• Generative machine learning to predict sets of values of uncertain characteristics. • Prior distributions of characteristics and energy are used to find optimal value sets. • Model inversion and generative methods with and without an error domain are compared. • Sets of values predicted by generative methods provide a robust energy use estimation. We use a significant amount of energy to provide thermal comfort in buildings, and predicting this energy use is challenging due to the nature of building characteristics. Many energy-related characteristics are either uncertain or difficult to measure. This uncertainty hinders analysis of the current state and potential improvements. We propose a novel approach based on generative machine learning (ML) to estimate the values of energy-related characteristics. A generative ML network is trained to predict the sets of values of characteristics within predetermined constraints, corresponding to historical energy use. In addition, we build on an error-domain approach to include systematic modelling uncertainties. Three approaches − model inversion, generative ML, and generative ML with an error domain – are compared to predict sets of values of energy-related characteristics. The predicted sets of values obtained using a generative ML approach provide the most precise estimates of energy. However, these estimates can differ significantly from the measured energy use. While the predicted sets of values using an error domain approach are statistically conservative, they allowed an approximately correct estimate of energy use. Thus, leveraging a generative ML approach, we enriched energy models with relevant information that facilitates reliable energy analyses of the current state and predicted energy use.

Occupant detection and recognition support functional goals such as security, healthcare, and energy management in buildings. Typical sensing approaches, such as smartphones and cameras, undermine the privacy of building occupants and inherently affect their behavior. To overcome these drawbacks, a non-intrusive technique using floor-vibration measurements, induced by human footsteps, is outlined. Detection of human-footstep impacts is an essential step to estimate the number of occupants, recognize their identities and provide an estimate of their probable locations. Detecting the presence of occupants on a floor is challenging due to ambient noise that may mask footstep-induced floor vibrations. Also, signals from multiple occupants walking simultaneously overlap, which may lead to inaccurate event separation. Signals corresponding to events, once extracted, can be used to identify the number of occupants and their locations. Spurious events such as door closing, chair dragging and falling objects may produce vibrations similar to footstep-impacts. Signals from such spurious events have to be discarded as outliers to prevent inaccurate interpretations of floor vibrations for occupant detection. Walking styles differ among occupants due to their anatomies, walking speed, shoe type, health and mood. Thus, footstep-impact vibrations from the same person may vary significantly, which adds uncertainty and complicates occupant recognition. In this paper, efficient strategies for event-detection and event-signal extraction have been described. These strategies are based on variations in standard deviations over time of measured signals (using a moving window) that have been filtered to contain only low-frequency components. Methods described in this paper for event detection and event-signal extraction perform better than existing threshold-based methods (fewer false positives and false negatives). Support vector machine classifiers are used successfully to distinguish footsteps from other events and to determine the number of occupants on a floor. Convolutional neural networks help recognize the identity of occupants using footstep-induced floor vibrations. The utility of these strategies for footstep-event detection, occupant counting, and recognition is validated successfully using two full-scale case studies.

Information regarding occupants inside buildings has the potential to improve security, energy management, and caregiving. Typical sensing approaches for occupant localization rely on mobile devices and cameras. These systems compromise privacy. Occupant localization using floor-vibration measurements, induced by footsteps, is a non-intrusive sensing method that requires few sensors (one per ~35 m2). Current occupant-localization methodologies that rely on vibration measurements are data-driven techniques. These techniques do not account for the structural behavior of floor slabs leading to ambiguous interpretations of vibrations measurement in the presence of obstructions and varying floor rigidities. In this paper, a model-based approach using error-domain model falsification (EDMF) is used to overcome these limitations. EDMF incorporates information related to physics-based models in the interpretation of vibration measurements to identify a population of possible occupant locations. EDMF accommodates systematic errors and model bias to reject models that contradict measurement data. Uncertainties from multiple sources such as modeling imperfection and walking-gait variability are included explicitly while estimating occupant locations using EDMF. A unique approach to identify footstep-contact dynamics is proposed and evaluated for its ability to improve the precision of occupant localization. The approach involves dividing the floor-slab into zones using knowledge of structural behavior. Clustering measured vibrations to define several footstep-contact severity levels helps reduce uncertainty in walking gait thus improving the accuracy of footstep-contact dynamics to use as loading input into model simulations. The utility of occupant localization using this approach is evaluated using a full-scale case study. Localization precision increased by more than 50% compared with non-zone-based strategies.

Optimal performance of civil infrastructure is an important aspect of liveable cities. A judicious combination of physics-based models with monitoring data in a validated methodology that accounts for uncertainties is explored in this paper. This methodology must support asset managers when they need to extrapolate current performance to meet future needs. Three model-based data-interpretation methodologies, residual minimization, Bayesian model updating and error-domain model falsification (EDMF), are compared according to their ability to provide accurate interpretations of monitoring data. These comparisons are made using a full-scale case study, a steel-concrete composite bridge in USA. Validation of data interpretation is carried out using cross-validation (leave-one-out and hold-out). A joint-entropy metric is used to evaluate the extent to which the data that is used for validation contains information that is independent of data used for interpreting structural behaviour. Once accurately updated and validated knowledge of structural behaviour is available, it is employed to make predictions of remaining fatigue-life of the bridge. Validated identification of structural behaviour helps ensure accurate predictions of capacity of bridges beyond their design lives. EDMF and a modified form of Bayesian model updating are analytically and numerically equivalent, while EDMF has several practical advantages. Both methods provide accurate identification and safe estimations of the remaining fatigue life of the bridge. Such enhanced understanding of structural behaviour leads to appropriate decisions regarding civil infrastructure assets.