Frictional properties of faults : from observation on the Longitudinal Valley Fault, Taiwan, to dynamic simulations

Faults can slip either aseismically or through episodic seismic ruptures, but we still do not understand the factors which determine the partitioning between these two modes of slip. This challenge can now be addressed thanks to the dense set of geodetic and seismological networks that have been deployed in various areas with active tectonics. The data from such networks, as well as modern remote sensing techniques, indeed allow documenting of the spatial and temporal variability of slip mode and give some insight. This is the approach taken in this study, which is focused on the Longitudinal Valley Fault (LVF) in Eastern Taiwan. This fault is particularly appropriate since the very fast slip rate (about 5 cm/yr) is accommodated by both seismic and aseismic slip. Deformation of anthropogenic features shows that aseismic creep accounts for a significant fraction of fault slip near the surface, but this fault also released energy seismically, since it has produced five M_w>6.8 earthquakes in 1951 and 2003. Moreover, owing to the thrust component of slip, the fault zone is exhumed which allows investigation of deformation mechanisms. In order to put constraint on the factors that control the mode of slip, we apply a multidisciplinary approach that combines modeling of geodetic observations, structural analysis and numerical simulation of the "seismic cycle". Analyzing a dense set of geodetic and seismological data across the Longitudinal Valley, including campaign-mode GPS, continuous GPS (cGPS), leveling, accelerometric, and InSAR data, we document the partitioning between seismic and aseismic slip on the fault. For the time period 1992 to 2011, we found that about 80-90% of slip on the LVF in the 0-26 km seismogenic depth range is actually aseismic. The clay-rich Lichi M\'elange is identified as the key factor promoting creep at shallow depth. Microstructural investigations show that deformation within the fault zone must have resulted from a combination of frictional sliding at grain boundaries, cataclasis and pressure solution creep. Numerical modeling of earthquake sequences have been performed to investigate the possibility of reproducing the results from the kinematic inversion of geodetic and seismological data on the LVF. We first investigate the different modeling strategy that was developed to explore the role and relative importance of different factors on the manner in which slip accumulates on faults. We compare the results of quasi dynamic simulations and fully dynamic ones, and we conclude that ignoring the transient wave-mediated stress transfers would be inappropriate. We therefore carry on fully dynamic simulations and succeed in qualitatively reproducing the wide range of observations for the southern segment of the LVF. We conclude that the spatio-temporal evolution of fault slip on the Longitudinal Valley Fault over 1997-2011 is consistent to first order with prediction from a simple model in which a velocity-weakening patch is embedded in a velocity-strengthening area.

Extreme learning machines for virtual metrology and etch rate prediction

Virtual metrology (VM) aims to predict metrology values using sensor data from production equipment and physical metrology values of preceding samples. VM is a promising technology for the semiconductor manufacturing industry as it can reduce the frequency of in-line metrology operations and provide supportive information for other operations such as fault detection, predictive maintenance and run-to-run control. Methods with minimal user intervention are required to perform VM in a real-time industrial process. In this paper we propose extreme learning machines (ELM) as a competitive alternative to popular methods like lasso and ridge regression for developing VM models. In addition, we propose a new way to choose the hidden layer weights of ELMs that leads to an improvement in its prediction performance.

Motor fault detection and diagnosis using a hybrid FMM-CART model with online learning

In this paper, a hybrid online learning model that combines the fuzzy min-max (FMM) neural network and the Classification and Regression Tree (CART) for motor fault detection and diagnosis tasks is described. The hybrid model, known as FMM-CART, incorporates the advantages of both FMM and CART for undertaking data classification (with FMM) and rule extraction (with CART) problems. In particular, the CART model is enhanced with an importance predictor-based feature selection measure. To evaluate the effectiveness of the proposed online FMM-CART model, a series of experiments using publicly available data sets containing motor bearing faults is first conducted. The results (primarily prediction accuracy and model complexity) are analyzed and compared with those reported in the literature. Then, an experimental study on detecting imbalanced voltage supply of an induction motor using a laboratory-scale test rig is performed. In addition to producing accurate results, a set of rules in the form of a decision tree is extracted from FMM-CART to provide explanations for its predictions. The results positively demonstrate the usefulness of FMM-CART with online learning capabilities in tackling real-world motor fault detection and diagnosis tasks. © 2014 Springer Science+Business Media New York.

Grid Global Behavior Prediction

Complexity has always been one of the most important issues in distributed computing. From the first clusters to grid and now cloud computing, dealing correctly and efficiently with system complexity is the key to taking technology a step further. In this sense, global behavior modeling is an innovative methodology aimed at understanding the grid behavior. The main objective of this methodology is to synthesize the grid's vast, heterogeneous nature into a simple but powerful behavior model, represented in the form of a single, abstract entity, with a global state. Global behavior modeling has proved to be very useful in effectively managing grid complexity but, in many cases, deeper knowledge is needed. It generates a descriptive model that could be greatly improved if extended not only to explain behavior, but also to predict it. In this paper we present a prediction methodology whose objective is to define the techniques needed to create global behavior prediction models for grid systems. This global behavior prediction can benefit grid management, specially in areas such as fault tolerance or job scheduling. The paper presents experimental results obtained in real scenarios in order to validate this approach.

Earthquake prediction: the interaction of public policy and science.

Earthquake prediction research has searched for both informational phenomena, those that provide information about earthquake hazards useful to the public, and causal phenomena, causally related to the physical processes governing failure on a fault, to improve our understanding of those processes. Neither informational nor causal phenomena are a subset of the other. I propose a classification of potential earthquake predictors of informational, causal, and predictive phenomena, where predictors are causal phenomena that provide more accurate assessments of the earthquake hazard than can be gotten from assuming a random distribution. Achieving higher, more accurate probabilities than a random distribution requires much more information about the precursor than just that it is causally related to the earthquake.

Intermediate- and long-term earthquake prediction.

Progress in long- and intermediate-term earthquake prediction is reviewed emphasizing results from California. Earthquake prediction as a scientific discipline is still in its infancy. Probabilistic estimates that segments of several faults in California will be the sites of large shocks in the next 30 years are now generally accepted and widely used. Several examples are presented of changes in rates of moderate-size earthquakes and seismic moment release on time scales of a few to 30 years that occurred prior to large shocks. A distinction is made between large earthquakes that rupture the entire downdip width of the outer brittle part of the earth's crust and small shocks that do not. Large events occur quasi-periodically in time along a fault segment and happen much more often than predicted from the rates of small shocks along that segment. I am moderately optimistic about improving predictions of large events for time scales of a few to 30 years although little work of that type is currently underway in the United States. Precursory effects, like the changes in stress they reflect, should be examined from a tensorial rather than a scalar perspective. A broad pattern of increased numbers of moderate-size shocks in southern California since 1986 resembles the pattern in the 25 years before the great 1906 earthquake. Since it may be a long-term precursor to a great event on the southern San Andreas fault, that area deserves detailed intensified study.

Scale dependence in earthquake phenomena and its relevance to earthquake prediction.

The recent discovery of a low-velocity, low-Q zone with a width of 50-200 m reaching to the top of the ductile part of the crust, by observations on seismic guided waves trapped in the fault zone of the Landers earthquake of 1992, and its identification with the shear zone inferred from the distribution of tension cracks observed on the surface support the existence of a characteristic scale length of the order of 100 m affecting various earthquake phenomena in southern California, as evidenced earlier by the kink in the magnitude-frequency relation at about M3, the constant corner frequency for earthquakes with M below about 3, and the sourcecontrolled fmax of 5-10 Hz for major earthquakes. The temporal correlation between coda Q-1 and the fractional rate of occurrence of earthquakes in the magnitude range 3-3.5, the geographical similarity of coda Q-1 and seismic velocity at a depth of 20 km, and the simultaneous change of coda Q-1 and conductivity at the lower crust support the hypotheses that coda Q-1 may represent the activity of creep fracture in the ductile part of the lithosphere occurring over cracks with a characteristic size of the order of 100 m. The existence of such a characteristic scale length cannot be consistent with the overall self-similarity of earthquakes unless we postulate a discrete hierarchy of such characteristic scale lengths. The discrete hierarchy of characteristic scale lengths is consistent with recently observed logarithmic periodicity in precursory seismicity.

Intermediate-term earthquake prediction.

An earthquake of magnitude M and linear source dimension L(M) is preceded within a few years by certain patterns of seismicity in the magnitude range down to about (M - 3) in an area of linear dimension about 5L-10L. Prediction algorithms based on such patterns may allow one to predict approximately 80% of strong earthquakes with alarms occupying altogether 20-30% of the time-space considered. An area of alarm can be narrowed down to 2L-3L when observations include lower magnitudes, down to about (M - 4). In spite of their limited accuracy, such predictions open a possibility to prevent considerable damage. The following findings may provide for further development of prediction methods: (i) long-range correlations in fault system dynamics and accordingly large size of the areas over which different observed fields could be averaged and analyzed jointly, (ii) specific symptoms of an approaching strong earthquake, (iii) the partial similarity of these symptoms worldwide, (iv) the fact that some of them are not Earth specific: we probably encountered in seismicity the symptoms of instability common for a wide class of nonlinear systems.

Rock friction and its implications for earthquake prediction examined via models of Parkfield earthquakes.

The friction of rocks in the laboratory is a function of time, velocity of sliding, and displacement. Although the processes responsible for these dependencies are unknown, constitutive equations have been developed that do a reasonable job of describing the laboratory behavior. These constitutive laws have been used to create a model of earthquakes at Parkfield, CA, by using boundary conditions appropriate for the section of the fault that slips in magnitude 6 earthquakes every 20-30 years. The behavior of this model prior to the earthquakes is investigated to determine whether or not the model earthquakes could be predicted in the real world by using realistic instruments and instrument locations. Premonitory slip does occur in the model, but it is relatively restricted in time and space and detecting it from the surface may be difficult. The magnitude of the strain rate at the earth's surface due to this accelerating slip seems lower than the detectability limit of instruments in the presence of earth noise. Although not specifically modeled, microseismicity related to the accelerating creep and to creep events in the model should be detectable. In fact the logarithm of the moment rate on the hypocentral cell of the fault due to slip increases linearly with minus the logarithm of the time to the earthquake. This could conceivably be used to determine when the earthquake was going to occur. An unresolved question is whether this pattern of accelerating slip could be recognized from the microseismicity, given the discrete nature of seismic events. Nevertheless, the model results suggest that the most likely solution to earthquake prediction is to look for a pattern of acceleration in microseismicity and thereby identify the microearthquakes as foreshocks.

A facilities maintenance management process based on degradation prediction using sensed data

Energy efficiency and user comfort have recently become priorities in the Facility Management (FM) sector. This has resulted in the use of innovative building components, such as thermal solar panels, heat pumps, etc., as they have potential to provide better performance, energy savings and increased user comfort. However, as the complexity of components increases, the requirement for maintenance management also increases. The standard routine for building maintenance is inspection which results in repairs or replacement when a fault is found. This routine leads to unnecessary inspections which have a cost with respect to downtime of a component and work hours. This research proposes an alternative routine: performing building maintenance at the point in time when the component is degrading and requires maintenance, thus reducing the frequency of unnecessary inspections. This thesis demonstrates that statistical techniques can be used as part of a maintenance management methodology to invoke maintenance before failure occurs. The proposed FM process is presented through a scenario utilising current Building Information Modelling (BIM) technology and innovative contractual and organisational models. This FM scenario supports a Degradation based Maintenance (DbM) scheduling methodology, implemented using two statistical techniques, Particle Filters (PFs) and Gaussian Processes (GPs). DbM consists of extracting and tracking a degradation metric for a component. Limits for the degradation metric are identified based on one of a number of proposed processes. These processes determine the limits based on the maturity of the historical information available. DbM is implemented for three case study components: a heat exchanger; a heat pump; and a set of bearings. The identified degradation points for each case study, from a PF, a GP and a hybrid (PF and GP combined) DbM implementation are assessed against known degradation points. The GP implementations are successful for all components. For the PF implementations, the results presented in this thesis find that the extracted metrics and limits identify degradation occurrences accurately for components which are in continuous operation. For components which have seasonal operational periods, the PF may wrongly identify degradation. The GP performs more robustly than the PF, but the PF, on average, results in fewer false positives. The hybrid implementations, which are a combination of GP and PF results, are successful for 2 of 3 case studies and are not affected by seasonal data. Overall, DbM is effectively applied for the three case study components. The accuracy of the implementations is dependant on the relationships modelled by the PF and GP, and on the type and quantity of data available. This novel maintenance process can improve equipment performance and reduce energy wastage from BSCs operation.