Self-Monitoring for Innovation
|1 October 2016|
|30 September 2020|
|More info (PDF):|
|[[media: | pdf]]|
Involved internal personnel
Involved external personnel
We are entering the era of the “internet of things” (IoT), when machines, vehicles, goods, household equipment, clothes and all sorts of items will be equipped with embedded sensors, computers and communication devices. The importance of generating new knowledge and value from these technical advancements is evident from the fact that Vinnova, the Swedish Energy Authority and Formas, together with Swedish industry, have started a number of strategic innovation programs.
Those new developments require, and at the same time enable, monitoring the operation of complex systems in real-time. The ability to diagnose malfunctions quickly and predict faults, minimizing costs and productivity loss, is important for financial reasons but also to conserve environmental resources. Today, the majority of diagnostic functions are created by human experts in a time-consuming and expensive process. Ubiquitous monitoring of complex systems will only become possible with new, cost-effective, autonomous methods that do not require extensive human supervision. We refer to equipment that monitors its own operation, learns over time what the problems are, how they are characterized, and how to automatically detect them as self-monitoring.
A common feature shared by many modern industrial systems, enabled by the IoT, is access to large and ubiquitous streams of data describing their operation. One way to take advantage of this and automatically detect faults and deviations is to identify groups of peers, or similar systems, and evaluate how well each individual fits the rest of the pack. This approach is based on the “wisdom of the crowd”, that is, the assumption that by understanding the similarities and differences in the operation of groups of systems, it is possible to detect malfunctioning individuals.
It is, however, neither realistic nor desired that self-monitoring systems be fully autonomous. In fact, domain experts from companies are a crucial resource in developing, evaluating and specialising diagnostic and monitoring functions. Their expertise goes beyond the technical specification of the systems themselves and includes business and societal aspects. Thus it is necessary for any automatically derived solutions to be able to interact with domain experts, by taking advantage of available a priori knowledge, by explaining and justifying the solutions given, as well as by accepting feedback and incorporating it into further processing. We refer to this combination of semi-autonomous learning from the data and from expert knowledge as joint human-machine learning.
Against this background we define the overall core research question of this project as: How to construct self-monitoring systems that use joint-human machine learning to adapt to specific domains, by taking advantage of groups of peers, and ubiquitous streams of data?