Elevator Car Load Estimation Using Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Elevator systems face challenges in accurately determining load data without access to the elevator control system, particularly in remote monitoring or third-party systems, and existing sensor solutions are costly and sensitive to installation and calibration errors.
Innovation Solution
A method using a reinforcement learning model to estimate elevator car load based on condition data from loading events and movement cycles, incorporating parameters like rope elongation, hoisting machine tilt, and ambient conditions, and utilizing sensor devices for input data processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high quality sensor measurements are used to ensure accuracy and reliability of load data, then measurement precision and reliability are improved, but sensor cost and installation complexity increase
Solution Approach 1:
The patent replaces complex mechanical load sensors with a reinforcement learning model that processes easily obtainable operational data (current, speed, position) to estimate load. This substitutes a mechanical measurement system with an intelligent software-based system that achieves accurate load estimation without requiring sophisticated sensor installation or calibration.
2Loss of information
If access to elevator control system is required to obtain load data, then load data availability is improved, but system accessibility and ease of operation worsen
Solution Approach 1:
The reinforcement learning model acts as an intermediary that processes publicly available operational data (current, speed, position) to derive load information without requiring direct access to the elevator control system's internal load data. This mediator approach enables load estimation in third-party and remote monitoring scenarios where control system access is restricted.
3Reliability
If sophisticated sensor structures are used to meet measurement requirements, then measurement reliability is improved, but manufacturing cost and tolerance sensitivity increase
Solution Approach 1:
The patent replaces sophisticated mechanical sensor structures with a software-based reinforcement learning model that processes standard electrical measurements. This substitution eliminates the need for precision-manufactured sensors with tight tolerance requirements, significantly simplifying the manufacturing and calibration processes while maintaining measurement reliability.
Data Source
Figure 1~4
Figure 5~7
AI summary
The invention relates to a method for producing loaddata (430) of an elevator car (110) of an elevator system (100). The method comprises: obtaining (310) condition data (410) comprising at least one loading condition parameter being affected by the load of the elevator car (110), wherein the condition data (410) is obtained during a loading event of the elevator car (110) at a loading landing or during an elevator car movement cycle between a loading landing and a destination landing; using (320) the obtained condition da-ta as input data of a reinforcement learning model (420); processing (330) the input data with the reinforcement learning model (420) to produce output data comprising the load data (430) of the elevator car(110) representing an estimate of the load of the elevator car (110); and using (340) the produced load data (430) of the elevator car (110) in controlling of the elevator system (100) and/or in condition monitoring of the elevator car (110). The invention relates also to an elevator computing unit (220), a load estimation system (200), a computer program product (725), and a computer-readable medium for producing load data(430) of an elevator car (110) of an elevator system (100).