Autoencoder Similarity Search for Minute Process Data Fluctuations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large-scale plants, monitoring and analyzing process data for anomalies is challenging due to the complexity of data fluctuations, leading to difficulties in identifying similar past data for current anomalies, especially when minute fluctuations are involved.
Innovation Solution
A similar data search system that utilizes an autoencoder to process differences between query and registration data sets, calculating similarity based on reconstruction errors to identify accurately similar data, including a training mechanism for machine learning models to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional similarity search methods are used to find past process data similar to current anomaly data, then large fluctuations can be detected, but minute fluctuations cannot be accurately identified due to complex data variations
Solution Approach 1:
The patent extracts the fluctuation pattern from the process data by calculating the difference between current and past data, then applies autoencoder to isolate and represent only the essential fluctuation characteristics. This extraction removes irrelevant information and focuses on the core similarity features, enabling accurate detection of minute fluctuations while handling data complexity.
Solution Approach 2:
The patent introduces an autoencoder as an intermediary component that transforms the complex process data into a compressed latent space representation. This intermediary model learns to capture the essential fluctuation patterns and serves as a bridge between the raw complex data and the similarity comparison process, enabling accurate minute fluctuation detection.
2Reliability
If all process data is monitored constantly to detect anomalies, then comprehensive monitoring is achieved, but operator workload becomes unmanageable
Solution Approach 1:
The patent replaces the manual mechanical monitoring process with an automated information processing system. The system automatically acquires process data, calculates fluctuations, applies autoencoder for pattern recognition, and identifies anomalies without operator intervention. This substitution maintains comprehensive monitoring coverage while eliminating the burden of constant manual observation.
Solution Approach 2:
The monitoring system performs self-service by automatically detecting anomalies and identifying similar past cases without requiring operator analysis. The autoencoder-based system autonomously processes the data, generates similarity scores, and presents relevant information, allowing operators to focus on decision-making rather than data analysis.
3Measurement precision
If detailed analysis of all process data fluctuations is performed to identify cause, then accurate cause analysis is possible, but time required for analysis increases significantly
Solution Approach 1:
The patent performs preliminary action by pre-training the autoencoder model on historical process data to learn normal fluctuation patterns. When an anomaly occurs, the pre-trained model can immediately process the new data and quickly identify similar past cases, avoiding the need for time-consuming analysis from scratch. This preliminary preparation enables rapid accurate cause analysis.
Solution Approach 2:
The patent segments the complex analysis task into distinct components: data acquisition, fluctuation calculation, autoencoder transformation, similarity scoring, and case identification. This segmentation allows each component to be optimized independently and processed efficiently, reducing overall analysis time while maintaining accuracy through systematic processing of each segment.
Data Source
AI summary
According to one embodiment, similar data search system includes a processor. The processor acquires a query data set including measurement values. The processor generates, based on the query data set and a registration data set, an input data set representing a difference between the query data set and the registration data set. The processor inputs the input data set to a trained model. The processor acquires an output data set output by the trained model or an intermediate output data set that is an intermediate output of the trained model. The processor calculates similarity between the query data set and the registration data set based on the output data set or the intermediate output data set. The processor searches a database based on the similarity.