Probabilistic Fault Detection for Sensor Noise
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional diagnostic frameworks for detecting faulty device components fail to accurately model noise in sensor data, leading to false indications and masking true faults, and are not generic across different types of devices.
Innovation Solution
A method that involves obtaining control samples from a device in a no-fault mode, performing exploratory data analysis, and using a measurement error model to identify faulty components through a likelihood ratio test, while determining an optimal remedy pattern based on historical data and cost information, with the option to include testing or not depending on cost considerations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional diagnostic frameworks use simple mathematical models and knowledge bases, then the device complexity is low, but the measurement precision of sensor data is poor leading to false fault indications
Solution Approach 1:
The patent transforms sensor data by modeling measurement errors as random variables with specific probability distributions. It changes the parameter representation from raw sensor values to probabilistic distributions, enabling statistical analysis that distinguishes true faults from noise while maintaining manageable complexity through standardized statistical models.
Solution Approach 2:
The patent introduces a probabilistic engine as an intermediary layer between sensor data and fault diagnosis. This engine uses measurement error models and statistical distributions to mediate the relationship between noisy sensor readings and component fault status, improving measurement precision without requiring direct complex analysis of raw sensor data.
2Reliability
If noise modeling is not implemented, then the device complexity remains low, but false fault indications increase due to environmental factors and component unreliability
Solution Approach 1:
The patent performs preliminary action by characterizing measurement errors before fault detection. It collects sensor data during known normal operation, models the measurement errors as probability distributions, and stores these as reference profiles. This preliminary error characterization enables reliable fault detection without complex real-time noise filtering, as the noise profile is pre-established.
Solution Approach 2:
The patent transforms the reliability problem by changing parameters from deterministic fault thresholds to probabilistic distributions. Instead of using fixed thresholds that are susceptible to noise, it models sensor readings and error terms as random variables with specific distributions, enabling statistical hypothesis testing that reliably distinguishes true faults from noise variations.
3Reliability
If comprehensive testing is performed on all faulty components, then the reliability of fault identification is high, but the cost and time consumption increase significantly
Solution Approach 1:
The patent applies partial action by performing statistical analysis on a subset of components rather than exhaustive testing of all components. The probabilistic engine calculates likelihood ratios for each component based on sensor data and measurement error models, allowing the system to identify the most probable faulty components with high confidence without testing every component, thus reducing time while maintaining reliability.
Solution Approach 2:
The patent applies local quality by focusing diagnostic resources on specific components with highest fault probability. The system calculates component-specific likelihood ratios and prioritizes investigation of components with values indicating higher fault probability, rather than uniformly testing all components. This localized approach maintains high reliability for critical components while reducing overall diagnosis time.
4Adaptability or versatility
If device-specific diagnostic models are created for each device type, then the measurement precision for that specific device is high, but the adaptability to different device types is reduced
Solution Approach 1:
The patent achieves universality by creating a device-agnostic probabilistic diagnostic framework. The measurement error models, probability distributions, and hypothesis testing methods are formulated in a general way that can be applied to any device type. The system collects device-specific sensor data and applies the same statistical principles universally, enabling adaptation to medical devices, communication devices, electronic devices, and other types without requiring device-specific model redesign.
Data Source
AI summary
A method, non-transitory computer readable medium, and anomaly detection computing apparatus that detects one or more of a plurality of symptoms associated with a device when the device is operating in a fault mode. One or more of a plurality of components of the device that are potentially faulty are identified based on the detected symptoms. One or more tests are performed on each of the one or more of the components to confirm that at least a subset of the one or more of the components is faulty. An optimal remedy pattern is determined for the subset of the one or more of the components based at least in part on historical data and cost data obtained from a database, and the optimal remedy pattern is output.