Motor-Driven Roller Belt Tension Inference Using Isolation Forest
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods cannot accurately measure dynamic belt tension during motor operation, affecting motor-driving performance due to the inability to assess tension changes in a rotating belt.
Innovation Solution
A machine learning-based dynamic belt tension inference method using the Isolation Forest algorithm trains a model to infer dynamic tension by processing real-time data from a motor driven roller system, determining whether the tension is within acceptable user settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If contact-type mechanical tension gauges or non-contact infrared/ultrasonic tension gauges are used to measure belt tension, then static tension measurement is achieved, but dynamic tension measurement capability is lost
Solution Approach 1:
The patent replaces mechanical tension gauges and physical sensors with a machine learning-based inference system. The system uses the Isolation Forest algorithm to analyze motor operating parameters (current, torque, speed) and infer dynamic belt tension without physical contact, thereby achieving both static and dynamic tension measurement capabilities while eliminating mechanical measurement limitations
Solution Approach 2:
The patent transforms the measurement approach by changing from direct physical measurement to indirect parameter-based inference. By monitoring changes in motor current, torque, and speed parameters and feeding them into the trained ML model, the system derives dynamic tension values that adapt to varying operating conditions, resolving the contradiction between static measurement precision and dynamic adaptability
2Measurement precision
If machine learning model training with multiple hyper-parameter sets is performed, then model performance is optimized, but training time and computational resources increase
Solution Approach 1:
The patent performs preliminary model training during system setup or maintenance periods when real-time performance is not critical. Multiple hyper-parameter sets are trained in advance, and the best-performing model is selected for deployment. This separates the time-consuming training phase from the real-time inference phase, allowing optimization without impacting operational response time
Solution Approach 2:
The patent implements a balanced approach by training multiple models with different hyper-parameter sets (excessive action during training) but deploying only the optimal model for real-time use (partial action during operation). This ensures high inference accuracy while managing training resources efficiently, as the exhaustive training is performed once rather than continuously
Data Source
AI summary
A dynamic belt tension inference method includes steps of: (A) performing a training process and (B) performing an inference process. Step (A) includes steps of: using Isolation Forest algorithm to train a tension inference model; using Isolation Forest algorithm under the same hyper-parameter set to perform multiple trainings to generate multiple tension inference models; respectively computing multiple model performances of the multiple tension inference models according to the anomaly score and a pre-recorded data label; computing an averaged model performance; determine whether multiple averaged model performances have been acquired; selecting one of the multiple hyper-parameter sets that corresponds to an optimal averaged model performance as a final hyper-parameter set for training to output an final model. Step (B) includes a step of: inferring processed data by the final model to generate the anomaly score and the dynamic tension corresponding to the anomaly score.