Factory Sensor Event Forecasting with Multidimensional Pattern Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current event forecasting methods for time-series tensor data lack the ability to perform long-term and highly accurate forecasting without prior knowledge of time-series patterns, and are unable to effectively capture multidirectional dynamic patterns in sensor data from manufacturing environments.
Innovation Solution
An event forecasting system that continuously extracts model parameters from time-series sensor data, featurizes the data into summary information including modeling and error information, and uses this information to output the probability of future event occurrences, enabling long-term and accurate event forecasting without requiring prior knowledge of the patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional event forecasting methods are used, then the system can process time-series sensor data, but the forecasting accuracy and long-term prediction capability are insufficient
Solution Approach 1:
The patent segments time-series sensor data into multiple time windows, extracting features from each window to capture both short-term and long-term patterns. This segmentation allows the system to analyze different temporal scales independently and combine their predictions, thereby improving both forecasting accuracy and extending the effective prediction horizon.
Solution Approach 2:
The patent transforms time-series data into multi-dimensional feature space by extracting temporal, spectral, and statistical features across different time windows. This dimensional transformation enables the forecasting model to capture complex patterns that are not apparent in the original time domain, improving prediction accuracy while maintaining long-term forecasting capability.
2Measurement precision
If complex data processing is performed to capture multidirectional dynamic patterns, then forecasting accuracy improves, but computational complexity increases
Solution Approach 1:
The patent extracts key temporal and statistical features from time-series data, separating essential pattern information from raw data. By extracting features such as mean, variance, and temporal correlations from segmented windows, the system captures multidirectional dynamic patterns while reducing computational complexity compared to processing complete raw datasets.
Solution Approach 2:
The patent processes data in partial segments rather than analyzing the entire time series at once. By applying feature extraction and forecasting to multiple smaller time windows, the system achieves comprehensive pattern recognition through partial actions, improving accuracy while keeping each processing step computationally manageable.
3Adaptability or versatility
If multiple features are extracted from time-series data, then the ability to capture dynamic patterns improves, but data processing time increases
Solution Approach 1:
The patent applies feature extraction periodically across segmented time windows rather than continuously processing the entire dataset. This periodic approach extracts multiple features at key intervals, capturing dynamic patterns effectively while reducing overall processing time compared to continuous analysis of all data points.
Solution Approach 2:
The patent performs preliminary feature extraction on segmented data before final forecasting. By pre-processing and extracting features from time windows in advance, the system prepares pattern information that can be quickly utilized during forecasting, reducing real-time processing time while maintaining comprehensive pattern detection capability.
Data Source
AI summary
An event forecasting system includes a feature amount extracting unit and a forecasting unit. The feature amount extracting unit continuously extracts model parameters {m, r, S, ⊝, F} of dynamic patterns in a time direction and a facility direction from a multidimensional time-series tensor X of time-series sensor data collected for every period n from a plurality of types d of sensors respectively disposed at a plurality w of facilities of a factory, and further sequentially featurizes the multidimensional time-series tensor X into summary information {Z, ε} including modeling information Z and error information ε of the modeling information by use of the model parameter {m, r, S, ⊝, F}. The forecasting unit outputs a probability p of occurrence of an alert label y at a predetermined time Is ahead by use of the summary information {Z, ε} as an input.