1 Introduction
-
Extended the review by adding the most relevant recently published papers. All schemes and charts have been updated accordingly;
-
Improved the guidelines by introducing decision and task nodes taking into account factors like the aim of the DT, the decisions it has to make, which functionalities it has to provide, etc.;
-
Realized a preliminary reference architecture to support the design of AI-Assisted DTs with a particular focus on predictive maintenance applications.
Acronym | Description |
---|---|
AI | Artificial Intelligence |
AR | Augmented Reality |
B5G | Beyond Fifth Generation |
BIM | Building Information Modelling |
CNN | Convolutional Neural Network |
CNs | Connections |
DD | Digital Twin Data |
DL | Data Layer |
DT | Digital Twin |
gDT | Geometrical DT |
HMI | Human–Machine Interface |
IEC | International Electrotechnical Commission |
IEEE | Institute of Electrical and Electronics Engineers |
IIoT | Industrial IoT |
IoT | Internet of Things |
ISO | International Organization for Standardization |
IT | Information Technology |
KNN | K-nearest Neighbour |
MAE | Mean Absolute Error |
ML | Machine Learning |
MQTT | Message Queue Telemetry Transport |
PCD | Point Cloud Data |
PD | Primary Data |
PdM | Predictive Maintenance |
PL | Physical Layer |
PLM | Product Lifecycle Management |
PS | Physical Space |
PT | Physical Twin |
RANSAC | Random Sample Consensus |
RMSE | Root Mean Squared Error |
RUL | Remaining Useful Life |
SD | Secondary Data |
SL | Service Layer |
Ss | Services |
VL | Virtual Layer |
VR | Virtual Reality |
VS | Virtual Space |
2 Background
2.1 Digital twins
AI category | Literature works |
---|---|
Machine learning | |
Intelligent Sensor Data Integration | |
Data mining | |
Adversarial search | Ambra and Macharis [2] |
Computer vision & Image Processing | |
Evolutionary computing | Ricondo et al. [52] |
Operations research | Magnanini and Tolio [43] |
Other | Boockmeyer et al. [9] |
2.2 Maintenance approaches
3 An overview of DTs and AI in railways
3.1 Machine learning
3.2 Intelligent sensor data integration
3.3 Data mining
3.4 Computer vision and image processing
4 Challenges and open issues
4.1 Interoperability
4.2 Connectivity
4.3 Lack of standards and frameworks
4.4 Data privacy and security
4.5 Scalability
5 Guidelines for DT design and development
5.1 Requirement specification
-
simulation mode, namely it simulates/emulates the behaviour of the physical system and thus it executes in a static way. For instance, when the DT is used in the design stage, it is not connected to a physical counterpart (that can not even exist), therefore it has no capability to evolve as the physical system changes with respect to time;
-
Replication mode, i.e., the DT is used online and its models are fed with real-time data, so the virtual replica evolves along with its physical counterpart throughout its life cycle. This means that any changes in either the physical or Digital Twin are reflected in its counterpart, creating a closed feedback loop (that it’s not true in the simulation mode execution) Segovia and Garcia-Alfaro [56], Ferko et al. [26] and Eramo et al. [21].
5.2 Process planning
5.3 Architectural design
5.4 Digital representation
-
control models that, based on control theory, apply physics laws and compare simulated results with known information, i.e., mathematical models;
-
data-dependent models, which can leverage AI and use data structures that store all the variables describing reality at a predefined abstraction level;
-
hybrid control-data models, which combine control and data-dependent models to obtain advantages from both of them;
-
physical models, representing physical properties and phenomena (e.g., deformation);
-
geometrical models, mirroring the geometry, shapes, sizes, positions, logic, and interfaces of the real system.