Prognostic Surveillance Parameter Tuning for Detection-Cost Tradeoffs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ML-based prognostic-surveillance systems face complex tradeoffs among fast anomaly detection, high prognostic accuracy, and low compute cost, making it impossible to optimize all three objectives simultaneously, as improving one objective often degrades the others.
Innovation Solution
A system that allows users to select a functional objective and optimizes operational parameters using Monte Carlo simulations and stochastic gradient descent to achieve the desired objective, with subordinate objectives becoming constraints during optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the prognostic-surveillance system is configured to improve fast anomaly detection, then the detection speed increases, but the compute cost increases and prognostic accuracy may decrease
Solution Approach 1:
The system dynamically adjusts operational parameters such as the number of signals, samples per signal, and training vectors based on the selected functional objective. When fast anomaly detection is prioritized, the system increases sampling rates and reduces signal filtering, which speeds up detection but increases compute cost. This dynamic configuration allows the system to adapt its resource consumption to match the prioritized objective.
Solution Approach 2:
The optimization system changes key parameters including the number of signals in the inferential model, the number of samples for each signal, signal-to-noise ratios, and the number of training vectors. By adjusting these parameters, the system can shift the balance between detection speed and compute cost. For example, increasing the number of samples improves detection speed but increases computational load, while the system selects optimal values based on the prioritized objective.
2Measurement precision
If the prognostic-surveillance system is configured to improve prognostic accuracy, then the accuracy of anomaly detection increases, but the detection speed decreases and compute cost increases
Solution Approach 1:
When prognostic accuracy is the prioritized objective, the system applies partial action by using a smaller number of signals in the inferential model and fewer samples per signal compared to fast detection mode. This reduces computational complexity and speeds up processing while maintaining sufficient accuracy for the application. The system performs only the necessary computations to achieve the required accuracy level without excessive processing.
Solution Approach 2:
The system adjusts parameters such as reducing the number of signals, decreasing samples per signal, and modifying training vector counts to optimize for accuracy. These parameter changes allow the system to achieve high prognostic accuracy while managing the tradeoff with detection speed by selecting parameter values that provide sufficient accuracy without excessive computational burden.
3Loss of energy
If the prognostic-surveillance system is configured to reduce compute cost, then the memory footprint and processing requirements decrease, but anomaly detection speed and prognostic accuracy deteriorate
Solution Approach 1:
When compute cost reduction is the prioritized objective, the system extracts and removes unnecessary computational elements by using fewer signals in the inferential model and fewer samples per signal. This reduces memory footprint and processing requirements while retaining the essential functionality for anomaly detection. The system removes only the excess computational burden while preserving the core detection capability.
Solution Approach 2:
The system employs simplified models with fewer signals and samples that require less computational resources and memory. These simplified configurations act as lightweight versions that consume less energy and computing power, accepting reduced detection effectiveness as a tradeoff. The system uses these resource-efficient configurations when compute cost is the primary constraint.
4Measurement precision
If the number of signals in the inferential model is increased to improve detection accuracy, then the prognostic accuracy improves, but the compute cost and model complexity increase
Solution Approach 1:
The optimization system directly controls the number of signals parameter in the inferential model. By adjusting this parameter, the system can balance between model complexity and prognostic accuracy. When optimizing for accuracy, the system increases the number of signals; when optimizing for low compute cost or fast detection, it reduces the number of signals. This parameter control allows flexible adjustment of model complexity to match operational requirements.
Data Source
AI summary
The disclosed embodiments relate to a system that optimizes a prognostic-surveillance system to achieve a user-selectable functional objective. During operation, the system allows a user to select a functional objective to be optimized from a set of functional objectives for the prognostic-surveillance system. Next, the system optimizes the selected functional objective by performing Monte Carlo simulations, which vary operational parameters for the prognostic-surveillance system while the prognostic-surveillance system operates on synthesized signals, to determine optimal values for the operational parameters that optimize the selected functional objective.