SMIP 2020 Seminar

Frequency-dependent horizontal-to-vertical spectral ratios (HVSR) can provide information on site resonant frequencies, which are potentially useful for predicting site amplification. We adapt a relational database developed to archive and disseminate VS data to include HVSR and investigate the consistency of HVSR derived from different measurements of ambient noise (temporary instruments, permanent instruments) and earthquake recordings. The database as a whole consists of 2,797 sites in California. HVSR consistency is analyzed using subsets of sites with multiple data sources; noise and seismic data are consistent for 60% of sites, whereas different noise measurements have about 75% consistency.

We summarize an analysis of ground motions recorded during three events that occurred during the July 2019 Ridgecrest earthquake sequence. We collected and uniformly processed 1,483 three-component recordings for the events from an array of 824 sensors spanning ten seismographic networks. Signal processing followed well-established NGA procedures. We developed site condition metadata from available geophysical data and multiple models. We computed intensity measures such as spectral acceleration at a number of oscillator periods and inelastic response spectra. We compared elastic and inelastic response spectra to seismic design spectra in building codes to evaluate ground motion damage potential at spatially-distributed sites.

Dynamic soil properties are key ingredients of analyses for predicting/assessing soil-structure interaction (SSI) and site response effects under seismic excitations. While there is already a large amount of valuable data recorded by numerous geotechnical arrays worldwide, there is no reliable technique that enables the extraction of dynamic nonlinear/hysteretic properties of soil layers. Herein, a stochastic filtering method devised for estimating the nonlinear soil properties from earthquake data recorded by geotechnical arrays.1D-3C finite element site models are used, and the soil layers’ constitutive model parameters and input excitation (bedrock or within motions) time-histories are identified using Unscented Kalman Filtering techniques. The method is first verified using synthetic examples and then validated using real-life data from centrifuge tests as well as the well-known Lotung site. Subsequently, the method is applied to earthquake data recorded by several CSMIP Geotechnical Arrays.

We use deep learning to predict surface-to-borehole Fourier amplification functions (AFs) from discretized shear-wave velocity profiles. Specifically, we train a fully connected and a convolutional neural network (NN) using observed mean AFs observed at ∼600 KiK-net and California Strong Motion Instrument Program (CSMIP) vertical array sites. Compared to predictions based on theoretical SH 1D amplifications, the NN results in up to 50% reduction of the mean squared log error between predictions and observations at sites not used for training. In the future, NNs may lead to a purely data-driven prediction of site response that is independent of proxies or simplifying assumptions.

Current provisions in ASCE-41 for performance-based assessment are applied to an existing three-story steel moment frame building that was designed and constructed prior to the 1961 UBC code revisions. A computer model of a perimeter frame that comprises the primary lateral system of the building was developed and validated against available instrumented data from two earthquakes. Both linear and nonlinear procedures were used in the assessment. Findings from the study indicate that the linear static and dynamic procedures produced consistent demand-to-capacity ratios. The nonlinear static procedure resulted in the most severe demands at the lowest level with two beams failing the Collapse Prevention limit state whereas the nonlinear dynamic procedure produced the lowest demands on the building; however, the fact that some individual motions caused some beams to exceed Life Safety or Collapse Prevention limits indicates that ground motion selection can play a major role in the outcome of the assessment when using the nonlinear dynamic procedure.

Our research puts the accidental torsion provisions in ASCE-7 for low-rise buildings in perspective; various combinations of plan aspect ratios, irregularity, and diaphragms rigidity are investigated. The presented work is based on simulations; however, the building models used in the study are proportioned to represent a wide range of code conforming buildings. 4-story building prototypes with a plan aspect ratio of 1:1, 1:2, 1:4, and 1:8 are modeled. The building models possess translational to rotational period ratios (Ω) ranging from 1.1 to 2.0. Type 1a (Torsional Irregularity) and Type 1b (Extreme Torsional Irregularity) – according to ASCE 7 – is considered as the measure of floor plan irregularity. Uncertainty in stiffness is treated as the source of accidental eccentricity. Results are compared with corresponding MDOF models having regular plans (i.e., symmetric) and rigid diaphragms. We conclude that the magnification in deformation demands due to accidental torsion in buildings with a semirigid diaphragm, or inherent plan irregularity, is smaller than building with regular floor plan and rigid diaphragm. Equivalent design eccentricities obtained from this body of work indicate that the 5% equivalent eccentricity rule is conservative to capture the deformation's magnification due to accidental torsion in low-rise buildings possessing floor plan irregularity or semirigid diaphragms if median estimates of all stories are the basis of code calibration.

The importance and relevance of structural health monitoring (SHM) for highway bridges in the United States is highlighted by the bridges’ poor condition and the growing amount of resources for data-driven condition assessment in the feld of artifcial intelligence (AI), particularly, machine learning (ML). To tackle this issue, a human-machine collaboration (HMC) framework for highway bridge SHM is developed in this study to take advantage of the strengths of both AI and engineering domain expertise. The H-MC framework uses a physics-based model (human) to conduct probability of exceedance (POE) analysis coupled with novelty detection ML model (machine) to establish a damage detection and assessment algorithm. To produce the training data for the model, nonlinear time history analyses (NTHA) are performed on analytical bridge model for vibration responses to many selected ground motions. Feature extraction and selection are performed using an Ordinal Fisher Score analysis with k-fold cross-validation parameter tuning to produce the input for training the ML model. A component capacity-based damage state model is used to produce the output for training the ML model. During model validation, the beneft of the H-MC is demonstrated through signifcant increase in the classifcation accuracy when the POE analysis is coupled with the ML model. Once the ML model is trained, it is tested on the seismic responses of two instrumented bridges, El Centro -Hwy8/Meloland Overpass and Parkfeld -Hwy46/Cholame Creek Bridge. The model accurately classifed all 14 undamaged events and one damaged event, including damage assessment consistent with reported visual inspection of the bridge after the damaging event.

The process of implementing a damage detection strategy for aerospace, civil and mechanical engineering infrastructure is referred to as structural health monitoring (SHM). The SHM process compliments traditional nondestructive evaluation by extending these concepts to online, in situ system monitoring on a more global scale. For long term SHM, the output of this process is periodically updated information regarding the ability of the structure to perform its intended function in light of the inevitable aging and degradation resulting from operational environments. After extreme events, such as earthquakes or blast loading, SHM is used for rapid condition screening and aims to provide, in near real time, reliable information regarding the integrity of the structure.

This presentation will briefly summarize the historical developments of SHM technology, which have been primarily driven by four applications: rotating machinery, offshore oil platforms, civil infrastructure, and aerospace structures. Next, the current state of the art is summarized where the SHM problem is described in terms of a statistical pattern recognition paradigm. In this paradigm, the SHM process can be broken down into four parts: (1) Operational Evaluation, (2) Data Acquisition and Cleansing, (3) Feature Extraction and Data Compression, and (4) Statistical Model Development for Feature Discrimination. This talk will then focus on recent developments related to both the sensing hardware and data analysis aspects of SHM. Some final comments will be made on outstanding technology development and validation needs that are necessary for more widespread adoption of SHM.

Advancements in sensor technologies and communication networks are creating new opportunities for advanced methods of measuring earthquake response and damage in critical infrastructure systems. Based on applied R&D sponsored by the U.S. Department of Energy (DOE), new optically-based sensor systems have been developed that provide for continuous measurement and rapid transmission of key infrastructure response observables immediately after an earthquake. The short latency of the underlying physics of optical sensors, and the ability to perform high resolution measurements across a broad frequency bandwidth are attributes that make optical-based measurement systems particularly attractive for applications in earthquake response measurement. Concurrently, transformational progress underway in wireless communications and the Internet of Things (IOT) are enabling new paradigms for expedient deployment of sensor systems and rapid extraction and analysis of time-critical data.​