Rotational Equipment Anomaly Detection Using Frequency-Domain Vibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for monitoring the condition of low-cost rotational equipment, such as pumps in industrial settings, require significant expertise and are unsuitable for large-scale deployment due to complexity and asset-specific calibration needs, making them inefficient for detecting anomalies in low-cost assets.
Innovation Solution
The development of automated systems that use trained anomaly detection models to process vibration data from rotational equipment, generating anomaly scores and alerts based on frequency domain analysis, allowing for asset-agnostic binary anomaly detection without human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning anomaly prediction models are used for detecting states of multi-sensor systems, then measurement precision is improved, but device complexity increases and requires subject matter expertise for calibration
Solution Approach 1:
The system performs self-calibration by automatically determining thresholds for abnormality versus normality through unsupervised learning on historical data, eliminating the need for manual expert calibration. The model learns normal operational patterns and automatically adapts to different assets without requiring subject matter expertise for threshold selection.
Solution Approach 2:
The system transforms the complex multi-parameter deep learning model into a simplified binary classification framework that outputs anomaly scores. By changing the output parameters from continuous predictions to discrete anomaly indicators with automatically determined thresholds, the system maintains high detection accuracy while reducing calibration complexity.
2Measurement precision
If asset-specific deep learning models are implemented, then measurement precision improves, but ease of operation deteriorates due to requiring subject matter expertise
Solution Approach 1:
The system automatically trains and deploys models without requiring subject matter expertise. Historical operational data is used to self-calibrate the models, which then automatically detect anomalies on new assets. This eliminates the need for experts to manually configure or calibrate models for each asset deployment.
Solution Approach 2:
The system creates a universal anomaly detection framework that can be applied across different asset types and locations. By using unsupervised learning on historical data from each asset, the same general model architecture can be deployed universally without asset-specific customization or expert intervention for each deployment scenario.
3Device complexity
If conventional physics-based models are used for monitoring, then device complexity is reduced, but measurement precision deteriorates for detecting abnormal operating states
Solution Approach 1:
The system replaces complex physics-based mechanical models with data-driven machine learning models that automatically learn operational patterns from historical data. This substitution maintains relatively simple system architecture while significantly improving the accuracy of detecting abnormal operating states through pattern recognition in multi-sensor data.
Solution Approach 2:
The system transitions from physics-based parameter thresholds to data-driven anomaly scores. By changing from predetermined physical thresholds to dynamically learned patterns in sensor data, the system achieves higher detection precision while maintaining manageable system complexity through automated model training.
4Measurement precision
If manual calibration and expert intervention are required, then measurement precision can be maintained, but productivity decreases due to time-consuming calibration processes
Solution Approach 1:
The system automatically calibrates itself using historical operational data from each asset without requiring manual expert intervention. This self-calibration process occurs automatically during initial deployment, enabling rapid scaling to multiple assets while maintaining high detection accuracy through data-driven threshold determination.
Solution Approach 2:
The system performs preliminary automated training and calibration using historical data before deployment. By pre-processing and self-calibrating with available historical operational data, the system eliminates the need for time-consuming manual calibration during deployment, thereby increasing productivity while maintaining detection precision.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables cost-effective, automated monitoring of large numbers of low-cost rotational equipment by distinguishing normal from abnormal operation, reducing the need for expert calibration and facilitating early detection of deviations in operational states.
Implementation Method 1
These sensors can be integrated into sensor packages that transform raw time series sensing data into a set of features in the frequency domain
Data Source
AI summary
A method of detecting deviation from an operational state of a device includes obtaining preprocessed data corresponding to data sensed by one or more sensor devices coupled to the device, where obtaining the preprocessed data includes applying a transform to the data sensed by the one or more sensor devices to generate a set of features in a frequency domain. The method also includes processing the preprocessed data using a trained anomaly detection model to generate an anomaly score. The method also includes processing the anomaly score using an alert generation model to determine whether to generate an alert.