Predictive maintenance techniques are used to help determine the condition of in-service equipment in order to estimate when maintenance should be performed. This approach promises higher availability and costs savings over traditional maintenance management. To use predictive maintenance, the data scientists at ZEISS need to extract valuable information from sensor data applying state-of-the-art machine learning methods to ensure that the provided information add true user value. For reliable machine learning models sensor data must be carefully labeled. However, the process of data labeling is a highly time-consuming task requiring domain expertise. Data scientists need to consult domain experts to make sure data can be correctly labeled.
The aim of this project was to develop an application that provides the visualization of time-series data together with highlighting the possible anomalies, which come from unsupervised anomaly detection methods. The data scientist can accept or reject each of the propositions and in case of uncertainty, write a comment and ask the experts for their opinion. Customers
Dr. Lydia Nemec, Kay-Uwe Clemens Project Leader
Sajjad Taheri Developers
Palle Klewitz, Nils Faulhaber, Vsevolod Stepanov, Aleksandr Karavaev Video
Client Acceptance Test