Digital Twins in Industry: Theory, Technology, and Challenges

By Samudrapom DamApr 3 2024Reviewed by Susha Cheriyedath, M.Sc.

In an article recently published in the journal Nature Computational Science, the authors reviewed the main pitfalls and challenges of digital twins in industry and discussed the role of artificial intelligence (AI) in this context. Digital twins share a close relationship with advanced technologies such as AI.

Background

Digital twins have received significant attention in the last decade as they represent an effective approach to achieving the fusion between physical and virtual spaces. Specifically, digital twins can evolve and iterate through the seamless fusion and connection between physical and virtual domains.

This synchronization and consistency can benefit several services, including dynamic optimization and real-time monitoring. In recent years, the rapid development of digital twins has resulted in their application in different fields, specifically in industry.

For instance, digital design, testing, and verification; guiding the implementation and configuration process; condition monitoring and operation optimization; and digital reconfiguration design and verification are the major digital twins' applications in the shop floor lifecycle. Similarly, process monitoring and quality management; operation optimization, prognostic and health management; and guidance for disassembly and remanufacturing are the key digital twins' applications in the product lifecycle.

Although digital twins have addressed several problems in many industrial application settings, some drawbacks and limitations must be addressed. In this paper, the authors reviewed the major challenges to overcome and pitfalls to avoid to improve digital twins’ maturity and facilitate their applications in the industry on a large scale.

Main pitfalls

Overly complex or overly simplistic models are one of the key drawbacks of digital twins. For instance, overly complex models need substantial resources like computing time and costs, which can be unnecessary in several situations. Similarly, complex problems cannot be effectively solved using simpler models in practical applications. Thus, accurately clarifying the user requirements and analyzing specific problems is crucial for guiding the model construction/reconstruction trade-off between simple models and complex models.

Most existing studies on digital twins primarily focus on big data and neglect small datasets. However, AI techniques and distributed computing are typically required for big data analysis and processing, while small data only requires conventional statistical analysis techniques and data-mining methods for analysis and processing.

In many scenarios, small data plays a more important role compared to big data. Thus, integrating the limitations and strengths of both big data and small data approaches to develop a hybrid approach is necessary to enable better industrial services. Underused AI is another key drawback in the field of digital twins. AI is effective for mining knowledge from vast quantities of data and can be used in several practical situations in industry.

Digital twins can utilize AI to provide different services, such as energy consumption optimization, equipment failure prediction, and dynamic scheduling, with high quality. For instance, digital twins combined with machine learning can make risk probability rate prediction of an oil pipeline system and evaluate the system's remaining useful life in process manufacturing.

Similarly, in discrete manufacturing, the hybrid particle swarm optimization (PSO) algorithm with a variable neighborhood search strategy can be employed to identify the optimal scheduling scheme. AI can also resolve the issue of allocating computing resources optimally to overcome the computing power limitations on the shop floor.

Moreover, embodying AI into edge hardware can significantly reduce the data transfer amount and better meet the digital twins' timeliness requirement. Recent neural network-based products and large language models (LLMs) like ChatGPT can also play a critical role in digital twins.

Major challenges

Accurate construction of models and their comprehensive validation is a major challenge of digital twins. This challenge can be tackled by combining physics-based models and data. Similarly, a comprehensive validation of models is necessary to determine the feasibility of models. However, no international benchmark or standard currently exists in the industry to guide the model validation implementation.

Many products, production lines, shop floors, and factories have high reliability and safety requirements as they contain explosive or flammable articles. Although AI techniques can improve the value and effects of several industrial applications of digital twins, assigning the responsibility for accidents is difficult due to the black-box nature of these techniques, which implies a lack of explainability in the workings of the AI models and the outcomes obtained from them.

Explainable AI for industrial applications can address this issue. However, establishing common ground for evaluating the explainability of AI models, creating an acceptable benchmark by factoring in both objective and subjective factors, and striking a balance between model interpretability and performance are essential for further progress in this field.

To summarize, digital twin development in industries faces consistent technical challenges, which must be resolved using diverse approaches, including those based on AI, for facilitating large-scale industrial applications.