Electric Power Converter Anomaly Detection With Clustered ML Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current electric power converter monitoring systems face challenges in detecting gradual changes and anomalies in real-time due to the large amount of data generated, which exceeds the processing capabilities of existing converters and requires significant computing power and storage, often leading to delayed detection of issues and increased maintenance costs.
Innovation Solution
A computer-implemented method using a machine learning model for condition monitoring of electric power converters, which involves clustering input data into sets to reduce the number of samples, determining representative samples, and training the model on these samples to detect anomalies, allowing for real-time monitoring without the need for additional sensors and enabling processing on edge devices with limited resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual inspection and continuous maintenance are used to monitor plant components, then the condition of plants and machines can be monitored, but changes or problems are only detected when machines and systems have failed, leading to high costs due to machine and plant downtime
Solution Approach 1:
The patent applies preliminary action by continuously monitoring multiple process parameters and training a machine learning model to detect anomalies before they cause machine failure. The system performs preliminary detection of gradual changes in the drive system, allowing maintenance to be scheduled before actual failure occurs, thus reducing unplanned downtime and associated costs.
2Measurement precision
If a large amount of data is collected for detailed assessment of drive system condition, then monitoring accuracy can be improved, but significant computing power and data storage are required, preventing evaluation close to the drive system
Solution Approach 1:
The patent extracts only the essential features and patterns from the large volume of collected process parameter data through machine learning model training. The model learns to identify relevant anomalies and gradual changes by processing historical data offline, then uses this compressed knowledge for real-time monitoring with minimal computational resources during operation.
Solution Approach 2:
The machine learning model is trained in advance using large datasets to capture complex patterns and relationships in drive system behavior. This preliminary training phase performs the computationally intensive data processing, allowing the deployed model to make accurate predictions with minimal real-time computing requirements.
3Reliability
If conventional monitoring functions and analysis options in the converter are used, then active protection can be provided, but the analysis is only performed retroactively and not during the occurrence of damage
Solution Approach 1:
The patent implements continuous feedback monitoring by constantly comparing current process parameter values against the trained machine learning model's expectations. The system provides real-time feedback about the drive system's health status, enabling detection of anomalies during their occurrence rather than retroactively, allowing for immediate corrective action.
Data Source
AI summary
A computer-implemented method of providing a machine learning model for condition monitoring of an electric power converter is provided. The method includes: obtaining a first batch of input data that includes a number of samples of one or more operating parameters of the converter during at least one operating state of the converter; reducing the number of samples of the first batch by clustering the samples of the first batch into a first set of clusters, (e.g., according to a first clustering algorithm, e.g., based on a clustering feature tree, such as BIRCH), and determining at least one representative sample for each cluster; providing the representative samples for training the machine learning model; and/or training the machine learning model based on the representative samples.