Time Series Data Correlation for Predictive Maintenance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems for anomaly detection and predictive maintenance in industrial systems lack the capability to identify new patterns in time series data and associate them with other types of data, such as alarm and event data, to effectively predict future events.
Innovation Solution
A system and method that processes time series data to identify sequences of interest, extracts these sequences as events, and calculates a confidence level to quantify the relationship between the time series data and known events, enabling real-time prediction of future events by comparing data patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If time series data is analyzed using traditional anomaly detection methods, then equipment behavior changes can be detected, but the system cannot identify new patterns or associate them with other data types to predict future events
Solution Approach 1:
The patent combines time series data analysis with discrete event data (alarms, maintenance records, operational events) into a unified predictive framework. By merging these previously separate data types and analyzing them together, the system can identify patterns that span multiple data sources, enabling more accurate future event predictions while managing complexity through integrated processing.
Solution Approach 2:
The system performs preliminary pattern identification and association building during periods when events are not occurring, pre-processing and storing relationship patterns between time series data and discrete events. This preliminary action allows the system to quickly apply pre-learned patterns for real-time prediction without excessive computational complexity during critical monitoring phases.
2Loss of information
If discrete alarm data and time series data are analyzed separately, then each data type can be processed independently, but the system cannot effectively correlate them to predict future events
Solution Approach 1:
The patent introduces an intermediary correlation layer that bridges time series data and discrete event data. This intermediary component identifies and stores relationships between temporal patterns in sensor data and discrete events (alarms, maintenance actions, operational changes), allowing the system to correlate information across data types without requiring direct complex interaction between all data sources, thus reducing overall processing complexity.
3Loss of time
If the system monitors only current equipment state, then real-time detection is achieved, but future events cannot be predicted
Solution Approach 1:
The system performs preliminary analysis of historical time series data and discrete events to identify predictive patterns before failures or critical events occur. By pre-processing historical data to extract meaningful relationships and storing these as correlation rules, the system maintains real-time monitoring capability while simultaneously building predictive models that can forecast future events based on current patterns matching historical precedents.
Solution Approach 2:
The system uses feedback from discrete events (such as alarms and maintenance records) to continuously refine and update the correlation models. When events occur, the system analyzes the preceding time series data patterns, updates the correlation knowledge base, and uses this refined information to improve future predictions, creating a continuous improvement loop that enhances prediction capability over time.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
A system for predicting events by associating time series data with other types of non-time series data can include a processor configured to receive a data stream including time series data transmitted from a sensor configured to measure an operating parameter of a component being monitored. The processor identifies sequences of interest in the time series data having predictive value. The processor compares the real-time data stream to a set of known historical patterns that act as effective leading indicators of different alarms and events. The processor extracts any identified sequences of interest from the time series data as an extracted event. The processor quantifies the relationship between the data of the extracted event and the known historical pattern by calculating a confidence level to denote a probability of occurrence of the event by comparing how closely the new time series data matches the data patterns associated with known events.