Data Mining for Service and Maintenance
-a workshop in conjunction with KDD 2011 -
System (or equipment, machine, instrument) maintenance or servicing is a crucial part of business for many industrial processes, especially high-cost and safety-critical processes such as power plant, oil/gas turbine or aircraft engine operations. It is often an industry standard to provide system service as part of the contractual agreement, and the quality of service is increasingly becoming the key differentiator among similar systems. Appropriate and timely maintenance can increase the safety and reliability of the system, detect failing components to reduce collateral damage to other healthy subsystems, and ultimately increase revenue and improve customer satisfaction.
Instead of the common practice of fixing the maintenance schedule for the equipment (e.g., by usage time) or doing reactive diagnostics maintenance (i.e., fixing the problem after it happens), increasingly industries are moving toward a so-called proactive service mode (which is often called condition monitoring, preventive maintenance, among other names). The idea is to monitor the system closely via service data analytics, and schedule maintenance timely (e.g., when some components need to be serviced or replaced) and proactively (before the system is down). For this purpose, there are much service data being generated lately, including sensory data, system error messages, event log files, service histories, and other types of machine status information. The key question here is how to analyze these data effectively and efficiently in order to provide good service to the systems, while minimizing service cost. Most of these data are time-series based, and some data (e.g., event log files) might be unstructured or semi-structured, requiring natural language processing and text mining. Data analysis is challenging, as one often needs to take into account different data types to effectively predict component failure proactively. Another aspect of great interest to industries is how to design the service data in the first place, to facilitate data analysis.
Service-data analytics is a challenging task due to several intrinsic characteristics of modern industrial systems. First, systems like aircraft engines are very complex and the measurable variables are far less than enough to model system dynamics. Secondly, one often observes significant system-to-system variances, due to various initial conditions, operation modes and operation environments. Thirdly, for a system composed of tens of thousands of components, there exist multiple potential failure modes, which present different signatures in the data. Last but not the least, missing data is common in service data because of practical constraints.
For its unique characteristics mentioned above, service data analytics is a new domain in which many data mining and knowledge discovery techniques (such as time series analysis and prediction, relational learning, multi-view learning, transfer learning or incomplete data analysis) can have a profound effect. However, this new domain has not received sufficient attention in the KDD community. There are separate communities which primarily focus on prognostics problems (such as the Prognostics and Health Management society, http://www.phmsociety.org/), but they mainly tackle the problem from engineering perspectives and with physical-based approaches.
The purpose of this multi-disciplinary workshop is to bring together industry experts and researchers from data mining, machine learning, text analysis and signal processing who share an interest in problems and applications of system service and maintenance. We believe that this is an important application domain for KDD, and will continue to organize this workshop in future KDD conferences if there is enough interest. The goal will be to formulate the relevant research questions and bridge the gap between data mining research and industry needs in this important field.
We plan to provide a platform for exchange of ideas, identification of important and challenging applications, and discovery of possible synergies. The difference between service and maintenance will be highlighted in multiple industries. It is our hope that this will spur vigorous discussions and encourage collaboration between the various disciplines, potentially resulting in collaborative projects. We will particularly emphasize the mathematical and engineering aspects of service data analysis.
We will try to address many of these topics through both invited and contributed talks. The workshop program will consist of presentations by invited speakers of both industry and academia, and by authors of extended abstracts submitted to the workshop. In addition, there will be a slot for a panel discussion to identify important problems, applications, and synergies between the two disciplines.