Electromechanical Fault Data Labeling with Adaptive Time Windows

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing data labeling methods for electromechanical devices are inaccurate due to delayed fault reporting, leading to mislabeling and ineffective training of prediction models, which are time-consuming and costly.

Innovation Solution

A method and apparatus for labeling data of electromechanical devices that estimate a second timestamp prior to a fault reporting event based on domain knowledge, iteratively adjust a time window, and determine an optimal evaluation metric to accurately label data as unhealthy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If data is labeled based on fault reporting events, then the labeling process is automated and scalable, but the labeling accuracy deteriorates due to delayed fault reporting

Engineering Contradiction:
Improvelabeling scalabilityVSAvoidlabeling accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The system performs preliminary action by estimating the second timestamp (fault occurrence time) before the first timestamp (fault reporting time) using domain knowledge. This allows the system to proactively identify and label the actual fault period, rather than waiting for delayed fault reports. The estimation process includes iteratively adjusting the time window between first and second timestamps to optimize labeling accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system implements feedback by using evaluation metrics to assess the quality of labeled data and iteratively improving the timestamp estimation. The domain knowledge is refined based on feedback from actual fault data patterns, allowing continuous improvement of the fault occurrence time estimation accuracy.

Inventive Principle:
Principle #23Feedback

2Reliability

If simulations and tests are performed to generate faulty data, then prediction models can be trained, but the data does not accurately represent field conditions and requires huge manual effort

Engineering Contradiction:
Improveprediction model trainingVSAvoidmanual effort
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system enables self-service by automatically generating accurately labeled faulty data from actual field operations data. The system uses the fault reporting events and domain knowledge to self-label the data without requiring manual simulation or testing, thereby eliminating the need for time-consuming and expensive manual data generation while maintaining high reliability for model training.

Inventive Principle:
Principle #25Self-service

3Quantity of substance

If the time window size is increased to capture more fault data, then more faulty samples are obtained for training, but the precision of fault detection deteriorates due to inclusion of healthy data

Engineering Contradiction:
Improvefaulty data samplesVSAvoidfault detection precision
Core Design Contradiction:
Quantity of substanceVSMeasurement precision

Solution Approach 1:

The system applies parameter changes by dynamically adjusting the time window parameters (first timestamp and second timestamp) based on domain knowledge and evaluation metrics. Instead of using a fixed time window, the system optimizes the window size and position to precisely capture the fault period while excluding healthy data, thereby maintaining both sufficient sample quantity and high detection precision.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4682660A1A method for labeling data of electromechanical devices and an apparatus thereof
Publication Date: 2026.01.21 ABB (SCHWEIZ) AG
  • EP4682660A1 patent drawingFigure 1~2
  • EP4682660A1 patent drawingFigure 3
  • EP4682660A1 patent drawingFigure 4a~4b

AI summary

A method and an apparatus (104) for labeling data of electromechanical devices is disclosed. The method includes obtaining fault reporting event present in data associated with an electromechanical device. The fault reporting event comprises first timestamp associated with fault of electromechanical device. The method includes estimating second timestamp prior to first timestamp based on domain knowledge of electromechanical device. Data in the time window formed between first timestamp and second timestamp are labeled as unhealthy. The method includes iteratively increasing size of time window by updating second timestamp. The method includes determining an evaluation metric of classifier (212) in classifying data as unhealthy and healthy for each iteration. The method includes determining value of second timestamp associated with optimal evaluation metric. The method includes labeling data in time window formed between first timestamp and updated second timestamp as unhealthy data.