A research article published in the Beijing-based National Science Review proposes an end-to-end diagnostic framework applicable to diverse manufacturing systems.
Artificial intelligence is envisioned to play an increasingly central role in the manufacturing industry due to its potential to predict faults before they occur. Sensory data measured in manufacturing processes, such as vibration, pressure, temperature and energy data, can be used to write up a predictive AI algorithm.
In this way, run-to-failure maintenance could be replaced by predictive maintenance which would reduce cost and increase reliability of the machinery. However, existing diagnosis and monitoring techniques are not generalizable to different manufacturing applications, and a universal framework that requires only “simple tuning of parameters” is desired.
The new approach, proposed by Ye Yuan from the School of Artificial Intelligence and Automation and Han Ding from the State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, exploits the predictive power of a deep learning technology called convolutional neural networks (CNNs) to automatically extract hidden degradation features from noisy time-course data.
It has been called a general end-to-end framework as CNNs can “extract features automatically and solve the problems accurately” without “any dependence on prior knowledge”, according to the paper.
The theory has been tested on ten representative data sets drawn from a wide variety of manufacturing applications. Results reveal that it performed well in examined benchmark applications and higher than 95% accuracy in diagnosing problems was achieved, indicating its potential role in advancing smart manufacturing. [APBN]