Research

Click on the above image to view a short video relating to our research

Research activities within the center are focused on interpreting condition-based sensor data streams, and developing the means for utilizing this information to coordinate system performance degradation with logistical and operational decisions.

Focus Areas

Specific research projects include the below, click within a section to learn more about each focus area.

Photograph of gears and pulleys

Adaptive Diagnostics and Prognostics

Research in Adaptive Diagnostics and Prognostics is designed to provide practitioners with advanced fault detection capabilities as well as technologies for accurate prediction of remaining useful lifetimes. At the Center for Predictive Analytics and Logistics Management, we focus on solving some of the key practical challenges the may compromise the effective implementation of Diagnostic and Prognostic solutions, such as data sparsity, dynamic environments, interdependencies among system components.

  • Prognostics for Systems with Sparse and Fragmented Data: Although sensor monitoring technologies are becoming increasingly inexpensive, many challenges still remain unaddressed when it comes to the consistency and continuity of data acquisition and communication. Such challenges have created a mixture of dense yet discontinuous signal observations that require specialized techniques capable of utilizing such data for fault detection and life prediction.
  • Diagnostics and Prognostics for Systems under Dynamic Operational Profiles: Increased flexibility has resulted in an array of operational regimes for the same machine. Prognostics solutions therefore need to capture the effects of historical as well as future operational conditions on the degradation processes that take place prior to failure.
  • Prognostics for Systems with Component Interactions: Systems are becoming increasingly complex, with increased levels of machine automation and flexibility. Diagnostic and prognostic systems now need to account for component interdependencies that exist among critical system components. These interdependencies require prognostic solutions that capture C2C (component-to-component) and C2S (component-to-system) interactions that take place at the degradation process level.
  • Predictions of Multiple Failure Modes: The traditional paradigm of a single dominant failure mode no longer represents a practical discourse for modern engineering systems. Due to their increased complexity, predictive analytic solutions need to identify, track and predict multiple highly probable failure modes in various engineering systems. Novel statistical and data mining methodologies are currently being developed for real-time analysis of in-situ data to enable the on-line identification of the most likely failure modes.
Photograph of data stream

Predictive Analytics and Big Data

Complex engineering systems are now being embedded with hundreds and perhaps thousands of sensors capable of monitoring different facets of physical degradation. We are therefore faced with a pressing need for methods and solutions capable of handling Big Data challenges, especially for analyzing and modeling high-dimensional multi-stream degradation-based sensor signals. Predictive Analytics is one of the main research focus areas at PARO. Predictive Analytics is the process of extracting information from massive data streams with the goal of predicting the future performance/degradation of complex engineering systems.

  • Dimensionality Reduction and Fusion of Multi-Stream Data: An important challenge in Big Data applications is the size of the data. We develop novel statistical-based techniques for reducing the dimensionality of multiple sensor data streams and developing new data fusion algorithms that can be sued for predicting system performance.
  • Intelligent Sensor Selection: Condition or process monitoring applications often involves observing signals from thousands of sensors. Utilizing all the observed signal streams is not necessarily always an optimal strategy due various degrees of correlation with the underlying physical processes. This project focuses on the development of technologies capable of identifying a subset of most informative sensors that provide sufficient yet accurate knowledge that captures all the system dynamics.
Photograph of automotive manufacturing line

Manufacturing Process Monitoring and Control

To gain global economic competitiveness, the US manufacturing industries are increasingly embracing Lean, Six Sigma, and Just-In-Time (JIT) production policies in their shop floor. However, these policies create significant vulnerabilities that, to date, have not been formally addressed. For example, while Lean and JIT advocate low levels of buffer stock and work-in-process, the sustainability and reliability of the manufacturing system is often overlooked. Consequently, the impact of any unexpected interruptions caused by quality degradation or machine breakdowns can be detrimental resulting in lost production that cannot be compensated due to low inventory levels, thus immediately compromising subsequent logistics, delivery schedules, and ultimately loss of competitive edge.

  • Stream of Variation for Multistage Manufacturing Processes: Multistage systems refer to system consisting of multiple units, stations, or operations used to finish a final product or service. The quality characteristics of one stage are not only influenced by the local variations at that stage, but also by the propagated variations from upstream stages. Given these challenges, multistage systems present tremendous research opportunities related to quality engineering. For example, how to model product and process design information with the goal of reducing variability, and how to identify root causes of variability in manufacturing systems, etc.
  • Causation-based Quality Control: Thousands of sensors are now being embedded in manufacturing systems to automatically collect production information. Though this data-rich environment provides a great opportunity for more effective process control, the data analysis techniques currently used in the manufacturing area are limited. Most of the current quality control researches focus on correlation or association, which concerns how to reliably and accurately predict some features of a system from other features of that system. However, for effective process control, there is a need to identify the cause-effect relationships among variables which go beyond correlation or association. This idea leads to “causation-based quality control”, which is built upon observational data, causal modeling and discovery, and causal inference and decision making.
Photograph of wind turbines in field

Maintenance and Service Logistics

Smart sensors and wireless communication technologies have enabled us to streamline and coordinate service logistics and maintenance operations with the goal of maximizing equipment availability. The dynamics that govern spare parts inventory differ significantly from those of raw materials and final products. With the increasing use of sensing technologies, new tools are required for integrating this information into decision making processes related to maintenance operations and logistics. Some examples of the ongoing research in this area include;

  • Autonomous Spare Parts Logistics: In service logistics, the cost of stockouts translates immediately to lost production in the manufacturing sector and costly delays in service industry. To address these challenges, we have developed solutions that combine condition-based sensor signals with maintenance scheduling incorporated spare parts availability and logistics into the decision process. This is achieved by providing practitioners with state-of-the-art technology capable of accurate predictions of remaining life. The tight integration of these predictions with optimization models for inventory management and control allows us to provide optimal condition-based inventory policies with unprecedented service levels, balancing stockout cost with the cost of capital tied in spare parts.
  • Optimal Condition-Based Fleet Maintenance: In applications involving fleets of assets, such as arrays of wind turbines, fleets of trucks or aircrafts, etc., optimal maintenance of individual units may not necessarily be economically feasible. In fact, it is widely believed that economy of scale effects are essential to bring overall operational costs down. This research project focuses on exploiting similarities in operational and environmental conditions to develop optimal maintenance policies that combine unit-level condition-based maintenance with optimal fleet maintenance policies. This is achieved by identifying an optimal balance between the opportunity costs of early component replacement with the savings incurred by sharing fixed maintenance costs among multiple units.
  • Condition-Based Supply Chains: In large supply chains, which are made of various independent contractors, it is important that condition-based changes downstream in supply chains as propagated upstream to ensure a proper incorporation of these changes into the operating profiles. For example, imminent failure of production facilities might prompt for holding more stock preemptively to overcomes downtimes. We provide methodologies that allow for a dynamic incorporation of varying conditions and imminent changes to resource availability by devising schedules that allow for easy adaptation as well as quick integration of changes will being minimally disruptive.