Machine Health Monitoring With Dynamic ML Anomaly Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional prognostic and health management systems face challenges in efficiently processing and merging large amounts of data in real-time, leading to delayed anomaly detection and reduced accuracy in failure prediction due to the complexity and time-consuming nature of data processing.
Innovation Solution
A machine-learning-based prognostic and health management system dynamically selects and applies three types of damage alert models - complete-life-cycle, failure-free, and value-to-image models - based on data completeness and anomaly risk, allowing for customized data processing and simplification, thereby improving data processing rates and anomaly judgment accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional data processing methods are used to merge and process large amounts of sensor data, then data completeness can be maintained, but data processing time increases significantly and real-time anomaly detection is delayed
Solution Approach 1:
The patent segments the data processing task by dividing sensor data into different completeness levels (first level: complete life cycle data, second level: failure-free data, third level: value-to-image data). This segmentation allows the system to process different types of data through specialized models rather than treating all data uniformly, thereby reducing overall processing time while maintaining detection accuracy.
Solution Approach 2:
The system dynamically selects which damage alert model to apply based on the completeness level of the input data. This dynamic adaptation allows the system to choose the most appropriate processing path for each data set, optimizing both processing speed and accuracy according to the specific characteristics of the data being analyzed.
2Measurement precision
If comprehensive damage alert models are trained on complete life cycle data, then anomaly detection accuracy is improved, but model training complexity and time consumption increase
Solution Approach 1:
The patent creates multiple specialized damage alert models segmented by data completeness level. Instead of training one comprehensive model on all data types, the system trains separate models for different data completeness scenarios (complete life cycle, failure-free, value-to-image). This segmentation reduces the complexity of each individual model while collectively maintaining high prediction accuracy across all data types.
Solution Approach 2:
The system applies partial training approaches by creating models trained on specific subsets of data (e.g., failure-free data for normal operation detection, complete life cycle data for comprehensive failure analysis). This partial action approach allows each model to specialize in particular detection scenarios, reducing overall training complexity while maintaining comprehensive coverage through model selection.
3Loss of information
If multiple types of sensor data are collected and stored for comprehensive analysis, then data completeness is improved, but data management burden and storage requirements increase
Solution Approach 1:
The patent segments stored data into different completeness levels and categories (complete life cycle operation records, failure-free operation records, value-to-image converted data). This segmentation allows the system to manage large volumes of data more efficiently by organizing them into distinct groups that can be processed by appropriate specialized models, reducing the management burden while preserving data completeness.
Solution Approach 2:
The system introduces an intermediary classification mechanism that automatically categorizes incoming sensor data into appropriate completeness levels and routes them to corresponding damage alert models. This intermediary layer simplifies data management by automating the organization and routing process, reducing manual intervention requirements while maintaining comprehensive data utilization.
Data Source
AI summary
A machine-learning-based prognostic and health management system comprises a machine sensor, an instruction receiver, a processor, and an annunciator. The machine sensor is configured to dynamically receive data of a machine under test associated with operations of the machine under test. The instruction receiver is configured to dynamically receive a model-assigning command. The processor is configured to dynamically apply a damage alert machine-learning model corresponding to the model-assigning command for processing the data of the machine under test to predict an anomaly probability of an anomaly occurrence of the machine under test. The processor also dynamically generates, according to the anomaly probability, a damage possibility warning on the machine under test, and determine whether to keep the machine under test running or not.