Aquatic Maintenance Prediction Using Multi-Sensor ML Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital aquatic installation control and monitoring systems are inaccurate and inefficient in predicting maintenance needs, leading to timely and costly issues, as they operate on a linear basis and fail to account for interactions between operational parameters, resulting in inadequate maintenance scheduling and resource management.
Innovation Solution
A predictive maintenance system utilizing a trained machine learning model that integrates data from various sensors, including those embedded in submersible and floating vehicles, to forecast maintenance requirements and optimize scheduling, reducing the risk of operational degradation and resource overuse.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If linear control and monitoring systems are used to compare operational parameters to thresholds, then the system is simple to operate, but the accuracy of maintenance prediction deteriorates
Solution Approach 1:
The patent replaces the mechanical linear threshold-comparison system with a machine learning-based predictive system. The ML model processes multiple operational parameters simultaneously, capturing complex interactions between parameters that linear systems miss. This substitution enables accurate maintenance prediction while maintaining ease of operation through automated model execution.
2Measurement precision
If alert threshold values are compensated to account for parameter interactions, then the accuracy of maintenance prediction is improved, but the system becomes overly conservative and generates false alerts
Solution Approach 1:
The patent transforms the approach from adjusting threshold values to changing the fundamental parameters processed by the system. The machine learning model uses multiple operational parameters as inputs and predicts maintenance needs based on learned patterns, rather than comparing single parameters against compensated thresholds. This eliminates false alerts while maintaining accuracy.
3Device complexity
If traditional control systems emit alerts based on single parameter thresholds, then the device complexity is low, but the reliability of maintenance scheduling deteriorates
Solution Approach 1:
The patent segments the maintenance scheduling problem into distinct operational phases: data collection from multiple sensors, ML model prediction of degradation events, and generation of optimized maintenance sequences. This segmentation allows the system to handle complexity in manageable parts while achieving high reliability in maintenance timing.
Solution Approach 2:
The system implements feedback loops where sensor data continuously feeds the ML model, which updates predictions based on actual system behavior. The maintenance scheduling is adjusted based on predicted degradation events, creating a closed-loop system that improves reliability over time while managing complexity through automated feedback processing.
4Reliability
If maintenance operations are performed based on conservative threshold alerts, then the reliability of preventing degradation is improved, but the resource consumption increases
Solution Approach 1:
The patent performs preliminary maintenance actions based on predicted degradation events rather than waiting for actual failures or using conservative thresholds. The ML model forecasts when maintenance will be needed, allowing operators to perform maintenance just in time, preventing degradation while avoiding premature maintenance that wastes resources.
Data Source
AI summary
The aquatic installation predictive maintenance system (100) comprises:at least one physical/chemical sensor (110, 115, 116, 117, 181, 182, 183, 184) interacting with water in at least one aquatic installation and configured to provide series of at least one sensed value representative of a physical/chemical parameter, andat least one processor (120) configured to execute instructions representative of the steps of:operating a trained machine learning model, said model being trained to associate, for at least one series of sensed value representative of a physical/chemical parameter, at least one aquatic installation operational degradation event associated with a date of event occurrence, anddetermining a sequence of maintenance operations to be performed on at least one aquatic installation as a function of at least one predicted aquatic installation operational degradation event and associated date of event occurrence.