Neural Network Outlier Detection for Predictive Maintenance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning models, such as neural networks, struggle to handle outliers during prediction, leading to undefined behavior and potential undetected errors, especially in critical systems like autonomous cars and power plant control, as they are not equipped to recognize new types of anomalies or provide reliable uncertainty estimates.
Innovation Solution
A method involving a deep neural network that uses a latent variable model to identify outliers by mapping intermediate output values to projected values in a lower-dimensional subspace, allowing for the detection of both known and unknown anomalies and assessing the reliability of predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep neural networks are used for anomaly detection, then prediction accuracy for known fault types is improved, but the system cannot handle previously unknown anomalies or provide reliable uncertainty estimates
Solution Approach 1:
The system segments the anomaly detection task into two independent components: a supervised deep neural network for detecting known fault types and an unsupervised outlier detection mechanism for identifying previously unknown anomalies. This segmentation allows each component to specialize in its respective strength without compromising the other, resolving the contradiction between prediction accuracy for known faults and adaptability to unknown anomalies.
Solution Approach 2:
The system introduces intermediate uncertainty estimates as a mediator between the deep neural network predictions and the final anomaly detection result. These uncertainty estimates serve as a bridge that enables the system to provide reliable confidence measures for predictions while simultaneously detecting both known and unknown anomalies, thus resolving the contradiction between precision and versatility.
2Loss of information
If supervised modeling is used to predict fault types, then actionable insights are provided, but the system fails when encountering previously unseen anomaly patterns
Solution Approach 1:
The system performs preliminary unsupervised outlier detection before applying supervised classification. By first identifying whether an input pattern is an outlier using unsupervised methods, the system prepares a reliability assessment in advance that prevents the supervised model from making unreliable predictions on unknown patterns, thus maintaining both actionable insights and reliability.
Solution Approach 2:
The system implements feedback through uncertainty estimates that are continuously generated during prediction. When the unsupervised outlier detection identifies high uncertainty for a given input, this feedback signal indicates that the supervised model's prediction should be treated with caution, thereby maintaining reliability even when actionable insights are generated for known patterns.
3Reliability
If routine maintenance is performed, then equipment reliability is maintained, but excessive maintenance costs are incurred
Solution Approach 1:
The system transitions from static routine maintenance schedules to dynamic condition-based maintenance. By continuously monitoring equipment state through both supervised fault detection and unsupervised outlier detection, the system dynamically adjusts maintenance timing based on actual equipment conditions, maintaining reliability while avoiding unnecessary maintenance costs.
Solution Approach 2:
The system changes the maintenance parameter from fixed time intervals to variable condition thresholds. By using uncertainty estimates and outlier detection results as dynamic parameters that trigger maintenance actions, the system optimizes the balance between maintaining equipment reliability and minimizing maintenance costs, performing maintenance only when actually needed.
Data Source
AI summary
An example method comprises receiving a new observation characterizing at least one parameter of an entity; inputting the new observation to a deep neural network having hidden layers; obtaining a second set of intermediate output values that are output from at least one of the hidden layers by inputting the received new observation to the deep neural network; mapping the second set of intermediate output values to a second set of projected values; determining whether or not the received new observation is an outlier with respect to the training dataset based on the latent variable model and the second set of projected values, calculating a prediction for the new observation; and determining a result indicative of the occurrence of at least one anomaly in the entity based on the prediction and the determination whether or not the new observation is an outlier.