Machine Fault Diagnosis Using Pattern Clustering and Confidence Vectors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine condition monitoring methods, particularly rule-based systems, face challenges in accurately detecting failures in complex systems with limited labeled data, requiring extensive time and effort to develop effective condition monitoring rule bases, and struggle with classifying new failure patterns due to rare fault occurrences and variations in signal patterns.

Innovation Solution

The approach employs supervised machine learning techniques with pattern-recognition-based methods, utilizing both labeled and unlabeled data for training, including signal and pattern clustering to generalize from limited training samples, extracting features, and creating confidence vectors to classify failures effectively, even with scarce annotated data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If rule-based systems are used for machine condition monitoring, then the approach is widely used and simple to implement, but it requires extensive time and effort (80 man years) to develop accurate rule bases for complex systems

Engineering Contradiction:
Improveease of implementationVSAvoidtime to develop rule base
Core Design Contradiction:
Ease of manufactureVSLoss of time

Solution Approach 1:

The patent replaces manual rule-based systems with automated machine learning models that learn fault patterns from data. Instead of experts manually creating rules through deliberation, the system automatically learns diagnostic rules from labeled and unlabeled sensor data, reducing development time from 80 man years to a fraction of that time while maintaining or improving accuracy.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent performs preliminary clustering of unlabeled data to identify potential fault patterns before final classification. By pre-processing and organizing the data structure in advance through clustering, the system prepares the groundwork for faster and more accurate rule generation, reducing the overall time required for rule base development.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If traditional classification approaches are used, then the system can classify known failure patterns, but it struggles to classify new failure patterns that differ from training signals

Engineering Contradiction:
Improveclassification accuracy for known patternsVSAvoidability to classify new failure patterns
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal classification system that handles both known and unknown failure patterns through a two-stage approach. The clustering stage identifies structural patterns in the data that are independent of specific fault labels, while the classification stage applies these universal patterns to both labeled training data and unlabeled test data, enabling the system to generalize to new failure patterns it has never seen before.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent uses unlabeled data to create copies or representations of fault patterns that complement the limited labeled training data. By clustering unlabeled data and using these clustered representations alongside labeled samples, the system effectively increases the diversity and coverage of training patterns, improving its ability to recognize both known and novel failure modes.

Inventive Principle:
Principle #26Copying

3Measurement precision

If only limited labeled data is available, then the classifier performance becomes highly variable and dependent on the reliability of labeled samples, but obtaining extensive labeled data is time-consuming and difficult

Engineering Contradiction:
Improveclassifier performanceVSAvoidtime for data annotation
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent merges labeled and unlabeled data into a unified classification framework. By combining the reliable but limited labeled data with the abundant unlabeled data through a two-stage process (clustering followed by classification), the system leverages the strengths of both data types: the accuracy of labeled samples and the volume/coverage of unlabeled samples, reducing variability and improving overall performance without requiring extensive annotation time.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces clustering as an intermediary step between raw data and final classification. This intermediary process organizes unlabeled data into meaningful groups that can then be integrated with labeled data for training. The clustering intermediary transforms the raw unlabeled data into a form that is more useful for classification, reducing the direct dependency on large amounts of labeled data while improving classifier stability.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP2859418B1Generalized pattern recognition for fault diagnosis in machine condition monitoring
Publication Date: 2020.07.29 SIEMENS AG
  • EP2859418B1 patent drawingFigure 1
  • EP2859418B1 patent drawingFigure 2
  • EP2859418B1 patent drawingFigure 3

AI summary

A generalized pattern recognition is used to identify faults in machine condition monitoring. Pattern clusters are identified in operating data. A classifier is trained using the pattern clusters in addition to annotated training data. The operating data is also used to cluster the signals in the operating data into signal clusters. Monitored data samples are then classified by evaluating confidence vectors that include substitutions of signals contained in the training data by signals in the same signal clusters as the signals contained in the training data.