ML Condition Monitoring with Adjusted Physical Quantities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Unreliable measured physical quantities used as input in machine learning models for condition monitoring of machines lead to inaccurate output data, which can cause premature component failure and reduced machine lifetime.
Innovation Solution
A data adjusting device and method that estimates and adjusts physical quantities using machine learning models to limit external influences, allowing for more accurate condition monitoring by replacing unreliable measured data with estimated data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If temperature threshold monitoring is used for safety, then machine protection is improved, but component lifetime is reduced due to premature shutdowns
Solution Approach 1:
The system performs preliminary estimation of true component temperatures using machine learning models before threshold violation occurs. By predicting temperature trends and compensating for measurement errors in advance, the system avoids premature shutdowns while still protecting the machine from actual thermal damage.
Solution Approach 2:
The patent introduces an intermediary machine learning model that acts as a mediator between the unreliable temperature sensor and the safety monitoring system. This model compensates for measurement errors and provides a more accurate estimate of the true component temperature, allowing the safety system to make better decisions.
2Device complexity
If ambient temperature is used as reference, then temperature estimation is simplified, but measurement accuracy deteriorates when fan failures occur
Solution Approach 1:
The machine learning model serves as an intermediary that processes the ambient temperature reading along with other sensor data to compensate for fan failure effects. This intermediary layer transforms the unreliable ambient temperature measurement into a more accurate estimate of component temperature without requiring complex hardware changes.
Solution Approach 2:
The system dynamically adjusts the temperature estimation by changing the reference parameters used in the calculation. When fan failure is detected, the model shifts from using ambient temperature as the primary reference to incorporating alternative parameters such as coolant temperature and thermal models, thereby maintaining accuracy despite the original reference becoming unreliable.
3Reliability
If multiple temperature sensors are deployed, then monitoring coverage is improved, but system complexity and cost increase
Solution Approach 1:
The machine learning model acts as a virtual sensor that estimates temperatures at locations where physical sensors are not installed. By using data from existing sensors and thermal models, the intermediary model provides comprehensive monitoring coverage without requiring additional hardware sensors, thus avoiding increased system complexity.
Solution Approach 2:
Instead of deploying physical sensors throughout the system, the patent creates virtual copies of temperature measurements through machine learning predictions. These virtual sensor readings are generated by modeling the thermal behavior of components based on data from a limited number of actual sensors, providing comprehensive monitoring coverage without the complexity of extensive sensor deployment.
Data Source
AI summary
A method of adjusting data used for detecting faulty conditions of a machine comprises estimating, in a first machine learning model, required output data based on input data comprising a measured first physical quantity associated with the machine and the output data comprising an estimated second physical quantity of the machine, processing at least some of the input data for obtaining an adjusted first physical quantity having limited external influences, and applying the adjusted first physical quantity together with the input data without the measured physical quantity in the first machine learning model for obtaining modified output data for use in detecting faulty conditions of the machine, which modified output data comprises an adjusted estimated second physical quantity having limited external influences.