A Data Mining Strategy For Predictive Maintenance In The Operations Industry/
Mina Magdy Mamdouh Kamel
A Data Mining Strategy For Predictive Maintenance In The Operations Industry/ Mina Magdy Mamdouh Kamel - 2022 - 100 p. ill. 21 cm.
Supervisor:
Nashwa Abdelbaki
Thesis (M.A.)—Nile University, Egypt, 2022 .
"Includes bibliographical references"
Contents:
CERTIFICATION OF APPROVAL ...................................................................... iii
DEDICATION ....................................................................................................... vii
ACKNOWLEDGEMENTS .................................................................................... ix
TABLE OF CONTENTS ........................................................................................ xi
LIST OF FIGURES ............................................................................................... xv
LIST OF TABLES ............................................................................................... xvii
ACRONYMS ........................................................................................................ xix
PUBLICATIONS .................................................................................................. xxi
ABSTRACT ........................................................................................................ xxiii
1. INTRODUCTION ............................................................................................ 1
1.1. Motivation ................................................................................................. 1
1.2. Thesis Outline ........................................................................................... 4
2. Background Overview and Related Work ........................................................ 5
2.1. Background ............................................................................................... 5
2.1.1. Mobile Network Definitions .................................................................. 6
2.1.2. Access Network Operation System ....................................................... 8
2.1.3. Mobile Network Alarm Types and Main Definitions ........................... 8
xii
2.1.4. Maintenance in Mobile Networks ......................................................... 9
2.1.5. Preventive Maintenance ...................................................................... 11
2.1.6. Predictive Maintenance ....................................................................... 12
2.2. Historical work ........................................................................................ 15
2.2.1. Classification State of Art.................................................................... 15
2.2.2. Long Short-Term Memory State of Art ............................................... 17
2.2.3. Association Rules State of Art............................................................. 18
2.2.4. Our Observations ................................................................................. 20
3. ASSOCIATION RULE MINING ALGORITHMS ....................................... 21
3.1. Principle of Association Rule Mining ..................................................... 22
3.2. Association Rules Efficiency Measurements .......................................... 22
3.3. Association Rule Models ........................................................................ 24
3.3.1. Apriori Algorithm Illustration ............................................................. 25
3.3.2. Frequent Pattern (FP) Growth Algorithm Illustration ......................... 25
3.4. Apriori vs FP Growth .............................................................................. 26
4. SYSTEM MODEL ......................................................................................... 27
4.1. Data Challenges ......................................................................................... 27
4.1.1. Dataset Size Challenge .......................................................................... 27
4.1.2. Flapping Challenge ................................................................................. 29
4.2. Model Structure Challenge...................................................................... 30
4.3. Model Requirements ............................................................................... 32
4.3.1 Time Challenge of The Event to Predict ............................................. 33
4.3.2. Problem Prediction with Recommendation Action ............................. 34
4.3.3. Computational Power Issue ................................................................. 35
4.4. Approach Structure ................................................................................. 35
4.5. Operation Flow ........................................................................................ 37
4.5.1 Data Collection .................................................................................... 37
4.5.2 Handling Outliers ...................................................................................... 37
4.5.3. Target Event and Hidden Interval ....................................................... 40
4.5.4. Select Historical Window .................................................................... 41
4.6. Feature Engineering ................................................................................ 42
4.6.1. Feature Engineering for Historical Alarms Behavior .......................... 43
4.6.2. Feature Engineering Over Historical and Target alarm ....................... 46
4.7. Running Over Spark FP-Growth ............................................................. 48
4.8. Iteration ................................................................................................... 48
4.9. Testing and Training Data Flow.............................................................. 48
5. EXPERIMENTS AND RESULT ................................................................... 51
5.1. Experiments related to previous work ..................................................... 51
5.1.1. Classification Experiments .................................................................. 51
5.1.2. Long Short-Term Memory Experiments ............................................. 53
5.1.3. Association Rules Experiments ........................................................... 55
5.2. New Model Experiments ......................................................................... 57
5.2.1. Phase 1 Build a Proof of Concept (POC) ............................................ 57
5.2.2. Phase 2 Building The logic over Network Scale ................................. 60
5.3. RESULTS................................................................................................ 62
6. CONCLUSION AND FUTURE WORK ....................................................... 67
References .............................................................................................................. 69
Abstract:
Service-providing companies such as mobile networks use a massive amount of different hardware devices to support their customers. They build a maintenance process to have better quality management and technical advancements. Risks from hardware failure, security, and environmental issues must be controlled. The industry focuses on this process since it is directly responsible for the effectiveness and sustainability of the industrial operation. Maintenance plays a main role in organizational structures, operation cost, and customer experience.
Nowadays, the industry directly focuses on data and knowledge discovery processes. Sources like the Internet of Things contain a massive number of sensors. Each of these devices generates a large amount of data in real-time. Big data clusters provided a platform to accommodate this large amount of data. Data mining in many types of research has proven to be a powerful tool that can analyze large data from industrial sources. This has an advantage over manual analytics since it easily merges data from different sources to clarify hidden relations for better data understanding.
The operational maintenance process collects data from different hardware components in the industry. This large number of components generates data in real-time. This leads to a massive amount of data. The modern operation process applies the knowledge discovery methods over this massive amount of data. This generates relations between these data sources and provides a solution to the existing problems. Big data systems used a complicated parsing process to transform the formats into standard ones.
Predictive maintenance is one of the top rising trends, especially after the recent vibe of Machine Learning (ML), which introduced many powerful tools that can be used in event prediction. Predictive maintenance increases the efficiency of
xxiv
involved machines in the system by predicting future faults. This is different from reactive maintenance which allows assets to run to failure.
Different techniques of Machine Learning like classification or data mining are used to predict future alarms. However, one of the missing aspects is to predict the future alarm and provide a recommendation for the required action.
In our thesis, we propose a methodology to apply data mining over the historical trend of generated events from the telecom system. This data mining process generates rules to predict future failures. Our methodology recommends the required actions to avoid future failure. It predicts alarms that will happen after a time period from 12 to 24 hours. This means our system guarantees to predict future failure with a sufficient time window to apply the recommended action. Our approach can deal with any sort of available logging system. We did a fine tuning to adjust the data formats and feature engineering steps to show hidden features required for a better model. We customized new parameters to explain the model gain.
Text in English, abstracts in English.
Informatics-IFM
Dissertation, Academic
610
A Data Mining Strategy For Predictive Maintenance In The Operations Industry/ Mina Magdy Mamdouh Kamel - 2022 - 100 p. ill. 21 cm.
Supervisor:
Nashwa Abdelbaki
Thesis (M.A.)—Nile University, Egypt, 2022 .
"Includes bibliographical references"
Contents:
CERTIFICATION OF APPROVAL ...................................................................... iii
DEDICATION ....................................................................................................... vii
ACKNOWLEDGEMENTS .................................................................................... ix
TABLE OF CONTENTS ........................................................................................ xi
LIST OF FIGURES ............................................................................................... xv
LIST OF TABLES ............................................................................................... xvii
ACRONYMS ........................................................................................................ xix
PUBLICATIONS .................................................................................................. xxi
ABSTRACT ........................................................................................................ xxiii
1. INTRODUCTION ............................................................................................ 1
1.1. Motivation ................................................................................................. 1
1.2. Thesis Outline ........................................................................................... 4
2. Background Overview and Related Work ........................................................ 5
2.1. Background ............................................................................................... 5
2.1.1. Mobile Network Definitions .................................................................. 6
2.1.2. Access Network Operation System ....................................................... 8
2.1.3. Mobile Network Alarm Types and Main Definitions ........................... 8
xii
2.1.4. Maintenance in Mobile Networks ......................................................... 9
2.1.5. Preventive Maintenance ...................................................................... 11
2.1.6. Predictive Maintenance ....................................................................... 12
2.2. Historical work ........................................................................................ 15
2.2.1. Classification State of Art.................................................................... 15
2.2.2. Long Short-Term Memory State of Art ............................................... 17
2.2.3. Association Rules State of Art............................................................. 18
2.2.4. Our Observations ................................................................................. 20
3. ASSOCIATION RULE MINING ALGORITHMS ....................................... 21
3.1. Principle of Association Rule Mining ..................................................... 22
3.2. Association Rules Efficiency Measurements .......................................... 22
3.3. Association Rule Models ........................................................................ 24
3.3.1. Apriori Algorithm Illustration ............................................................. 25
3.3.2. Frequent Pattern (FP) Growth Algorithm Illustration ......................... 25
3.4. Apriori vs FP Growth .............................................................................. 26
4. SYSTEM MODEL ......................................................................................... 27
4.1. Data Challenges ......................................................................................... 27
4.1.1. Dataset Size Challenge .......................................................................... 27
4.1.2. Flapping Challenge ................................................................................. 29
4.2. Model Structure Challenge...................................................................... 30
4.3. Model Requirements ............................................................................... 32
4.3.1 Time Challenge of The Event to Predict ............................................. 33
4.3.2. Problem Prediction with Recommendation Action ............................. 34
4.3.3. Computational Power Issue ................................................................. 35
4.4. Approach Structure ................................................................................. 35
4.5. Operation Flow ........................................................................................ 37
4.5.1 Data Collection .................................................................................... 37
4.5.2 Handling Outliers ...................................................................................... 37
4.5.3. Target Event and Hidden Interval ....................................................... 40
4.5.4. Select Historical Window .................................................................... 41
4.6. Feature Engineering ................................................................................ 42
4.6.1. Feature Engineering for Historical Alarms Behavior .......................... 43
4.6.2. Feature Engineering Over Historical and Target alarm ....................... 46
4.7. Running Over Spark FP-Growth ............................................................. 48
4.8. Iteration ................................................................................................... 48
4.9. Testing and Training Data Flow.............................................................. 48
5. EXPERIMENTS AND RESULT ................................................................... 51
5.1. Experiments related to previous work ..................................................... 51
5.1.1. Classification Experiments .................................................................. 51
5.1.2. Long Short-Term Memory Experiments ............................................. 53
5.1.3. Association Rules Experiments ........................................................... 55
5.2. New Model Experiments ......................................................................... 57
5.2.1. Phase 1 Build a Proof of Concept (POC) ............................................ 57
5.2.2. Phase 2 Building The logic over Network Scale ................................. 60
5.3. RESULTS................................................................................................ 62
6. CONCLUSION AND FUTURE WORK ....................................................... 67
References .............................................................................................................. 69
Abstract:
Service-providing companies such as mobile networks use a massive amount of different hardware devices to support their customers. They build a maintenance process to have better quality management and technical advancements. Risks from hardware failure, security, and environmental issues must be controlled. The industry focuses on this process since it is directly responsible for the effectiveness and sustainability of the industrial operation. Maintenance plays a main role in organizational structures, operation cost, and customer experience.
Nowadays, the industry directly focuses on data and knowledge discovery processes. Sources like the Internet of Things contain a massive number of sensors. Each of these devices generates a large amount of data in real-time. Big data clusters provided a platform to accommodate this large amount of data. Data mining in many types of research has proven to be a powerful tool that can analyze large data from industrial sources. This has an advantage over manual analytics since it easily merges data from different sources to clarify hidden relations for better data understanding.
The operational maintenance process collects data from different hardware components in the industry. This large number of components generates data in real-time. This leads to a massive amount of data. The modern operation process applies the knowledge discovery methods over this massive amount of data. This generates relations between these data sources and provides a solution to the existing problems. Big data systems used a complicated parsing process to transform the formats into standard ones.
Predictive maintenance is one of the top rising trends, especially after the recent vibe of Machine Learning (ML), which introduced many powerful tools that can be used in event prediction. Predictive maintenance increases the efficiency of
xxiv
involved machines in the system by predicting future faults. This is different from reactive maintenance which allows assets to run to failure.
Different techniques of Machine Learning like classification or data mining are used to predict future alarms. However, one of the missing aspects is to predict the future alarm and provide a recommendation for the required action.
In our thesis, we propose a methodology to apply data mining over the historical trend of generated events from the telecom system. This data mining process generates rules to predict future failures. Our methodology recommends the required actions to avoid future failure. It predicts alarms that will happen after a time period from 12 to 24 hours. This means our system guarantees to predict future failure with a sufficient time window to apply the recommended action. Our approach can deal with any sort of available logging system. We did a fine tuning to adjust the data formats and feature engineering steps to show hidden features required for a better model. We customized new parameters to explain the model gain.
Text in English, abstracts in English.
Informatics-IFM
Dissertation, Academic
610