Active Learning Input Selection for Safe Physical System Operation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for operating physical systems are inefficient and costly due to redundant measurements and lack of effective strategies for selecting informative inputs, leading to unsafe operating modes and increased wear.
Innovation Solution
A computer-implemented method using multi-output Gaussian processes to determine informative inputs based on information gain and uncertainty, ensuring safe operation by selecting inputs that maximize information gain and probability of safe operation, thereby reducing redundant measurements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional measurement methods are used to operate physical systems, then comprehensive data collection is achieved, but redundant measurements increase cost and time consumption
Solution Approach 1:
The patent changes the parameter of measurement selection from exhaustive to selective by introducing an acquisition function that evaluates information gain. This function dynamically determines which inputs warrant measurement based on their potential to reduce uncertainty about system behavior, thereby eliminating redundant measurements while preserving essential data quality.
Solution Approach 2:
The system performs self-service by autonomously selecting which measurements to take through the acquisition function. The trained multi-output Gaussian process model evaluates potential inputs and automatically identifies those that will provide maximum information gain, enabling the system to self-optimize its measurement strategy without external intervention.
2Reliability
If more inputs are measured to ensure safe operation, then system safety is improved, but measurement cost and wear increase
Solution Approach 1:
The patent transforms the safety assurance approach from measuring all inputs to selectively measuring high-value inputs. The acquisition function incorporates safety considerations by evaluating how each potential measurement reduces uncertainty about safe operating conditions, changing the parameter of measurement selection from quantity-based to value-based.
Solution Approach 2:
The multi-output Gaussian process model serves as an intermediary between the physical system and measurement apparatus. It processes potential input evaluations and translates them into informed measurement decisions, acting as a mediator that identifies which measurements will most effectively ensure safety without requiring exhaustive measurement of all system inputs.
3Loss of information
If exhaustive measurement is performed to evaluate all operating modes, then complete system understanding is achieved, but productivity decreases due to redundant measurements
Solution Approach 1:
The patent changes the parameter of measurement evaluation from uniform to differential by introducing the acquisition function. This function assigns different information gain values to different potential inputs based on the current state of knowledge, enabling the system to prioritize measurements that will most improve understanding while skipping redundant ones, thus maintaining completeness while boosting productivity.
Solution Approach 2:
The system implements feedback through the iterative process where measurements are taken, the multi-output Gaussian process model is updated with new data, and the acquisition function is re-evaluated. This feedback loop ensures that system understanding continuously improves while automatically adapting the measurement strategy to avoid redundancy, thereby maintaining productivity.
Data Source
AI summary
Active learning for operating a physical system. The method includes: providing a data set that comprises data points each comprising an input for operating the physical system, and a first and second observation of the physical system; training a multi-output Gaussian process for predicting the first observation for a given input with the data set; training a Gaussian process for predicting the second observation for a given input with the data set; determining with the data set an input for operating the physical system; determining the first and second observations that result from operating the physical system with the determined input; and adding a data point to the data set that comprises the determined input and the determined first and second observations.