Method for setting up a data processing chain
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing predictive maintenance systems suffer from false positives, leading to unnecessary and costly on-site interventions due to their inability to accurately predict anomalies in monitored environments, particularly in challenging locations, thereby compromising the profitability of IoT projects.
Innovation Solution
A method utilizing a reinforcement learning algorithm to optimize an artificial intelligence model for anomaly prediction by selecting and training on the most accurate datasets, replacing or updating the model based on experimental accuracy scores, and enriching or generating datasets for improved reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a predictive maintenance system uses sensor data to predict anomalies, then the ability to detect potential failures is improved, but false positives increase leading to unnecessary interventions
Solution Approach 1:
The system implements a feedback mechanism where operators validate predicted anomalies and provide corrections. This feedback loop allows the system to learn from false positives and improve its prediction accuracy over time, reducing unnecessary interventions while maintaining reliable anomaly detection
Solution Approach 2:
The system automatically adjusts its prediction algorithms and parameters based on accumulated feedback data without requiring manual reconfiguration. This self-service capability enables the system to continuously reduce false positives while maintaining high detection accuracy
2Measurement precision
If on-site interventions are performed to verify predicted anomalies, then the accuracy of anomaly detection can be improved, but operational costs and downtime increase
Solution Approach 1:
Instead of performing full on-site interventions for all predicted anomalies, the system performs only partial verification actions. Operators selectively visit sites only for predictions with high confidence scores or critical severity, reducing travel and intervention costs while maintaining sufficient verification accuracy
Solution Approach 2:
The system introduces an intermediary validation layer that uses additional remote sensing data and historical patterns to verify predictions before dispatching operators. This intermediary step filters out many false positives, reducing unnecessary on-site interventions while maintaining verification accuracy
3Reliability
If the artificial intelligence model is continuously updated with new datasets, then the predictive accuracy is improved, but the system complexity increases
Solution Approach 1:
The system uses a universal data processing framework that handles multiple dataset types and model versions through a single standardized interface. This multi-functional architecture simplifies model management by providing consistent methods for data ingestion, processing, and deployment across different AI models and update cycles
Solution Approach 2:
The system manages model complexity by dynamically adjusting parameters such as training frequency, dataset selection criteria, and model architecture configurations based on performance metrics and resource availability. This adaptive parameter tuning allows continuous improvement of predictive accuracy while preventing excessive complexity accumulation
Data Source
Figure 1~2
AI summary
The invention relates to a method for parameterizing a data processing chain (4) implementing a current artificial intelligence model (12) associated with a current accuracy score, the parameterization method comprising an implementation of a reinforcement algorithm including: - selection of a dataset (18) based on an associated reward; - training of an artificial intelligence model on the basis of the selected dataset (18) to obtain an experimental artificial intelligence model (20); - calculation of an experimental accuracy score of the experimental artificial intelligence model (20); and - based on a result of a comparison between the current accuracy score and the experimental accuracy score: • replacement or not of the current artificial intelligence model (12) by the experimental artificial intelligence model (20); and • updating of the reward associated with the selected dataset (18).