Failure Prediction Algorithm with Descriptive Log File Analysis

• The key business value will be is to predict lock failures at the early stages and enable cost saving. If use case owner will know in advance which lock in which station will likely to fail in advance, they can arrange priorities accordingly. Additionally they can optimise maintenance calendar according to their resources.
• What makes our classification approach different is that, feature engineering is performed on the log data by using sliding window. The effects of window size and shift size between windows are also tested.

Sliding Window Model Creation

Input(s):

• Time series log data
• Previous failure data

Main feature(s):

• Our machine learning algorithm is designed as a binary classification model and the model tries to redict lock failures in advance by using a sliding windows feature extraction approach

Output(s):

• To predict whether a failure will occur within N periods (e.g. days, weeks, hours etc.) by analysis of logs of StockArt devices that established in stations provided by TriaTech

• Time series log data should be provided
• Since this is a supervised machine learning algorithm, previous failure data (medical cabinet lock failures in this example) should be provide

End users that are using any industrial product (medical cabinets in this example).

Commercial licence to be negotiated.