Supervised Vibration Diagnosis for More Reliable Machine Fault Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive maintenance programs for machines, such as motors and pumps, rely on periodic data collection and rule-based automated diagnosis, often resulting in inaccurate and unreliable fault predictions due to the lack of specialized technicians and the complexity of analyzing vibration, temperature, and rotation speed data.
Innovation Solution
A machine diagnostic system employing supervised learning and Bayesian probability models, including naive Bayesian probability sub-networks and family neural networks, to analyze machine data at both test point and overall machine levels, providing accurate fault diagnosis and health rating with confidence levels, and allowing for user feedback to improve model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If rule-based automated diagnosis is used to reduce the need for specialized technicians, then the need for specialized technicians is reduced, but the accuracy and reliability of diagnosis results deteriorate due to false positive and false negative results
Solution Approach 1:
The system implements feedback loops where expert technicians review automated diagnosis results and provide corrections. These corrections are fed back to retrain and refine the machine learning models, continuously improving diagnosis accuracy while maintaining automated operation. The feedback mechanism allows the system to learn from expert judgments and reduce false positives and negatives over time.
Solution Approach 2:
The patent introduces machine learning models as intermediaries between the automated diagnosis system and final diagnosis results. These models process sensor data and provide probabilistic fault assessments that combine automated efficiency with improved accuracy. The intermediary layer filters and refines raw automated diagnosis outputs before presenting final results, reducing false positives and negatives while maintaining automation.
2Ease of operation
If periodic data collection is used for predictive maintenance, then maintenance scheduling is simplified, but the timeliness of fault detection deteriorates leading to unexpected failures
Solution Approach 1:
The system implements periodic data collection at optimized intervals based on machine operating conditions and fault risk assessments. Rather than fixed scheduling, the periodic action adapts to actual machine states, collecting data more frequently when anomalies are detected and less frequently during normal operation. This maintains ease of scheduling while improving timely fault detection through condition-based triggering.
Solution Approach 2:
The diagnostic system performs self-monitoring and automatic alerting, enabling the machine to effectively detect and report its own faults. The system continuously analyzes sensor data and automatically triggers maintenance alerts when faults are detected, eliminating the need for manual monitoring while maintaining timely detection. This self-service capability bridges the gap between simplified scheduling and timely fault detection.
3Productivity
If automated diagnosis systems are implemented to avoid unexpected failures, then production continuity is improved, but the system complexity increases requiring sophisticated algorithms and models
Solution Approach 1:
The diagnostic system is segmented into modular components: sensor data acquisition modules, machine learning model modules, analysis modules, and alerting modules. Each module performs a specific function and can be independently configured and maintained. This segmentation reduces overall system complexity by breaking down the sophisticated diagnosis system into manageable, interchangeable units that can be deployed incrementally.
Solution Approach 2:
The patent implements universal machine learning models that can diagnose multiple fault types across different machine configurations using the same core architecture. The system is designed to handle various sensor types, machine types, and fault conditions through a unified framework, reducing complexity by avoiding the need for separate specialized systems for each application scenario.
Data Source
AI summary
Supervised learning is implemented to improve the accuracy of automated diagnoses performed by monitoring units installed at a machine. The monitoring units perform indicator acquisition and automated diagnoses based on a Bayesian model derived in accordance with the machine's known configuration. Raw data is collected, including machine vibration data and other diagnostic data. The data is analyzed to diagnose for specific fault defect assumptions so as to generate the automated diagnoses results and a rating for overall health of the machine. The results are uploaded to an external environment that can be accessed by an expert for review and correction. Based upon the expert's corrections, the Bayesian model is adjusted using supervised learning to improve the automated diagnoses performed by the monitoring units.