Shiyu Zhou, Ph.D. - US grants

PI: Shiyu Zhou
Proposal Number: 9985310
Institution: U of Pennsylvania

Abstract

Randomized computation has become one of the major research fields
in Computer Science since its emergence in the late 1970's.
In spite of its seemingly enhanced power, as far as we know,
randomness may not provide any computational advantage
(by more than a polynomial factor) over determinism.
The first objective of this project is to investigate the issue of
reducing randomness in computation via explicit combinatorial constructions.

Derandomization in space-bounded computation is the major subject to
study here, with the focus on pseudorandom generator constructions.
In particular, the constructions of pseudorandom generators for constant-width
read-once branching programs, and of discrepancy sets for combinatorial
rectangles are to be examined.

Derandomizing the well-known randomized log-space algorithm for
undirected graph connectivity problem will be investigated further.
In the meantime, a variant of the connectivity problem
in the model of directed graphs with tree structures will also be examined,
hoping this can provide a better understanding of the relationship between
non-deterministic and deterministic computations.

An advanced topic course (graduate or undergraduate) on explicit
combinatorial constructions and their applications will be designed.
The goal of the course is to enable the students to attain a solid
understanding of the mathematical foundation of explicit
constructions as well as their applications to algorithm design
and computational complexity.

Mobile computing over wireless channels has emerged as a rapidly growing
technology and gained a large amount of attention in recent years.
The highly asymmetric nature of the communication environment
in this context gives rise to many new challenges concerning the
issues of information dissemination and retrieval. Indexed data
scheduling is to design data broadcast schemes that
minimize both the average waiting time and energy consumption
of the clients in retrieving information on air.
Another objective of this project is to investigate
the possibility of applying the ideas from indexed data scheduling
to the design of communication protocols for multimedia applications
in wireless mobile communication networks.

This grant provides funding for the development of a methodology for modeling, analysis, and control of variation propagation in complicated manufacturing processes. Process variation and inconsistency are the major quality concerns in a manufacturing process. For a complicated manufacturing process involving multiple operation steps, the process variation at different steps will be accumulated on the product and propagates along the process. This project aims at developing a systematic methodology to describe and reduce the process variation and hence improve the process quality. A quantitative variation propagation model will be developed. Both analytical and empirical methods based on product/process design and engineering knowledge are used to link the key process variation sources and key product quality characteristics in this model. This quantitative model allows system theory and advanced statistical techniques (e.g., variance component analysis of linear mixed models) to be adopted in order to conduct forward and backward analysis of variation propagation. The forward analysis can identify important process stages and provide guidelines for design improvement, while the backward analysis can quickly identify the root causes of quality variations. The research results will be finally validated in industrial settings.

If successful, this research project will contribute to the science base of process control and quality improvement for manufacturing processes. Using the developed methodology, vast amounts of information from product design, process design, in-process sensing, and product quality inspection will be integrated under a quantitative model. This integrated model lays a foundation to develop effective techniques of variation propagation analysis and quick variation root cause identification. Effective implementation of the developed methodology in industry will provide a set of powerful tools for computer-aided product/process design and process monitoring and diagnosis for variation reduction, and thus provide a competitive boost to U.S. industry. The research accomplishments will be transferred into undergraduate and graduate curricula as well, which will generate long-term impact on the education of quality engineering and manufacturing.

The objective of this award is to develop a sensing methodology that enables highly reliable and accurate monitoring and diagnosis for manufacturing processes. This proposed research is to use a system approach to study the design, fabrication, optimization, assessment, and applications of distributed micro sensors embedded in metal tooling that is fabricated by Rapid Tooling manufacturing processes. A multidisciplinary research team seeks to advance fundamental knowledge in sensor technologies, including sensor design and fabrication for high-temperature strain and temperature measurements, the embedding of sensors into metal manufacturing tooling, and the interpretation and use of sensor data in decision-making for manufacturing process monitoring and control. Three interrelated research tasks are planned. Task 1 will focus on the design, fabrication, embedding, and optimization of distributed micro thermo-mechanical sensors in metal tooling. Task 2 focuses on the research on a new process monitoring and diagnostics methodology that can quickly identify both sensor and process faults. Task 3 will address issues related to the sensing system implementation and testing. The validation and testing of the system will be conducted on an industrial testbed. Wireless system implementation will also be explored.

New course development, existing course improvement, mentoring and outreach activities will attract and engage students as well as industry to smart tooling technologies. Embedded sensing systems will provide measurement with high spatial and temporal resolution at critical process locations, thus enabling a much more reliable and accurate monitoring and diagnosis system. The proposed embedded sensing system would also be extremely appealing in practice, striking to enhance the competitiveness of US Tooling industry as well as numerous manufacturing processes.

This Faculty Early Career Development (CAREER) research proposes to provide funding to develop, implement, and teach a multilevel, self-improving variation modeling and diagnosis methodology for complex manufacturing processes. The growing demand for products with improved functionality and time to market puts an enormous strain on production systems, resulting in ever-growing multilevel (i.e., both process level and station level) complexity in manufacturing processes. Targeting on these complexities, the research consists of several key steps. First, an efficient iterative model-building technique will be developed to identify the complex process-level variation flow. With the process-level model, the propagated variation and station-level local variation can be separated. Then, the spatial and temporal patterns of the quality data due to local variation sources will be gradually learned from the data and accumulated to form a self-improving signature library. Finally, the variation source diagnosis and process design evaluation are achieved based on this model. In addition to research, this project includes a substantial education component that includes curriculum and lab development, student advising, involvement of students from underrepresented groups, and various outreach activities including industry participation, high school involvement, and international collaboration.

If successful, the results of this research will fill the research gap in the control of complex processes by providing holistic modeling of process- and station- level complexities, effective diagnostic capability, and generic applicability to various processes, and thus provide a substantial boost to the overall competitiveness of US industries. The integrated education activities will contribute to manufacturing workforce training. Beyond manufacturing, the success of the project will also provide generic modeling and analysis tools for systems with complex flows of information. Broad dissemination of the developed methodologies could lead to diffusion to other fields vital to the nation's economic growth and security.

This Sensors and Sensor Networks (NSF 05-526), Sensors Small Team (SST)/Grant Opportunity for Academic Research (GOALI)/Collaborative Research grant provides funding to establish, validate, and implement a comprehensive multi-sensor planning, integration, distribution, and decision-making methodology for effective dimensional quality control of complex manufacturing processes. The research will establish and integrate: (i) A new dimensional sensing system to provide spatially- and temporally-dense dimensional measurements of intermediate and final products. The basic approach is to integrate different coordinate measurement sensors with such characteristics as touch-probe point sensor with high accuracy but low speed; and area optical sensor with low accuracy but high speed; (ii) A math-based decision making methodology for effective root cause identification of process variation in complex manufacturing processes by integrating sensing data and a vast array of product and process design information; and (iii) A system-level optimal sensor distribution strategy for sensor distribution to achieve optimal diagnosability and inspectability for quality and productivity improvement. This project will be carried out in close collaboration with the University of Wisconsin - Madison, Illinois Institute of Technology and Dimensional Control System, Inc. The methodology development will be based on, and the resulting technology will be tested and implemented in the DCS process simulation software.

The multi-sensor planning, integration, and analysis techniques will link such varied areas as system theory, computer aided design, optimization, and advanced statistics to solve problems on manufacturing process control. As a result, a new sensor and multi-sensor network system will be developed to help manufacturers considerably reduce process variation while at the same time significantly improve productivity and quality. This technique, if successfully developed, will provide a substantial boost to the overall competitiveness of US industries. The project will also significantly contribute towards the development of new curriculum and educational efforts as it will provide multidisciplinary training for students in the areas of mechanical and industrial engineering, system science, and statistics. Research accomplishments will be transferred into undergraduate and graduate curricula and also result in laboratory development.

GOALI/Collaborative Research: Event-Log-Based Failure Prediction and Maintenance Service for After-Sales Engineering Systems
Principle Investigators: Shiyu Zhou and Yong Chen
ABSTRACT
The objective of this collaborative research project is to test if the system event logs contain enough information to enable statistically sound and accurate prediction of the occurrence of failure events, and if yes, establish a generic event log analysis methodology for failure prediction and condition-based optimal maintenance of after-sales engineering systems. With the rapid development of information technology, an abundance of data that record the events occurred in a system (e.g., machine activities, critical system failures, operator/user actions, task status) are now collected automatically when the system is in use. Targeting the profusion of event logs, this research consists of four components: (1) fitting a system survival model using event logs to quantify the associations between various system events and the key failure event; (2) monitoring discrete events sequence to statistically test if the survival model fitted from historical data can fully represent the present system characteristics; (3) developing robust condition-based service policy based on the survival model and Semi-Markov decision processes; and (4) implementing and validating the established methodology for maintenance service of medical imaging diagnostic systems at the healthcare unit of General Electric company.
If successful, this research will advance fundamental knowledge in the planning and control of maintenance service operations for after-sales equipment by fully exploiting the current data-rich environment. The results of this research will help after-sales service industry to evolve from ad hoc experience-based operations into efficient optimized operations. In addition, the interdisciplinary nature of this collaborative research project can provide students a unique opportunity to obtain training in reliability, operations research, data mining, and statistics. Given the ubiquitous existence of system event logs, the established methodologies are potentially applicable to a broader spectrum of after-sales service applications such as manufacturing, communication, and computer network systems.

The objective of this collaborative research is to understand and control the variation propagation between product geometry and its dynamic response in periodic structures. Manufacturing processes are inherently imprecise, producing products that vary in geometry. The geometry variation detrimentally impacts not only the fitness of the final assembly, but also in many cases the functionalities of the final product. In this research, we will model and analyze the variation propagation in periodic structure from the manufactured part geometry to its dynamic response and then to provide guidelines on mitigating the undesirable dynamic response via controlling the mean and variation of part geometry. The proposed methodology includes variation propagation analysis in two directions: (i) In the forward prediction, the geometry variation of the manufactured products is characterized based on dense measurements, and then the probabilistic distributions of the dynamic responses are computed. (ii) In the backward analysis, using the abnormal dynamic response as input, a group of local geometric features that, upon modification, can mitigate the undesirable dynamic response are identified.
If successful, this research will positively impact a variety of periodic structural products, including bladed-disks in aero-engines and power generation equipment, cutting tools in high speed machining, and space antenna, etc. Through the collaboration with General Electric, this research will directly benefit a host of industries such as aerospace, automotive and manufacturing industries where these periodic structures are used. Through its integrated research, education and outreach activities, this project will provide advanced knowledge in variation propagation and reduction for students from high schools to graduate schools and will enhance domestic students? interest in science and engineering and therefore strengthen our competitiveness in the global workforce.

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).

This project focuses on process control and variation reduction issues of the ultrasonic cavitation based nanocomposite fabrication process and targets bringing this process from lab environment to a scale-up industrial environment. The ultrasonic cavitation based dispersion of nanoparticles in aluminum and magnesium alloy melts has been shown to be a very promising process of producing metal matrix nano-composites. The scientific objective is to discover the fundamental processing/microstructure/property relationship in this fabrication process through the integration of statistical methods and physical analysis, and then utilize the relationship for process optimization and control. This research will focus on the specific tasks as follows: (i) Quantitative assessment of the nano-particle dispersion within the microstructure. (ii) In-situ process sensing signal processing and characterization. (iii) In-situ process monitoring and optimization. The knowledge generated in this project will reveal the influence of non-linear effects (cavitation and streaming) on the nanoparticle dispersion, micro/nano-structures, and mechanical properties of aluminum matrix nanocomposites and will enable the production of high performance bulk Al matrix nanocomposites at the scale-up industrial level.

The success of the project will catalyze a transition from traditional process control techniques to a generic model-based diagnostic paradigm and contribute to a new scientific base for scale-up nano-manufacturing. Successful implementation of this project will result in providing our nation's manufacturing base with new, more energy-efficient production methods while at the same time enabling new products (e.g., automotive engine block) that themselves are more energy efficient in comparison to products available today. This project can also provide students the unique opportunity to obtain interdisciplinary training in various fields including mechanical engineering, material science, system science, and statistics.

The research objective of this Grant Opportunity for Academic Liaison with Industry (GOALI) collaborative research award is to establish a series of quality control methodologies on modeling, monitoring, and diagnosis of spatial point patterns. A spatial point pattern is a set of locations randomly distributed within a designated region. It is a natural way to model many critical quality characteristics in various manufacturing processes, such as the surface defects on steel bars, slabs and semiconductor wafer, and the distribution of reinforce particles in composite materials. The research approach is in three focus areas: (i) Modeling and monitoring of replicated spatial point patterns by integrating the deterministic point pattern alignment methods and spatial statistics techniques; (ii) Identification of the impacts of covariates on replicated point patterns based on a nonparametric functional regression model; and (iii) Geometric and three-dimension point pattern detection by bringing the Hough Transform method, an interesting computer vision method, into the quality control area.

If successful, the results of this research will provide a novel set of quality control tools to various relevant industries such as steel rolling, semiconductor manufacturing, and composite fabrication. These tools will take advantage of the increasingly available spatial point data and help to significantly improve the process productivity and quality. The close collaboration with OG Technologies in this Grant Opportunities for Academic Liaison with Industry project will lead to a realistic testbed and quick dissemination of research results among practitioners, as well as initiation of technology transfer. The interdisciplinary nature of this project can provide students the unique opportunity to obtain training in spatial statistics, manufacturing quality control, and computer vision.

The research objective of this Grant Opportunity for Academic Liaison with Industry (GOALI) collaborative project is to establish a series of data-driven modeling, failure prognosis, and service decision making methodologies that are tailored for both the opportunities and the needs of teleservice systems. In a teleservice system, the historical off-line records of failure events and the condition monitoring signals collected from a large number of units are available. At the same time, the condition monitoring signals from the in-service units are collected in real time as well. This unprecedented data availability provides us significant opportunities to develop accurate and robust algorithms to predict the remaining useful life and make optimal service decisions. The research consists of the following components: (i) a new state space formulation and nonlinear filtering approach for multi-phase condition monitoring signal modeling and estimation; (ii) a unified framework to jointly model the condition monitoring signals and the time-to-failure data for failure prognosis; (iii) a condition-based predictive service policy based on the joint prognosis model; and (iv) implementation and validation through collaboration with the General Motors.

If successful, the results of this research will enhance the science base of teleservice systems and catalyze a transition from reactive/preventive service to an integrative model-based predictive paradigm. The research is particularly timely for the booming teleservice industry, helping them to evolve from experience-based operations into efficient optimized operations. Given the information explosion and the ubiquitous existences of data, the research results can be applied to the teleservice of a broad spectrum of products such as manufacturing systems and communication systems. The synergistic nature of this project can provide students the unique opportunity to obtain training in various fields related with teleservice systems, including reliability, signal processing, vehicle engineering, statistics, and operations research.

Asthma is a common lung disease with acute and chronic manifestations that impacts more than 22.2 million Americans or 7.9% of the population, including over 6.7 million children younger than 18 years of age. The cost of asthma is significant both for individuals and for the society as a whole. It is highly desirable to establish transformative technologies to improve the patient quality of life and reduce the cost of asthma management. The recent development in sensor and mobile computing technology provide great opportunities to establish Smart Asthma Management (SAM) systems and achieve a quantum leap in asthma management. Leveraging on the fast development of information infrastructure, patients can create a detailed temporal log recording their symptoms, medicine usage, and possibly vital physiological signals through an easy access to a website or their smart phones in SAM systems. This unprecedented continuous stream of patient-generated data in SAM systems provides us significant opportunities to better estimate patient condition and make clinical intervention decisions. However, since the information infrastructure of SAM has not become available until recently, very limited work is available for SAM systems. Against this background, this collaborative project aims to develop a suite of statistical modeling, monitoring, prognosis, and clinical intervention decision making methodologies based on a flexible yet rigorous multistate model to describe the evolving of patient conditions. The true underlying state of the patient is assumed unknown; however, there is reason to expect that it could be inferred from patient generated data such as the frequency of the rescue inhaler usage (the time and frequency of the rescue inhaler use is an important indicator of asthma control).

Some anticipated advances include: (i) Multistate model with event intensity function as observations. The proposed methodology brings the mixed effect model and the multistate model into a unified framework to integrate the population information embedded in the historical records of multiple patients and the individual information collected in real-time in a quantitative way. (ii) Stochastic filtering approach for individual patient condition modeling and updating. The novel state space formulation enables efficient stochastic filtering algorithms to estimate and update the states and parameters in the multistate model. (iii) Clinical intervention decision support for patients and clinicians. The salient features of the proposed policy are that it is based on a condition-based policy and incorporates uncertainties in the patient condition model through a Partially Observable Markov Decision Process (POMDP) framework which has been widely used and proven to be very effective in the management of industrial systems. Plans are in place to evaluate the effectiveness of the resulting technologies in collaboration with clinical experts.


The project is likely to contribute predictive technologies that could help reduce the cost and improve the quality of healthcare in the US, especially as it relates to effective management of chronic illnessess. Additional broader impacts of the project include enhanced research-based training opportunities for graduate and undergraduate students (including members of under-represented minorities) in healthcare engineering, statistics, and operation research; enrichment of the curricula in health systems in industrial engineering and operations research at the University of Wisconsin-Madison and the University of Iowa.

Cloud-based platforms for accessing, sharing, and visualizing manufacturing-enterprise-level data are becoming available. In a cloud-based quality-data-management system, the quality-characteristics of different devices, products, and facilities are accumulated in a centralized database. These data pertain to multiple machines and multiple facilities, offering opportunities to achieve more effective quality control and productivity improvements. However, most cloud-based platforms are as yet unable to exploit the information contained in such data to make better decisions for production-system control and quality improvement. The objective of this project is to advance a series of methodologies that enable modeling of a large number of quality characteristics, timely change detection, accurate root cause diagnosis, and optimal repair decision-making. The project will also contribute to workforce training by offering students opportunities to engage in interdisciplinary research dealing with manufacturing, computing, sensing, and machine learning.

The reason why cloud-based platforms may not as yet exploit manufacturing-enterprise-level data lies in the dearth of techniques to (1) describe the quality characteristics and their relationships, and (2) make decisions informed by such descriptive models. To enable cloud-based quality-data-management systems of the future, the investigators will first advance methodology needed for a flexible, yet rigorous, hierarchical graphical model, which will describe the inter-relationships among different quality characteristics. The hierarchical structure of the model will enable information sharing across different facilities within an enterprise. Based on this descriptive model, the investigators will next develop methodologies for process monitoring and diagnosis via likelihood based risk-adjustment and Bayesian-factor theory, and for optimal repair decisions via Partially Observable Markov Decision Processes (POMDP) framework. The developed methodologies will be tested on data obtained from an industrial collaborator.

Low-cost sensors for real-time monitoring of contaminants in water, such as toxic heavy metal ions, could provide early warning of contamination, thereby improving drinking water safety and protecting public health. A graphene-based water sensor platform is thus explored for rapid, sensitive, and selective detection of various water contaminants, overcoming limitations of current sensing technologies such as slow detection and inadequate sensitivity. However, the commercialization of such a sensor system is limited by its relatively high manufacturing cost due to the batch processing that involves traditional lithographic electrode fabrication and multiple manual post-electrode fabrication processes. This award explores a low-cost customized inkjet printing process for manufacturing of graphene-based water sensors. The research entails engineering various inks and modifying the standard inkjet printing process to produce the complete sensor system, continuously. High throughput manufacturing of the nano-enabled water sensing systems reduces their cost and enhances market acceptance. The research outcomes provide the rationale for substrate selection and treatment, scalable methods for producing various inks suitable for inkjet printing, and process models for customized inkjet printing. Project results could be used for many other applications such as solar cells, lithium-ion batteries, and supercapacitors, enabling low-cost manufacturing of a wide range of printable electronic devices. The project trains diverse student populations including women and minorities on scalable nanomanufacturing, nanodevice design and real-time water-sensing technologies through hands-on research experience, a course module, and enriching existing curricula.

The sensor platform is based on a field-effect transistor structure with reduced graphene oxide as the sensing channel and gold nanoparticles as anchoring sites of selective chemical probes. A major challenge for inkjet printing is the customization of the inkjet printing process for a specific device or system architecture. Customization involves engineering suitable inks, modifying the standard printing process parameters, and integrating components at different scales. The research team aims to close this knowledge gap by exploring inkjet printing of the entire graphene-based sensor system to enable the large-scale production via high throughput roll-to-roll nanomanufacturing of the sensor devices, which should result in low cost. The scalable nanomanufacturing of inks for all sensor components: electrode, sensing material, and probe, and their printing and integration into water sensor systems are investigated, together with methods for selecting and treating polymer substrates and customizing inkjet printing parameters. The sensor performance is validated in industrial testbeds through collaboration with A. O. Smith Corporation and NanoAffix Science, LLC. The project leads to a low-cost, high-yield scalable nanomanufacturing platform for graphene-based water sensor systems and other flexible electronic systems that can be readily commercialized by industrial partners.

The timely and accurate diagnosis and prognosis of fault conditions in mechanical structures and civil infrastructure using real-time measurements can play a critical role in ensuring the safe and sustainable operation of these structures. This, however, is inherently difficult because structural degradations and faults usually have very subtle characteristic signature with infinitely many possible patterns and profiles, which is further compounded by various uncertainties. The existing techniques fall short in addressing these challenges. The overarching goal of this research is to create a new framework of fault diagnosis and prognosis enabled by physics-guided data. This framework is built upon the integration of computational intelligence with high-fidelity modeling and analysis and the adaptation of a highly promising, non-contact sensor-structure interaction mechanism. The new modeling framework will lead to useful diagnostic and prognostic tools in many areas such as aerospace, marine, transportation, infrastructure, energy and power. This project will contribute significantly to the workforce training by promoting the interdisciplinary research of computing, sensing, and statistical analysis, and by promoting the concepts of resilient and sustainable systems.

The research encompasses a series of inter-related components. High-fidelity multi-scale physical models capable of characterizing high-frequency dynamic responses of complex structural systems with high efficiency will be created. Data-driven calibration of the physic-guided model to address the model inadequacy and bias issues will be formulated and established. Fault diagnosis algorithm through compressed sensing technique based on the calibrated physics-guided model will be developed. Fault prognosis through statistically rigorous mixed effects models and multivariate Gaussian process models will be synthesized. Combined with the adaptive sensor-structure integration mechanism, collectively these contributions form a new framework that can lead to orders-of-magnitude enhancement in sensitivity and robustness of structural fault diagnosis and prognosis.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.