Prof. Weidong Zhu

University of Maryland Baltimore Country, USA

  Modal parameters of structures, such as natural frequencies and mode shapes, have been widely used for vibration-based structural damage detection. A model-based damage detection method that uses changes in natural frequencies to detect damage has advantages over conventional nondestructive tests in detecting various types of damage using minimum measurement data. Two major challenges associated with applications of the model-based method to practical engineering structures are addressed: accurate modeling of test structures and the development of a robust inverse algorithm to detect damage, which are defined as the forward and inverse problems associated with the damage detection method, respectively. To resolve the forward problem, new physics-based finite element modeling techniques for fillets in thin-walled beams and bolted joints are developed, so that complex structures with thin-walled beams and/or bolted joints can be accurately modeled with a reasonable model size. To resolve the inverse problem, a robust iterative algorithm using a trust-region method, called the Levenberg-Marquardt (LM) method is developed to accurately detect locations and extent of damage using a minimum number of measured natural frequencies. The LM method can ensure global convergence of iterations in solving severely under-determined system equations and deal with damage detection problems with relatively large modeling error and measurement noise. The vibration-based damage detection method developed is applied to various structures including lightning masts, a space frame structure, and a pipeline. The locations and extent of damage can be successfully detected in experimental damage detection. In the numerical simulation where there are no modeling error and measurement noise, exact locations and extent of damage can be detected.

  Besides the model-based method, non-model-based methods that use vibration shapes to identify damage in beams and plates are introduced. Curvature mode shapes and continuous wavelet transforms of mode shapes of a damaged beam are compared with those from polynomial fits with proper orders to yield curvature damage indices and continuous wavelet transform damage indices, respectively, to identify damage. A non-model-based method that uses mode shapes and one that uses principal, mean and Gaussian curvature mode shapes are introduced to identify damage in plates. A new multi-scale differential geometry scheme is developed to calculate curvature mode shapes. Comparing a mode shape of a damaged plate with that from a polynomial that fits the mode shape can yield a mode shape damage index to identify damage, and comparing curvature mode shapes associated with a mode shape of a damaged plate with those from a polynomial that fits the mode shape can yield four curvature damage indices to identify damage. Two non-model-based methods that use vibration shapes measured by a continuously scanning laser Doppler vibrometer system are introduced to identify damage in beams. Spatially detailed vibration shapes can be measured by the system in a rapid and accurate manner. Curvature damage indices can be obtained using curvature vibration shapes from the demodulation method to identify damage. All of the above non-model-based methods are robust against measurement noise and do not require any a priori information of undamaged structures that are usually not available in practice.

Prof. Zhimin Xi

University of Tennessee - Knoxville, USA

Data-Driven Based Battery Health Prognosis with Diagnosis Uncertainties and Insufficient Training Data Sets

  This paper investigates data-driven based battery prognosis with diagnosis uncertainties and insufficient training data sets. Four types of data-driven prognosis methods are investigated including the neural network, similarity-based approach, relevance vector machine, and a recently developed copula-based approach. The remaining useful life (RUL) predictions of lithium-ion battery capacity are compared with capacity estimation error due to the fact that onboard lithium-ion battery capacity estimation is difficult and almost always contains estimation errors. Thus, robustness of each prognosis methods can be studied for real time capacity RUL estimation. Furthermore, collection of sufficient run-to-failure training data sets for lithium-ion batteries is almost impossible even though it is desirable for all data-driven based methods. Therefore, robustness of these methods in terms of the insufficient training data sets is also studied. These insightful results will help designers choose appropriate prognosis algorithms in designing battery management systems (BMS) for lithium-ion batteries.

Prof. Qingkai Han

Dalian University of Technology, China

Health Monitoring and Vibratory Fault Prediction of Rotating Machinery

The major rotating machinery including large centrifugal or axial flow compressor, gas turbine and aero-engine are in the value chains of high-end and the core aspects of the industry factories, regarded as important embodiments of the national industrial core competence and high-technical level. These machines in industrial plant as key equipment are very expensive costing a few million dollars, and a single day's loss of shutdown may be very huge. Maintenance is of high importance but very difficult even many researchers and companies have made a lot of efforts and contributions.
However, there is no comprehensive and effective technology to completely solve the problem until now. Recently, maintenance programs for major rotating machinery are developing into preventive maintenance and predictive maintenance. In order to truly implement the preventive or predictive maintenance on them, several advanced and practical technologies, i.e. mostly associated with health monitoring and fault prediction, are prompted to be broken. On the other hand, the various indicators used to study the health of the machine, especially to deal with vibratory faults often occurred on the machine, are predominantly vibratory related; after all, any change in the condition of the machine affects its dynamic conditions and therefore the vibratory behaviors.
The health monitoring and fault prediction of rotating machinery include the following aspects: 1) health monitoring strategy and fault prediction principles; 2) fault mechanisms of rotor systems and structures; 3) advanced measurement technologies; 4) advanced signal data processing technologies; 5) vibration fault detection and diagnosis; 6) fault prediction and life estimation.
Taking the aero-engine rotor system and its blades as examples, several vibratory faults are presented to reveal new mechanism and give new vibration behaviors, such as blade rubbing-impact, rotor cavity oil induced instability, elastic-supporting misalignment and high order resonance induced fatigue of blade. Optical fiber-Bragg sensors and wireless strain transducers are introduced into the measurements of bearing deformations or blade cracks. The obtained vibration signals of machine and structure are processed to extract the feature parameters to indicate sensitively the healthy or fault conditions by using of time and/or frequency domain analyses. The fault diagnoses are classified as data driven and model based, either statistical or artificial intelligent ones. At last, the main difficulty of fault prediction lies in the evolution process of a fault and the happening time of failure, and the estimation of fault remaining life length. Both evolution speed based and state model based prediction technologies are investigated by using an example of bearing damage fault.
Some successful examples and cases are introduced. The most important contribution is to identify what is truly effective for practical plant maintenance among these proposed technologies.

Prof. Nam-Ho Kim

University of Florida, USA

Physics-based prognostics-promises and challenges

Traditionally, prognostics algorithms have been developed in physics-based and data-driven approaches. Since the two approaches have their own pros and cons, it may not be informative to compare these two approaches. Instead, this presentation will focus on the characteristics of physics-based prognostics algorithms so that researchers can understand and interpret the outcomes of the algorithms better. The presentation starts with the discussions on how to select physics-based methods depending on model complexity, the number of available data, and the level of noise. Then, the presentation shows how to rank different prognostics algorithms when the true degradation model is not available. It will be shown that the rank of algorithms depends on the particular realization of noise. Since most physics-based algorithms are based on statistical inference, the presentation lastly addresses statistical correlation, which includes the correlation between model parameters, between bias and initial damage, and between model parameters and loading conditions. It is important to consider all these effects in order to properly use physics-based prognostics algorithms and properly interpret the outcomes.

Prof. Christian Gogu

Universite Toulouse III, France

New prognostics-based structural maintenance strategies for civil aircraft

Currently, structural maintenance of civil aircraft is based on the scheduled maintenance strategy, where all the aircraft of a same model are inspected according to a predetermined schedule. Embedding sensors in aeronautical structures allowing to monitor their structural health can also open up new strategies for more efficient maintenance planning. A first possibility is to move from scheduled maintenance to condition based maintenance, where the current condition of the aircraft triggers specific maintenance policies. A further step is to not only use the current state of the damage but to also predict future damage growth and determine the maintenance policy accordingly. Both condition based as well as prognostics based strategies can be carried out independently or can be coordinated with other, scheduled, non-structural maintenances, which leads to a total of four new possible maintenance strategies. An application of these maintenance strategies to fuselage panels of a short range commercial aircraft is presented and their effectiveness and cost-efficiency is compared between them and with that of traditional scheduled maintenance. We show that large savings can be achieved, particularly when large variabilities are present.

Prof. Bruno Castanier

University of Angers, France

What are the effects of the reliability model uncertainties in the maintenance decisions?

While the risk due to the quality and quantity of the available data is one of the major concerns in the product design and qualification processes, this issue seems not to be tackled in the operational phases and especially for the optimization of maintenance policies. Indeed, the vast majority of the works proposed in the literature concentrates on the definition of a maintenance rule for the maximization of a long-term economic profitability by assuming well-defined and stationary reliability or degradation models. However, the convergence to these stationary states can be really slow and strongly related to the knowledge level of the failure modes and therefore to the data collected during its operation. In order to reduce this uncertainty, it is necessary to integrate some of knowledge acquired during the various product qualification and endurance tests. This remains, from our point of view, one of the major areas of improvement in industrial practices, especially for the development of the condition-based maintenance approaches.

The context of this work is the definition of optimization criteria which embrace the uncertainty on the reliability models and, in a longer term, to strengthen the relationships between the product design and qualification processes to the operating and maintenance phases. The focus of this paper is to highlight, based on numerical examples, the impact of the level of the available data on the efficiency of the maintenance decision, especially in condition-based maintenance, and, subsequently, to discuss on possible extensions in the maintenance decision criteria.

Prof. Jung Ryul Lee

KAIST, Korea

Recent Advances In In-process NDE and Smart Hangar for Aerosapce SHM

Recently, we are in transition from metallic to composite in operation and manufacturing of aerospace structures. NDE, SHM, and PHM are all over used, in design, manufacturing, operation and overhau stepsl in their lifes. In this invited talk, opto-electronic systems capable of in-process NDE, in-situ NDE and integrated SHM and thier real world applications are introduced. In addition, their field implementation scenario are proposed and advances in Smart Hangar are introduced.

Prof. Chao Hu

Iowa State University, USA

Ensemble Learning with Degradation-Dependent Weights for Remaining Useful Life Prediction

Remaining useful life (RUL) prediction is critical to implement predictive maintenance. While significant research has been conducted in model-based and data-driven prognostics, very limited research has been done to investigate the prediction of RUL using an ensemble learning method that combines prediction results from multiple learning algorithms. The objective of this research is to introduce a new ensemble prognostics method with degradation-dependent weights. Specially, this method assigns an optimized, degradation-dependent weight to each learner (i.e., learning algorithm) such that the weighted sum of the prediction results from all the learners predicts the RUL of mechanical components with better accuracy. The ensemble prognostic algorithm is demonstrated using an experimental data collected from an engine simulator. Experimental results have shown that the predictive model trained by the ensemble learning algorithm outperform the existing methods.

Prof. Myong-Kee Jeong

Rutgers University, USA

Deep Learning Based Virtual Metrology (VM) and Yield Prediction in Semiconductor Manufacturing Processes

In this talk, we present the new deep learning based supervised autoencoder to extract meaningful features from massive in-line sensor functional signals of semiconductor manufacturing processes. Based on those extract features, we build the virtual metrology (VM) model to predict important quality characteristics of the process and the yield prediction model.

Mr. Mark Henss

University of Stuttgart, Germany

Framework for a Uniform Description of Prognostics and Health Management

The successful application of Prognostics and Health Management (PHM) systems increase world-wide steadily due to an increase of smart products on the market. Today PHM systems are developed and applied in different engineering disciplines using different models and methods. The diversity regarding models, methods and types of products lead to highly complex systems. To handle the complexity in an efficient way there is a need for a framework of PHM systems.

This paper describes the mathematical framework of PHM based on a generic model and a clear formalization (notation and semantic). The model includes modeling, diagnostics, prognostics and optimization. By the kind of application (e.g. the used lifetime method) the characteristics of the model are defined. Lifetime methods to estimate the Remaining Useful Life (RUL) are separated from the models to make the framework manageable for all disciplines. Combined with probability mathematics the uncertainties in PHM systems (model, prognostics, RUL etc.) are shown.

On the one hand, a common notation, semantic and a clear framework is necessary to manage highly complex systems. On the other hand, the cost and benefit of a PHM system are connected to the uncertainties. Both points are presented in this paper as the base to choose the right PHM system design.

Prof. Young Joo Lee

Ulsan National Institute of Science and Technology (UNIST), Korea

Probabilistic Fatigue Life Prognosis for Steel Railway Bridges after Local Inspection and Repair

Steel railway bridges are exposed to repeated train loads which often cause fatigue failure. To guarantee the target fatigue life, bridge maintenance such as local inspection and repair should be properly provided based on accurate fatigue life prognosis, but it is a challenging task because there are various sources of uncertainty associated with bridges, train loads, environment, and maintenance work. For the optimal risk-based maintenance, it is thus essential to predict the probabilistic fatigue life of a steel railway bridge and update the life prognosis information based on the results of local inspection and repair. In this research, a probabilistic approach is proposed to estimate the fatigue failure risk of steel railway bridges and update the prior information of fatigue life prognosis after bridges are inspected and repaired. The proposed method is applied to a generic steel railway bridge, and the effects of local inspection and repair on the probabilistic fatigue life prognosis is discussed through parametric studies.

Dr. Miso Kim

Korea Research Institute of Standards and Science, Korea

Metamaterial-based Enhancement of Elastic Wave Energy Harvesting

Enhancement of metamaterial-based energy harvesting will be presented from design, analysis towards experimental demonstration. Metamaterials, artificially engineered materials, exhibit unique properties including bandgap and negative refractive index and thus enable us to manipulate mechanical wave propagations. In order to amplify input mechanical wave energy into energy harvesting systems, metamaterials can be utilized to guide and localize acoustic and elastic waves towards the desired position for harvesting. Recently, several research efforts on metamaterial-based enhancement of energy harvesting have been reported, but mostly based on intuitive design or with little experimental support. We propose several metamaterial-based energy harvesting systems including phononic crystals with defect and acoustic metamaterials with local resonances. Systematic design through geometric and bandgap optimization process is performed and followed by experimental demonstration. Drastic enhancement of energy harvesting performance via metamaterials is demonstrated and thoroughly investigated both analytically and experimentally.

Prof. Byeng D. Youn

Seoul National University, Korea

PHM Frontiers: Physics-based Approach to Deep Learning

Journal bearing rotor systems are widely used in various industrial applications. Due to uncertainties, the rotor may not behave as it is expected, and thus accidents may happen, which causes economic losses. To prevent such incidents, diagnosing the health state of rotor systems is of critical importance. Conventional health diagnosis approaches heavily rely on the physical features of the rotor systems, which require large amount of resources to develop. To minimize the amount of time and effort required for the feature engineering process of the rotor systems, the deep learning based feature engineering method is proposed in this paper. Among various deep learning techniques, convolutional neural networks (CNN), which is known to be powerful in recognizing images, is used. To use CNN, vibration signals are transformed into images using the omni-directional regeneration (ODR) method. The ODR method can generate virtual vibration signals in any direction from the actual vibration signals acquired from the two orthogonally placed proximity sensors. Then, the virtual ODR signals are transformed into images that sketch vibration signals acquired at all directions. Lastly, the image patterns are recognized by CNN, which replaces the expensive feature engineering step used in the conventional methods. The proposed deep learning based feature engineering is validated with the datasets from the journal bearing rotor testbed. It is demonstrated that the deep learning based anomaly detection of the journal bearings can substantially improve the efficiency and accuracy, compared to the conventional anomaly detection methods.

Prof. Diego Galar

Lulea University of Technology, Sweden

Maintenance Analytics, industrial data science and virtual commissioning

Industrial systems are complex with respect to technology and operations with involvement in a wide range of human actors, organizations and technical solutions. For the operations and control of such complex environments, a viable solution is to apply intelligent computerized systems, such as computerized control systems, or advanced monitoring and diagnostic systems. Moreover, assets cannot compromise the safety of the users by applying operation and maintenance activities. Industry 4.0 is a term that describes the fourth generation of industrial activity which is enabled by smart systems and Internet-based solutions. Two of the characteristic features of Industry 4.0 are computerization by utilizing cyber-physical systems and intelligent factories that are based on the concept of "internet of things". Maintenance is one of the application areas, referred to as maintenance 4.0, in form of self-learning and smart systems that predicts failure, makes diagnosis and triggers maintenance by making use of "internet of things".
Thus, for complex assets, much information needs to be captured and mined to assess the overall condition of the whole system including the one from design and manufacturing which obviously contains the physical knowledge. Therefore the integration of asset information during the entire lifecycle is required to get an accurate health assessment of the whole system, and determine the probability of a shutdown or slowdown avoiding black swans and other unexpected or unknown asset behaviors.
Moreover, the asset data are not only huge but often dispersed across independent systems that are difficult to access, fuse and mine due to disparate nature and granularity. If the data from these independent systems are combined into a common correlated data source, these new sets of information will add value to the individual data sources.
This talk will discuss the possibilities that lie within applying the maintenance analytics concept by the means of virtualization i.e virtual commissioning of the assets through data fusion and integration from a systems perspective.

Prof. Jacob Bortman

Ben-Gurion University of the Negev, Israel

Using physical based models for machinery RUL estimation

Machinery prognosis is the process of forecasting the remaining operational life (RUL), future condition, or probability of failure based on the acquired condition monitoring data. RUL forecast becomes crucial since it allows planning of efficient maintenance actions.

In order to estimate the remaining usefull life it is needed, first, to estimate the damage severity or size, and then, to forecast the damage growth model.

Damage size estimation is not yet developed and needs additional research. Classic theories assume that energy dissipation is proportional to fault size. For small faults, it is not correct. In general the energy-fault size relation depends on the transmission path from the defect to the snsor which differs for every case of defect location, machine and sensor. Therefore the estimation of the fault size is crucial. State of the art algorithm for bearings size estimation will be presented. The algorithm approach is based on insights from a general dynamic bearing model and an analytical model for RE-spall interaction. The outline of these models will be presented. Other machine components such as, gears and shafts will be discussed.

Prof. Bogdan Epureanu

University of Michigan, USA

An evolution of the dynamic responses and its effects on serviceability of offshore wind turbines with stochastic loads and soil degradation

Novel methods combined with an integrated simulation platform are proposed and used to ensure a twenty-year lifespan of an offshore wind turbine and substructure (OWT&S). These methods and the platform enable the estimation of the long-term evolution of dynamic responses of the OWT&S due to the degradation of soil modulus under stochastic loading conditions. Specifically, stress power spectrums and two probability distribution functions of soils (a Rayleigh distribution function and a Gaussian distribution function) are employed to calculate the probability distribution of degradation for all locations in soils. Moreover, a new method of using the derivative of degradation functions and inverse functions of the degradation functions is proposed to calculate the mean degradation index. These methods significantly decrease computational effort, which is a critical drawback of previous methods. Case studies demonstrate that the dimensions of the substructures significantly affect the evolution of the dynamic responses, suggesting that the evolution of dynamic responses should be considered in the design step to secure the serviceability of OWTs.

Dr. Przemyslaw Gromala

Robert Bosch GmbH, Automotive Electronics, Germany

Novel PHM concept for future use in safety relevant electronics for harsh environment

Electronic systems are and will be the major factor that will increase the safety on the road. According to the World Health Organization up to 2020 it is expected that number of accidents will be reduced by 50%. This will be realized through introduction of more advanced safety relevant systems that will slowly take control as well as responsibility for steering the car. As of today the safety relevant systems such as ABS/ESP saved only in Germany 8500 lives, and prevent 260000 accidents. Rear-end collisions that are among the worst, can be reduced by 72% due to the emergency braking system. In the urban traffic, up to 40 km/h the Bosch emergency braking system can completely prevent collisions with stationary vehicles. It is expected that autonomous driving will completely revolutionize the transportation system that will finally lead to 0 casualties in 2050.

In order to fulfill social expectancy, the safety relevant electronics systems that will be used in future automotive industry must be more complex. The future ECUs used in self-driving cars will be typically smart systems of the 3rd generation, which will perform human like operations. The 3rd generation smart systems will act independently in respect to control and decision making. In addition, these systems will be able to self-testing, self-calibration and self-healing.

Last but not least, the concept of Internet of Things will bring the components that are traditionally developed for consumer electronic market under the engine hood. All these aspects will required new approach regarding reliability and quality assurance. It is already observed that the lifetime requirements for embedded electronics used in automotive increases from 15 towards 25 years and for avionics systems towards 35 years. At the same time the time of qualification test is expected to be reduced by 30% with the cost of reliability tests to be reduced by 25%.

All these challenges and requirements can be realized by development of the new reliability concept that is strongly supported through numerical simulation and product optimization at the very early development stage. In the seminar I will present the novel approach for reliability assessment of the future electronic control units and smarty systems. I will present concept of simulation driven design that we are using during the development phase as well as the application of the hybrid prognostics and health management concept for the future safety relevant electronic control modules.

Prof. Bongtae Han

Mechanical Engineering Department University of Maryland, USA

Condition Monitoring of Automotive Smart Systems Utilizing Piezoresistive Stress Sensor

It is expected that advanced electronic components and smart systems would dictate the level of innovations in nearly all industrial sectors, including but not limited to automotive and other transport/logistics solutions, production equipment, energy and other infrastructure, etc. One such component is automotive electronic control unit (ECU), which controls the electrical system or subsystems in motor vehicles. In a conventional ECU, a protective metal case is used to ensure reliability under harsh environmental conditions. Recently, an epoxy molding compound (EMC) was adopted to replace the metal case. The EMC technology reduced the manufacturing cost significantly, yet the presence of a large amount of outer EMC increased the stresses of ECUs during transfer molding process and operations (Gromala, Fischer, Zoller, Andreescu, Duerr, Rapp & Wilde, 2013, Kim, Han, Yadur & Gromala, 2014). The long-term reliability assessment of this newly adopted manufacturing technology should be conducted to make the technology a more viable alternative to various ECUs.

More recently, a silicon-based IForce piezoresistive stress sensor (Gromala, Fischer, Zoller, Andreescu, Duerr, Rapp & Wilde, 2013) was employed to assess the reliability of EMC-based advanced ECUs. In spite of serval advantages, most notably in-situ stress measurements during operations, the stress sensor provides only a local stress state around the sensor. It is inevitable, thus, to establish the relationship of the stresses between the stress sensor and the critical parts of ECUs such as solder joints, wire bonds, chips, etc., to be able to use the stress sensor in prognostics and health monitoring (PHM) systems.

Finite element analysis (FEA) has been used widely to predict the stresses and strains field inside the electronic devices, and it can be employed to develop the relationship. However, an accurate relationship can be obtained only when the FEA model is verified and calibrated until numerical predictions match to experimental data (Han, Guo, Lim & Caletka, 1996).

In this paper, a model/sensor hybrid approach was implemented to conduct failure prognostics of an automotive electronic control unit (ECU). A 3-D finite element model simulating a complex ECU was built, and its predictability was calibrated and verified by an optical displacement measurement technique called moiré interferometry.

The stress state of the ECU during thermal cyclic loadings are documented by piezoresistive-based stress sensors embedded in test vehicles. The silicon chip consists of two stress sensors, and it is packaged in a standard land grid array (LGA) package. In each sensor, there is a whole matrix of 12 sensing cells, being placed in a 4 × 4 array. Four cells in the corners are inactive.
The in-situ loading history was obtained using the data obtained from a stress sensor in conjunction with a numerical metric that converted the stress signal into in-situ temperature excursion. The verified model was then utilized to develop a quantitative relationship between the stress senor data and the stress of the most critical locations in the ECU.

The results demonstrated that the proposed approach (predictive modeling with a stress sensor) will be an effective way to conduct failure prognostics of ECUs subjected to various operating conditions.