Abstract:Digital twin-enabled building operations heavily rely on a building sensor network to collect operational data. Virtual sensing, in combination with physical sensors, has great potential in establishing a reliable and informative sensing environment. However, physical sensors can experience hardware errors or systematic errors due to the working environment within building systems. Virtual sensors are also subject to their model uncertainties arising from input physical sensor errors and changes in system operations. These virtual sensor errors have a negative effect on the in-situ calibration accuracy of corresponding physical sensors. To address these issues, this study proposes a simultaneous in-situ calibration (SIC) method designed to calibrate multiple physical and virtual sensors with varying errors in the context of digital twins. The SIC problem is mathematically formulated, including two types of distance functions (direct and indirect) and defining correction functions to calibrate various types of virtual sensors (such as backup, validation, replacement/virtualization, and observation sensors). Importantly, the proposed method tackles the challenges posed by the unknown calibration environment in operational systems. A field application study was conducted for two physical sensors and three different virtual sensors operating in a district heating substation serving real residential buildings in Korea. The case study shows how the different calibration accuracies for each sensor depend on the calibration approaches, such as physical-targeted (existing Case 1), virtual-targeted (existing Case 2), and simultaneous calibration (novel Case 3). Under the simultaneous errors, only Case 3 satisfied the reference accuracy for all sensors. Specifically, the average MAEs in Case 3 were improved by approximately 79.3%, 75.3%, 63.9%, and 87.5% for the target physical sensor, backup virtual sensor, virtualized sensor, and validation virtual sensor, respectively, compared to the accuracies before calibration. These results demonstrate the importance and effectiveness of simultaneous in-situ calibration of physical and virtual sensors in digital twin-enabled building operations. The proposed calibration can contribute to the synchronization of physical and virtual building systems, thus achieving reliable and informative virtual sensing-driven digital twins.

Neural network-based nonintrusive calibration for an unobserved model in digital twin-enabled building operations

A Robot Calibration Method Based on Joint Angle Division and an Artificial Neural Network

A Method for Predicting Relative Position Errors in Dual-Robot Systems Via Knowledge Transfer from Geometric and Nongeometric Calibration

Simultaneous in-situ calibration for physical and virtual sensors towards digital twin-enabled building operations

Power Control for NN-based Wireless Distributed Inference with Improved Model Calibration

A Soft Calibration Technique for Infrared Imaging Based on Physical Model of Thermal Radiation and Variational Bayesian Inference from Small Dataset

Calibrated Prediction Intervals for Neural Network Regressors

Kalibre: Knowledge-based Neural Surrogate Model Calibration for Data Center Digital Twins

A Benchmark Study on Calibration

A novel in-situ sensor calibration method for building thermal systems based on virtual samples and autoencoder

Deep neural operator-driven real-time inference to enable digital twin solutions for nuclear energy systems

Operational Calibration: Debugging Confidence Errors for DNNs in the Field

A calibration method for enhancing robot accuracy through integration of an extended Kalman filter algorithm and an artificial neural network

Calibrating Deep Neural Networks using Explicit Regularisation and Dynamic Data Pruning

Kinematic and Nonparametric Calibration of a Parallel Machining Robot Based on an Artificial Neural Network

Computer Model Calibration with Time Series Data using Deep Learning and Quantile Regression

A hybrid neural network for real-time OD demand calibration under disruptions

Modeling of an inductive displacement sensor based on 1DCNN-LSTM-AT

Two Sides of Miscalibration: Identifying Over and Under-Confidence Prediction for Network Calibration

Improved Trainable Calibration Method for Neural Networks on Medical Imaging Classification

Calibrating Deep Neural Network using Euclidean Distance