O&M Control Using Dissimilar AI Models for Robust Predictions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current artificial intelligence systems are 'greedy, brittle, opaque, and shallow', requiring vast amounts of data, prone to breaking in new contexts, unexplainable, and lacking innate knowledge, leading to fragility and inefficiency in large complex systems.
Innovation Solution
Implement a system and method using dissimilar models to generate a final prediction through dynamic weightings, incorporating sensor data and executing operational or maintenance decisions based on these predictions, while ensuring robustness and explainability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single AI model is used for predictions, then the system is simpler to implement, but the system becomes brittle and prone to breaking in new contexts
Solution Approach 1:
The patent combines multiple dissimilar AI models (e.g., neural networks, decision trees, support vector machines) into an ensemble system where each model contributes to the final prediction. This merging of different model types creates a more robust system that can handle new contexts better than any single model alone, directly resolving the contradiction between simplicity and robustness.
Solution Approach 2:
The ensemble of dissimilar AI models functions like a composite material, where each model type contributes different strengths and properties. Just as composite materials combine different substances to achieve superior properties, the ensemble combines different model architectures to achieve superior robustness and adaptability while maintaining reasonable system complexity.
2Measurement precision
If vast amounts of data are used for training, then model accuracy improves, but the system becomes greedy and inefficient
Solution Approach 1:
The patent applies partial action by using only the necessary amount of data for training each model in the ensemble, rather than exhaustively training on all available data. Each dissimilar model is trained on appropriate subsets, achieving sufficient accuracy without the excessive data consumption that would reduce system efficiency and productivity.
Solution Approach 2:
The data training process is segmented across multiple models, where different portions or aspects of the data are used to train different model types in the ensemble. This segmentation allows the system to achieve high prediction accuracy through diverse model perspectives without requiring any single model to consume vast amounts of data, thereby maintaining efficiency.
3Measurement precision
If complex AI models are deployed, then prediction capability improves, but the system becomes opaque and unexplainable
Solution Approach 1:
The ensemble system segments the prediction task across multiple dissimilar models, where each model's decision process can be individually examined and explained. This segmentation allows stakeholders to understand which specific model contributed which aspect of the prediction, maintaining explainability while achieving high prediction capability through the collective strength of multiple models.
Solution Approach 2:
The ensemble mechanism acts as an intermediary that aggregates predictions from multiple dissimilar models in a transparent manner. This intermediary layer provides explainability by showing how individual model predictions are combined, allowing the system to maintain high prediction capability while preserving understanding of the decision-making process.
Data Source
AI summary
Disclosed are systems and methods for operations and maintenance (O&M) systems for controlling the operation of a second system, wherein the second system includes a plurality of objects (i.e., components and/or subsystems). The O&M system comprises means for evaluating at least one of the plurality of objects using a first model to produce a first prediction of a characteristic of the object; evaluating at least one of the plurality of objects using a second model to produce a second prediction of the characteristic of the object, the second model being dissimilar to the first model; generating a final prediction of the characteristic of the object as a function of dynamic weightings of the first prediction and the second prediction; and, executing an operational or maintenance decision with respect to the second system as a function of the final prediction of the characteristic of the at least one of the plurality of objects.