Fluid Flow Analysis Using Unsupervised Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fluid flow measurement and detection systems require user intervention for installation and data labeling, and fail to effectively apply unsupervised and supervised machine learning techniques for predictive modeling and leak detection in plumbing systems.
Innovation Solution
A system and method that uses unsupervised and supervised machine learning to derive time series of fluid flow events, disaggregate compound events, and communicate findings to users or network addresses, employing techniques like subset sum problems and waveform matching to identify and classify fluid flow patterns without relying on user labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If user labeling and manual teaching phase are used for flow detection systems, then measurement precision can be achieved, but device complexity and ease of operation deteriorate due to extensive user intervention requirements
Solution Approach 1:
The system performs self-labeling by automatically identifying and classifying flow events using machine learning algorithms. The device autonomously teaches itself to recognize different flow patterns (showers, baths, sink usage, toilet flushing) without requiring users to manually label each flow event, thereby eliminating the tedious teaching phase while maintaining measurement precision
Solution Approach 2:
The system pre-trains machine learning models with extensive flow data before deployment. This preliminary training enables the device to automatically recognize and classify flow events upon installation, eliminating the need for on-site user labeling and teaching phases while maintaining accurate measurement capabilities
2Reliability
If traditional flow detection systems are deployed, then basic flow measurement is achieved, but productivity and automation level worsen due to lack of predictive modeling and automated leak detection
Solution Approach 1:
The system continuously monitors flow patterns and uses machine learning to predict expected water usage based on historical data and user behavior patterns. When actual flow deviates from predictions, the system automatically generates leak alerts, providing continuous feedback that improves reliability without requiring manual intervention
Solution Approach 2:
The system replaces manual leak detection methods with automated machine learning-based predictive modeling. The ML algorithms automatically analyze flow patterns, identify anomalies, and detect leaks without human intervention, thereby improving both reliability and productivity simultaneously
3Measurement precision
If comprehensive flow data collection is implemented, then predictive modeling accuracy improves, but loss of time increases due to extensive data labeling requirements
Solution Approach 1:
The system automatically labels flow data using unsupervised machine learning techniques that cluster flow patterns without pre-defined categories. This self-labeling process eliminates the time-consuming manual annotation of flow events while still achieving high predictive modeling accuracy through automated feature extraction and pattern recognition
Solution Approach 2:
The system performs preliminary unsupervised clustering and automated labeling of flow data before predictive modeling. This advance processing of data labeling eliminates the need for time-consuming manual annotation during deployment while ensuring high-quality labeled datasets for accurate predictive models
Data Source
AI summary
A system and method for observing fluid flow behavior at a selected site, deriving judgments and recommendations, and further communicating data, judgments and recommendations. The system receives flow rate information, derives a time series of fluid flow events therefrom, identifies compound events consisting of contemporaneous events, disaggregates compound events by application of an unsupervised model, and applies the unsupervised model to derive a solution space of a subset sum problem-type, wherein historical data of the observed fluid flow is not necessarily accessed. The system derives a prior probability of events associated with an event conditional upon event features and attributes, and thereupon estimates prior probabilities based upon user-derived labels for events from many external sites; and/or derives a posterior probability of labels associated events, conditional upon event features and attributes, and estimates posterior probabilities based upon both prior updated information relating to the selected site and a priori calculated probabilities.


