Resource Allocation Model for IP-Backhaul Maintenance using Risk Analysis and Convolutional Neural Network
| dc.contributor.advisor | Rosa, Tsegaye (PhD) | |
| dc.contributor.author | Tamrat, Demena | |
| dc.date.accessioned | 2022-02-10T11:19:44Z | |
| dc.date.accessioned | 2023-11-04T15:13:03Z | |
| dc.date.available | 2022-02-10T11:19:44Z | |
| dc.date.available | 2023-11-04T15:13:03Z | |
| dc.date.issued | 2021-11 | |
| dc.description.abstract | resources is vital. Over- or under-used maintenance resources such as automotive, tool, technician, and others are the main causes of extended maintenance times and dissatisfied customers. Ethiotelecom uses IP backhaul transport network technology mainly using IP-MPLS enabled routers and power systems to deliver Fourth Generation (4G) cellular and Fixed Mobile Convergence (FMC) services. Due to a lack of a failure risk analysis as a metric in the maintenance of the IP backhaul and unavailability of the services, Ethiotelecom loses 2.073 billion ETB every month. In this study, to assist the decision-making framework in daily resource sharing between IP and Power maintenance teams optimally at Ethiotelecom, we developed a knowledge-based optimal failure mode classifier model by fusing IP backhaul network alarm and Key Performance Indicators (KPIs) data obtained from Network Management Systems (NMS) at the network (topology) and feature level by using a design matrix and sporadic data aggregation process (DAP). To create a labeled design matrix based on a rational decision, we use weak-supervision data programming based on user-defined criteria functions obtained from reliability analysis using Failure Mode and Effect Analysis (FMEA) and the Decision-Making and Trial Evaluation Laboratory (DEMATEL) on the network failure history data with the help of experts from the two sections. In addition, we adopted a Windows slicing method in the time domain to obtain daily sliced grayscale images of the two failure modes on Python. We trained the two-dimensional convolutional neural networks (2D-CNN) to classify the images using grid search and k-fold cross-validation and tested the model using F1-Score and ROC-AUC. We have a 97 percent accurate baseline model. Besides, we test the model separately by selecting the alarm and KPI frames to capture the confidence of the model and to avoid uncertainty in the decision-making framework in IP-backhaul maintenance resource allocation. We then have a 98 percent and a 46 percent fair model, respectively. The study then concludes with some validation challenges and pitfalls | en_US |
| dc.identifier.uri | http://etd.aau.edu.et/handle/123456789/30001 | |
| dc.language.iso | en | en_US |
| dc.publisher | Addis Ababa University | en_US |
| dc.subject | IP-backhaul | en_US |
| dc.subject | Maintenance | en_US |
| dc.subject | Resource allocation | en_US |
| dc.subject | Decision support | en_US |
| dc.subject | 2D | en_US |
| dc.subject | CNN | en_US |
| dc.subject | KPI | en_US |
| dc.subject | Alarm | en_US |
| dc.subject | Domain Knowledge | en_US |
| dc.subject | FMEA | en_US |
| dc.subject | DEMATEL | en_US |
| dc.title | Resource Allocation Model for IP-Backhaul Maintenance using Risk Analysis and Convolutional Neural Network | en_US |
| dc.type | Thesis | en_US |