Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Pretrain, Prompt, and Transfer: Evolving Digital Twins for Time-to-Event Analysis in Cyber-physical Systems

Qinghua Xu, Tao Yue, Shaukat Ali, Maite Arratibel

TL;DR

This work addresses evolving Digital Twins (DTs) for time-to-event (TTE) analysis in cyber-physical systems (CPS) by introducing PPT, an uncertainty-aware transfer-learning framework that uses pretraining, prompt tuning, and uncertainty quantification (UQ). PPT constructs source and target DTs (DTM+DTC) and aligns their hidden representations through a transfer-learning objective, while selecting the most informative samples via UQ. The approach is evaluated on elevator (vertical transport) and autonomous driving system (ADS) datasets, showing consistent improvements in Huber loss over baselines, with transfer learning, prompt tuning, and UQ each contributing to performance and efficiency. The results highlight the practical potential of PPT for DT evolution in CPS where data from updated or new deployments are limited, enabling safer, more reliable time-to-event analyses across domains, including power grids and railways in future work. The methodology emphasizes robust evaluation via repeated trials and statistical testing, and provides insights into method selection (e.g., CS vs. Ensemble UQ) based on resource constraints and domain needs.

Abstract

Cyber-Physical Systems (CPSs), e.g., elevator systems and autonomous driving systems, are progressively permeating our everyday lives. To ensure their safety, various analyses need to be conducted, such as anomaly detection and time-to-event analysis (the focus of this paper). Recently, it has been widely accepted that digital Twins (DTs) can serve as an efficient method to aid in the development, maintenance, and safe and secure operation of CPSs. However, CPSs frequently evolve, e.g., with new or updated functionalities, which demand their corresponding DTs be co-evolved, i.e., in synchronization with the CPSs. To that end, we propose a novel method, named PPT, utilizing an uncertainty-aware transfer learning for DT evolution. Specifically, we first pretrain PPT with a pretraining dataset to acquire generic knowledge about the CPSs, followed by adapting it to a specific CPS with the help of prompt tuning. Results highlight that PPT is effective in time-to-event analysis in both elevator and ADSs case studies, on average, outperforming a baseline method by 7.31 and 12.58 in terms of Huber loss, respectively. The experiment results also affirm the effectiveness of transfer learning, prompt tuning and uncertainty quantification in terms of reducing Huber loss by at least 21.32, 3.14 and 4.08, respectively, in both case studies.

Pretrain, Prompt, and Transfer: Evolving Digital Twins for Time-to-Event Analysis in Cyber-physical Systems

TL;DR

This work addresses evolving Digital Twins (DTs) for time-to-event (TTE) analysis in cyber-physical systems (CPS) by introducing PPT, an uncertainty-aware transfer-learning framework that uses pretraining, prompt tuning, and uncertainty quantification (UQ). PPT constructs source and target DTs (DTM+DTC) and aligns their hidden representations through a transfer-learning objective, while selecting the most informative samples via UQ. The approach is evaluated on elevator (vertical transport) and autonomous driving system (ADS) datasets, showing consistent improvements in Huber loss over baselines, with transfer learning, prompt tuning, and UQ each contributing to performance and efficiency. The results highlight the practical potential of PPT for DT evolution in CPS where data from updated or new deployments are limited, enabling safer, more reliable time-to-event analyses across domains, including power grids and railways in future work. The methodology emphasizes robust evaluation via repeated trials and statistical testing, and provides insights into method selection (e.g., CS vs. Ensemble UQ) based on resource constraints and domain needs.

Abstract

Cyber-Physical Systems (CPSs), e.g., elevator systems and autonomous driving systems, are progressively permeating our everyday lives. To ensure their safety, various analyses need to be conducted, such as anomaly detection and time-to-event analysis (the focus of this paper). Recently, it has been widely accepted that digital Twins (DTs) can serve as an efficient method to aid in the development, maintenance, and safe and secure operation of CPSs. However, CPSs frequently evolve, e.g., with new or updated functionalities, which demand their corresponding DTs be co-evolved, i.e., in synchronization with the CPSs. To that end, we propose a novel method, named PPT, utilizing an uncertainty-aware transfer learning for DT evolution. Specifically, we first pretrain PPT with a pretraining dataset to acquire generic knowledge about the CPSs, followed by adapting it to a specific CPS with the help of prompt tuning. Results highlight that PPT is effective in time-to-event analysis in both elevator and ADSs case studies, on average, outperforming a baseline method by 7.31 and 12.58 in terms of Huber loss, respectively. The experiment results also affirm the effectiveness of transfer learning, prompt tuning and uncertainty quantification in terms of reducing Huber loss by at least 21.32, 3.14 and 4.08, respectively, in both case studies.
Paper Structure (76 sections, 61 equations, 8 figures, 20 tables, 4 algorithms)

This paper contains 76 sections, 61 equations, 8 figures, 20 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Background and Definitions
  3. CPS evolution
  4. TTE analysis in industrial elevators and ADSs
  5. Digital Twin
  6. Uncertainty Quantification
  7. Approach
  8. Overall Architecture
  9. Data Processing Component
  10. Digital Twin Component
  11. Transfer Learning Component
  12. Training Process of PPT
  13. Pretraining Phase
  14. Prompt Tuning Phase
  15. Experiment Design
  16. ...and 61 more sections

Figures (8)

  • Figure 1: Digital Twin for Cyber-physical System
  • Figure 2: Overall architecture of PPT
  • Figure 3: Structure of the data processing module. Dark purple squares represent the selected most uncertain samples, light purple squares signify their contexts, and white squares denote unselected samples.
  • Figure 4: PPT's prompt tuning. Note that we only display components related to prompt tuning and omit the detailed DT structure and UQ strategies for brevity since it is identical to those in the pretraining phase.
  • Figure 5: Digital Twin for Cyber-Physical System
  • ...and 3 more figures

Theorems & Definitions (8)

  • Definition 2.1: Industrial Elevator TTE Analysis
  • Definition 2.2: Industrial Elevator Evolution
  • Definition 2.3: ADS TTE Analysis
  • Definition 2.4: ADS Evolution
  • Definition 9.1: Industrial Elevator TTE Analysis
  • Definition 9.2: Industrial Elevator Evolution
  • Definition 9.3: ADS TTE Analysis
  • Definition 9.4: ADS Evolution