Sensor Waveform Segmentation for Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In facility maintenance, generating an appropriate normal waveform pattern is challenging due to the use of inexpensive sensors with low performance, which leads to false anomaly detection and is exacerbated by operations with low frequency and random timings, such as emergency stops, making it difficult to learn and detect anomalies effectively.
Innovation Solution
An information processing apparatus that acquires transition data, divides sensor waveforms into sections based on operation patterns, and generates an estimation model using partial waveforms to create a normal waveform pattern, excluding sections where anomalies are unlikely, thereby improving anomaly detection accuracy even with low-performance sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If inexpensive sensors are used to reduce cost, then device cost decreases, but measurement precision deteriorates leading to waveform fluctuations and false anomaly detection
Solution Approach 1:
The waveform is divided into multiple sections based on operation types (acceleration, constant speed, deceleration, retention). By segmenting the waveform according to operational phases, the system can analyze each section with appropriate criteria, reducing the impact of sensor fluctuations on overall detection accuracy.
Solution Approach 2:
Different evaluation criteria are applied to different waveform sections based on their operational characteristics. For example, acceleration phases use different anomaly thresholds than constant speed phases. This localized approach improves detection precision without requiring high-performance sensors across the entire waveform.
2Measurement precision
If a large number of training waveforms are required to compensate for low sensor performance, then measurement precision improves, but loss of time increases due to extended learning requirements
Solution Approach 1:
By dividing the waveform into operation-specific sections, the learning process focuses on characteristic patterns of each operation type rather than requiring extensive general training data. This segmentation enables effective anomaly detection with fewer training waveforms.
Solution Approach 2:
The system applies evaluation criteria selectively to specific waveform sections rather than uniformly across the entire waveform. This partial application of analysis reduces the amount of training data needed while maintaining detection accuracy in critical sections.
3Measurement precision
If waveform analysis includes all operation sections, then measurement precision improves, but device complexity increases due to handling low-frequency random operations
Solution Approach 1:
The waveform is segmented into distinct operation sections (acceleration, constant speed, deceleration, retention) with clear boundaries. This segmentation simplifies processing by allowing the system to focus analysis on specific sections rather than treating the entire waveform uniformly, reducing complexity while maintaining precision.
Solution Approach 2:
The system applies comprehensive evaluation only to critical waveform sections where anomalies are most likely to occur, such as acceleration and deceleration phases. Less critical sections receive simplified or no evaluation, reducing processing complexity while maintaining detection accuracy where needed.
Data Source
AI summary
According to one embodiment, an information processing apparatus includes processing circuitry configured to acquire transition data representing a transition of a plurality of operations of a monitoring object, divide, based the transition data, a first waveform of a sensor with respect to the monitoring object into a plurality of sections corresponding to the plurality of operations, and generate a first estimation model related to a state of the monitoring object based on partial waveforms of the plurality of sections in the first waveform.