abstract
- © 2003-2012 IEEE.The force-displacement profile is a key parameter in manufacturing electric motor thermo-protectors, hence its accurate estimation helps preventing malfunctions due to overheating and/or short-circuits. In this research, we propose a novel force profile characterizer based on a Machine Learning algorithm (ML), the Extreme Learning Machine (ELM). Here, we combine the ELM with a Partial Average Test filter (PAT) to predict the behavior of thermostatic bimetallic strips. The computational efficiency inherent to ELMs allows the use of the algorithm (PAT-ELM) in real manufacturing environments, where computational resources tend to be limited and response time is of the utmost importance. The algorithm results were compared with actual measurements taken from production samples following ASTM B106-08, and the force-displacement profile of the thermostatic bimetallic strips measurements. The results show a correlation in excess of 86% including batches smaller than 50 samples. This result was constant even in cases where measurements were affected by noise present in industrial environments. The time required to obtain the strength profile was significantly lower than alternative methods, making this algorithm suitable for IoT systems.