What is the problem?

There exists a vast amount of approaches and techniques that could be applied for real-time failure prediction in industrial IoT data. The most advanced ones apply machine learning to get insights from the data and make high quality predictions. A variety of algorithms, data characteristics and technical possibilities should be considered. Practitioners and researchers alike, who are confronted with this situation spend tedious amount of time in order to identify all aspects that need to be considered for implementing the right solution for their specific case. This is labor intensive and costs resources and money.

What is the research question?

• What are the machine-learning techniques that can be applied to the problem of failure prediction?
• How real-time equipment failure prediction solutions can be implemented? What aspects should be considered, comparison of possible solutions?

What is the aim?
In order to address this problem a reference model of a solution that predicts equipment failures from real-time industrial IoT data should be created.

What is the method?
By applying a literature review, existing techniques used to failure prediction are first collected, evaluated and then assessed based on their applicability to the real-time data processing. A holistic framework is created that brings together data characteristics, failure prediction approaches, and possible data processing technologies. A case study underlines the application feasibility of the framework.

80% theoretical review
20% case study