Industrial Anomaly Detection Using Semi-Supervised Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection models in industrial environments face challenges in effectively detecting abnormal datapoints, especially with large unlabeled datasets and limited labeled data, requiring significant training effort and expert validation.
Innovation Solution
The implementation of a predictive maintenance system that iteratively applies anomaly detection models to operation data, updates the training dataset with classified datapoints, and re-trains the models, enabling continuous learning and robust anomaly detection even with a small number of normal datapoints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing anomaly detection models are trained with a large amount of pre-labelled data, then detection accuracy is improved, but training effort and data labeling cost increase significantly
Solution Approach 1:
The system performs preliminary action by initially training the anomaly detection model with a small amount of labeled data before deployment. This preliminary training establishes a baseline model that can then operate in semi-supervised mode, eliminating the need for extensive pre-labeling of entire datasets while still achieving effective anomaly detection.
Solution Approach 2:
The system implements self-service through semi-supervised learning where the model automatically adapts to new data distributions and anomaly patterns without requiring manual re-labeling. The model serves itself by continuously learning from a small labeled seed set and unlabeled operational data, reducing dependency on expert annotators while maintaining detection accuracy.
2Productivity
If anomaly detection models are trained with only a small number of labeled datapoints, then training effort is reduced, but detection reliability deteriorates
Solution Approach 1:
The system applies dynamics by implementing a semi-supervised learning framework that dynamically adapts to varying data conditions. The model transitions from initial supervised training on small labeled sets to continuous adaptation using unlabeled operational data, maintaining reliability while improving training efficiency through dynamic learning rate adjustment and progressive unlabeled data incorporation.
Solution Approach 2:
The system utilizes parameter changes by modifying model training parameters such as learning rates, confidence thresholds, and data weighting factors during the semi-supervised learning process. These parameter adjustments enable the model to effectively learn from limited labeled data while incorporating unlabeled data, maintaining detection reliability without requiring extensive labeled datasets.
3Measurement precision
If extensive data labeling and expert validation are performed, then model accuracy is improved, but engineering effort and time consumption increase
Solution Approach 1:
The system applies partial action by performing data labeling and validation only on a small seed set of representative samples rather than the entire dataset. This partial labeling approach, combined with semi-supervised learning, achieves sufficient model accuracy while dramatically reducing the engineering effort and time required compared to comprehensive data annotation.
4Reliability
If anomaly detection models are retrained frequently to maintain accuracy, then detection performance is improved, but computational resources and training time are consumed
Solution Approach 1:
The system implements periodic action by retraining the anomaly detection model at scheduled intervals or triggered by specific conditions such as performance degradation thresholds or data distribution shifts. This periodic retraining strategy maintains detection performance while optimizing computational resource usage, avoiding continuous retraining that would consume excessive energy and processing power.
Data Source
AI summary
System, Device and Method of detecting at least one abnormal datapoint in operation data (U) associated with an industrial environment (610) is disclosed. The method comprising iteratively applying one or more anomaly detection models (fi) to at least one subset (S) of the operation data (U), wherein the anomaly detection models (fi) are trained based on a training dataset (L) consisting of datapoints labeled as normal; classifying subset-datapoints in the subset (S) as one of normal datapoints (N) and abnormal datapoints (A) using the anomaly detection models (fi); updating the training dataset at least with the normal datapoints; retraining the anomaly detection models (fi) with the updated training dataset after expiration of a threshold time, wherein the threshold time is based on the number of updates to the training dataset; and detecting the at least one abnormal datapoint in the operation data (U) using the anomaly detection models (f′i).