Elevator Sensor Calibration Using Transfer Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Elevator sensor systems face challenges in precise calibration due to variations in sensor manufacturing and installation, leading to inconsistent responses and reduced fault detection accuracy.
Innovation Solution
The method involves applying a known excitation to the elevator system and using transfer learning to calibrate a trained model based on response changes, shifting baseline and fault designations, and adjusting fault detection boundaries to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If sensor systems are manufactured and installed with standard tolerances, then manufacturing cost and installation ease are improved, but sensor system response precision deteriorates due to variations in sensor characteristics and installation effects
Solution Approach 1:
The patent applies parameter changes by using transfer learning to transform the sensor response data through learned parameter adjustments. The system learns optimal parameter transformations that map varied sensor responses to a standardized reference space, effectively compensating for manufacturing and installation variations without requiring precise physical measurements during calibration
Solution Approach 2:
The patent uses copying by creating a reference sensor system response model that represents ideal sensor behavior. Actual sensor responses are then compared against this copied reference model, and transfer learning is applied to align the actual responses with the reference, enabling calibration without physical access to the original reference sensor
2Ease of operation
If sensor system calibration is performed without transfer learning, then calibration process simplicity is improved, but fault detection accuracy deteriorates due to inconsistent sensor responses across different installations
Solution Approach 1:
The patent implements self-service by enabling the sensor system to automatically calibrate itself through transfer learning. The system uses readily available operational data and pre-trained models to perform self-calibration without requiring external expert intervention or complex manual calibration procedures, thus maintaining simplicity while improving reliability
Solution Approach 2:
The patent applies preliminary action by pre-training sensor models using reference data collected during manufacturing or from ideal conditions. These pre-trained models serve as a foundation that is then adapted to specific installations through transfer learning, eliminating the need for extensive on-site calibration while ensuring accurate fault detection from the outset
3Measurement precision
If transfer learning is applied for model calibration, then fault detection precision is improved, but computational complexity and calibration time increase
Solution Approach 1:
The patent applies partial action by implementing transfer learning in a staged manner, performing calibration only on the most critical sensor parameters that have the greatest impact on fault detection accuracy. This selective approach achieves sufficient precision improvement without the full computational burden of calibrating all sensor parameters, thus reducing calibration time while maintaining effectiveness
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of elevator sensor system (220) calibration includes collecting, by a computing system, a plurality of data from one or more sensors (214) of an elevator sensor system while a calibration device (222) applies a known excitation. The computing system compares an actual response to an expected response to the known excitation using a trained model. The computing system performs analytics model calibration to calibrate the trained model based on one or more response changes between the actual response and the expected response.