AUTOMATICALLY DETERMINING A ROOT-CAUSE INDICATOR OF A TECHNICAL SYSTEM CONCERNING AN OUTPUT OF A MACHINE LEARNING MODEL

• Main Authors: LEBACHER, Michael, BRONNER, Johanna
• Format: Patent
• Language: English; French; German
• Published: 24.04.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

Computer-implemented method for automatically determining a root-cause indicator (L) of a technical system (10) concerning an output of a machine learning model (f) trained to analyse the technical system (f) based on sensor data of different sensors measuring parameters of the technical system (10), comprising- receiving (S1) measured sensor data (X) of the technical system (10) and an output (y) of the machine learning model (f) for the received measured sensor data (X) as input, wherein the machine learning model (f) is used to evaluate the technical system (10),- determining (S2) an explanation information (h) for the output (y) of the machine learning model (f), comprising a set of feature values (θ) indicating an impact of the feature on the output (y) of the machine learning model (f),- training (S3) a root-cause model (q) based on root-cause labels (Li) assigned to the output of the machine learning model (f) to map the set of feature values (θ) of the explanation information (h) onto a root-causes indicator (L),- determining (S4) one of the root-cause labels by inputting a new set of feature value (θ) determined for an output of the machine learning method (f) resulting from inputting a new measured sensor data into the root-cause model (q), and- outputting (S5) the determined root-cause label as root-cause indicator (L) to a user interface.