Failure Pattern Ranking for Multisensor Time-Series Diagnosis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for monitoring technical systems, such as oil and gas production equipment, struggle to reliably and efficiently detect subtle and rapid changes in sensor data, often requiring extensive historical data and manual analysis, which can lead to missed anomalies and inefficient maintenance.
Innovation Solution
An assistance apparatus that analyzes multivariate time-series sensor data by assigning symbolic representations to specific temporal courses, calculates similarity measures, and ranks failure patterns to automatically identify potential issues, allowing for early and accurate detection of failures without the need for machine learning training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual monitoring by operation staff is used, then expertise and experience can be applied, but subtle changes and rapid changes can be missed and analysis depth is limited
Solution Approach 1:
The patent introduces an intermediary system (the assistance apparatus with processor) that mediates between the raw sensor data and the domain expert. This intermediary automatically performs the tedious work of comparing sensor data temporal courses against stored failure patterns, allowing the domain expert to focus on high-level decision-making while ensuring no subtle or rapid changes are missed.
Solution Approach 2:
The system enables self-service by allowing the monitoring system to automatically analyze its own data without requiring constant human intervention. The processor autonomously compares temporal courses, calculates similarity measures, and generates failure probability assessments, freeing operation staff from manual monitoring while maintaining high reliability.
2Extent of automation
If machine learning approaches are used, then automated analysis can be achieved, but extensive historical data containing different types of failures is needed for training
Solution Approach 1:
The patent applies preliminary action by pre-storing failure patterns and their associated temporal courses in a database before actual failure detection is needed. These patterns represent predetermined knowledge about how sensor data behaves during various failure types, allowing the system to immediately compare against new data without requiring extensive historical training data collection and processing.
Solution Approach 2:
The system creates simplified copies of failure scenarios in the form of stored temporal courses and failure patterns. Instead of needing vast amounts of actual historical failure data, the system uses representative copies (patterns) that capture the essential characteristics of different failure types, enabling automated analysis with minimal data requirements.
3Ease of operation
If fixed threshold rules are used, then general rules can be applied, but it is hard to judge if a rule can be applied in a certain situation and subtle changes are missed
Solution Approach 1:
The patent transitions from static fixed thresholds to dynamic temporal course analysis. Instead of comparing sensor values against fixed thresholds, the system analyzes the temporal evolution of sensor data, comparing how values change over time against stored failure patterns. This dynamic approach adapts to different situations automatically while maintaining ease of operation through automated pattern matching.
Solution Approach 2:
The system changes the parameter being analyzed from absolute sensor values to temporal courses (rates of change, trends, patterns over time). By transforming the data representation from static values to dynamic temporal patterns, the system can detect subtle changes and rapid transitions that fixed thresholds would miss, while maintaining simple operation through automated comparison algorithms.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Assistance apparatus for automatically identifying failure types of a technical system by analysing monitored time series of more than one different sensor data, each sensor data representing a different parameter of the technical system, comprising at least one processor configured to - determine for each sensor data a set of specific temporal courses of first time series of said sensor data of said sensor and assign a symbolic representation to each of the different specific temporal courses, - provide at least one failure pattern, each failure pattern representing one failure type out of several failure types of the technical system and each failure pattern consisting of a failure-type-specific combination of specific temporal courses of the first time series of at least a subset of sensor data in the same segment of time, wherein each specific temporal course is represented by the respective symbolic representation, - obtain more than one monitored time series of sensor data of the technical system, each of them divided into a sequence of time segments, and automatically assign to each time segment a symbolic representations (20, 27) according to the temporal course of the sensor data in the time segment, - calculate a similarity measure for the set of symbolic representations of a selected time interval of the obtained more than one monitored time series of sensor data and all failure patterns, - determine a ranking of the failure pattern depending on decreasing values of the calculated similarity measure, and - output the ranking via a user interface.