow-Induced Vibrations. - New York : Springer-Verlag, 1980. -P.320-322.

95.Pratiwi W. P., Indriawati K. Fault-tolerant control strategy based on reliability and cost analysis in heat recovery steam generator plant // 2020 International Conference on Smart Technology and Applications (ICoSTA), IEEE, 2020. - P. 1-6.

96.Ioannou E. The Effects of Temperature Distortion on Aerodynamics and Low Engine Order Forced Response in Axial Turbines : дис. ... - London : City University London, 2015.

97.Abouelanouar B., Elkihel A., Gziri H. Experimental study on energy consumption in rotating machinery caused by misalignment // SN Appl. Sci. - 2020. - Vol. 2, № 7. -P. 1-6.

98.Sridhar V., Chana K. S. Development of a combined eddy current and pressure sensor for gas turbine blade health monitoring // Proceedings of the ASME Turbo Expo 2017: Turbomachinery Technical Conference and Exposition. Volume 6: Ceramics; Controls, Diagnostics and Instrumentation; Education; Manufacturing Materials and Metallurgy. Charlotte, North Carolina, USA. June 26-30. - P. V006T05A015.

99.Wang Q., Qi H., Liu F. Time series prediction of E-nose sensor drift based on deep recurrent neural network // Chinese Control Conference (CCC), Guangzhou, China, 27 July-30 July 2019. - P. 3479-3484. - IEEE.

100. Chen Y., Zhang H., Wang Y., Yang Y., Zhou X., Wu Q. J. MAMA net: multi-scale attention memory autoencoder network for anomaly detection // IEEE Trans Med Imaging. - 2020. - Vol. 40, № 3. - P. 1032-1041.

101. Amari S. I. Backpropagation and stochastic gradient descent method // Neurocomputing. - 1993. - Vol. 5, № 4-5. - P. 185-196.

102. Soydaner D. A comparison of optimization algorithms for deep learning // Intern J Pattern Recognit Artif Intell. - 2020. - Vol. 34, № 13. - P. 2052013.

103. Sagheer A., Kotb M. Time series forecasting of petroleum production using deep LSTM recurrent networks // Neurocomputing. - 2019. - Vol. 323. - P. 203-213.

104. Wang Z., Liu Q., Chen H., Chu X. A deformable CNN-DLSTM based transfer learning method for fault diagnosis of rolling bearing under multiple working conditions // Int J Prod Res. - 2021. - Vol. 59, № 16. - P. 4811-4825.

105. Sherstinsky A. Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) network // Phys D: Nonlinear Phenom. - 2020. - Vol. 404. - P. 132306.

106. Portnoy L., Eskin E., Stolfo S. Intrusion detection with unlabeled data using clustering // In Proceedings of ACM CSS workshop on data mining applied to security (DMSA-2001), 2001.

107. Amer M., Goldstein M., Abdennadher S. Enhancing one-class support vector machines for unsupervised anomaly detection // In Proceedings of the ACM SIGKDD workshop on outlier detection and description. - 2013. - P. 8-15.

108. Xu D., Wang Y., Meng Y., Zhang Z. An improved data anomaly detection method based on isolation forest // 2017 10th international symposium on computational intelligence and design (ISCID). - 2017. - Vol. 2. - P. 287-291. -IEEE.

109. Xu L., Yeh Y. R., Lee Y. J., Li J. A hierarchical framework using approximated local outlier factor for efficient anomaly detection // Procedia Computer Science. -2013. - Vol. 19. - P. 1174-1181.

110. Wang Z., Zhou Y., Li G. Anomaly detection by using streaming K-means and batch K-means // 2020 5th IEEE international conference on big data analytics (ICBDA). - 2020. - P. 11-17. - IEEE.