Blind Prediction Research Articles

Seismic performance of sheet-pile wall supporting liquefiable backfill: Blind predictions of centrifuge model tests using Ta-Ger constitutive model

Blind numerical simulations of seven centrifuge tests with “identical” models of a sheet-pile wall supporting medium dense liquefiable backfill and subjected to seismic excitation were performed as part of the LEAP 2020 project. The analyses were conducted with FLAC using the Ta-Ger soil constitutive model. The Ta-Ger model parameters were calibrated against available laboratory DSS data on Ottawa sand with similar relative density with a focus on capturing at an element level: i) the liquefaction triggering resistance, ii) the post-liquefaction rate of shear-strain accumulation, iii) overburden effects on liquefaction triggering resistance and iv) realistic shear stress-strain responses. A single numerical model was built in prototype scale and analyses were performed by only varying the input seismic motion recorded at the base of each of seven centrifuge tests. Comparisons of numerical predictions with measured centrifuge test responses indicate that the analyses successfully capture the primary mechanisms of the system response. These included liquefaction in the free-field and development of negative pore pressures behind the wall, with accurate predictions of outward wall displacement for the majority of the tests. Most centrifuge tests and numerical predictions consistently exhibit a systematic linear trend of increase in wall displacements with spectral acceleration at the predominant frequency of the system. The numerical analyses overpredicted wall displacements only for centrifuge tests not following this trend, indicating that the variations maybe due to experimental variations from their specifications that were not considered in the blind predictions.

The use of fuzzy linear regression for the selection of the most appropriate fuzzy implication in a fly ash-based concrete model

In this research, fuzzy linear regression (FLR) method combined with three well-known fuzzy implications was implemented for evaluating the relation among the amount of fly ash in concrete mixture and the compressive strength of concrete. More specifically, 267 experimental data 40 of which were used for testing the validation of the process were subjected to FLR method for calculating the truth values, which indicated the degree of how the experimental outputs belong to the predicted ones. Also, the degree of fuzziness was calculated for performing the sensitivity analysis of the model. The truth values that emerged were used for applying three basic fuzzy implications such as Lukasiewicz, Reinchenbach, and Kleene-Dienes implication. By evaluating and comparing the results of every fuzzy implication, it was concluded that Lukasiewicz was the most appropriate implication method as it yielded the smallest deviation of truth values (σ = 4.00) in contrast to the theoretical ones (σ = 4.83 in Reinchenbach and σ = 12.31 in Kleene-Dienes fuzzy implication). The accuracy of the FLR method was also validated for calculating the coefficient of the mean absolute percentage error level (MAPE = 5.56%) of the blind prediction process, and the results revealed that the application of fuzzy linear regression method is suitable for evaluating the truth values of experimental data in order to be used in fuzzy implications. Thus, it is a satisfactory procedure for making inferences between concrete parameters.

A Thermal Conductivity Model for Deformable and Unsaturated Soils to Assess the Thermal Behaviour of Energy Piles

This paper presents an empirical model that predicts the thermal conductivity of soils by accounting for the effect of both the degree of saturation and void ratio on the heat exchange capacity of shallow geothermal reservoirs. The model is generated by the product of two terms: the former accounts for the effect of void ratio on the dry thermal conductivity, whereas the latter describes the influence of the degree of saturation on the moisture-dependent thermal conductivity. The model is a function of three parameters, which are easy to calibrate based on their physical meaning. Model predictions are validated against five different sets of experimental data from literature by means of two alternative approaches: blind prediction of thermal conductivity measurements not employed during calibration and numerical simulations of thermal tests performed on energy piles. Results show that the proposed model is capable of accurately predicting both the thermal conductivity of deformable unsaturated soils as well as reproducing the thermal behaviour of energy piles.

Benchmark VERCORS 2022: blind prediction of time-dependent behavior of concrete containment building with low and high-fidelity models

The VERCORS program aims to acquire an extensive experimental dataset collected to provide a solid basis for numerical modeling of concrete containment buildings (CCBs). In order to cover the entire life-span of a real containment, the measurements are done on 3× smaller mock-up which leads to 9-fold acceleration of all processes related to drying. The goal for the participants of the third VERCORS benchmark was to predict the behaviour of the CCB based on standard laboratory measurements on VERCORS concrete. The previous paper [1] presented the calibration procedure of material models for moisture transport and time-dependent behavior of concrete and summarized the results obtained with a computationally efficient low-fidelity model (LFM). The present paper compares the responses of the LFM and a high-fidelity model (HFM) with a detailed geometry of the entire containment and presents a comparison with the experimental data collected over the last 8 years on the VERCORS mockup.

From ambient vibration data analysis to 1D ground-motion prediction of the Mj 5.9 and the Mj 6.5 Kumamoto earthquakes in the Kumamoto alluvial plain, Japan

We present horizontal ground motion predictions at a soft site in the Kumamoto alluvial plain for the Mj 5.9 and Mj 6.5 Kumamoto earthquakes of April 2016, in the framework of an international blind prediction exercise. Such predictions were obtained by leveraging all available information which included: (i) analysis of earthquake ground motions; (ii) processing of ambient vibration data (AMV); and (iii) 1D ground response analysis. Spectral analysis of earthquake ground-motion data were used to obtain empirical estimates of the prediction site amplification function, with evidence of an amplification peak at about 1.2 Hz. Horizontal-to-vertical spectral ratio analysis of AMV confirmed this resonance frequency and pointed out also a low-frequency resonance around 0.3 Hz at the prediction site. AMV were then processed by cross-correlation, modified spatial autocorrelation and high-resolution beamforming methods to retrieve the 1D shear-wave velocity (Vs) structure at the prediction site by joint inversion of surface-wave dispersion and ellipticity curves. The use of low frequency dispersion curve and ellipticity data allowed to retrieve a reference Vs profile down to few thousand meters depth which was then used to perform 1D equivalent-linear simulations of the M 5.9 event, and both equivalent-linear and nonlinear simulations of the M 6.5 event at the target site. Adopting quantitative goodness-of-fit metrics based on time–frequency representation of the signals, we obtained fair-to-good agreement between 1D predictions and observations for the Mj 6.5 earthquake and a poor agreement for the Mj 5.9 earthquake. In terms of acceleration response spectra, while ground-motion overpredictions were obtained for the Mj 5.9 event, simulated ground motions for the Mj 6.5 earthquake severely underestimate the observations, especially those obtained by the nonlinear approach.Graphical

Estimation of weak and strong ground motions based on diffuse field concept for earthquake for steps 2 and 3 of blind prediction exercise

For the accurate prediction of earthquake ground motions, the investigation of three elements in its process, namely the source, path, and site effects, is crucial, and their accuracy and reliability should be evaluated. The Committee of the 6th Effect of Surface Geology on Seismic Motion Symposium conducted a blind prediction exercise to investigate the accuracy and reliability of estimating the subsurface velocity structure related to the site effect and for predicting weak and strong ground motions. We participated in a blind prediction exercise and estimated the weak and strong ground motions that occurred during the 2016 Kumamoto earthquake sequence. First, the velocity structures at a target site KUMA and our reference site EEB were identified based on the diffuse field concept for earthquake (DFCe). The identified velocity structures at KUMA and EEB reproduced the observed horizontal-to-vertical spectral ratios of earthquakes well and were considered to be able to reproduce the horizontal and vertical site amplification factors based on DFCe. Second, the incident waves at the seismic bedrock of EEB were estimated. For weak and strong ground motions, the spectral amplitudes of the three components of the incident waves were estimated from the vertical motions observed on the surface at EEB assuming the generation of a diffuse field for a single earthquake record and the linearity of the vertical amplification factor, even during strong shaking. Finally, the ground motions on the surface at KUMA were estimated using the incident waves estimated at the seismic bedrock, the inverted velocity structure, and the dynamic deformation characteristics distributed by the organizer of the blind prediction exercise for shallow soil sediments. The spectral amplitudes of the estimated strong ground motions were smaller at around 1 Hz and larger at around 10 Hz than the observed ones. However, the estimated overall spectral levels of all components corresponded to the observed ones. Although several parameters require further research, such as the damping factor of soil sediments, we have shown that the proposed method can estimate the overall seismic ground motions at a target site in the blind.Graphical

Use of Fourier phase characteristics and effective stress analyses for post-earthquake ground motion estimation: application to ESG6 blind prediction steps 2&3 dataset and JMA accelerometric data

After the occurrence of a large earthquake, engineering seismologists are often requested to estimate strong ground motions at a site where strong motion data were not obtained. The goal of this study was to test the ability of a class of methods that uses Fourier phase characteristics for the post-earthquake ground motion estimation, making use of the precious opportunity provide by the ESG6 Blind Prediction Steps 2&3. It was also part of the goal of this study to test the performance of the effective stress analyses to account for soil nonlinearity. In addition to the dataset provided by the organizer of the blind prediction, the author used additional accelerometric data from a nearby JMA site. To simulate ground motions for an M5.9 earthquake at the target site “KUMA”, the Fourier amplitude spectrum was estimated from the spectral ratio between KUMA and the nearby JMA site. The Fourier phase spectrum was approximated by the spectrum of another event at KUMA. Comparison between the estimated and recorded ground motions after the blind prediction revealed that the estimated ground motions were fairly consistent with the observed ground motions, indicating the effectiveness of the method when the rupture process of the target event is simple and the soil nonlinearity at the target site is not significant. To simulate ground motions at KUMA for the M6.5 foreshock and the M7.3 mainshock of the 2016 Kumamoto earthquake sequence, the author conducted effective stress analyses using a program called “FLIP” to account for soil nonlinearity. Comparison between the estimated and recorded ground motions after the blind prediction indicated that the low-frequency components were overestimated and the high-frequency components were underestimated. The strong soil nonlinearity considered in the effective stress analyses was the main cause of the discrepancy. One explanation for this result could be that the nonlinear soil behavior at KUMA during the foreshock and the mainshock was not a strong one. Another explanation could be that the effect of soil nonlinearity was already included in the records at JMA and the effect of soil nonlinearity was double counted in the results submitted by the author.Graphical

Modelling the structures of frameshift-stimulatory pseudoknots from representative bat coronaviruses.

Coronaviruses (CoVs) use -1 programmed ribosomal frameshifting stimulated by RNA pseudoknots in the viral genome to control expression of enzymes essential for replication, making CoV pseudoknots a promising target for anti-coronaviral drugs. Bats represent one of the largest reservoirs of CoVs and are the ultimate source of most CoVs infecting humans, including those causing SARS, MERS, and COVID-19. However, the structures of bat-CoV frameshift-stimulatory pseudoknots remain largely unexplored. Here we use a combination of blind structure prediction followed by all-atom molecular dynamics simulations to model the structures of eight pseudoknots that, together with the SARS-CoV-2 pseudoknot, are representative of the range of pseudoknot sequences in bat CoVs. We find that they all share some key qualitative features with the pseudoknot from SARS-CoV-2, notably the presence of conformers with two distinct fold topologies differing in whether or not the 5' end of the RNA is threaded through a junction, and similar conformations for stem 1. However, they differed in the number of helices present, with half sharing the 3-helix architecture of the SARS-CoV-2 pseudoknot but two containing 4 helices and two others only 2. These structure models should be helpful for future work studying bat-CoV pseudoknots as potential therapeutic targets.

The Magnificent Five Images of Supernova Refsdal: Time Delay and Magnification Measurements

In late 2014, four images of supernova (SN) “Refsdal,” the first known example of a strongly lensed SN with multiple resolved images, were detected in the MACS J1149 galaxy-cluster field. Following the images’ discovery, the SN was predicted to reappear within hundreds of days at a new position ∼8″ away in the field. The observed reappearance in late 2015 makes it possible to carry out Refsdal’s original proposal to use a multiply imaged SN to measure the Hubble constant H 0, since the time delay between appearances should vary inversely with H 0. Moreover, the position, brightness, and timing of the reappearance enable a novel test of the blind predictions of galaxy-cluster models, which are typically constrained only by the positions of multiply imaged galaxies. We have developed a new photometry pipeline that uses DOLPHOT to measure the fluxes of the five images of SN Refsdal from difference images. We apply four separate techniques to perform a blind measurement of the relative time delays and magnification ratios between the last image SX and the earlier images S1–S4. We measure the relative time delay of SX–S1 to be days and the relative magnification to be . This corresponds to a 1.5% precision on the time delay and 17% precision for the magnification ratios and includes uncertainties due to millilensing and microlensing. In an accompanying paper, we place initial and blind constraints on the value of the Hubble constant.

Strong ground motion simulations of the 2016 Kumamoto earthquakes using corrected empirical Green’s functions: methods and results for ESG6 blind prediction Steps 2 and 3 with improved parameters

This paper describes the methods and results of the strong ground motion simulations for three earthquakes from the 2016 Kumamoto earthquake sequence using corrected empirical Green’s functions. The target earthquakes were an aftershock (Mw 5.5), the largest foreshock of the sequence (Mw 6.1), and the mainshock (Mw 7.1). The corrected empirical Green’s function method was used in the simulations. This simulation method combines simple source and path factors with empirical site amplification and phase factors to generate realistic site-specific strong motions. Simulations were originally conducted to participate in blind prediction exercises in ESG6. Although the simulations performed in this study were based on the models submitted to the blind prediction committee, several modifications were made after the blind prediction exercise. First, the observed records at the target site of the blind prediction called KUMA were used to compare observed and synthetic strong ground motions. In addition, a regional spectral inversion was conducted to obtain a more appropriate Q-value and site amplification factor. Synthetic strong motions were found to explain the observed strong ground motions at KUMA and other stations. Comparisons with predictions by other methods and the sensitivity to the rupture scenario were also discussed. These results provide useful information for applying the corrected Green’s function method to strong ground motion simulations.Graphical

SAMPL9 blind predictions for toluene/water partition coefficients using nonequilibrium alchemical approaches.

We present our blind prediction of the toluene-water partition coefficients in the context of the SAMPL9 challenge. For the calculation of the solvation free energies in water, toluene, and 1-octanol, we used an efficient MD-based nonequilibrium alchemical technique relying on the GAFF2 non-polarizable force field. The method is based on the fast-growth of an initially decoupled solute. Canonical sampling of the associated end-state is efficiently obtained by performing a Hamiltonian replica exchange simulation of the gas-phase solute molecule alone, combined with equilibrium configurations of the solvent. Before submitting the prediction, a pre-assessment of the method and of the force field was made by comparing with the known experimental counterpart the calculated octanol-water partition coefficients using different set of atomic charges. The analysis allowed to optimize our blind prediction for the toluene-water partition coefficients, providing at the same time valid clues for improving the performance and reliability of the non-polarizable force field in free energy calculations of drug-receptor systems.

Seismic analysis of failure mechanisms in adjacent interacting stone masonry buildings via rigid block modeling

Groups of contiguous unreinforced stone masonry buildings are a common type of housing seen in old European downtowns. However, assessing their response to earthquakes poses several challenges to the analysts, especially when the housing units are laid out in compact configurations. In fact, in those circumstances a modeling technique that allows for the dynamic interaction of the units is required. The numerical study carried out in this paper makes use of a rigid block modeling approach implemented into in-house software tools to simulate the static behavior and dynamic response of an aggregate stone masonry building. Said approach is used to reproduce the results of bi-axial shake-table tests that were performed on a building prototype as part of the activities organized within the Adjacent Interacting Masonry Structures project, sponsored by the Seismology and Earthquake Engineering Research Infrastructure Alliance for Europe. The experimental mock-up consisted of two adjacent interacting units with matching layout but different height. Two rigid block models are used to investigate the seismic response of the mock-up: a 3D model allowing for the limit analysis of the building on one hand, and a 2D model allowing for the non-linear static pushover and time-history analysis on the other. The 3D model was built for the blind prediction of the test results, as part of a competition organized to test different modeling approaches that are nowadays available to the analysts. The 2D model was implemented once the experimental data were made available, to deepen the investigation by non-linear static pushover and time-history analysis. In both models, the stonework is idealized into an assemblage of rigid blocks interacting via no-tension frictional interfaces, and mathematical programming is utilized to solve the optimization problems associated to the different types of analysis. Differences between numerical and experimental failure mechanisms, base shears, peak ground accelerations, and displacement histories are discussed. Potentialities and limitations of the adopted rigid block models for limit, pushover and time-history analyses are pointed out on the basis of their comparisons with the experimental results.

Simulation of blind pre-diction and post-diction shaking table tests on a masonry building aggregate using a continuum modelling approach

Masonry buildings of historical centres are usually organized within aggregates, whose structural performance against seismic actions is challenging to predict and constitutes still an open issue. The SERA—AIMS (Seismic Testing of Adjacent Interacting Masonry Structures) project was developed to provide additional experimental data by testing a half-scale, two-unit stone masonry aggregate subjected to two horizontal components of dynamic excitation. In this context, this paper investigates the reliability of the modelling approach and the assumptions adopted to generate a three-dimensional continuum finite element model. The work involves two stages, namely a blind pre-diction and a post-diction phase, and proposes a series of simulation analyses including a strategy to shorten the actual records and save computation costs. The study was performed to investigate the extent of uncertainty in modelling for such masonry aggregates in relation to the experimental outcomes. Pre-diction results were proven to be not accurate in terms of predicted displacements and damage patterns. The upgrades introduced for the post-diction analyses, including the calibration of the elastic modulus and the introduction of a non-linear interface between the two units, allowed to improve the outcomes, with reasonable results in terms of predicted base shear force, displacements along Y-direction and damage pattern for the non-linear stage. The overall approach showed to be appropriate for the structural analysis of existing masonry aggregates, but the accurate modelling of this type of structure remains challenging due to the high level of uncertainties.

Shake-table testing of a stone masonry building aggregate: overview of blind prediction study

City centres of Europe are often composed of unreinforced masonry structural aggregates, whose seismic response is challenging to predict. To advance the state of the art on the seismic response of these aggregates, the Adjacent Interacting Masonry Structures (AIMS) subproject from Horizon 2020 project Seismology and Earthquake Engineering Research Infrastructure Alliance for Europe (SERA) provides shake-table test data of a two-unit, double-leaf stone masonry aggregate subjected to two horizontal components of dynamic excitation. A blind prediction was organized with participants from academia and industry to test modelling approaches and assumptions and to learn about the extent of uncertainty in modelling for such masonry aggregates. The participants were provided with the full set of material and geometrical data, construction details and original seismic input and asked to predict prior to the test the expected seismic response in terms of damage mechanisms, base-shear forces, and roof displacements. The modelling approaches used differ significantly in the level of detail and the modelling assumptions. This paper provides an overview of the adopted modelling approaches and their subsequent predictions. It further discusses the range of assumptions made when modelling masonry walls, floors and connections, and aims at discovering how the common solutions regarding modelling masonry in general, and masonry aggregates in particular, affect the results. The results are evaluated both in terms of damage mechanisms, base shear forces, displacements and interface openings in both directions, and then compared with the experimental results. The modelling approaches featuring Discrete Element Method (DEM) led to the best predictions in terms of displacements, while a submission using rigid block limit analysis led to the best prediction in terms of damage mechanisms. Large coefficients of variation of predicted displacements and general underestimation of displacements in comparison with experimental results, except for DEM models, highlight the need for further consensus building on suitable modelling assumptions for such masonry aggregates.

Blind study site assessment of shear-wave velocity at Kumamoto City, Japan, using direct-fitting SPAC methods

The study used data acquired by the ESG6 Blind Prediction Step BP1 Working Group for purposes of facilitating a comparison of interpretation methods for obtaining shear-wave velocity profiles (Vs) from array observations of microtremor (passive seismic) noise. This work uses the direct-fitting MMSPAC method and the krSPAC method on passive seismic data supplied from four seven-station nested triangular arrays with apertures ranging from 1 to 962 m, located within Kumamoto City, Japan. The data allow a useful frequency range of 38 Hz down to 0.3 Hz, giving depth sensitivities from 2 m to > 1000 m. Results are presented as a seven-layer model which has time-averaged shear wave velocities for top 30 m and 300 m of Vs30 = 189 m/s and Vs300 = 584 m/s, respectively. HVSR spectra show two significant peaks at 1.2 and 0.35 Hz which are indicative of major Vs contrasts at depths 26 m and 750 m. The MMSPAC method (and its krSPAC variant) also proved viable on one asymmetric array where four of the seven stations were corrupted by incoherent low-frequency noise. Indications of a lateral variation in Vs could be detected due to the non-concentric geometry of the four arrays, and also from variations in HVSR spectra at stations of the largest array. Further analysis in step 4 of the blind trials, making use of geological data and a Preferred model supplied to participants, showed apparent discrepancies between the Preferred and our BP1 model for the upper 40 m where a supplied PS log appears to be inconsistent with geological data and the blind BP1 model. At low frequencies 0.5–2.5 Hz dispersion data and the BP1 model suggest that use of the Rayleigh effective mode is superior to use of the fundamental mode in deducing the Vs model at depths below 100 m. The method of direct fitting of model and observed SPAC spectra used in MMSPAC also enabled the use of a bandwidth 0.5–38 Hz for interpretation, which is a wider bandwidth than that achieved by other participants for use of passive seismic data alone.Graphical

Subsurface structure identification at the blind prediction site of ESG6 based on the earthquake-to-microtremor ratio method and diffuse field concept for earthquakes

We participated in the blind prediction exercise organized by the committee of the blind prediction experiment during the 6th International Symposium on Effects of Surface Geology on Seismic Motion (CBP-ESG6). In response to the committee's request, we identified the ground velocity structure from microtremors observed at a target site as the first step of the exercise. First, we calculated the horizontal-to-vertical spectral ratio of microtremors (MHVR) at the target site from the distributed microtremor data collected in the vicinity of the target site in Kumamoto Prefecture. Then, we converted the MHVR into a pseudo horizontal-to-vertical spectral ratio of earthquake (pEHVR) using the previously proposed and validated earthquake-to-microtremor ratio (EMR) method, where an empirically obtained EMR is used to convert MHVR into pEHVR. Next, we inverted the S-wave and P-wave velocity structures based on the pEHVR and the diffuse field concept for earthquakes. The theoretical EHVR calculated from the identified velocity structure reproduced the pEHVR quite well in the frequency range of 0.1–22 Hz. After the collection of the blind prediction results by all the participants, the CBP-ESG6 released the observed earthquake records, a preferred model based on the P–S logging data from the in-situ borehole measurement combined with the generic deeper structure, and the average of all the predicted structures by the participants. Notably, our inverted structure was found to be close to the preferred model and the averaged one of all the blind prediction participants, despite some minor differences in the horizontal site amplification factor around the maximum peak frequency at 0.8–1 Hz.Graphical

Protein structure prediction using the evolutionary algorithm USPEX.

Protein structure prediction is one of major problems of modern biophysics: current attempts to predict the tertiary protein structure from amino acid sequence are successful mostly when the use of big data and machine learning allows one to reduce the "prediction problem" to the "problem of recognition". Compared with recent successes of deep learning, classical predictive methods lag behind in their accuracy for the prediction of stable conformations. Therefore, in this work we extended the evolutionary algorithm USPEX to predict protein structure based on global optimization starting with the amino acid sequence. Moreover, we compared frequently used force fields for the task of protein structure prediction. Protein structure relaxation and energy calculations were performed using Tinker (with several different force fields) and Rosetta (with REF2015 force field) codes. To create new protein structure models in the USPEX algorithm, we developed novel variation operators. The test of the new method on seven proteins having (for simplicity) no cis-proline (with ω ≈ 0°) residues, and a length of up to 100 residues, revealed that our algorithm predicts tertiary structures of proteins with high accuracy. The comparison of the final potential energies of the predicted protein structures obtained using the USPEX and the Rosetta Abinitio approach showed that in most cases the developed algorithm found structures with close or even lower energy (Amber/Charmm/Oplsaal) and scoring function (REF2015). While USPEX has clearly demonstrated its ability to find very deep energy minima, our study showed that the existing force fields are not sufficiently accurate for accurate blind prediction of protein structures without further experimental verification.

SpinDrop: Dropout-Based Bayesian Binary Neural Networks With Spintronic Implementation

Neural Networks (NNs) provide an effective solution in numerous application domains, including autonomous driving and medical applications. Nevertheless, NN predictions can be incorrect if the input sample is outside of the training distribution or contaminated by noise. Consequently, quantifying the uncertainty of the NN prediction allows the system to make more insightful decisions by avoiding blind predictions. Therefore, uncertainty quantification is crucial for a variety of applications, including safety-critical applications. Bayesian NN (BayNN) using Dropout-based approximation provides a systematic approach for estimating the uncertainty of predictions. Despite such merit, BayNNs are not suitable for implementation in an embedded device or able to meet high-performance demands for certain applications. Computation in-memory (CiM) architecture with emerging non-volatile memories (NVMs) is a great candidate for high-performance and low-power acceleration BayNNs in hardware. Among NVMs, Magnetic Tunnel Junction (MTJ) offer many benefits, but they also suffer from various non-idealities and limited bit-level resolution. As a result, binarizing BayNNs is an attractive option that can directly implement BayNN into a CiM architecture and able to achieve benefits of both CiM architecture and BayNNs at the same time. Conventional in-memory hardware implementations emphasize conventional NNs, which can only make predictions, and do not account for both device and input uncertainty, thus, reducing both reliability and performance. In this paper, we propose for the first time Binary Bayesian NNs (BinBayNN) with an end-to-end approach (from algorithmic level to device level) for their implementation. Our approach takes the inherent stochastic properties of MTJs as a feature to implement Dropout-based Bayesian Neural Networks. We provide an extensive evaluation of our approach from the device level up to the algorithmic level on various benchmark datasets.

Towards a predictive, physics-based friction model for the dynamics of jointed structures

Bolted connections are ubiquitous in mechanical designs and pose a significant challenge to understanding and predicting the vibration response of assembled structures. The present paper develops a physics-based rough contact model of the frictional interactions within a joint. This model sums over the probable interactions of asperities – defined as locally maximum surface features – to determine the contact forces. Here, the tangential contact forces vary smoothly between sticking and slipping and allow the model to better capture the qualitative trends of experimental amplitude dependent frequency and damping than previous studies. Furthermore, the novel model is generalized to allow for arbitrarily varying normal pressure including potential separation to better represent the interfacial dynamics. This includes developing a new, computationally tractable approximation to the analytical Mindlin partial slip solution for tangential loading of contacting spheres. The results highlight the importance of accurately characterizing the as-built topology of the interface, the plastic behavior of the contacting asperities, the relevant length scale of asperities, and the eccentricity of asperities. A predictive friction coefficient based on plasticity provides a poor match to experiments, so fitting the friction coefficient is also considered. Numerical results are compared to experiments on the Brake-Reuß Beam to assess the predictive potential of the models. While blind predictions over-predict the slip limit, the current model presents a significant improvement in physics-based modeling and highlights areas for ongoing research.

Preliminary nonlinear analysis of a scaled masonry building under shaking test for the blind prediction of the SERA AIMS project

Experimental tests performed on scaled masonry buildings provide key information to improve the knowledge under seismic actions. Masonry buildings behavior is strongly influenced by their physical and mechanical parameters under dynamic actions. In fact, the actual structural behavior is very complex to predict due to the significant variability of the input parameters and the strong heterogeneity of masonry. Furthermore, the behavior of masonry buildings is often influenced by the interaction with adjacent building units. In this context, the SERA AIMS project aims to improve knowledge on the interaction between adjacent buildings. In order to assess the seismic capacity of masonry structures and their damage evolution, nonlinear models often require a numerical calibration of nonlinear parameters. Simplified Finite Element (FE) models, with some very simple assumptions, can be more suitable for complex problems like as the interaction between adjacent building aggregates. The low initial knowledge level in the SERA AIMS blind competition favored simple assumptions. The availability of simple models allowed to perform consecutive time histories including the cumulative effects of previous signals. In fact, the tested prototype was subjected to many replicas. The masonry structure and the crucial interfaces between the units have been modelled by means of nonlinear elements according to the reduced knowledge level at the blind prediction stage. The main goal was to estimate the key information of a masonry building under seismic action like as: triggering and the type of damage at the most stressed areas and therefore the load threshold at which evident damage is expected. Global FE model provides information on the global behaviour (in plane behaviour), while, according to the failure models typically found in masonry buildings, kinematic analyses have been performed to assess the out of plane (local) behaviour, too. The numerical results obtained by the preliminary analysis have been compared with the experimentally detected damages. The simplified approach, based on limited information without calibration, discussed in this paper, represents a useful support tool to design dynamic tests on full-scale or scaled masonry buildings, but also to assess the vulnerability of real masonry structures.