Manufacturing Facility Abnormality Detection Using LSTM Autoencoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting abnormalities in manufacturing facilities are labor-intensive, costly, and lack objectivity, making it difficult to determine whether a manufacturing facility is functioning properly, especially as facilities become more diversified and produce smaller quantities of varied products.
Innovation Solution
A method and apparatus using a long short-term memory (LSTM) autoencoder learning model to detect abnormalities by receiving data from multi-sensors, generating a learning model, determining thresholds, and switching control modes based on abnormal state detection, allowing for real-time monitoring and reduced costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual monitoring by factory manager is used to detect abnormalities, then detection capability is provided, but manpower cost increases and objectivity/accuracy deteriorates
Solution Approach 1:
The manufacturing facility performs self-diagnosis by comparing its own operational data against learned normal patterns. The system enables the facility to automatically detect its own abnormalities without external human intervention, thereby improving reliability while eliminating manpower requirements.
Solution Approach 2:
The patent replaces the mechanical human monitoring system with an automated learning model based on neural networks. This substitution transforms manual detection into an automated computational process, maintaining detection capability while removing dependency on human resources.
2Adaptability or versatility
If diverse manufacturing facilities are deployed to produce varied products, then production flexibility improves, but abnormality detection difficulty increases
Solution Approach 1:
The learning model adapts to different manufacturing facilities by learning from their specific operational data patterns. Each facility's unique parameters and operational characteristics are captured during the learning phase, enabling the system to handle diversity while maintaining consistent detection accuracy across varied production environments.
3Speed
If real-time monitoring is implemented to enable immediate response, then response speed improves, but system complexity increases
Solution Approach 1:
The system performs preliminary learning during normal operation to establish baseline patterns of normal behavior. This pre-learning phase enables the system to immediately detect deviations from normal operation without requiring complex real-time analysis algorithms, thus achieving fast response with manageable complexity.
Data Source
AI summary
A method and apparatus for detecting an abnormality of a manufacturing facility is disclosed. According to an example embodiment of the present disclosure, a learning model generating method for manufacturing facility abnormality detection may include receiving a measured value for a normal state of a manufacturing facility collected through a multi-sensor on a time-by-time basis, generating a learning model including a predetermined weight set and training the learning model using the measured value, and determining, using the learning model, a threshold corresponding to a boundary between the normal state and an abnormal state of the manufacturing facility and a criterion for determining the abnormal state in a local window representing a predetermined time interval.