Architecting a Digital Twin-Based Predictive Maintenance System for Modelling Cable Joint Degradation

The large scale adoption of wind turbines and solar panels in the Netherlands places new demands on the medium voltage power grid. For example, highly varying loads can cause failures in certain cables. Cable joints are natural weak spots prone to faults due to varying currents, creating downtime challenges for public utility companies. Predictive maintenance (PdM) practices are necessary to minimize downtime for users. We present a Model-Based System Engineering approach using formal models and UML views to provide a scalable PdM design ontology for modeling cable joint degradation. We aim to monitor cable joint degradation from different manufacturers under varying conditions throughout the Netherlands in real-time using a Digital Twin (DT) approach. Our design provides high-resolution, real-time synchronization between the DT-based PdM system and the cable joints. The proposed architecture is scalable, robust, and flexible, and the software implementation is publicly available in an open-source repository.  


INTRODUCTION
The ongoing energy transition in the Netherlands from a centralized to a decentralized system, driven by the increasing adoption of wind turbines and solar panels, has resulted in significant changes to the Dutch Medium Voltage (MV) elec-tricity grid.As a result of the increasing adoption of sustainable energy sources, MV grids are experiencing increasing variations in load and bi-directionality.Cable joints, the connections between two power cables, are natural weak spots constructed manually on-site and prone to faults due to varying currents.These weak spots create challenges for public utility companies as they aim to minimize user downtime.Therefore, effective predictive maintenance (PdM) or Prognostics and Health Management (PHM) practices are essential to tackle these growing downtime challenges.Partial discharges (PDs) are a symptom of a weak spot in a power cable joint that could lead to a fault.Diagnosing PDs is a proven method to assess the condition of underground power cable joints.A PD induces a small pulse in the conductor(s) and earth screen that propagates through the cable in both directions (Wagenaars, 2010).Alliander, a public utility company in the Netherlands, uses a non-intrusive system to detect PDs in real-time using sensors on two ends of the MV circuit.
Many previous studies have attempted to model cable joint degradation using mathematical, physical, or documentcentric engineering approaches.However, using these methods in isolation is a limitation, as they often model a single component or process.The complex "systems-of-systems" behavior that make up the entire MV grid can be represented by integrating models with different viewpoints into a single system.In this study, we present a novel Model-Based System Engineering (MBSE) approach using formal models for modeling the ontology of the MV grid and for providing a smart and scalable PdM solution for cable joint degrada-1 4th Asia Pacific Conference of the Prognostics and Health Management, Tokyo, Japan, September 11 -14, 2023 R10-02 tion.To model the views of the different concerns of cable joint degradation, we use UML views by Kruchten (1995) and Mall (2018).The ambition level of our PdM system is to monitor the degradation of various types of cable joints from different manufacturers under varying conditions throughout the Netherlands in real-time.The eventual goal is to provide a Remaining Useful Life module to accurately predict when a fault in a cable joint will occur.We can access high-quality historical weather and PD data, providing information on the joint's condition and environmental influences.As such, we can apply a Digital Twin (DT) approach (Tiddens, 2018).Our MBSE application provides high-resolution, real-time synchronization between the DT-based PdM system and the cable joints, providing a more effective and efficient method of monitoring and predicting the health of cable joints in the MV context.As a first application, we focus on joints that show PDs due to either (1) temperature-related expansion or contraction, (2) water ingress, or (3) both.These defects have a high incidence rate, and when detected early, it is possible to intervene before the fault occurs.Our approach has the potential to significantly improve the PHM of cable joints and reduce the costs associated with PD-related downtime.

RELATED WORK
This study is related to numerous other studies.Tiddens (2018) defines a framework for defining a business case and selecting the optimal PdM approach.Cocheteux, Voisin, Levrat, and Iung (2009) provide a methodology for selecting failure modes and prognostic models for the design process.Li, Verhagen, and Curran (2018) developed a functional architecture for PHM using a system engineering approach (Li et al., 2018).Aizpurua andCatterson (2015, 2016) showcase a methodology for designing PdM systems.Vogl, Weiss, and Donmez (2014) provide an overview of standards for PHM systems.Previous work on math-based computations to predict power outages due to PD and weather has been conducted by van Osch (2021).

METHODOLOGY
We adopted a MBSE approach to design the DT-based PdM system, which allowed for systematic and efficient system development.The MBSE approach involved using UML to create a detailed software design for the system using four views: User view, Implementation view, Behavioral view, and Environmental view (Kruchten, 1995;Mall, 2018).The Structural View was omitted, as it provides too much detail for this paper.The User view included a context diagram and use case diagram to give a high-level understanding of the system from the user's perspective.The Implementation view included a component diagram, which specified the system's structural organization and the interrelationships between its components.The Behavioral view included an activity diagram describing the system's dynamic behavior and the interactions between its components.Finally, the Environmental view included a deployment diagram illustrating the physical hardware components that make up the system and their connections.The MBSE approach began with defining the business case and requirements of the DT-based PdM system in collaboration with all stakeholders, including Alliander, the primary stakeholder.These requirements were then transformed into a software design specification linked to the Reference Architecture's feature model by van Dinter, Tekinerdogan, and Catal (2023) to determine the necessary components.The UML models were created based on the software design specification, providing a detailed representation of the system's structure and behavior.In addition to the UML models, code and unit tests were developed and made publicly available by van Dinter, Ekmekci, et al. (2023).

User View
The PdM software can be executed locally and can, for instance, forecast PDs or predict the Remaining Useful Life (RUL) of cable joints.The Object-Oriented software uses input arguments from a technical user.The context diagram is shown in Figure 1, and the use case diagram is shown in Figure 2. The following paragraphs will provide more detail on the design of both diagrams.The technical user, typically a data scientist, can provide a train or test command, which uses a set of failed joints and joints with a seasonal PD pattern, or a predict command, together with a list of circuit IDs and joint locations, to get predictions of the specific joints.The train command always calls the test command after training the model.After providing a command, the software launches the PredictiveMaintenanceJob, and loads (1) information on failed joints, (2) joints with seasonal PD patterns, and (3) circuit coordinates.We load each circuit's coordinates, request circuit-level weather data from Alliander's weather API (Alliander, n.d.), and request circuit configurations (i.e., configs) and circuit-level PD data from local storage or Cloud Environment.The circuit config provides information to build a Joint object with location-specific PD.The models use PD and circuit weather data to train, and their results are returned to the user.The model parameters are saved to local storage when the train command is called.The latest stored model is loaded from storage when the test or predict command is called.When the user passes the visualization flag, analysis results are shown.

Implementation View
The component diagram in Figure 3 shows a layered design following the 5C architecture for cyber-physical systems (Lee, Bagheri, & Kao, 2015).Note that our architecture has not implemented components for the Configuration layer, as we do not handle asset maintenance optimization in this case  For future work, we will focus on (1) evaluating the performance of the architecture in different case studies globally, (2) deploying the software, (3) adding components in the Configuration layer, (4) using other available data, such as joint age, manufacturer, connecting cable types, soil type, or load, (5) assessing the design's performance in practice and (6) optimizing the predictive maintenance models.

Figure 1 .
Figure 1.Context diagram of the predictive maintenance software

Figure 4 .
Figure 4. Activity diagram of predictive maintenance software

Figure 5 .
Figure 5. Deployment diagram of data files on local disk