Industry-driven KPIs: The USU Perspective

USU AG is the European leader in IT-Service-Management software and one of only two suppliers worldwide, who are certified in all 16 ITIL®-processes [1]. IT service management (ITSM) refers to the entirety of activities that are performed by an organization to plan, design, deliver, operate and control information technology (IT) services [2]. For the later, control of IT services, key performance indicators (KPI) play a crucial role, as they can express the status of services and processes in measurable, objective and by graphical representation on dashboards intuitive way. Additionally, USU Software AG provides solutions for industrial big data analysis, applying machine learning and complex-event-processing technologies on sensor data (historical data and streams) from production and machine tools [3]. E.g. we realized Germany’s largest project for predictive maintenance at our customer Heidelberger Druckmaschinen AG, predicting the status of way over 10,000 printing machines, analyzing more than 1,500 sensors per each machine [4].

In both domains, production industry and IT, deriving service actions from data analysis is producing considerable costs for service provisioning, notably inferred labor costs for internal and external service employees, travel costs for onsite services and in particular in the production industry opportunity cost for lost time of production. That means false alarms are extremely expensive. Therefore several HOBBIT benchmarks are of prominent importance for us.

First, the Sensor Streams Benchmark and the Unstructured Streams Benchmark both aim to extract structured data from semi- or unstructured streams. While one could believe, that machine generated log data might be structured per se, that is not true, as these log data streams contain many syntactical and textual information. To extract and recognize them with high accuracy is of major importance in both domains.

Second, the aforementioned algorithms to extract and recognize textual information from streams have to be able to process a huge amount of data in short time, if online monitoring should become an option. Consequently data throughput of the two benchmarks is of interest as well.

Third, once you have derived structured data in good quality from the methods above, machine learning algorithms will have to a) learn and b) recognize abnormal events and problems within the (now) structured data streams. In HOBBIT the Structured Machine Learning Benchmark is addressing this need. So, KPIs measuring the performance of supervised and unsupervised methods to finally recognize abnormal behavior and problems is an important indicator how good certain algorithms are to e.g. avoid false alarms.




Spread the word. Share this post!

Leave A Reply

Your email address will not be published. Required fields are marked *