Time Series Analysis Device for Dynamic Behavior Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing time series data from apparatuses, such as [PTL 1] and [PTL 2], are inadequate for diagnosing abnormalities in specific behaviors that occur outside regular observation periods or cannot display waveforms effectively, particularly in scenarios like a dump truck's unloading process or an elevator's movement between floors.
Innovation Solution
An analysis device and method that accumulate sensor, operation, and control data with time information, recognize behaviors using specific algorithms, and display or output data associated with these behaviors, allowing for diagnosis algorithm changes based on recognized behaviors to improve accuracy and reduce false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If waveform extraction is performed at regular time intervals (daily or weekly), then data processing is simplified and storage requirements are reduced, but the system cannot capture behavior changes occurring outside these regular intervals
Solution Approach 1:
The patent transitions from static, fixed-interval waveform extraction to dynamic, event-driven waveform extraction. The system automatically determines extraction intervals based on detected behavior changes, adjusting the sampling rate to match the actual operational characteristics of the apparatus. This resolves the contradiction by making the data collection frequency adaptive rather than fixed.
Solution Approach 2:
The patent implements preliminary behavior recognition algorithms that continuously monitor sensor data to detect behavior changes before they occur or as they begin. By identifying the start of a behavior change event in advance, the system can trigger waveform extraction at the optimal moment, ensuring capturing critical transitions without requiring continuous high-frequency sampling of all parameters.
2Speed
If threshold processing is used for real-time state recognition, then rapid response to state changes is achieved, but the system cannot diagnose abnormalities that occur outside observed time periods or were not present at observation moments
Solution Approach 1:
The patent accumulates and stores sensor data, operation data, and control data with time information before behavior changes occur. This preliminary data accumulation ensures that when a behavior change is detected through threshold processing, the system already has the necessary data context to accurately diagnose abnormalities, even if they develop after the initial observation point.
Solution Approach 2:
The patent segments the time series data into distinct behavior sections based on recognized behavior changes. Each segment is then analyzed with appropriate diagnosis algorithms tailored to that specific behavior type. This segmentation allows the system to maintain rapid response through threshold processing while improving diagnosis reliability by applying context-appropriate analysis to each segmented period.
3Loss of information
If all sensor data is displayed continuously, then complete information is provided for analysis, but the display becomes overwhelming and difficult to interpret for long time series data
Solution Approach 1:
The patent segments long time series data into meaningful behavior sections based on recognized behavior changes. Each segment represents a distinct operational phase with its own characteristics. This segmentation preserves all necessary information while making the data interpretable by organizing it into logical, manageable units that reflect actual operational patterns.
Solution Approach 2:
The patent applies different display qualities and levels of detail to different segments based on their local characteristics. Critical behavior changes receive enhanced display treatment with detailed parameter information, while stable periods are displayed more concisely. This local quality approach ensures complete information is available where needed while maintaining ease of interpretation overall.
4Device complexity
If a single diagnosis algorithm is used for all operating conditions, then the system structure is simplified, but diagnosis accuracy decreases when applied to diverse behavior types
Solution Approach 1:
The patent implements a universal diagnosis framework that automatically selects and applies appropriate diagnosis algorithms based on the recognized behavior type. The system maintains a library of specialized algorithms but presents a unified interface and selection mechanism. This multi-functionality approach preserves system structure simplicity through automated algorithm selection while achieving high diagnosis accuracy by applying the most suitable algorithm for each specific behavior section.
Data Source
AI summary
An analysis device for time series data from an apparatus to be diagnosed is provided with an accumulation device which accumulates sensor data, operation data, or control data, obtained from the apparatus, while accumulating time information, an algorithm accumulation unit which accumulates algorithms for recognizing behavior of the apparatus, a behavior recognition unit which recognizes behavior of the apparatus by using a recognition algorithm, and a specification unit which specifies a behavioral item to be recognized. A behavior recognition algorithm corresponding to the specified behavioral item is selected from the algorithm accumulation unit; sensor data, operation data, or control data is selected from the accumulation device; start and end times of a selected behavior are recognized by the behavior recognition unit; and the recognized start and end times are associated with time information about data accumulated in the accumulation device.