DynNet: Physics-based neural architecture design for nonlinear structural response modeling and prediction

Wavelets as a Powerful Signal Processing Tool The principles of wavelets can be applied to a range of problems in civil engineering structures, such as earthquake-induced vibration analysis, bridge vibrations, and damage identification. This book is particularly useful for graduate students and researchers in vibration analysis, especially those dealing with random vibrations. Wavelet Analysis in Civil Engineering explains the importance of wavelets in analyzing nonstationarities in ground motions. The example of a tank is considered to develop the problem and the model (based on linear assumptions) and several case studies are explored—fixed base, flexible base, lateral and rocking motions of foundations, with and without fluid—to explain how to account for ground motion nonstationarities. Bridge vibrations caused by vehicle passage are explored, as is structural damage identification. Wavelet analytic techniques starting from single degree of freedom systems to multiple degree of freedom systems are set out and detailed solutions of more complicated problems involving soil and fluid interactions are presented. Separate chapters have been devoted to explaining the basic principles of the wavelet-based random nonstationary vibration analysis of nonlinear systems, including probabilistic analysis. Comprised of seven chapters, this text: Introduces the concept and utility of wavelet transform Describes the discretization of ground motions using wavelet coefficients Explains how to characterize nonstationary ground motions using statistical functionals of wavelet coefficients of seismic accelerations Develops the formulation of a linear single-degree-of-freedom system Shows stepwise development of the formulation of a structure idealized as a linear multi-degree-of-freedom system in terms of wavelet coefficients Defines wavelet domain formulation of a nonlinear single-degree-of-freedom system Introduces the concept of probability in wavelet-based theoretical formulation of a nonlinear two-degree-of-freedom system Covers a variety of case studies highlighting diverse applications Wavelet Analysis in Civil Engineering explains the importance of wavelets in terms of non-stationarities of ground motions, explores the application of wavelet analytic techniques, and is an excellent resource for users addressing wavelets for the first time.

Pseudo-wavelets were proposed for system identification of linear systems to determine their natural frequencies and modal damping ratios. The approach provides excellent results for single-degree-of-freedom (SDOF) oscillators but significant errors might be observed for multiple-degree-of- freedom (MDOF) systems due to interference between different modes if the pseudo-wavelet transform (PWT) is directly applied. IN this paper a truncated pseudo-wavelet transform was employed to improve the accuracy for MDOF systems. A peak of the Fourier amplitude spectrum of the response was located and truncated in its neighborhood to isolate the peak. The frequency amplitude response function of a linear SDOF system was selected as the pseudo-wavelet. A truncated pseudo-wavelet transform was performed and its maximum value was located on a contour map of the PWT coefficients. The associated frequency and bandwidth of the MDOF system may be identified by the matching values of the scale and shift parameters. The methodology is illustrated for linear MDOF systems to identify their system parameters and good estimates were obtained. Noise effects were discussed. The approach can be applied for structural health monitoring. Change in system parameters using any two segments of response data may suggest occurrence of structural damages.

Deep long short-term memory networks for nonlinear structural seismic response prediction

Risks to cultural heritage and related losses should be mitigated before disasters such as earthquakes happen. Risks can be addressed by various means, from raising cultural attention of authorities to documenting artistic or historical value of an object. The main contribution of structural engineering to cultural heritage concerns regular maintenance and monitoring for risk reduction. Risk mitigation of historical buildings, as a part of more general concept of conservation, involves different disciplines. Teams need to be multidisciplinary and information deriving from historical, metric, stylistic, structural, seismic, geotechnical and physical analysis may contribute to achievement of an overall comprehension of cultural assets. The synergic action of characterization and monitoring techniques are essential factors to understand, on one hand mechanisms and consequences of degradation and, on other hand, to provide reliable and well-grounded guidelines for definition of technical interventions to prevent/ to stop degradation phenomena, to restore functionality and use of historical building/ artifact, or to predict, mitigate and even control response to accidental events, including strong motions. In this field, an important role is played not only by analytical aspects, but also by development and validation of innovative materials and systems for conservation. International deontological guidelines on conservation of cultural heritage define structural rehabilitation of heritage structures as cure of a sick person, hence the heritage structures require anamnesis, diagnosis, therapy and controls, corresponding respectively to searches for significant data and information, individuation of causes of damage and decay, choice of remedial measures and control of efficiency of interventions. Moreover, same codes state that: the best therapy is preventive maintenance, which can only be achieved via monitoring of structure. In this thesis work a few topical issues of structural modelling, monitoring and assessment of historic masonry buildings were addressed, with particular emphasis on dynamic testing and identification. The possible connections with other disciplines are analysed and discussed throughout text. In this framework, outline of thesis includes an introductive first chapter in which context established by most recent codes and guidelines concerning architectural heritage conservation is duly reviewed and analysed. The importance of attaining a knowledge of structure is also discussed. The second chapter sets up scene, in which it introduces principal issues of seismic risk and safety assessment of architectural heritage. Firstly, a brief overview is given of seismic risk and of geological and geotechnical aspects as related to ancient heritage. Successively, viability of performance-based approaches, for application to seismic assessment of architectural heritage, is discussed also in light of a few recent proposals. In this context, fundamentals of structural health monitoring are also reported. Chapter 3 is intended to stress importance of modal testing as an effective tool for ancient structures characterisation, so it starts with a state-of-the-art on linear system identification methods with emphasis on output-only techniques. In particular, time domain and joint time-frequency domain identification techniques are introduced and deeply analysed. Model updating is then addressed and its connection with operational modal analysis is underlined. Finally, a few noteworthy examples of linear identification and model updating of architectural heritage structures are reported. Chapter 4 is about dynamic and seismic behaviour of domes. The coverage focuses on three ideal benchmarks on reconciling geometric survey with dynamic monitoring. The analyses concerned structures with oval shape domes, such as Sanctuary of Vicoforte, S. Caterina in Casale Monferrato and S. Agostino in L'Aquila. The final products are virtual models which were enabled to predict linear dynamic response under earthquake excitation. Chapter 5 inspects modelling strategies suited for masonry under intense seismic excitations. The state-of-the-art covers both models for equivalent static analysis and models which operates in dynamics. A model allowing for stiffness degradation, pinching and hysteresis is then proposed, whose formulation admits extensions to multiple degree of freedom systems. The proposal originates from well-known Bouc-Wen model. Chapter 6 deals with non-linear identification methods. In perspective, also non-linear identification is expected to become a powerful tool in context of structural and seismic reliability assessment, especially in light of increasing levels of knowledge and prediction capabilities which recent standards strive for. Unfortunately, non-linear identification is to date a specialized and challenging matter, and it has been seldom applied to full-scale structures. In this chapter, special emphasis is given to on-line implementations, with several numerical examples showing potential of non-linear as well as hysteretic system identification. The last chapter presents an experimental application of non-linear identification. A scaled model of a two-span masonry arch bridge has been artificially damaged and monitored at each damage step. A non-linear identification has been performed from shaker tests data. Results of experimental campaign will be used to corroborate a non-linear and hysteretic model of bridge endowed with prediction capabilities

Improving classification accuracy using data augmentation on small data sets

In order to enhance the fidelity, convenient and flexibility of swinging motion, the structure of incompletely restrained cablesuspended system controlled by two drums was proposed, and the dynamic response of the system under swinging and heaving motion were investigated in this paper. The cables are spatially discretized using the assumed modes method and the system equations of motion are derived by Lagrange equations of the first kind. Based on geometric boundary conditions and linear complementary theory, the differential algebraic equations are transformed to a set of classical difference equations. Nonlinear dynamic behavior occurs under certain range of rotational velocity and frequency. The results show that asynchronous motion of suspension platform is easily caused imbalance for cable tension. Dynamic response of different swing frequencies were obtained via power frequency analysis, which could be used in the selection of the working frequency of the swing motion. The work will contribute to a better understanding of the swing frequency, cable tension and posture with dynamic characteristics of unilateral geometric and kinematic constraints in this system, and it is also useful to investigate the accuracy and reliability of instruments in future.

An improvement over Fourier transform to enhance its performance for frequency content evaluation of seismic signals

The main goal of this research was to develop and present a general, efficient, mathematical, and theoretical based methodology to model nonlinear forced-vibrating mechanical systems from time series measurements. A system identification modeling methodology for forced dynamical systems is presented based on a dynamic system theory and a nonlinear time series analysis that employ phase space reconstruction (delay vector embedding) in modeling dynamical systems from time series data using time-delay neural networks. The first part of this work details the modeling methodology, including background on dynamic systems, phase space reconstruction, and neural networks. In the second part of this work, the methodology is evaluated based on its ability to model selected analytical lumped-parameter forced-vibrating dynamic systems, including an example of a linear system predicting lumped mass displacement subjected to a displacement forcing function. The work discusses the application to nonlinear systems, multiple degree of freedom systems, and multiple input systems. The methodology is further evaluated on its ability to model an analytical passenger rail car predicting vertical wheel∕rail force using a measured vertical rail profile as the input function. Studying the neural modeling methodology using analytical systems shows the clearest observations from results, providing prospective users of this tool an understanding of the expectations and limitations of the modeling methodology.

This paper presents a new family of explicit time integration algorithms with controllable numerical dissipation for structural dynamic problems by utilizing the discrete control theory. Firstly, the equilibrium equation of the implicit Yu-[Formula: see text] algorithm is adopted, and the recursive formulas of velocity and displacement for the explicit CR algorithm are used in the algorithms. Then, the transfer function and characteristic equation of the algorithms with integration coefficients are obtained by the [Formula: see text] transformation. Furthermore, their integration coefficients are derived according to the poles condition. It was indicated that the proposed algorithms possess the advantages of second-order accuracy, self-starting, and unconditional stability for linear systems and nonlinear systems with softening stiffness. The numerical dissipation of the algorithms is controlled by the spectral radius at infinity [Formula: see text]. It was also shown that the proposed algorithms have the same poles as the Yu-[Formula: see text] algorithm, and thus the same numerical properties. Compared with the implicit Yu-[Formula: see text] algorithm, the proposed algorithms are explicit in terms of both the displacement and velocity formulas. Finally, the effectiveness of the proposed algorithms in reducing the undesired participation of higher modes for solving the dynamic responses of linear and nonlinear systems has been demonstrated.

We investigate the average response to small external perturbations for discrete dynamical systems with chaotic attractors. The average linear response satisfies a fluctuation theorem, and in general diverges exponentially in the long-time limitt→∞. It vanishes identically for allt>0 only in a number of special cases including the logistic model with bifurcation parameter α=4. The nonlinear response turns out to be crucial. Its average is analyzed for a time-localized (pulse) perturbation. Near the onset of chaos it exhibits universal scaling behaviour expressed by two critical exponents. For static perturbations the resulting dynamics is extremely sensitive to the perturbation strength.

Predicting nonlinear dynamic response of internal cantilever beam system on a steadily rotating ring

Among the irregular responses of nonlinear dynamic systems, chaotic responses of nonlinear systems are probably the most attractive phenomena along with the new observations in the last decades. A nonlinear deterministic system may behavior chaotically under regular such as periodic excitations. Regular motion of a system subjected to periodic exertions is usually periodic. In contrast with regular motions, final states of chaotic vibrations are extremely nonperiodic. This research is to analyzing the irregular behavior of dynamic systems with implementation of a newly developed criterion named Periodicity-Ratio. The development of a methodology for diagnosing the irregular motions from the regular motions of a dynamic system is presented. The Periodicity-Ratio describes the degree of periodicity of motion and can be conveniently used to distinguish a nonperiodic motion from a regular vibration or oscillation and to diagnose whether or not a motion is chaotic and the other irregular responses of the nonlinear dynamic systems, without plotting any figures. The analyses on the irregular behavior of nonlinear dynamic systems with the implementation of the Periodicity-Ratio will be demonstrated.

In order to determine the pile internal forces of wharf under simultaneous seismic excitation in orthogonal horizontal directions with ease, the total internal forces can be obtained by multiplying the pile forces of two‐dimensional model under transverse earthquake loading with a magnification factor. The factors were computed from the seismic response of wharves having various aspect ratios (length to width) and eccentricity ratios when subjected to 30 pairs of bi‐directional earthquake records. Mean magnification factors and their corresponding dispersion are presented for all ground motion ensembles. The influence of aspect ratio and eccentricity ratio is evaluated and discussed. It is concluded that the magnification factor can be expressed as the function of aspect ratio and eccentricity ratio, and follows lognormal distribution. The proposed equation of magnification factor can be an alternative for internal force analysis of wharves under bi‐directional ground motion.

Effective computational prediction of complex or novel molecule syntheses can greatly help organic and medicinal chemistry. Retrosynthetic analysis is a method employed by chemists to predict synthetic routes to target compounds. The target compounds are incrementally converted into simpler compounds until the starting compounds are commercially available. However, predictions based on small chemical datasets often result in low accuracy due to an insufficient number of samples. To address this limitation, we introduced transfer learning to retrosynthetic analysis. Transfer learning is a machine learning approach that trains a model on one task and then applies the model to a related but different task; this approach can be used to solve the limitation of few data. The unclassified USPTO-380K large dataset was first applied to models for pretraining so that they gain a basic theoretical knowledge of chemistry, such as the chirality of compounds, reaction types and the SMILES form of chemical structure of compounds. The USPTO-380K and the USPTO-50K (which was also used by Liu et al.) were originally derived from Lowe’s patent mining work. Liu et al. further processed these data and divided the reaction examples into 10 categories, but we did not. Subsequently, the acquired skills were transferred to be used on the classified USPTO-50K small dataset for continuous training and retrosynthetic reaction tests, and the pretrained accuracy data were simultaneously compared with the accuracy of results from models without pretraining. The transfer learning concept was combined with the sequence-to-sequence (seq2seq) or Transformer model for prediction and verification. The seq2seq and Transformer models, both of which are based on an encoder-decoder architecture, were originally constructed for language translation missions. The two algorithms translate SMILES form of structures of reactants to SMILES form of products, also taking into account other relevant chemical information (chirality, reaction types and conditions). The results demonstrated that the accuracy of the retrosynthetic analysis by the seq2seq and Transformer models after pretraining was significantly improved. The top-1 accuracy (which is the accuracy rate of the first prediction matching the actual result) of the Transformer-transfer-learning model increased from 52.4% to 60.7% with greatly improved prediction power. The model’s top-20 prediction accuracy (which is the accuracy rate of the top 20 categories containing actual results) was 88.9%, which represents fairly good prediction in retrosynthetic analysis. In summary, this study proves that transferring learning between models working with different chemical datasets is feasible. The introduction of transfer learning to a model significantly improved prediction accuracy and, especially, assisted in small dataset based reaction prediction and retrosynthetic analysis.

Accurate prediction of the internal corrosion rates of oil and gas pipelines is an urgent need in the oil industry at present, and laboratory-scale experiments are of great significance in achieving this goal. However, obtaining large amounts of experimental data is both costly and time-consuming. Here, the possibility of using ensemble learning methods to reduce the corrosion rate prediction errors based on small datasets is investigated. The work presented in this paper focuses on the prediction accuracy and the influencing factors of bagging and boosting methods under 99 sets of laboratory data. The research we have done suggests that the bagging algorithm outperforms the boosting algorithm in scenarios where small samples of discrete data are used, and that the number, dimensionality, and dispersion of the training samples all have an impact on the prediction results. Further, the prediction error values obtained via ensemble learning methods are smaller compared to the results obtained using traditional empirical models such as Norsok M506. It is reasonable to conclude that the use of bagging algorithms in ensemble learning is suitable for obtaining accurate corrosion rate predictions for small laboratory datasets.