Anomaly Detection Control Using Compressed Sensor Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing anomaly predictive diagnostic devices face challenges in collecting input data in milliseconds or microseconds due to multidimensional sensor data transmission via communication networks, limiting diagnosis to slow degradation anomalies.
Innovation Solution
A control device and method that includes data compression, anomaly detection using machine learning, and a database for collecting and storing data, allowing for asynchronous processing and updating of anomaly detection parameters, enabling faster data collection and anomaly detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multidimensional sensor data is transmitted via communication network (LAN/WAN), then data can be collected from distributed sources, but data collection speed is limited and cannot achieve millisecond or microsecond order
Solution Approach 1:
The patent segments the data collection architecture into two parts: (1) a high-speed local collection unit that gathers data from multiple sensors at millisecond/microsecond speeds, and (2) a communication unit that transmits the collected data via network. This segmentation allows the system to maintain high-speed local data acquisition while still utilizing network communication for data transmission, thereby resolving the contradiction between network-based data collection capability and data collection speed.
2Reliability
If data is collected and processed synchronously (waiting for response), then processing completeness is ensured, but processing speed decreases due to waiting time
Solution Approach 1:
The patent implements preliminary action by pre-defining anomaly detection parameters and processing rules before data arrives. The system prepares detection algorithms and thresholds in advance, so when data is collected, the processing can proceed immediately without waiting for complex runtime decisions, thereby improving processing speed while maintaining completeness through pre-configured comprehensive detection rules.
Solution Approach 2:
The patent implements asynchronous processing with feedback mechanisms where the main processing flow continues without waiting for anomaly detection results, but the detection results are fed back for record-keeping and subsequent analysis. This feedback loop ensures processing completeness is maintained for audit and analysis purposes while the main system productivity is not blocked by detection waiting time.
3Measurement precision
If anomaly detection parameters are updated frequently, then detection accuracy improves, but system complexity and processing overhead increase
Solution Approach 1:
The patent implements periodic action by updating anomaly detection parameters at predetermined intervals rather than continuously or with every data cycle. This periodic update mechanism maintains detection accuracy by regularly refreshing parameters with new learning results, while reducing system complexity by limiting the frequency of parameter updates to manageable intervals, thereby resolving the contradiction between detection accuracy and system complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A control device includes feature amount generating means for generating a feature amount suitable for detecting an anomaly that occurs in a control target from data that relates to the control target, machine learning means for carrying out machine learning using the feature amount generated by the feature amount generating means, anomaly detecting means for detecting the anomaly, based on the feature amount generated by the feature amount generating means and an anomaly detection parameter determined based on a learning result of the machine learning and used in detection of the anomaly that occurs in the control target, instructing means for instructing the anomaly detecting means to perform detection of the anomaly, and data compressing means for data-compressing the feature amount generated by the feature amount generating means, and providing the data-compressed feature amount to the machine learning means and the anomaly detecting means. The instructing means transmits a request required in detection of the anomaly to the anomaly detecting means, and the anomaly detecting means carries out detection of the anomaly, without sending a response to the request to the instructing means.