Anomaly Detection With Bayesian Failure Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing PHM algorithms struggle to accurately predict failure in complex engineering assets like maritime vessels, particularly due to transient events that cause false positives and early intervention, and lack the capability to estimate remaining useful life with associated probability values, leading to inefficient maintenance and potential increased risk of failure.
Innovation Solution
A system and method utilizing a trained machine learning algorithm and Bayesian model to predict failure probability based on real-time sensor data, incorporating a look-back time and horizon time, which generates a probability of failure and triggers alerts or self-protection protocols when exceeding a threshold, while using rolling averages to mitigate transient effects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If classic machine learning algorithms are used for anomaly detection, then the system can detect faulty states, but transient events cause false positives leading to early intervention and loss of operator confidence
Solution Approach 1:
The system performs preliminary classification of transients as benign or malicious using a trained machine learning algorithm before triggering failure predictions. This preliminary action filters out benign transients that would otherwise cause false positives, allowing the system to maintain high reliability while avoiding premature maintenance interventions
Solution Approach 2:
A Bayesian model acts as an intermediary between the anomaly detector and the failure prediction system. It combines the anomaly probability with historical transient data to calculate a refined failure probability, mediating the interaction between detection sensitivity and false positive reduction
2Reliability
If anomaly-based fault detection is used, then faulty states can be identified, but the system cannot estimate remaining useful life with probability values
Solution Approach 1:
The system merges anomaly-based fault detection with prognostic estimation by integrating the Bayesian failure probability model with the anomaly detector. This combination enables simultaneous detection of faulty states and estimation of remaining useful life with associated probability values, providing comprehensive reliability information
Solution Approach 2:
The system uses feedback from historical failure data and transient patterns to continuously refine the Bayesian model parameters. This feedback mechanism enables accurate RUL estimation by learning from past system behavior and updating probability predictions based on accumulated operational experience
3Device complexity
If non-deep learning ML anomaly detectors are used in maritime domain, then computation is simpler, but they trigger far in advance of system failure causing excessively early intervention
Solution Approach 1:
The system dynamically adjusts the prediction horizon and probability thresholds based on system state and historical patterns. By changing these parameters adaptively, the system maintains simple non-deep learning algorithms while achieving accurate maintenance timing that avoids excessively early intervention
Solution Approach 2:
The Bayesian model incorporates dynamic adjustment of failure probability calculations based on real-time anomaly patterns and historical transient data. This dynamic approach allows the system to maintain computational simplicity while accurately predicting failure timing, preventing premature maintenance actions
Data Source
AI summary
A method and apparatus for predicting failure of an engineering asset based on real-time data. The system comprises a plurality of sensors for measuring data on the engineering asset. The method of predicting failure comprises receiving a data record comprising data on the engineering asset collected from the plurality of sensors at time t, generating, using a trained machine learning algorithm, a probability PF that the received data record indicates that the engineering asset is in a faulty state; determining what number of data records received in a look-back time Lt are indicative of the engineering asset being in a faulty state, wherein the look-back time Lt is a time period occurring before the time t at which the data record was collected; and predicting a probability of the engineering asset failing during a horizon time Ht, wherein the horizon time Ht is a time period after time t at which the data record was collected. The predicting step implements a Bayes forecasting model to predict the probability of failure based on the generated probability PF and the number of data records which were determined to be faulty within the look-back time Lt.